Classification of Indian Medicinal Flowers using MobileNetV2

Image-based classification of medicinal flowers is a crucial and game-changing work that can help in developing the traditional medical practices and healthcare techniques that have been followed in countries including India and China for centuries. Accurate work can help enhance traditional practic...

Full description

Saved in:
Bibliographic Details
Published in2024 11th International Conference on Computing for Sustainable Global Development (INDIACom) pp. 1512 - 1518
Main Authors Bipin Nair, B J, Arjun, B, Abhishek, S, Abhinav, N M, Madhavan, Varun
Format Conference Proceeding
LanguageEnglish
Published Bharati Vidyapeeth, New Delhi 28.02.2024
Subjects
classification
Deep learning
Flowering plants
Image processing
Medical services
Medicinal flowers
Mobile handsets
MobileNet
Object recognition
Performance evaluation
Reviews
Online AccessGet full text
DOI10.23919/INDIACom61295.2024.10498274

Cover

More Information
Summary:Image-based classification of medicinal flowers is a crucial and game-changing work that can help in developing the traditional medical practices and healthcare techniques that have been followed in countries including India and China for centuries. Accurate work can help enhance traditional practices, making them faster and more efficient. Further studies on this topic can help in drug discovery. On doing a thorough review of the previous works we have found a lack of works regarding the classification of Indian Medicinal flowers and to overcome this, our work presents a robust methodology for classifying Indian medicinal flowers using the MobileNetV2 light-weight network. We have worked with a dataset of 3343 flowers that are divided across 13 classes consisting of images of flowers taken in their natural environment under varying backgrounds, light, and climatic conditions which were then preprocessed to make it fit for our task. The images in the dataset are used with MobileNetV2 to classify each of the images into the individual class of the flower. While doing a comparative study, we inferred the best parameters for optimal results with our dataset and measured its performance using various reference metrics. The highest accuracy that we obtained with MobileNetV2 was 98.23%. It was also observed that the performance of the deep learning model is affected by the distribution and representation of each flowers in the dataset.
DOI:10.23919/INDIACom61295.2024.10498274