Detecting Structural Components of Building Engineering Based on Deep-Learning Method

Abstract

Detecting engineering structural components is the basis for intelligently managing construction engineering quality, scheduling, and costs. However, the detection of engineering structural components still cannot be done reliably and effectively by any technical means. Following a detailed analysis of existing object detection algorithms, an automatic method for building structural component detection based on the Deeply Supervised Object Detector (DSOD) is proposed. Compared with other algorithms, DSOD only needs limited data and can obtain the highest level of object detection by training from scratch. To verify the effectiveness of the method, based on the entity-scale reduction model of a building structure, a combined image data set of engineering structural components is established by multilayer, polymorphic, multidirectional, multiangle, structural data acquisition. Following the definitions of true positive, false positive, and false negative, the precision and recall rate of structural component detection at different shooting angles, different visual ranges, and different occlusion degrees were tested with a confidence threshold of 0.7. The experimental results show that the method has high detection precision, high recall rate, and high speed. It can effectively solve the problem of the detection of structural component of building engineering and provide practical guidance on how to scientifically collect engineering structural component images at construction sites.