Defect Identification and Classification of Fair-faced Concrete based on Feature Enhancement

Yan Li^*Bing LingZhexiong Shang

• Hainan University, Haikou, China

The visual quality and structural integrity of fair-faced concrete—a construction material prized for its aesthetic finish—are heavily dependent on the timely and accurate detection of surface defects. These defects, if left undetected, can compromise not only the appearance but also the long-term durability and safety of buildings and infrastructure. However, traditional inspection methods rely heavily on manual labor, which is time-consuming, error-prone, and inconsistent. Automated detection systems have emerged as a promising alternative, yet they face significant challenges in real-world deployment: limited availability of high-quality labeled data, wide variability in defect appearances, and poor image quality due to environmental factors such as lighting changes and surface stains. According to the above problem, this paper proposes a defect recognition method based on Feature enhancement, with the introduction of uncertainty random Feature enhancement Module (Feature Augmentation Module, FAM) characteristics of dynamic disturbance model to extend the data distribution of diversity, Combined with variance adaptive learning mechanism automatically adjust according to the data distribution characteristic dimension turbulence intensity, and design integration of regularization of variance and the multiplicity of L2 regularization loss function to suppress fitting. Experiments show that the proposed method effectively improves the training efficiency and generalization performance of convolutional neural networks in Fair-faced Concrete defect classification, and provides a reliable solution for automated inspection in engineering scenarios.