A YOLOv8‐based method for detecting tea disease in natural environments

Abstract

Abstract Tea ( Camellia sinensis ) has a long history in China, and the tea industry plays a crucial role in the national economy. Tea diseases can lead to the reduction of tea yield and reduce the quality of tea. Accurate and rapid identification of these diseases can help prevent and manage them effectively, significantly reducing production losses. However, manual recognition of tea diseases is costly, slow and subject to subjective factors. This paper proposes a deep learning‐based tea disease recognition method in natural environment: referred to as YOLOv8‐tea disease. The tea disease dataset in natural environment was made by ourselves. YOLOv8s is the baseline model. The VoVGSCSP module and efficient multi‐scale attention module were introduced into YOLOv8s to improve the training speed and recognition accuracy of the model. To reduce the number of model parameters, Cross Stage Partial GhostNet Layer was used in the backbone network instead of C2f. Wise‐IoU loss is used as a loss function to solve the problem of inaccurate detection caused by low image quality and improve the generalization ability of the model. Finally, in the dataset of tea diseases, the proposed method achieved an mAP@0.5 (where mAP is mean average precision) of 96.34%. The number of model parameters was reduced to 8.81 M, and the number of floating point operations was reduced to 20.3 G. Compared to the original YOLOv8s model, mAP@0.5 increased by 5.08%, the number of parameters decreased by 26.14%, and the detection speed was the fastest, with the frame per second reaching 153.3.

Abstract from the paper’s OpenAlex record, shown for indexing and research. Copyright remains with the authors or the publisher. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.