Incremental learning aims to continuously acquire new knowledge from data streams while maintaining previously learned knowledge. Existing incremental learning methods typically assume that the training (source domain) and testing (target domain) data are identically distributed. However, differences in sensor parameters and imaging conditions inevitably lead to distribution gaps between data collected from different satellites (domains). The ensuing domain shift problem substantially impairs the generalization of continuously learned knowledge from source domains to unseen ones. To tackle this problem, we propose adaptive mixture-of-experts distillation (AMoED) for cross-satellite generalizable incremental remote sensing scene classification (CSGIRSSC). Specifically, AMoED adopts a high-level semantic learning pipeline, in which new knowledge is acquired through the coordinated guidance of multiple domain-specific experts, rather than directly from raw data. This pipeline prevents the model from being exposed to large volumes of newly emerging data, thereby alleviating the erasure of previous knowledge when adapting to new data distributions. Besides, the adaptive mixture of domain-specific experts facilitates the formation of universal class concepts, which exhibit strong generalizability across different domains. During the learning process, an equi-partite subset is constructed for knowledge acquisition and consolidation, accompanied by a shallow style-mixing operation to mitigate the interference of domain discrepancies. Extensive experiments are conducted on four remote sensing scene classification datasets, and the proposed method consistently achieves state-of-the-art performance across various scenarios and settings. The code is released at https://github.com/fuyimin96/AMoED.

Black-box adversarial attacks remain challenging due to limited access to model internals. Existing methods often depend on specific network architectures or require numerous queries, resulting in limited cross-architecture transferability and high query costs. To address these limitations, we propose JAD, a latent diffusion model framework for black-box adversarial attacks. JAD generates adversarial examples by leveraging a latent diffusion model guided by attention maps distilled from both a convolutional neural network (CNN) and a Vision Transformer (ViT) models. By focusing on image regions that are commonly sensitive across architectures, this approach crafts adversarial perturbations that transfer effectively between different model types. This joint attention distillation strategy enables JAD to be architecture-agnostic, achieving superior attack generalization across diverse models. Moreover, the generative nature of the diffusion framework yields high adversarial sample generation efficiency by reducing reliance on iterative queries. Experiments demonstrate that JAD attack offers improved attack generalization, generation efficiency, and cross-architecture transferability compared to existing methods, providing a promising and effective paradigm for black-box adversarial attacks.

Few-Shot Learning (FSL) methods have made significant advancements in natural optical image recognition. These methods rely on fine-tuning models that have already been extensively trained on large-scale datasets. However, pre-trained knowledge can mislead subsequent recognition processes. In the case of Synthetic Aperture Radar (SAR) FSL tasks, the pretraining stage usually uses self-supervised techniques to mitigate the issue of limited data, which can worsen the misguided judgment resulting from pre-trained knowledge. In this paper, a new framework, named Causal Meta-Learning (CML), is proposed to tackle this issue. Firstly, the FSL task is modeled as a causal graph from a causal inference perspective to clearly identify the bias introduced by pre-trained knowledge. Secondly, two distinct backdoor interventions are designed: intra-task and extra-task adjustments, to sever the direct linkage between pretrained knowledge and feature representations. Finally, a dedicated dataset, named mini-MSTAR, is reconstructed based on the Moving and Stationary Target Acquisition and Recognition (MSTAR) dataset to evaluate our framework. It's important to note that the contributions of CML are independent of existing meta-learning-based FSL methods, enabling CML to enhance all of them. Experiments conducted on mini-MSTAR demonstrate the improved recognition capabilities of several baseline models in 3 -way 1 -shot and 3 -way 5 -shot scenarios.

In recent years, visible-infrared object detection has achieved significant progress. However, most existing methods primarily emphasize the shared features between the two modalities while overlooking their feature differences. To address this limitation, we propose the Difference-Guided Modality Fusion Network, which can effectively improve the fusion and detection performance of modalities. Specifically, we propose a cross-modal data augmentation strategy to overcome the limitations of single-modality reliance by exchanging the partial modal information. To further capture and analyze feature differences between modalities, we introduce a differential attention fusion approach that models a difference matrix across modal channels, thereby quantifying and strengthening the salient features of the two modalities. Additionally, we develop a modality-aware dynamic learning mechanism that employs a loss function that can simultaneously focus on the differences and common parts of the modalities, guiding the model to adaptively learn features between the modalities. Experimental results on FLIR, LLVIP and M3FD datasets demonstrate the effectiveness of the proposed method, with mAP reaching 42.3%, 67.5% and 59.0% respectively.

Orthogonal time frequency space (OTFS) modulation has garnered significant interest for its robust performance in fast time-varying channels, making it suitable for mobile underwater acoustic (UWA) communication system. This article introduces OTFS modulation to the UWA system and proposes a low-complexity minimum mean-squared error (MMSE) turbo equalization method. Leveraging the characteristics of UWA channels in the delay-Doppler (DD) domain, the method employs symbol-level MMSE equalization. By focusing processing on signals within the DD domain’s interference range, it reduces the channel matrix size, thereby lowering complexity. Given the long delay spread and large Doppler shift of UWA channels, symbol-level MMSE equalization inherently involves high complexity. To mitigate this, we propose two methods to further reduce the computational load associated with matrix inversion. First, we utilize common blocks in the channel matrix and employ a block iterative matrix inversion algorithm to retain computational results, thereby avoiding repeated inversions of the large dimensional matrix. Additionally, we enhance the diagonal dominance property of the channel matrix using the discrete Fourier transform (DFT) matrix. Subsequently, we approximate the inversion using the second-order Neumann series decomposition, further lowering computational complexity. Simulation results and experimental validations at Danjiangkou Lake demonstrate the efficacy of the proposed low-complexity iterative equalization algorithm.

Open set recognition (OSR) effectively enhances the reliability of pattern recognition systems by accurately identifying samples of unknown classes. However, the decision-making process in most existing OSR methods adheres to an ill-considered pipeline, where classification probabilities are inferred directly from overall feature representations, neglecting the reasoning about inherent relations. Besides, the handling of identified unknown samples is typically restricted to the assignment of a generic "unknown" class label but fails to explore underlying category information. To tackle the above challenges, we propose a new paradigm for OSR, entitled Reason and Discovery (RAD), which comprises two main modules: the Reason Module and the Discovery Module. Specifically, in the Reason Module, the distinction between known and unknown is performed from the perspective of reasoning the matching relations between topological information and appearance characteristics of discriminative regions. Then, the mixture and recombination of relation representations across classes are employed to provide diverse estimations of unknown distribution, thereby recalibrating OSR decision boundaries. Moreover, in the Discovery Module, the identified unknown samples are semantically grouped through a biased deep clustering process for discovering novel category information. Experimental results on various datasets indicate that the proposed method can achieve outstanding OSR performance and good novel category discovery efficacy.

Synthetic aperture radar (SAR) images obtained from multi-sensor systems usually exhibit significant shift in data distribution, known as the domain shift. It is challenging to utilize the relevant knowledge of multi-sensor datasets for SAR image cross-domain classification with general supervised learning methods. While domain adaptation (DA) methods can alleviate the domain shift by aligning distributions, they are limited in handling the scale/resolution variations observed in multi-sensor SAR images. These methods mainly focus on feature representations at a single scale for distribution alignment, which may fail to fully align the distributions across different scales. To solve this problem, we propose a new multi-scale alignment domain adaptation network (MSADAN) for SAR ship cross-domain classification. MSADAN explicitly considers the scale factor, enabling us to overcome the weak generalization observed in existing DA methods when dealing with SAR images exhibiting significant scale/resolution variations. Specifically, we develop a scale-aware feature extractor to effectively capture the multi-scale information present in the datasets, facilitating comprehensive representation learning. Furthermore, we propose a multi-level bilinear fusion (MLBF)-based adversarial learning strategy to overcome the limitations of single-scale feature extraction for distribution alignment, aiming to enhance the generalization ability of the model across domains. In addition, a customized class contrastive loss is designed to improve the inter-class separability and intra-class compactness by penalizing cross-class confusion. Experimental results on datasets demonstrate the superiority of MSADAN in SAR ship cross-domain classification.Note to Practitioners—Ship classification using Synthetic aperture radar (SAR) images proves to be a promising strategy in modern maritime monitoring systems. The primary motivation of this paper is to develop a cross-domain classification system tailored for SAR ship target. In real-world scenarios, SAR data often presents significant resolution variations due to the different imaging modes and conditions across multiple sources. Existing DA methods fail to effectively address this unique challenge in the remote sensing field, resulting in poor cross-domain classification performance. The proposed method considers the scale or resolution variations of the targets during feature learning. Furthermore, explicitly incorporating resolution into the distribution alignment enhances the generalization ability of the model to target tasks when dealing with the significant scale or resolution variations across datasets. Experimental results confirm the practicality and robustness of our SAR ship cross-domain classification method in real scenarios. This is of significant importance in promoting the application of machine learning approaches in practical scenarios. In the future, we plan to integrate the physical scattering information of SAR images for the interpretable cross-domain classification for SAR targets.

