为解决缫丝时绪下茧粒与工作背景辨识度较低、茧粒分布密集以及茧粒之间相互遮挡而漏检的问题,课题组提出了一种基于改进YOLOv5s的缫丝机绪下茧粒数检测算法.该算法在Backbone中引入RFB-SE(receptive field block-squeeze and excitation)模块,实现了对与工作背景辨识度较低茧粒的检测;使用空间增强注意力模块(spatially enhanced attention module,SEAM)来改进网络的颈部(Neck),解决了由于茧粒遮挡而造成漏检的问题;引入Soft-NMS代替非极大值抑制(non-max suppression,NMS),改变了原始模型对于预测框的处理方式,更好地改善了漏检问题.实验结果表明:该算法在数据集上召回率达到了 98.3％;平均精度均值达到了 98.8％,相比原始模型提高了 3.3％.该算法解决了茧粒与工作背景辨识度低、茧粒间相互遮挡而造成的漏检问题.
Detection of Cocoon Number under Thread of Silk Spinning Machine Based on Improved YOLOv5s
To solve the problems of low recognition between the cocoons and the working background,dense distribution of cocoons,and mutual occlusion between cocoons during the silk reeling process,research group proposed an improved YOLOv5s based algorithm for detecting the number of cocoons in the silk reeling machine.This algorithm introduced the receptive field block-squeeze and excitation(RFB-SE)module in Backbone to detect cocoons with low recognition of work background.Using spatially enhanced attention module(SEAM)to improve the neck of the network solved the problem of missed detections caused by cocoon occlusion.The introduction of Soft-NMS instead of non-max suppression(NMS)changed the way of original model deal with the prediction box and better improved the problem of missed detections.The experimental results show that the algorithm has a recall rate of 98.3％and an average accuracy of 98.8％on the dataset in this paper,which is 3.3％higher than the original model.This algorithm solves the problem of missed detections caused by low recognition of cocoons and working background and mutual occlusion between cocoons.
object detectionimproved YOLOv5sSEAM(Spatially Enhanced Attention Module)squeeze and excitation attention mechanismSoft-NMS
万方数据
维普
