Performance Comparison of Deep CNN Models for Disease Diagnosis on Apple Leaves

Abstract

The apple is one of the most cultivated fruits in the world. They are round in shape and their color varies from green to red. Apple Orchards face constant threats from a large number of insects and pathogens and the early detection of these diseases can help in mitigating these harmful effects. An apple tree takes around six to ten years to mature and produce fruit and therefore, the production costs are high and there is no room for such diseases to get a healthy fruit and a profitable yield. Delayed or incorrect diagnosis of these diseases can lead to using either inadequate or more than required chemicals or using a wrong chemical altogether to treat the plant. Historically, this problem was solved using conventional machine learning algorithms like SVMs, Decision Trees and Random Forests. However, in recent times, the approach to solve this problem has shifted to deep learning, specifically Convolutional Neural Networks. CNN’s are powerful tools that can be used for image classification. We can get state-ofthe-art results even by using small amounts of data and little to no data preprocessing. In this work, we are going to compare some of the state of the art CNN architectures on the task of accurately classifying a given image into different categories of diseases or as a healthy leaf. Finally, experimental results are conveyed and performance analysis of these various architectures has been done.