Pest species identification algorithm based on improved YOLOv4 network

Pest species recognition suffers from the problems of easy loss of small targets, dense pest distribution, and low individual recognition rate. To improve the efficiency of pest monitoring, this paper proposed a pest species recognition algorithm called DF-YOLO based on the YOLOv4 network. First, the DenseNet network is introduced into the YOLOv4 backbone network CSPDarknet53 to enhance the feature extraction capability of the model and improve the individual recognition rate of densely distributed targets. Second, the focal loss function is used to improve the effect of sample imbalance on training and optimize the mining process of complex samples. The algorithm was tested on a homemade pest dataset. And the results show that the mAP (mean average precision) of the method is 94.89%, which is 4.66% higher than that before improvement, and the detection speed is 18.92 f/s, which is better than the mainstream algorithm. Experiments were conducted under different light sources, different backgrounds, and different densities. The mAP of our method is 93.96% with an accuracy within 1%, which proves the high robustness of the method. This study improves the intelligence of pest control, and the proposed method is more effective in pest species identification compared with other methods, which can meet the needs of practical use.