Research on Pest Identification Algorithm Based on Deep Learning

In order to solve the problems of difficulty, slow speed and low accuracy in the process of pest identification, In this paper, the IP102 dataset is used as the experimental object, the dataset was trained using the original Faster-rcnn model. The following improvements are made to the original Faster-rcnn model: (1) The backbone network ResN et-50 was replaced with ResN eSt-50, and the split-attention mechanism was introduced, fusion of feature channels, reconstruction of convolution structure, improve accuracy; (2) The loss function Focal-loss is introduced to improve the imbalance of positive and negative samples, and the accuracy and robustness of the model are improved. By comparing the average accuracy of the original model with the improved model, the accuracy and recall of different species of pests were compared, the average accuracy of the proposed method is increased by 2.6%, The feasibility of the proposed method is verified, contribute to the further development of smart agriculture.