DiseSniper: A potato disease identification system based on the ResNet model.

Potato is one of China most promising high-production cash crops. However, the frequent occurrence of potato diseases and insect pests directly leads to a considerable crop reduction or low yield, restricting the agricultural economy development. A variety of potato diseases and insect pests dramatically impact potatoes quality and production and pose a significant threat to human health. Therefore, it is necessary to use image processing methods to detect potato leaves growth state at early stages. This study mainly implements the accurate identification of potato blight through the analysis of RGB images and is divided into three parts: data set construction, model construction, and system development. First, this study collected the images of potato blight from the Internet as part of the data set. Blurred and unidentifiable images were removed during data set construction. Then we consulted agricultural experts on crop disease to label the disease types of potato disease image data to construct the labeled potato disease data set. Second, various image filtering algorithms were used to enhance the input image data. A potato blight recognition model was built based on deep learning algorithms. Specifically, the Restnet50 model was used to train the blight classification model. The convolution operation was performed on the input image of 3 channels containing 50 Conv2D functions to distinguish the types and the degree of diseases. Finally, the disease identification software, DiseSniper, was developed based on the trained model. The system was constructed by the agile development software engineering method. Specifically, the software’s front end was designed using PyQT5 and implemented the algorithm in the software. Based on the image data set of potato leaves, the accuracy of the test data set reaches more than 98%.

Download paper here

Recommended citation: Y. Wang et al., “DiseSniper: A potato disease identification system based on the ResNet model,” 2022 10th International Conference on Agro-geoinformatics (Agro-Geoinformatics), 2022, pp. 1-4, doi: 10.1109/Agro-Geoinformatics55649.2022.9859214..