To improve the detection efficiency and effectiveness and achieve automated detection of isolation details,three types including the isolation details with horizontal isolation seams,the ones with vertical isola-tion seams and the ones whose pipeline were with flexible joints were classified according to construction de-tails.The isolation details with horizontal isolation seams were selected.Based on the detection requirements and defect characteristics,a deep learning-based method for detecting isolation details was proposed.By taking on-site photos of isolation details of 128 isolated buildings in China,a data set of isolation details was estab-lished and calibrated.The isolation details detection model was constructed by combining the residual network model with transfer learning technology.The results show that the recognition accuracy of this model reaches 98.4％and the F1 score is 0.984 on the test sets.The proposed model can extract the characteristics of hori-zontal isolation seams,and accurately determine the presence of horizontal isolation seam defects in the isola-tion details,and realize automatic detection of the isolation details with horizontal isolation seams.
deep learningisolated buildingshorizontal isolation seamdefect detection
国家科技期刊平台
NSTL
万方数据
