Optimized hybrid transformer-extreme gradient boosting model for black pepper disease classification

Abstract

Black pepper (Piper nigrum) is a vital spice crop cultivated in several tropical countries and is highly susceptible to various leaf diseases that significantly affect crop yield and quality. Traditional manual disease identification methods are time-consuming, labor-intensive, and error-prone. While machine learning (ML) and deep learning (DL) techniques have shown promise in plant disease detection, limited work exists specifically on black pepper leaf diseases. Existing research often relies on small, imbalanced, and non-standardized datasets, leading to poor generalization and reduced accuracy. To address this challenge, the proposed study presents a novel approach utilizing deep attention-based transformer-optimized weighted hyper-tuned extreme gradient boosting network (DAT-OWH-XGBNet) for early black pepper leaf disease detection and classification. The model was trained using the black pepper leaf disease dataset, which includes anthracnose, yellowing, quick wilt, slow wilt, and healthy leaf images. The proposed model combines attention-based transformers with a class-weighted and hyperparameter-optimized extreme gradient boosting (XGBoost) classifier to extract high-level spatial features during feature extraction and optimized XGBoost for final classification. The findings show that the proposed model achieved 99.88% prediction accuracy and 99.82% classification accuracy, outperforming existing convolutional neural network (CNN) methods. Our model is designed to support farmers and agricultural specialists in detecting pepper leaf diseases at the early stage.