Medicinal Plant Classification using Convolutional Neural Network and Transfer Learning

Abstract

Medicinal plants are not only an essential source of therapeutic compounds but also an alternative source of medications used by most people around the world. Due to recent advances in computer vision, plant identification from images has become a rapidly developing research field. Various results showed good accuracy, precision, and real-world applications. This paper aimed to investigate an accurate and precise automated identification of medicinal plants. We present a new medicinal plant dataset containing images of ten (10) classes of medicinal plant species and one (1) class containing a mixed variety of weeds, vines, and non-medicinal plants. Then we proposed a model based on MobileNetV3 architecture for a low-cost, reliable, and efficient medicinal plant classification. Using the proposed model and Transfer Learning, results revealed a 97.43% accuracy on the challenging task. Overall, the findings revealed the feasibility of an efficient and reliable medicinal plant classifier for real-world applications.