IMAGE SEGMENTATION AND DEEP LEARNING FOR PLANT DISEASE DETECTION

Abstract

We reviewed numerous research and survey paper to check the techniques applied by other but they do not mention using advanced image segmentation methods. So, to test various methods for image processing and image segmentation we tried various methods like HSV segmentation, color correction, segmentation using flood fill algorithm, etc. At the end, by extracting the leaf portion and using various image processing techniques, we can increase the accuracy for detecting the plant disease. We trained a deep learning CNN based image segmentation model on more than 30k leaf images of more than 5 plants. After training and segmenting the dataset with the trained model, we used various image processing techniques to check for the increase in the accuracy. Also, to get the best result with least computational load, we experimented with various parameters while training the model. By using these mentioned methods, we are able to achieve 0.9795% validation accuracy. After training the model to classify the disease and plant, we converted it into tflite so that it can be directly loaded on the smartphone.