Abstract With the rapid advancement of deep neural networks in wireless communications, applications such as signal modulation recognition and target detection face threats from adversarial example attacks. To enhance system robustness against adversarial attacks, adversarial example detection holds a unique position and role as a complementary approach to conventional adversarial defense methods. This paper investigates the spatial and frequency domain attribute differences between clean and adversarial signal examples, proposing a joint spatial-frequency domain adversarial example detection method for signal modulation recognition networks. In the frequency domain, we extract time-shifted autocorrelation features that capture the peak width differences between clean and adversarial examples, where adversarial perturbations exhibit wider autocorrelation peaks due to their signal-like energy distribution. In the spatial domain, we characterize the inter-layer feature propagation patterns through DNN layers by computing cosine similarities between layer-wise activations and class centers, revealing that adversarial examples exhibit progressive deviation from their true class in deeper layers. These complementary dual-domain features are then fused and classified through a Random Forest ensemble to achieve robust adversarial detection. Experimental results show that the proposed method achieves an adversarial detection rate of 90.32% with an AUC of 0.9475 under PGD attacks, substantially outperforming autoencoder-based and KL-divergence-based baseline detectors by 22.20% and 4.36% respectively. The detector also maintains robust performance across different attack types, achieving detection rates of 98.82% against FGSM and 99.36% against CW attacks. These results validate that the proposed method serves as an effective frontline defense to enhance the adversarial robustness of signal modulation recognition networks.

The printed circuit board (PCB) stands as the cornerstone of electronic equipment, with its schematic holding paramount importance for system performance and reliability. In light of the pervasive use of electronic devices in society, concerns regarding maintenance, safety, backdoors, and other latent issues have garnered significant attention. Automatic schematic generation (ASG), with its distinct capability for generating circuit schematics autonomously, not only plays a pivotal role in electronic design automation (EDA) but also aids in deciphering the fundamental principles of PCB equipment to effectively address these underlying issues. However, constrained by the increasingly sophisticated manufacturing processes of PCBs and the inherent legal and ethical controversies surrounding reverse engineering, the development of related technologies faces notable bottlenecks. To break through technical barriers and advance technological progress, this paper comprehensively combs through the existing ASG, offers in-depth description of the core algorithms of the technology—layout and routing, and for the application of the technology in PCB reverse engineering, analyzes in detail the current challenges and the faced problems. Around these challenges, feasible solutions are discussed in this paper, with the aims of promoting the research of automatic PCB schematic generation technology and contributing new strength to EDA and PCB reverse engineering automation.

• The first work to apply eye-movement attention mechanisms in the task of image cross-view matching. • Gaze information from the human brain is utilized to guide model training and learning; they are not required during testing. • The fusion of the eye-movement and self-attention mechanisms using sub-image block-level can significantly guide model to focus more on the key image regions. • The implementation of image adaptive cropping and resolution enhancement strategies effectively removes the interference of redundant information and reduces the waste of computational resources. The goal of cross-view matching between street and aerial images is to retrieve aerial-view images that correspond to a given street-view image from a database of GPS-tagged aerial images. This task relies on image cross-view matching technology, focusing on the extraction and alignment of features representing the same location in both image types. The significant differences in perspective and appearance between street-view images and aerial-view images present a challenge. Aerial-view images cover a broader area, while street-view images focus on specific locations, creating information asymmetry that complicates the matching process. To tackle these challenges, this paper proposes a gaze-guided hybrid attention Vision Transformer, which uses gaze information to guide the model to focus on and align task-related features. Furthermore, inspired by the human visual cognitive process of "focus and zoom," we develop a hybrid attention module alongside an image adaptive cropping and resolution enhancement module. The hybrid attention module utilizes gaze information to guide the model to focus on relevant regions, while the image adaptive cropping strategy uses gaze information to guide the model to eliminate irrelevant regions. Techniques for improving image resolution allow for the magnification of important regions, thereby aiding the extraction of fine-grained features. We evaluate the model's performance on benchmark datasets and conduct ablation study experiments to assess the contributions of each module. Experimental results show that the method achieves a top-1 accuracy of 75.56% on the CVACT dataset, representing state-of-the-art performance. This study provides valuable insights into incorporating human experience into computational models, particularly through gaze-guided learning of visual task networks to enhance model performance.

Dense convolutional network has been continuously refined to adopt a highly efficient and compact architecture, owing to its lightweight and efficient structure. However, as the current dense-like architectures are mainly designed manually, it becomes increasingly difficult to adjust the channels and reuse level based on past experience. As such, we propose an architecture search method called dense optimizer that can search high-performance dense-like network automatically. In dense optimizer, we view the dense network as a hierarchical information system, maximizing the network's information entropy while constraining the effectiveness and the distribution of the entropy across each stage via a power law, thereby constructing an optimization problem. We also propose a branch-and-bound optimization algorithm that tightly integrates power-law principle with search space scaling to solve the optimization problem efficiently. The superiority of dense optimizer has been validated on different computer vision benchmark datasets. Our searched model DenseNet-OPT achieved a top-1 accuracy of 84.3% on CIFAR-100, which is 5.97% higher than the original one. Specifically, dense optimizer achieves high-quality search results while only requiring 4 h of computation time on a single CPU.

