Foreign object detection (FOD) in railway catenary systems is crucial for ensuring operational safety and preventing catastrophic failures. However, current detection frameworks encounter two significant challenges. First, the infrequency of fault events leads to severe data scarcity, hampering the training and validation of robust detection models. Second, although lightweight networks (e.g., MobileNet, YOLO) achieve compactness by compressing channel factors, they struggle to balance local feature extraction with global dependency modeling. To address these challenges, we propose a solution that includes the following: 1) RailFOD23, a publicly available dataset created using generative AI to mitigate data scarcity; and 2) EPRepSADet, a compact detection framework that utilizes a re-parameterizable bottleneck (Re-bottleneck) and lightweight self-attention (LSA) module for efficient FOD. The Re-bottleneck consolidates multi-branch structures into a single-path representation, whereas LSA facilitates element-wise attention modeling to effectively reduce computational complexity. In addition, the efficient detection head further minimizes model complexity through hierarchical semantic modeling. Extensive experiments demonstrate that EPRepSADet achieves a mean Average Precision (mAP) of 92.5% on the RailFOD23 test set, requiring only 1.7G FLOPs, thus outperforming several state-of-the-art baseline models.

The computational power required for training deep learning models has been skyrocketing in the past decade as they scale with big data, and has become a very expensive and scarce resource. Therefore, distributed training, which can leverage distributed available computational power, is vital for efficient large-scale model training. However, most previous distributed training frameworks like DDP and DeepSpeed are primarily designed for co-located clusters under homogeneous computing and communication conditions, and hence cannot account for geo-distributed clusters with both computing and communication heterogeneity. To address this challenge, we develop a new data parallel based distributed training framework called Co-Adaptive Data Parallelism (C-ADP). First, we consider a data owner and parameter server that distributes data to and coordinates the collaborative learning across all the computing devices. We employ local training and delayed parameter synchronization to reduce communication costs. Second, we formulate a data parallel scheduling optimization problem to minimize the training time by optimizing data distribution. Third, we devise an efficient algorithm to solve this scheduling problem, and formally prove that the obtained solution is optimal in the asymptotic sense. Experiments on the ImageNet100 dataset demonstrate that C-ADP achieves fast convergence in heterogeneous distributed training environments. Compared to Distributed Data Parallel (DDP) and DeepSpeed, C-ADP achieves 21.6 times and 26.3 times improvements in FLOPS, respectively, and a reduction in training time of about 72% and 47%, respectively.

3D perception in train operating environments presents significant challenges, as it must ensure both precise distance estimation and computational efficiency to meet stringent braking requirements. To date, existing 3D detection architectures, which employ dense voxel or pillar representations, encounter challenges of computational inefficiency and accuracy degradation when processing large-scale railway Light Detection And Ranging (LiDAR) data. To address this challenge, we propose RailVoxelDet, a railway-optimized 3D detector integrating the Multi-factor Dynamic Voxel Feature Encoder (MDVFE) and efficient backbone. Specifically, MDVFE converts point clouds to 2D sparse voxels, reducing computational complexity. The backbone employs residual bottlenecks with shared full connected layers and sparse convolutions, enhanced by the SimAM-Point module. Additionally, the Feature Query and Matching Module (FQMM) is proposed to establish a bottom-up multi-level feature fusion architecture. Experimental results show RailVoxelDet reaches 71.29% mAP on OSDaR23 and 61.94% mAP on AirR24, with 6.42G FLOPs and a 71.42ms inference time. It outperforms 12 comparison models, delivering state-of-the-art results.

System vulnerability can be exposed through adversarial attack, facilitating the improvement of system robustness. However, existing research has predominantly focused on white-box attacks, which typically require access to system parameter information, thus limiting the attack conditions. In contrast, research on black-box attacks is relatively scarce, and there is a lack of innovative adversarial attack methods specifically tailored for structured data. Therefore, in conjunction with Explainable AI, we propose a novel simple, fast and efficient black-box technique to generate adversarial example, focusing on the attack perspective. The proposed method constructs new traffic data by performing a crossover operation between the Non-Saliency portion of the original traffic and the corresponding portion of the candidate traffic. A combination that successfully alters the prediction results of the original sample is obtained through a search strategy based on random sampling and iterative update, and the initial adversarial example is generated. To make the number of modified features and the magnitude of perturbations minimized, the integration feature selection and perturbation control strategies were performed, generating the final adversarial example. In order to verify the performance of the method in evading IDS, we compared it against six state-of-the-art attack algorithms. The experimental results show that the proposed method outperforms the compared black-box methods in terms of Attack Success Rate (ASR) and query count on the three datasets, and in most cases, it even surpasses the compared white-box attack methods. Specifically, against the NSL-KDD dataset, when the target model is a DNN, the ASR increases by 29%-33%, and the query count is reduced by 23-33 times compared to the baseline black-box methods; the average number of modified features decreases by 90% compared to the white-box methods.

The number and complexity of network attacks and intrusion events are constantly increasing, timely detection of abnormal intrusion behavior is an important challenge in the field of network security. To this end, this article proposes an ensembled intrusion detection model based on heterogeneous data augmentation and mutual knowledge distillation (KD), KDEHDA. To enhance the diversity of data, the training set is divided into multiple data subsets using the bootstrap sampling method, and different data augmentation methods are used on each subset to obtain multiple balanced data subsets. To enhance the generalization ability of the model, different base CNN models are trained using balanced data subsets, and KD is used to transfer the weight information of CNN. The final result is the voting ensemble of each CNN model. Experimental results show that the proposed model is superior to the existing optimal model.

