摘要：

为促进道面地下领域知识和目标检测算法的深度融合, 缓解不同病害样本间的特征复杂性和相似性导致的特征畸变问题, 提升病害的自动化检测效果, 提出了融合关联关系推理的机场道面地下病害检测算法。首先, 所提算法结合残差网络和多尺度特征金字塔网络(FPN)提取目标特征信息; 其次, 通过挖掘机场道面地下病害关联关系矩阵, 结合图推理设计地下病害关联关系推理模块, 以区域生成网络(RPN)生成的特征向量作为输入特征, 利用自我学习的变换矩阵设定图的传播权重, 实现特征信息传播并构建有效的关联关系推理模块。实验结果证明, 融合关联关系推理的机场道面地下病害检测算法可以有效地利用地下病害之间的关联关系, 消除病害之间的相互干扰并且检测效果达到最优, 检测的平均准确率达到87.38%。

关键词: 机场道面, 地下病害, 领域知识, 目标检测, 关联关系推理

Abstract:

To promote the deep integration of domain knowledge on underground pavement with object detection algorithms, alleviate feature distortion caused by feature complexity and similarity among different disease samples, and enhance automatic disease detection, an airport pavement underground disease detection algorithm integrating association reasoning is proposed. First, the proposed method combines a residual network and a multi-scale Feature Pyramid Network (FPN) to extract target feature information. Second, a module for underground disease association reasoning is designed based on graph inference, leveraging the correlation matrix of airport pavement underground diseases. The feature vectors generated by the Regional Proposal Network (RPN) are used as input features, and self-learning transformation matrices are employed to set the propagation weight of the graph. This enables feature information propagation and constructs an effective association reasoning module. Experimental results demonstrate that the airport pavement underground disease detection algorithm integrating association reasoning effectively utilizes the correlation relationships between underground diseases, eliminates mutual interference among defect samples, and achieves optimal detection performance, with an average accuracy of 87.38%.

Key words: airport pavement, underground disease, domain knowledge, object detection, association reasoning