Multiobject real-time automatic detection method for production quality control of prefabricated laminated slabs

The production quality assessment of prefabricated laminated slabs (PLS) is required to ensure the installation of components on the construction site. However, existing inspection methods are difficult to realize PLS production quality inspection during the production process, and the inspection speed cannot match the PLS production speed. To address these issues, a deep learning (DL) method based on You Only Look Once version 4 (YOLOv4) is proposed for the quality detection of PLS in production in this research. Five new measures are introduced to improve the performance of YOLOv4. The ResNet50vd backbone and deformable convolution (DCN) are used to enhance the ability of the model to capture the feature information of small targets with drastically changing morphologies in complex contexts. The k-means++ clustering algorithm is used to recalculate anchor box specifications to overcome the challenge of large differences in scales across targets as well as within the same target. Intersection over Union (IoU)-aware loss is used to optimize the loss function to improve the localization accuracy of small targets. Geometric median filtering pruning (FPGM) is used to simplify the network, and the model parameters are reduced to 44.1%, which significantly improves the detection speed of the components. The results show that the improved YOLOv4 can quickly and accurately detect laminated panels, PVC boxes, and steel bars in PLS when IoU=0.75, which is more suitable for industrial applications, and the model achieves a mean average precision (mAP), F1 score, and FPS of 91.38%, 94.54%, and 22.91 frame rate (f/s), respectively, which is superior to the comparison method. The proposed method can be directly applied to PLS in production to preemptively detect errors and discrepancies in the components and improve the qualification rate and quality management efficiency of the components, which has a positive effect on the rapid development of industrialized construction.