[1]王 诏*,王 燕,苏国辉,等.一种阶梯型多尺度神经网络条带噪声降噪模型[J].计算机技术与发展,2023,33(07):68-74.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 010]
　WANG Zhao*,WANG Yan,SU Guo-hui,et al.A Staircase Multi-scale Neural Network Model for Stripe Noise Reduction[J].,2023,33(07):68-74.[doi:10. 3969 / j. issn. 1673-629X. 2023. 07. 010]

点击复制

一种阶梯型多尺度神经网络条带噪声降噪模型()

分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年07期

页码:

68-74

栏目:

媒体计算

出版日期:

2023-07-10

文章信息/Info

Title:

A Staircase Multi-scale Neural Network Model for Stripe Noise Reduction

文章编号:

1673-629X(2023)07-0068-07

作者:

王 诏* ; 王 燕; 苏国辉; 史升凯

青岛海洋地质研究所,山东 青岛 266071

Author(s):

WANG Zhao* ; WANG Yan; SU Guo-hui; SHI Sheng-kai

Qingdao Institute of Marine Geology,Qingdao 266071,China

关键词:

降噪; 阶梯型多尺度; 混合空洞卷积; 卷积神经网络; 遥感影像; 条带噪声

Keywords:

denoise; staircase multi-scale; hybrid dilated convolution; convolutional neural network; remote sensing image; stripe noise

分类号:

TP751

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 07. 010

摘要:

针对遥感影像上存在的复杂条带噪声问题,提出一种阶梯型多尺度卷积神经网络降噪模型( SmCNN) 。 该模型采用一种阶

梯型网络结构轻量化设计,高度融合高、低层等多层网络特征,对影像上的非均质、高覆盖、多尺度条带噪声具备优秀的识别、去除能力。 模型主要通过 3*3 空洞卷积构建多尺度网络和残差网络融合多尺度特征信息,增强多尺度条带噪声的检测

能力、弥补深度网络退化缺点;采用锯齿状混合空洞卷积网络结构解决图像信息不连续问题;设计阶梯型多尺度网络结构、

引入 1*1 卷积以轻量化神经网络模型,降低模型复杂度。 实验结果表明,SmCNN 降噪性能明显优于传统图像降噪方法,比经典的前馈去噪卷积神经网络( DnCNN) 降噪模型在均方误差、峰值信噪比、结构相似性三项图像降噪质量评价指标上,分别提高了 61. 2% 、11. 8% 和 0. 7% ,且网络结构轻量化效果显著,节约了 53. 11% 的模型训练时间。

Abstract:

In view of the complex stripe noise in remote sensing images, a staircase multi - scale convolutional neural network noisereduction model ( SmCNN) is proposed. The model adopts a lightweight design of the staircase network structure,　highly integrates multi-layer network features such as high and low layers, and has excellent recognition and removal capabilities for heterogeneous, highcoverage,multi - scale stripe noise on images. The model mainly uses 3 *3 dilated convolution to construct multi - scale network andresidual network to integrate multi-scale feature information,enhance the detection ability of multi-scale stripe noise and make up for thedegradation of deep network. The jagged hybrid dilated convolution network is used to solve the problem of image informationdiscontinuity. The multi-scale network structure of staircase type is designed,and the 1 *1 convolutional lightweight neural networkmodel is introduced to?

reduce the complexity of the model. The experimental results show that SmCNN has better denoising performance

than traditional image denoising methods. Compared with the classical DnCNN model,the denoising quality evaluation indexes of meansquare error,peak signal - to - noise ratio and structural similarity are improved by 61. 2% , 11. 8% and?

0. 7% respectively. Thelightweight effect of network structure is remarkable,and 53. 11% of the model training time is saved.

常用功能