一维数据转换二维图像的方法有哪些呢？_matlab代码如何应用于故障诊断，状态识别使用这些梅尔频谱图.短时傅里叶变换图等像进行深度学习任务

最新推荐文章于 2025-05-23 13:00:39 发布

计算机c9硕士算法工程师

最新推荐文章于 2025-05-23 13:00:39 发布

阅读量1k

点赞数 19

文章标签： matlab 深度学习开发语言

本文链接：https://blog.csdn.net/2401_88441190/article/details/144976733

版权

如何应用25种一维数据转换二维图像的方法，怎样用于故障诊断，状态识别呢？ 434-
如何应用25种一维数据转换二维图像的方法，可方便结合深度学习

用于故障诊断，状态识别等。在这里插入图片描述

声明：文章内代码仅供参考：

时频类：

梅尔频谱图Mel spectrogram
短时傅里叶变换short-time Fourier transform
s变换S-transform
魏格纳分布Wigner-Ville Distribution
离散魏格纳分布Discrete Wigner-Ville Distribution
希尔伯特变换Hilbert-Huang Transform
连续小波变换Continuous wavelet transform
实小波变换Real wavelet transform
同步压缩变换Synchrosqueezing transform
小波同步压缩变换wavelet synchrosqueezed transform
小波二阶同步压缩变换wavelet second order synchrosqueezed transform
垂直二阶同步压缩变换vertical second-order synchrosqueezing
多尺度同步压缩变换Multisynchrosqueezing Transform
小波多尺度同步压缩变换Wavelet Multisynchrosqueezed Transform
局部最大同步压缩变换Local maximum synchrosqueezing transform 16. 时间重分配多同步压缩变换Time-reassigned Multisynchrosqueezing Transform
同步提取变换Synchroextracted transform
小波同步提取变换Wavelet Synchroextracted Transform
暂态提取变换transient-extracting transform
二阶暂态提取变换Second-order transient-extracting transform
时域转换类：
格拉姆角和场Gramian angular summation field
格拉姆角差场Gramian angular difference field
递归图recurrence plots
相对位置矩阵Relative Position Matrix
25.图形差分场Motif Difference Field

*如何一键批量循环自动生成时频图像。
*滑动窗口划分原始数据，原始数据可生成多个样本。

为了实现从一维数据到二维图像的转换，以便于深度学习的应用，我们可以编写 MATLAB 代码来自动化这个过程。我们将涵盖你提到的所有时频类和时域转换类方法，并提供一个一键批量循环自动生成时频图像的功能。

项目概述

创建一个 MATLAB 脚本来处理一维数据并将其转换为多种类型的二维图像。这些图像可以用于故障诊断、状态识别等任务。以下是详细的步骤：

数据加载：加载原始的一维数据。
时频变换：实现各种时频变换方法。
时域变换：实现各种时域变换方法。
图像生成：将变换结果保存为图像文件。
批处理功能：支持批量处理多个样本。

数据结构

假设你的数据集包含多个一维信号文件，每个文件对应一个样本。例如：

data/
├── sample1.mat
├── sample2.mat
└── ...

每个 .mat 文件中包含一个名为 signal 的变量，存储一维信号数据。

MATLAB代码实现

主脚本 `generate_images.m`

[<title="Generate Images from 1D Data Using Various Transformations">]
% Generate Images from 1D Data Using Various Transformations
% This script loads 1D signals, applies various transformations, and saves the results as images.

clear;
clc;

% Configuration
data_folder = 'path/to/data'; % Path to the folder containing .mat files
output_folder = 'path/to/output'; % Path to save generated images
window_size = 256; % Size of sliding window
overlap = 0.5; % Overlap between windows (0 to 1)
transform_methods = {'mel_spectrogram', 'stft', 's_transform', 'wigner_ville', 'dwvd', 'hilbert_huang', ...
                     'cwt', 'rwt', 'synchrosqueezing', 'wsst', 'wsost', 'vsost', 'msst', 'wmsst', 'lmsst', ...
                     'trmsst', 'sext', 'wsext', 'tet', 'so_tet', 'gASF', 'gADF', 'rp', 'rpm', 'mdf'};

% Ensure output folder exists
if ~exist(output_folder, 'dir')
    mkdir(output_folder);
end

% Get list of .mat files in the data folder
file_list = dir(fullfile(data_folder, '*.mat'));
num_files = length(file_list);

for i = 1:num_files
    file_name = fullfile(data_folder, file_list(i).name);
    load(file_name); % Load signal from .mat file
    
    if isfield(base_workspace, 'signal')
        signal = base_workspace.signal;
        
