Attention-based Hybrid Deep Learning Framework for Plumbago Zeylanica Identification

The identification of medicinal plants based on leaf image classification is an important application area for medicinal plants. The application areas of medicinal plants are medicine, agriculture, and biodiversity conservation. In order to improve the accuracy of leaf image classification for medicinal plants, there are challenges to be addressed. This paper proposes a hybrid deep learning framework based on the benefits of using CNNs and transformer networks for the classification of medicinal plants. This paper proposes a hybrid deep learning framework based on the benefits of using CNNs and transformer networks for the classification of medicinal plants.This proposed framework uses EfficientNet, ResNet50, and Vision Transformer as parallel feature extractors to classify medicinal plants from photos of their leaves. Prior to classification, distinct importance weights are assigned to each model in an attention-based fusion mechanism for feature fusion. In order to identify Plumbago leaves, the fused features are subsequently run through fully connected layers for binary classification. An Explainable AI module is integrated into the suggested framework for creating saliency maps for the proposed model in order to increase its transparency. When compared to individual backbone models, the suggested hybrid deep learning framework increases the classification accuracy for medicinal plants from leaf images.