Fusing Channel Pruning and YOLOv5s's Bin Label Detection Algorithm
ln order to efficiently extract the key information of labels,a method for accurately detecting labels under the condition of low floating point operation and low parameter number is designed.Taking YOLOv5s as the baseline model,in order to obtain more abundant gradient flow information,an adaptive fusion R2C2f-ECA module with multi-scale information is proposed.Using heavy parameter convolution for feature fusion and neuron activation by LeakyReLU,the complex fea-tures of input data can be better modeled and the representation ability of the model can be enhanced while the dead neurons are avoided.To solve the problem of regression target mismatch,WloULoss,a boundary frame loss function based on dynamic non-monotonic focusing mechanism,is used to improve the regression performance of the model.The experi-ment shows that the improved model can improve the data set of material bin label mAP_0.5 and MAP_5:0.95 by 1.9%and 2.1%respectively,reaching 99.1%and 94%.
万方数据
维普
