YOLO-based segmented dataset for drone vs. bird detection for deep and machine learning algorithms.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Increasing use of unmanned aerial vehicles (UAVs) raises safety and security concerns.
• Existing UAV detection systems often fail or misclassify objects like birds.
• A standardized dataset is crucial for training robust UAV detection models.

Purpose of the Study:

• To present a novel dataset of drone and bird images for UAV detection model training.
• To provide a comprehensive resource for researchers and developers in the UAV field.
• To improve the accuracy and reliability of automated UAV detection systems.

Main Methods:

• Dataset creation using Roboflow software with AI-assisted annotation.
• Manual, edge-to-edge image segmentation for detailed training data.
• Inclusion of both training and testing sets for model evaluation.

Main Results:

• A dataset featuring diverse images of drones and birds in various environments.
• Detailed annotations enabling high-accuracy training for models like YOLO.
• Demonstrated advantages over existing datasets by including commonly confused objects.

Conclusions:

• The presented dataset is a valuable resource for advancing UAV detection and classification.
• Its comprehensive nature and diverse scenarios enhance model training effectiveness.
• This work contributes to improved safety and security in professional and recreational UAV use.