基于改进YoloX的输电通道工程车辆小目标检测识别  

Detection and Recognition of Small Targets of Vehicles in Transmission Channel Based on Improved YoloX

作　　者：张智坚 焦良葆[1] 高阳 邹辉军 孙宏伟 王彦生 ZHANG Zhijian;JIAO Liangbao;GAO Yang;ZOU Huijun;SUN Hongwei;WANG Yansheng(AI Industrial Technology Research Institute,Nanjing Institute of Technology,Nanjing 211167;Jiangsu Intelligent Perception Technology and Equipment Engineering Research Center,Nanjing 211167)

机构地区：[1]南京工程学院人工智能产业技术研究院,南京211167 [2]江苏省智能感知技术与装备工程研究中心,南京211167

出　　处：《计算机与数字工程》2025年第2期415-421,485,共8页Computer & Digital Engineering

基　　金：江苏省自然科学基金项目(编号:BK20201042)资助。

摘　　要：针对输电通道下存在的工程车辆小目标严重威胁输电通道安全,原始YoloX算法对小目标漏检严重的现象,在原始单阶段目标检测算法YoloX的基础上添加视觉注意力机制SK来增大感受野;使用空洞卷积模块替换原始网络中的SPP模块,进一步融合不同感受野信息;在网络Neck部分添加ASFF模块,提高底层特征中的细粒度特征;最后利用二次识别后处理方式,进一步降低小目标的漏检率。实验结果表明,提出的算法提高了检测的准确率,与传统的YoloX算法相比,mAP提高了8.46%,小目标的识别效果明显提升,证明了新算法的有效性。In view of the phenomenon that small targets of engineering vehicles under the transmission channel seriously threaten the safety of the transmission channel,and the original YoloX algorithm has seriously missed the detection of small targets,the visual attention mechanism SK is added to the original single-stage target detection algorithm YoloX to increase the receptive field.The hole convolution module is used to replace the SPP module in the original network to further fuse different receptive field information.ASFF module is added in the network Neck part to improve the fine-grained features in the underlying features.Final⁃ly,the second recognition post-processing method is used to further reduce the miss detection rate of small targets.The experimen⁃tal results show that the proposed algorithm improves the detection accuracy.Compared with the traditional YoloX algorithm,the map is improved by 8.46%,and the recognition effect of small targets is significantly improved,which proves the effectiveness of the new algorithm.

关 键 词：YoloX 小目标 注意力机制 ASFF 二次识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]