The Research Of Pavement Crack Automatic Identification Method Based On Improved Mask R-CNN Model

Pavement crack is a common road disease,which reduces the pavement service life and impair vehicle safety.Therefore,it is significant to detect and identify the pavement crack.With the development of artificial intelligence,using deep learning algorithms to detect pavement crack has also made good achievements.Compared to conventional crack identification methods,deep learning algorithms has higher detection accuracy and stronger generalization ability,which can be used in various pavement conditions.Therefore,this paper proposes a pavement crack identification method via improved Mask R-CNN model.Firstly,in order to solve problems of limited scenes and single task in the existing crack identification algorithm,this paper uses Mask R-CNN model for crack recognition.It distinguishes different kinds of cracks and different individual cracks of the same kind by the form of instance segmentation.It can also complete the task of crack detection and segmentation simultaneously.Then it is applied to the collected crack dataset.Through the training and optimization of the model,the crack pixels in the generated detection box are segmented while the crack is located.Secondly,to solve the problems of inaccurate detection of crack edge and low accuracy of Mask R-CNN model,an improved C-Mask RCNN is designed,which improves the quality of crack region proposal generation by cascading multi threshold detectors and achieves accurate positioning under high threshold.After that,the improved model is optimized and compared to achieve the optimal effect while improving the model detection accuracy and segmentation accuracy.Experimental results show that the m AP of C-Mask RCNN model detection part is95.4%,which is 9.7% higher than that of the conventional model,and the m AP of the segmentation part is 93.5%,which is 13.0% higher than that of the conventional model.Finally,after the crack is identified,this paper divides the crack image into a grid of0.1m×0.1m,and uses the automatic evaluation method of crack based on JTG 5210-2018 standard to calculate the pavement crack rate.Then the geometric parameters of crack are calculated by combining the results of grid.At the same time,to solve the problem of detection box deviation in the existing crack detection,several corresponding adjustment methods are proposed according to the types of cracks.In summary,the C-Mask RCNN model proposed in this paper can locate and extract cracks with high identification accuracy.What’s more,the research on the quantification of crack parameters in this paper can also be applied to the automatic identification of pavement cracks and road maintenance.