Herbal Species Identification using Deep Learning with Image Processing

Sorting medicinal plant leaves according to their classification can make it easier to link those leaves to certain illnesses and, in turn, help find possible treatments. This paper examines both the front and rear aspects of leaf images along with addition to structural traits in an effort to find the unique, optimal feature combination that increases the recognition rate. To accurately identify medicinal plants, a dataset of scanned pictures of leaves collected from various angles is created. The leaves are categorized based on sizes and shapes. Identification of medicinal plant leaves will have a significant positive impact on the health care pharmaceutical production business. Though there were many methods and techniques developed by various researches using different Machine learning algorithms like KNN and SVM, there is still a need for enhancement of such developed methods. This paper introduces an automated plant detection system utilizing deep convolutional neural network (DCCN), which comprises of image pre-processing, feature extraction, and recognition stages. The classifier built upon our model employs layers such as dropout, convolutional, fully connected layers and max pooling to learn intricate leaf features, that achieved 98.1% accuracy with DCNN and Advanced Xception model, which is hyperparameter tuned, observed with 99.25% accuracy on the Indian medicinal leaf’s dataset. The results are compared to models like Mobilenet, MobilenetV2 and RESNET50 on the dataset comprising of around 2000 different medicinal leaf images by leveraging knowledge from various plant species. The proposed Advanced Xception system achieved remarkable accuracy with minimal losses in predicting the category of leaf images.