Deep Learning for Plant Diseases: Detection and Saliency Map Visualisation

Abstract

Recently, many researchers inspired from the success of deep learning in computer vision to improve the performance of plant diseases detection systems. Unfortunately, most of these studies did not leverage the recent deep architectures and based essentially on AlexNet, GoogleNet or similar architectures. Moreover, the deep learning visualization methods are not taken advantage of, which makes these deep classifiers not transparent and qualified as black boxes. In this chapter, we have tested multiple state-of-the-art Convolutional Neural Network (CNN) archi- tectures using three learning strategies on a public dataset for plant diseases classi- fication.These new architectures outperform the state-of-the-art results of plant dis- eases classification with an accuracy that reached 99.76%. Furthermore, we have proposed the use of saliency maps as visualization method to understand and in- terpret the CNN classification mechanism. This visualization method increases the transparency of deep learning models and gives more insight about the symptoms of plant diseases.