Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Local Herb Identification Using Transfer Learning: A CNN-Powered Mobile Application for Nepalese Flora

Prajwal Thapa, Mridul Sharma, Jinu Nyachhyon, Yagya Raj Pandeya

TL;DR

The study tackles the challenge of identifying 60 Nepalese herb species from images using transfer learning with CNNs, aiming for real-time, on-device classification on smartphones. It compares six architectures (DenseNet121, ResNet50, VGG16, InceptionV3, EfficientNetV2, ViT) and finds DenseNet121 superior after leveraging ImageNet and plant-dataset pretraining, with 12,000 labeled images split into train/validation/test (9000/1500/1500). The method integrates data augmentation and regularization to combat overfitting, and deploys the model via a Flutter app with TensorFlow Lite for offline mobile inference, enabling users to identify herbs and access related information. This work advances accessible, accurate herb recognition in resource-limited Nepal, supporting education, conservation, and sustainable utilization of flora.

Abstract

Herb classification presents a critical challenge in botanical research, particularly in regions with rich biodiversity such as Nepal. This study introduces a novel deep learning approach for classifying 60 different herb species using Convolutional Neural Networks (CNNs) and transfer learning techniques. Using a manually curated dataset of 12,000 herb images, we developed a robust machine learning model that addresses existing limitations in herb recognition methodologies. Our research employed multiple model architectures, including DenseNet121, 50-layer Residual Network (ResNet50), 16-layer Visual Geometry Group Network (VGG16), InceptionV3, EfficientNetV2, and Vision Transformer (VIT), with DenseNet121 ultimately demonstrating superior performance. Data augmentation and regularization techniques were applied to mitigate overfitting and enhance the generalizability of the model. This work advances herb classification techniques, preserving traditional botanical knowledge and promoting sustainable herb utilization.

Local Herb Identification Using Transfer Learning: A CNN-Powered Mobile Application for Nepalese Flora

TL;DR

The study tackles the challenge of identifying 60 Nepalese herb species from images using transfer learning with CNNs, aiming for real-time, on-device classification on smartphones. It compares six architectures (DenseNet121, ResNet50, VGG16, InceptionV3, EfficientNetV2, ViT) and finds DenseNet121 superior after leveraging ImageNet and plant-dataset pretraining, with 12,000 labeled images split into train/validation/test (9000/1500/1500). The method integrates data augmentation and regularization to combat overfitting, and deploys the model via a Flutter app with TensorFlow Lite for offline mobile inference, enabling users to identify herbs and access related information. This work advances accessible, accurate herb recognition in resource-limited Nepal, supporting education, conservation, and sustainable utilization of flora.

Abstract

Herb classification presents a critical challenge in botanical research, particularly in regions with rich biodiversity such as Nepal. This study introduces a novel deep learning approach for classifying 60 different herb species using Convolutional Neural Networks (CNNs) and transfer learning techniques. Using a manually curated dataset of 12,000 herb images, we developed a robust machine learning model that addresses existing limitations in herb recognition methodologies. Our research employed multiple model architectures, including DenseNet121, 50-layer Residual Network (ResNet50), 16-layer Visual Geometry Group Network (VGG16), InceptionV3, EfficientNetV2, and Vision Transformer (VIT), with DenseNet121 ultimately demonstrating superior performance. Data augmentation and regularization techniques were applied to mitigate overfitting and enhance the generalizability of the model. This work advances herb classification techniques, preserving traditional botanical knowledge and promoting sustainable herb utilization.
Paper Structure (18 sections, 6 figures, 5 tables)

This paper contains 18 sections, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Data
  4. Data Collection
  5. Data Cleaning and Preprocessing
  6. Data Labeling
  7. Datasets Creation
  8. Transfer Learning
  9. Regularization
  10. Experimental Setup and Evaluation
  11. Network Architecture
  12. Evaluation
  13. Results
  14. Evaluation on DenseNet121
  15. Network Implementation
  16. ...and 3 more sections

Figures (6)

  • Figure 1: Manual and automatic herb recognition
  • Figure 2: Overall architecture of our model used
  • Figure 3: AUC score of the model with each AUC score of 60 different herb species
  • Figure 4: Features learned by our model in each layer
  • Figure 5: Activity diagram of the mobile application
  • ...and 1 more figures