Existing hyperspectral anomaly detection methods primarily focus on spectral or spatial differences between anomalies and the background. However, these methods often overlook spectral aliasing around anomalous targets and the interference of anomalous pixels in background reconstruction. To address these issues, an adaptive Taylor decomposition and polynomial kernel differential fusion network is proposed. First and foremost, the adaptive Taylor decomposition method is proposed to perform spectral decomposition of the hyperspectral image. Moreover, the decomposed hyperspectral image is fed into a polynomial kernel differential fusion module to enhance multi-scale feature processing capabilities. Then, a weakened diagonal convolution module is constructed to reduce the influence of diagonal pixels on the tested pixels. At last, the residual is computed to obtain the final anomaly detection results. The key advantage of this method lies in enhancing the distinction between the background and anomalous targets, while effectively mitigating the degradation of background reconstruction quality caused by anomalies. Experimental results show that the method improves background reconstruction and network robustness. The proposed method also outperforms ten state-of-the-art methods.

Semi-supervised learning (SSL) faces significant challenges when applied to highly imbalanced datasets, often leading to biased pseudo-labels and degraded model performance. To address this, we introduce a novel framework that integrates Balanced Mini-Mixup (BMM) and Dynamic Distribution Alignment (DDA) for robust SSL under class imbalance. The proposed method mitigates majority-class dominance by employing adaptive thresholding for pseudo-label selection and encourages fair representation through a balanced mixup strategy that preserves class diversity during training. Without increasing model complexity, our approach achieves superior generalization. Extensive evaluations on CIFAR-10, STL10, and a real-world fish dataset demonstrate consistent improvements in class-wise F1-scores, outperforming state-of-the-art baselines in imbalanced settings.

Convolutional neural networks (CNNs) are a powerful tool for various computer vision tasks, demonstrating exceptional performance in image classification, object detection, and segmentation. However, traditional training methods often require meticulous hyperparameter tuning, architectural adjustments, or the introduction of additional data through techniques such as data augmentation to achieve optimal accuracy. This letter introduces an innovative training strategy that leverages layer freezing to enhance the training process while keeping the model's architecture and hyperparameters unchanged. By selectively and progressively freezing certain hidden layers in the CNN, we prevent the model from reaching a saturation point. This approach effectively reduces the backpropagation parameter space, facilitating more focused and efficient learning in the remaining layers.

This letter introduces an innovative approach for image restoration. Our model is primarily motivated by the integration of curvature constraints into a self-supervised convolutional neural network (CNN), which combine the hand-crafted prior with CNN structure without training on external dataset. Firstly, it is argued that detail loss may be induced by sparsity-based models, which eliminate bases with low coefficients. Therefore, a geometric approach is proposed to preserve details, with the incorporation of Gaussian curvature as a regularization term for both noise suppression and detail preservation. Secondly, the deep image prior is used to establish the optimization backbone. This framework harnesses the translation in-variance and smoothness properties of CNN as a tightly supervised recovery mechanism for effective noise suppression. By combining a simple data-fitting term with curvature-based regularization terms, we develop a self-supervised model for image restoration that maintains fine details. Experimental results demonstrate the effectiveness of our proposed algorithm in addressing various image restoration problems.

Abstract One of the important factors in deep feature learning is their loss function design which highly influences the network performance. In this paper, we proposed a classroom (CR) learning approach along with arcface loss for contactless palmprint recognition to obtain high-level discriminative features without any extra load made to the network architecture. CR loss allows the network to learn the best possible feature representations for palmprint images. To validate our concept, we performed extensive experimental evaluations on various popular benchmark palmprint databases where our methods outperform the state-of-the-art methods. We also introduced a challenging contactless palmprint database called Harbin Institute of Technology-Network & Information Security Research Center contactless palmprint database version 1.0 (HIT-NIST-V1), as a new contribution to this domain. The result proves that the proposed CR loss consistently outperforms the SOTA methods for all the considered databases and especially for HIT-NIST-V1.

Abstract On-line gambling is the key illegal behaviour of public security department in most countries due to the potential threat to cyberspace security and social stability. Hence, the research on gambling domain names (GDN) classification is quite important and in great demand for academia and industry. Till now, there is very little research work on this topic. Most of the GDN training datasets in previous work were chosen from GDN blacklists provided by publicly available data sources, and the authors did not verify the authenticity and accuracy of these datasets, and the classification results are not particularly satisfactory. In this paper, certificated and textual analysis-based classification method CT-GDNC is proposed to get GDN training data set with an accuracy of 0.9776 and significantly improve the classification results of GDN. The exhaustive comparative experiments on 10K GDN obtained via Bert fine-tuning model and 10K benign data collected from Alex Top 1 million list show that the proposed method achieves new baseline result for GDN classification with classification accuracy 0.9936, precision 0.9936, F1 0.9936 and recall 0.9939.

