手机知网 App
24小时专家级知识服务
打 开
物理学
High-Quality Sonar Image Generation Algorithm Based on Generative Adversarial Networks
In this paper,we propose a high-quality sonar image generation model based on generative adversarial networks,which is used to generate user-controlled high-resolution and high-quality sonar images,so that the targets in these images have more obvious features and can be further used in engineering applications such as target detection and image classification.As a result of the development of underwater detection technology,sonar target detection technology is widely used in various engineering fields.High resolution and high-quality underwater sonar images have become an indispensable part of underwater detection.However,due to the high cost of sonar and the high cost of marine experiments,sonar datasets are scarce.Even then,few sonar data can be obtained,and most images have no target and have more noise so that they cannot be used for underwater detection.In recent years,using generative adversarial networks to generate images has gradually become an important method of solving the problem of data scarcity.But there are still some shortcomings—the images produced are poor and the resolution is low.To overcome the issue of poor image quality and low resolution caused by gradient disappearing and sudden change in the training process,we,therefore,use a controllable multi-layer transposed convolutional layer and gradient correction term to improve image resolution and image quality.A simple network structure does not make the model difficult to train,and the user may change the size of the generated image by simply changing the convolution parameter i to achieve the objective of improving the resolution.Moreover,the gradient correction term added in the training phase will always restrict the gradient to a certain value.To a certain degree,the problems of gradient disappearing and gradient sudden change in model training can be overcome,thereby providing a guarantee for the production of high-quality images.Experiments demonstrate that this method will successfully increase the resolution and image quality of the GAN-based image and make the image target more apparent.
0 0
手机阅读本文
下载APP 手机查看本文
第40届中国控制会议论文集(6)
2021年
相似文献
图书推荐
相关工具书

搜 索