In Graph Convolutional Networks (GCNs), the aggregation of node features in graph convolutional learning is typically guided solely by the topology of the graphs. However, both network topology and node features provide unique and valuable information. Relying solely on topology cannot yield entirely accurate and comprehensive neighborhood information. This paper proposes Output Correction for GCNs (OC-GCN), which advocates for representation learning incorporating feature and topology views. Specifically, we employ two GCN encoders to extract node embeddings in both the feature and topology spaces. We first identify consistent and inconsistent nodes by comparing the pseudo-labels generated by the encoders. Subsequently, we regenerate representations of inconsistent nodes by aggregating the representations of consistent nodes within their respective neighborhoods. Our experiments demonstrate that OC-GCN significantly enhances the classification accuracy of inconsistent nodes. We conducted extensive experiments on benchmark datasets and observed that OC-GCN outperforms state-of-the-art baselines across various label rates.

Abstract Single image super‐resolution (SISR) is a promising research direction in computer vision and image processing for improving the visual perception of low‐quality images. In recent years, deep learning algorithms have driven tremendous development in SR, and SR methods based on various network architectures have significantly improved the quality of reconstructed images. Although there has been a large amount of reviews focusing on SISR, few studies have focused specifically on network architectures for SISR. This paper aims to provide a systematic overview of the design ideas of SISR using multiple architectures, including Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), Transformer, and Diffusion model. In addition, an experimental analysis and comparison of state‐of‐the‐art SR algorithms have been performed on publicly available quantitative and qualitative datasets. Finally, some future directions are discussed that may help other community researchers.

PCB reverse engineering plays a crucial role in verifying circuit design, detecting hardware Trojans, and maintaining outdated devices. The reverse generation of PCB schematics, a vital aspect of this engineering, heavily relies on manual design due to the challenge of objectively evaluating schematic quality. This paper introduces a novel aesthetic evaluation metric to assess the quality of PCB schematics. Based on this metric, a PCB schematic reverse generation method using reinforcement learning is proposed. Experimental results demonstrate the metric and method’s reliability and effectiveness, as they can automatically generate PCB schematics comparable to those designed by human engineers.

Abstract Transferability of adversarial samples under different convolutional neural network (CNN) models is one of the metrics indicators to assess the efficiency of adversarial examples and an important research direction in defense of that. Transferability isolate models employ a particular alternative model to avoid black-box attacks. Meanwhile, recent research has revealed that adversarial transferability across sub-models may be utilized to express the diversity requirements of sub-models under ensemble robustness abstractly. Due to the lack of mathematical description for this adversarial transferability, it was utilized to be abstractly described as the diversity of different hypotheses. This paper employs the Jacobians matrix’s singular value decomposition (SVD) to provide a more accurate mathematical description of the transferability of adversarial samples between models and proposes a corresponding evaluation metric. Based on this metric, a new regularization constraint is introduced into the ensemble training, and the adversarial transferability between the sub-models is isolated optimally without the prior information of the adversarial samples. Based on the proposed metric accurately defining the transferability, further ensemble robustness experiments under small-scale dataset disentangle the correlation between transferability and robustness, indicating that the transferability isolation can only achieve robustness under an alternative transfer-based attack with partial sub-models of the ensemble.

Abstract Element detection is a key step in non‐destructive testing of printed circuit board (PCB) based on computed tomography (CT). In recent years, some image segmentation methods based on deep learning have shown great potential in the element segmentation task of PCB CT images, and greatly improved the efficiency and accuracy. However, since image segmentation is based on pixel‐level classification, the annotation of training data is difficult and costly. Aiming at this problem, the authors proposed a new active learning method based on the integration of relevant information about segmentation tasks and data variance. In this method, the Result Quality Evaluation Module (RQEM) proposed by us is used to generate task‐related information, and an adversarial network is used to generate the difference information between samples and the initial labelled data. Finally, the two parts of information are fused and used as the standard of data selecting. In the PCB CT image element segmentation task, the authors only need to select 12.7% of the whole training set with their proposed method to make Mean Intersection Over Union (MIOU) reach 79.7, which has reached 95% of the optimal performance of 83.7. The in‐depth analysis also verifies the effectiveness and stability of the authors’ method.

