Detection and Classification of Types of Cracks Using Deep Learning

The identification and classification of cracks are vital for maintaining the safety and integrity of buildings and transportation infrastructure. Conventional manual inspection methods are often time-consuming and labor-intensive, underscoring the necessity for automated systems leveraging computer vision techniques. This study proposes two distinct methodologies: one based on digital image processing and the other on deep learning, to facilitate the development of crack inspection systems. The deep learning approach employs a pre-trained model serving as a feature extractor, coupled with a fully connected neural network classifier, to facilitate the detection and classification of cracks. Specifically, detection is accomplished utilizing Convolutional Neural Networks (CNNs), while classification is conducted employing ResN et, addressing the distinct tasks of identifying cracks and categorizing them accurately. Evaluation on publicly available crack datasets demonstrates the superior accuracy of the proposed method in detecting surface cracks, surpassing existing state-of-the-art techniques. Furthermore, a supervised deep learning model is introduced for the detection and classification of pavement cracks, adept at tackling the complexities associated with diverse backgrounds and variations in crack patterns. Results from this model exhibit promise accurately classifying individual image patches within collected datasets. These methodologies hold considerable potential for practical implementation in non-destructive testing, enabling the detection and classification of various types of cracks across diverse structural components.