Deep Learning Based Medicinal Plant Identification for Enhanced Botanical Conservation

Plants used in medicine are an essential part of the human health system with various natural medicines and health properties. The right identification of medicinal plants will support the conservation of these natural resources and enhancement of these traditional medical practices. Medicinal plants can now be identified and classified more precisely and reliably by use of leaf and plant pictures using the technology of artificial intelligence and machine learning, especially deep learning. We used Convolutional Neural Networks (CNNs) deep learning models with transfer learning VGG11, ResNet34, and DenseNet121. The novelty of our study is that we combine DenseNet121 with the Multi-Trend Binary Code (MTBC) feature descriptor to perform better and extend features representation. These models have been tested on two benchmark datasets, which include the Indonesian Medicinal Plants Dataset as well as the Indonesian Herb Leaf Dataset. Although all CNN models performed well in terms of accuracy, the proposed hybrid model, DenseNet121+MTBC, performed better than the remaining, attaining its best accuracy of 94.51%, and offering better precision, recall, and F1-score metrics. The results note the usefulness of the combination of the traditional texture descriptors and deep learning features, thus, the synergistic trait of the hybrid approach. The hand-crafted features combined with DenseNet121 give a more effective solution to the repetitive phenomenon of medicinal plant identification than just any CNN. The method offers a convenient and efficient method of alternative relying on conventional methods of identification, offering proficient, exact, and advantageous rapid medication identification of plants.