A New Concrete Crack Detection based on Deep Feature Extraction

Abstract

Concrete is one of the most intensively used materials in the construction industry. This building material, which is basically obtained by mixing additives such as cement, sand and water, is used extensively in the construction industry. In this paper, we propose a new machine learning method that can automatically classify cracks on the surface of concrete material that is frequently used in the construction industry. For this purpose, a feature vector was obtained using deep and textural feature extraction methods. The most significant features are selected from this feature vector using neighborhood component analysis method. The selected features are classified using support vector machines. Deep feature extraction was achieved using the AlexNet architecture, which is a well-known method in the literature. The developed model has reached 99.9% accuracy in classifying crack images and the results obtained clearly demonstrate the performance of the model in automatic concrete crack classification.