Partially occluded target recognition is a pressing issue in synthetic aperture radar (SAR) target recognition. Occlusion causes the loss of crucial information, like target structure details. This paper proposes a new structural relation extraction framework to address partial occlusion. It is achieved through the tailored design of counterfactual samples synthesizing and jigsaw mutual learning (CSS-JML). The ASC model parameters have clear physical meanings, aiding in understanding local structural changes. By integrating SAR and ASC images, effective structural relationship representations are extracted, mitigating occlusion effects. The CSS module is designed to generate occluded counterfactual SAR and ASC images using pairs of target data. There is no longer a requirement for further annotation information because this new generation process is limited by recognition tasks. The JML module employs mutual learning to complete jigsaw puzzle tasks in both modalities. And in this process, we design two types of similarity constraints to facilitate the extraction of unified structural information across different modalities. The FA module interacts with recognition features, facilitating the classification and identification of partially obscured targets. Experimental results on MSTAR-based and FUSARShip-based test datasets with three occlusion patterns demonstrate the method’s superiority in most occluded conditions, confirming its effectiveness in SAR occluded target recognition. Note to Practitioners—The motivation for this paper stems from the need to design a robust and efficient SAR target recognition method under partial occlusion. The structural relationships within samples can provide valuable information for recognizing occluded targets. We designed a counterfactual generation module to generate occluded target samples using paired targets in an unsupervised manner. This module is jointly optimized with subsequent recognition tasks, without the need for additional information. Recognizing the interaction bottleneck in mutual learning tasks, we designed two similarity constraints to promote the extraction of unified representations. Finally, we designed a feature affinity module to introduce structural representation information into the recognition branch. The proposed method achieves excellent classification performance in various occlusion environments.

Multi-expert networks have shown great superiority for imbalanced data classification tasks due to their complementary and diverse. We have summarized two aspects for further explorations:(1)uncontrollable results, arising from the performance differences of individual experts and variations in sample difficulty;(2)insufficient exploration of the internal data structure. These factors result in inconsistent model performance across different data distributions, thereby impact the model’s generalization ability. To address the above issues, we propose a Collaborative Global-Local Structure Network (CGL-Net) with knowledge distillation for imbalanced data classification. Firstly, CGL-Net, as a new framework, decouples the representation learning of imbalanced data into global and local structure, enhancing the controllability of integration model in a hierarchical manner. Secondly, CGL-Net innovatively combines knowledge distillation, data augmentation, and multiple expert networks, efficiently extracting the internal structure of the data and improving robust recognition on imbalanced data. In particular, the global structure learning introduces an independent student network that integrates knowledge from diverse experts, enabling the model to achieve comprehensive and balanced performance across categories in imbalanced data. The local structure learning incorporates augmented data, allowing the model to focus on discriminative regional learning of individual objects, thereby enhances the robust representation for imbalanced data. After completing these two sequential learning stages, the model hierarchically integrates knowledge to achieve robust recognition performance on imbalanced data. Extensive experiments on six benchmark datasets demonstrate that the proposed CGL-Net significantly outperforms recent state-of-the-art methods.

Fuzzy clustering is still a hot topic because it can calculate the support degrees of an object belonging to different clusters to characterize uncertainty. However, it remains a challenge to detect clusters of arbitrary shapes, sizes, and dimensionality. What's worse, some objects are indistinguishable (imprecise) when they are in the overlapping regions of different clusters. To address such issues, this paper investigates a belief-based fuzzy and imprecise clustering (BFI) method, which can detect arbitrary clusters and provide the behavior (support) of objects to these clusters. Moreover, BFI can assign each imprecise object to a meta-cluster, defined as the union of specific clusters, to characterize (partial) imprecision. The proposed BFI can significantly reduce the risk of misclassification, and the effectiveness is validated in image processing (e.g., image segmentation and classification) and several benchmark datasets by comparing it with some typical methods.

Infrared small target detection (ISTD) aims to precisely capture the location and morphology of small targets under all-weather conditions. Compared with generic objects, infrared targets in remote fields of view are smaller in size and exhibit lower signal-to-clutter ratios. This poses a significant challenge in simultaneously preserving low-level target details and understanding high-level contextual semantics, forcing a trade-off between reducing miss detection and suppressing false alarms. In addition, most existing ISTD methods are designed for specific target types under certain infrared platforms, rather than as a unified framework broadly applicable across diverse infrared sensing scenarios. To address these challenges, we propose a unified self-prompt learning framework for ISTD under the guidance of the Segment Anything Model (SAM). Specifically, the model is incorporated with SAM in the encoding stage through a consult-guide manner, adapting the general knowledge to facilitate task-specific contextual understanding. Then, shallow-layer features are employed to generate self-derived prompts, which bidirectionally interact with encoded latent representations to complement subtle low-level details. Moreover, the semantic inconsistency during resolution recovery is mitigated by integrating a mutual calibration module into skip connections, ensuring coherent spatial-semantic fusion. Extensive experiments are conducted on four public ISTD datasets, and the results demonstrate that the proposed method consistently achieves superior performance across different infrared sensing platforms and target types. The code is released at https://github.com/fuyimin96/SAM-SPL.

The distribution of SAR targets generally conforms to a long-tailed distribution. Due to the existence of sample distribution bias and sample selection bias, training classifiers on this distribution of data often introduces spurious correlations between samples and classes. To address this issue, we propose a two-stage causal intervention framework. The core is that structural causality allows for independent interventions on multiple biases, thereby ensuring high-quality tail class predictions while maintaining unbiased performance for head classes. Firstly, we construct a structural causal graph for the long-tailed recognition task from causal perspective. Based on this graph, the causal paths underlying the two types of biases are identified. Secondly, we design a data augmentation method named DiagPatch-M, which identifies causal features within samples. In this process, these generated patches randomly integrate causal and non-causal features from two different samples, disrupting the original recognition process and effectively eliminating biases induced by sample selection. Thirdly, we design an unbiased structural risk minimization (USRM) optimization strategy, which eliminates the “head preference” of conventional models and the “tail preference” of modified models. This strategy reduces the bias introduced by the model’s dependence on the original sample distribution, and achieves stable recognition under different sample distributions. Experimental results on two long-tailed and two balanced datasets demonstrate that the effectiveness of our model surpasses the state-of-the-art (SOTA) methods, indicating the efficacy of our proposed framework in tackling the challenges posed by the long-tailed distribution in SAR target recognition.

Quantification and reduction of uncertainty in deep-learning techniques have received much attention but ignored how to characterize the imprecision caused by such uncertainty. In some tasks, we prefer to obtain an imprecise result rather than being willing or unable to bear the cost of an error. For this purpose, we investigate the representation of imprecision in deep-learning (RIDL) techniques based on the theory of belief functions (TBF). First, the labels of some training images are reconstructed using the learning mechanism of neural networks to characterize the imprecision in the training set. In the process, a label assignment rule is proposed to reassign one or more labels to each training image. Once an image is assigned with multiple labels, it indicates that the image may be in an overlapping region of different categories from the feature perspective or the original label is wrong. Second, those images with multiple labels are rechecked. As a result, the imprecision (multiple labels) caused by the original labeling errors will be corrected, while the imprecision caused by insufficient knowledge is retained. Images with multiple labels are called imprecise ones, and they are considered to belong to meta-categories, the union of some specific categories. Third, the deep network model is retrained based on the reconstructed training set, and the test images are then classified. Finally, some test images that specific categories cannot distinguish will be assigned to meta-categories to characterize the imprecision in the results. Experiments based on some remarkable networks have shown that RIDL can improve accuracy (AC) and reasonably represent imprecision both in the training and testing sets.

A recursive minimum upper bound estimator (UBE) is proposed in this article for stochastic systems with colored measurement noise (CMN) in the presence of generalized unknown disturbance (UD), which is motivated by noncooperative target tracking in the environment of continuous external interference. The CMN causes system noises to be correlated in time dimension, and the UD makes online calculation of estimate error covariance intractable, both of which give rise to deterioration of classical Kalman-like filtering and smoothing. By considering that constructing the upper bound of estimate error covariance requires looser conditions than directly calculating the theoretical covariance, an UBE is first defined, to obtain the filtered estimate and smoothed estimate together. Then, based on the reconstructed measurement model containing multiple state vectors due to measurement differencing to whiten system noises, the recursive structure of the defined UBE is derived in the case of CMN (CUBE) by introducing a free parameter to be optimized, and the existence condition of CUBE is also discussed. Finally, the minimum UBE with CMN, i.e., CMUBE, is presented by pursuing the minimum upper bound of estimate error covariance online through parameter optimization, in order to further suppress the peak of estimate errors. The advantages of estimation accuracy of the proposed CMUBE over Kalman filter (KF)/smoother, KF with CMN and minimum upper bound filter (MUBF) are demonstrated by an example of noncooperative target tracking in persistent interference environment, in terms of filtering versus smoothing, different values of the initial estimate error covariance and the positive-definite matrix setting a priori, sensor accuracies and different levels of UD.

