Abstract: Aimed at the problem that the disease area in the apple leaf image is small and the background environment is complex, which brings about the low accuracy of apple leaf disease identification, a method to improve the detection of apple leaf diseases by YOLOv5 is proposed. This method used the k-means algorithm to recalculate the anchor box to speed up network convergence. Coordinate attention mechanisms were introduced to enhance the characteristic expression of diseases in complex backgrounds. Through the BiFPN feature fusion network, the multi-scale feature aggregation ability of the network was enhanced. EIoU was used as the loss function of the target box to improve the regression accuracy. Experimental results show that the mAP value of the improved method is increased by 1. 4 percentage points compared with YOLOv5, and the detection accuracy is significantly improved compared with different object detection algorithms. The identification algorithm can be deployed on mobile to provide technical guidance for the early prevention and control of apple disease.