View-based methods have achieved state-of-the-art performance in 3D object retrieval. However, view-based methods still encounter two major challenges. The first is how to leverage the inter-view correlation to enhance view-level visual features. The second is how to effectively fuse view-level features into a discriminative global descriptor. Towards these two challenges, we propose a multi-range view aggregation network (MRVANet) with a vision transformer based feature fusion scheme for 3D object retrieval. Unlike the existing methods which only consider aggregating neighboring or adjacent views which could bring in redundant information, we propose a multi-range view aggregation module to enhance individual view representations through view aggregation beyond only neighboring views but also incorporate the views at different ranges. Furthermore, to generate the global descriptor from view-level features, we propose to employ the multi-head self-attention mechanism introduced by vision transformer to fuse the view-level features. Extensive experiments conducted on three public datasets including ModelNet40, ShapeNet Core55 and MCB-A demonstrate the superiority of the proposed network over the state-of-the-art methods in 3D object retrieval.

Recently, autonomous vehicles (AVs) have received a lot of attention from the automotive industry as well as the AV research community across the globe. To increase the safety of the flow of traffic, they are anticipated to help or perhaps take the place of human drivers when it comes to vehicle manoeuvring at various levels of autonomy. But before they can be widely used, AVs must first be developed to overcome their inherent security and road safety issues. The adoption of autonomous vehicles depends on the results of the driving test and their safety validation on the roads. This paper examines the associated autonomous vehicle testing and validation methodologies such as autonomous vehicle functional testing, integrated vehicle testing, and system verification across many architectures. In addition, the paper presents some of the state-of-the-art machine learning algorithms used for AV operation on roads. Knowledge of such recent trends in the validation and verification techniques for road safety will be helpful for the development of autonomous vehicles.

Detecting defects buried inside chips is critical for failure analysis in semiconductor manufacturing. In this paper, we perform 3D voxel segmentation on 2.5D semicon chips to locate and identify defects that may be present in them. We integrate tree based Ensemble method with the Cascaded Anisotropic Convolutional Neural Networks to employ component structural characteristics of semicon 3D object in voxel segmentation process. We fabricate custom 2.5D chips purposely creating defective regions by using a specific fabrication and assembling process. Thereafter, use commercial 3D XRM tools for 3D imaging of these chips. We perform accurate 3D Object localization for each 3D x-ray scan by using a slice and fuse approach. Then, we perform voxel segmentation on logic die (integral component of semicon chip) to detect Cu-pillar, solder, and void regions (if any). The results show that we achieve state-of-the-art voxel segmentation dice scores for all three sub-components.

Defect segmentation from 2D images plays a critical role in industrial product quality assessment. In practice, it is common that there are sufficient normal (defect-free) images but a very limited number of anomalous (defective) images. The existing works proposed several UNet variants (e.g., CAM-UNet) by incorporating normal images into the training process to improve the defect segmentation performance. In this paper, we propose Multi-Level Spatial Attention UNet (MLSA-UNet) to address the industrial defect segmentation task. MLSA-UNet is trained in an end-to-end manner to simultaneously classify normal/anomalous images and segment out defective regions from anomalous images. The classification process is conducted by Spatial Attention Learning Module (SALM) to generate multi-level spatial attention maps which are exploited by Spatial Attention Guided Decoding Module (SADM) to provide the guidance in the decoding process of UNet. Extensive experiments on MVTec AD dataset demonstrate the superiority of the proposed MLSA-UNet over multiple state-of-the-art UNet variants on defect segmentation.

With the COVID-19 pandemic, one critical measure against infection is wearing masks. This measure poses a huge challenge to the existing face recognition systems by introducing heavy occlusions. In this paper, we propose an effective masked face recognition system. To alleviate the challenge of mask occlusion, we first exploit RetinaFace to achieve robust masked face detection and alignment. Secondly, we propose a deep CNN network for masked face recognition trained by minimizing ArcFace loss together with a local consistency regularization (LCR) loss. This facilitates the network to simultaneously learn globally discriminative face representations of different identities together with locally consistent representations between the non-occluded faces and their counterparts wearing synthesized facial masks. The experiments on the masked LFW dataset demonstrate that the proposed system can produce superior masked face recognition performance over multiple state-of-the-art methods. The proposed method is implemented in a portable Jetson Nano device which can achieve real-time masked face recognition.

