Design of a Crop Disease Detection Model using Multi-parametric Bio-inspired Feature Representation and Ensemble Classification

Abstract

Crop disease detection has become an integral part of smart farming models. To perform this task, various intrusive & non-intrusive models are proposed by researchers. Intrusive models have higher deployment cost, higher complexity & contaminate underlying crops, due to which they are limited to clinical use cases. For non-intrusive methods, it is observed that most of these models are capable of achieving better performance under application-specific datasets, and cannot be scaled for larger datasets. To overcome this limitation, a novel crop disease detection & yield prediction model via multi-parametric bio-inspired feature representation is proposed in this text. The proposed model initially uses a crop-specific adaptive thresholding technique, which assists in efficient segmentation for different crop types. The segmented imagery is processed via multiple feature extraction units, which extract colour, shape, texture & convolutional features. These features are further processed via use of Genetic Algorithm (GA) based feature selection model, that implements feature variance maximization to identify optimal feature sets. The selected feature sets are classified using ensemble classification model that combines Support Vector Machines (SVMs), Multilayer Perceptron (MLP), Linear Regression (LR), Decision Tree (DT), and Naïve Bayes (NB) classifiers. These classifiers were selected based on their accuracy performance under different crop types. It was observed that SVM & LR had better performance for Soybean & Squash crops, MLP & LR had better performance for Potato & Pepper crops, while NB had better accuracy for Apple & Raspberry crops. Due to a combination of these adaptive classifiers, the proposed model is capable of achieving an accuracy of 99.5% across multiple datasets, which makes it highly useful for a wide variety of classification scenarios.