Automatic Classification of Medicinal Plants Using State-Of-The-Art Pre-Trained Neural Networks

Abstract

Now a days every mankind is suffering due to infections. Ayurveda, the science of life helped to take preventive measures which boost our immunity. It is plant-based science. Many medicinal plants found useful in daily life of common people for boosting immunity. Identifying the plant species having medicinal plant is challenging, it requires botanical expert. In the process of manual identification, botanical experts use various plant features as the identification keys, which are examined adaptively and progressively to identify plant species. The shortage of experts and trained taxonomist created global taxonomic impediment problem which is one of the major challenges. Various researchers have worked in the field of automatic classification of plants since the last decade. The leaf is considered as primary input as it is available throughout the whole year. The research paper mainly focuses on the study of transfer learning approach for medicinal plant classification, which reuse already developed model at the starting point for model on a second task. Transfer learning approach is a black box approach used for image classification and many more applications by extracting features from an image. Some of the transfer learning models are MobileNet-V1, VGG-19, ResNet-50, VGG-16. Here it uses Mendeley dataset of Indian medicinal plant species which is freely available. Output layer classifies the species of leaves. The result provides evaluation and variations of above listed features extracted models. MobileNetV1 achieves maximum accuracy of 98%.