Deep learning has been widely used in the field of radar image classification because of its powerful feature extraction capabilities. In the land-sea clutter classification of sky-wave over-the-horizon radar (OTHR), deep learning methods perform poorly due to the radar receiver noise and the ionosphere. Addressing this challenge, a dual attention aided residual neural networks (DAAResNet) is proposed for OTHR land-sea classification. Leveraging prior knowledge that landsea clutter features predominantly cluster around the 0 Hz frequency, two attention mechanisms are introduced. Firstly, a channel attention module (CAM) is proposed, which directs the network’s focus towards critical channels. Secondly, a frequency attention module (FAM) is proposed, which directs attention towards pivotal frequencies. The classification performance of DAAResNet is validated on the original dataset and the scarce dataset. Experimental results show that DAAResNet outperforms state-of-the-art methods.

The localization of underwater acoustic sensor networks (UASNs) has emerged as a critical research area in the marine information fusion field. Generally, the convex optimization method is adopted to solve the localization problem. However, this method has limitations in complex underwater environments, since it is difficult to transform the nonconvex optimization problem into a convex optimization problem under such conditions. Recently, deep reinforcement learning (DRL) has shown great potential and promise in solving intricate optimization tasks. Motivated by this, we propose to adopt DRL for UASNs localization to improve accuracy and robustness. The key challenge is that existing DRL-based methods require discretization of the environment, which leads to a compromise between search time and localization precision. To address this challenge, we first model the localization problem as a Markov decision process (MDP) with continuous state and action spaces and subsequently introduce a continuous control DRL framework to solve the localization problem. Within this framework, we develop three continuous control DRL-based localization estimators to address the localization problem in unsupervised, supervised, and semisupervised scenarios. Comprehensive simulations demonstrate the effectiveness of our approach, as the proposed solutions exhibit several advantageous features compared to traditional methods, such as: 1) compared with the convex optimization-based method, the convex relaxation is not required; 2) compared with the least squares method, the proposed estimators are capable of converging to a global optimal state; and 3) compared with the discrete control DRL method, the proposed estimators reduce localization time and enhance localization accuracy significantly.

Evidential clustering is a promising clustering framework using Dempster–Shafer belief function theory to model uncertain data. However, evidential clustering needs to estimate more parameters compared with other clustering algorithms, and thus the clustering performance of evidential clustering will be greatly affected if data is insufficient or contaminated. In addition, the existing evidential clustering algorithms can not well deal with high-dimensional data such as texts and images. To solve the above problems, an evidential clustering algorithm based on transfer learning and deep autoencoder (TDEC) is proposed. The TDEC utilizes deep autoencoder to obtain evidential clustering-friendly representations of the original data, and applies the maximum mean discrepancy (MMD) constraint between the source network and the target network, so that the network can learn domain-invariant features. The algorithm jointly trains the deep evidential clustering networks in the source domain and the target domain, and realizes the deep feature representations of high-dimensional data in the target domain for evidential clustering by minimizing reconstruction loss, entropy-based evidential clustering loss, MMD loss and the regular penalty term of the network parameters. In addition, an iterative optimization method to solve the TDEC objective function is proposed. Extensive experiments were conducted to evaluate the clustering performance of the proposed TDEC algorithm compared with the existing shallow transfer clustering algorithms and deep clustering algorithms. For both image and text clustering tasks, the proposed TDEC achieved approximately 5% performance improvement over the comparison algorithms on average. In addition, the practical application value of the proposed TDEC algorithm was demonstrated in unsupervised remote sensing image scene classification.

Adversarial attacks serve as an efficient approach to investigating model robustness, providing insights into internal weaknesses. In real-world applications, the model deployment typically adheres to a black-box setting, necessitating the transferability of adversarial examples crafted on a source model to others. Attack methods in the general computer vision field often employ global input transformations in individual spatial or frequency domains to boost adversarial transferability. However, the recognition of remote sensing objects primarily relies on target-related discriminative regions, whose determination exhibits significant model specificity. Besides, the coupling between objects and background further exacerbates the gap between models. Consequently, the transferability of adversarial examples is limited due to overfitting to the source model. To tackle this problem, we propose a spatial-frequency co-transformation (SFCoT) to improve adversarial transferability for remote sensing object recognition. Specifically, the input image is decomposed into blocks and components in the spatial and frequency domains, respectively. Then, a selective frequency transformation (SFT) is performed on the low-frequency components to narrow intermodel gaps. Subsequently, modular spatial transformations (MSTs) are adopted in blocks to enhance target-related diversity. Incorporating transformations across domains effectively mitigates the overfitting to model-specific information, leading to better adversarial transferability. Extensive experiments have been conducted on FGSCR-42 and MTARSI datasets, and the results demonstrate that the proposed method achieves state-of-the-art performance across various model architectures. The code will be released athttps://github.com/fuyimin96/SFCoT.

Cross-scene classification aims to transfer knowledge acquired from a label-rich source domain to an unlabeled target domain with a distribution shift. With the large amount of available remote sensing data from diverse satellite platforms, it prompts us to utilize the knowledge from extensive datasets to address target tasks in small-scale domains, known as partial domain adaptation (PDA). However, the PDA setting poses significant challenges for remote sensing scene images. Existing methods often fail to sufficiently explore both task-specific and transferable knowledge across domains based on the representations of entire samples, potentially resulting in the amplification of negative transfer brought by irrelevant knowledge. To address this, we propose a new selective alignment transformer (SAT) designed to distinguish transferable and untransferable knowledge across domains for cross-scene classification in RSIs under the PDA scenario. Specifically, a new bi-level reweighting strategy that incorporates transferability-aware patch selection and class-wise reweighting is developed to emphasize the transferable image patches and classes. Based on the aforementioned reweighting strategy, we further introduce a patch-weighted maximum mean discrepancy (PMMD) loss, which selectively aligns the distributions from the patch-level perspective, facilitating the learning of transferable domain-invariant representations. The experimental results of SAT demonstrate its effectiveness and superiority in addressing this practical domain adaptation (DA) task, outperforming state-of-the-art methods in PDA tasks on four datasets.

Object detection is a fundamental task in various applications ranging from autonomous driving to intelligent security systems. However, recognition of a person can be hindered when their clothing is decorated with carefully designed graffiti patterns, leading to the failure of object detection. To achieve greater attack potential against unknown black-box models, adversarial patches capable of affecting the outputs of multiple-object detection models are required. While ensemble models have proven effective, current research in the field of object detection typically focuses on the simple fusion of the outputs of all models, with limited attention being given to developing general adversarial patches that can function effectively in the physical world. In this paper, we introduce the concept of energy and treat the adversarial patches generation process as an optimization of the adversarial patches to minimize the total energy of the “person” category. Additionally, by adopting adversarial training, we construct a dynamically optimized ensemble model. During training, the weight parameters of the attacked target models are adjusted to find the balance point at which the generated adversarial patches can effectively attack all target models. We carried out six sets of comparative experiments and tested our algorithm on five mainstream object detection models. The adversarial patches generated by our algorithm can reduce the recognition accuracy of YOLOv2 and YOLOv3 to 13.19% and 29.20%, respectively. In addition, we conducted experiments to test the effectiveness of T-shirts covered with our adversarial patches in the physical world and could achieve that people are not recognized by the object detection model. Finally, leveraging the Grad-CAM tool, we explored the attack mechanism of adversarial patches from an energetic perspective.

Multisource unsupervised domain adaptation (MUDA) is an important and challenging topic for target classification with the assistance of labeled data in source domains. When we have several labeled source domains, it is difficult to map all source domains and target domain into a common feature space for classifying the targets well. In this article, a new progressive multisource domain adaptation network (PMSDAN) is proposed to further improve the classification performance. PMSDAN mainly consists of two steps for distribution alignment. First, the multiple source domains are integrated as one auxiliary domain to match the distribution with the target domain. By doing this, we can generally reduce the distribution discrepancy between each source and target domains, as well as the discrepancy between different source domains. It can efficiently explore useful knowledge from the integrated source domain. Second, to mine assistance knowledge from each source domain as much as possible, the distribution of the target domain is separately aligned with that of each source domain. A weighted fusion method is employed to combine the multiple classification results for making the final decision. In the optimization of domain adaption, weighted hybrid maximum mean discrepancy (WHMMD) is proposed, and it considers both the interclass and intraclass discrepancies. The effectiveness of the proposed PMSDAN is demonstrated in the experiments comparing with some state-of-the-art methods.

This article presents two-filter smoothing (TFS) by maximizing the correntropy rather than minimizing the mean square error, for nonlinear systems with non-Gaussian noises, such as heavy-tailed distributed noises or outliers, motivated by high-precision noncooperative target backtracking. The maximum correntropy (MC) recursive TFS, abbreviated as MRTFS, is first derived, where the smoothed estimate is obtained by fusing forward and backward filtering results step by step through maximizing the correntropy. Then, the information filtering form of the above MRTFS is proposed in order to loose the initial conditions of forward and backward filtering and enhance the structural conciseness of MRTFS. Considering that the fixed-point iteration is adopted to realize MC-based forward-time filtering, backward-time filtering, and two-filter fusion, its convergence is shown in the premise that the interested state vector is bounded and the kernel bandwidth of the correntropy is larger than a special threshold. Meanwhile, the computational complexity of MRTFS is analyzed, which is similar to that of extended Kalman-like TFS. An experiment of noncooperative target backtracking shows that estimation accuracy of the proposed method is superior to that of extended Kalman filter (EKF), TFS, Rauch–Tung–Striebel smoother (RTS) and MC-based EKF/RTS, in terms of estimation confidence ellipses, different levels of kernel bandwidths and iterations.

