Application Research and Analysis of Computer Vision in the Field of Construction Engineering

Aiming at the problems of low degree of automation, frequent safety accidents and difficult quality control in construction engineering, this paper studies the application of computer vision, and focuses on the core idea of algorithm, underlying logic, current application and future development of Yolo series, SC crack+and efficientnet models. Yolo is good at real-time multi-target detection, which is suitable for component management and crack detection; SC crack+realizes accurate crack segmentation in complex environment through unsupervised mechanism, making it an ideal choice for traditional buildings with scarce samples. Efficientnet stands out in image classification and multimodal data fusion through composite scaling and lightweight design. Challenges remain, including Yolo’s limited small target accuracy, the hyperspectral data cost of SC crack+and the generalization limitation of efficientnet. The improvement directions such as multi-scale feature fusion, spatial constraint optimization and multi-modal data integration are proposed to support the intelligent transformation in construction monitoring and structural health assessment.