Agricultural pest detection algorithm based on attention mechanism

Recently, the Yolov5s algorithm in agricultural pest detection is often compromised by upsampling of information, resulting in the loss of detailed information in small target features. To solve this problem, an algorithm for agricultural pest detection in natural environment based on attenuation factors is proposed in this paper. The backbone of Yolov5s was used as a backbone network, and the attention mechanism SE was added to the bottleneck structure of the network to solve the problems such as the disappearance of the gradient in the training process and improve the detection accuracy of small targets. Experimental results show that the MAP0.5 and MAP0.5:0.95 of the improved Yolov5s in IP102 reach 58.76% and 35.53%, respectively. Compared with the original algorithm, the MAP0.5 and MAP0.5:0.95 is improved by 0.83% and 0.4%, respectively.