A lightweight small-target detection method for grape leaf diseases based on improved YOLOv10n

Grapevines are susceptible to various diseases that threaten yield and quality. The diverse morphology, dense distribution, and small size of grape leaf lesions make accurate detection challenging, particularly on resource-constrained mobile platforms. To address these challenges, we propose YOLOv10n-RLF, a lightweight detector based on YOLOv10n. First, the C2f-RepViT Block (C2f-RVB) module separates token and channel mixing to improve feature-extraction efficiency and reduce redundancy, decreasing parameters by 21.74% and giga floating-point operations (GFLOPs) by 19.54%. Second, the proposed lightweight shared-convolution detection head (v10_LSCD) uses shared convolution to further reduce the parameter count and computational cost. Third, the Feature Pyramid Shared Convolution (FPSC) module replaces Spatial Pyramid Pooling-Fast (SPPF) to enhance multi-scale feature fusion and preserve the representation of small lesion regions. Finally, the SIoU loss is adopted to improve bounding-box regression and accelerate convergence. Experimental results show that YOLOv10n-RLF reduces parameters by 30.4%, GFLOPs by 26.2%, and model size by 36.4%, with only marginal decreases of 0.3% points in mean average precision at an intersection over union (IoU) threshold of 0.5 (mAP@0.5) and 0.2% points in mAP@0.5:0.95. This favorable efficiency-accuracy trade-off demonstrates the potential of the proposed model for resource-constrained mobile deployment and provides technical support for precision grapevine disease management.