基于离散剪切波与优化深度卷积神经网络的图像降噪方法
Image Denoising Method Based on Discrete Shearlet Transform and Optimized Deep Convolutional Neural Network
白华军 1李荣昌 1司洁戈 1张义 1张景熙1
作者信息
- 1. 中国电子科技集团公司第三研究所,北京 100015
- 折叠
摘要
海洋试验图像通常受到海洋气象条件、海水光照折射和海洋深度等因素的影响,导致在海洋中采集的图像包含严重的噪声.为了提高海洋试验图像的清晰度和降噪性,提出一种基于离散剪切波与优化深度卷积神经网络相结合的海洋试验图像降噪方法.采用离散剪切波变换分解海洋试验图像,能有效从图像中提取不同方向和频率的特征.利用优化深度卷积神经网络强大的图像特征提取能力,经网络模型训练后,能获取图像中的关键特征,达到降噪的目的.在验证实验中,所提方法与传统图像降噪方法相比,能有效保留图像的纹理和细节特性,获得了较好的降噪效果,有助于提高海洋试验图像的清晰度和降噪性.
Abstract
Marine test images are often affected by factors such as marine meteorological conditions,seawater light refraction,and ocean depth,resulting in severe noise in the images collected in the ocean.In order to improve the clarity and denoising performance of marine test images,this paper proposes a marine test image denoising method based on discrete shearlet transform combined with an optimized deep convolutional neural network.The discrete shearlet transform is used to decompose the marine test image,which can effectively extract features of different directions and frequencies from the image.By utilizing the powerful feature extraction ability of the optimized deep convolutional neural network,after training the network model,key features in the image can be obtained,thus achieving the purpose of denoising.In the verification experiment,compared with traditional image denoising methods,the proposed method can effectively retain the texture and detail characteristics of the image,achieve good denoising effect,and improve the clarity and denoising performance of marine test images.
关键词
离散剪切波变换/降噪方法/深度卷积神经网络/海洋试验Key words
discrete shearlet transform/denoising methods/deep convolutional neural networks/ocean experiments引用本文复制引用
出版年
2024
NETL
NSTL
万方数据