摘要
针对由于光在水中传播所带来的影响,导致所获得的水下图像不清晰以及颜色失真的问题,提出了一种基于条件生成对抗网络(CGAN)的水下图像增强算法。为了达到更好的增强效果,利用完全配对的水下图像与清晰图像进行模型的训练,通过端到端的方式获取增强图像。在生成网络模型中,采用U-Net网络结构进行网络的信息减负,同时为了捕捉到更多的低频特征,在损失函数中引入L 1损失,让生成的结果更加真实和清晰。通过最后的实验结果表明:训练的模型有效解决了水下图像的颜色失真与模糊问题,对水下图像有不错的增强效果。
An underwater image enhancement algorithm based on conditional generation adversarial networks(CGAN)is proposed to solve the problem of unclear underwater image and color distortion caused by light propagation in water.In order to achieve better enhancement effect,fully matched input and output images are used for model training to obtain enhanced images end-to-end.In generated network model,U-Net network structure is adopted to reduce network information load.Meanwhile,in order to capture more low-frequency features,L 1 loss function is introduced to make the generated results more real and clear.Experimental results show that the training model can effectively solve the problem of color distortion and blur in the underwater image,and has a good enhancement effect on the underwater image.
作者
杨国亮
王杨
赖振东
YANG Guoliang;WANG Yang;LAI Zhendong(College of Electrical Engineering and Automation,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《传感器与微系统》
CSCD
北大核心
2021年第5期121-123,共3页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(51365017)。
关键词
水下图像增强
条件生成对抗网络
深度学习
underwater image enhancement
conditional generation adversarial networks(CGAN)
deep learning
