基于尺度融合的机场跑道地下病害检测算法

Scale Fusion Based Airport Runway Subsurface Defect Detection Algorithm

及时准确地检测机场跑道地下病害对保障飞行安全至关重要,由于机场道面结构层复杂,其电磁波传播环境复杂、噪声强度大,致使地下病害的探地雷达数据特征被严重干扰。为此,提出了一种基于尺度融合的机场跑道地下病害检测算法SF-SSD。首先,在VGG16主干网络上设计具有更宽感受野的RFB模块,抑制了病害周围的噪声干扰,提取更多原始雷达数据特征;然后,使用尺度融合的方式融合网络浅层特征,获取不同类型病害间的细微差异,形成高辨识度的病害细节纹理特征;最后,根据6种不同分辨率的特征图生成6种尺度先验框进行类别预测和位置回归,通过非极大值抑制的方式筛除冗余的先验框。在真实机场跑道地下病害数据集上进行了测试,并与目前5种经典目标检测算法进行对比。实验结果表明,SF-SSD算法可以较好地从辨识度低、噪声强度大的雷达数据中完成病害自动检测,并取得了最高的平均准确率,达到了82.18%。

Timely and accurate detection of airport runway subsurface diseases is crucial to ensure flight safety. Due to the complex structure layer of airport road surface, the complicated electromagnetic wave propagation environment and the high noise intensity, the ground-penetrating radar would be seriously disturbed. A scale fusion-based airport runway subsurface disease detection algorithm SF-SSD was proposed. Firstly, the RFB module with wider perceptual field was designed on the VGG16 backbone network, which suppressed the noise interference around the disease and extracted more features from the original radar data. Then, the shallow features of the network were fused using scale fusion to obtain the subtle differences between different types of disease and form highly discriminative detailed texture features of the disease. Finally, six scaled prior frames were generated for category prediction and location regression, and the redundant prior frames were screened out by non-maximal value suppression. It was tested on real airport runway subsurface disease dataset, and compared with five current classical target detection algorithms. The experimental results showed that the SF-SSD algorithm could accomplish automatic disease detection better from radar data with low discrimination and high noise intensity, and achieved the highest average accuracy of 82.18%.