基于改进YOLOv5的牧群检测方法

Detection Methods for Herds Based on the Improved YOLOv5 Model

在农牧业生产和生态环境监测领域,精确识别牧群对于管理和监测工作至关重要。然而,传统的牧群目标检测方法存在着检测精度低和效率低等问题。针对这些问题,提出了一种改进的YOLOv5牧群(以羊和牛为检测对象)识别算法。首先,在网络中嵌入SA注意力机制模块,通过组卷积来减少计算负担,同时采用Channel Shuffle操作来促进不同组之间的信息交流。其次,引入CoordConv卷积,有助于优化神经网络在包含坐标信息的任务中的表现,同时提升精度和召回率。最后,用EIOU损失函数替代原模型中的CIOU,在提高模型收敛速度的同时获得更好的定位效果。经过实验证明,优化后的模型在自行制作的数据集上的平均精度达到了92.3%,相较于原始YOLOv5模型提升了1.4%。改进后的模型在检测精度和速度方面都有明显提升,可以快速而准确地进行牧群的检测和识别。

The accurate identification of herds is essential for management and monitoring in the fields of agricultural 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 proposes 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 networks 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 improving 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.