Comparative Evaluation of Machine Learning and Deep Learning Models for Mango Disease Detection

Abstract

Mango is among the most valuable and popular fruits in Bangladesh and most other tropical nations in commercial terms. Mango farmers, however, experience serious losses in crop year after year because of numerous fruit diseases and in particular, Anthracnose, Mango Scab, and Stem-End Rot. To minimize the damage and enhance the quality of fruits, these diseases should be detected early and correctly. Conventional detection systems are based on human observation by specialists and are not always time-consuming and not always accessible to all farmers. As artificial intelligence in the agricultural sector evolves, machine learning (ML) and deep learning (DL) methods are more and more applied to identify diseases through the use of fruit images. In this thesis, we set our target on creating an automated disease classification system of mango fruit by utilization of image analysis. It will aim at comparing the performance of five well-known models namely Convolutional Neural Network (CNN), VGG16, InceptionV3, K-Nearest Neighbors (KNN), and Random Forest. There was a formation of a dataset of four classes; Anthracnose, Mango Scab, Stem-End Rot, and Healthy mango; based on both field images and open-source agricultural data. Resizing, normalization and augmentation were the preprocessing functions that were used to enhance the quality of the dataset. Standard performance measures, which included accuracy, precision, recall, and F1-score, were used to train and evaluate each model. The ultimate result of this research is to determine what model is the most effective in detecting mango disease using fruit images. The study helps in the creation of intelligent farming implements that would assist farmers to make timely decision, minimize production losses and enhance sustainable farming in Bangladesh and other countries.