Pavement crack detection method based on deep learning

Konferenz: CIBDA 2022 - 3rd International Conference on Computer Information and Big Data Applications
25.03.2022 - 27.03.2022 in Wuhan, China

Tagungsband: CIBDA 2022

Seiten: 4Sprache: EnglischTyp: PDF

Autoren:
Huang, Jian; Wu, Dandi (School of Communication and Information Engineering , Xi 'an University of Science and Technology, China)

Inhalt:
Pavement crack detection is an important step to achieve the task of pavement maintenance. Only by detecting pavement cracks in time can ensure road safety. Road crack detection has very important practical significance. This paper proposes to detect and extract road surface cracks based on deep learning, and based on the segmentation results, the crack types are classified and the crack area is quantitatively calculated. In this paper, after improving the ASPP module of the DeepLabV3+ network, the crack image is extracted, and then the extracted crack image is classified by the MobilenetV2 network, and finally it is quantified according to the quantization algorithm proposed in this paper. The experimental results show that the improved segmentation network has better extraction effect on crack images, the mIoU reaches 77.48%, the classification accuracy of the classification network for various types of cracks reaches more than 85%, and the quantization algorithm can accurately quantify the crack area and other information.