Highway Pavement Repair Detection and Length Calculation Based on Deep Learning and Least Squares

For recognizing and calculating the length of a pavement repair, this paper proposes an algorithm based on deep learning and least squares. Firstly, the pavement repair lesions were detected using the improved YOLOv5. Then the regions of pavement repair were cut from the row image. Secondly, using the instance segmentation algorithm, the repair lesions were cut out from the image. Its corresponding pixel points were extracted as (x, y) coordinates. Finally, the scatter was fitted to a curve using the least squares. The length of the pavement repair was calculated by curve integration of the curves. Target detection and strength segmentation models were trained on open-source data sets. The accuracy of the model repair length calculation was actually tested on a section of roadway in Jinan, Shandong Province.