Concrete Crack Classification Using Deep Convolutional Neural Network

Abstract

Concrete is one of the most used materials in civil construction, and like any other material, it is subject to pathological degradation, which is often overlooked. The lack of proper inspection is correlated with 66% of building accidents. The maintenance operation of structures typically involves a visual inspection to assess their condition and subsequently seek solutions. This damage often manifests in the form of cracks or fissures, which, if left untreated, can lead to partial or total structural collapse. This article aims to propose a new architecture of deep convolutional neural network for identifying cracks on concrete surfaces. The proposed classifier achieves 94.04% accuracy in diagnosing the pathology, having been trained, validated, and tested on 25,000 image samples from three databases.