In order to solve the problems of difficulty,high risk,low efficiency and low detection accuracy of traditional debris flow source identification in artificial field investigation,an improved YOLOv5s-GCE debris flow source target detection method has been proposed.Firstly,Ghost convolution is used to replace common convolution to complete lightweight feature extraction and reduce the number of parameters and computation;Secondly,CBAM dual channel attention mechanism was added to the backbone network of YOLOv5s model to pay attention to important features of key regions and effectively improve model performance;Final-ly,CIOU is replaced by EIOU loss function to improve the detection accuracy of the algorithm.Compared with YOLOv5s model,mAP@0.5 increased by 8.6%,mAP@0.5:0.95 increased by 10.5%,parameter number decreased by 10.6%,and computation amount decreased by 10.1%.It can effectively detect debris flow sources,provide theoretical reference and data support for the accurate identification and positioning of debris flow sources,and further serve the risk assessment and prevention of debris flow in Lidaosi Yaogou area.
关键词
注意力机制/泥石流物源/目标检测/深度学习/Ghost卷积/损失函数
Key words
attention mechanism/debris flow sources/object detection/deep learning/Ghost convolution/loss function
万方数据