This repository contains the implementation of an aircraft detection model using YOLO (You Only Look Once). The model is trained on a custom dataset of various aircraft types and tested on aircraft videos to detect and classify aircraft effectively.
The goal of this project is to detect and classify aircraft in video footage using a deep learning-based object detection approach. The YOLO model was trained on a dataset containing images of multiple aircraft types, including Il-76, J10, J15-J16, KJ-2000, and Mig-21.
The YOLO model used for this project is YOLOv10l, which consists of:
- 461 layers
- 25,727,162 parameters
- 126.4 GFLOPs
The model was evaluated on a validation dataset, and the results are as follows:
| Class | Images | Instances | Precision (P) | Recall (R) | mAP@50 | mAP@50-95 |
|---|---|---|---|---|---|---|
| All | 44 | 54 | 0.864 | 0.739 | 0.850 | 0.736 |
| Il-76 | 7 | 7 | 0.754 | 0.714 | 0.786 | 0.786 |
| J10 | 10 | 17 | 0.691 | 0.529 | 0.649 | 0.541 |
| J15-J16 | 9 | 9 | 1.000 | 0.743 | 0.922 | 0.757 |
| KJ-2000 | 10 | 10 | 0.994 | 0.800 | 0.949 | 0.875 |
| Mig-21 | 8 | 11 | 0.882 | 0.909 | 0.942 | 0.723 |
- Preprocessing Time: 0.2ms per image
- Inference Time: 7.0ms per image
- Postprocessing Time: 0.1ms per image
The trained model was tested on various aircraft videos. The following video outputs demonstrate the model's ability to detect and classify aircraft in real-time:
output_videoA.mp4
AircraftDetection.ipynb: Jupyter Notebook containing the model training and testing code.bestMulti.pt: The trained YOLO model weights.- Video outputs of model inference on test data.