本文主要研究内容
作者郑升(2019)在《基于RCSST和CNN的微弱地震信号噪声压制方法研究》一文中研究指出:地震勘探是处理油气勘探的主要方法,经过观察和剖析人工激发的地震波在地下的传播规律,帮助人们探寻到地下的油气资源。但是在检波器检测到的地震信号中,常伴有很多的干扰噪声,这严重地影响到有效地震信号的识别。同时由于常规油气田已经到达了开采末期,人们逐步的把目光转向非常规油气田,由此获得的勘探信号中有效信号能量更低,噪声能量更高,地震信号的噪声压制问题也变得更加困难,这极大地增加了地震资料的解释难度。因而抑制地震信号中的干扰噪声,提升地震资料的信噪比(signal to noise ratio,SNR)是地震勘探数据处理的关键环节。本文研究重点是山地地区地震信号和沙漠地区地震信号的噪声消除问题,针对两种地震信号的不同特征,本文基于Shearlet去噪算法提出了两种改进思路,并通过合成记录和实际记录的仿真实验,验证了改进算法的有效性。基于硬阈值的Shearlet变换充分利用了Shearlet变换和硬阈值的优点,在处理地震信号干扰噪声的过程中取得了一定的效果。但是这种方法是通过选取固定的阈值来压制噪声的,因此在去除噪声的同时很容易过度扼杀有效信号。而且去噪后的地震信号会出现虚假同相轴,影响去噪效果。本文提出了一种基于自适应阈值的递归循环平移shearlet变换(recursive cycle spinning shearlet transform,RCSST)去噪算法。首先,我们采用了递归循环平移(recursive cycle spinning,RCS)与Shearet变换相结合的方法变换分解地震资料,接着采用根据Shearlet系数的能量大小而自适应变化的阈值处理变换系数防止系数被过度扼杀,保护有效信号的幅度,同时有效抑制随机噪声,增强有效信号的连续性。实验结果表明,当信噪比较低的时候,该方法可以比传统的方法更好的压制随机噪声和保护有效信号。近年来,随着计算机水平的持续提高,卷积神经网络(Convolutional Neural Network,CNN)得到了迅速的发展。作为一种先进的深度学习算法,卷积神经网络已经在图像信号处理和语音信号处理等领域取得了突破性的进展。本文根据沙漠地震信号低信噪比的特点,提出了一种基于Shearlet变换的深度残差卷积神经网络(Deep Residual Convolutional Neural Network for Shearlet Transform,ST-CNN)模型,完成对沙漠地震数据的噪声消减。本文模型分为两个过程:训练过程和测试过程。在训练过程中,我们把变换后的沙漠地震数据Shearlet系数输入到网络模型中,把变换后的噪声Shearlet系数当成训练标签,利用卷积神经网络学习拟合到输入和标签间的映射关系。在最后的测试过程,我们就可以使用这个映射关系,在地震数据Shearlet系数的基础上,获取到干扰噪声的Shearlet系数,然后得到纯净信号的Shearlet系数,再经过逆变换获得去噪后的沙漠地震信号。为了验证其有效性,我们把此算法应用到沙漠地震合成记录和实际记录的仿真实验中,实验结果表明,基于ST-CNN模型的去噪算法在同相轴(有效信号)的恢复和噪声的消减方面有较大进步,信噪比也有很大的提高。
Abstract
de zhen kan tan shi chu li you qi kan tan de zhu yao fang fa ,jing guo guan cha he pou xi ren gong ji fa de de zhen bo zai de xia de chuan bo gui lv ,bang zhu ren men tan xun dao de xia de you qi zi yuan 。dan shi zai jian bo qi jian ce dao de de zhen xin hao zhong ,chang ban you hen duo de gan rao zao sheng ,zhe yan chong de ying xiang dao you xiao de zhen xin hao de shi bie 。tong shi you yu chang gui you qi tian yi jing dao da le kai cai mo ji ,ren men zhu bu de ba mu guang zhuai xiang fei chang gui you qi tian ,you ci huo de de kan tan xin hao zhong you xiao xin hao neng liang geng di ,zao sheng neng liang geng gao ,de zhen xin hao de zao sheng ya zhi wen ti ye bian de geng jia kun nan ,zhe ji da de zeng jia le de zhen zi liao de jie shi nan du 。yin er yi zhi de zhen xin hao zhong de gan rao zao sheng ,di sheng de zhen zi liao de xin zao bi (signal to noise ratio,SNR)shi de zhen kan tan shu ju chu li de guan jian huan jie 。ben wen yan jiu chong dian shi shan de de ou de zhen xin hao he sha mo de ou de zhen xin hao de zao sheng xiao chu wen ti ,zhen dui liang chong de zhen xin hao de bu tong te zheng ,ben wen ji yu Shearletqu zao suan fa di chu le liang chong gai jin sai lu ,bing tong guo ge cheng ji lu he shi ji ji lu de fang zhen shi yan ,yan zheng le gai jin suan fa de you xiao xing 。ji yu ying yu zhi de Shearletbian huan chong fen li yong le Shearletbian huan he ying yu zhi de you dian ,zai chu li de zhen xin hao gan rao zao sheng de guo cheng zhong qu de le yi ding de xiao guo 。