Fast identification of concrete cracks using 1D deep learning and explainable artificial intelligence-based analysis

The present paper discusses a computationally efficient Deep Learning (DL) model for real-time classification of concrete crack/non-crack and investigates the 'black-box' nature of the proposed DL model using eXplainable Artificial Intelligence (XAI). The state-of-the-art DL models like semantic segmentation require labor-intensive labeling for pixel-level classification. The proposed framework combines image binarization and a Fourier -based 1D DL model for fast detection and classification of concrete crack/non-crack features. Image binariza-tion as a precursor to DL extracts possible Crack Candidate Regions (CCR) and eliminates the plane structural background during DL training and testing. Metadata within the 1D DL model was generated and analyzed using local XAI, wherein t-distributed Stochastic Neighborhood Embedding (t-SNE) was used to visualize the knowl-edge transfer within the hidden layers. The proposed model enables real-time pixel-level classification of crack/ non-crack at the rate of 2 images/s on a mobile platform with limited computational facilities.

Automated summary

This paper presents a computationally efficient deep learning model for real-time classification of concrete crack/non-crack. It also investigates the 'black-box' nature of the model using explainable artificial intelligence. The proposed framework combines image binarization and a Fourier-based 1D deep learning model for fast detection and classification of concrete crack/non-crack features. The model enables real-time pixel-level classification at a rate of 2 images per second on a mobile platform with limited computational resources.