Image Processing based Damage Detection of Offshore Infrastructural Elements using Texture Information

Paper #229International audienceThe current approach to Structural Health Monitoring (SHM) for offshore infrastructural elements is largely based on feedback provided by divers who may lack the desired level of inspection training, resulting in an undependable perspective of the damage present. There is thus a requirement for a robust, reliable and reproducible means for non-destructive testing. This paper presents a texture analysis based image segmentation using Support Vector Machine (SVM) classification for the purposes of detecting and classifying damaged regions on the surface of infrastructural elements. For the texture analysis stage, a Grey Level Co-occurrence Matrix (GLCM) was employed to calculate seven texture measures, namely; contrast, homogeneity, correlation, angular second moment (ASM), skewness, entropy and kurtosis. SVM was then used to classify regions as being either damaged or undamaged. As an example, the technique was applied to images of a corroding metallic surface in marine environment. Texture analysis is especially useful when regions of interest in an image are more distinguishable from the background based on their texture rather than pixel intensity. The effectiveness of this Non-Destructive Testing (NDT) technique is evaluated through the use of performance coordinates in the Receiver Operating Characteristic (ROC) space