The current state-of-the-art (SOTA) methods validate the role of shape in object categorization, however, except few, most of them neglect object shape information. Motivated by low-shot learning and increasing synthetic data in vision tasks, we investigated how image-based embedding generalization can be improved by the data itself. We propose a new data augmentation approach for low-shot object generalization regime based on image-only. The proposed method learns a discriminative embedding space using SIFT shape points for 3D objects, such that it’s easier to map images and point clouds into one. Numerous experiments show that the proposed approach is superior to the existing low-shot SOTA methods.

Abstract As the backbone of the domain name system (DNS), DNS resolvers are essential to the Internet. Nowadays, the measurement of DNS resolvers, especially open DNS resolvers, has become a research focus. Previous research works show that DNS responses returned from some open DNS resolvers are not expected for clients and the Internet. We call these DNS responses ‘unexpected DNS responses’. Research on unexpected DNS responses is beneficial to the research, usage and management of open DNS resolvers. This paper explores unexpected DNS responses returned from open DNS resolvers in terms of identification and classification to better understand the behaviours of open DNS resolvers. First, an identification method is proposed to identify all kinds of DNS responses from each section of DNS messages. Second, a classification method is proposed to classify unexpected DNS responses by their influences on clients and the Internet. Furthermore, an efficient identification and classification method is proposed to simplify the above process. Among about 9 million responding open DNS resolvers in the experiments on the IPv4 address space, about 40% return unexpected DNS responses. Experimental results show that the proposed methods can identify and classify all kinds of DNS responses returned from open DNS resolvers.

Abstract Palmprint is an important key player in biometric family and also informs some extra basic personality details of an individual. In this paper, we utilize these extra information and designed an automated mobile vision (MV) system to extract principal lines from human palm and analyze them for behavioral significances. Hence, the main concern of this paper is to come up with a simple yet powerful low-level MV solution to extract the complex challenging features from palmprint. In the proposed system, the computational tasks are offloaded to a dedicated palmistry server and efficiently minimizes the energy consumption of mobile device after performing some preliminary computational low-level tasks. The implementation is divided into four major phases: (i) hand-image acquisition and pre-processing, (ii) region-of-interest extraction from the palm images, (iii) post-processing to extract principal lines and (iv) features computation for behavior analysis. The basic palmistry uses line lengths, angles, curves and branches to identify a person’s behavior. The exhaustive experiments show that the proposed system achieves an average accuracy of 96%, 92% and 84% for heart, life and head line detection and personality prediction, respectively. Finally, mapping the extracted results with the original palmprint is augmented back to the use for better visualization.

3D object recognition is a well studied 2D multi-view object classification task that achieves high accuracy if the object textures are distinctive. However, if objects are texture-less and are only differentiable by their shapes but at certain viewpoints. Thus, the problem is still very challenging. Furthermore, the existing methods are mostly based on supervised learning with lots of images per object which are difficult to collect and label them for training. In this letter, we introduced a multi-loss view invariant stochastic prototype embedding to minimize and improve the recognition accuracy of novel objects at different viewpoints by using a progressive multi-view learning approach. An extensive experimental results show that the proposed method outperforms the state-of-the-art methods on different types datasets and also on different backbones.

Defect segmentation is critical in real-wold industrial product quality assessment. There are usually a huge number of normal (defect-free) images but a very limited number of annotated anomalous images. This poses huge challenges to exploiting Fully-Convolutional Networks (FCN), e.g., UNet, as they require sufficient anomalous images with defect annotations during training. To further leverage the information from normal data, a novel CAM-guided U-Net with adversarial regularization (CAM-UNet-AR) is proposed. We first modify the existing CAM-UNet to incorporate the CAMs for both normal and anomalous classes and fine-tune the segmentation network using a combined loss which jointly considers pixel-wise classification, foreground segmentation and boundary segmentation. Secondly, an auxiliary adversarial regularization module (ARM) is proposed to facilitate the segmentation network to encode the “normal components” from training images into consistent representations. Extensive experiments on MVTec AD dataset show the superiority of our proposed network over multiple state-of-the-art U-Net variants.

To reduce the risk of infecting or being infected by the recent COVID-19 virus, wearing mask is enforced or recommended by many countries. AI based system for automatically detecting whether individuals are wearing face mask becomes an urgent requirement in high risk facilities and crowded public places. Due to lacking of existing masked face datasets and the urgent low-cost application requirement, we propose a progressive semi-supervised learning method – called maskedFaceNet to minimize the efforts on data annotation and letting deep models to learn by using less annotated training data. With this method, the detection accuracy is further improved progressively while adapting to various application scenarios. Experimental results show that our maskedFaceNet is more efficient and accurate compared to other methods. Furthermore, we also contribute two masked face datasets for benchmarking and for the benefit of future research.

