Enhancing Bridge Deck Delamination Detection Based on Aerial Thermography Through Grayscale Morphologic Reconstruction: A Case Study

TL;DR

This paper introduces a grayscale morphologic reconstruction pre-processing method to improve UAV-based thermography detection of bridge deck delamination, addressing environmental temperature variations that hinder conventional techniques.

Contribution

It proposes an iterative background estimation technique that enhances delamination detection accuracy in thermographic images from UAV surveys.

Findings

Detection performance improved with the proposed method

Raw image post-processing caused over- and under-estimation

Method shows promise for generalization with further tuning

Abstract

Environmental-induced temperature variations across the bridge deck were one of the major factors that degraded the performance of delamination detection through thermography. The non-uniformly distributed thermal background yields the assumption of most conventional quantitative methods used in practice such as global thresholding and k-means clustering. This study proposed a pre-processing method to estimate the thermal background through iterative grayscale morphologic reconstruction based on a pre-selected temperature contrast. After the estimation of the background, the thermal feature of delamination was kept in the residual image. A UAV-based nondestructive survey was carried out on an in-service bridge for a case study and two delamination quantization methods (threshold-based and clustering-based) were applied on both raw and residual thermal image. Results were compared and evaluated based on the hammer sounding test on the same bridge. The performance of detectability was noticeably improved while direct implementation of post-processing on raw image exhibited over- and under-estimation of delamination. The selection of pre-defined temperature contrast and stopping criterion of iteration were discussed. The study concluded the usefulness of the proposed method for the case study and further evaluation and parameter tuning are expected to generalize the method and procedure.