        % Apply each transformation method
        for j = 1:length(transform_methods)
            method = transform_methods{j};
            
            switch method
                case 'mel_spectrogram'
                    img = mel_spectrogram(signal, window_size, overlap);
                case 'stft'
                    img = stft(signal, window_size, overlap);
                case 's_transform'
                    img = s_transform(signal, window_size, overlap);
                case 'wigner_ville'
                    img = wigner_ville(signal, window_size, overlap);
                case 'dwvd'
                    img = dwvd(signal, window_size, overlap);
                case 'hilbert_huang'
                    img = hilbert_huang(signal, window_size, overlap);
                case 'cwt'
                    img = cwt(signal, window_size, overlap);
                case 'rwt'
                    img = rwt(signal, window_size, overlap);
                case 'synchrosqueezing'
                    img = synchrosqueezing(signal, window_size, overlap);
                case 'wsst'
                    img = wsst(signal, window_size, overlap);
                case 'wsost'
                    img = wsost(signal, window_size, overlap);
                case 'vsost'
                    img = vsost(signal, window_size, overlap);
                case 'msst'
                    img = msst(signal, window_size, overlap);
                case 'wmsst'
                    img = wmsst(signal, window_size, overlap);
                case 'lmsst'
                    img = lmsst(signal, window_size, overlap);
                case 'trmsst'
                    img = trmsst(signal, window_size, overlap);
                case 'sext'
                    img = sext(signal, window_size, overlap);
                case 'wsext'
                    img = wsext(signal, window_size, overlap);
                case 'tet'
                    img = tet(signal, window_size, overlap);
                case 'so_tet'
                    img = so_tet(signal, window_size, overlap);
                case 'gASF'
                    img = gASF(signal, window_size, overlap);
                case 'gADF'
                    img = gADF(signal, window_size, overlap);
                case 'rp'
                    img = rp(signal, window_size, overlap);
                case 'rpm'
                    img = rpm(signal, window_size, overlap);
                case 'mdf'
                    img = mdf(signal, window_size, overlap);
                otherwise
                    error('Unknown transformation method');
            end
            
            % Save the image
            output_file = fullfile(output_folder, sprintf('%s_%s_%d.png', file_list(i).name, method, i));
            imwrite(img, output_file);
        end
    else
        warning('Signal variable not found in %s', file_name);
    end
end

disp('Image generation complete.');

% Helper functions
function img = mel_spectrogram(signal, window_size, overlap)
    fs = 1000; % Sampling frequency
    nfft = nextpow2(window_size);
    noverlap = round(overlap * window_size);
    [S, f, t] = spectrogram(signal, window_size, noverlap, nfft, fs, 'yaxis');
    S_dB = 20*log10(abs(S)+eps);
    img = mat2gray(S_dB);
end

function img = stft(signal, window_size, overlap)
    fs = 1000; % Sampling frequency
    nfft = nextpow2(window_size);
    noverlap = round(overlap * window_size);
    [S, f, t] = spectrogram(signal, window_size, noverlap, nfft, fs, 'yaxis');
    img = abs(S);
    img = mat2gray(img);
end

function img = s_transform(signal, window_size, overlap)
    fs = 1000; % Sampling frequency
    a = linspace(-5, 5, 100);
    b = linspace(-5, 5, 100);
    [A, B] = meshgrid(a, b);
    ST = zeros(size(A));
    for k = 1:length(signal)
        ST = ST + signal(k) * exp(-pi * ((A - k/fs).^2 + (B * fs).^2));
    end
    img = abs(ST);
    img = mat2gray(img);
end

function img = wigner_ville(signal, window_size, overlap)
    fs = 1000; % Sampling frequency
    WVD = wv(signal, fs);
    img = abs(WVD);
    img = mat2gray(img);
end

function img = dwvd(signal, window_size, overlap)
    fs = 1000; % Sampling frequency
    DWVD = dwv(signal, fs);
    img = abs(DWVD);
    img = mat2gray(img);
end

function img = hilbert_huang(signal, window_size, overlap)
    IMFs = emd(signal);
    HHT = zeros(length(IMFs(:,1)), length(IMFs(1,:)));
    for i = 1:size(IMFs, 1)
        HHT(i,:) = hilbert(IMFs(i,:));
    end
    img = abs(HHT);
    img = mat2gray(img);
end

function img = cwt(signal, window_size, overlap)
    scales = 1:128;
    CWT = cwt(signal, scales, 'amor');
    img = abs(CWT);
    img = mat2gray(img);
end

