Hybrid CNN-SVM with Borderline SMOTE for Imbalance Class Cabbage Plants

Abstract

Cabbage farming is highly vulnerable to diseases and pests, leading to substantial yield losses if not properly managed. Traditional diagnostic methods, reliant on manual assessment, are often time-consuming and inaccurate. This study introduces a hybrid approach combining Convolutional Neural Networks (CNN) and Support Vector Machines (SVM) to address these challenges, specifically focusing on improving classification accuracy in imbalanced cabbage image datasets. CNNs are leveraged for their powerful feature extraction, while SVM, optimized using a One-vs-All strategy, enhances multi-class classification. To handle data imbalance, Borderline SMOTE (Synthetic Minority Over-sampling Technique) is applied, generating synthetic samples to balance underrepresented classes. The SqueezeNet architecture is employed for feature extraction, with SVM hyperparameters fine-tuned via grid search. Results demonstrate that the integration of CNN, SVM, and Borderline SMOTE significantly improves classification performance, particularly for minority classes, achieving an accuracy of 99%. This approach offers a more reliable and efficient tool for early detection of cabbage diseases and pests, contributing to better agricultural management and reduced crop losses.