Method for enhancing forward modeling data of underground pipeline of residential area

Abstract

The invention discloses a residential area underground pipeline forward modeling data enhancement method, which comprises the following steps: improving a WGAN-GP model: adopting U-net + + as a backbone network of a generator, adopting a SwinTransform module as a feature extraction module of a discriminator, and introducing structural similarity loss into a loss function to obtain an underground pipeline image generation model; inputting the pipeline label as a condition vector into an underground pipeline image generation model, and outputting a simulated underground pipeline radar echo image by the model; a residual dense block RDB is adopted to replace a convolution module in the U-net model, a double attention mechanism is introduced into jump connection, a dynamic normalization strategy is adopted for features of different levels, a multi-scale gradient sensitive loss function is adopted as a loss function of the model, and an underground pipeline image denoising model is obtained; and taking the actually measured underground pipeline radar echo image as the input of the underground pipeline image denoising model, and outputting the denoised underground pipeline radar echo image by the model.

Images