The proposed U-HDN architecture consists of three components: U-net architecture, multidilation module, and hierarchical feature learning module. The red dotted box presents the modified U-net; the green dotted box is a multi-dilation module; the blue dotted box shows the hierarchical feature learning module.

Contexts in source publication

Context 1

... by above observations, in this paper a new network called U-HDN, to fuse multi-scale features in encoder-decoder network based on U-net for crack detection is proposed. The flowchart and the proposed U-HDN architecture are shown in Figure 2 and Figure 3, and the proposed method consists of three components: U-net architecture, multi-dilation module (MDM), and hierarchical feature (HF) learning module. Firstly, an U-net is divided into encoder and decoder networks, which have the same scale at each stage. ...

Context 2

... the proposed method is designed and calculated the number of feature maps. In this paper, we employ spatial domain to calculate the feature maps, and the number of the feature maps are shown in Figure 3 (shown on the green boxes). ...

Context 3

... is shown in Figure 3, the red dotted box presents the modified U-net. Contracting path consists of two 3 × 3 convolution layers, each followed by the activation function rectified linear unit (ReLU) [70], and a 2 × 2 max pooling layers for down-sampling. ...