Advanced Multi-Crop Disease Classification: A Stacking Ensemble Approach for Solanaceae and Malvaceae in Tirupattur District

Abstract Leaf diseases pose a severe and multifaceted threat to the productivity of key cash crops in the Tirupattur district, notably Tomato (Solanum lycopersicum), Brinjal (Solanum melongena), Chilli (Capsicum annuum), and Lady's Finger (Abelmoschus esculentus). Traditional detection methods fail to scale effectively across this crop diversity. This paper proposes a novel Stacked Generalization Ensemble Deep Learning (DL) model to establish a single, robust diagnostic architecture for simultaneously classifying multiple diseases across all four species. The architecture utilizes Transfer Learning on three distinct Convolutional Neural Network (CNN) feature extractors: EfficientNetB3, ResNet101, and DenseNet169. Their unique, high-dimensional feature vectors are concatenated and input to a non-linear Support Vector Machine (SVM), which acts as the discriminative Meta-Learner. Tested on a comprehensive, locally augmented dataset of 18,000 images, the ensemble model achieved an outstanding 99.25% overall classification accuracy and an F1-Score of 99.23%. This performance significantly surpassed the best individual baseline model, DenseNet169, by 0.87%. Detailed feature visualization using t-SNE confirms the ensemble's ability to resolve inter-species symptom ambiguities, validating the method’s efficiency and superior generalization for complex agricultural environments. Keywords: Deep Learning, Ensemble Learning, Stacking Generalization, Multi-Crop, Solanaceae, Malvaceae, Support Vector Machine, Transfer Learning.