🍎 Fruit Recognition ML

High-performance GPU-accelerated machine learning for fruit classification

Features • Quick Start • Documentation • Performance • Contributing

Features

Performance 95%+ accuracy on balanced datasets GPU acceleration with CUDA support Fast inference - 394+ images/second 12-second training on RTX 4090 Technical ML.NET 4.0 with SDCA algorithm Anti-overfitting measures Cross-platform (Windows/Linux) Self-contained executables

Flexible Data 8 image formats supported Recursive folder scanning Any directory structure Automatic class detection Easy to Use No dependencies required JSON configuration Batch processing Professional logging

Quick Start

Option 1: Pre-built Releases

# 1. Download from releases
curl -L https://github.com/luangrezende/fruit-recognition-net-ml/releases/download/v1.0.0/fruit-recognition-training-win-x64.zip

# 2. Extract and run
./fruit-recognition-training    # Train your model
./fruit-recognition-testing     # Make predictions

Option 2: Build from Source

git clone https://github.com/luangrezende/fruit-recognition-net-ml.git
cd fruit-recognition-net-ml/src
dotnet restore
dotnet build --configuration Release

Project Structure

fruit-recognition-net-ml/
├── 📂 src/
│   ├── Fruit.Recognition.MachineLearning.Domain/     # Core ML models
│   ├── Fruit.Recognition.MachineLearning.Training/   # Training app
│   ├── Fruit.Recognition.MachineLearning.Testing/    # Testing app
│   ├── data/
│   │   ├── training/    # Your training images
│   │   └── test/        # Images for testing
│   └── models/       # Trained ML models
└── .github/         # CI/CD workflows

Supported Fruits

Fruit	Status	Notes
🍎 Apples	✅ Built-in	Red, Green, Yellow varieties
🍌 Bananas	✅ Built-in	Various ripeness stages
🍊 Oranges	✅ Built-in	Multiple sizes and types
🍇 Custom	✅ Flexible	Add any fruit by creating folders

Configuration

Training Configuration (click to expand)

{
  "ModelConfiguration": {
    "ImageWidth": 128,
    "ImageHeight": 128,
    "TestFraction": 0.3,
    "ValidationFraction": 0.2,
    "UseGpu": true,
    "FallbackToCpu": false,
    "Architecture": "Anti-Overfitting-SDCA"
  },
  "PathConfiguration": {
    "DatasetPath": "..\\data\\training",
    "ModelOutputPath": "..\\models\\fruit-recognition-model.zip"
  }
}

Testing Configuration (click to expand)

{
  "PathConfiguration": {
    "ModelPath": "..\\models\\fruit-recognition-model.zip",
    "TestImagesPath": "..\\data\\test"
  }
}

Usage

Training a Model

cd src/Fruit.Recognition.MachineLearning.Training
dotnet run

What happens:

1. Scans training directory recursively
2. Loads and validates all images
3. Trains GPU-optimized model
4. Saves model for predictions
5. Shows accuracy metrics

Making Predictions

cd src/Fruit.Recognition.MachineLearning.Testing
dotnet run

What happens:

1. Loads the trained model
2. Processes test images
3. Classifies with confidence scores
4. Displays batch results

Performance

Metric	Value
Accuracy	95.4%
Training Time	~12 seconds
Inference Speed	394+ images/sec
Dataset Size	4,637 images
Hardware	RTX 4090, 32GB RAM

System Requirements

Minimum

• Windows 10/11 x64 or Linux x64
• 4GB RAM
• 1GB free disk space

Recommended (GPU Training)

• NVIDIA RTX 3060+
• 8GB+ RAM
• CUDA 12.x toolkit
• SSD storage

Development

# Restore dependencies
dotnet restore src/Fruit.Recognition.MachineLearning.sln

# Build solution
dotnet build src/Fruit.Recognition.MachineLearning.sln --configuration Release

# Run training
dotnet run --project src/Fruit.Recognition.MachineLearning.Training

# Run testing
dotnet run --project src/Fruit.Recognition.MachineLearning.Testing

Troubleshooting

GPU Issues

# Check CUDA installation
nvcc --version
nvidia-smi

# Enable CPU fallback
"FallbackToCpu": true

Low Accuracy

• Add more training images (100+ per fruit)
• Balance dataset between classes
• Increase image resolution
• Check for mislabeled images

CI/CD Pipeline

This project uses GitHub Actions for:

• Continuous Integration
• Cross-platform builds
• Automated releases
• Documentation updates

Create a Release

git tag v1.0.0
git push origin v1.0.0

License

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

Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Open a Pull Request

Support

---

Built with ❤️ using ML.NET 4.0 and .NET 8

⭐ If this project helped you, please consider giving it a star!