Computer Engineering and Applications ›› 2024, Vol. 60 ›› Issue (16): 319-324.DOI: 10.3778/j.issn.1002-8331.2305-0351

• Engineering and Applications • Previous Articles     Next Articles

Multi-Scale Feature Fusion Detection and Recognition Algorithm for Airport Runway Foreign Object

GUO Xiaojing, ZOU Songlin   

  1. 1.Engineering Technology Training Center, Civil Aviation University of China, Tianjin 300300, China
    2.College of Electronic Information and Automation, Civil Aviation University of China, Tianjin 300300, China
  • Online:2024-08-15 Published:2024-08-15

多尺度特征融合的机场跑道异物检测与识别算法

郭晓静,邹松林   

  1. 1.中国民航大学 工程技术训练中心,天津 300300
    2.中国民航大学 电子信息与自动化学院,天津 300300

Abstract: Foreign objects on the airport runway pose a fatal threat to the take-off and landing of the aircraft, and the way of human inspection of foreign objects is costly and inefficient. It is necessary to use deep learning detection and measurement methods for the detection of foreign objects on the airport runway. Aiming at the problems of missed detection and inaccurate positioning of foreign objects with different sizes and difficult feature extraction, a foreign object detection and recognition algorithm based on multi-scale feature fusion is proposed. Taking multi-scale feature extraction and fusion as the starting point, it proposes multi-point support for spatial attention, strengthens the feature extraction of foreign objects, and can pay enough attention to the abnormalities of different sizes. Using the BiFPN network for feature fusion can fully integrate feature information of different sizes. The experimental results show that the average accuracy reaches 94.7%, which is 5.9 percentage points higher than YOLOv5, and also surpasses YOLOv6, YOLOv7 and Faster R-CNN, and verifies that this algorithm has good application value in the field of foreign object detection on airport runways.

Key words: foreign objects on airport runway, object detection, spatial attention, feature fusion

摘要: 机场跑道异物对飞机起飞降落存在着致命威胁,而人工巡检异物的方式成本高,效率低,因此将深度学习检测算法用于机场跑道异物检测是必要的。针对异物尺寸大小不一、特征提取难度大而导致的漏检与定位不准问题,提出一种基于多尺度特征融合的机场跑道异物检测与识别算法。以多尺度特征提取与融合为切入点,提出多分支空间注意力,加强对异物的特征提取,同时能够关注不同尺寸大小的异物。采用BiFPN网络进行特征融合,能够高效融合不同尺度的特征信息。实验结果表明,改进后算法平均精度达到94.7%,相比于YOLOv5提高5.9个百分点,也超越了YOLOv6、YOLOv7以及Faster R-CNN,从而验证了该算法在机场跑道异物检测领域有较好的应用价值。

关键词: 机场跑道异物, 目标检测, 空间注意力, 特征融合