App-Based Solutions to Detect Medicinal Plants/Crops using Machine Learning

Abstract

In this study, we present a deep learning-based approach for the classification of Indian medicinal plant leaves using transfer learning with the Xception convolutional neural network architecture. TensorFlow's high-level API is used to preprocess and load the dataset which consists of labeled photos of different species of medicinal leaves. The model employs a pre-trained Xception base (excluding the top layers) with weights loaded manually to accommodate the Kaggle execution environment. A custom classification head is appended to the frozen base model, incorporating dense and dropout layers for feature abstraction and regularization. The dataset is partitioned into training, validation and test sets and the model is trained for 25 epochs using the Adam optimizer and sparse categorical cross-entropy loss. Post-training, the model's performance is evaluated on the test set and its predictive capabilities are demonstrated using a sample input image. Finally, the training history is visualized to assess model convergence and the trained model is serialized for future inference. The proposed pipeline demonstrates the efficacy of transfer learning in the automated classification of medicinal leaf images with potential applications in botanical research and herbal medicine identification.