Detection Methods for Herds Based on the Improved YOLOv5 Model
The accurate identification of herds is essential for management and monitoring in the fields of agricul-tural and animal husbandry production and ecological environment monitoring.Nevertheless,there are issues in the conventional target recognition method of herds,like low detection accuracy and low efficiency.This paper pro-poses an improved-YOLOv5 recognition algorithm of herds(sheep and cattle are used as detection objects)to solve these issues.First,group convolution is used to reduce the computational load by embedding the SA attention mechanism module in the network,and the information exchange among different groups is facilitated by Channel Shuffle operation.Next,the CoordConv convolution is introduced to help optimize the performance of neural net-works in tasks that include coordinate information while enhancing precision and the recall rate.Lastly,the CIOU in the original model is substituted with the EIOU loss function to obtain better localization effects while improv-ing the model's rate of convergence.Experimental results show that the average accuracy of the optimized model on the self-produced dataset reaches 92.3%,which is 1.4%higher than that of the original YOLOv5 model.The improved model have significantly improved detection accuracy and speed,allowing it to detect and identify herds rapidly and accuracely.
YOLOv5 modelHerd recognitionAttention mechanismConvolutional neural networkLoss function
NETL
NSTL
万方数据
