🌿 EdgeSoybeanNet

An Edge-AI Framework for Real-Time, High-Accuracy Soybean Pod Counting

---

🧠 Overview

EdgeSoybeanNet is an AI-powered framework designed for real-time, high-accuracy soybean pod counting. Optimized for edge deployment, it enables efficient field-level analysis without heavy computational resources.

This repository provides:

• Implementation of the EdgeSoybeanNet model
• Pre-trained model weights
• Evaluation and inference scripts
• Configuration examples for deployment

---

📦 Repository Structure

EdgeSoybeanNet/
├── dataset/              # Contains only test images
│   └── ...               # (e.g., sample1.jpg, sample2.jpg)
│
├── json/                 # Contains JSON annotation files for test images only
│   └── ...               # (e.g., sample1.json, sample2.json)
│
├── codes/                # All Python scripts (training, inference, utils, etc.)
│   ├── Pre_processing/   # Codes for the preprocessing stage (Stage 1 of the framework)
│   │   └── ...   
│   ├── UNetLite.py       # Lightweight UNet model definition and visualization
│   ├── engine.py         # Core training loop and evaluation functions
│
├── results/              # Sample outputs from each stage of the framework
│   └── ...    
│
├── trained_models/       # Pre-trained model weights (.pth files)
│   └── ...
│
└── requirements.txt      # Python dependencies

Note: The json/ directory includes only the JSON annotation files for the test dataset used in evaluation. For access to the complete set of JSON files (training, validation, and testing) used in this research, please refer to the EdgeSoybeanNet Research Paper for dataset access links and preparation instructions.

---

🧰 Installation

1. Clone the repository

git clone https://github.com/Goecol/EdgeSoybeanNet.git
cd EdgeSoybeanNet

2. Create and activate a virtual environment

python3 -m venv env
source env/bin/activate    # On macOS/Linux
env\Scripts\activate       # On Windows

3. Install dependencies

pip install -r requirements.txt

---

📂 Dataset Information

The repository includes only the test dataset for demonstration. The training and validation datasets are large and not uploaded here. Please refer to the EdgeSoybeanNet Research Paper for dataset access links and preparation instructions.

---

🚀 Running EdgeSoybeanNet

You can run the EdgeSoybeanNet model directly from the main model file:

python UNetLite.py

This script automatically imports the engine module and executes the model using:

import engine
engine.run(UNetLite)

Inside the run() function of engine.py, you can specify whether to train, test, or evaluate the model by adjusting the logic in that function. This gives you full flexibility to switch between different modes or datasets.

(Optionally, you can rename engine.py to edge_runner.py later for better clarity when managing multiple models.)

---

⚙️ Edge Deployment

The framework is optimized for low-power edge devices such as:

• Raspberry Pi (with Coral Edge TPU): Raspberry PI 5
• NVIDIA Jetson Nano / Xavier
• Mobile devices using TFLite

Deployment configurations and scripts will be added soon.

---

🔄 Future Updates

More procedures, training instructions, and optimization details for EdgeAI deployment will be added soon. Stay tuned for updates on:

• Full dataset preparation guide
• Model quantization and pruning scripts
• Edge deployment benchmarks

---

📜 Citation

If you use this work, please cite the EdgeSoybeanNet research paper.

---

🔗 Repository

GitHub: https://github.com/Goecol/EdgeSoybeanNet