占道经营目标检测算法研究

Research on the object detection algorithm of occupied business

为辅助城管部门治理占道经营现象,解决传统监管方式效率低的问题,基于固定摄像头与巡查车,结合深度学习的方法,提出了一种改进的YOLOv5s目标检测模型,首先使用SPPCSPC模块替换原SPP模块,增大了感受野,其次采用BiFPN结构,减少了对微小目标以及尺寸差异较大目标存在漏检现象的问题,最后使用SIoU作为定位损失函数,使得预测框的定位与回归更加精准。实验结果表明,改进后的YOLOv5s模型mAP达到了59.70%,比原模型提高了3.02%,具有更好的检测效果,并部署至服务器试运行,提高了城管部门管理占道经营现象的效率。

In order to assist the urban management department to manage the phenomenon of occupied business and solve the problem of low efficiency of traditional supervision methods,based on fixed cameras and inspection vehicles,combined with deep learning methods,this paper adopts a deep learning approach and proposes an improved YOLOv5s target detection model,firstly,using the SPPCSPC module to replace the original SPP module to increase the perceptual field,secondly,using the BiFPN structure to reduce the problem of missing detection for tiny targets and targets with large size differences Finally,the SloU is used as the localization loss function,which makes the localization and regression of the prediction frame more accurate.The experimental results show that the improved YOLOv5s model mAP reaches 59.70%,which is 3.02%higher than the original model,with better detection effect,and is deployed to the server for trial operation,which improves the efficiency of the urban management department in managing the phenomenon of occupied business.