Eureka AIR delivers breakthrough ideas for toughest innovation challenges, trusted by R&D personnel around the world.

Mars image augmentation method based on generative adversarial network

A technology for generating images and Mars, which is applied in the field of image augmentation in computer vision, can solve the problems of unstable training, low resolution of Mars images, uncontrollable image categories, etc., and achieve the effect of both diversity and scale expansion

Active Publication Date: 2022-04-22
HARBIN INST OF TECH
View PDF1 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] Aiming at the problems of low resolution of Mars images obtained by existing Mars image augmentation methods, uncontrollable image categories and unstable training, the present invention provides a Mars image augmentation method based on generative confrontation network

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Mars image augmentation method based on generative adversarial network
  • Mars image augmentation method based on generative adversarial network
  • Mars image augmentation method based on generative adversarial network

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment approach 1

[0045] Specific implementation mode 1. Combination Figure 1 to Figure 5 As shown, the present invention provides a Mars image augmentation method based on generation confrontation network, including,

[0046] The image generation network based on DCGAN is set, including an image generator and an image discriminator, and the image generator includes a feature layer processing module one to a feature layer processing module eight and a convolution processing module;

[0047] The characteristic layer processing module performs AdaIN operation, convolution operation and AdaIN operation on a pair of input random hidden codes in turn; the network structure of the characteristic layer processing module 2 to the characteristic layer processing module 8 is the same, and the Mars output from the previous characteristic layer processing module is respectively The image is subjected to deconvolution operation, AdaIN operation, convolution operation and AdaIN operation in sequence; the co...

specific Embodiment

[0071] Below in conjunction with specific embodiment the inventive method is further described:

[0072] Such as figure 1 As shown, the training samples are firstly prepared according to the actual needs. In order to facilitate the comparison with the existing methods, in this embodiment, the publicly available mars32k Mars image data set of NASA is selected. Then, the close-range Mars image in mark32k is selected as the training data sample. This is because when evaluating the quality of the generated Mars image, more attention is paid to the details of the Mars image, and the training samples when training the network should also be rich and rich. details. To this end, the close-range Mars image samples in the mars32k Mars image dataset are selected as training samples to train the image generation network based on DCGAN. Each part is described in detail below:

[0073] Prepare training samples. Training sample images can be collected according to actual needs, and then ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a Mars image augmentation method based on a generative adversarial network, and belongs to the technical field of image augmentation in computer vision. The Mars image augmentation method aims at solving the problems that an existing Mars image augmentation method is low in resolution, image types cannot be controlled, and training is unstable. Comprising the following steps: an image generation network based on a DCGAN performs fine image generation by using a progressive network training method aiming at the problem that the DCGAN is difficult to generate a high-quality high-resolution Mars image; aiming at the problem that the GAN cannot control the generation of a target type Mars image, a style migration technology is introduced to design a controllable Mars image generation method; in order to solve the problem that the Mars rover is difficult to autonomously complete training due to unstable DCGAN training, a new loss function is designed based on a Wasserstein distance. The method is used for mars image augmentation.

Description

technical field [0001] The invention relates to a Mars image augmentation method based on a generative confrontation network, and belongs to the technical field of image augmentation in computer vision. Background technique [0002] Mars is a terrestrial planet next to the Earth in the solar system, and it is also the terrestrial planet most similar to the Earth in the solar system. The discovery of water made Mars considered one of the most likely planets to harbor life, and it also made Mars one of the main targets for space exploration. At present, the aerospace field has achieved world-renowned achievements in earth satellite and manned spaceflight projects. The development of deep space exploration will be the follow-up focus, which will be of great significance to scientific and technological progress and social development. [0003] Due to the complex and changeable environment of Mars, often accompanied by weather such as sand and dust, it is more difficult to obtai...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06V10/774G06V10/764G06V10/44G06V10/54G06V10/56G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/088G06N3/048G06N3/045G06F18/2155G06F18/24Y02T10/40
Inventor 张永强丁明理张子安张印田瑞王骢
Owner HARBIN INST OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com