This paper proposes a novel interpretable framework for fine-grained aircraft classification in high-stakes remote sensing applications. Our approach addresses three key challenges: small inter-class variance, large intra-class variance, and the need for model interpretability. Specifically, our framework is built on the Swin Transformer (SwinT) backbone and includes three main modules. First, we present the Dynamic Attention Fusion Module (DAFM), which adaptively fuses multi-stage attention maps from the SwinT backbone. By leveraging a dispersion-based weighting mechanism, DAFM balances the contributions of coarse and fine-grained features, capturing both global structures and localized details. Second, we propose the Adaptive Image Pair Interaction Module (AIPI), which dynamically adjusts feature interaction strategies based on intra-class and inter-class similarity, effectively enhancing informative regions and improving robustness. To further optimize discriminative power, we incorporate an AIPI loss function that enforces intra-class consistency and inter-class separability. Finally, we develop a Binary Neural Tree Module (BNTM) to hierarchically select and propagate informative image patches, enhancing both feature refinement and interpretability through explicit path-based decision-making. Extensive experiments on benchmark datasets demonstrate that our framework significantly improves classification accuracy and interpretability, making it well-suited for applications requiring transparent and reliable decision-making. The Code can be found at https://github.com/StarmanGzx/BNTM.

Land cover classification (LCC) based on remote sensing image segmentation is a prominent task of remote sensing data interpretation. The commonly used optical data is susceptible to the weather, so it has the potential to utilize complementary features from the supplementary synthetic aperture radar (SAR) data to enhance segmentation performance. However, current multi-modal segmentation methods focus on the deep fusion of features, which usually ignores the significance of structural consistency information. In order to make use of the mutual correction and information exchange between multi-modal data, we propose DCIFNet, a dual-stream correction-interaction-fusion multi-modal LCC network. Specifically, we design a differential feature correction and enhancement module (DF-CEM) that leverages bidirectional differential features to correct multi-modal features. In addition, for corrected feature pairs, we deploy a parallel attention interaction module (PAIM) to focus on the pixel-level feature correlation and achieve effective information exchange in both channel and spatial dimensions. Through the expert fusion module (EFM), DCIFNet leverages the gate network to attain a flexible and compact feature fusion between multi-modal features. Experimental results show that our method achieves a superior performance compared with other multi-modal fusion segmentation methods on three optical-SAR datasets. The source code of DCIFNet is publicly available at https://gitee.com/asdwer2046/dcifnet.

With the exponential growth of satellite remote sensing data, land cover classification models must adapt continuously to new classes. However, conventional incremental learning methods face critical challenges: catastrophic forgetting degrades recognition of old classes, and the softmax function further suppresses old-class probabilities due to ”class crowding.” Existing distillation techniques also struggle to transfer features in irregular geospatial regions. To address these issues, we propose Soft Labels and Subregion Distillation (SLSRD). SLSRD mitigates class crowding by employing soft labels instead of hard labels, derived from a hybrid of softmax and sigmoid outputs that preserve richer probabilistic information. Concretely, the soft label is a convex combination of softmax- and sigmoid-based probabilities that preserves inter-class relations while relaxing over-confident exclusivity for newly introduced categories, and it supervises all pixels across stages. In parallel, a breadth-first search identifies subregions within each image, which are weighted by probability and size, and similarity between corresponding subregions of the old and new models is maximized. This dual strategy effectively transfers fine-grained knowledge and overcomes the limitations of conventional distillation methods, particularly for large-scale remote sensing imagery. Experiments on three benchmark datasets-Vaihingen, GID, and FBP-demonstrate that SLSRD outperforms traditional methods, significantly improving incremental land cover classification.

Land cover classification (LCC) is an important application in remote sensing data interpretation. As two common data sources, SAR images can be regarded as an effective complement to optical images, which will reduce the influence caused by single-modal data. But common LCC methods are focusing on designing advanced network architectures to process single-modal remote sensing data. Few works have been oriented toward improving segmentation performance through fusing multi-modal data. In order to deeply integrate SAR and optical features, we propose SwinTFNet, a dual-stream deep fusion network. Through the global context modeling capability of Transformer structure, SwinTFNet models teleconnections between pixels in other regions and pixels in cloud regions for better prediction in cloud regions. In addition, a Cross-Attention Fusion Module (CAFM) is proposed to fuse features from optical and SAR data. Experimental results show that our method improves greatly in the classification of clouded images compared with other excellent segmentation methods and achieves the best performance on multi-modal data.

