Multi-Layer Convolutional Neural Network Based Identification of Plant Category and Leaf Diseases

Abstract

Agriculture plays a major role in developing and improving the economy of a nation. It is the backbone of the global economy, the world's largest food source, and is responsible for ensuring a decent income for millions of homes worldwide. Pakistan is one of those countries in which agriculture plays a vital role in the development by assisting 19 percent of the Gross Domestic Product (GDP) and 42 percent of the labor employment approximately. As the world is going through major technological reforms in all its sectors including agricultural. Artificial Intelligence (AI) is one of the leading technologies which has been adopted by the most of the world in various domains. Machine Learning (ML) and Computer Vision (CV) has eased up the process of visualizing all types of data and providing the best outcomes from it. Agricultural sector growth is effected by various factors. Plant diseases are one of the leading factors. Plant diseases reduce crop yield and reduce production quality. There are various leaf diseases which cannot be identified through naked eye and it is a very challenging task for the farmers to keep information of all these diseases eventually leading to reduction of quality and overall production. Various research has been made in the field of agriculture using CV, ML and Deep Learning (DL) which includes crops disease detection, plant category identification, Leaf Disease Detection etc. No such study has been made on the combined identification of plant category and their leaf diseases. In this study we present a DL method by using a Multi-Layer Convolutional Neural Network for the identification of plants leaf diseases and their categories. Convolutional Neural Network (CNN) is a method which is being widely used for the classification of images and produces best results for many classification problems. Dataset was collected by two online available resources, Plant viii Village and Fruits 360. Both the datasets were combined by taking the common plants classes including different images of Fruits, Vegetables and their Leaf diseases eventually leading to a final dataset of 70 classes having 167k images approximately. Dataset has been preprocessed according to the requirements of the proposed Convolutional Neural Network. A novel CNN has been proposed in this study for the classification of the acquired dataset. Several CNN configurations have been used for training, validating and testing the data. We have achieved an overall training and validation accuracies of 99.95 and 99.53 percent respectively. Our model is also tested on a batch of test images providing the best test accuracy of 99 percent.