Health diagnosis of mango trees using image processing techniques

Abstract

A Mango disease detection artificial intelligent model needs robust and effective newlinefeature extraction methods. The machine vision system has been designed for the newlineidentification of disease in plants from color leaf images. The research done proposes newlinenovel algorithms to extract color features Pseudo Color Regions and Texture Features newlineusing Pseudo Color Co-Occurrence Matrix. A new Mango dataset has been created and newlinealgorithms tested on it. An artificial intelligence model has also been created and tested on an existing disease dataset of Apple and Tomato plants. Results were compared with existing methods in the literature. The effectiveness of each statistical function was studied in classifying the pattern using a Support Vector Machine. For textures that are newlinedifferent like smooth new leaves, dry leaves, growth a Gray Level Co-occurrence based newlinestatistics was effective but values failed to discriminate in certain diseases. The proposed and implemented novel method which uses second-order statistics on a pseudo-color-based co-occurrence matrix has resulted in better classification. Pseudo Color Region feature is created using a novel intermediate data structure and found to be more effective than hue-based color features. It identifies dots, spots, patches and regions of different colors on the leaf and uses that as a feature vector to classify plant diseases. This generic method can be applied for early disease detection for plants and help farmers take corrective measures to avoid loss of yield.