A Cascade Broad Neural Network for Concrete Structural Crack Damage Automated Classification

Abstract

Crack is the earlier indication of concrete structural severe damage; it plays an important role in structure health monitoring (SHM) of industrial civil infrastructures (such as buildings, bridges, roads, dams, etc.). Crack damage classification is the first and critical stage for concrete SHM. However, commonly used human visual classification is costly, labor-intensive, and unreliable, other machine learning based classification methods also have some drawbacks. To address these problems, this article proposes a cascade broad neural network architecture for concrete surface structural crack damage automated classification, which generates an effective and efficient framework with much less hyper-parameters than deep neural networks, and sufficiently explores the advantages of multilevel cascades of classifier ensemble. Experimental results on four challenging datasets demonstrate that its performance is quite more excellent than current mainstream classification methods (both in testing accuracy and training time).