To address the challenges of large model size, limited detection accuracy and poor adaptability to complex environments in tomato leaf disease detection tasks, this paper proposes a lightweight and efficient detection method based on an improved YOLO11n. First, a dynamically refined intersection over union loss function is introduced to optimize bounding box regression quality across different training stages. Subsequently, an adaptive multiscale fusion module is designed to enhance feature extraction adaptability to varying scales. To further strengthen spatial perception across lesions of different sizes, a progressive receptive field via dilated convolutions module is proposed. Finally, a detail enhanced detection head is incorporated to improve detection performance on small-scale and blurred-boundary disease regions. Extensive experiments validate the effectiveness of the proposed approach, achieving a 2.1% improvement in mean Average Precision at an Intersection-over-Union (IoU) threshold of 0.5 (mAP50) and a 2.9% improvement in mean Average Precision (mAP) averaged over Intersection-over-Union thresholds from 0.5 to 0.95 (mAP50–95) compared with the YOLO11n baseline, while boosting inference speed to 482 frames/second. The proposed method demonstrates excellent accuracy, real-time performance and lightweight deployment capability, providing a novel technical solution and practical support for intelligent agricultural disease detection.
Yuan et al. (Thu,) studied this question.