dan shi zhe chong fang fa shi tong guo shua qu gu ding de yu zhi lai ya zhi zao sheng de ,yin ci zai qu chu zao sheng de tong shi hen rong yi guo du e sha you xiao xin hao 。er ju qu zao hou de de zhen xin hao hui chu xian xu jia tong xiang zhou ,ying xiang qu zao xiao guo 。ben wen di chu le yi chong ji yu zi kuo ying yu zhi de di gui xun huan ping yi shearletbian huan (recursive cycle spinning shearlet transform,RCSST)qu zao suan fa 。shou xian ,wo men cai yong le di gui xun huan ping yi (recursive cycle spinning,RCS)yu Shearetbian huan xiang jie ge de fang fa bian huan fen jie de zhen zi liao ,jie zhao cai yong gen ju Shearletji shu de neng liang da xiao er zi kuo ying bian hua de yu zhi chu li bian huan ji shu fang zhi ji shu bei guo du e sha ,bao hu you xiao xin hao de fu du ,tong shi you xiao yi zhi sui ji zao sheng ,zeng jiang you xiao xin hao de lian xu xing 。shi yan jie guo biao ming ,dang xin zao bi jiao di de shi hou ,gai fang fa ke yi bi chuan tong de fang fa geng hao de ya zhi sui ji zao sheng he bao hu you xiao xin hao 。jin nian lai ,sui zhao ji suan ji shui ping de chi xu di gao ,juan ji shen jing wang lao (Convolutional Neural Network,CNN)de dao le xun su de fa zhan 。zuo wei yi chong xian jin de shen du xue xi suan fa ,juan ji shen jing wang lao yi jing zai tu xiang xin hao chu li he yu yin xin hao chu li deng ling yu qu de le tu po xing de jin zhan 。ben wen gen ju sha mo de zhen xin hao di xin zao bi de te dian ,di chu le yi chong ji yu Shearletbian huan de shen du can cha juan ji shen jing wang lao (Deep Residual Convolutional Neural Network for Shearlet Transform,ST-CNN)mo xing ,wan cheng dui sha mo de zhen shu ju de zao sheng xiao jian 。ben wen mo xing fen wei liang ge guo cheng :xun lian guo cheng he ce shi guo cheng 。zai xun lian guo cheng zhong ,wo men ba bian huan hou de sha mo de zhen shu ju Shearletji shu shu ru dao wang lao mo xing zhong ,ba bian huan hou de zao sheng Shearletji shu dang cheng xun lian biao qian ,li yong juan ji shen jing wang lao xue xi ni ge dao shu ru he biao qian jian de ying she guan ji 。zai zui hou de ce shi guo cheng ,wo men jiu ke yi shi yong zhe ge ying she guan ji ,zai de zhen shu ju Shearletji shu de ji chu shang ,huo qu dao gan rao zao sheng de Shearletji shu ,ran hou de dao chun jing xin hao de Shearletji shu ,zai jing guo ni bian huan huo de qu zao hou de sha mo de zhen xin hao 。wei le yan zheng ji you xiao xing ,wo men ba ci suan fa ying yong dao sha mo de zhen ge cheng ji lu he shi ji ji lu de fang zhen shi yan zhong ,shi yan jie guo biao ming ,ji yu ST-CNNmo xing de qu zao suan fa zai tong xiang zhou (you xiao xin hao )de hui fu he zao sheng de xiao jian fang mian you jiao da jin bu ,xin zao bi ye you hen da de di gao 。
论文参考文献
论文详细介绍
论文作者分别是来自吉林大学的郑升,发表于刊物吉林大学2019-06-25论文,是一篇关于微弱地震信号论文,变换论文,递归循环平移论文,卷积神经网络论文,残差模型论文,吉林大学2019-06-25论文的文章。本文可供学术参考使用,各位学者可以免费参考阅读下载,文章观点不代表本站观点,资料来自吉林大学2019-06-25论文网站,若本站收录的文献无意侵犯了您的著作版权,请联系我们删除。
标签:微弱地震信号论文; 变换论文; 递归循环平移论文; 卷积神经网络论文; 残差模型论文; 吉林大学2019-06-25论文;