CNN-based Leaf Image Classification for Bangladeshi Medicinal Plant Recognition

Classifying plant species has taken much attention in the research area to help people recognizing plants easily. In recent years, the convolutional neural networks (CNN) have achieved tremendous computer vision results, especially in image classification. Usually, humans find it difficult to recognize proper medicinal plants. It requires the intuition of an expert botanist, which is a time consuming manual task. In this research, we proposed an automated system for the medicinal plant classification, which will help people identify useful plant species quickly. A new dataset of 10 medicinal plants of Bangladesh is introduced, collected from different regions across the country, and some state-of-the images collected from different sources. After that, a three-layer convolutional neural network is employed to extract the high-level features for the classification trained with the data augmentation technique. The training process was done on 34123 images, and the experimental result on another 3570 images proved that this method is quite feasible and effective, which gave by a 71.3% accuracy rate.