Zhenbing Zhao

Transmission tower re-identification (ReID) supports the removal of redundant unmanned aerial vehicle (UAV) inspection images but remains challenging because towers have similar structures, viewpoints vary significantly, and backgrounds are complex. This paper proposes a two-stage framework combining YOLO11n-Seg foreground segmentation with a Contrastive Language–Image Pre-training (CLIP)-based ReID model using a ResNet-50 visual backbone enhanced by foreground-aware cropping (FG-CROP), a depthwise-separable Convolutional Block Attention Module (DS-CBAM), and cross-scale gated fusion (CSGF). YOLO11n-Seg was trained on an independently collected dataset with no image or tower-identity overlap with the re-identification data, and the selected validation checkpoint was used to process all 1772 ReID images. The ReID dataset contains 1437 training images from 464 identities, while the test set contains 106 query images and 229 gallery images from 106 disjoint identities. Across five random seeds, the complete framework achieved 62.34 ± 0.87% mAP and 60.20 ± 1.42% Rank-1, compared with 56.90 ± 0.89% mAP and 54.52 ± 1.40% Rank-1 for the black-background CLIP-ReID baseline with a ResNet-50 visual backbone. The selected YOLO11n-Seg checkpoint achieved a 97.26% mask mAP50 and 71.61% mask mAP50-95 on the independent validation set.

High voltage equipment, as a core component of power systems, plays an indispensable role in ensuring the reliability of power supply through its safe and stable operation. Traditional visual defect detection for high voltage equipment, however, often relies on manual inspection and experience‐based judgement, which struggles to meet the growing monitoring demands. Consequently, the introduction of more intelligent and precise large‐model technologies into the domain of high voltage equipment visual defect detection information processing is urgently required. The large‐scale foundational visual models, with their robust data processing capabilities, complex pattern recognition and reasoning abilities, are progressively penetrating various industries, serving as a key driver for industrial upgrades and transformation. In the field of high voltage equipment, the application prospects of foundational visual models are particularly promising. In this paper, we provide a comprehensive review of the research progress on visual large models in the specific context of high voltage equipment. It summarises a concise overview of the engineering challenges that foundational visual models need to tackle. Furthermore, based on research achievements from the era of smaller models, we thoroughly examine the application potential of large‐model technologies in the high voltage equipment domain. Additionally, we also identify and scrutinise the scientific issues associated with visual large models. At present, research on vision foundation models for high voltage equipment is still in its early stages both domestically and internationally. Significant gaps remain in areas such as dataset construction, pretraining and fine‐tuning, all of which present promising directions with immense application potential. The primary objective of this paper is to offer a comprehensive and systematic review of large pretrained models for high voltage equipment defect detection methods, providing valuable reference for researchers exploring this field.

Transmission and distribution networks are evolving into distributed smart grids, making accurate defect localization of line components increasingly important for power inspection. However, traditional manual inspection is inefficient and unreliable in complex scenarios, while most existing deep learning methods rely only on two-dimensional visible-light images and struggle to capture spatial structure and occlusion relationships. To address these limitations, this paper proposes a defect detection network that integrates monocular relative-depth information with adaptive progressive feature fusion. The framework adopts a dual-branch architecture, where visible-light images provide appearance information and relative-depth maps generated from the corresponding RGB frames using Depth Anything V2 encode auxiliary spatial relationships. Because both modalities originate from the same image frame, no additional depth sensor or cross-sensor temporal synchronization is required. An adaptive progressive fusion module performs coarse-to-fine cross-modal interaction, and a coordinate attention mechanism is introduced to enhance positional encoding and suppress background interference. Experiments on a self-constructed dataset of 9838 RGB images paired with estimated relative-depth maps covering five typical defect categories show that the proposed method achieves 92.4% mAP@50 and 67.1% mAP@50–95. The results demonstrate its effectiveness in improving defect localization, small-object detection, and robustness across the evaluated transmission and distribution line scenes.

