Classification and Identification of Tea Diseases Based on Improved YOLO V7 Model of MobileNeXt

Abstract

To overcome the constraints associated with conventional approaches used in the classification and detection of tea diseases, which are characterized by their limited accuracy and sluggish responsiveness, this study introduces an enhanced YOLOv7 lightweight model algorithm integrated with MobileNeXt. This refinement not only bolsters the model's capacity for extracting and processing features but also effectively lightens the computational load, expedites recognition, and integrates a dual-layer routing attention mechanism visual converter to enhance the capture of crucial details and textures within disease images. Consequently, these enhancements lead to improved model performance and computational efficiency, ensuring precise and rapid identification of tea diseases. Furthermore, this model incorporates the more appropriate SIoU as the loss function, mitigating losses, minimizing omissions, and reducing misclassifications, thus resulting in superior recognition, even in complex image backgrounds. Based on the training outcomes, the enhanced model attains Precision, Recall and mean Average Precision scores of 93.5%, 89.9%, and 92.1%, respectively, marking substantial enhancements of 5.06%, 2.16%, and 2.91% compared to the original YOLOv7 model. Additionally, the model's size is reduced by 19.12%, and its detection speed accelerates by 11.13%. This improved model excels in accurately and expediting.

Plant phenotyping relevance

茶の病害画像から病気を分類・検出するYOLOv7改良モデルの開発と性能評価が中心であり、植物の病害状態を画像ベースで推定するフェノタイピング手法に該当する。

Code and data availability

The paper's Data Availability statement links a public GitHub repository containing the authors' code and tea disease image dataset used to train and evaluate the improved YOLOv7 model.