Medi-Plant: A Deep Learning Approach for Medicinal Plant Classification with Pix2Pix Generative Adversarial Network

Abstract Background Biodiversity conservation is crucial due to the risk of extinction faced by many plant species. Traditional medicinal systems heavily rely on the diverse array of plants, offering an alternative to manufactured medications and promoting healthy living. Despite the significance of these plants, datasets for therapeutic herbs are not readily available. To address this gap, this study proposes an automatic system for identifying medicinal plants based on computer vision and deep learning techniques, utilizing various neural network approaches. Results The study introduces the Medi-plant dataset, comprising 6000 leaf images from 50 different Indian plant species. To validate the dataset, pre-trained deep convolutional neural network architectures including MobileNetV2, ResNet-50, and Xception were employed. The proposed Medi-Plant model, leveraging all three architectures, achieved an impressive accuracy of 97.96%. Conclusions The findings demonstrate the effectiveness of the Medi-plant dataset and the proposed Medi-Plant model in accurately identifying medicinal plants. Additionally, a cross-platform application named Medi-Plant Identification was developed, capable of swiftly identifying herb images and providing pertinent information from the database. By continuing to expand the dataset, this research aims to benefit stakeholders and society at large by fostering awareness and understanding of herbs and their therapeutic properties.