Automated Tree Leaf Disease Detection System Using Deep Learning and Image Classification

Abstract

Plant leaf diseases remain one of the persistent challenges in agricultural production. Diseases have rapid spread, are often hard to detect at an early stage, and when the symptoms manifest, it may be too late to take necessary corrective measures. At the moment, detection relies on a field survey performed by experienced personnel who visually inspect the state of each plant. This process is both lengthy and highly inaccurate. Our solution is a system capable of automatically identifying various plant diseases using Convolutional Neural Networks to analyze leaf images. We designed an app which was able to successfully operate with leaf pictures taken in natural conditions. Upload the image — get your diagnosis. This was our goal. The neural network we used was initially trained using 54,303 images (healthy as well as those exhibiting symptoms). Combining basic image preprocessing methods along with more sophisticated deep features extraction allowed us to build a model capable of detecting the first stages of diseases, which could otherwise go unnoticed. Classification accuracy amounted to 93.2%, and precision was 94.5%.