Tomato Leaf Disease Detection and Classification Using Convolutional Neural Network and Machine Learning

Abstract

Tomato is one of the most important horticulture crops in the world, however, it faces many important threats due to different diseases which may deteriorate the yield quality and loss. For management of efficient crop and reasonable yield, rapid and precise disease identification is very important. This study has made elaborate comparison between the conventional machine learning and deep learning for the classification of Tomato leaves images. This dataset has a total of 30000 images of 10 different types of diseases and healthy condition of Tomato leaves. The segmentation has been made for each disease as Bacterial-Spot (BS), Early-Blight (EB), Healthy, Late-Blight (LB), Leaf-Mold (LM), Septoria-Leaf-Spot (SLS), Spider-Mites-Two-spotted-Spider-Mite (SMTSSM), Target-Spot (TS), Tomato-Mosaic-Virus (TMV) and Tomato-Yellow-Leaf-Curl-Virus (TYLCV). A Convolutional Neural Network (CNN) is applied which consists of four convolutional layers, max pooling, a fully connected layer, dropout for regularization and a softmax layer for class probability. The classic machine learning techniques such as adaBoost classifier (ABC), K-nearest neighbor (KNN), random forest classifier (RFC), and naïve bayes (NB) has also been examined. Our experiment showed that the CNN is significantly better than the classic classifiers, it was able to able to obtain 96% accuracy whereas the classic classifiers gave poor accuracy, ABC 52%, KNN 56%, RFC 71% and NB 49%.