Research on LGYOLO Bridge Tie Cable Surface Inspection and Recognition Methods

Aiming at the inefficiency of manually detecting surface defects of bridge ties, the operation is difficult and time-consuming. We propose a Bridge YOLO algorithm based on the improved YOLO v5s, which introduces the improved PPLC-Net as the feature extraction network, significantly reduces the number of parameters and computation, and improves the detection speed; GCC3Block helps to pay attention to the global information, expands the sensing field, reduces the loss of information, and realizes the finer multiscale feature fusion, and is lightweight. C3-P replaces the Neck part of C3, which ensures that the accuracy is not degraded and reduces the number of parameters and computational overhead; Shape IOU is used to reduce the impact of the sample's own shape and anchor frame deviation on the regression. Compared with the original model, the precision is improved by 2.6%, the recall is improved by 3.1 %, the m AP is improved by 0.8%, the number of parameters is reduced by 26.6%, and the weights are reduced by 0.8 M. The improved model is more friendly to deploy on mobile devices, and the identification of samples is more accurate.