Tree Species Classification Using Deep Learning and Machine Learning

An intelligent system for automated tree species classification using leaf images. This project leverages deep learning (ResNet34) for feature extraction and ensemble methods (StackingClassifier, RandomForest, and ...) for classification.

---

Table of Contents

1. Introduction
2. Features
3. Dataset
4. Methodology
5. Results
6. Usage
7. System Architecture
8. Technologies Used
9. Contributors

---

Introduction

The goal of this project is to simplify and automate the classification of tree species based on leaf images. This system can aid:

• Ecologists in biodiversity monitoring.
• Farmers in tree species identification for better care.
• Tourists in learning about native trees.

---

Features

• Feature Extraction: Pretrained ResNet34 used to extract 512 features from images.
• Ensemble Classification: Combines SVM and Random Forest using a stacking approach.
• Custom Dataset: Includes manually collected and preprocessed images of 28 tree species.
• User Interface: Simple web interface for users to upload leaf images and get predictions.

---

Dataset

• Sources:
  • Manually collected leaf images.
  • Crawled from the web using a custom scraper.
  • Frame extractions from leaf videos.
• Data Stats: ~32,000 images representing over 30 tree species.
• Preprocessing: Images normalized using mean [0.485, 0.456, 0.406] and std [0.229, 0.224, 0.225].

---

Methodology

Steps:

1. Data Collection and Preprocessing:
   • Crawling images from the web.
   • Manual video frame extraction.
   • Dataset splitting into training and testing sets.
2. Feature Extraction:
   • Using ResNet34 pretrained on ImageNet.
3. Classification:
   • Classifier models: SVM, Random Forest, and Naive Bayes.
   • Stacking method for the final classification.
4. Evaluation:
   • Metrics used: Accuracy, Precision, Recall, F1-Score.

---

Results

Model	Accuracy	Precision	Recall	F1-Score
SVM	93.41%	92.5%	91.8%	92.1%
Random Forest	83.61%	84.3%	83.6%	83.9%
StackingClassifier	94.98%	93.5%	92.9%	93.2%

Confusion matrices and detailed metrics are available in the repository.

---

Usage

1. Clone the repository:

   git clone https://github.com/A-Darvish/Tree-Species-Identification.git
   cd Tree-Species-Identification

2. Run the web interface:

   python app.py

3. Access the interface: Open http://127.0.0.1:5000 in your browser.

4. Upload a leaf image and get predictions.

---

System Architecture

Data Collection -> Preprocessing -> Feature Extraction -> Classification -> Web Interface

Key Modules:

• Feature Extraction: ResNet34.
• Classification Models: SVM, Random Forest, StackingClassifier.
• Web Framework: Flask.
• UI: HTML + CSS for a simple user interface.

---

Technologies Used

• Python Libraries: PyTorch, scikit-learn, Flask, NumPy, Pandas, Matplotlib.
• Deep Learning Model: ResNet34 pretrained on ImageNet.
• Web Framework: Flask.

---

Contributors

• Arvand Darvish: Developed the project as part of a BSc thesis.
• Supervisor: Dr. Ahmad Nickabadi.