Deep Learning Based Analysis and Detection of Potato Leaf Disease

Abstract

Potato is an essential crop worldwide, and its leaves are prone to numerous illnesses, including early and late blight. Accurately detecting these diseases can help farmers prevent their spread and minimize yield loss. In this research paper, we propose a deep learning approach to classify potato leaves into three categories: early blight, late blight, and healthy. Our dataset consists of images of potato leaves with different diseases and healthy leaves. A collection of 4072 images, including healthy potato leaves and leaves infected with Early blight, Late blight served as the basis for our analysis. To increase the dataset size, we pre-processed the images by scaling them to 256 x 256 pixels and used data augmentation methods. Our findings show the CNN model's ability to accurately classify potato leaf diseases and its potential to help with early diagnosis and prevention of these diseases. Future research may examine the illnesses of potato leaves categorized using larger datasets and perform the evaluation of various additional machine learning algorithms. There are several challenges in existing techniques like dataset size, labeling accuracy, class imbalance, generalization to new disease strains and some which cannot be overcome like Environmental Variability. The efficiency and production of potato farming could be increased with the development of automated methods for the identification and prevention of potato leaf disease. We used 4072 images total, of which 3251 were used for training, 405 for testing, and 416 for validation in order to analyse the model performance. In the study of the results, the model provides an accuracy of 98.52% for identifying various potato leaf diseases.