去噪学习记录

luoxinrui7

已于 2024-03-27 21:30:26 修改

阅读量132

点赞数 1

文章标签： python 人工智能机器学习

于 2024-03-27 16:50:38 首次发布

本文链接：https://blog.csdn.net/Albuquerque7/article/details/137082033

版权

文章介绍了使用方差比率来确定纯背景噪声范围的方法，以及基于累积分布函数(CDF)进行信号去噪的过程，包括硬阈值处理和保存处理后的信号和时间频率谱。

摘要由CSDN通过智能技术生成

ROV

利用方差比率找到纯背景噪声范围：通过计算在不同位置的方差比率来确定噪声的范围。

noise  = zeros(1, id);
start_s = floor(id / nrs);
end_s = start_s;

for i = start_s + 1: id - end_s
    noise(i) = var(s(1:i)) / var(s(i:end));
end
    [~,noise_t] = min(noise(start_s + 1: end - end_s));
    noise_tt  = noise_t + start_s + 1;  % The range of pure background noise： 0~ noise_tt
dn.rov = noise; % saving the ROV curve

nrs：表示用于计算方差比率的窗口的大小。

通过循环遍历信号中每个位置，计算当前位置之前和之后子信号的方差并计算它们的比率，在noise向量的对应位置存储。最后找到noise向量中最小值的索引，以确定纯背景噪声的范围，这个索引加上srart_s+1得到了纯背景噪声的结束索引noise_tt.

CDF基于累积分布函数

%%  The denoising method based on CDF thresholding
f_x = zeros(size(Tx, 1), size(Tx, 2));
thres = zeros(size(Tx, 1), 1);
for i = fi1 : 1 : fi2
    scal = Tx(i,:);
    x = abs(scal);
    y = sort(x);
    len = id;
    ii = floor(id * nnum);  %
    x1 = x(1: noise_tt);
    [ecdf_f, xx] = ecdf(x1);
    ii = find(ecdf_f >= ECDF_threshold);
    thre = xx(min(ii));
    thres(i) = thre;
    x(find(abs(x) < thre)) = 0;
    x(find(abs(x) >= thre)) = 1;
    
    f_x(i,:) = x;
end
dn.nr = noise_tt;
dn.thres = thres; % saving wavelet coefficient threshold for each frequency
Tx1 = Tx.*f_x;
s_0 = iwsst(Tx1, f, [f(1), f(end)]); % denoising waveform after hard thresholding based on ECDF threshold
dn.ecdf_tf = Tx1; % saving denoising time-frequency spectrum after hard thresholding based on ECDF threshold
dn.ecdf_dw = s_0; % saving denoising waveform  after hard thresholding based on ECDF threshold