Research on vehicle detection method based on improved YOLOv5
Based on YOLOv5s,the detection of vehicles ahead on traffic roads is investigated,and the original YOLOv5s net-work model is improved in order to improve the vehicle detection accuracy.First,the Mish activation function is applied to the YO-LOv5s model instead of the original ReLU function,and the smoother Mish activation function is used to effectively avoid the prob-lem of gradient disappearance.Secondly,BiFPN is used as the feature pyramid,which increases the information channel for feature transfer and improves the model's perceptual ability and the ability to correlate contextual information.Finally,EIoU is introduced as part of the loss function to accurately represent the positional relationship between the predicted and real frames,which indi-rectly makes the convergence speed increase.Experiments on the D2-City dataset show that the improved YOLOv5s has an average accuracy mAP@0.5 of 84.2%,which is a 2 percentage point improvement over the original YOLOv5s algorithm.
NETL
NSTL
万方数据
