RSA-YOLO: tomato leaf disease and pest detection model based on improved YOLOv8

Abstract

Abstract Currently, the detection of diseases and pests on tomato leaves presents several critical challenges, including ambiguous multi-scale target recognition, significant background interference, and limited adaptability to varying lighting conditions. To address these issues, this study proposed an efficient detection model for tomato leaf diseases and pests based on YOLOv8n, named RSA-YOLO. First, a Receptive-Field Attention (RFA) mechanism was integrated into the backbone network to overcome the limitations of conventional convolutional kernel parameter sharing by dynamically adjusting the spatial features within the convolutional kernel’s receptive-field. This enhancement effectively improves the model’s capability to extract features from complex patterns. Second, to enhance both efficiency and accuracy in multi-scale feature fusion, the original Spatial Pyramid Pooling Fast (SPPF) module in YOLOv8n was replaced with the Spatial Pyramid Pooling with Efficient Layer Aggregation Network (SPPELAN) module. Finally, an Adaptive Spatial Feature Fusion (ASFF) mechanism was incorporated into the detection head to strengthen the scale invariance of features, thereby better addressing variations in target size. Experimental results demonstrated that RSA-YOLO significantly outperformed the original YOLOv8n, with Precision (P) increasing by 1.6%, mean Average Precision (mAP@0.5) improving by 2.6%, and F1-score rising by 2.4%. Furthermore, compared to YOLOv3-tiny, YOLOv5, YOLOv6, YOLOv9, YOLOv11, YOLOV12, and YOLOv13, RSA-YOLO achieved mAP@0.5 improvements of 6.8%, 4.1%, 4.5%, 2.4%, 2.6%,3.5% and 1.9%, respectively. The results indicate that RSA-YOLO significantly improves detection accuracy while maintaining a moderate and controllable computational cost and model size. This demonstrates its feasibility for deployment on portable or resource-constrained devices in agricultural scenarios and provides technical support for monitoring and controlling tomato leaf diseases and pests.

Plant phenotyping relevance

トマト葉の病害・害虫状態を画像から検出するYOLOベース手法を開発・比較評価しており、植物状態の取得・推定が研究の中心である。

Code and data availability

The paper uses a third-party Roboflow/Kaggle tomato leaf dataset and reports no public code, model checkpoints, or data deposit. The authors explicitly state the data are not publicly available and only available upon request, so no paper-specific public asset qualifies.

No evidence-backed public reproduction asset is currently recorded.