Abstract Due to the large fluctuations of the boundaries and internal variations of the lesion regions in medical image segmentation, current methods may have difficulty capturing sufficient global contexts effectively to deal with these inherent challenges, which may lead to a problem of segmented discrete masks undermining the performance of segmentation. Although self‐attention can be implemented to capture long‐distance dependencies between pixels, it has the disadvantage of computational complexity and the global contexts extracted by self‐attention are still insufficient. To this end, the authors propose the GFRNet, which resorts to the idea of low‐rank matrix factorization by forming global contexts locally to obtain global contexts that are totally different from contexts extracted by self‐attention. The authors effectively integrate the different global contexts extract by self‐attention and low‐rank matrix factorization to extract versatile global contexts. Also, to recover the spatial contexts lost during the matrix factorization process and enhance boundary contexts, the authors propose the Modified Matrix Decomposition module which employ depth‐wise separable convolution and spatial augmentation in the low‐rank matrix factorization process. Comprehensive experiments are performed on four benchmark datasets showing that GFRNet performs better than the relevant CNN and transformer‐based recipes.

Abstract Colonoscopy is a common method for the early detection of colorectal cancer (CRC). The segmentation of colonoscopy imagery is valuable for examining the lesion. However, as colonic polyps have various sizes and shapes, and their morphological characteristics are similar to those of mucosa, it is difficult to segment them accurately. To address this, a novel neural network architecture called CrossFormer is proposed. CrossFormer combines cross‐attention and multi‐scale methods, which can achieve high‐precision automatic segmentation of the polyps. A multi‐scale cross‐attention module is proposed to enhance the ability to extract context information and learn different features. In addition, a novel channel enhancement module is used to focus on the useful channel information. The model is trained and tested on the Kvasir and CVC‐ClinicDB datasets. Experimental results show that the proposed model outperforms most existing polyps segmentation methods.

Abstract Aiming at the problems of high cost and low accuracy of scene details during the depth map generation in 3D reconstruction, we propose an unsupervised monocular image depth prediction algorithm based on Fourier domain analysis. Generally speaking, small‐scale images can better display depth details, while large‐scale images can more reliably display the depth distribution value of the entire image. In order to take advantage of these complementary properties, we crop the input image with different cropped image ratios to generate multiple disparity map candidates, and then use Fourier frequency domain analysis algorithms to fuse disparity mapping candidates into left and right disparity maps. At the same time, we propose a loss function based on MSSIM to compensate the difference between left and right views and realize unsupervised monocular image depth prediction model training. Experimental results show that our method has good performance on the KITTI dataset.

Jigsaw puzzle algorithm is important as it can be applied to many areas such as biology, image editing, archaeology and incomplete crime-scene reconstruction. But, still, some problems exist in the process of practical application, for example, when there are a large number of similar objects in the puzzle fragments, the error rate will reach 30%–50%. When some fragments are missing, most algorithms fail to restore the images accurately. When the number of fragments of the jigsaw puzzle is large, efficiency is reduced. During the intelligent puzzle, mainly the Sum of Squared Distance Scoring (SSD), Mahalanobis Gradient Compatibility (MGC) and other metrics are used to calculate the similarity between the fragments. On the basis of these two measures, we put forward some new methods: 1. MGC is one of the most effective measures, but using MGC to reassemble the puzzle can cause an error image every 30 or 50 times, so we combine the Jaccard and MGC metric measure to compute the similarity between the image fragments, and reassemble the puzzle with a greedy algorithm. This algorithm not only reduces the error rate, but can also maintain a high accuracy in the case of a large number of fragments of similar objects. 2. For the lack of fragmentation and low efficiency, this paper uses a new method of SSD measurement and mark matrix, it is general in the sense that it can handle puzzles of unknown size, with fragments of unknown orientation, and even puzzles with missing fragments. The algorithm does not require any preset conditions and is more practical in real life. Finally, experiments show that the algorithm proposed in this paper improves not only the accuracy but also the efficiency of the operation.

Impact crater is the major geomorphic feature on the lunar surface. More than 8.2 million valid data samples were obtained by the Chang'E-1 Laster Altimeter(LAM) and the craters are hidden among them. In order to visualize the topographical structure of these craters with high accuracy and computational stability, a novel scattered data fitting method was proposed in this paper. Based on the characteristics analyses of the data, an new class of orthogonal spline functions named as GF-system was used in our method. To verify the effectiveness of the proposed fitting algorithm, a typical surface models was chosen to be approximated by a series of random sampling data sets. Finally, we provide numerical results for two typical lunar crater data sets, demonstrating that our algorithm works efficiently and accurate for large data sets with highly non-uniform sampling densities.

To provide immersive freehand painting experience, we proposed a flexible airbrush model making use of the hands tracking capability of Leap Motion Controller. The airbrush model uses a common screen as the painting canvas. When the user moves hands over the screen, the brush model continually acquires his/her hands movement data and extracts multiple control signals which describes multiple gestures. The virtual airbrush moves along with the user's hands movement as if it is fixed on his/her finger, and its properties change with gestures' change. When the virtual airbrush intersects with the screen, it continually exerts paints onto the screen. User test shows that the user can easily create multifarious brush stroke effects by directly operating over the screen.