Deep learning-based crack detection with a focus on masonry and concrete earth retaining structures

Crack detection is crucial for proactive maintenance and ensuring structural safety in civil infrastructure. Conventional crack detection techniques were limited to manual assessments, which are often labor-intensive, time-consuming, and prone to subjective bias. Later, due to technological advancements, it has been automated. This paper proposes a comparative study of deep learning techniques, such as LeNet-5, GoogLeNet, improved GoogLeNet, and ResNet50, for crack detection in masonry and concrete structures with a focus on earth retaining structures (ERS). The study combined publicly available masonry and concrete crack image for classification (CCIC) datasets to examine the performance of four deep learning techniques for crack detection. The accuracy obtained from four deep learning techniques for crack detection in masonry and concrete structures is presented in this work. The findings of this work highlight the potential development of an advanced automated defect detection system in similar civil infrastructure monitoring, including ERS of road and railway networks.