An Accurate Mango Pest Identification Employing the Gaussian Mixture Model and Expectation-Maximization (EM) Algorithm

Abstract

Mangoes are fruits that originated in South and Southeast Asia. Mango fruits are consumed in high volume worldwide. However, fruit, stem, root, and mango leaves are damaged by pests, and it has a significant negative impact on mango production. Many works were developed for the pest detection of mango; but, none of the works concentrated on the practical applicability and learning efficiency of the model. Furthermore, the boundary features of the leaf regions were not analyzed in depth for accurate identification of different mango pests. Therefore, in this study, Machine Learning (ML) algorithms are used to assess mango fields and identify pests to address the requirement for an early-stage pest identification system. This study presents a novel method for mango plant disease and pest identification and classification using a combination of machine learning, IoT, computer vision, and drone technology. The proposed system is designed to analyze large mango fields and detect biological threats in an early stage and the problems faced by farmers in mango crops. The system utilizes a Dense Net architecture for feature extraction and a custom Corner Net model employing the Gaussian Mixture Model and Expectation Maximization (GMM-EM) algorithm for effective pest classification. The proposed system is tested using the IP102 dataset, which is a large challenging benchmark database for pest identification. The experimental results represented the enhanced pest detection of mango crops with an accuracy of 89.90%, precision of 79.94%, recall of 75.82%, and F1-score of 77.80% by using the proposed model. Furthermore, the proposed system is robust and can reliably locate and categorize different pests even in complicated backgrounds and various insects and their color, size, and brightness. The proposed approach can also send SMS notifications to farmers concerning diseases and pests.