Herbal Plant Identification Using Deep Learning

Abstract

From traditional medicine to today’s research in pharmacology, herbal plants are seen as very important. Yet, correctly identifying herbal species is challenging since many species share the same features and must be classified by experienced taxonomists. Technological advances such as deep learning have provided a way to automate this work with improved accuracy. The proposed system identifies herbal plants by analyzing their images using Convolution Neural Networks (CNNs), which are known for being effective in computer vision. To ensure the dataset is strong, I used thousands of clear leaf pictures from various herbal plant species that were taken in many environmental settings. Before training, the images were processed in stages by normalizing them, creating variations, and separating important objects. To find the most suitable CNN, VGG16, ResNet50, and MobileNetV2 were assessed based on their accuracy, how efficient they are, and whether they could be used on mobile phones. By using transfer learning, the model could take advantage of previously trained models on huge image collections.