MapGen-GAN: A Fast Translator for Remote Sensing Image to Map Via Unsupervised Adversarial Learning

Abstract

Map is an essential medium for people to understand our changing planet. Recently, research on generating and updating maps through remote sensing images has been an important and challenging task in geographic information. Traditional methods for map generation are time-consuming and labor-intensive. Besides, most supervised learning methods for map generation lack labeled training samples. It is challenging to generate maps quickly and efficiently for emergency rescue operations such as earthquakes, fire disasters, or tsunami. In this paper, we propose an unsupervised domain mapping model based on adversarial learning called MapGen-GAN. MapGen-GAN is a generative adversarial network that can do end-to-end translation from remote sensing images to general map quickly, and trained with no human annotation data. In order to improve the fidelity and the geometry precision of generated maps, we employ circularity-consistency and geometrical-consistency constraints as a part of the loss function of the proposed model. And then, an improved residual block Unet is designed and adopted as the generator of MapGen-GAN to capture the geographic structure information of buildings, roads, and topography outlines under different resolutions in the map generation. By applying the proposed model to two distinct datasets, experiments demonstrate that our model can generate maps efficiently and quickly and outperform the state-of-the-art approaches.

RSMT: A Remote Sensing Image-to-Map Translation Model via Adversarial Deep Transfer Learning

“…In essence, generating maps from remote sensing images is deemed an image-to-image translation task, which learns to map an image in a specific domain to an analogous image in a different domain. The current image-to-image translation methods such as Gc-GAN [1], CycleGAN [2] and MapGen-GAN [3] can translate remote sensing images to maps for a specific area. However, it is well known that the style of urban construction infrastructure varies greatly in different regions.…”

“…In some metropolises of China, high-rise office buildings and apartments are relatively typical and dense, while in the United States, the majority of the population live in single-family houses with private gardens. Specifically, if we train an image translation model (such as MapGenGAN [3]) to generate maps of some areas in Beijing, we should use remote sensing images and maps of the areas in or around Beijing as the training set. If we use a well-trained translator (MapGen-GAN) that adopts Beijing images as training data to generate maps of Los Angeles (LA), the results will be distorted and blurred, as illustrated in Figure 1.…”

“…The proposed model RSMT is an extension of our previous work MapGen-GAN [3], which solves the limitation that the map generation model is only applied to the datasets of the same region and improves the applicability of map generation model. RSMT is based on an adversarial transfer training scheme coupled with attention mechanism design.…”

“…MapGen-GAN: MapGen-GAN [3] is our previous work for map translation tasks. The framework is based on circularity and geometrical consistency constraints, transforming remote sensing images to maps directly and reducing the translation's semantic distortions.…”

Intelligent Matching Method for Heterogeneous Remote Sensing Images Based on Style Transfer

“…continue to emerge. Li et al [23] used deep translation to convert optical images into SAR images and Song et al [24] used GAN to convert optical images into maps, reflecting the feasibility and practicality of generative adversarial network application to remote sensing image processing.…”