Detection of Plant Leaf Diseases using Image Processing

Koay, Kah Leong (2020) Detection of Plant Leaf Diseases using Image Processing. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Koay Kah Leong.pdf Restricted to Registered users only Download (2MB)

Abstract

Agriculture is no doubt a very crucial industry to Malaysia in terms of the amount of GDP it is contributing. It would definitely be a nightmare if anything bad happens to the industry. As we all may know, plant is prone to diseases and it is often inevitable. The best way to keep the impact as minimal as possible is to be able to detect the diseases as quickly as possible. The traditional way of identifying plant diseases is done by the manual observation of experts of the field. This method is very expensive, tedious, inaccurate and consumes a lot of time as well. With the advent of image processing throughout the years, automatic detection of plant disease has been made possible by capturing and processing the image of the plant leaves. By going through the few image processing phases, like acquisition of image, pre-processing, segmentation, extraction of features and classification, the condition of the plant, whether infected or not, will be identified. There are several techniques that has emerged recently to carry out the process for each of the phases like SVM, GLCM, Otsu’s thresholding and etc. MATLAB is a powerful software that are often used for image processing and machine learning algorithms. In this project, the system is being implemented using MATLAB to detect plant diseases through processing image of the plant leaves. In this project, comparison between using k-means clustering and Color Thresholding has been done to evaluate whether which performs better on the target dataset. Moreover, feature extraction of the image will be done using a pretrained CNN model which is ResNet-50 and GLCM to compare whether which performs better as a feature extractor to extract features of the leaf images. SVM are being used as a classifier in this project to classify the diseases on the leaf. Overall the project was able to achieve a maximum classification accuracy of 96.63%.