Model-based target decomposition method has been widely applied due to its clear physical scattering significance. However, after establishing decomposition basis, the process of solving the scattering components and parameters is usually underdetermined, which will lead to the issues such as component negative power and overestimation. For this problem, this letter examines the target decomposition task from the perspective of deep learning and proposes an adversarial decomposition feature learning (ADFL) model. This model could learn decomposition features suitable for current terrain characteristics according to input data. At the same time, the model-based adversarial feature prior is embedded in ADFL to maintain the physical scattering meanings. On real PolSAR datasets, the learned features of proposed model are well correlated with real terrain scattering characteristics. Further, it avoids negative decomposition features and make more accurate fitting of scattering components, effectively alleviating the above problems.

The task of fine-grained aircraft recognition is crucial in the field of remote sensing. Despite some progress achieved by traditional deep learning methods in addressing this challenge, they are often perceived as a “black box,” lacking transparent explanations for model decisions. Current interpretable methods based on attention mechanisms, although providing some interpretability, do not align with human thought logic. Therefore, we propose a multiscale rotation-invariant prototype network (MSRIP-Net). Our approach simulates the intuitive reasoning process of humans in identifying objects by segmenting them into multiple components. Importantly, MSRIP-Net has the capability to automatically recognize rigid components on aircraft targets without relying on additional part annotations, using only image-level class labels. In addition, our approach effectively addresses challenges presented by noise, deformations, and multiscale variations in remote sensing targets and has been comprehensively evaluated on datasets FAIR1M1.0 and Rareplane. Our results demonstrate that MSRIP-Net achieves higher accuracy compared with existing fine-grained recognition methods. Furthermore, we provide insights into the model’s decision-making process to illustrate the interpretability of our approach.

Aerial image objects are usually orientated arbitrarily, with a large scale range, and densely distributed. Traditional horizontal bounding box (HBB) detectors tend to filter out densely distributed objects leading to missed detections, such as ship (SH) and vehicle. Therefore, oriented object detection has become a mainstream solution in recent years. The 2-D Gaussian distribution representation of the oriented bounding boxes (OBBs) solves the problem of angular discontinuity and boundary discontinuity and thus gets more attention. However, as the aspect ratio of the object gradually decreases, its predicted angular performance continues to decrease. We find that the angular gradient of an object decreases sharply as the aspect ratio decreases, resulting in a large gradient gap between a small aspect ratio object (SARO) and a large aspect ratio object (LARO). It makes the detector prefer to ignore SARO during training, which weakens the high precision performance of SARO. We call this phenomenon shape imbalance. To solve the problem, we proposed a simple gradient rebalancing strategy named shape balance. Since the shape imbalance is only related to the aspect ratio of the object, we designed a modulation function with an inverse aspect ratio to calculate the balance coefficient. The principle of the function is that the larger the aspect ratio, the smaller the balance coefficient; the smaller the aspect ratio, the larger the balance coefficient. We aim to get the balance coefficients for objects with different aspect ratios. Location loss multiplied by a balance coefficient can directly adjust the gradient gap between objects with different aspect ratios to achieve a rebalancing effect. Extensive experiments conducted on DOTA-v1.0 dataset and DIOR-R dataset verify the effectiveness of our proposed method. Our method improves the detection performance of Gaussian location loss by an average of 2.08%/1.01%(AP75/mAP) metrics on the DOTA-v1.0 dataset and 1.17%/0.82%(AP75/mAP) improvements for DIOR-R dataset.

Deep learning works normally in PolSAR image classification because the complex terrain scattering characteristic results in large intraclass differences and high interclass similarity. Deep metric learning (DML) aims to make the features keep a closer intraclass and a farther interclass distance. Therefore, we introduce DML and then propose an N-cluster generative adversarial net (N-cluster GAN) framework for PolSAR image classification. However, existing DML losses mainly focus on the relationship between individual samples in feature space. Hence, we propose N-cluster loss that pays more attention to the overall structure of all samples. Meanwhile, traditional hard negative sample mining methods occupy lots of computational resources. In addition, the hard level of the negative samples will affect the model’s performance. Therefore, we explore a new method based on a GAN framework to replace the sample mining. Positive N-cluster loss is added to the discriminator ($D$), and a negative one is added to the generator ($G$). In this way,$D$will possess better classification ability, and$G$can produce hard negative samples for$D$. Then, the hard level of the generated negative samples will change with the discrimination of$D$, which is appropriate for the proposed model. N-cluster loss can be directly calculated through the extracted features rather than redundant data preparation. The proposed model is verified on four PolSAR datasets from two aspects of the loss function and negative samples mining. Then, it achieves competitive performance compared with state-of-the-art algorithms.

