Image visual detection method for rebar engineering quality in concrete structure based on deep learning
Aiming at the problems of low detection efficiency and incomplete coverage caused by the current manual sampling method for quality inspection of rebar engineering,an improved YOLOX object detection algorithm was proposed to achieve intelligent detection of rebar spacing and diameter.The method realized rebar spacing and diameter detection based on accurate rebar intersection identification.The coordinate attention mechanism module was added to YOLOX object detection algorithm,and the original loss function for bounding box regression was replaced with the complete intersection over union loss function,which significantly improved the accuracy of the object detection algorithm in detecting both the rebar intersection and its central coordinates.According to the pixel coordinate information from the rebar intersection prediction box,along with depth information collected by the RGBD camera,the rebar spacing detection was realized.The Holistically-nested Edge Detection(HED)edge detection algorithm was used to eliminate the interference of rebar rib edges in the image on the number of pixels included in the statistical rebar diameter,to realize the detection of rebar diameter.The results show that the maximum error of rebar spacing detected by the improved algorithm is 4.04 mm,and the average error is less than 2.8 mm,which meets the requirements of construction specifications.When the improved algorithm is used to detect 8,10,16,20,25 mm rebar diameters,the average accuracy of the detection value in the range of standard value d0±1 mm is 97.22％.
万方数据
维普
