An Method for Increase the Target Recognition Rate of Synthetic Aperture Radar Images based on Improved Generative Adversarial Networks

• Main Authors: CHANG, HONG-EN, 張宏恩
• Other Authors: CHANG, YANG-LANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/j56ukn

碩士 === 國立臺北科技大學 === 電機工程系 === 107 === Current research shows that synthetic aperture radar (SAR) signals have been proven to be highly detectable and recognizable in surface environments and in specific target monitoring applications, as synthetic aperture radars are weather- and day-night-proof and weather-proof. It has now been widely used in defense, telemetry and disaster prevention and other fields. Therefore, automatic target recognition of SAR images is becoming more and more important. Combining today's more popular deep learning models, we can achieve good results in SAR image recognition. The generative adversarial networks model mentioned in this paper is composed of two networks: generator and discriminator. The Generator is responsible for generating radar images of low-dimensional random numbers, while the Discriminator needs to learn how to distinguish between real images and fake images generated by the Generator. These two networks will be trained during the training process. Fight against each other to significant features of images. However, there are still many improvements in the traditional generative adversarial networks. In this study, Wasserstein generative adversarial networks is used in this architecture. In addition to solving the above problems, it can produce high-quality SAR images. In order to provide effective and complete training in the case of insufficient training samples.