In industrial scenarios, bearing fault diagnosis often suffers from data scarcity and class imbalance, which significantly hinders the generalization performance of data-driven models. While generative adversarial networks (GANs) have shown promise in data augmentation, their efficacy deteriorates in the presence of multi-category and structurally complex fault distributions. To address these challenges, this paper proposes a novel fault diagnosis framework based on a Multi-Domain Feature Transformer GAN (MDFT-GAN). Specifically, raw vibration signals are transformed into 2D RGB representations via joint time-domain, frequency-domain, and time–frequency-domain mappings, effectively encoding multi-perspective fault signatures. A Transformer-based feature extractor, integrated with Efficient Channel Attention (ECA), is embedded into both the generator and discriminator to capture global dependencies and channel-wise interactions, thereby enhancing the representation quality of synthetic samples. Furthermore, a gradient penalty (GP) term is introduced to stabilize adversarial training and suppress mode collapse. To improve classification performance, an Enhanced Hybrid Visual Transformer (EH-ViT) is constructed by coupling a lightweight convolutional stem with a ViT encoder, enabling robust and discriminative fault identification. Beyond performance metrics, this work also incorporates a Grad-CAM-based interpretability scheme to visualize hierarchical feature activation patterns within the discriminator, providing transparent insight into the model’s decision-making rationale across different fault types. Extensive experiments on the CWRU and Jiangnan University (JNU) bearing datasets validate that the proposed method achieves superior diagnostic accuracy, robustness under limited and imbalanced conditions, and enhanced interpretability compared to existing state-of-the-art approaches.

Haze is a natural phenomenon that can cause discomfort to the visual system. Since the essential information of hazy images, i.e., the channel features of the images, has been neglected in most previous studies, it leads to the degradation of the generalization ability of the model. In this paper, we propose a multi-scale image dehazing neural network model based on the channel attention mechanism. In this paper, this model consists mainly of a network structure consisting of multi-scale encoding-decoding module, channel attention mechanism module, and multi-scale residual module. We demonstrate the robustness of the proposed dehazing network model in this paper by conducting qualitative as well as quantitative analysis on synthetic datasets and real-world datasets.

In order to address the problem of defogging images of foggy scenes, this paper proposes a parameter-tunable image defogging algorithm based on the FC-MSPCNN model. Firstly, we simplify the traditional PCNN model and also propose the Limited-FCMSPCNN (LFC-MSPCNN) model. Secondly, we give the five key parameters in the model. There are five key parameters in the model: α, β, V, G and a modulation parameter q, and how they are set. Thirdly, we verified that the algorithm in this paper has good defogging effect on foggy sky images by comparing with the Dark Channel Prior algorithm, the Retinex algorithm and the FC-MSPCNN algorithm, showing that the LFC-MSPCNN method proposed in this paper has robustness.

Bearing fault diagnosis in complex industrial settings presents significant challenges due to weak signal characteristics, multi-modal semantic misalignment, and limited interpretability. To address these issues, this paper proposes a Bio-Inspired Multi-Modal Graph Attention Network (BI-MMGAN), a novel intelligent diagnostic framework that integrates Pulse-Coupled Neural Networks (PCNN) with multi-modal contrastive learning. The framework incorporates three key innovations: (1) A Fault-Aware PCNN (FA-PCNN) spectral enhancement system: Unlike traditional denoising methods, this bio-inspired module leverages pulse synchronization dynamics to actively amplify weak transient fault features while preserving spectral integrity. (2) A Contrastive Graph Attention mechanism: This transforms heterogeneous time-frequency representations (STFT, CWT, MEL) into structured graph formats, capturing complex spectral topologies and dependencies beyond the reach of conventional convolutions. (3) An Adaptive Cross-Modal Fusion strategy: By employing attention-weighted integration, this strategy aligns semantic features across multi-source spectra to resolve scale and resolution discrepancies. Extensive experiments on three benchmark datasets (CWRU, Paderborn, and University of Ottawa) demonstrate that BI-MMGAN significantly outperforms state-of-the-art approaches. Notably, under harsh noise conditions (-6 dB), the proposed method achieves a diagnostic accuracy of 91.45%, surpassing the second-best baseline by 4.20%. Interpretability analysis further elucidates the decision-making process through attention visualization and PCNN firing rate evaluation. The code is available at https://github.com/chenxiG/BI-MMGAN.

Image enhancement has been a hot research topic in image processing in recent years, and Pulse-Coupled Neural Network (PCNN) based image processing has become one of the important channels for image processing. Recently, we have applied this network to image enhancement based on Fire-controlled Pulse-coupled NeuralNetwork (FC-MSPCNN) without changing its model structure by combining the adaptive parameter setting method to redefine the link coefficients β. On the one hand, and reprocessing the mapping matrix of the extracted three channels on the other hand. The experimental results of the improved FC-MSPCNN are proved to be more effective and robust compared with the traditional classical algorithm, both in terms of competent analysis and objective comparison.