基于多分辨率网络的桥梁裂缝检测方法研究  

Method for Bridge Crack Detection Based on Multiresolution Network

作　　者：李良福[1] 武彪 王楠[1] Li Liangfu;Wu Biao;Wang Nan(School of Computer Science,Shaanxi Normal University,Xi′an,Shaanxi 710119,China)

机构地区：[1]陕西师范大学计算机科学学院,陕西西安710119

出　　处：《激光与光电子学进展》2021年第12期95-104,共10页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61573232,61401263)。

摘　　要：针对传统的桥梁裂缝检测算法具有抗噪能力差和难以处理复杂背景的裂缝图像,以及常规深度学习图像分割算法存在空间精确度低的问题,提出一种基于多分辨率且具有较高空间精确度的桥梁裂缝检测方法。首先使用无人机采集桥梁图像,通过图像增强处理得到桥梁裂缝数据集。接着利用并行连接多分辨率子网和重复的多尺度融合,使检测模型在整个过程中保持高分辨率表示,同时在相同深度和相似水平的低分辨率表示的帮助下执行重复的多尺度融合以提升高分辨率表示,使得高分辨率表示中具有很强的高级语义特征。最后在数据集上对所提算法进行训练及验证。结果表明,所提算法的各项分割指标都有较为显著的提升,裂缝检测准确率高达93.8%,平均交互比达到85.48%。Aiming at the problem that traditional bridge crack detection algorithms have poor antinoise ability and difficulty processing crack images with complex backgrounds,and the conventional deep learning image segmentation algorithm has low spatial accuracy,a bridge crack detection method based on multi-resolution and high spatial accuracy is proposed.First,the unmanned aerial vehicle is used to collect bridge images.The bridge crack dataset is obtained through image enhancement processing.Then,the parallel connection is used to connect multiresolution subnets and repeated multi-scale fusions,so that the detection model maintains high-resolution representations throughout the process, while performing repeated multiscale fusion using low-resolution representations of the same depth and similar level.This is to improve the high-resolution representation so that high-resolution representation also exhibits strong high-level semantic features.Finally,the proposed algorithm is trained and verified on the dataset.The results show that all the segmentation indexes of the proposed algorithm are significantly improved,the accuracy of crack detection is as high as 93.8%,and the average interaction ratio reaches85.48%.

关 键 词：图像处理 桥梁裂缝检测 语义分割 无人机数据采集 高分辨率网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]