Mango Leaf Disease Detection: Deep Hybrid Model With Optimal Tuned Weights

ABSTRACT An ML system is meant to automatically upload and match new images of afflicted leaves with training data to recognise the mangoes' leaf illness. With the advancement of these methods, the deep learning concept exists. In this way, this paper takes the work of disease identification via a hybrid deep concept, as the hybridization of the models could make the detection even stronger than the prior models. Thereby, the suggested Mango Leaf Disease detection system has 4 major levels: preprocessing, segmentation, feature extraction, and disease detection. In the preprocessing step, image pre‐processing is done by contrast enhancement and histogram equalisation. In the second step, the segmentation process takes place via the Improved Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm. The feature extraction stage comes next, where a variety of features were extracted, including colour features, pixel features, Improved Center Symmetric Local Ternary Pattern (ICS‐LTP), and the Center Symmetric Local Derivative Pattern (CS‐LDP). A hybrid classifier that combines Bidirectional Long Short Term Memory (Bi‐LSTM) and Deep‐Maxout is suggested for the detection stage, where training is carried out by a new Spider Monkey with Modified Global and Local Leader Update (SMMGL) optimization algorithm via tuning the weights of Bi‐LSTM. Furthermore, the suggested model shows exceptional performance and achieved 93.01% of accuracy and an F‐measure of 96.82%, which surpasses the results of the traditional approaches.