PDMBM: Design of a High-Efficiency Plant Disease Classification Method using Multiparametric Bio Inspired Modelling

A wide variety of machine learning and deep learning models are proposed by researchers, which assist in effective plant segmentation, feature extraction & selection, disease classification and post-processing operations. But these models are designed for disease-specific applications, which limits their scalability & performance when applied to larger plant datasets. To improve this scalability, a novel two stage bioinspired model that assists in high-efficiency feature selection & classifier selection is proposed in this text. To reduce these redundancies, a particle swarm optimization (PSO) model is applied to the feature set. This model aims at inter-class feature variance maximization, and intra-class feature variance minimization, which assists in high-density feature selection & control. The selected features are given to an ensemble classification layer, which is controlled via a customized Genetic Algorithm (GA) model. This model selects optimum configuration of support vector machine (SVM), logistic regression (LR), Naïve Bayes (NB), and random forest (RF) classifiers for the given dataset. These configurations are initialized from the training set, and then fine-tuned via cross-validation operations. Due to combination of PSO & GA for feature selection & classification, the proposed model is able to identify diseases from Maize, Wheat, Rice, Cotton, and Apple plants, which assists in deploying it for large-scale disease detection scenarios. It was also observed that the proposed model was capable of reducing the delay needed for classification by 8.3% when compared with various state-of-the-art methods. This delay reduction was possible due to use of PSO, that assists in high-efficiency feature reduction across multiple datasets.