Vehicle target detection method based on improved YOLO V3 network model.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Existing network models struggle with detecting small vehicle targets in aerial photography.
• Accurate and efficient vehicle detection is crucial for various applications, including surveillance and traffic monitoring.

Purpose of the Study:

• To propose an improved YOLO V3 network model for enhanced small vehicle target detection.
• To increase the accuracy and efficiency of detecting small vehicles in aerial imagery.

Main Methods:

• Optimization and adjustment of anchor boxes within the YOLO V3 architecture.
• Enhancement of the network's residual module to improve feature extraction for small targets.
• Integration of rectangular prediction frames with orientation angles for precise vehicle localization.

Main Results:

• The improved YOLO V3 model achieved an accuracy rate of 89.3%, a 15.9% increase over the original YOLO V3.
• Recall rate improved by 16%, and F1 score increased by 15.9%, demonstrating superior detection performance.
• Reduced instances of incorrect and missed vehicle detections were observed.

Conclusions:

• The proposed method effectively enhances the detection capability for small vehicle targets in aerial images.
• The integration of anchor box optimization, improved residual modules, and oriented bounding boxes offers a robust solution for vehicle detection.
• This research provides valuable insights and a potential framework for addressing challenges in small object detection.