CSR-YOLOv11: Detection Algorithms for Diseases of Soybeans, Tomatoes and Strawberries

To address the challenges of low accuracy, susceptibility to background interference, and difficulty in detecting target diseases in existing algorithms for strawberry, tomato, and soybean disease detection in agricultural settings, this study proposes an improved disease detection model, CSR-YOLOv11, based on YOLOv11. The model demonstrates enhanced generalization capabilities. In the backbone network, a novel CSCMViTBv3 module is introduced by integrating MobileViTv3 with <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1 \times 1$</tex> convolutions. This module reduces model parameters while improving feature representation. For the neck network, the RepELAN2 module is designed using RepConvN as the foundational component. By leveraging reparameterized convolutions, this module addresses the low inference speed and limited feature learning capacity inherent in multi-branch architectures, thereby enhancing detection accuracy for small-target lesions. Experimental results demonstrate that on a selfconstructed dataset of soybean, tomato, and strawberry diseases in agricultural environments, the CSR-YOLOv11 model achieves improvements of 2.3% in mAP 50(%) and 3.4% in detection accuracy compared to the baseline YOLOv11 model.