ECDBMFA: Design of an Ensemble engine for Crop Disease detection via Bioinspired Multidomain Feature Analysis

Abstract Crop disease detection models require integration of multimodal tasks including contextual segmentation, feature representation & selection, classification & post-processing operations. A wide variety of machine learning & deep learning techniques are proposed by researchers to design such models, but most of them are either highly complex, or have limited efficiency when applied to real-time farm fields. To overcome these issues, this text proposes design of an efficient & novel ensemble engine for crop disease detection via bioinspired multidomain feature analysis. The proposed model uses Fuzzy Saliency Maps for identification of contextual disease regions, which are represented into multidomain feature sets. These sets use Fourier, Gabor, Convolutional, Wavelet and Cosine operations, which assist in extraction of multiple high-density features. The extracted features are reduced via a Bacterial Foraging Optimization (BFO) Model, which retains only higher variance feature sets. The retained features are classified by an ensemble model, that combines Naïve Bayes (NB), Multilayer Perceptron (MLP), Logistic Regression (LR), k Nearest Neighbors (kNN) and Support Vector Machine (SVM) classifiers. These classifiers are combined using a boosting model, that assists in retaining unique-correct class training information sets. Due to which the proposed model is able to improve the classification accuracy for Wheat, Tomato, Rice & Corn diseases by 9.4%, while improving precision by 3.5%, recall by 5.9% and reducing delay by 4.3% when compared with existing classification models under similar disease types.