Classification of missing data based on estimation is still challenging since existing methods relying on one imputation strategy fail to consider the diversity of different attribute distributions. In this case, there are inevitably some “bad” estimations at the attribute level, reducing the performance of classification. This article proposes a mixed-type imputation method (MTI) to classify missing data under the theory of belief functions (TBF) via two quality matrices to address this problem. The proposed MTI method has the advantages of making estimations as close to the truth as possible at the attribute level while reducing the negative impact of possible bad estimations on the classification. Specifically, the first matrix used to impute missing values can characterize the different supports of multiple imputation methods for estimating various attributes. The other matrix used to perform the classification task can extract the reliabilities of estimations on the different classes. The validity has been demonstrated in the final decision support based on the TBF, famous for characterizing uncertainty and imprecision, for example, caused by missing values.

This paper presents recursive two-filter smoothing (TFS) in the criterion of maximizing the correntropy (MC) instead of minimizing the mean square error, to pursue robustness for outlier rejections caused by non-Gaussian noises and obtain high-precision state estimate, which is motivated by non-cooperative target backtracking. Here, non-cooperative target tracking often needs to consider non-Gaussian noises. The MC-based recursive TFS (abbreviated as MRTFS) is put forward, where both the forward and backward filters are performed independently and recursively in the criterion of MC. Meanwhile, an MC-based fusion rule is further designed to obtain the final smoothed estimate by fusing the forward filtered estimate and backward predicted estimate step by step, in order to improve estimation accuracy. A target backtracking example with non-Gaussian noises is simulated to show the advantage of estimation accuracy of the proposed MRTFS over Kalman filter/smoothers, MC-based Kalman filter/Rauch-Tung-Striebel smoother, in terms of different kernel bandwidths and levels of process noises.

Visual navigation plays an important role for Unmanned Aerial Vehicles(UAVs). In some applications, the landmark image and the real-time image may be heterogeneous, like near-infrared and visible images. In this work, we propose a multimodal image registration method to deal with near-infrared and visible images so that it can be applied to visual navigation system for the localization of UAVs in GPS-denied environments. At first, a new feature extraction strategy is developed to embed different modalities of images into the common feature space based on disentangled representations. Such common space is independent of the image modality, and this can eliminate the modality differences. Meanwhile, an intensity loss is introduced to measure the similarity of mono-modal images. In the proposed method, we can directly predict the transformation parameters and thus accelerates the localization of UAV s. Extensive experiments on synthetic datasets are conducted to demonstrate the validity of our method, and the experimental results show that the proposed method can effectively improve the localization accuracy.

Open set recognition (OSR) aims to correctly recognize the known classes and reject the unknown classes for increasing the reliability of the recognition system. The distance-based loss is often employed in deep neural networks-based OSR methods to constrain the latent representation of known classes. However, the optimization is usually conducted using the nondirectional euclidean distance in a single feature space without considering the potential impact of spatial distribution. To address this problem, we propose orientational distribution learning (ODL) with hierarchical spatial attention for OSR. In ODL, the spatial distribution of feature representation is optimized orientationally to increase the discriminability of decision boundaries for open set recognition. Then, a hierarchical spatial attention mechanism is proposed to assist ODL to capture the global distribution dependencies in the feature space based on spatial relationships. Moreover, a composite feature space is constructed to integrate the features from different layers and different mapping approaches, and it can well enrich the representation information. Finally, a decision-level fusion method is developed to combine the composite feature space and the naive feature space for producing a more comprehensive classification result. The effectiveness of ODL has been demonstrated on various benchmark datasets, and ODL achieves state-of-the-art performance.

In pattern classification, there may not exist labeled patterns in the target domain to train a classifier. Domain adaptation (DA) techniques can transfer the knowledge from the source domain with massive labeled patterns to the target domain for learning a classification model. In practice, some objects in the target domain are easily classified by this classification model, and these objects usually can provide more or less useful information for classifying the other objects in the target domain. So a new method called distribution adaptation based on evidence theory (DAET) is proposed to improve the classification accuracy by combining the complementary information derived from both the source and target domains. In DAET, the objects that are easy to classify are first selected as easy-target objects, and the other objects are regarded as hard-target objects. For each hard-target object, we can obtain one classification result with the assistance of massive labeled patterns in the source domain, and another classification result can be acquired based on the easy-target objects with confidently predicted (pseudo) labels. However, the weights of these classification results may vary because the reliabilities of the used information sources are different. The weights are estimated by mean difference reflecting the information source quality. Then, we discount the classification results with the corresponding weights under the framework of the evidence theory, which is expert at dealing with uncertain information. These discounted classification results are combined by an evidential combination rule for making the final class decision. The effectiveness of DAET for cross-domain pattern classification is evaluated with respect to some advanced DA methods, and the experiment results show DAET can significantly improve the classification accuracy.

Open set recognition (OSR) aims to classify known classes and recognize unknown classes simultaneously. Existing OSR methods have primarily focused on learning decision boundaries based on overall feature representations, and have achieved good performance on various coarse-grained image datasets. However, the overall feature representations of objects in fine-grained image datasets are highly similar, making it difficult to distinguish between known and unknown classes by overall feature-based decision boundaries. To address this problem, we propose a progressive learning vision transformer (PLViT) with a coarse-to-fine optimization strategy. In PLViT, the overall feature representations are first optimized in the distance space to learn the initial decision boundaries. Then, a context-aware patch selection module is designed to locate the discriminative part regions. Afterwards, the multi-layer representations of each selected patch are aggregated according to the self-attention weights, and input into the last transformer layer to extract local feature representations. Finally, overall and local feature representations are adaptively fused and optimized in the angular space to further refine the decision boundaries. Experimental results on four fine-grained remote sensing object recognition datasets show that PLViT outperforms state-of-the-art methods.

With the rapid development of supervised learning-based SAR target recognition technology, it is easy to find that the recognition performance is proportional to the amount of training samples. However, the biased data distribution and under-representation of the model caused by incomplete data within categories exacerbate the challenge of SAR interpretation. In this paper, we propose a new view-semantic transformer network (VSTNet) that generates synthesized samples to complete the statistical distribution of training data and improve the discriminative representation of the model. First, SAR images from different views are encoded into a disentangled latent space, which allows us to synthesize data with more diverse views by manipulating view-semantic features. Second, the synthesized data as a complement effectively expands the training set and alleviates the overfitting problem of limited data in sparse views. Third, the proposed method unifies SAR image synthesis and SAR target recognition into an end-to-end framework to boost their performance against each other. Experiments conducted on moving and stationary target acquisition and recognition (MSTAR) data demonstrate the robustness and effectiveness of the proposed method.

Deep learning algorithms have achieved state-of-the-art progress in synthetic aperture radar (SAR) automatic target recognition (ATR) tasks. They theoretically assume that training and test samples are independent and identically distributed (i.i.d.) for generalization, but it is intractable for practical ATR scenarios. In this paper, we propose a novel contrastive feature disentangling framework termed ConFeDent to learn features with improved generalization performance under a condition of a weaker distribution consistency. More specifically, ConFeDent aims to describe the semantic interactions between two arbitrary SAR training samples instead of treating them independently. It can implicitly disentangle features encoding the pose and identity knowledge from the whole samples with a semi-parametric geometric transformation model and a second-order energy model. In particular, except for the identity label, we use deductive-based geometry knowledge as supervision to teach the model to learn the concept of aspect angle variation. A progressively amortized inference scheme is constructed for efficient feature learning and recognition in an end-to-end manner. Finally, we further release a strengthened version, called ConFeDent+, which can explicitly utilize and learn more information from cross-category samples. Experimental results on the moving and stationary target acquisition and recognition (MSTAR) benchmark demonstrate the effectiveness of our proposed models in the SAR ATR. In particular, we validate the algorithms in a more challenging scenario where the range of aspect angles for training and testing samples is permitted to be disparate. Our model can achieve much higher recognition accuracy than other SAR ATR algorithms.

It is still a challenging problem to characterize uncertainty and imprecision between specific (singleton) clusters with arbitrary shapes and sizes. In order to solve such a problem, we propose a belief shift clustering (BSC) method for dealing with object data. The BSC method is considered as the evidential version of mean shift or mode seeking under the theory of belief functions. First, a new notion, called belief shift, is provided to preliminarily assign each query object as the noise, precise, or imprecise one. Second, a new evidential clustering rule is designed to partial credal redistribution for each imprecise object. To avoid the “uniform effect” and useless calculations, a specific dynamic framework with simulated cluster centers is established to reassign each imprecise object to a singleton cluster or related meta-cluster. Once an object is assigned to a meta-cluster, this object may be in the overlapping or intermediate areas of different singleton clusters. Consequently, the BSC can reasonably characterize the uncertainty and imprecision between singleton clusters. The effectiveness has been verified on several artificial, natural, and image segmentation/classification datasets by comparison with other related methods.

