Double Channel CNN Based Tomato Plant Leaf Disease Detection

Abstract

Early blight, late blight, bacterial spot, leaf mold, spider mites, yellow leaf curl virus and Septoria leaf spot are the most common diseases of tomato plant which cause major loss to agriculture sector. Diagnosis and prognosis at early stage are prime challenges to save life of tomato crop. This research presents a state-of-the-art Double- Channel Convolution Neural Network (D-CNN) model for diagnoses of 8 different Classes of Tomato Plant Leaves. The proposed model integrates two separate Convolutional Neural Network (CNN) channels with similar number of Conv Layers and Parameters. One Channel processes the pre-processed Red-Green-Blue (RGB) data, whereas the second channel processes the region segmented data obtained by applying a Multi-Otsu Thresholding algorithm.
We verified the proposed model on the Plant Village dataset which consists of more than 16,000 images of tomato plant leaves. The proposed model attains an overall accuracy of 94% after 200 iterations, comparatively 3.2% faster than Support Vector Machine (SVM), 2.2% faster than Probabilistic Neural Network (PNN), and 8% faster than Residual Neural Network (ResNet-50)