Classification of Indian Herbal Plants based on powder microscopic images using Transfer Learning

The objective of this paper is to proficiently classify the microscopic images of powder of Indian Herbal plants. Since they hold great importance in medicine industry and their identification is only done by experts for the powdered form, we have eluded the need for an expert by automating the process, yielding decent results. Although, attempts have been made to perform this task but the methodologies used do not provide the results with high accuracy. Inspired from the state-of-the-art deep learning techniques we have performed the classification by fine-tuning 4 pre-trained models provided by the Keras library which have provided with great results on ImageNet dataset. Out of the 4 models used, VGG16 provides the highest accuracy, Precision, Recall and F1 score but is the slowest to train. MobileNet is fastest but is mediocre in other parameters while Xception is 2nd fastest but with lowest accuracy and InceptionV3 with mediocre results.