A lightweight deep learning method for medicinal leaf image classification using feature fusion.

Area of Science:

• Computer Vision
• Machine Learning
• Botany

Background:

• Accurate identification of medicinal plant leaves is crucial for their nutritional properties but challenging for human observers.
• Automated systems are needed to assist researchers and farmers in efficient and accurate leaf identification.

Purpose of the Study:

• To develop a novel federated learning-based Feature Fusion deep learning model for classifying medicinal plant leaves.
• To enhance feature integration and classification accuracy using a hybrid approach.

Main Methods:

• Utilized Neighborhood Component Analysis-Convolutional Neural Network (NCA-CNN) framework for feature integration.
• Extracted hybrid features (handcrafted LBP, HOG, and deep features) from RGB images.
• Fused features using canonical correlation analysis (NCA) and classified using a CNN classifier.

Main Results:

• The proposed model achieved an exceptional accuracy of 98.90% on the test dataset.
• Demonstrated robust performance in processing diverse image features across multiple resolutions.
• Successfully trained and evaluated a client-side model using federated learning.

Conclusions:

• The novel federated learning-based Feature Fusion model shows superior performance for medicinal plant leaf classification.
• The approach holds significant potential for advancing academic research and agricultural applications.
• Highlights the effectiveness of integrating handcrafted and deep features for improved accuracy.