International Journal of Computing and Digital Systems | 2026
Authors: Chowdhury N.S.; Limon M.S.I.K.; Rafee M.M.S.
DOI: 10.12785/ijcds/1571115120
Journal: International Journal of Computing and Digital Systems
Year: 2026
Publisher: University of Bahrain
Document Type: Article
Open Access: All Open Access; Gold Open Access
Cited by: 0
Proper medicinal plant classification and identification are essential for biodiversity preservation, healthcare, and pharmaceutical research. Accurate identification ensures the safe use of herbal remedies, prevents the depletion of endangered species, and supports the discovery of novel bioactive compounds for drug development. Many plant species have similar morphologies; traditional identification techniques can be labor-intensive, require a great deal of expertise, and result in mistakes. To address these issues, our work uses deep learning and transfer learning approaches to classify medicinal plants in Bangladesh.We analyze the performance of six pre-trained convolutional neural networks (CNNs)—VGG16, VGG19, ResNet101, DenseNet121, DenseNet201, and MobileNetV2—on the ‘BDMediLeaves’ dataset, which includes 2,029 original and 38,606 augmented images of ten medicinal plant species captured under natural lighting conditions. All images were resized to 224×224 pixels and normalized to the range [0,1] before being split into 70% training, 20% validation, and 10% testing sets. The models are assessed based on classification accuracy, feature extraction efficiency, and computational performance. Among the models evaluated, ResNet101 achieves the highest accuracy of 99.07%, demonstrating superior feature extraction and generalization capabilities. Other models showed efficient learning curves but converged more slowly, while MobileNetV2 had the lowest accuracy due to its lightweight architecture. Our study highlights how deep learning and transfer learning may be used to create effective and scalable plant identification systems. Through the utilization of superior datasets and computational methods, our methodology bridges the gap between contemporary artificial intelligence (AI) applications and conventional botanical knowledge. The findings have significant consequences for pharmaceutical research, medicinal plant conservation, and healthcare applications, especially in areas where traditional medicine is frequently used. © 2026, University of Bahrain. All rights reserved.
convolutional neural networks; Deep learning; medicinal plants; plant identification; transfer learning