With the rapid development of UAV technology and aerial surveillance systems, efficient and accurate aircraft inspection techniques have become increasingly important. In this paper, we propose an improved YOLOv8-based model for improving the accuracy and efficiency of aircraft detection. We integrate the Double Attention mechanism to capture the key features of the aircraft more efficiently; employ the Adaptive Feature Pyramid Network (AFPN) for multi-scale feature processing to improve the detection capability for aircraft of different sizes; and introduce the Multi-Path Decoupled Intersection over Union (MPDIOU) to optimize the accuracy of the bounding box. Through a series of experiments on a standard aircraft inspection dataset, our model can achieve an average precision mean of 96.8% in real-time aircraft inspection, while the recall rate can reach 95%. Our model demonstrates a significant improvement in various key performance metrics, which proves the potential and effectiveness of the proposed method in practical applications.
Aerospace Aircraft Inspection Based on Improved YoloV8
17.11.2023
862223 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch