HerbaVisionNet: Optimized CNN And Resnet50v2 Model For Enhanced Medicinal Plant Identification And Application Design

Authors

  • Harshit Goyal Department of Computer Science& Engineering, BV (DU)COE, Pune, Maharashtra, India-411043
  • Bindu Garg Department of Computer Science & Engineering, BV (DU)COE, Pune, Maharashtra, India-411043
  • Ashok Kumar Goyal Department of Mathematics, M.S.J. Government PG College, Bharatpur, Rajasthan, India-321001

DOI:

https://doi.org/10.15379/ijmst.v10i3.3702

Keywords:

Medicinal plants, Plant identification, Deep learning, Convolutional Neural Network(CNN), ResNet50v2, Image processing, Feature extraction, Classification

Abstract

Addressing the escalating demand for medicinal plants, this study undertakes the critical task of establishing an efficient identification system. Employing Convolutional Neural Networks (CNNs) with ResNet50v2 architecture, the project endeavours to develop a robust model capable of classifying 30 distinct medicinal plant types from high-resolution images. This classification relies on the precision of feature extraction to ensure accurate identification. Notably, the dataset used is meticulously curated to ensure its adaptability to the diverse botanical characteristics inherent in medicinal plants. Central to achieving high accuracy is the ResNet50v2 model, which forms the cornerstone of the project. Integrated into a real-time web application implemented with Flask, this model facilitates swift and accessible plant identification for various users, including researchers, herbalists, and enthusiasts. The study underscores the efficacy of deep learning algorithms in medicinal plant identification, showcasing remarkable accuracy and reliability in classification tasks. Beyond mere identification, the model demonstrates an ability to precisely categorize plant species, thereby bridging the gap between identification and exploration. This precision is underscored by an exceptional 99% accuracy rate in distinguishing between 30 medicinal plant species, attributed to meticulous model training with a diverse, high-resolution dataset capturing intricate plant features. Continual efforts are directed towards expanding the dataset to encompass variations in real-world scenarios. Strategies to address class imbalance through oversampling techniques and class weight adjustments are being rigorously pursued. Additionally, the model undergoes refinement through fine-tuning its architecture and integrating advanced data augmentation techniques. The research significantly contributes to the broader field of AI applications in sustainable practices, particularly in the conservation of medicinal plant resources. Serving as a potent tool for accurate identification, it holds promising implications for future advancements in deep learning and botanical research.

Downloads

Download data is not yet available.

Downloads

Published

2023-07-22

How to Cite

[1]
H. . Goyal, B. . Garg, and A. K. . Goyal, “HerbaVisionNet: Optimized CNN And Resnet50v2 Model For Enhanced Medicinal Plant Identification And Application Design”, ijmst, vol. 10, no. 3, pp. 3778-3791, Jul. 2023.