Rollable Latent Space for Azimuth Invariant Sar Target Recognition

Sagi, Kazutoshi; Toizumi, Takahiro; Senda, Yuzo

doi:10.1109/igarss.2018.8519366

Cited by 1 publication

(1 citation statement)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Bao et al [24] introduced the Cycle-GAN algorithm, which transformed the image from the simulation domain to the real domain, and realized the data enhancement of the target domain image. Sagi et al [25] putted forward the idea of rotatable hidden space. Through inputting two SAR images of different azimuth angles, the encoder-decoder structure is looked on as the hidden space feature representation of the input image, and a rotation transformation matrix is obtained by using the feature, and then SAR images of various azimuth angles is generated on the ground of the learned matrix.…”

Section: Related Workmentioning

confidence: 99%

Target Recognition of SAR Image Based on CN-GAN and CNN in Complex Environment

et al. 2021

View full text Add to dashboard Cite

In recent years, with the rapid development of deep learning, the research of radar image automatic target recognition (ATR) has made great progress. However, because of the complex environments and special imaging principles, Synthetic Aperture Radar (SAR) image still have the problems of sample scarcity and strong speckle noise, which affects the target recognition performance. To solve the above problems, we proposed a target recognition method of SAR image based on Constrained Naive Generative Adversarial Networks (CN-GAN) and Convolutional Neural Network (CNN). Combining Least Squares Generative Adversarial Networks (LSGAN) and Image-to-Image Translation (Pix2Pix), CN-GAN can overcome these problems of low Signal-to-Clutter-Noise Ratio (SCNR), model instability and the excessive freedom degree of the output, which are produced by conventional naive GAN. Besides, we adopted a shallow network structure design in CNN, which can effectively improve the generalization ability of the model and avoid the problem of model overfitting. The experimental results in this paper demonstrate that CN-GAN has achieved the data generation and data enhancement, the SCNR of generated data is higher than the origin data set and data sets gained by other forms of GANs, the recognition performance based on the extended data set is better than the origin data set, and the recognition rate of data set enhanced by CN-GAN is higher than that of other common data enhancement methods. INDEX TERMS Convolutional Neural Network, Generative Adversarial Networks, Synthetic aperture radar, Target recognition I. INTRODUCTION C OMPARED to optical, infrared and other sensors, SAR is not influenced by the weather, light or other conditions, can achieve continuous observation in all weather, and has a certain surface penetration ability [1]. It has been widely used in civil and military fields [2].At the same time, SAR technology has developed rapidly, with SAR image getting better quality and higher resolution, while the ATR development based on SAR image is relatively slower [3]. These difficulties of SAR ATR mainly focus on three aspects: (1) These complex environments including non-target clutter, occlusion, stacking, concealment, camouflage, electronic countermeasures, electromagnetic interference, etc., lead to the low SCNR problem; (2) The variation of the target itself and the difference of the same type target, for example, the variation of structure and connection, and the variants caused by the target damage, account for sample scarcity; (3) The influence of imaging parameters includes elevation angle, frequency, imaging mode, polarization mode, number of sights, signal-to-noise ratio, resolution, etc., results in poor correlation among different targets, which makes the recognition method for one or several SAR targets cannot be quickly applied to other targets, that is, the effectiveness and adaptability of current recognition methods need to be improved [4], [5]. In recent years, the target recognition technology based on artificial ...

show abstract

Section: Related Workmentioning

confidence: 99%