Hierarchical pruning of lightweight multiscale edge-aware transformer for real-time insulator defect detection model

Hui Chen, Changsheng Zhu, Hongwei Bai, X. H. Feng, Tianyu Li, Lulu Zhu, Jingjie Li, Yongxin WangMeasurement 2025 · DOI

Abstract

Abstract Insulator defect detection is an important part of the normal and stable operation of the power system. To address the problems of poor detection ability of traditional detection methods for small targets, poor detection ability in complex scenes and high computational overhead, we propose hierarchical pruning of lightweight multiscale edge-aware Transformer (HLME-DETR). The backbone network uses lite context-guided dual block to reduce the computational overhead of the model and uses global contextual information to understand the scenario and enhance the model’s detection ability in complex scenarios. In addition, the CCFM structure uses the Dual domain Selection Fusion module for feature splicing, which repairs the loss of features caused by feature splicing by capturing spatial and frequency information. Since most of the insulator defects are small targets, we designed the cross-stage partial multi-scale edge enhancement module, which enhances the detection of small targets by enhancing the edge details of the objects to be detected. The Lightweight Adaptive Extraction module is used for downsampling to increase the sensory field of the model. In addition, to reduce the computational overhead of the Transformer architecture and improve the detection efficiency in real applications, we use layer-adaptive pruning to further compress the model, thus effectively improving the deployment performance and inference speed. The experimental results show that HLME-DETR outperforms other models in terms of detection performance and lightweighting, with the number of parameters and computation reduced by 57.8% and 41.4%, respectively, while <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mrow> <mml:mtext>A</mml:mtext> </mml:mrow> <mml:mrow> <mml:msub> <mml:mrow> <mml:mtext>P</mml:mtext> </mml:mrow> <mml:mrow> <mml:mrow> <mml:mtext>50</mml:mtext> </mml:mrow> </mml:mrow> </mml:msub> </mml:mrow> </mml:mrow> </mml:math> and <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mrow> <mml:mtext>A</mml:mtext> </mml:mrow> <mml:mrow> <mml:msub> <mml:mrow> <mml:mtext>P</mml:mtext> </mml:mrow> <mml:mrow> <mml:mrow> <mml:mtext>50:95</mml:mtext> </mml:mrow> </mml:mrow> </mml:msub> </mml:mrow> </mml:mrow> </mml:math> rise by 1.9% and 1.2%, respectively, compared with the baseline model. Especially in small target detection, <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mrow> <mml:mrow> <mml:mtext>A</mml:mtext> </mml:mrow> <mml:mrow> <mml:msub> <mml:mrow> <mml:mtext>P</mml:mtext> </mml:mrow> <mml:mi>S</mml:mi> </mml:msub> </mml:mrow> </mml:mrow> </mml:math> rises 5.6% compared with the baseline model, which fully proves the effectiveness of the method in the practical application of insulator defect detection.
BibTeX
@inproceedings{chen2025hierarchical,
  author = {Hui Chen and Changsheng Zhu and Hongwei Bai and X. H. Feng and Tianyu Li and Lulu Zhu and Jingjie Li and Yongxin Wang},
  title = {{Hierarchical pruning of lightweight multiscale edge-aware transformer for real-time insulator defect detection model}},
  booktitle = {Measurement},
  year = {2025},
  doi = {10.1088/1361-6501/ae2d80},
}

Loading related content...

Figure