Plant Health Monitoring System Using Machine Learning and Fuzzy Logic Approach

Abstract

Survival of human beings in various continents across the globe primarily depends upon agriculture. The Food and Agriculture Organisation estimates that pathogens, insects, and weeds together reduce 20-40% of the total agricultural productivity. As it is, with increase in population at a fast pace, the production is unable to meet the food requirements of the countries, especially in the Asian and African continent. Adding to the irony, losses from diseases affect the economy of these regions by reducing the income of farmers and price rises for consumers. Thus, there is a need to minimize the factors responsible for loss of productivity. This can be achieved with plant disease diagnosis as early as possible. newlineThe utilization of image processing and computer vision-based concepts have aided the researchers to provide promising solutions of accurate and early diagnosis to the farmers. The plant disease diagnosis is broadly performed into disease identification, and disease quantification. Most of the research works have been addressing one of these two issues. The popularity of neural networks has drawn a lot of attention from both academia as well as industry for its utilization in developing an effective disease diagnosis system. Some of the distinct advantages of neural networks are its ability to learn by themselves, store data on the network itself, and parallel processing. However, one of the important facts, that having a limited database restricts the utilization of neural networks for varied crop species. newlineThis thesis develops a single plant disease diagnostic system to perform disease identification and estimate the severity of the infection at the same time. The corn, tomato, and grapevine leaf images from the Plant Village database has been utilized for all experiments. The identification has been performed with two kinds of neural networks. Initially, the Probabilistic Neural Networks have been utilized to train the color and texture feature database of the leaf images. The color and texture features ha