Aircraft Classification from Aerial Imagery.
Completed:
.This project tackles the challenge of identifying military aircraft from aerial images using computer vision techniques. We created a labeled aircraft dataset with RoboFlow and applied data augmentation to enhance model performance. Three models were evaluated: YOLOv5, YOLOv8 (by Ultralytics), and a custom ConvNet built with PyTorch. The custom model aimed to balance computational efficiency and detection accuracy, addressing shortcomings of the baseline models. By comparing these models, we identified performance trade-offs and gained insights into optimizing aircraft detection for realistic datasets.
⚪ This project is the result of a semester project completed in Fall 2024 for the University of Central Florida. See the university website here.
⚪ The project repository can be located using this provided link. It contains the project composition, repository navigation, models' information, and evaluations completed. Click here.
▱▰▱ Project Goals. ▰▱▰
- 1. Create a labeled dataset of military aircraft using RoboFlow to train and evaluate computer vision models.
- 2. Compare the performance of YOLOv5, YOLOv8, and a custom deep learning model (ConvNet) for aircraft detection.
- 3. Design a custom ConvNet to balance detection accuracy and computational efficiency, addressing issues found in baseline models.
- 4. Use standard evaluation metrics (precision, recall, F1 score, and mAP) to assess model effectiveness and identify performance trade-offs.
- 5. Gain insights into improving computer vision models for accurate aircraft detection in a challenging real-world dataset.
