Abstract
Medicinal plants offer a wealth of essential nutritional properties, yet identifying their leaves is a compound and time-consuming task which often challenges human observers. An automated computer vision system is essential to support researchers and farmers in accurately and efficiently identifying these leaves. This study introduces a novel federated learning-based Feature Fusion deep learning model for classifying medicinal plant leaves ( https://data.mendeley.com/datasets/nnytj2v3n5/1 ). The proposed approach employs an NCA-CNN (Neighborhood Component Analysis-Convolutional Neural Network) framework to integrate features effectively. Using RGB images, the model extracts hybrid handcrafted features, such as Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG), collective with deep features. These features were fused into a cohesive feature vector through canonical correlation analysis (NCA), enhancing key characteristics while reducing noise. A CNN classifier then categorizes the medicinal leaf images. The model efficiently processes diverse image features to train and evaluate a client-side model across multiple resolutions. The proposed method accomplished an exceptional accuracy of 98.90% on the test dataset. The proposed approach demonstrated superior performance, underscoring its robustness and potential for advancing both academic research and agricultural applications.