Ashegh-Sad-Warrior/yolo_aerial_detection_

README File Content for Hugging Face

YOLO Aerial Detection Model

This repository contains a YOLO-based model for aerial object detection. The model is trained to detect and classify various objects from aerial images, such as planes, ships, vehicles, and more. Below are the details and visualizations regarding the performance of the model.

Model Description

• Model Name: YOLO Aerial Mine Detection
• Framework: Ultralytics YOLOv11n-obb
• Languages: English, Persian
• Classes Detected:
  • Plane (هواپیما)
  • Ship (کشتی)
  • Storage Tank (مخزن ذخیره)
  • Baseball Diamond (زمین بیسبال)
  • Tennis Court (زمین تنیس)
  • Basketball Court (زمین بسکتبال)
  • Ground Track Field (زمین دو و میدانی)
  • Harbor (بندرگاه)
  • Bridge (پل)
  • Large Vehicle (خودرو بزرگ)
  • Small Vehicle (خودرو کوچک)
  • Helicopter (هلیکوپتر)
  • Roundabout (میدان)
  • Soccer Ball Field (زمین فوتبال)
  • Swimming Pool (استخر شنا)

Training Details

• Dataset: Custom aerial images annotated for object detection.
• Metrics: Precision, Recall, mAP@0.5, F1 Score
• Training Environment: Kaggle, GPU-accelerated environment
• Optimizer: SGD
• Libraries Used:
  • Ultralytics: YOLOv11n-obb (version 8.0.0)
  • Gradio: For creating the user interface (version 3.1.4)
  • Pandas: For data handling (version 1.3.3)
  • Pillow: For image manipulation (version 8.4.0)
  • OpenCV: For video processing (version 4.5.3)

Evaluation Results

Below are the various evaluation results obtained during the training and testing phases of the model.

F1-Confidence Curve

Precision-Confidence Curve

Precision-Recall Curve

Recall-Confidence Curve

Confusion Matrix

Labels Correlogram

Labels Distribution

How to Use

1. Clone this repository.
2. Load the model using the Ultralytics YOLO library.
3. Use the model for object detection on aerial images.

python from ultralytics import YOLO

Load the trained model

yolo_model = YOLO('yolo11n-obb.pt')

Perform detection

yolo_model('test_image.jpg')

License

This model is open-sourced under the MIT License.

Acknowledgements

Special thanks to the Kaggle community and Hugging Face for providing tools and platforms for developing and sharing this project.