MWD-YOLO11n: A Tomato Leaf Disease Detection Method

Most tomato diseases begin on the leaves and then spread throughout the plant. Therefore, quickly and accurately identifying the type of disease is crucial for timely prevention and control. This study proposes a tomato leaf disease detection method based on an improved YOLO11 architecture, trained on a collected tomato leaf dataset using a single-stage object detection model. First, a Multi-Scale Dynamic Attention (MSDA) mechanism is introduced into the YOLO11 model to enhance its capability in extracting disease-related features. Second, Wavelet Convolution (WTConv) is employed to replace part of the traditional convolution operations, improving the model's ability to capture edge and texture details of the disease. Finally, a Dynamic Sampling operator (DySample) is utilized to optimize the resolution of feature maps, further boosting the model's detection accuracy. The study classifies eight different conditions of tomato leaves. Experimental results on the tomato leaf dataset show that the accuracy reaches 89.3% and the recall improves to 83.2%. The improved model achieves a mAP50 of 88.8% and a mAP50-95 of 77.9%, significantly enhancing the accuracy of tomato leaf disease recognition. This research contributes to the rapid identification and detection of tomato leaf diseases, helping to minimize economic losses.