基于改进YOLOv5的平贝母检测模型

Fritillaria ussuriensis Maxim Detection Model Based on Improved YOLOv5

现有平贝母表土剥离机智能化程度不高,在收获平贝母时存在造成平贝母损伤的问题,故需要研制一款智能化的平贝母表土剥离机,而实现智能表土剥离的第一步就是实现智能识别,因此,提出一种基于改进YOLOv5的平贝母检测模型YOLOv5-Swin-L。通过引入Swin transformer,在骨干网络中替换C3,减少序列长度和降低计算复杂度,从而简化模型的参数量;同时用ACON激活函数将原来网络结构中的激活函数替换,可提高模型的精确度,增加模型的鲁棒性。实验结果表明,改进后的YOLOv5-Swin-L对平贝母识别的准确率最高可达96.39%,召回率最高可达95.76%,优于YOLOv5系列的其他网络模型。

The existing Fritillaria ussuriensis Maxim topsoil stripping machine is not highly intelligent and may cause Fritillaria ussuriensis Maxim damage when harvesting Fritillaria ussuriensis Maxim,so it is necessary to develop an intelligent Fritillaria ussuriensis Maxim topsoil stripping machine,and the first step to realize intelligent topsoil stripping is to realize intelligent identification.Therefore,a Fritillaria ussuriensis Maxim detection model based on improved YOLOv5-YOLOv5-Swin-L is proposed.Swin Transformer is introduced to replace C3 in the backbone network to reduce the sequence length and computational complexity,thus simplifying the number of parameters in the model.At the same time,the activation function in the original network structure is replaced by ACON activation function,which can improve the accuracy and robustness of the model.Experimental results show that the improved YOLOv5-Swin-L can identify Fritillaria ussuriensis maxim with the highest accuracy of 96.39%and the highest recall rate of 95.76%,which is superior to other network models of YOLOv5 series.