A4Aerospace

💎 Idea: MAVEX Maximus    – Landmine Detection and Deactivation Solution

Introduction:

Landmines in post-conflict regions are silent threats that claim lives, hinder development, and damage ecosystems. Traditional detection methods are often slow, hazardous, and unsuitable for rugged terrains.

MAVEX Maximus introduces a dual-drone system combining:

• MAVEX Mothership – The central operational hub.
• MAVEX Model-X – A compact detection drone for precision demining.

This innovative solution leverages European Space technologies for safer, faster, and sustainable demining operations.

Problem and Solution:

Problem:

• Millions of undetected landmines in hazardous terrains (forests, urban ruins).
• Manual demining methods are slow, dangerous, and labor-intensive.

Solution:

• MAVEX Mothership: Provides mission support using Copernicus and Galileo for precise navigation and data relay.
• MAVEX Model-X: Detects and deactivates landmines using advanced sensors and autonomous navigation.

🛰️ EU Space Technologies

1. Copernicus Earth Observation Data:

• Use Case: Provides high-resolution imagery to identify potential minefields, analyze terrains, and map vegetation density.
• Value Added:
  • Streamlines mission planning with accurate terrain data.
  • Reduces false-positive detections by pre-mapping target zones.

2. Galileo GNSS Navigation:

• Use Case: Precision navigation and positioning in GPS-denied environments.
• Value Added:
  • Ensures accurate payload deployment for deactivation.
  • Improves autonomous drone operations in complex terrains.

🔒 EU Space for Defence and Security

Challenge Addressed:

• "Landmine Detection and Deactivation"

Contribution to EU Defence and Security:

1. Minimizing Human Risk: Autonomous drones keep operators safe.
2. Efficiency: Autonomous systems cover inaccessible areas quickly.
3. Scalability: Modular systems adapt to various terrains and missions.

🤼 Meet the Team

1. Vasanth Muruganantham – Drone Systems Specialist and Project Lead

• Expert in UAV design and payload optimization.
• Recognized as a 2024 Otto Lilienthal Award Winner.
• Researcher in Space Engineering, TU Berlin.

2. Nithyashree Karunakar – System Engineering and AI Specialist

• Expert in AI-driven anomaly detection and system engineering.
• Winner of Best Performance Award, ERC24 Droning Task.
• Master’s Candidate in AI, IU Berlin.

3. Salman Shariff – Operations Lead

• Experienced in business development and outreach.
• Administrator at Citizens Group of Institutions.

4. Christian Janke – Advisor

• 14+ years in UAV research and development.
• Faculty at Embry-Riddle Aeronautical University.

🚁 System Details: MAVEX Mothership and MAVEX Model-X

1. MAVEX Mothership:

• Role: A long-endurance drone acting as the operational hub.
• Capabilities:
  • Deploys MAVEX Model-X to precise locations.
  • Relays satellite, drone, and ground data.
  • Stores additional payloads for extended missions.

2. MAVEX Model-X:

• Role: A compact, spherical drone for close-proximity detection and deactivation.
• Capabilities:
  • Detects landmines using GPR, multi-spectral imaging, and infrared sensors.
  • Deploys foam hardeners and cyanoacrylate adhesives for contactless deactivation.
  • Operates autonomously in confined environments.

🔬 Payload and Technology Integration

Detection Sensors:

• Ground Penetrating Radar (GPR): Identifies buried anomalies.
• Magnetic Gradiometer: Detects metallic landmines.
• Infrared Sensors: Identifies heat signatures and anomalies.

Deactivation Mechanisms:

• Foam Technology: Expanding foam hardens on triggers to neutralize them.
• Cyanoacrylate Adhesives: Aerosol glue blocks trigger activation.

📊 Impact and Benefits

1. Safety:
   • Autonomous systems ensure zero human risk in hazardous zones.
2. Efficiency:
   • Covers larger areas faster than manual methods.
   • Reduces false positives with multi-sensor verification.
3. Adaptability:
   • Operates in varied terrains, including dense forests and arid regions.
   • Modular payloads accommodate diverse missions.
4. Sustainability:
   • Leaves no ecological footprint.
   • Integrates findings into EU demining efforts.

