摘要
针对传统方法压缩图像时出现的编码增益状态不佳与压缩效率较低的问题,提出了基于深度学习网络的激光光谱图像压缩方法。通过DPCM预测算法消除激光光谱图像的谱间冗余,再利用SPIHT算法消除剩下的空间冗余,进而使图像的残差值缩小。在此基础上使用深度学习网络中的卷积神经网络把多种卷积层与非线性激活层叠加在一起来完成对复杂反射函数的处理,进而实现大量数据训练。最后经过多层卷积的感受野对数据进行压缩,从而实现对激光光谱图像的压缩。实验证明,与传统方法对比,本文方法的编码增益状态更佳,且图像压缩效率较高,能够有效对激光光谱图像进行压缩,具有较高的应用价值。
In order to solve the problems of poor coding gain and low compression efficiency in traditional methods,a multi-spectral image compression method based on deep learning algorithm is proposed.The DPCM prediction algorithm is used to eliminate the spectral redundancy of multi-spectral image,and the SPIHT algorithm is used to eliminate the remaining spatial redundancy,so that the residual value of the image is reduced.On this basis,the convolution neural network in deep learning algorithm is used to stack a variety of convolution layers and nonlinear activation layers to complete the processing of complex reflection functions.And then,a large number of data training is realized.Finally,the multi-spectral image is compressed by the multi-layer convolution receptive field.Experiments show that compared with the traditional methods,the coding gain state of this method is better,and the image compression efficiency is higher,which can effectively compress the multi-spectral image,and has higher application value.
作者
蒋媛
魏瑞
卢超
JIANG Yuan;WEI Rui;LU Chao(School of Physical and Telecommunications Engineering,Shaanxi University of Technology,Hanzhong Shaanxi 723000,China)
出处
《激光杂志》
北大核心
2020年第12期176-180,共5页
Laser Journal
基金
国家自然科学基金(No.61972239,61772398)
陕西省重点研发项目(No.2019SF-257)
陕西理工大学科学研究计划项目(No.SLGKY16-09,SLG1808)。
关键词
深度学习
卷积神经网络
激光光谱图像
压缩效率
deep learning
convolution neural network
multi-spectral image
compression efficiency
