Transfer Learning-Based Deep Learning Framework for Medicinal Plant Leaf Analysis and Disease Detection

Abstract

Medicinal plants play a significant role in both conventional and modern medical procedures, it is imperative that their leaves be accurately observed. Examining medicinal plant leaves by hand is time-consuming and prone to human error, especially when trying to spot early signs of illness.  This research suggests using a transfer learning-based deep learning approach based on a pre-trained ResNet50 model to analyze photos of medicinal plant leaves in order to solve this issue. The experimental evaluation of this work was conducted using a standard dataset of leaf images from medicinal plants. Three classes representing healthy leaves and two main illness states were created from the dataset. All the images were scaled prior to the application of any preprocessing technique like data augmentation and normalization. The final classification layers were used to train the classification layers for a three-class classification problem with the convolutional part of the pre-trained ResNet50 network. The transfer learning approach based on feature extraction can prevent overfitting and improve the efficiency of learning for datasets of moderate sizes. Experimental findings demonstrate that the developed framework effectively captures meaningful feature representations and achieves reliable classification performance for medicinal plant analysis. The work has verified that the transfer learning approach using ResNet50 is a robust and efficient method for the automatic medicinal plant leaf monitoring and intelligent plant health evaluation.