🌱 SIGNET

This study presents a signal-to-image classification approach for analyzing plant bioelectrical responses under varying irrigation conditions. The proposed framework, SIGNET aims to enhance physiological interpretation and multi-class prediction performance in controlled environment agriculture.

✨ Features

• Modular Architecture: Three independent Python modules
• Signal Preprocessing: Automated outlier detection and cleaning
• Multi-Modal Encoding: MTF, GAF, RP time-series to image conversion
• Deep Learning: Multi-modal CNN based backbone
• Easy Execution: Single command pipeline with built-in configuration

🚀 Quick Start

Installation

pip install torch torchvision pandas numpy matplotlib scikit-learn Pillow tqdm scipy
git clone https://github.com//ISW-LAB/SIGNET.git
cd SIGNET

Usage

1. Update data paths in main_pipeline.py:

SIGNAL_FILE_PATHS = [
    "data/morning.csv",
    "data/noon.csv", 
    "data/evening.csv"
]

2. Run pipeline:

python main_pipeline.py

📁 Files

signet/
├── signal_preprocessing.py      # Signal cleaning
├── signal_to_image.py          # MTF/GAF/RP conversion
├── deep_learning_model.py      # Multi-modal CNN
├── main_pipeline.py           # Main execution
└── quick_start.py             # Simple test run

⚙️ Key Features

• Signal Preprocessing: Outlier detection and cleaning
• Image Encoding: MTF, GAF, RP time-series to image conversion
• Multi-Modal CNN: 6 modalities → CNN_based backbone → 3-class output
• One-Command Execution: Complete pipeline in single run

📊 Data Format

CSV files should contain:

irrigated_1,irrigated_2,irrigated_3
2.45,2.37,2.41
2.46,2.38,2.42
...

🔧 Configuration

Edit SIGNETConfig in main_pipeline.py:

PREPROCESSING = {
    'initial_trim': 1000,        # Remove first N samples
    'target_samples': 25000,     # Signal length
    'iqr_multiplier': 2.0        # Outlier threshold
}

IMAGE_CONVERSION = {
    'window_size': 64,           # Window size
    'stride': 32,                # Window step
    'save_format': SaveFormat.PNG
}

TRAINING = {
    'num_epochs': 30,
    'batch_size': 32,
    'learning_rate': 0.001
}

📈 Architecture

Raw Signals → Preprocessing → MTF/GAF/RP Images → Multi-Modal CNN → 3 Classes

• Input: 6 modalities (MTF, GAF, RP × original/scaled)
• Backbone: Custom CNN based Backbone
• Output: irrigated_1, irrigated_2, irrigated_3

🛠️ Commands

python main_pipeline.py                    # Complete pipeline
python main_pipeline.py --step preprocessing  # Individual step
python main_pipeline.py --config           # View settings
python quick_start.py                      # Quick test

📁 Output

signet_output/
├── processed_signals/    # Cleaned CSV files
├── encoded_images/       # MTF/GAF/RP images
└── trained_models/       # Model checkpoints

🔧 Individual Modules

# Signal preprocessing only
from signal_preprocessing import preprocess_signals

# Image conversion only  
from signal_to_image import convert_signals_to_images

# Model training only
from deep_learning_model import train_signet_model

🧪 Quick Test

For testing with smaller data:

# Edit in main_pipeline.py or quick_start.py
PREPROCESSING['target_samples'] = 5000
IMAGE_CONVERSION['window_size'] = 32  
TRAINING['num_epochs'] = 10

📞 Contact Us

If you have any questions or provide your cell images, please contact us by email

• Hongseok Oh: hs.oh-isw@cbnu.ac.kr
• Yeongyu Lee: yg.lee-isw@chungbuk.ac.kr