基于YOLOv5s的矿区地质灾害检测轻量化改进算法  

Lightweight improvement algorithm for geological hazard detection in mining areas based on YOLOv5s

作　　者：贾永森 林珍锋 高建宏 张喆 续欣莹 JIA Yong-sen;LIN Zhen-feng;GAO Jian-hong;ZHANG Zhe;XU Xin-ying(Huajin Coking Coal Co.,Ltd.,Lvliang 033000,Shanxi Province,China;Taiyuan University of Technology of College of Electrical and Power Engineering,Taiyuan 030024,China)

机构地区：[1]华晋焦煤有限责任公司,山西吕梁033000 [2]太原理工大学电气与动力工程学院,太原030024

出　　处：《信息技术》2025年第2期46-54,60,共10页Information Technology

基　　金：山西省自然科学基金(202103021224056)。

摘　　要：针对目前地质灾害检测领域存在检测算法复杂、计算量大与数据集匮乏等问题,文中提出了基于YOLOv5s改进的轻量化网络,并将所提网络在自制地质灾害数据集GDD1上进行训练测试。文中的最优改进方案是YOLOv5s-GhostNet-EIoU,采用Ghost Module替换YOLOv5s网络中的部分卷积模块、C3Ghost替换YOLOv5s网络中的C3模块,大幅降低网络参数,并改进损失函数来弥补轻量化带来的精度损失。实验结果表明,YOLOv5s-GhostNet-EIoU网络mAP值达78.9%,相较于YOLOv5s基网络,模型体积压缩47.1%,计算量减小48.8%,参数量降低47.8%。该算法在保持YOLOv5s高检测精度的前提下,能够满足在无人机平台上的部署与实时检测的要求,实现了模型的轻量化。A lightweight network based on the improved YOLOv5s has been proposed to address the issues of complex detection algorithms,high computational costs,and limited datasets in the field of geological disaster detection.In this paper,the optimal improvement solution,YOLOv5s-GhostNet-EIoU,was developed by replacing some convolution modules in the YOLOv5s network with Ghost Modules and replacing the C3 module with C3Ghost.This significantly reduces the network parameters,and the loss function was also improved to compensate for the accuracy loss caused by the lightweight design.Experimental results demonstrate that the YOLOv5s-GhostNet-EIoU network achieves an mAP value of 78.9%on the self-made geological disaster dataset GDD1.Compared to the base YOLOv5s network,the model size is compressed by 47.1%,computational costs are reduced by 48.8%,and parameter quantity is decreased by 47.8%.This algorithm meets the requirements of deployment and real-time detection on unmanned aerial vehicle(UAV)platforms while maintaining the high detection accuracy of YOLOv5s.It successfully achieves model lightweightization.

关 键 词：矿区 地质灾害 目标检测 YOLOv5s 轻量化 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]