Abstract

During the growth phase, pumpkin crops are susceptible to a range of diseases. However, persistent bottlenecks remain in practical field operations, particularly the difficulty of detecting lesions at multiple scales and low identification efficiency. To address this, this study proposes an improved disease detection method based on YOLO11. Specifically, data augmentation techniques—including random flipping, scaling, brightness adjustment, and color jittering—are employed to diversify the training samples and enhance the model’s generalization capability under complex field conditions. Furthermore, by integrating the C2f-RepNCSPFPN structure to strengthen global semantic representation, incorporating the CBAM attention mechanism to suppress interference from complex backgrounds, and modifying the pyramid architecture with SimSPPF to increase sensitivity to small-scale lesions, a high-performance and lightweight detection model named YOLO11n-CCS is constructed. Experimental results demonstrate that the YOLO11n-CCS model achieves significant improvements in key metrics such as mAP@0.5 and recall. It effectively handles challenging scenarios involving leaf occlusion, varying illumination, and overlapping lesions. Concurrently, the model maintains a compact size of 3.6 M parameters and 14.0 G FLOPs, making it suitable for deployment on edge devices. The findings of this research offer a practical technical solution for real-time, precise disease identification in field crops and the development of crop protection robots.

IPC Classification

Keywords