基于改进的YOLOv5s目标实时检测方法

Realtime Detection Method of Parcel Based on the Improved YOLO v5s Algorithm

针对包裹单件分离领域存在的包裹识别不准确、实时性差及效率低等问题,本文提出一种基于改进的YOLOv5s算法的包裹检测模型。将RepVGG模块融入特征提取网络,降低网络参数的计算量,将损失函数CIOU优化为SIOU,引入真实框和预测框之间的向量角度,提升模型准确性。实验结果表明,在包裹检测任务中,该模型的准确率可达到95.2%,召回率达到90.3%,检测速度达到136.9帧每秒(framespersecond,FPS),可实时精确地检测传送带上的各类包裹,包括检测难度较大的异形件,能够满足实际需求。该研究具有一定的实际应用价值。

To address the challenges inherent in the domain of single-parcel separation,characterized by inaccuracies in parcel recognition,suboptimal real-time performance,and inefficient processing,this paper introduces an enhanced parcel detection model based on the YOLO v5 algorithm.The model seamlessly integrates RepVGG modules into the feature extraction network,thereby significantly reducing computational overhead.Furthermore,it refines the loss function from CIOU to SIOU,incorporating vector angle considerations between real and predicted bounding boxes to enhance model precision.The empirical results of comprehensive experimentation underscore the model's exceptional performance in parcel detection tasks.It attains an impressive accuracy rate of 95.2%,a robust recall rate of 90.3%,and a rapid detection speed of 136.9 FPS(Frames Per Second).This model excels in efficiently and precisely detecting a wide array of parcels on a conveyor belt,even those with complex irregular shapes,fully meeting the practical demands of real-world applications.This research bears significant practical implications in the realm of computer vision.