基于生成对抗网络的声呐图像超分辨率算法

Sonar image super-resolution algorithm Based on Generative Adversarial Network

摘要:

由于受到复杂水声信道影响,侧扫声呐生成图像一般存在分辨率低、细节模糊等问题,这给基于侧扫声呐图像的小目标检测与识别过程带来了一定难度。常见的图像超分辨率处理算法大致可以分为插值算法、稀疏表示的重建算法等。传统插值算法存在图像边缘模糊,细节不清楚的问题,本文提出了一种改进型超分辨率生成对抗网络(SRGAN)结构,该结构可在保留图像中的小目标细节的基础上提升细节特征显示度,并且可使目标边缘更加清晰,可为后续的小目标检测和识别在一定程度上提供技术支持。

Abstract:

Due to the influence of complex underwater acoustic channels, the image generated by side scan sonar have problems such as low resolution and blurred details, which brings some difficulties to the small target detection and recognition process based on side scan sonar images. The common image super-resolution processing algorithms can be roughly divided into interpolation algorithm, sparse representation reconstruction algorithm, etc. The traditional interpolation algorithm has the problems of blurred image edges and unclear details. This paper proposes an improved super resolution generation countermeasure network (SRGAN) structure. This structure can improve the display of detail features on the basis of retaining the details of small targets in the image, and make the target edges clearer, which can provide technical support for the subsequent small target detection and recognition to a certain extent.