Intelligent Detection of Old Tunnel Apparent Defects Via Semantic Segmentation
To improve the defect segmentation effect of old tunnels,an ECA-Unet++detection algo-rithm is proposed on basis of the attention mechanism and multi-feature fusion.Firstly,Unet++network is used as the framework to effectively extract defect features through multi-scale feature fu-sion.Then,the attention module ECAnet is applied in the decoding part to suppress the weight of ir-relevant channels and improve the anti-interference ability of the model in the complex background of old tunnels.Finally,a complex dataset of the apparent defects of the old tunnel is constructed by tak-ing the four defects of lining cracking,leaking,segment breakage and lining shedding as objects.To verify the performance of the proposed algorithm,four classical networks,Unet,PSPnet,Deeplabv3+and Unet++,are selected for comparison.The test results show that the ECA-Unet++model has a good detection effect under complex background.On the data set constructed in this paper,mI-oU is 55.33％and mPA is 70.47％,which is suitable for the accurate detection of apparent complex defects in old tunnels.
old tunnelsapparent defectsUnet++semantic segmentation
万方数据
维普
