Identification of Philippine Therapeutic Leave using Deep Learning

Abstract

Despite the advancements in many chemical medications used to treat diseases, medicinal plants have also been very effective in doing so. The global use of herbal products is increasing due to the growing contribution of science and technology to its ethical and scientific growth. In the Philippines, medicinal plants or herbs are popular in treating illness. Although the traditional identification method is effective and requires the expertise of skilled practitioners, it is time-consuming and prone to mistakes. Studies identifying the therapeutic use of naturally occurring plant chemicals have now been publicly disclosed that automatically detect the species of medicinal plants and leaves using machine vision and deep learning. This led to the first phase of this work, which used the readily available Philippine medicinal plants to identify the leaves automatically. With transfer learning using YOLOv3, this study seeks to identify Philippine medicinal plants in real-time. Only segmented leaves of Basella alba (alugbati), Mentha (mint), Moringa oleifera (malungay), Nerium oleander (adelfa), and Psidium guajava (bayabas) are used in the identification of medicinal leaves. The identification of the images utilized an inference approach in the real-time application based on the extracted features using YOLOv3. The result of the test performed illustrates an optimal outcome in the detection of different medicinal leaves. The optimal model performance results to mAP 98.63%. The test summary demonstrates a high detection accuracy and produces results at a fast speed. Thus, various model inferences using input images, live feed, and video inputs exemplify the model’s effectiveness in detecting the inputs medical leaves classes.