A Novel Approach For the Detection of Tea Leaf Disease Using Deep Neural Network

Abstract

India is among the biggest tea exporter in the world. However, tea leaf diseases caused by persistent pathogen exposure result in considerable crop yield losses around the world. Detection of the disease of tea leaves at early stages can reduce the damage of tea output. Detecting the disease with the naked eye can be inefficient and counterproductive. Convolutional Neural Networks (CNNs) are commonly used to implement an effective method for the image classification. In detection of plant disease, the use of CNN is widespread. Therefore, in the proposed work, a Deep CNN having multiple hidden layers is considered for the classification of diseased tea leaves into different categories. This helps the network in detecting more number of features and thereby improving the accuracy in disease detection. The classification is done consisting of the following categories of leaves; Gray Blight, Algal Spot, Brown Blight, Helopeltis, Healthy Leaves and Red Spot. Further, a labeled dataset consisting of 5867 diseased and healthy tea leaf images have been created and uploaded on Kaggle. The suggested method demonstrates that the model is able to accurately detect the kind of persistent tea leaf disease with a 96.56% accuracy. The accuracy of the following disease classes are as follows, Algal Spot has an accuracy of 98.23%, Brown Blight has an accuracy of 97.98%, Gray Blight has an accuracy of 93.46%, Healthy classes of leaves has an accuracy of 99.10%, the Helopeltis disease class has an accuracy of 98.98% and Red Spot has an accuracy of 92% The model that is proposed in this literature is far superior than the existing methods in terms of accuracy. Furthermore this model can be adopted to work with various IoT devices to deploy in real world applications and the architecture of this model can be used to train different crop images to classify their diseases.