Road surface crack detection method based on MobileNet-PSPNet neural network model

Abstract

The invention provides a campus road crack detection method based on a Mobilenet-PSPNet neural network model, and the method comprises the following steps: S1, collecting a crack image data set, manually calibrating a label, and converting the label into a corresponding mask bitmap; S2, designing a MobileNet-PSPNet neural network, and training the MobileNet-PSPNet neural network by using the image processed in the step S1; S3, collecting campus road surface images and transmitting the campus road surface images to the terminal in real time; S4, extracting global features of the data set image acquired in the step S3 by using MobileNet; S5, extracting local features of the global feature map obtained in the step S4 through a pyramid adaptive average pooling module in the MobileNet-PSPNet network; S6, performing up-sampling operation on the local features obtained in the step S5, and then performing feature fusion on the local features and the global features to obtain new features including the global features and the local features; and S7, obtaining a final prediction result through convolution and up-sampling operation. According to the method, the crack of the campus road is detected in real time through the improved Mobilenet-PSPNet based on the PSPNet convolutional neural network, the method is accurate and efficient, and wrong detection and missing detection are not easy to generate.

Description

technical field [0001] The invention relates to the field of crack detection based on a deep learning convolutional neural network, in particular to a method for detecting cracks on a road surface based on a MobileNet-PSPNet neural network model. Background technique [0002] According to the official website of the Ministry of Transport, the total mileage of my country's highways has exceeded 5 million kilometers, ranking first in the world. However, at present, road crack detection is mainly performed manually in practice. This method is costly, labor-intensive and low-efficiency. At the same time, it is mainly detected by manual methods, which is not only subjective, but also may vary depending on the supervisor's physical condition or The reason for lack of concentration will lead to missed detection or misclassification. Therefore, the main problem to be solved by the present invention is to quickly and timely grasp road pavement information and realize automatic detec...

AI Technical Summary

Problems solved by technology

[0003] In order to solve the above technical problems, especially the low accuracy of crack detection and the loss of detailed information, the present invention provides a campus road crack detection method based on the MobileNet-PSPNet neural network model in view of the defects or deficiencies in the prior art

Images