Detection and Application of Apple Leaf Diseases Based on Object Detection

Abstract

The healthy development of apple leaves is closely related to apple yield. There are three common symptoms of apple leaves: leaf spot, leaf rust and powdery mildew. Apple leaves are very susceptible to these viruses and are highly infectious. If the disease management of apple leaves is improper, it will lead to heavy losses in apple production and directly affect the economic income of farmers. Early detection of apple leaf diseases is the basis for timely prevention, which can inhibit the spread of diseases and minimize serious economic losses. Nowadays, the existing target detection technology can be used in apple leaf detection. Therefore, this paper will use the yolov5 algorithm based on deep learning to detect apple leaves and determine whether apple leaves are diseased. The experimental data comes from the Kaggle AI Challenge competition data set and the public data set on Kaggle. The disease area is manually marked by labeling tools to form a training apple leaf disease data set. In data processing, the Mosaic data enhancement method is used for data expansion, and the image is transferred into the HSV color space for tone transformation, so that the disease features have better highlights and achieve the purpose of data enhancement. The experimental model uses the YOLOv5 s network model. In order to improve the performance of the model, the CBAM attention mechanism is added to the backbone network of YOLOv5 s. The experimental data show that the accuracy (P) of the YOLOv5 s network model is 0.849, the recall rate ( R ) is 0.788, and the average detection accuracy (mAP) is 0.843. The accuracy (P) of YOLOv5s + CBAM was 0.869, the recall rate (R) was 0.798, and the average detection accuracy (mAP) was 0.848. At the end of the experiment, the web application design based on Flask framework is designed. In the last chapter, this paper looks forward to the future and summarizes the experiment.