An Ensemble-Based Deep Learning Approach Using ConvNeXtV2 and BEiT Models for Watermelon Disease Classification

Abstract

Purpose – This study aims to develop a framework named MelonDx for early classification of watermelon diseases. The goal is to overcome the limitations and subjectivity of traditional methods by applying advanced deep learning models, thereby contributing to improving accuracy and productivity in agriculture.

Design/Methodology/Approach – MelonDx integrates ConvNeXtV2, a state-of-the-art convolutional neural network (CNN), and BEiT, a transformer encoder model (like BERT) through ensemble learning, to exploit their complementary strengths in feature extraction and disease classification. Extensive experiments were conducted on a watermelon disease dataset to evaluate the performance of the proposed framework. Metrics such as Accuracy, F1 score, and AUC were used to assess the effectiveness of the ensemble model compared to individual models and other state-of-the-art methods.

Findings – The experimental results demonstrate that MelonDx achieves superior performance, with an accuracy of 99.43%, an F1 score of 99.43%, and an AUC score of 99.99%. These results highlight the framework’s ability to outperform individual models, such as ConvNeXtV2 and BEiT, as well as existing state-of-the-art methods on the experimental dataset.

Originality/Value – This study presents a novel ensemble-based framework for watermelon disease classification, combining the strengths of CNN and transformer-based models for enhanced accuracy and robustness. The MelonDx framework has the potential to transform precision agriculture by reducing crop losses and optimizing production efficiency through accurate disease detection.