Image Fusion for Improved Detection of Near-Surface Defects in NDT-CE Using Unsupervised Clustering Methods

The capabilities of non-destructive testing (NDT) methods for defect detection in civil engineering are characterized by their different penetration depth, resolution and sensitivity to material properties. Therefore, in many cases multi-sensor NDT has to be performed, producing large data sets that require an efficient data evaluation framework. In this work an image fusion methodology is proposed based on unsupervised clustering methods. Their performance is evaluated on ground penetrating radar and infrared thermography data from laboratory concrete specimens with different simulated near-surface defects. It is shown that clustering could effectively partition the data for further feature level-based data fusion by improving the detectability of defects simulating delamination, voids and localized water. A comparison with supervised symbol level fusion shows that clustering-based fusion outperforms this, especially in situations with very limited knowledge about the material properties and depths of the defects. Additionally, clustering is successfully applied in a case study where a multi-sensor NDT data set was automatically collected by a self-navigating mobile robot system.