基于轻量化YOLOv7的井下高压场景安全识别研究
Safety identification in underground high-voltage scene based on lightweight YOLOv7
柏跃屹 1华心祝1
作者信息
- 1. 安徽理工大学 安全科学与工程学院,安徽 淮南 232001
- 折叠
摘要
为了使杨柳煤矿安全监测平台精准快速地识别机电人员高压作业场景中存在的不安全行为,以井下中央变电所为例,聚焦绝缘护具佩戴情况设计安全识别框架.基于YOLOv7 目标检测算法引用部分卷积(PConv),提高模型在处理遮挡或缺失画面的鲁棒性和泛化能力;融合快速神经网络结构(FasterNet),降低计算冗余优化检测性能;最后融合时间空间注意力模块(CBAM),提高算法的特征提取能力.实验结果表明:轻量化处理后较原模型体积缩小 30.5%,计算量减少23.7%,识别平均精度可达97.3%,单张图片检测速度提升 38.1%.在复杂背景下小目标检测任务中有效地解决了漏检问题.
Abstract
In order to enable the safety monitoring platform of Yangliu Coal Mine to accurately and quickly identify the unsafe behaviors in the high-voltage operation scene of electromechanical personnel,taking the underground central substation as an example,the safety identification framework was designed by focusing on the wearing of insulation protectors.Based on the YOLOv7 target detection algorithm,the partial convolution(PConv)was used to improve the robustness and generalization ability of the model in dealing with occluded or missing images.The fast neural network structure(FasterNet)was fused to reduce the computational redundancy and optimize the detection performance.Finally,the convolutional block attention module(CBAM)was fused to improve the feature extraction ability of the algorithm.The experimental results showed that,the volume of the lightweight model was reduced by 30.5%compared with the original model,the calculation amount was reduced by 23.7%,the average recognition accuracy was up to 97.3%,and the detection speed of a single image was increased by 38.1%.The problem of missed detection was effectively solved in the small target detection task under complex background.
关键词
井下高压作业/煤矿机电人员/YOLOv7-tiny/绝缘护具/部分卷积Key words
underground high voltage operation/coal mine electromechanical personnel/YOLOv7-tiny/insulation protector/partial convolution引用本文复制引用
出版年
2024
NETL
NSTL
万方数据