A Lightweight Pest Detection Algorithm based on the PI-Net basic Model

Insect infestation is a very serious natural disaster in agricultural production. Therefore, the detection of pest species is of great significance for agricultural insect control and production. However, the current traditional pest detection methods have the problem that they cannot guarantee the accuracy of the equipment under low computing power. Therefore, this paper proposes a lightweight pest detection algorithm based on the PI-Net basic model, in which the Fuses-mbconv structure is introduced into ResNet and combined with the high-efficiency network to obtain the EfficientNet_b3_WB backbone network. The selected dataset is then tested and compared with the PI-Net base model. The experimental results show that the improved model achieves 68.9% on the AP50 evaluation standard compared to the traditional PI-Net model, significant improvements were achieved on several key performance indicators (4.8 points for the average precision AP metric, 3.2 points for the AP50 metric, and The AP75 indicator also improved by 3.9 points).