In target recognition, the information about the target usually exists in several domains captured by different sources (sensors). However, it is difficult for us to obtain the perfect target information as the source domain data due to the sensors' limitations sometimes. For the target classification of visible and infrared paired images, we assume that some classes of visible and infrared paired images and other classes of visible images can be obtained, whereas other classes of unseen infrared images need to be classified. This problem is actually a zero-shot deep domain adaptation (ZDDA) problem which divides the data into task-relevant (T-R) data and task-irrelevant (T-I) data. Moreover, the classes of T-R data require recognition, while the classes of T-I data do not need. The traditional ZDDA method sacrifices the classification accuracy of T-R data in the target domain for the generalization ability of T-I data in the source domain. So we propose a method to solve the problem in another way. More precisely, we first use the image-to-image translation network to learn the mapping between the source domain (visible images) T-I data and the target domain (infrared images) T-I data, and convert the visible T-R images to pseudo-infrared images. Then the pseudo-infrared images and the inverted grayscale T-R images are combined to construct a new hybrid domain (source domain I). Meanwhile, we also construct a hybrid domain (source domain II) of T-I images similarly. Besides, we use the infrared T-I images to construct the third domain (source domain III). Finally, we design a deep domain generalization method for cross-domain infrared image classification. And the total loss consists of the classification loss of the source domain I and the distribution alignment loss between the source domains II and III. We evaluate our method using VAIS ship and RGB-NIR scene datasets. The experimental results demonstrate the effectiveness of the proposed method.

Traditional pattern classification methods can handle the objects whose categories are contained in the given (known) categories of training data. In open-set scenarios, objects to classify may belong to the ignorant (unknown) classes that is not included in training data set. Open-Set Recognition (OSR) tries to detect these unknown class objects and classify known class objects. In this paper, we propose a simple OSR method based on k-Nearest Neighbors (k-NNs). The test data (objects to classify) is put together with the labeled training data, so that the labeled training instances and the unlabeled objects to classify can appear in the k-NNs of other objects. Then, the probability of object lying in the given classes can be determined according to the k-NNs of this object. If the labeled training data is the majority of k-NNs, this object most likely belongs to one of the given classes, and the distances between the object and its neighbors are taken into account here. Then the objects with high probability are marked with the estimated probability. However, if most of k-NNs are the unlabeled test objects, the class of object cannot be classified in this step because the k-NNs of the object is uncertain. In this paper, the probability of the other uncertain objects belonging to known classes is re-calculated based on the labeled training instances and the objects marked with the estimated probability. Such iteration will not stop until all the probabilities of objects belonging to known classes are not changed. Then, the Otsu's method is employed to obtain the optimal threshold for the final recognition. If the probability of object belonging to known classes is smaller than this threshold, it will be assigned to the unknown class. The other objects will be considered as known classes and then committed to a specific class by a pre-trained classifier. The effectiveness of the proposed method has been validated using some experiments.

In clustering process, fuzzy partition performs better than hard partition when the boundaries between clusters are vague. Whereas, traditional fuzzy clustering algorithms produce less interpretable results, limiting their application in security, privacy, and ethics fields. To that end, this paper proposes an interpretable fuzzy clustering algorithm—fuzzy decision tree-based clustering which combines the flexibility of fuzzy partition with the interpretability of the decision tree. We constructed an unsupervised multi-way fuzzy decision tree to achieve the interpretability of clustering, in which each cluster is determined by one or several paths from the root to leaf nodes. The proposed algorithm comprises three main modules: feature and cutting point-selection, node fuzzy splitting, and cluster merging. The first two modules are repeated to generate an initial unsupervised decision tree, and the final module is designed to combine similar leaf nodes to form the final compact clustering model. Our algorithm optimizes an internal clustering validation metric to automatically determine the number of clusters without their initial positions. The synthetic and benchmark datasets were used to test the performance of the proposed algorithm. Furthermore, we provided two examples demonstrating its interest in solving practical problems.

Change detection result is usually obtained by clustering or classifying; however, the spatial information of pixels is rarely considered during the classification process. In this letter, we propose a practical method to improve the performance of existing change detection algorithms on remote-sensing images without prior information. First, the existing detection result is regarded as an initial result. Second, it takes advantage of this initial result with neighborhood information of pixels to select the training data, then a random forest classifier is trained for precise classification. Finally, the median filtering is used to eliminate singular points for further improvement of detection performance. Corresponding experiments on three real synthetic aperture radar (SAR) data sets demonstrate the effectiveness of the proposed method.

In the above letter[1], the images ofFigs. 2(a),3(b), and4(a)given in Section IV are wrong. The errors make the preevent image (a) the same as the postevent image (b) inFigs. 2–4. The corrected data sets are shown in the following figures.

In pattern classification, we may have a few labeled data points in the target domain, but a number of labeled samples are available in another related domain (called the source domain). Transfer learning can solve such classification problems via the knowledge transfer from source to target domains. The source and target domains can be represented by heterogeneous features. There may exist uncertainty in domain transformation, and such uncertainty is not good for classification. The effective management of uncertainty is important for improving classification accuracy. So, a new belief-based bidirectional transfer classification (BDTC) method is proposed. In BDTC, the intraclass transformation matrix is estimated at first for mapping the patterns from source to target domains, and this matrix can be learned using the labeled patterns of the same class represented by heterogeneous domains (features). The labeled patterns in the source domain are transferred to the target domain by the corresponding transformation matrix. Then, we learn a classifier using all the labeled patterns in the target domain to classify the objects. In order to take full advantage of the complementary knowledge of different domains, we transfer the query patterns from target to source domains using the K-NN technique and do the classification task in the source domain. Thus, two pieces of classification results can be obtained for each query pattern in the source and target domains, but the classification results may have different reliabilities/weights. A weighted combination rule is developed to combine the two classification results based on the belief functions theory, which is an expert at dealing with uncertain information. We can efficiently reduce the uncertainty of transfer classification via the combination strategy. Experiments on some domain adaptation benchmarks show that our method can effectively improve classification accuracy compared with other related methods.

In synthetic aperture radar (SAR) automatic target recognition (ATR), there are mainly two types of methods: physics-driven model and data-driven network. The physics-driven model can exploit electromagnetic theory to obtain physical properties, while the data-driven network will extract deep discriminant feature of targets. These two types of features represent the target characteristics in scattering domain and image domain, respectively. However, the representation discrepancy caused by the different modalities between them hinders the further comprehensive utilization and fusion of both features. In order to take full advantage of physical knowledge and deep discriminant feature for SAR ATR, we propose a new feature fusion learning framework SDF-Net to combine scattering and deep image features. In this work, we treat the attributed scattering centers (ASC) as set-data instead of multiple individual points, which can well mine the topological interaction among scatterers. Then multi-region multi-scale sub-sets are constructed at both component and target levels. To be specific, the most significant scattering intensity and overall representation in these sub-sets are exploited successively to learn permutation-invariant scattering features according to a set-oriented deep network. The scattering representations can provide mid-level semantic and structural features that are subsequently fused with the complementary deep image features to yield an end-to-end high-level feature learning framework, which helps enhance the generalization ability of networks especially under complex observation conditions. Extensive experiments on Moving and Stationary Target Acquisition and Recognition database verify the effectiveness and robustness of the SDF-Net compared against both typical SAR ATR networks and ASC-based models.

It is quite an important and challenging problem for change detection (CD) from heterogeneous remote sensing images. The images obtained from different sensors (i.e., synthetic aperture radar (SAR) & optical camera) characterize the distinct properties of objects. Thus, it is impossible to detect changes by direct comparison of heterogeneous images. In this article, a new unsupervised change detection (USCD) method is proposed based on image translation. The cycle-consistent adversarial networks (CycleGANs) are employed to learn the subimage to subimage mapping relation using the given pair (i.e., before and after the event) of heterogeneous images from which the changes will be detected. Then, we can translate one image (e.g., SAR) from its original feature space (e.g., SAR) to another space (e.g., optical). By doing this, the pair of images can be represented in a common feature space (e.g., optical). The pixels with close pattern values in the before-event image may have quite different values in the after-event image if the change happens on some ones. Thus, we can generate the difference map between the translated before-event image and the original after-event image. Then, the difference map is divided into changed and unchanged parts. However, these detection results are not very reliable. We will select some significantly changed and unchanged pixel pairs from the two parts with the clustering technique (i.e.,$K$-means). These selected pixel pairs are used to learn a binary classifier, and the other pixel pairs will be classified by this classifier to obtain the final CD results. Experimental results on different real datasets demonstrate the effectiveness of the proposed USCD method compared with several other related methods.

