Neural-network-based damage classification of bridge infrastructure using texture analysis

Damage in concrete structures can be assessed by analyzing the texture of surface deterioration using optical concrete imagery. This research proposes the application of an enhanced method of texture analysis, based on the signal processing technique of Haar's wavelet transform in combination with the grey level co-occurrence matrix statistical approach, to characterize and quantify damage. Three different types of imagery, colour, greyscale, and thermography are evaluated for their effectiveness in representing surface deterioration. The multilayer perceptron artificial neural network classifier is applied on three different datasets: spatial, spectral, and a combination spatial-spectral dataset. Results show that the combination of textural and spectral data produced the highest overall accuracies;; the thermography provided better classifications than the other types of imagery. Classifications based on the combination datasets were used to determine the different levels of damage in the concrete.