🛠️ Future Plans and Developments

1. Swarm Technology:

• Coordinated swarm operations to cover large areas and share real-time data.
• Redundancy and efficiency: If one drone fails, others take over its tasks..

2. AI-Powered Threat Analysis:

• Uses historical and terrain-specific data from Copernicus.
• Guides MAVEX drones to high-risk areas, minimizing time and effort.

3. MAVEX-X Modular Upgrades:

• Chemical Sniffer Sensors: Detect explosive residues in the air.
• Environmental Data Collection: Improves terrain and weather planning.

4. MAVEX-Diffuser for Bomb Disposal:

• Robotic manipulators for neutralizing IEDs.
• Remote detonation systems for larger threats.

5. Advanced Camouflage for MAVEX Model-X:

• Inspired by BAE Systems' Active Camouflage Technology.
• Features:
  • Adaptive surfaces change color and texture for stealth.
  • Infrared suppression and radar-absorbing materials.
  • Holographic projection for blending into surroundings

🌍 Vision for the Future

MAVEX drones aim to revolutionize humanitarian and defense missions. Combining state-of-the-art technologies like camouflage, swarming, and AI analytics, MAVEX will redefine safety, efficiency, and sustainability in global demining and disaster management efforts.

🚀 Why MAVEX Maximus?

1. Autonomous Flight in Confined Spaces: Ideal for dense forests and urban ruins.
2. Satellite Integration: Uses Copernicus and Galileo for unparalleled precision.
3. Safety First: Non-contact deactivation ensures risk-free operations.
4. Scalable Innovation: Modular design adapts to missions of varying complexity.
5. Eco-Friendly Operations: Leaves zero ecological footprint.

MAVEX Maximus is not just a drone – it’s a game-changing solution to the global landmine crisis.

    CODING SECTION:

"Landmine Detection and Classification Using Convolutional Neural Networks (CNN) with Augmented GPR Data"

import numpy as np
import tensorflow as tf
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, confusion_matrix
import matplotlib.pyplot as plt


# Step 1: Load the .npy file
file_path = "20170621_deg0_HHVV.npy"  # Update the file path if necessary
data_dict = np.load(file_path, allow_pickle=True).item()
data = data_dict['data']  # 3D GPR data
ground_truth = data_dict['ground_truth']  # Ground truth labels


print("Original Data Shape:", data.shape)
print("Ground Truth Shape:", ground_truth.shape)


# Augment the dataset
def augment_dataset(data):
    augmented_data = []
    for i in range(data.shape[0]):  # Iterate through slices
        augmented_data.append(data[i])  # Original
        augmented_data.append(data[i] + np.random.normal(0, 0.05, data[i].shape))  # Noisy
    return np.array(augmented_data)


augmented_data = augment_dataset(data)
augmented_labels = np.repeat(ground_truth, 2)  # Repeat labels for augmented data


print("Augmented Data Shape:", augmented_data.shape)
print("Augmented Labels Shape:", augmented_labels.shape)


# Step 2: Preprocess data for CNN
# Normalize data
augmented_data = (augmented_data - np.min(augmented_data)) / (np.max(augmented_data) - np.min(augmented_data))
# Reshape data for CNN: (samples, height, width, channels)
augmented_data = augmented_data[..., np.newaxis]


# Train-test split
X_train, X_test, y_train, y_test = train_test_split(augmented_data, augmented_labels, test_size=0.3, random_state=42)


# Step 3: Define the CNN Model
model = tf.keras.Sequential([
    tf.keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=(170, 440, 1)),
    tf.keras.layers.MaxPooling2D((2, 2)),
    tf.keras.layers.Conv2D(64, (3, 3), activation='relu'),
    tf.keras.layers.MaxPooling2D((2, 2)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(128, activation='relu'),
    tf.keras.layers.Dense(1, activation='sigmoid')  # Binary classification
])


# Step 4: Compile the Model
model.compile(optimizer='adam',
              loss='binary_crossentropy',
              metrics=['accuracy'])