It remains an important research topic for the classification of imbalanced data. There exist some methods to solve this problem, such as hybrid-sampling, over-sampling, and under-sampling. Each method has its own advantage, and different methods generally provide some complementary knowledge. We want to combine these three methods at the decision level in an appropriate way for achieving as good as possible classification performance. Evidence theory is expert at representing and combining uncertain information. So a new method called an evidential combination of classifiers (ECC) is proposed for dealing with imbalanced data. The classification result generated by different strategies (i.e., hybrid-sampling, over-sampling, or under-sampling) may have different reliabilities for query patterns. A cautious reliability evaluation rule is developed for each classification result based on the close neighborhoods. After that, the classification result is revised with a new belief redistribution way according to the reliability evaluation, and the probability/belief of one class can be partially transferred to other classes as well as the total ignorance, which is defined by the whole frame of classes. By doing this, we can reduce the error risk of each classification method. Then, the revised classification results from different methods are combined by evidence theory to make the final class decision. The effectiveness of the ECC method has been demonstrated using several experiments, and it shows that ECC can effectively improve the classification performance comparing with other related methods.

Classifier fusion remains an effective method to improve classification performance. In applications, the classifiers learnt using different attributes may work with various frames of discernment (FoD) of classification. There generally exist more or less complementary knowledge among these classifiers. However, how to efficiently combine such classifiers under different FoD is a challenging problem. In this article, we propose a new method for classifier fusion with different FoD based on the belief functions, which allow to well represent and deal with uncertain information. The credal transformation rules are developed to map the various FoD into a common one. It allows to transfer the probability (or mass of belief) of one class in the given FoD not only to several singleton classes but also to the metaclasses (i.e., disjunction of several classes) and the ignorance in other chosen FoD according to a transformation matrix, which is estimated based on the training (pairwise) data by minimizing a certain error criteria. Thus, we can well characterize the uncertainty and imprecision during the transformation of FoD. After that, the outputs of different classifiers represented by basic belief assignments (BBAs) can be transformed to a common FoD. Then, the well-known Dempster's rule is employed to combine these transformed BBA to obtain final classification result under the chosen FoD. Several real data sets are used in the experiment to evaluate the performance of the proposed method. Our experimental results show that this new method can efficiently improve the classification accuracy with respect to other related methods.

Raw polarimetric images are captured by a focal plane polarimeter which is covered by a micro-polarizer array (MPA). The design of the MPA plays a crucial role in polarimetric imaging. MPAs are predominantly designed according to expert engineering experience and rules of thumb. Typically, only one optimization criterion, maximizing bandwidth, is used to design the MPA. To select a design, an exhaustive search is usually performed on a very limited set of available polarizing patterns, which must be constrained in order to make the search tractable. In contrast, this paper proposes a fully automated and optimal MPA design method (AO-MPA) which generates significantly improved MPAs. Instead of the single criterion of bandwidth, we propose six design principles, and show how they can be utilized to mutually optimize the MPA design by formulating a tri-objective optimization problem with multiple constraints. A much larger set of possible MPA patterns is rapidly and automatically searched by applying advanced multi-objective optimization techniques. We have tested AO-MPA using two groups of experiments, in which AO-MPA is compared against several other leading MPA design methods, and the patterns generated by AO-MPA are compared against state-of-the-art patterns from the literature. The results, obtained using a public benchmark dataset, show that the AO-MPA method is very computationally efficient, and can find all optimal MPA patterns for all array sizes. Moreover, for each size, AO-MPA obtains all optimal layouts simultaneously. AO-MPA generates designs which require fewer polarization orientations, while also yielding better performance in estimating intensity measurements, Stokes vector and the degree of linear polarization. This results in MPAs which are easier to manufacture while also being more robust to noise.

The scattering signatures of a synthetic aperture radar (SAR) target image will be highly sensitive to different azimuth angles/poses, which aggravates the demand for training samples in learning-based SAR image automatic target recognition (ATR) algorithms, and makes SAR ATR a more challenging task. This paper develops a novel rotation awareness-based learning framework termed RotANet for SAR ATR under the condition of limited training samples. First, we propose an encoding scheme to characterize the rotational pattern of pose variations among intra-class targets. These targets will constitute several ordered sequences with different rotational patterns via permutations. By further exploiting the intrinsic relation constraints among these sequences as the supervision, we develop a novel self-supervised task which makes RotANet learn to predict the rotational pattern of a baseline sequence and then autonomously generalize this ability to the others without external supervision. Therefore, this task essentially contains a learning and self-validation process to achieve human-like rotation awareness, and it serves as a task-induced prior to regularize the learned feature domain of RotANet in conjunction with an individual target recognition task to improve the generalization ability of the features. Extensive experiments on moving and stationary target acquisition and recognition benchmark database demonstrate the effectiveness of our proposed framework. Compared with other state-of-the-art SAR ATR algorithms, RotANet will remarkably improve the recognition accuracy especially in the case of very limited training samples without performing any other data augmentation strategy.

In applications of domain adaptation, there may exist multiple source domains, which can provide more or less complementary knowledge for pattern classification in the target domain. In order to improve the classification accuracy, a decision-level combination method is proposed for the multisource domain adaptation based on evidential reasoning. The classification results obtained from different source domains usually have different reliabilities/weights, which are calculated according to domain consistency. Therefore, the multiple classification results are discounted by the corresponding weights under belief functions framework, and then, Dempster's rule is employed to combine these discounted results. In order to reduce errors, a neighborhood-based cautious decision-making rule is developed to make the class decision depending on the combination result. The object is assigned to a singleton class if its neighborhoods can be (almost) correctly classified. Otherwise, it is cautiously committed to the disjunction of several possible classes. By doing this, we can well characterize the partial imprecision of classification and reduce the error risk as well. A unified utility value is defined here to reflect the benefit of such classification. This cautious decision-making rule can achieve the maximum unified utility value because partial imprecision is considered better than an error. Several real data sets are used to test the performance of the proposed method, and the experimental results show that our new method can efficiently improve the classification accuracy with respect to other related combination methods.

It remains a challenging problem for data classification without training patterns. In many applications, there may exist some labeled data in other related domains (called source domain), and such labeled data can be helpful to solve the classification problem in the target domain. It is considered that the source domain and target domain are heterogeneous here and they represent the distinct feature spaces. A new transfer classification method for heterogeneous data is proposed based on the evidence theory. Some pattern pairs in the source domain and target domain are given to predict the link of these two domains. For each pattern in the target domain, we estimate its possible mapping value in the source domain using these pattern pairs with a self-organizing map (SOM) technique, and then the mapping value is classified using the labeled data in the source domain. However, the patterns with close values in the target domain may have more or less different values in the source domain due to the distinct characteristics of these two domains. So the mapping value can be very uncertain sometimes. In such a case, the target pattern is allowed to have multiple mapping values with different weights/reliabilities in the source domain. These mapping values can produce different classification results. The evidence theory is good at characterizing and combining uncertain information. In order to improve the classification accuracy, a new evidence-based weighted fusion method is developed for combining these classification results, which are discounted by the corresponding weights under the belief functions framework, and the final class decision is made according to the combination result. In experimental applications, some heterogeneous remote sensing data and UCI data are used to test the performance of new method with respect to several other methods, and it shows that the new method can efficiently improve the classification accuracy.

Evidence theory, also called belief function theory, provides an efficient tool to represent and combine uncertain information for pattern classification. Evidence combination can be interpreted, in some applications, as classifier fusion. The sources of evidence corresponding to multiple classifiers usually exhibit different classification qualities, and they are often discounted using different weights before combination. In order to achieve the best possible fusion performance, a new credal belief redistribution (CBR) method is proposed to revise such evidence. The rationale of CBR consists of transferring belief from one class not just to other classes, but also to the associated disjunctions of classes (i.e., meta-classes). As classification accuracy for different objects in a given classifier can also vary, the evidence is revised according to prior knowledge mined from its training neighbors. If the selected neighbors are relatively close to the evidence, a large amount of belief will be discounted for redistribution. Otherwise, only a small fraction of belief will enter the redistribution procedure. An imprecision matrix estimated based on these neighbors is employed to specifically redistribute the discounted beliefs. This matrix expresses the likelihood of misclassification (i.e., the probability of a test pattern belonging to a class different from the one assigned to it by the classifier). In CBR, the discounted beliefs are divided into two parts. One part is transferred between singleton classes, whereas the other is cautiously committed to the associated meta-classes. By doing this, one can efficiently reduce the chance of misclassification by modeling partial imprecision. The multiple revised pieces of evidence are finally combined by the Dempster-Shafer rule to reduce uncertainty and further improve classification accuracy. The effectiveness of CBR is extensively validated on several real datasets from the UCI repository and critically compared with that of other related fusion methods.

In this paper, we newly suggest that more attention should be paid on learning rotation-equivariant and label-invariant features for each target instead of the conventional rotation-invariant ones. To achieve this goal, we present a novel rotation awareness based self-supervised learning (RR-SSL) deep model to recognize the behavior of target rotation, which is also benefit from the discriminative training scheme without manual labeling. Then this model is incorporated into another deep discriminative model of target recognition to form a dual-task learning framework, where their bottom layers are shared to capture the expected features. Sufficient experimental results on moving and stationary target acquisition and recognition (MSTAR) database demonstrate the effectiveness of our proposed model. The overall framework can achieve a better or comparative recognition accuracy compared with other state-of-the-art SAR-ATR algorithms.