Background: Rotenone is a widely used environmental pesticide, and epidemiological studies suggest that exposure is associated with an increased risk of Parkinson’s disease (PD); however, the molecular toxicological basis of this association remains incompletely defined. Ferroptosis is an iron-dependent, lipid peroxidation-driven form of regulated cell death that is relevant to PD and other neurodegenerative disorders. In this study, we provide disease-contextual functional evidence linking ferroptosis to rotenone-induced PD-like neurotoxicity. Methods: We combined network toxicology, human PD substantia nigra transcriptomic analysis using GSE7621, and SH-SY5Y cell-based validation. Rotenone-associated targets were predicted and analyzed for ferroptosis-related enrichment, PD transcriptomic signatures were used for disease-contextual candidate prioritization, and selected findings were validated using qPCR, CCK-8, Western blotting, C11-BODIPY lipid peroxidation staining, and transmission electron microscopy. Results: By further integrating a human PD substantia nigra transcriptomic dataset (GSE7621), we prioritized an 11-gene, PD-contextualized ferroptosis-associated candidate module (LIPF, FAM170A, MCHR1, IL17A, MYB, GFAP, ARMC3, GKN1, GATA3, IL17F, and TEKT1). In SH-SY5Y cells, rotenone exposure consistently upregulated this candidate transcriptional module, and this induction was broadly attenuated by the ferroptosis inhibitor ferrostatin-1 (Fer-1). In parallel, orthogonal functional assays supported an iron- and lipid peroxidation-driven injury state under rotenone exposure that was suppressible by ferroptosis inhibition and iron chelation. Finally, we further performed an exploratory drug–gene association screen to prioritize clinically available candidates, and a limited qPCR check suggested that several selected compounds partially attenuated representative hub-gene induction under rotenone exposure. Conclusions: Collectively, these findings provide disease-contextual and experimentally supported evidence linking rotenone exposure to ferroptosis-associated neurotoxicity, and identify a ferroptosis-responsive transcriptional module for future hypothesis-driven mechanistic investigation.

With the rapid development of high-resolution aerial photography facilitated by unmanned aerial vehicles (UAVs), identifying transformer bushings defects from aerial images has become a pivotal technique for the intelligent inspection of power distribution lines. However, this task currently faces challenges such as the high complexity of detection backgrounds, the loss of small-scale target features, and the insufficient incorporation of appearance knowledge during feature fusion. To address these issues, this paper proposes an intelligent detection method based on appearance knowledge, termed AK-RT-DETR. Specifically, an Appearance-Knowledge Feature Extraction Network (AK-FEN) is constructed to adaptively capture detailed characteristics of transformer bushings and insulating caps, enabling a hierarchical representation of appearance knowledge, while a Cascaded Group Attention (CGA) module is introduced to enhance the network’s ability to extract fine-grained features. Additionally, a FS-IoU loss function is designed to improve the precision of bounding-box regression. Experimental results demonstrate that AK-RT-DETR achieves a 3.6\% improvement in mean average precision (mAP) compared with the baseline model, providing a new approach for improving the accuracy of transformer bushings defect detection in power distribution lines.

Bolt defect detection (BDD) is critical to ensure the safety of transmission lines. However, the scarcity of defect images and imbalanced data distributions significantly limit detection performance. To address this issue, we propose a segmentation-driven bolt defect editing (SBDE) method for defect generation and data augmentation. The proposed framework comprises three main components. First, a bolt attribute segmentation (BAS) model is constructed, which integrates contrast-based enhancement and frequency-domain filtering to improve edge feature representation. A multipart-aware optimization strategy is introduced to handle the noncontiguous structures of bolt attributes, enabling the generation of high-quality structural masks for subsequent editing. Second, a bolt attribute editing module based on image inpainting is designed, which removes specific bolt attributes by applying mask boundary optimization and context-aware reconstruction, thereby transforming normal bolts into defective ones. Finally, an editing recovery augmentation strategy is proposed to restore the edited bolts into the original inspection images, constructing complete defect samples for expanding the detection dataset. Extensive experiments were conducted on a self-built bolt dataset and the public MVTec AD dataset. The results show that the proposed method significantly outperforms existing state-of-the-art image editing approaches in generation quality and effectively improves defect classification and detection performance across multiple mainstream models, thereby validating the effectiveness and practical potential of the proposed approach. The code of the project is available at https://github.com/Jay-xyj/SBDE