Enhancing Rice Leaf Disease Classification: A Combined Algorithm Approach for Improved Accuracy and Robustness

This research addresses the problem of improving image classification accuracy, given the importance of classification accuracy in applications such as disease diagnosis and object recognition. This research aims to explore various deep-learning architectures and ensemble methods to improve classification accuracy effectively. We use a comprehensive approach for methodology, evaluating training data from various architectures using 17 transfer learning models. Our method incorporates models such as EfficientNetB0, AlexNet, and MobileNetV2, utilizing their unique strengths. We also designed a two-stage model that progressively combines architectures between low-accuracy and two high-accuracy models to create a more accurate classifier. The dataset contains 5932 images of rice leaf diseases distributed into four classes: Bacterial Leaf Blight, Blast, Brown Spot, and Tungro. The outstanding results showed a substantial increase in accuracy from 0.35 to 0.97 in the ensemble model. This significant improvement underscores the potential of integrating different architectures to utilize complementary features, ultimately improving classification accuracy. This research provides insights into image classification and offers practical solutions to improve accuracy in various domains. As for the implications, this study shows the promise of a blended approach to deep learning architecture in significantly improving image classification performance in various domains.