基于深度学习的隧道超高分辨率图像病害检测框架  

Deep learning-based disease detection framework for ultra-high resolution images of tunnels

作　　者：马海志[1] MA Haizhi(Beijing Urban Construction Survey and Design Institute Co.,Ltd.,Beijing 100101,China)

机构地区：[1]北京城建勘测设计研究院有限责任公司,北京100101

出　　处：《测绘通报》2025年第3期105-110,共6页Bulletin of Surveying and Mapping

摘　　要：现有隧道检测技术采集的数据通常会获得超高分辨率图像,而隧道中病害的实际面积较小,使得图像经过简单预处理(如缩放)后会发生病害信息丢失,且在有限的计算资源下训练的深度学习模型可能会出现物体检测率降低、训练不稳定等现象。针对上述问题,本文提出了一种基于深度学习和隧道超高分辨率图像的病害检测框架,通过对超高分辨率图像进行预处理,将原始图像分割成更小的补丁图像,将超高分辨率图像调整到合适的大小,以提高检测模型的性能,该框架适用于任何深度学习模型。试验结果表明,相较于常规检测流程,本文所提框架下的模型性能提高了约77.19%;且该框架适用于一般的超高分辨率图像,可以有效识别隧道以外的一般结构的损坏。The data collected by existing tunnel detection techniques usually obtains ultra-high resolution images,and the actual area of the disease in the tunnel is small,which makes the loss of disease information occur after the image has been simply pre-processed(e.g.,scaled),and the deep learning model trained under limited computational resources may have a reduced detection rate of the object,unstable training,and other phenomena.To address the above problems,this paper proposes a framework for disease detection based on deep learning and ultra-high resolution images of tunnels,which is applicable to any deep learning model by pre-processing the ultra-high resolution image,segmenting the original image into smaller patch images,and resizing the ultra-high resolution image to a suitable size to improve the performance of the detection model.The experimental results show that the performance of the model under the proposed framework improve by about 77.19%compared with conventional detection process.And the proposed framework is applicable to general ultra-high resolution images,which can effectively identify the damages of general structures other than tunnels.

关 键 词：隧道检测 超高分辨率图像 深度学习 病害检测 图像预处理 

分 类 号：P237[天文地球—摄影测量与遥感]