Paper Details
Abstract
Bean leaf disease poses a significant threat to crop yield and quality, which requires reliable and accurate detection methods. In this paper, we propose an ensemble deep learning architecture that integrates feature-level representations from three pre-trained Convolutional Neural Networks: EfficientNetB0, DenseNet121 and ResNet50 through a multi-head attention mechanism. In our framework, each network independently extracts high-level features from the input images. To effectively fuse the complementary information from these diverse models, we employ a multi-head attention module that computes Query, Key, and Value matrices, enabling the model to learn attention weights that dynamically determine the contribution of each branch. Experimental results on a bean leaf disease dataset demonstrate that our proposed ensemble architecture significantly outperforms individual models, achieving high accuracy and strong generalization. This framework offers a promising solution for the automated and accurate detection of bean leaf disease and can be extended to other plant disease recognition tasks.