CDCGAN-wc-based target SAR (Synthetic Aperture Radar) image generation method and device

Abstract

The invention discloses a target SAR image generation method and device based on CDCGAN-wc, and the method comprises the steps: obtaining a noise vector, and enabling the noise vector to obey Gaussian distribution; inputting the noise vector and a preset category label vector into an image generation model to obtain an SAR image of a preset category target; wherein the image generation model is a generator network in a conditional generation network CDCGAN model obtained through pre-training. According to the method, the sample label information is utilized, an unsupervised model is converted into a supervised generation model, and then the SAR image of the specified category of target is generated to serve as a training sample for subsequently realizing SAR image target recognition, so that the problem of insufficient sample data is solved, and different categories of sample images are not required to be generated by using a plurality of network models.

Description

technical field [0001] The invention belongs to the technical field of radar, in particular to a method and device for generating a target SAR image based on CDCGAN-wc. Background technique [0002] Because of its unique imaging characteristics, SAR (Synthetic Aperture Radar) plays an important role in reconnaissance, navigation, guidance, remote sensing and other fields. Among them, SAR image target recognition is an important part of many SAR applications. [0003] The main purpose of SAR image target recognition is similar to that of optical images, that is, to accurately determine the target type from unknown image data. However, different from optical image data, SAR image data is relatively expensive due to the equipment required for its acquisition, the high cost of the vehicle carrying these acquisition equipment, and the non-cooperative nature of targets in military applications. In addition, due to the characteristics of SAR imaging technology, the SAR image data...

AI Technical Summary

Problems solved by technology

However, different from optical image data, SAR image data is comparatively difficult due to the equipment required for its acquisition, the high cost of vehicles equipped with these acquisition equipment, and the non-cooperative nature of targets in military applications. In addition, due to the characteristics of SAR imaging technology, SAR image data will show certain differences under the conditions of different azimuth angles and elevation angles. The recognition rate of the data is seriously affected, so it is necessary to effectively expand the SAR image data

[0004] In related technologies, with the wide application of GAN (Generative Adversarial Network, Generative Adversarial Network), Bao Xianjie expanded the MSTAR data set based on DCGAN, WGAN and WGAN-GP. However, the above methods are all for a single category of targets. Image expansion, if you want to generate multi-category targets, you can only train multiple networks

Images