Robust multi-target multi-scale tomato leaf disease detection for precision agriculture applications.

Abstract

The tomato is one of the most important economic crops worldwide; frequent occurrences of foliar diseases can severely affect its quality and yield, resulting in substantial economic losses. However, state-of-the-art methods still struggle with multi-target, multi-scale disease detection in complex scenarios, lacking accuracy and speed for tomato leaf diagnosis. A novel improved YOLO v8s model is proposed in this study to achieve high-precision and fast identification of multi-target and multi-scale tomato leaf diseases. First, a multi-target, multi-scale image dataset encompassing seven typical tomato diseases was developed to effectively enhance the model's robustness under complex practical scenario by integrating multiple public datasets and employing diverse data augmentation techniques. Second, a transfer learning strategy was employed to transfer high-quality features from a pretrained model to the disease detection task, thereby improving convergence speed and generalization ability. Finally, the CBAM (Convolutional Block Attention Module) channel-spatial attention mechanism was introduced into the YOLO v8s network, enabling the model to adaptively focus on critical regions and significantly enhance feature extraction and target localization performance. Experimental results demonstrate that the improved YOLOv8s-CBAM model achieves superior performance in complex scenarios, with a precision of 96.9%, recall of 97.3%, F1 score of 97.0%, and mAP@0.5 of 99.1%, representing improvements of 2.5%, 2.0%, 2.2%, and 1.8%, respectively, over the original YOLO v8s model. Moreover, the model size was reduced to 24.8 MB, a decrease of 11.7 MB compared to the original, achieving an effective balance between accuracy and lightweight design. These results indicate that the proposed method exhibits enhanced feature extraction and localization stability in multi-target, multi-scale disease identification tasks, providing an effective technical solution for automated detection in complex agricultural disease scenarios.

Plant phenotyping relevance

トマト葉の病徴を画像から検出・識別するYOLOv8s-CBAM手法を開発し、データセット構築と性能比較検証を行っており、植物病害状態の画像ベース表現型計測が中心である。

Code and data availability

The paper describes a custom 9,000-image tomato disease dataset built from PlantVillage with mosaic synthesis and augmentation, and a YOLOv8s-CBAM model, but no block contains a data or code availability statement, deposit, or authors' public URL for the dataset, trained model, or analysis code. PlantVillage is cited,

No evidence-backed public reproduction asset is currently recorded.