# Step 5: Train the Model
history = model.fit(X_train, y_train, epochs=10, validation_split=0.2, batch_size=16)


# Step 6: Evaluate the Model
test_loss, test_accuracy = model.evaluate(X_test, y_test)
print(f"Test Accuracy: {test_accuracy:.2f}")


# Step 7: Generate Classification Report and Confusion Matrix
y_pred_probs = model.predict(X_test)
y_pred = (y_pred_probs > 0.5).astype(int).flatten()
print("Classification Report:\n", classification_report(y_test, y_pred))
print("Confusion Matrix:\n", confusion_matrix(y_test, y_pred))


# Step 8: Visualize Training History
plt.figure(figsize=(12, 6))
plt.plot(history.history['accuracy'], label='Training Accuracy')
plt.plot(history.history['val_accuracy'], label='Validation Accuracy')
plt.title("Model Accuracy")
plt.xlabel("Epochs")
plt.ylabel("Accuracy")
plt.legend()
plt.show()


plt.figure(figsize=(12, 6))
plt.plot(history.history['loss'], label='Training Loss')
plt.plot(history.history['val_loss'], label='Validation Loss')
plt.title("Model Loss")
plt.xlabel("Epochs")
plt.ylabel("Loss")
plt.legend()
plt.show()


# Step 9: Visualize Predictions
test_slice_idx = 10
prediction = y_pred_probs[test_slice_idx][0]
true_label = y_test[test_slice_idx]


plt.figure(figsize=(10, 6))
plt.imshow(X_test[test_slice_idx].squeeze(), aspect='auto', cmap='viridis')
plt.title(f"Prediction: {'Anomaly' if prediction > 0.5 else 'No Anomaly'} (True: {'Anomaly' if true_label == 1 else 'No Anomaly'})")
plt.colorbar()
plt.show()


# Step 10: Ground Truth vs Predictions Visualization
plt.figure(figsize=(10, 4))
plt.plot(y_test, label="True Labels", marker='o', linestyle='-', color='blue')
plt.plot(y_pred, label="Predicted Labels", marker='x', linestyle='--', color='red')
plt.title("Ground Truth vs Predicted Anomalies")
plt.xlabel("Sample Index")
plt.ylabel("Presence of Object (1=Yes, 0=No)")
plt.legend()
plt.grid()
plt.show()

  Model Performance Overview:

• Test Accuracy: 93%, showcasing reliable performance in detecting anomalies.
• Validation Accuracy: Stabilizes at 94.74% by the final epochs, indicating good generalization.
• Loss: Both training and validation losses decrease steadily, with validation loss stabilizing lower than training loss, demonstrating effective learning.

  Classification Metrics:

• Precision:
  • Anomalies (1): 91% (Correctly predicted anomalies out of all anomaly predictions).
• Recall:
  • Anomalies (1): 100% (All actual anomalies were detected).
• F1-Score:
  • Anomalies (1): 95%, reflecting a balance between precision and recall.
• Overall Accuracy: 93%, showing consistent performance across the test dataset.

  Confusion Matrix:

• True Positives: 31 (Correctly detected anomalies).
• True Negatives: 6 (Correctly detected non-anomalies).
• False Positives: 0 (No incorrect anomaly detections).
• False Negatives: 3 (Missed anomalies).

  Visualizations:

1. Training and Validation Metrics:

   • Accuracy Graph: Training accuracy gradually increases; validation accuracy stabilizes at 94.74%.
   • Loss Graph: Both training and validation losses drop significantly, stabilizing as the model learns.

2. GPR Data Visualization:

   • Test slice correctly classified as "Anomaly" with a matching true label, illustrating model reliability.

3. Ground Truth vs Predictions:

   • Close alignment between true labels and predicted labels across samples, showcasing high recall and precision with minimal deviations.

  Key Insights:

• The model demonstrates high recall (100%) for anomalies, ensuring no anomalies are missed, with strong precision (91%) minimizing false alarms.
• Improved test accuracy and metrics compared to previous runs, likely due to enhanced augmentation or hyperparameter tuning.
• Slight differences in training and validation metrics suggest good generalization with no significant overfitting concerns.