Potatoes are among the most widely cultivated crops globally but are highly vulnerable to bacterial and fungal diseases, particularly early and late blight, which can severely impact yield. Early detection during the budding stage is essential for mitigating losses and improving productivity. To address this challenge, computer vision and machine learning techniques have been extensively explored for automated disease detection. This paper presents a deep learning model that integrates image processing and convolutional neural networks (CNN) to classify potato leaves into three categories: early blight, late blight, and healthy. To enhance feature extraction and model performance, we incorporate the Mish activation function, which improves the learning capacity of the network. Our experiments utilize a dataset of over 1500 labeled images of healthy and diseased potato leaves from the publicly available Plant Village database. The results demonstrate that our CNN model, enhanced with Mish activation, achieves a classification accuracy of 98%, outperforming conventional models. These findings highlight the effectiveness and reliability of our approach, contributing to the development of an automated system for real-time potato leaf disease detection.
Full publication URL
