Map vectorization sample enhancement method and system based on generative adversarial network

Abstract

The invention provides a map vectorization sample enhancement method and system based on a generative adversarial network. The method comprises the following steps: S1, preprocessing an image map; s2,constructing a generative adversarial network and training a sample generation model; s3, calibrating a self-sample; s4, generating a sample substrate; and s5, carrying out sample combination enhancement. According to the invention, the lightweight sample set of the self data of the image map is manufactured; and the space of the effective sample set is greatly expanded by constructing a deep learning model of the generative adversarial network, so that the space can meet the data volume requirement of a deep learning training image vectorization model, and the method described by the invention is implemented by constructing a corresponding system. Compared with the traditional process of realizing vectorization of the image map manually / semi-automatically, the method provided by the invention only needs to plot a small amount of initial samples manually, and subsequently generates massive training samples meeting requirements in a full-automatic manner, thereby providing powerful technical support for research and application of vectorization of the image map.

Description

technical field [0001] The invention relates to the intersection field of geographic information and artificial intelligence, in particular to a map vectorization sample enhancement method and system based on a generative confrontation network. Background technique [0002] The vectorization of remote sensing image maps can realize the vectorized storage of ground and object elements in the map, which can not only effectively save storage space, but also greatly improve the loading speed of map data to meet the growing demand for map network use; another On the one hand, the vectorized feature elements can be better superimposed and analyzed with other similar data, and the service functions of spatial geographic information can be expanded and improved. [0003] On the other hand, for the collection and production of map data, at the current stage, image data is still the most convenient map data to collect among all kinds of data. Sensors mounted on various satellites, dro...

AI Technical Summary

Problems solved by technology

[0005] 1) Manually implemented vector production cannot produce absolutely consistent operating standards according to different map production requirements, and evaluation deviations will inevitably occur;

[0006] 2) Due to the strong timeliness of map products, each map update needs to redraw or modify the vectorized data, which is time-consuming and labor-intensive, and the operation cost is high;

[0007] 3) The manual operation method is affected by the individual differences of the operators, so it is impossible to achieve the same standard without differences. At the same time, human errors and errors will also affect the final image quality

[0010] 1) The deep learning framework based on massive samples cannot obtain sufficient data support in the application field of remote sensing image vectorization, and the lack of labeled samples cannot effectively converge the training model, so that high-precision image map vectorization cannot be achieved Model

[0011] 2) The image map is affected by many factors such as resolution, map quality, and shooting environment, and the use of public sample sets often cannot adapt to the specified image map to be vectorized

But it will also bring new problems, including the over-fitting problem of the model, the limited sampling space and the inability to collect enough samples, etc.

Images