Target decomposition features are the cornerstone of subsequent analyses for PolSAR images. Generally, adopting single or several decomposition algorithms limits the representation ability for original terrain characteristics. Using all the existing decomposition features, however, will definitely increase computational complexity. Besides, some features even have a negative effect on the following tasks. To address these problems, a sparse variational autoencoder feature selection framework (SVAE-FS) is proposed in this article. In detail, the encoder transforms the original feature set into latent space and then decoder reconstructs the corresponding pseudo set on this latent space. Similarly, a pseudo subset is subsequently obtained by the SVAE. The discrepancy, namely reconstruction error, between the pseudo set and the pseudo subset is taken as an evaluation criterion which reflects the feature representation ability of pseudo subset. Sparse constraint in the encoder makes the representative features stand out. Meanwhile, the linear feature transformation layer of the encoder enables the SVAE to evaluate different scale subsets without repeated training. Finally, a greedy selection approach with search scale$K$is proposed to find the suboptimal subset. This procedure not only reduces time consumption, but also ensures the performance of the subset. The selected features are analyzed on four real PolSAR datasets according to the terrain scattering characteristics. Furthermore, these features have achieved competitive performance on three PolSAR image tasks.

Model-based polarimetric target decomposition (TD) generally solves scattering components and parameters under pre-set decomposition base, then decomposition features are also obtained. However, pre-set base could not be adjusted according to different scenes. Furthermore, solving the polarimetric parameters needs to explore additional information or consider limiting conditions to build equations, which is hard and easily to bring negative effects into decomposition features. To this end, we regard the TD as a process of learning decomposition base and features by deep learning. Then, the polarimetric decomposition feature learning (PDFL) model is proposed in this paper. Strictly, this model is not an incoherent TD method but a learning-based method. It dose not need to construct the parameter solution equations or fixed base. Then, the decomposition base and feature can be adaptively learned according to scattering characteristics of current dataset. Due to the characteristics of unsupervised reconstruction, deep auto encoder (DAE) is used as the model foundation. Then, some adjustments and constraints are utilized to make the DAE fit closely with TD. The encoder extracts latent vector from PolSAR data, then the decoder reconstructs pseudo data on this latent vector. The reconstruction can be regarded as the inverse process of TD, so the base matrix of decoder and the latent vector indicate the learned decomposition base and features when the model converges. The effectiveness of PDFL is verified on simulated and real PolSAR datasets. Compared with representative algorithms, proposed model gains more discriminative features and achieves competitive performance on terrain classification and segmentation tasks.

In order to better interpret polarimetric synthetic aperture radar (PolSAR) images, many scholars tend to do target decomposition for PolSAR images and utilize the obtained features to perform subsequent classification. These target decomposition features play an important role in terrain classification but completely utilizing them produces a high computational complexity. Furthermore, some features have a negative impact on the classification task. Therefore, selecting the appropriate amount of high-quality features is of great significance to the classification task. In this paper, we propose a convolutional neural network (CNN)-based feature selection algorithm for PolSAR image classification. First, we design a 1-D CNN for feature selection, then train the designed network with all the decomposition features to obtain a trained model. Second, the Kullback-Leibler distance (KLD) between different features is utilized as a standard to select feature subsets. Third, feature subsets with excellent performance form the final results. Due to the special structure of the 1-D CNN, repetitively training model is avoided when the input changes. Different from traditional feature selection methods, our method considers the performance of features combination rather than single feature contribution. To this end, the feature subsets selected by the proposed method are more useful to the classification task. Innovatively introducing KLD in the selection stage avoids random selection and improves the selection efficiency. Finally, we validate the performance of selected feature subsets in traditional and deep learning classification frameworks. Experiments demonstrate that features selected by the proposed method have a good performance comparing with others on three real PolSAR data sets.

Compared with traditional pixel-based polarimetric synthetic aperture radar (PolSAR) image classification methods, superpixel-based methods take advantages of the spatial information of pixels, so they can overcome the influence of speckle noise on the classification result. Since traditional superpixel methods do not utilize the scattering characteristics of a PolSAR image, the boundaries of the superpixels are poorly preserved. The inaccuracy of superpixel segmentation boundaries has a negative impact on the subsequent classification. In this letter, we propose a decomposition-feature-iterative-clustering (DFIC) superpixel segmentation method for PolSAR images. The DFIC method innovatively introduces the decomposition features in generating superpixels, so the superpixel segmentation boundaries are well preserved. Because we selectively utilize superpixel information to classify the PolSAR images by setting a threshold, the effect of superpixel segmentation inaccuracy on the classification results is reduced. Experiments on two real PolSAR images demonstrate that the proposed method outperforms several state-of-the-art superpixel methods, and that the DFIC superpixel-based classification obtains better results than the other pixel-based methods.