Automatic Crack Detection on Concrete Surfaces Using LightweightDeep Learning Models

Abstract

Detecting cracks in concrete surfaces is critical for structural health monitoring. However, traditional methods show limited effectiveness due to their high costs, time-consuming processes, and vulnerability to human error. This situation reveals the need for innovative methods to produce faster, more economical, and more reliable results. This study developed an optimized convolutional neural network (CNN) model that works on low-resolution images and has a four-layer lightweight architecture. The proposed model demonstrated superior performance with an accuracy rate of 98.1% and provided distinct advantages over traditional methods regarding computational efficiency. In addition, using image segmentation techniques, crack areas are visually highlighted, and users are offered easy evaluation. The proposed model provides economical, fast, and accessible monitoring by eliminating the need for expensive hardware. In this way, structural health monitoring processes have become more effective and applicable on a larger scale. The study proposes an innovative solution that saves both time and cost in engineering applications by adopting modern artificial intelligence techniques for crack detection of concrete surfaces.