Plant Leaf Disease Identification and Classification using Transfer learning

Abstract

—Modern agriculture techniques have given the ability to create sufficient yield in the farm to fulfill
the expanding food requirement of people. Environmental change has consistently remained string to food
security and unfortunate for little farm holders in developing nation like India, over 25% potential crop yield
destroyed because of weeds, pest and diseases, this reports a large loss to small farm holder. An overview of
biotic diseases presented, biotic diseases are to be controlled with the adoption of image based leaf analysis of
plant. Many researchers have attempted to automate the leaf infection recognition process using Image based
identification. In this study, we suggested a transfer learning approach to detect leaf diseases using images of
the tomato leaves. By using pre-trained network results in the analysis of the problems presented in the terms of
validation graphs by comparing VGG16 and AlexNet. If we let the model un-learn it turns out its error rate is
considerably higher. Along these line, one should locate an ideal learning rate at which the model should to
learn. by applying the fine-tuning, here we compared AlexNet & VGG16 in our experimentation with a finite
set of images and also tested the model for real-time identification of leaf disease using Raspberry Pi &
webcam.