We present a distributed Bernoulli filter for tracking a target that may be present or absent in the cluttered surveillance area in unknown time intervals by using a decentralized sensor network. As a key feature of the Bernoulli filter, a parameter referring to the target existence probability is online updated jointly with the target state probability density function. We propose to fuse them in parallel, both in an arithmetic average fusion manner via the standard consensus or flooding scheme. Alternatively, one may communicate and fuse merely target existence probabilities, leading to a communication-inexpensive protocol. We experimentally compare the proposed approaches, based on the Gaussian mixture implementation of the Bernoulli filter, with the cutting-edge geometric average fusion approach based on a Doppler shift sensor network. Advantages are observed in computing efficiency and in dealing with local missed detection.

Feature learning-based polarimetric synthetic aperture radar (PolSAR) classification model will generally suffer from the challenge of deficient labeled pixels. In this paper, we propose a novel generative-discriminative network for PolSAR polar-spatial feature fusion learning and classification, which comprises of a deep generative network and a discriminative network with their bottom layers shared. With this architecture, it enables to make use of both labeled and unlabeled pixels in a PolSAR image for model learning in a semisupervised way. Moreover, the proposed network imposes a Gaussian random field prior and a conditional random field posterior on the learned fusion features and the output label configuration, respectively. Without the need of the complicated recurrent iterations, our network can still efficiently produce the structured fusion feature as well as a smoothed classification map by involving some auxiliary variables, and it is specifically optimized via variational inference within an alternating direction method of multipliers iteration scheme. Extensive experiments on different benchmark PolSAR imageries demonstrate the effectiveness and superiority of the proposed network. Compared with other state-of-the-art algorithms of PolSAR feature learning and classification, our model can achieve a much better performance in terms of the visual quality of the label map and overall classification accuracy, facilitating the much less labeling pixels.

OWA operator is an effective aggregation method in multi-criteria decision making problem. However, in some multi-criteria decision making cases, the criteria satisfactions have some uncertainty, for instance, which is a set of interval values at a series of different levels. For multi-criteria decision making problem, it is necessary to aggregate criteria satisfactions. But the linear ordering of criteria satisfactions is unknown at a specific level in these cases. Therefore, OWA operator cannot be applied to aggregate the satisfactions directly. In this paper, a new method, named as the Interval Value Exceedance Method (IVEM), is proposed. By using the proposed method, the domination relationship of criteria satisfactions for each level can be obtained. Then OWA operator can be used to aggregate these satisfactions based on the domination relationship, even if the linear ordering of satisfaction is unknown.

It is a challenging problem for pattern classification with few labeled instances. Transfer learning provides an efficient solution to improve the classification accuracy using some training knowledge in the related domain (called source domain). Nevertheless, the single transformation in one direction may be uncertain in some cases, and this is harmful for classification. So we propose a new classification method based on the fusion of data transformations in different directions between source domain and target domain. At first, the mapping of target in the source domain is estimated by K-nearest neighbor technique using some one-to-one instance pairs, and the estimated mapping instance (pattern) can be classified in the source domain according to the available training data. Then, the credibility of classification result is evaluated. If the credibility achieves the expected threshold, the classification result is directly output. Otherwise, it indicates that the transformation may be not very reliable, and the labeled instances in source domain will be transferred to target domain for the classification of target. The two versions of classification results will be fused with different weights based on evidential reasoning, and the weighting factors are optimized using the available training instances. By doing this, we can efficiently reduce the uncertainty of transformation and improve the classification accuracy. Some real data sets from UCI have been employed to validate the effectiveness of the proposed by comparing with other related methods.

Classifier fusion is an efficient strategy to improve the classification performance for the complex pattern recognition problem. In practice, the multiple classifiers to combine can have different reliabilities and the proper reliability evaluation plays an important role in the fusion process for getting the best classification performance. We propose a new method for classifier fusion with contextual reliability evaluation (CF-CRE) based on inner reliability and relative reliability concepts. The inner reliability, represented by a matrix, characterizes the probability of the object belonging to one class when it is classified to another class. The elements of this matrix are estimated from the -nearest neighbors of the object. A cautious discounting rule is developed under belief functions framework to revise the classification result according to the inner reliability. The relative reliability is evaluated based on a new incompatibility measure which allows to reduce the level of conflict between the classifiers by applying the classical evidence discounting rule to each classifier before their combination. The inner reliability and relative reliability capture different aspects of the classification reliability. The discounted classification results are combined with Dempster-Shafer's rule for the final class decision making support. The performance of CF-CRE have been evaluated and compared with those of main classical fusion methods using real data sets. The experimental results show that CF-CRE can produce substantially higher accuracy than other fusion methods in general. Moreover, CF-CRE is robust to the changes of the number of nearest neighbors chosen for estimating the reliability matrix, which is appealing for the applications.

In pattern classification problem, different classifiers learnt using different training data can provide more or less complementary knowledge, and the combination of classifiers is expected to improve the classification accuracy. Evidential reasoning (ER) provides an efficient framework to represent and combine the imprecise and uncertain informations. In this paper, we want to focus on the weighted combination of classifiers based on ER. Because each classifier may have different performance on the given dataset, the classifiers to combine are considered with different weights. A new weighted classifier combination method is proposed based on ER to enhance the classification accuracy. The optimal weighting factors of classifiers are obtained by minimizing the distances between fusion results obtained by Dempster's rule and the target output in training data space to fully take advantage of the complementarity of the classifiers. A confusion matrix is additionally introduced to characterize the probability of the object belonging to one class but classified to another class by the fusion result. This matrix is also optimized using training data jointly with classifier weight, and it is used to modify the fusion result to make it as close as possible to truth. Moreover, the training patterns are considered with different weights for the parameter optimization in classifier fusion, and the patterns hard to classify are committed with bigger weight than the ones easy to deal with. The pattern weight and the other parameters (i.e., classifier weight and confusion matrix) are iteratively optimized for obtaining the highest classification accuracy. A cautious decision making strategy is introduced to reduce the errors, and the pattern hard to classify will be cautiously committed to a set of classes, because the partial imprecision of decision is considered better than error in certain case. The effectiveness of the proposed method is demonstrated with various real datasets from UCI repository, and its performances are compared with those of other classical methods.

This paper proposes a novel method based on deformable dictionary learning for detecting the regions of change between multitemporal image pairs. We build on our previous work, which constructed a pair of dictionaries. The main shortcoming of this method was its dependence on a large amount of training data. In practice, there is often a shortage of ground-truthed training images, which limits the expression capability of the resulting dictionaries. This paper overcomes this challenge by incorporating the concept of deformation, wherein each atom of a dictionary is no longer a simple image patch, but instead is a flexible image deformation function. This enables the creation of more expressive dictionaries, capable of generalizing to a far greater variety of image patterns, while using a far smaller amount of ground-truthed images for supervised dictionary training. Deformation similarity is employed for patch matching to find the best set of atoms in the difference image (DI) dictionary for reconstructing image patches for a new input DI. Each such atom can be deformed to achieve a better match, thus extending generality while reducing the number of atoms needed in the dictionary. Multiple deformed atoms are weighted and combined to best reconstruct the input DI patch. Then, the same set of deformations and weights is projected to the corresponding atoms in the CD dictionary to obtain the output change-detection map. Experiments in six realistic synthetic aperture radar data sets demonstrate the robustness and efficiency of the proposed method in comparison with five other state-of-the-art methods from the literature.

The change detection in heterogeneous remote sensing images remains an important and open problem for damage assessment. We propose a new change detection method for heterogeneous images (i.e., SAR and optical images) based on homogeneous pixel transformation (HPT). HPT transfers one image from its original feature space (e.g., gray space) to another space (e.g., spectral space) in pixel-level to make the pre-event and post-event images represented in a common space for the convenience of change detection. HPT consists of two operations, i.e., the forward transformation and the backward transformation. In forward transformation, for each pixel of pre-event image in the first feature space, we will estimate its mapping pixel in the second space corresponding to post-event image based on the known unchanged pixels. A multi-value estimation method with noise tolerance is introduced to determine the mapping pixel using -nearest neighbors technique. Once the mapping pixels of pre-event image are available, the difference values between the mapping image and the post-event image can be directly calculated. After that, we will similarly do the backward transformation to associate the post-event image with the first space, and one more difference value for each pixel will be obtained. Then, the two difference values are combined to improve the robustness of detection with respect to the noise and heterogeneousness (modality difference) of images. Fuzzy-c means clustering algorithm is employed to divide the integrated difference values into two clusters: changed pixels and unchanged pixels. This detection results may contain some noisy regions (i.e., small error detections), and we develop a spatial-neighbor-based noise filter to further reduce the false alarms and missing detections using belief functions theory. The experiments for change detection with real images (e.g., SPOT, ERS, and NDVI) during a flood in U.K. are given to validate the effectiveness of the proposed method.

