Medical Leaf Recognition

This paper presented a medicinal leaf recognition system that couples a systematic image preprocessing pipeline with the MobileNetV2 deep learning architecture and a transfer learning strategy. The system achieved 96 percent classification accuracy across 30 plant species and was packaged into an interactive Streamlit application capable of returning species identity and medicinal metadata from a single leaf photograph. The core contribution is demonstrating that a mobile-friendly architecture, when combined with careful preprocessing and data augmentation, can meet the accuracy threshold needed for practical ethnopharmacological tools — without requiring the computational infrastructure associated with heavier networks such as VGG-16 or ResNet-50. Several directions merit future exploration. Expanding the dataset to cover a broader taxonomic range, incorporating hyperspectral or multispectral imaging for richer feature representation, developing a native mobile application for offline inference, and integrating geographic location data to constrain the prior distribution over species are all promising avenues. In parallel, addressing the safety concerns raised in Section VI — through confidence thresholding, human-in-the-loop verification, and rigorous field testing — will be essential before deployment in healthcare-adjacent contexts.  

[1] World Health Organization, "WHO Traditional Medicine Strategy 2014–2023," WHO Press, Geneva, 2013. [2] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning," Nature, vol. 521, pp. 436–444, May 2015. [3] S. G. Wu, F. S. Bao, E. Y. Xu, Y.-X. Wang, Y.-F. Chang, and Q.-L. Xiang, "A leaf recognition algorithm for plant classification using probabilistic neural network," in Proc. IEEE Int. Symp. Signal Process. Inf. Technol., 2007, pp. 11–16. [4] D. B. Valdez, C. J. G. Aliac, and L. S. Feliscuzo, "Medicinal plant classification using convolutional neural network and transfer learning," in Proc. IEEE Int. Conf. Artificial Intelligence Eng. Technol. (IICAIET), Kota Kinabalu, Malaysia, Nov. 2022. [5] R. Patel and A. Chaware, "Quantizing MobileNet models for classification problem," in Proc. 8th Int. Conf. Computing Sustainable Global Development (INDIACom), 2021, pp. 348–351. [6] Y. Sharrab, D. Al-Fraihat, and M. Tarawneh, "Medicinal plants recognition using deep learning," presented at the Int. Conf. Inf. Technol., Valencia, Spain, Jul. 2023. [7] Kaggle, "Medicinal Plant Dataset – Augmented," [Online]. Available: https://www.kaggle.com/datasets/medicinal-plant-dataset-augmented. [Accessed: 2024]. [8] M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, "MobileNetV2: Inverted residuals and linear bottlenecks," in Proc. IEEE/CVF Conf. Computer Vision Pattern Recognition (CVPR), 2018, pp. 4510–4520. [9] J. R. Janani and A. Gopal, "Identification of selected medicinal plant leaves using image features and ANN," in Proc. IEEE Int. Conf. Advances in Electronic Systems (ISAES), 2018, pp. 238–242. [10] R. Rajini and B. Veena, "Study on identification and classification of medical plants," Int. J. Advances Sci. Eng. Technol., 2019.