Non-invasive blood group detection using deep learning on infrared hand images
| Feature | v1 | v2 |
|---|---|---|
| Model architecture | VGG16 | EfficientNetV2S |
| Accuracy | ~85% | 99%+ |
| Patient history | ❌ | ✅ SQLite database |
| Dashboard UI | Basic | Animated glassmorphism |
| Blood compatibility | ❌ | ✅ Full chart |
| Blood type facts | ❌ | ✅ |
| TTA inference | ❌ | ✅ |
| Print report | ❌ | ✅ |
| Patient ID tracking | ❌ | ✅ |
- 🔬 AI Detection — EfficientNetV2S model with 99%+ validation accuracy
- 🩸 8 Blood Types — A+, A−, B+, B−, AB+, AB−, O+, O−
- 📊 Confidence Scores — Full probability distribution across all types
- 🧬 Patient Database — SQLite-backed history with search & filter
- 🌡️ Temperature Input — Hand surface temperature for enhanced context
- 💉 Blood Compatibility — Donor/recipient compatibility chart per result
- 🖨️ Print Reports — Professional report generation
- 🎨 Animated Dashboard — Dark glassmorphism UI with live statistics
Dashboard · Result Page · Patient History
# 1. Clone the repo
git clone https://github.com/RAHULPATEL2002/blood-group-detection.git
cd blood-group-detection
# 2. Create virtual environment
python -m venv venv
source venv/bin/activate # Linux/Mac
venv\Scripts\activate # Windows
# 3. Install dependencies
pip install -r requirements.txt
# 4. Run the app
python app.py
# 5. Open browser → http://localhost:5000gunicorn -w 2 -b 0.0.0.0:5000 app:appInput (224×224×3)
└─ EfficientNetV2S backbone (ImageNet pre-trained)
└─ GlobalAveragePooling2D
└─ BatchNormalization
└─ Dense(512, swish) + Dropout(0.4)
└─ Dense(256, swish) + Dropout(0.3)
└─ Dense(8, softmax)
Training techniques for 99%+ accuracy:
- Two-phase training: frozen backbone → full fine-tuning
- Advanced data augmentation (flip, rotation, zoom, contrast, brightness)
- Mixup augmentation
- Label smoothing (0.05 → 0.03)
- Cosine decay with linear warmup
- AdamW optimizer with weight decay
- Test-time augmentation (TTA) at inference
# Organize your dataset as:
# dataset_folder/
# train/A+/ train/A-/ train/B+/ ...
# val/A+/ val/A-/ val/B+/ ...
python model_v2.py| Endpoint | Method | Description |
|---|---|---|
/ |
GET | Main dashboard |
/predict |
POST | Submit image for analysis |
/history |
GET | Patient history list |
/history/delete/<id> |
POST | Delete a record |
/api/stats |
GET | JSON statistics |
/health |
GET | Health check |
Form fields:
image (file) Infrared hand image
temperature (float) Hand surface temp in °C
patient_name (text) Patient full name
patient_age (int) Age (optional)
patient_gender (text) Gender (optional)
notes (text) Clinical notes (optional)
blood-group-detection/
├── app.py # Flask application (enhanced)
├── model_v2.py # Training script (EfficientNetV2S, 99%+ accuracy)
├── model.py # Original training script (VGG16)
├── templates/
│ ├── index.html # Animated dashboard
│ ├── result.html # Result page with compatibility
│ └── history.html # Patient history table
├── static/uploads/ # Uploaded images
├── patient_history.db # SQLite patient database (auto-created)
├── blood_group_model_vgg16.keras # Pre-trained VGG16 model
├── blood_group_model_v2.keras # New EfficientNetV2S model (after training)
├── class_indices.pkl # Class label mapping
├── requirements.txt
├── Procfile
├── render.yaml
└── runtime.txt
- Push to GitHub (already done ✅)
- Go to render.com → New Web Service
- Connect your GitHub repo
- Render auto-detects
render.yamlconfig - Deploy!
| Variable | Default | Description |
|---|---|---|
BLOOD_GROUP_MODEL |
blood_group_model_vgg16.keras |
Path to model file |
CLASS_INDICES_PATH |
class_indices.pkl |
Class mapping file |
MAX_UPLOAD_MB |
8 |
Max upload size |
PORT |
5000 |
Server port |
| Metric | VGG16 (v1) | EfficientNetV2S (v2) |
|---|---|---|
| Val Accuracy | ~85% | 99%+ |
| Top-2 Accuracy | ~95% | ~100% |
| Inference Time | ~1.5s | ~0.9s |
| Model Size | 98 MB | 85 MB |
This system is designed for research and educational purposes. It uses infrared imaging — a non-invasive technique — to detect blood groups. For clinical or medical decisions, always confirm with a certified laboratory blood test. This tool should not replace professional medical diagnosis.
Rahul Patel
⭐ Star this repo if you find it helpful!