In the complex pattern classification problem, the fusion of multiple classification results produced by different attributes is able to efficiently improve the accuracy. Evidence theory is good at representing and combining the uncertain information, and it is employed here. Each attribute (set) can be considered as one source of evidence (information). In some applications, the observation of target attributes can be costly, and some unreliable information sources may harm the fusion result. Therefore, we want to use as few as possible sources of information with high quality to achieve the admissible classification accuracy. So we propose a new fusion method based on the adaptive selection of the information sources for pattern classification. For each pattern, the attribute (set) producing the highest accuracy among the various ones will be chosen to classify the pattern at first. If the reliability of classification result, which is evaluated by the K-nearest neighbors (K-NN) technique using training data, cannot satisfy the request, the next attribute source will be chosen according to its classification performance on the selected neighborhoods of the object. In the fusion, the classification results corresponding to different attributes are assigned different weights because of their different classification abilities, and the weighted evidence combination method is adopted to produce the best possible classification performance. Several real data sets from UCI have been used for the evaluation of the proposed method by comparison with other related fusion methods, and it shows that our new method can produce higher accuracy with smaller number of information sources than the other fusion methods which are directly used to combine all the sources of information.

In the complex pattern classification problem, the reliability of classifier output for the patterns located at different regions of the data set may be different. In order to efficiently improve the classification accuracy, we propose a new method to correct the original classifier output using the local knowledge of the classifier performance in different regions. The training data set can be divided into some small clusters corresponding to different regions. The prior knowledge of the classifier performance on each cluster is characterized by a confusion matrix representing the conditional probability of the pattern belonging to one class but committed to another class by the classifier. The matrix associated with each cluster is learnt by minimizing an error criteria using training data, which is assigned different weights to achieve the highest possible accuracy. If the classification accuracy of the training data in one cluster can be improved according to the corrected classification results, the associated confusion matrix becomes valid. Otherwise, the confusion matrix is invalid and patterns in this cluster cannot be modified any more. For each object, if it lies in the cluster with valid confusion matrix, its classification result will be corrected by the matrix before making the class decision. The above correction process can be regarded as the fusion of local and global information. Several experiments are given to test the performance of the proposed method using real data sets, and it shows that the new method is able to efficiently improve the classification accuracy compared with other related methods.

A new change detection method for heterogeneous remote sensing images (i.e. SAR & optics) has been proposed via pixel transformation. It is difficult to directly compare the pixels from heterogeneous images for detecting changes. We propose to transfer the pixels in different images to a common feature space for convenience of comparison. For each pixel in the 1stimage, it will be transferred to the 2ndfeature space associated with the 2ndimage according to the given unchanged pixel pairs. In fact, this transformation is done assuming that the pixel is not affected by the events. Then the difference value between the estimation of transferred pixel and the actual one in the same location of the 2ndimage can be calculated. The bigger difference value, the higher possibility of change happening. We can similarly do the opposite transformation from the 2ndimage to the 1stimage, and one more difference value is obtained in the 1stfeature space. Change occurrences will be detected using Fuzzy C-means clustering method based on the sum of two difference values. The flood detection in the SAR and optical images is given in the experiments, and it shows that the proposed method is able to efficiently detect changes.

In classification problem, several different classes may be partially overlapped in their borders. The objects in the border are usually quite difficult to classify. A hybrid classification system (HCS) is proposed to adaptively utilize the proper classification method for each object according to the K-nearest neighbors (K-NNs), which are found in the weighting vector space obtained by self-organizing map (SOM) in each class. If the K-close weighting vectors (nodes) are all from the same class, it indicates that this object can be correctly classified with high confidence, and the simple hard classification will be adopted to directly classify this object into the corresponding class. If the object likely lies in the border of classes, it implies that this object could be difficult to classify, and the credal classification working with belief functions is recommended. The credal classification allows the object to belong to both singleton classes and sets of classes (meta-class) with different masses of belief, and it is able to well capture the potential imprecision of classification thanks to the meta-class and also reduce the errors. Fuzzy classification is selected for the object close to the border and hard to clearly classify, and it associates the object with different classes by different membership (probability) values. HCS generally takes full advantage of the three classification ways and produces good performance. Moreover, it requires quite low computational burden compared with other K-NNs-based methods due to the use of SOM. The effectiveness of HCS is demonstrated by several experiments with synthetic and real datasets.

In this work, a new prototype-based clustering method named Evidential C -Medoids (ECMdd), which belongs to the family of medoid-based clustering for proximity data , is proposed as an extension of Fuzzy C -Medoids (FCMdd) on the theoretical framework of belief functions . In the application of FCMdd and original ECMdd, a single medoid (prototype), which is supposed to belong to the object set, is utilized to represent one class. For the sake of clarity, this kind of ECMdd using a single medoid is denoted by sECMdd. In real clustering applications, using only one pattern to capture or interpret a class may not adequately model different types of group structure and hence limits the clustering performance. In order to address this problem, a variation of ECMdd using multiple weighted medoids, denoted by wECMdd, is presented. Unlike sECMdd, in wECMdd objects in each cluster carry various weights describing their degree of representativeness for that class. This mechanism enables each class to be represented by more than one object. Experimental results in synthetic and real data sets clearly demonstrate the superiority of sECMdd and wECMdd. Moreover, the clustering results by wECMdd can provide richer information for the inner structure of the detected classes with the help of prototype weights.

The classification of incomplete patterns is a very challenging task because the object (incomplete pattern) with different possible estimations of missing values may yield distinct classification results. The uncertainty (ambiguity) of classification is mainly caused by the lack of information of the missing data. A new prototype-based credal classification (PCC) method is proposed to deal with incomplete patterns thanks to the belief function framework used classically in evidential reasoning approach. The class prototypes obtained by training samples are respectively used to estimate the missing values. Typically, in a c -class problem, one has to deal with c prototypes, which yield c estimations of the missing values. The different edited patterns based on each possible estimation are then classified by a standard classifier and we can get at most c distinct classification results for an incomplete pattern. Because all these distinct classification results are potentially admissible, we propose to combine them all together to obtain the final classification of the incomplete pattern. A new credal combination method is introduced for solving the classification problem, and it is able to characterize the inherent uncertainty due to the possible conflicting results delivered by different estimations of the missing values. The incomplete patterns that are very difficult to classify in a specific class will be reasonably and automatically committed to some proper meta-classes by PCC method in order to reduce errors. The effectiveness of PCC method has been tested through four experiments with artificial and real data sets.

We present a multidimensional evidential reasoning (MDER) approach to estimate change detection from the fusion of heterogeneous remote sensing images. MDER is based on a multidimensional (M-D) frame of discernment composed by the Cartesian product of the separate frames of discernment used for the classification of each image. Every element of the M-D frame is a basic joint state that allows to describe precisely the possible change occurrences between the heterogeneous images. Two kinds of rules of combination are proposed for working either with the free model, or with a constrained model depending on the integrity constraints one wants to take into account in the scenario under study. We show the potential interest of the MDER approach for detecting changes due to a flood in the Gloucester area in the U.K. from two real ERS and SPOT images.

A credal classification rule (CCR) is proposed to deal with the uncertain data under the belief functions framework. CCR allows the objects to belong to not only the specific classes, but also any set of classes (i.e. meta-class) with different masses of belief. In CCR, each specific class is characterized by a class center. Specific class consists of the objects that are very close to the center of this class. A meta-class is used to capture imprecision of the class of the object that is simultaneously close to several centers of specific classes and hard to be correctly committed to a particular class. The belief assignment of the object to a meta-class depends both on the distances to the centers of the specific class included in the meta-class, and on the distance to the meta-class center. Some objects too far from the others will be considered as outliers (noise). CCR provides the robust classification results since it reduces the risk of misclassification errors by increasing the non-specificity. The effectiveness of CCR is illustrated by several experiments using artificial and real data sets.

Theories of evidence have already been applied more or less successfully in the fusion of remote sensing images. These attempts were based on the classical evidential reasoning which works under the condition that all sources of evidence and their fusion results are related to the same invariable (static) frame of discernment. When working with multitemporal remote sensing images, some change occurrences are possible between two images obtained at a different period of time, and these changes need to be detected efficiently in particular applications. The classical evidential reasoning is adapted for working with an invariable frame of discernment over time, but it cannot efficiently detect nor represent the occurrence of change from heterogeneous remote sensing images when the frame is possibly changing over time. To overcome this limitation, dynamic evidential reasoning (DER) is proposed for the sequential fusion of multitemporal images. A new state-transition frame is defined in DER, and the change occurrences can be precisely represented by introducing a statetransition operator. Two kinds of dynamical combination rules working in the free model and in the constrained model are proposed in this new framework for dealing with the different cases. Moreover, the prior probability of state transitions is taken into account, and the link between DER and Dezert–Smarandache theory is presented. The belief functions used in DER are defined similarly to those defined in the Dempster–Shafer theory. As shown in the last part of this paper, DER is able to estimate efficiently the correct change detections as a postprocessing technique. Two applications are given to illustrate the interest of DER: The first example is based on a set of two SPOT images acquired before and after a flood, and the second example uses three QuickBird images acquired during an earthquake event.