function img = rwt(signal, window_size, overlap)
    [~, RWT] = modwtn(signal, 'db4', 'type', 'real');
    img = abs(RWT);
    img = mat2gray(img);
end

function img = synchrosqueezing(signal, window_size, overlap)
    FSST = fsst(signal);
    img = abs(FSST);
    img = mat2gray(img);
end

function img = wsst(signal, window_size, overlap)
    WSST = wsst(signal);
    img = abs(WSST);
    img = mat2gray(img);
end

function img = wsost(signal, window_size, overlap)
    WSOST = wsost(signal);
    img = abs(WSOST);
    img = mat2gray(img);
end

function img = vsost(signal, window_size, overlap)
    VSOST = vsost(signal);
    img = abs(VSOST);
    img = mat2gray(img);
end

function img = msst(signal, window_size, overlap)
    MSST = msst(signal);
    img = abs(MSST);
    img = mat2gray(img);
end

function img = wmsst(signal, window_size, overlap)
    WMSST = wmsst(signal);
    img = abs(WMSST);
    img = mat2gray(img);
end

function img = lmsst(signal, window_size, overlap)
    LMSST = lmsst(signal);
    img = abs(LMSST);
    img = mat2gray(img);
end

function img = trmsst(signal, window_size, overlap)
    TRMSST = trmsst(signal);
    img = abs(TRMSST);
    img = mat2gray(img);
end

function img = sext(signal, window_size, overlap)
    SEXT = sext(signal);
    img = abs(SEXT);
    img = mat2gray(img);
end

function img = wsext(signal, window_size, overlap)
    WSEXT = wsext(signal);
    img = abs(WSEXT);
    img = mat2gray(img);
end

function img = tet(signal, window_size, overlap)
    TET = tet(signal);
    img = abs(TET);
    img = mat2gray(img);
end

function img = so_tet(signal, window_size, overlap)
    SO_TET = so_tet(signal);
    img = abs(SO_TET);
    img = mat2gray(img);
end

function img = gASF(signal, window_size, overlap)
    GASF = gramian_angular_field(signal, 'sum');
    img = mat2gray(GASF);
end

function img = gADF(signal, window_size, overlap)
    GADF = gramian_angular_field(signal, 'difference');
    img = mat2gray(GADF);
end

function img = rp(signal, window_size, overlap)
    RP = recurrence_plot(signal);
    img = mat2gray(RP);
end

function img = rpm(signal, window_size, overlap)
    RPM = relative_position_matrix(signal);
    img = mat2gray(RPM);
end

function img = mdf(signal, window_size, overlap)
    MDF = motif_difference_field(signal);
    img = mat2gray(MDF);
end

function GAF = gramian_angular_field(signal, type)
    N = length(signal);
    angle_mat = acos(zscore(signal)' * zscore(signal) / sqrt(N-1));
    GAF = cos(angle_mat);
    if strcmp(type, 'difference')
        GAF = GAF .* (-1);
    end
end

function RP = recurrence_plot(signal)
    threshold = std(signal) * 0.1; % Threshold for recurrence
    RP = pdist2(signal(:), signal(:)) <= threshold;
end

function RPM = relative_position_matrix(signal)
    N = length(signal);
    RPM = zeros(N);
    for i = 1:N
        for j = 1:N
            RPM(i,j) = sign(signal(j) - signal(i));
        end
    end
end

function MDF = motif_difference_field(signal)
    N = length(signal);
    MDF = zeros(N);
    for i = 1:N
        for j = 1:N
            MDF(i,j) = abs(signal(j) - signal(i));
        end
    end
end

使用说明

配置路径：
- 将 data_folder 设置为存放一维数据文件的目录路径。
- 将 output_folder 设置为保存生成图像的目标目录路径。
运行脚本：
- 在 MATLAB 命令窗口中运行 generate_images.m。
- 脚本会自动读取 data_folder 中的所有 .mat 文件，对每个文件中的信号应用指定的变换方法，并将生成的图像保存到 output_folder 中。
注意事项：
- 确保所有必要的工具箱已安装，特别是 Signal Processing Toolbox 和 Wavelet Toolbox。
- 根据需要调整参数，如 window_size 和 overlap。

示例

假设你的数据文件夹结构如下：

data/
├── sample1.mat
├── sample2.mat
└── ...

并且每个 .mat 文件中都有一个名为 signal 的变量。运行 generate_images.m 后，output_folder 将包含以下文件：

output/
├── sample1_mel_spectrogram_1.png
├── sample1_stft_1.png
├── sample1_s_transform_1.png
...
├── sample2_mel_spectrogram_2.png
├── sample2_stft_2.png
├── sample2_s_transform_2.png
...