矿区输电线路螺母和U型挂环的轻量化锈蚀检测算法  

Lightweight rust detection algorithm for nuts and U-shaped hanging rings of transmission lines in mining areas

作　　者：徐志宗 梁鹏举 王新良 Xu Zhizong;Liang Pengju;Wang Xinliang(School of Physics&Electronic Information Engineering,Henan polytechnic University,Jiaozuo 454000,China;Hebi Coal Industry(group)Co.,Ltd.,Hebi 458000,China)

机构地区：[1]河南理工大学物理与电子信息学院,河南焦作454000 [2]鹤壁煤业(集团)有限责任公司,河南鹤壁458000

出　　处：《能源与环保》2025年第2期182-190,202,共10页CHINA ENERGY AND ENVIRONMENTAL PROTECTION

基　　金：河南省高等学校青年骨干教师培养计划(2019GGJS060);河南省高等学校重点科研基金项目(21B413005);河南省科技攻关项目(172102210274);国家自然科学基金项目(62101176)。

摘　　要：为了解决矿区输电线路螺母和U型挂环锈蚀检测任务中,小目标多、背景复杂、检测精度低的问题,提出了一种矿区输电线路螺母和U型挂环的轻量化锈蚀检测算法。首先,引入部分卷积PConv与ELAN结构相结合构建ELAN-P模块,轻量化主干网络,增加模型检测精度;其次,利用GhostConv对MP和SPPCSPC结构进行轻量化,降低模型参数量;然后,使用Slim-neck结构,轻量化特征融合网络,减少模型参数量,提高模型检测速率;最后,采用NAM注意力机制(Normalization-based Attention Module)增强网络对锈蚀目标的识别能力,提高模型检测精度。实验结果表明,改进后网络的平均检测精度均值(mean Average Precision,mAP)从88.0%提升到89.5%,同时模型的参数量从36.49 MB下降到22.86 MB,降低37.3%。检测速度从31.1帧/s提升到58.8帧/s,在保证检测精度的同时提高了模型的检测速度,改进后的算法在螺母和U型挂环检测任务中具有广泛的应用前景。In order to solve the problems of multiple small targets,complex backgrounds,and low detection accuracy in the rust detection task of transmission line nuts and U-shaped hanging rings in mining areas,a lightweight rust detection algorithm for transmission line nuts and U-shaped hanging rings in mining areas was proposed.Firstly,a combination of partial convolution PConv and ELAN structure was introduced to construct an ELAN-P module,which reduces the weight of the backbone network and increases the accuracy of model detection;secondly,using GhostConv to lightweight the MP and SPPCSPC structures and reduce the number of model parameters;then,using the Slim-neck structure,a lightweight feature fusion network was used to reduce the number of model parameters and improve the detection rate of the model;finally,the Normalization-based Attention Module(NAM)was adopted to enhance the network′s ability to recognize rust targets and improve the accuracy of model detection.The experimental results showed that the mean average precision(mAP)of the improved network increased from 88.0%to 89.5%.At the same time,the parameter count of the model decreased from 36.49 MB to 22.86 MB,a decrease of 37.3%.The detection speed has been increased from 31.1 frames per second to 58.8 frames per second,which not only ensures detection accuracy but also improves the detection speed of the model.The improved algorithm has broad application prospects in nut and U-shaped hanging ring detection tasks.

关 键 词：深度学习 锈蚀检测 YOLOv7 轻量化 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]