电力施工场景下基于知识蒸馏的安全帽佩戴检测算法研究

贾海江; 黄晓宏; 熊敏; 段丹; 王璇

欢迎来到能源电力期刊网！

English Version 读者登录期刊登录

您当前的位置：

首页 >

文章列表页 >

电力施工场景下基于知识蒸馏的安全帽佩戴检测算法研究

更新时间：2026-01-12

- 电力施工场景下基于知识蒸馏的安全帽佩戴检测算法研究
- Research on lightweight safety helmet detection algorithm based on knowledge distillation
- 电力大数据 2025年第10期
- 作者机构：
  
  中国电建集团江西省电力建设有限公司西安工程大学计算机科学学院
- 作者简介：
- 基金信息：
- DOI：
  中图分类号：
- 纸质出版：2025
- 稿件说明：
移动端阅览
贾海江, 黄晓宏, 熊敏, 等. 电力施工场景下基于知识蒸馏的安全帽佩戴检测算法研究[J]. 电力大数据, 2025,(10).

贾海江, 黄晓宏, 熊敏, et al. Research on lightweight safety helmet detection algorithm based on knowledge distillation[J]. 2025, (10).
贾海江, 黄晓宏, 熊敏, 等. 电力施工场景下基于知识蒸馏的安全帽佩戴检测算法研究[J]. 电力大数据, 2025,(10). DOI：

贾海江, 黄晓宏, 熊敏, et al. Research on lightweight safety helmet detection algorithm based on knowledge distillation[J]. 2025, (10). DOI：

摘要

针对复杂电力施工场景下安全帽检测模型在边缘设备部署时难以同时满足高精度和高速度需求的问题，本文提出了一种轻量化模型SDS-YOLOv8-tiny。首先利用LAMP剪枝算法移除SDS-YOLOv8模型中的冗余参数来直接大幅降低模型参数量和计算复杂度；再结合提出的逻辑与特征信息双重融合的BCCKD蒸馏策略，强化模型对上下文语义关联的建模能力，帮助学生模型提高特征的表达能力，从而在减少计算复杂度的同时提高模型的检测性能。实验表明，相比于原始的SDS-YOLOv8模型，轻量化后的SDS-YOLOv8-tiny模型在参数量和计算量上分别减少了74.9%和50.4%，模型大小压缩到2.1MB，mAP值为84.8%，轻量化程度明显且性能稳定，为在边缘设备中部署提供了方法。

Abstract

To address the problem that the helmet detection model in complex power construction scenarios is difficult to meet the demand for both high accuracy and high speed when deployed in edge devices

this paper proposes a lightweight model

SDS-YOLOv8-tiny

which firstly utilizes the LAMP pruning algorithm to remove redundant parameters in the SDS-YOLOv8 model to directly and significantly reduce the number of model parameters and computational complexity

and then combines with the proposed The proposed BCCKD distillation strategy with dual fusion of logic and feature information is used to strengthen the model's ability to model contextual semantic associations

which helps the student model to improve the expression of features

thus improving the model's detection performance while reducing the computational complexity. Experiments show that compared with the original SDS-YOLOv8 model

the lightweighted SDS-YOLOv8-tiny model reduces the number of parameters and computation by 74.9% and 50.4%

respectively

and the model size is compressed to 2.1MB with a mAP value of 84.8%

which is a significant degree of lightness and stable performance

and provides a way to deploy in edge devices.

关键词

Keywords

references

Redmon J, Divvala S, Girshick R, et al. You only look once: Unified, real-time object detection[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 779-788.

Wang L, Zhang X, Yang H. Safety helmet wearing detection model based on improved YOLO-M[J]. IEEE Access, 2023, 11: 26247-26257.

周孟然,王皓.基于改进YOLOv7的安全帽佩戴检测算法[J].软件,2024,45(08):14-17.

Lin B. Safety helmet detection based on improved YOLOv8[J]. IEEE Access, 2024.

Wang Z, Zhang J, Zhao Z, et al. Efficient yolo: A lightweight model for embedded deep learning object detection[C]//2020 IEEE International Conference on Multimedia & Expo Workshops (ICMEW). IEEE, 2020: 1-6.

Wan F, Sun C, He H, et al. YOLO-LRDD: A lightweight method for road damage detection based on improved YOLOv5s[J]. EURASIP Journal on Advances in Signal Processing, 2022, 2022(1): 98.

Zeng T, Li S, Song Q, et al. Lightweight tomato real-time detection method based on improved YOLO and mobile deployment[J]. Computers and electronics in agriculture, 2023, 205: 107625.

高东,刘丽娟.基于轻量化YOLOv5s的安全帽佩戴检测算法[J].电视技术,2024,48(06):88-94+98.

汝洪芳,梁一乐,王国新.基于剪枝算法改进YOLOv5的煤矿井下安全帽检测方法[J].黑龙江科技大学学报,2024,34(03):452-456+468.

Lee J, Park S, Mo S, et al. Layer-adaptive sparsity for the magnitude-based pruning. arxiv 2020[J]. arxiv preprint arxiv: 2010.07611.

Yang L, Zhou X, Li X, et al. Bridging cross-task protocol inconsistency for distillation in dense object detection[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2023: 17175-17184.

Shu C, Liu Y, Gao J, et al. Channel-wise knowledge distillation for dense prediction[C]//Proceedings of the IEEE/CVF international conference on computer vision. 2021: 5311-5320.

Liu Z, Li J, Shen Z, et al. Learning efficient convolutional networks through network slimming[C]//Proceedings of the IEEE international conference on computer vision. 2017: 2736-2744.

Yang C, Yang Z, Khattak A M, et al. Structured pruning of convolutional neural networks via l1 regularization[J]. IEEE Access, 2019, 7: 106385-106394.

Liu S, Chen T, Chen X, et al. The unreasonable effectiveness of random pruning: Return of the most naive baseline for sparse training[J]. arxiv preprint arxiv:2202.02643, 2022.