Feature Extraction from Plant Leaves and Classification of Plant Health Using Machine Learning

Abstract

Feature extraction (FE) is an important method that contributes dimensionality reduction especially when extracting the features from images. It is the method of converting the input images into a set of features understandable by the system or software. While processing images, it is desirable to extract features that are oriented toward discernment between two or more classes. In this proposed method, two classes are considered as infected and healthy, and data is distributed into six cases. Features are extracted from the plant images using feature descriptors Hu moments, Haralick texture and color histogram. After feature extraction, the images are classified as infected or healthy based on the feature descriptors. The extracted data are trained with the various classifier models such as LR, LDA, KNN, CART and RF, and validation is performed. 10 k cross-validation is used, and the three best algorithms are chosen to classify and predict the model. Among these models, RF, CART and KNN have given better performance, respectively. Finally, accuracy score above 95% and r2 score above 85% is found for various cases, and the results are visualized for all six cases.