README_EN.md

IndustrialDigitDatasetGenerator

IndustrialDigitDatasetGenerator is a tool specifically designed for generating digital image datasets in industrial scenarios. It can extract digits (0-9) from system fonts and generate synthetic datasets suitable for industrial environments through various data augmentation techniques, supporting YOLO format object detection annotations. This tool supports multithreading, has simple dependencies, a modular design, and many adjustable parameters, making it easy to use.

Keywords

Industrial Digital Image, Industrial Instrument Detection, Instrument Automation, Dial Reading Recognition, Industrial Instrument Reading Recognition, Data Augmentation, Object Detection, YOLO Annotation, Font Extraction, Caliper Reading、 Image Classification

Main Features

Font Extraction

• Automatically scan and extract font files from the system (default support for Windows and WSL, modify font directory for other systems)
• Filter out unsuitable special fonts
• Automatically crop and align digit images

Data Augmentation

• Noise Augmentation

  • Gaussian noise
  • Salt and pepper noise
  • Speckle noise
  • Poisson noise

• Image Transformations

  • Random rotation
  • Perspective distortion
  • Aspect ratio adjustment
  • Grayscale adjustment
  • Random occlusion

• Industrial Background

  • Perlin noise generation
  • Real industrial image backgrounds (NEU-DET), can be added manually
  • Random stripes and stains

• Random Pattern Interference Augmentation

  • Circular patterns (solid/hollow)
  • Rectangular patterns (solid/hollow)
  • Triangular patterns (solid/hollow)
  • Hexagonal patterns (solid/hollow)
  • Vertical stripe patterns
  • Horizontal stripe patterns
  • Support for custom pattern colors
  • Adjustable pattern size and quantity
  • Adjustable pattern transparency
  • Random position and rotation angle

• Letter Interference Augmentation

  • Extract letters from font files
  • Random letter interference or annotation (for countering dial units)

Annotation Generation

• Automatically generate YOLO format annotations, including bounding boxes and class information

Installation Requirements

Dependencies

Ensure the following Python libraries are installed:

• numpy
• Pillow
• matplotlib
• tqdm

Install them using the following command:

pip install numpy pillow matplotlib tqdm

Or, use our environment configuration:

conda create -n IDDG_env python=3.10
conda activate IDDG_env
pip install -r requirements.txt

Usage

0. Download Real Industrial Background Dataset

We use the NEU-DET dataset as the real industrial background. Please download it yourself and place it according to the Project Structure.

1. Font Extraction

python font_extractor.py

Extract digit fonts from the system and save them in PNG format.

Tips:

• Manually review the extracted fonts before generating the dataset to avoid unsuitable fonts.
• If you know which font is used in your application scenario, you can modify the font directory in default_config.py to generate the specified font and improve dataset accuracy.

2. Dataset Generation

python font_png_augmentation.py

Generate a digital image dataset with industrial backgrounds using the extracted fonts.

3. Visualization Tools

python visualize_annotations.py

Visualize the generated images and their YOLO format annotations.

python test_noise_pattern.py

Visualize the effect of noise pattern generation.

python test_digit_augmentation.py

Visualize the effect of digit augmentation.

Project Structure

IndustrialDigitDatasetGenerator/
├── NEU-DET/                 # Real industrial background image directory
│   └── IMAGES/             # Image directory
├── font_numbers/           # Extracted font image directory
│   ├── 0/
│   ├── 1/
│   ├── 2/
│   ├── 3/
│   ├── 4/
│   ├── 5/
│   ├── 6/
│   ├── 7/
│   ├── 8/
│   └── 9/                 
├── augmented_dataset/     # Generated dataset directory
├── docs/                  # Documentation directory
│   └── images/           # Documentation image directory
|—— default_config.py     # Default configuration parameters
├── font_extractor.py      # Font extraction tool
├── font_png_augmentation.py # Main program for dataset generation
├── test_digit_augmentation.py # Digit augmentation test tool
├── test_noise_pattern.py    # Noise pattern test tool
├── visualize_annotations.py # Annotation visualization tool
├── requirements.txt       # Dependency list
└── README.md             # Project description document

Contribution Guidelines

Feel free to submit Issues and Pull Requests.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Demonstration

Data Augmentation Effects

Pattern Interference Sample

Examples of different types of random pattern interference

YOLO Annotation Visualization