Research On Detection Of Tea Tree Leaf Blight Based On Deep Learning

Tea tree leaf blight,a disease that severely damages tea plantation production,leads to a decline in tea yield and quality.In recent years,deep learning algorithms have shown great potential in disease detection and identification,bringing new solutions to tea plantation disease management.However,existing deep learning models still have limitations in terms of real-time and accuracy.To address these issues,this paper proposes a deep learning-based tea tree leaf blight detection method for the YOLOXNano and YOLOv5 models of deep learning,with the following main work:(1)A tea tree leaf blight dataset was constructed in this study.In this paper,images of tea tree leaf blight disease were collected using DJI DJI Avata UAV in Jurong tea farm,Jiangsu province in May 2020.Through screening,cutting and processing,a total of 182 high-quality images of tea tree leaf blight were obtained in this paper,and these images were annotated in detail.This dataset provides an important basis for the research and application of tea tree leaf blight detection models.(2)In this study,a high-precision detection model TLG-YOLOv5 is designed for tea tree leaf blight disease.this paper improves the detection framework based on YOLOv5 by introducing Transformer self-attention mechanism to enhance the global perceptual field of the model;using BiFPN to improve the robustness of multi-scale feature fusion;integrating SA attention mechanism to realize channel random mixing operation to improve the detection accuracy;and incorporating ASFF technique to improve the feature scale invariance.improve detection accuracy;incorporate ASFF technique to improve feature scale invariance.By optimizing the loss function and adopting the migration learning strategy,this paper achieves high accuracy tea tree leaf blight disease detection,which provides strong support for tea tree disease control.(3)In this study,a lightweight YOLOX-based tea tree leaf blight detection model,LSYOLOX,is proposed to improve the YOLOX-Nano benchmark model.The joint attention mechanism is introduced to effectively strengthen the model’s focus on location information and global features.In addition,the T-NET network was used to enhance the model’s ability to capture long-distance dependent and angular region features,enabling the model to extract features more comprehensively and distinguish diseased leaves and soils more accurately.Optimizing the simOTA sampling strategy further improves the recognition accuracy of the model.Translated with www.Deep L.com/Translator(free version)(4)In this study,a binocular camera was used to assess the severity of tea tree leaf blight in real time.The images captured by the binocular camera are first processed by a deep learning model to identify and segment the diseased leaves.The depth information provided by the binocular camera is then used to convert the pixel area of the diseased leaves into the actual area.Based on the actual area of the leaves,the severity of tea tree leaf blight is classified into three levels: mild,moderate and severe,providing an accurate basis for disease control.(5)In this study,a PyQt5-based tea tree leaf blight detection system was designed,integrating a high-precision model with a lightweight model into a single application that supports three detection methods: picture,video and camera.This provides a practical and efficient tool for the diagnosis of tea tree leaf blight,which will help optimize and expand the application in the field of tea tree disease detection in the future.(5)A Qt-based tea leaf blight detection system was designed in this research,integrating high-precision models and lightweight models into a single application,supporting image,video,and real-time camera detection modes.This provides a practical and efficient tool for diagnosing tea leaf blight,contributing to future optimization and application expansion in the field of tea pest and disease detection.