Mango Leaf Disease Detection Based on Deep Learning Approach

Abstract

This study introduces a new method of disease prediction for mango leaves by breaking it down into four main steps: preprocessing, image segmentation, feature extraction, and disease prediction. Firstly, noise and other undesired artifacts are removed from the acquired raw image by median filtering & histogram equalization to improve the image's quality. The Otsu Threshold Method is then used to segment the preprocessed images. Then, from the segmented images, the most pertinent Texture Features Extraction are made, such as the Upgraded local binary pattern (ULBP) and grey level co-occurrence matrix (GLCM), colour features and pixel features. The framework for detecting mango leaf disease uses these features as input, and it is represented by an improved recurrent neural network (RNN). Additionally, the weight function of the improved RNN will be fine-tuned by employing Arithmetic Operators Customized with Dingoes Optimization (AOCDO) to improve the accuracy of illness identification. The traditional Arithmetic Optimization Algorithm (AOA) and the dingo optimizer are combined to create the new hybrid optimization model (DOX). A comparative assessment is also conducted to confirm the effectiveness of the proposed AOCDO+RNN model.