In order to realize the rapid and accurate detection of hydraulic concrete cracks, a semantic segmentation method based on improved Deeplab V3+ network is proposed. In this method, the Mobilenetv2 network is used to replace the original backbone network, and the hole convolution of the hole convolution pyramid pooling module (ASPP) is replaced by the hole depth separable convolution, so as to improve the operation speed, reduce the sampling multiple of deep features and decrease the loss of semantic information. The experimental results show that the frame rate can reach 51.11 frame/s, which is 23.33 frame/s higher than that of the original network, and the reasoning speed is greatly improved. The mean intersection over union and mean pixel accuracy are 89.45 and 95.19, respectively, with high segmentation accuracy. The segmentation effect of typical concrete cracks is also better than the comparison method, indicating a strong generalization ability.