Detection and Identification of Mango Disease Leaves Based on Improved YOLOv7

Detection and identification of diseased mango leaves is an effective means of scientific disease prevention and control, and is crucial to mango production. In order to improve the ability to accurately locate and detect mango leaf disease characteristics, a SimAM-YOLOv7 target detection algorithm was proposed based on the YOLOv7 target detection algorithm. First of all, considering that the characteristics of the types of mango disease leaves are relatively small and the recognition rate is low, which is prone to missed detection and false detection problems, a module MP-S (MPConv-SimAM) was constructed to improve the model’s attention to small targets, add the SimAM (Selective Image Attention Mechanism) no-parameter attention mechanism to the down-sampling stage at some Necks to improve the model’s attention to small disease targets and prevent missed detections and false detections; Secondly, add a small target detection layer, Reduce the loss of small target features and improve the model’s ability to detect small targets. Finally, the SIOU loss function based on angle vector regression is selected as the border loss function to further improve the detection performance of the model. On the self-made data set, the mAP of SimAM-YOLOv7 increased by $3.62 \%$ compared to the original YOLOv7 model, reaching $92.1 \%$, and the detection speed reached 46.67 frames/second.