In image classification of deep learning, adversarial examples where input is intended to add small magnitude perturbations may mislead deep neural networks (DNNs) to incorrect results, which means DNNs are vulnerable to them. Different attack and defense strategies have been proposed to better research the mechanism of deep learning. However, those researches in these networks are only for one aspect, either an attack or a defense. There is in the improvement of offensive and defensive performance, and it is difficult to promote each other in the same framework. In this paper, we propose Cycle-Consistent Adversarial GAN (CycleAdvGAN) to generate adversarial examples, which can learn and approximate the distribution of the original instances and adversarial examples, especially promoting attackers and defenders to confront each other and improve their ability. For CycleAdvGAN, once the Generator A and D are trained, GA can generate adversarial perturbations efficiently for any instance, improving the performance of the existing attack methods, and GD can generate recovery adversarial examples to clean instances, defending against existing attack methods. We apply CycleAdvGAN under semiwhite-box and black-box settings on two public datasets MNIST and CIFAR10. Using the extensive experiments, we show that our method has achieved the state-of-the-art adversarial attack method and also has efficiently improved the defense ability, which made the integration of adversarial attack and defense come true. In addition, it has improved the attack effect only trained on the adversarial dataset generated by any kind of adversarial attack.

Printed circuit board wire segmentation based on computed tomography (CT) image can help subsequently locate and estimate inner faults of circuit in an automatic and non‐destructive manner. However, CT imaging is prone to suffer from scattered artefacts, metal artefacts and other interference, destroying compact boundary structures of wires. Wires have the characteristic of dense local distribution, and massive vias, pads, and coppers can appear close to wires, resulting in mazy recognition surroundings. The above‐mentioned problems bring great difficulty for high‐accuracy recognition and location of wire segmentation. In this study, considering that deep convolutional neural network (DCNN) with powerful feature representation can recognise wires in confused surroundings, and graph cut (GC) model relying on grayscale and local texture information specialises in protecting edge structures of wires, the authors propose an effective framework called DCNN‐GC that employs DCNN to obtain global semantic prior to guide the GC model to accomplish satisfactory wire segmentation. The authors qualitative and quantitative results demonstrate outstanding performance, and achieve overwhelming intersection over union compared with traditional and DCNN‐based methods.

Emotion recognition, as an important part of human-computer interaction, has been extensively researched. Various studies have already verified the relationship between emotion and the event-related potentials (ERPs). In this paper, a new methodology for emotion recognition is investigated by detecting single-trial ERPs related to some specific level of emotions. First, a spatial filter is constructed to estimate the ERP components. Then the most discriminative spatial and temporal features of the entire ERP waveform are extracted with linear discriminant analysis. The performance of this method is tested by classifying the emotional valence on three levels, the extremely negative, the moderately negative and the neutral, with the support vector machine (SVM). The result shows that the proposed method is effective.

Image segmentation can be used in non-destructive testing, tracking and recognition. Level set method for image segmentation has poor performance on efficiency. In this paper, we propose to use max-flow algorithm to optimize a locally improved Chan-Vese model for image segmentation in the presence of intensity inhomogeneity. The energy function of local Chan-Vese model is introduced firstly. This model consists of global term, local term and penalty term and the local term contributes the segmentation for images with intensity inhomogeneity. Then, we convert this energy function to the frame of Graph Cut whose energy function can be efficiently minimized by max-flow algorithm. As a result, the process of optimization of local Chan-Vese model can be accelerated by using max-flow algorithm. The experiments demonstrate that the proposed method can achieve satisfactory segmentation for images with intensity inhomogeneity as well as very high efficiency.

This study proposes an improved level set method for segmenting images with intensity inhomogeneity. One of the improvements is to consider the difference between an original image and an estimated image without bias field in the image model. Apart from using this difference, Gaussian distribution with means and variance is utilised as the local intensity descriptor to map the original image into another domain so the object and the background can be better separated in the transformed domain. Then, an improved level set energy function that combines the image term, local variance, and the above difference is defined. The minimisation of the function can be processed by level set evolution. The proposed method is compared with existing methods, and experiments on both synthetic and real images demonstrate that authors’ method has superior performance.

Parallel computing is applied in computed tomography to shorten the imaging time and achieves outstanding effects. But image reconstruction algorithm remains to be significantly time-consuming when faced with the large data sets which decreases the efficiency of imaging process. Single graphic card cannot solve this problem because of the limitation on video memory. This paper presents a parallel method for FDK algorithm using multi-GPU based cluster system. In the method, the cluster system is divided into three levels: nodes, GPUs and single GPU. Corresponding parallel strategies are designed for different levels according to the algorithm characteristics and hardware structures. The experiments show that the multi-GPU based cluster system is able to achieve the same precision with the single node meanwhile it will gain higher speedup ratio with the increasing number of nodes or GPUs. The computing time of whole reconstruction reduces to almost half of the origin by doubling the computing devices.