This paper tackles the task of defect segmentation by exploiting sufficient normal (defect-free) training images and limited annotated anomalous images. We propose a class activation map guided UNet (CAM-UNet) with feedback refinement mechanism for accurate defect segmentation. We first modify and pretrain the encoder of a VGG-16 backboned UNet to classify normal and anomalous training images. Then, for each of the anomalous training images, a CAM is generated as the prior segmentation information. Based on the CAM, we propose a feedback refinement process to train two decoder networks to progressively improve the segmentation output. Extensive experiments conducted on MVTEC AD dataset show that the proposed method significantly outperforms multiple benchmarking UNet methods in terms of mean IOU.

This paper demonstrates the use of 3D Anisotropic Convolutional Neural Network (CNN) with predict-refine mechanism for 3D brain tumor segmentation. We propose two networks that utilize multi-scale feedback and saliency maps respectively to segment three critical regions involved in automated brain tumor segmentation. The proposed networks are formulated to predict feature maps at different resolutions during the prediction phase. These networks perform refinement process using the saliency or feature maps as feedback information for the refinement process. The recurrent architecture allows the network to automatically rectify errors in saliency map of the previous prediction phase resulting in more reliable final predictions. Our experimental results on the BraTS2017 dataset demonstrate the superior performance of our proposed predict-refine architecture than current state of the art approaches improving results by up to 8% without any additional increase in the 1.9M model parameters.

The performance of deep learning models are driven by various parameters but to tune all of them every time, for every given dataset, is a heuristic practice. In this paper, unlike the common practice of decaying the learning rate, we propose a step-wise training strategy where the learning rate and the batch size are tuned based on the dataset size. Here, the given dataset size is progressively increased during the training to boost the network performance without saturating the learning curve, which is seen after certain epochs. We conducted extensive experiments on multiple networks and datasets to validate the proposed training strategy. The experimental results proves our hypothesis that the learning rate, the batch size and the data size are interrelated and can improve the network accuracy if an optimal progressive step-wise training strategy is applied. The proposed strategy also reduces the overall training cost compared to the baseline approach.

A new region-based active contour model in level set formulation is proposed to segment medical images with poorly defined boundaries. From literature, it is observed that the traditional methods often fail to detect weak boundaries for images with intensity inhomogeneity. However, the machine learning (ML) algorithms are highly effective for such images but due to the noise most pixels are misclassified. Therefore in this paper, we propose a region-based active contour model using ML. In this paper, we consider an active contour driven by local Gaussian distribution (LGD) fitting energy which is known as LGD model. Further, this active contour LGD model is integrated with fuzzy k-nearest neighbor (k-NN) for added accurate segmentation. Also the energy stop function (ESF) of LGD model is modified to combine with k-NN. The results obtained are compared with the existing state-of-the-art models and the proposed method is clearly a triumph. The experimental results proves that the proposed model provides higher accuracy results for medical image segmentation and is robust compared to the other existing methods.

This paper addresses the task of automatic cleaning up of photo collections on low computing devices like mobile phones. That is, the task of automatically detecting and rejecting the blur, near duplicate and aesthetically poor images. To this end, a compact signature is presented with the constraints of being compact, simple to compute and use. One key characteristic of the proposed signature is that the same representation is used for all three different tasks that reduces the battery consumption and the storage requirements of a mobile device. The effectiveness of our proposed MIQA signature is experimentally validated by various experiments on several public datasets with subjective image quality evaluation. The proposed representation is optimal for all three IQA.

Close monitoring, proper control and management of plant diseases are essential in the efficient cultivation of crops. This paper presents a scheme that uses mobile phones for real-time on-field imaging of diseased plants followed by disease diagnosis via analysis of visual phenotypes. A threshold based offloading scheme is employed for judicious sharing of the computational load between the mobile device and a central server at the plant pathology laboratory, thereby offering a trade-off between the power consumption in the mobile device and the transmission cost. The part of the processing carried out in the mobile device includes leaf image segmentation and spotting of disease patch using improved k-means clustering. The algorithm is simple and hence suitable for Android based mobile devices. The segmented image is subsequently communicated to the central server. This ensures reduced transmission cost compared to that in transmitting full leaf image.

This paper presents a fast and efficient hand gesture based mobile augmented reality (MAR) system for interactive classroom. It provides a complex visual augmented layer over static slides to understand the concepts more clearly without touching the computer devices or using whiteboard. Simple hand gestures are used to interact with slides while presenting in the classroom or in any conference room with high accuracy and efficiency without any expensive hardware. The gesture path is tracked continuously using a color tracking algorithm proposed. A decision tree is used to make the decisions based on the gestures. The preliminary result indicates that the gesture recognition rate is near about approximately 94% and it is mostly acceptable. This enhances the user's interaction level with immersive feeling in immersive environment.