Fruits Disease Classification using Machine Learning Techniques

Abstract

Due to increased population, there is a high demand for agricultural products these days and therefore, effective growth and increased fruit production have become critical. Consequently, for better fruit yield cultivators employ traditional methods for monitoring fruit yield from harvest till ripening of fruit. However, manual monitoring and visual inspection doesn’t always bring the actual identification of fruit disease due to variety of reasons, such as less knowledge about pathogens, requiring more time for disease analysis and that too with less accuracy and so on, consequently, leaving for the need of a professional assistance and expertise. Moreover, the task also becomes difficult as various fruits demonstrate their gesticulation by changing the colour of their skin which can come from nature and resulting in various black or dark brown spots on the fruit skin indicating various diseases. As a result, it is necessary to propose an efficient smart farming strategy that will aid in increased productivity while at the same time involving less human effort. The proposed research work attempts to classify the fruit disease at its early stage by using machine learning techniques. For this purpose, fruit’s texture, and skin colour have been utilized. The approach fundamentally employs three machine learning classifier algorithms - KNN, Decision Tree, and Random Forest. Whereas the features have been determined by using three prominent feature extractors - Haralick, Hu Moments and colour histogram. Finally, the system has been evaluated by utilizing the k-fold cross validation method. Specimen dataset was divided into two groups — the training subset and the test subset. As a rule, four-fold cross-validation, three-fourths of the images were used for training the models whereas, the remaining one-fourth were used for testing purposes. Assessment results for suggested methodology after conducting experimentation on publicly available dataset and drawn confusion matrix and learning cure shows that Random Forest classifiers achieves accuracy about 99% while for K-Means accuracy statistics stands at 98.67% and for Decision trees it is about 97.75% - for colour histogram features.