Design and Analysis of Tomato Leaf Disease Identification System Using Improved Lightweight Customized Deep Convolutional Neural Network

Agriculture plays a significant role in survival of human. Agriculture production also plays important role in economy of any country. The disease is major issue which affects the production of crops in agriculture. Early detection of disease is required to increase production in agriculture field. Different type of pesticidesis also available to prevent disease but large of amount can have extensive effect on our health and also, it’s not possible to detect disease at early stage which may cause severe damage to the crop. Early disease detection aids farmers in reducing pesticide use. The detection system will help to cut the time and cost of manual processing. In this work we aim to classify tomato plant disease using improved lightweight customized Deep convolution neural network of plant village dataset. The objective of the work is to implement Mobile Net and customized lightweight model based on MobileNetV2 for image classification and comparing their performance. The proposed model for disease classification in tomato leaf places an emphasis on the implementation of a model that is not only lightweight but also fast and accurate to a significant level. The dataset contains total 14529 of tomato leaf images which include 10 classes. Out of these 2905 images were used for testing and for training purpose 11623 images were used. The model presentedachieves the accuracy of 99.26 % and model size is 3.05 MB in size.