Scholarlink

<p>Currently, tomato leaf disease detection faces critical challenges, including ambiguous multi-scale target recognition, strong background interference, and limited robustness under varying lighting conditions. Although recent state-of-the-art (SOTA) YOLO-based detectors have achieved notable progress, their performance remains constrained under complex agricultural scenarios, mainly due to limited receptive-field adaptability, insufficient multi-scale feature fusion, and weak scale awareness. To address these limitations, an efficient tomato leaf disease detection model based on YOLOv8n, termed RSA-YOLO, is proposed. A Receptive-Field Attention (RFA) mechanism is introduced into the backbone to enable dynamic receptive-field adaptation and improve feature discriminability. Meanwhile, the Spatial Pyramid Pooling with Efficient Layer Aggregation Network (SPPELAN) is employed to strengthen multi-scale contextual representation, and an Adaptive Spatial Feature Fusion (ASFF) mechanism is incorporated into the detection head to enhance feature scale invariance. Experimental results show that RSA-YOLO consistently outperforms the original YOLOv8n, achieving improvements of 1.6% in Precision, 2.6% in mAP@0.5, and 2.4% in F1-score. Compared with YOLOv3-tiny, YOLOv5n, YOLOv6n, YOLOv9C, YOLOv11n, YOLOv12n, and YOLOv13n, RSA-YOLO achieves mAP@0.5 gains of 6.8%, 4.1%, 4.5%, 2.4%, 2.6%, 3.5%, and 1.9%, respectively. These results demonstrate that RSA-YOLO improves detection accuracy over existing YOLO-based detectors while maintaining a moderate and controllable computational cost (11 GFLOPs) and model size (4.61M parameters), supporting its practical applicability in agricultural disease monitoring scenarios.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RSA-YOLO: an improved YOLOv8-based model for tomato leaf disease detection

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Qin Dong,
  • Fei Wang,
  • Mingping Tang,
  • Wei Wu

摘要

Currently, tomato leaf disease detection faces critical challenges, including ambiguous multi-scale target recognition, strong background interference, and limited robustness under varying lighting conditions. Although recent state-of-the-art (SOTA) YOLO-based detectors have achieved notable progress, their performance remains constrained under complex agricultural scenarios, mainly due to limited receptive-field adaptability, insufficient multi-scale feature fusion, and weak scale awareness. To address these limitations, an efficient tomato leaf disease detection model based on YOLOv8n, termed RSA-YOLO, is proposed. A Receptive-Field Attention (RFA) mechanism is introduced into the backbone to enable dynamic receptive-field adaptation and improve feature discriminability. Meanwhile, the Spatial Pyramid Pooling with Efficient Layer Aggregation Network (SPPELAN) is employed to strengthen multi-scale contextual representation, and an Adaptive Spatial Feature Fusion (ASFF) mechanism is incorporated into the detection head to enhance feature scale invariance. Experimental results show that RSA-YOLO consistently outperforms the original YOLOv8n, achieving improvements of 1.6% in Precision, 2.6% in mAP@0.5, and 2.4% in F1-score. Compared with YOLOv3-tiny, YOLOv5n, YOLOv6n, YOLOv9C, YOLOv11n, YOLOv12n, and YOLOv13n, RSA-YOLO achieves mAP@0.5 gains of 6.8%, 4.1%, 4.5%, 2.4%, 2.6%, 3.5%, and 1.9%, respectively. These results demonstrate that RSA-YOLO improves detection accuracy over existing YOLO-based detectors while maintaining a moderate and controllable computational cost (11 GFLOPs) and model size (4.61M parameters), supporting its practical applicability in agricultural disease monitoring scenarios.