Abstract:

Crack disease is the most common disease of bridges， DeepLabv3+ segmentation model puts forward a new Encoder-Decoder structure among deep learning methods. It combines the high-level semantic information and shallow features of the target， and adopts the method of deep separation and convolution， which achieves superior image segmentation effect. However， in the training process of the coding module， the spatial dimension of the input data is gradually reduced， resulting in the loss of useful information， which brings some limitations to the recognition of small targets with different scales. In order to improve the segmentation performance of the network， this paper proposes an image segmentation method based on improved DeepLabv3+.By adding Yolof module and Resnet module， the receptive field is further expanded and more accurate crack feature map is obtained at the same time. In order to verify the effectiveness of the improved algorithm， a large number of actual bridge crack images are taken as the original data set， which is compared with the current representative image segmentation models such as Mask R-CNN and DeepLabv3+ on the same dataset. The results show that the algorithm in this paper improves the accuracy of crack pixels by 12% and 8% respectively compared with Mask R-CNN and DeepLabv3+. The average pixel accuracy is 91.99%， and Mean Intersection over Union is 81.43%， which is more suitable for the task of bridge crack segmentation and has practical engineering application significance.

Key words: bridge engineering, bridge crack detection, DeepLabv3+, pixel accuracy, image segmentation