tadikamallavaibhav/GuardianShieldCombatDetector

Guardian-Shield Combat Impact Detector

Model Description

AI-powered combat impact classification system achieving 95.46% accuracy on 6 impact classes:

• 💥 Blast
• 🔫 Gunshot
• 🎯 Artillery
• 🚗 Vehicle Crash
• 🤕 Fall
• ✅ Normal

Model Architecture

ResNet-inspired Deep Learning Model

• Dilated convolutions (rates: 1, 2, 4)
• Multi-head attention (16 heads)
• Squeeze-and-Excitation blocks
• Dual pooling (Average + Max)
• Parameters: ~8.4M
• Format: TensorFlow Lite (optimized for edge deployment)

Intended Use

Primary Use: Real-time combat impact detection on wearable military devices

Input: 13-channel sensor data (200 timesteps)

• 3-axis accelerometer
• 3-axis gyroscope
• 3-axis magnetometer
• Heart rate, SpO2, breathing rate, temperature

Output:

• Impact classification (6 classes)
• Severity score (0-1)

Training Data

• Size: 60,000 synthetic samples
• Method: Physics-based signal modeling with physiological responses
• Split: 70/15/15 (train/val/test)
• Augmentation: Time shifting, scaling, noise injection

Performance Metrics

Metric	Value
Test Accuracy	95.46%
Validation Accuracy	~95.8%
Inference Time	<10ms (GPU T4)
Model Size	6.64 MB (TFLite)

Usage Example

import tensorflow as tf
import numpy as np

# Load model
interpreter = tf.lite.Interpreter('impact_classifier.tflite')
interpreter.allocate_tensors()

# Load normalization parameters
norm = np.load('norm.npz')
mean, std = norm['mean'], norm['std']

# Prepare sensor data (shape: [1, 200, 13])
sensor_data = your_sensor_reading  # Your 13-channel sensor data
normalized = (sensor_data - mean) / std

# Run inference
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()

interpreter.set_tensor(input_details[0]['index'], normalized)
interpreter.invoke()

# Get predictions
impact_type = interpreter.get_tensor(output_details[0]['index'])
severity = interpreter.get_tensor(output_details[1]['index'])

# Decode results
classes = ['blast', 'gunshot', 'artillery', 'vehicle_crash', 'fall', 'normal']
predicted_class = classes[impact_type.argmax()]
severity_score = severity[0][0]

print(f"Impact: {predicted_class}")
print(f"Severity: {severity_score:.2f}")

Files in This Repository

• impact_classifier.tflite - Optimized TensorFlow Lite model (6.64 MB)
• norm.npz - Normalization parameters (mean and std)
• README.md - This file

Deployment

Supported Platforms:

• ✅ Raspberry Pi (TensorFlow Lite)
• ✅ Android devices (TFLite Android SDK)
• ✅ Embedded systems (ARM Cortex-M7+)
• ✅ Edge TPU devices (Google Coral)

Limitations

⚠️ Important Considerations:

• Trained on synthetic data only
• Requires calibrated sensors (IMU + vitals)
• Performance may vary with real-world sensor noise
• Needs field validation with actual combat scenarios
• Environmental factors (temperature, altitude) may affect accuracy

Training Details

Framework: TensorFlow 2.x
Hardware: Google Colab GPU (T4)
Training Time: ~2-3 hours
Optimizer: Adam (lr=0.001, clipnorm=1.0)
Loss: Categorical Crossentropy (label smoothing=0.1) + MSE
Callbacks: Early stopping, learning rate reduction, model checkpointing

Ethical Considerations

This model is designed for defensive military applications to:

• Improve soldier safety and survivability
• Enable faster medical response
• Reduce combat casualties

Intended Users: Military medical personnel, combat medics, field commanders

Citation

@software{guardian_shield_2025,
  author = {Tadikamalla, Vaibhav},
  title = {Guardian-Shield Combat Impact Detector},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/tadikamallavaibhav/guardian-shield-combat-detector}
}

Related Links

• GitHub Repository: https://github.com/vaibhav-tadikamalla/tids-combat-detection
• Training Notebook: Available in GitHub repo (GUARDIAN_SHIELD_FINAL_TRAINABLE.ipynb)
• Full TIDS System: See GitHub for complete edge device implementation

Contact

Author: Vaibhav Tadikamalla
Project: TIDS - Tactical Impact Detection System

For questions, issues, or collaboration opportunities, please open an issue on GitHub.

License

MIT License - See LICENSE file in GitHub repository