Model Card: Military Aircraft Detection with Vision Transformer (ViT)

Model Information

• Model Name: Military Aircraft Image Detection
• Model Type: Vision Transformer (ViT)

Model Overview

• Purpose: The model is designed for the detection and classification of military aircraft in images.
• Intended Use: Military surveillance, object recognition, and security applications.

Model Training

• Training Data: Dataset of military aircraft images collected from Illia56/Military-Aircraft-Detection.
• Data Preprocessing: Random oversampling for class balance, data augmentation (rotation, flip, sharpness adjustment).
• Model Architecture: Vision Transformer (ViT) for image classification.
• Pre-trained Model: google/vit-base-patch16-224-in21k.

Model Evaluation

• Evaluation Metrics:
  • Accuracy
  • F1 Score
  • Confusion Matrix
• Evaluation Dataset: Split from the original dataset for testing.
• | Class | Precision | Recall | F1-Score | Support | |------------|-----------|--------|----------|---------| | A10 | 0.6716 | 0.7368 | 0.7027 | 247 | | A400M | 0.6217 | 0.6748 | 0.6472 | 246 | | AG600 | 0.4512 | 0.9919 | 0.6203 | 247 | | AV8B | 0.6618 | 0.7287 | 0.6936 | 247 | | B1 | 0.9000 | 0.6194 | 0.7338 | 247 | | B2 | 0.7862 | 0.9231 | 0.8492 | 247 | | B52 | 0.9528 | 0.4089 | 0.5722 | 247 | | Be200 | 0.8333 | 0.8300 | 0.8316 | 247 | | C130 | 0.8600 | 0.1748 | 0.2905 | 246 | | C17 | 0.5556 | 0.0405 | 0.0755 | 247 | | C2 | 0.5845 | 0.8543 | 0.6941 | 247 | | C5 | 0.3776 | 0.7490 | 0.5020 | 247 | | E2 | 0.8447 | 0.9028 | 0.8728 | 247 | | E7 | 0.6000 | 0.9595 | 0.7383 | 247 | | EF2000 | 1.0000 | 0.0364 | 0.0703 | 247 | | F117 | 0.6005 | 0.9433 | 0.7339 | 247 | | F14 | 0.9773 | 0.1741 | 0.2955 | 247 | | F15 | 0.2919 | 0.2186 | 0.2500 | 247 | | F16 | 0.8333 | 0.0203 | 0.0397 | 246 | | F18 | 0.9355 | 0.2348 | 0.3754 | 247 | | F22 | 0.4624 | 0.4980 | 0.4795 | 247 | | F35 | 0.5373 | 0.2915 | 0.3780 | 247 | | F4 | 0.4317 | 0.2429 | 0.3109 | 247 | | J10 | 0.8711 | 0.6842 | 0.7664 | 247 | | J20 | 0.5049 | 0.6301 | 0.5606 | 246 | | JAS39 | 0.4535 | 0.4737 | 0.4634 | 247 | | KC135 | 0.8957 | 0.7683 | 0.8271 | 246 | | MQ9 | 0.7358 | 0.8943 | 0.8073 | 246 | | Mig31 | 0.6080 | 0.4899 | 0.5426 | 247 | | Mirage2000 | 0.3245 | 0.6478 | 0.4324 | 247 | | P3 | 0.9423 | 0.3968 | 0.5584 | 247 | | RQ4 | 0.7166 | 0.8907 | 0.7942 | 247 | | Rafale | 0.3063 | 0.3968 | 0.3457 | 247 | | SR71 | 0.7824 | 0.7571 | 0.7695 | 247 | | Su25 | 1.0000 | 0.3618 | 0.5313 | 246 | | Su34 | 0.5340 | 0.8583 | 0.6584 | 247 | | Su57 | 0.6143 | 0.7317 | 0.6679 | 246 | | Tornado | 0.6883 | 0.2146 | 0.3272 | 247 | | Tu160 | 0.8000 | 0.8421 | 0.8205 | 247 | | Tu95 | 0.8340 | 0.8543 | 0.8440 | 247 | | U2 | 0.9371 | 0.6032 | 0.7340 | 247 | | US2 | 0.7074 | 0.6559 | 0.6807 | 247 | | V22 | 0.7212 | 0.9109 | 0.8050 | 247 | | Vulcan | 0.3343 | 0.8947 | 0.4868 | 247 | | XB70 | 0.6657 | 0.9676 | 0.7888 | 247 | | YF23 | 0.5490 | 0.7967 | 0.6501 | 246 | | Accuracy | | | 0.6082 | 11353 | | Macro Avg | 0.6804 | 0.6082 | 0.5787 | 11353 | | Weighted Avg| 0.6803 | 0.6082 | 0.5787 | 11353 |

Potential Bias

• Bias in Training Data: Possible biases related to the data collection process.
• Limitations: Potential biases due to the nature of the dataset and model architecture.

Ethical Considerations

• Fairness: Address any concerns regarding fairness and potential bias in model predictions.
• Privacy: Describe any privacy considerations related to the model's deployment and use.

Model Usage Guidelines

• Recommended Use Cases: Military surveillance, security applications.
• Limitations: Clearly outline model limitations and potential failure scenarios.
• Legal and Ethical Considerations: Compliance with legal and ethical standards.