【GRNN情绪识别】基于GRNN神经网络的情绪识别算法matlab仿真

article/2025/9/29 0:36:56

1.软件版本

matlab2021a

2.本算法理论知识

GRNN，即General Regression Neural Network，中文全称为广义回归神经网络，是由The Lockheed Palo Alto研究实验室在1991年提出的。GRNN是一种新型的基于非线性回归理论的神经网络模型[43,44]。GRNN是建立在非参数核回归基础之上的，该神经网络是以测试样本为后验条件，并从观测样本中计算得到自变量和因变量之间的概率密度函数，然后在计算出因变量关于自变量的回归值。由于GRNN不需要规定模型的类型，只需要设置神经网络的光滑因子参数，GRNN神经网络的光滑因子参数的取值对神经网络的输出影响较大，当光滑因子参数较大的时候，其对应的神经元所覆盖的输入区域就越大；当光滑因子参数较小的时候，神经网络对应的径向基函数曲线较陡，因此神经网络输出结果更接近期望值，但此时光滑度越差。由于GRNN广义回归神经网络是基于非线性核回归分析的神经网络，因此，对于任意一个非独立变量y，其相对于独立变量x的回归分析的过程是计算具有最大概率值y。现假设随机变量x和y的联合概率密度函数为f (x ,y)，已知x的观测值为X，则y相对于X的回归，即条件均值为：

从图的结构图可知，GRNN神经网络的输入层神经元数目和输入样本的维度是相同的，即每一个神经元将输入信号直接传递给GRNN神经网络的隐含层中。GRNN神经网络的模式层的神经元数目和学习训练样本的数目相同，即每一个神经元都分别对应着一个不同的学习训练样本，模式层中神经元的传递函数为：

3.核心代码

clc;
clear;
close all;
warning off;
addpath 'func\'
addpath 'func\Grnn\'
addpath 'func\PCA\'
%是否需要重新训练，1，重新训练，0不训练，直接测试
SEL                  = 1;
nPerson              = 9;
nExpressionPerPerson = 7;TEST_DATA            = 2;%1：选择测试样本集合1,2：选择测试样本集合2DIM                  = 120;SET = [];
for flag = 1:3
flagif  SEL == 1display('读入训练集合...');if TEST_DATA == 1%第一组样本TrainData  = readfaceFeature(nExpressionPerPerson, nPerson,0,130);trainLabel = [1,1,2,2,3,3,3,4,4,4,5,5,6,6,7,7,1,1,2,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,3,3,4,4,4,5,5,6,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,4,5,5,6,6,7,1,1,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,3,4,4,5,5,6,6,7,7]';%降低维度pcaFaces     = fastPCA(TrainData,DIM); % 主成分分析PCAendif TEST_DATA == 2%第二组样本jaffeTrainData  = readfaceFeature2(nExpressionPerPerson, nPerson,0,140,flag);trainLabel1 = [1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7]';trainLabel = [trainLabel1;trainLabel1];          %降低维度pcaFaces     = fastPCA(TrainData,DIM); % 主成分分析PCAendX          = pcaFaces; TrainData  = X;[X,A0,B0]  = scaling(X);net        = multiGRNNTrain(TrainData',trainLabel',0.0015);save NN_model.mat net A0 B0display('..............................');display('训练结束。');
enddisplay('读入测试集合...');
%降低维度
load NN_model.mat
load PCA.matif TEST_DATA == 1%第一组样本TestData0  = readfaceFeature(nExpressionPerPerson, nPerson,1,130);TestLabel  = [1,1,2,2,3,3,3,4,4,4,5,5,6,6,7,7,1,1,2,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,3,3,4,4,4,5,5,6,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,4,5,5,6,6,7,1,1,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,4,4,5,5,6,6,7,7,1,1,2,2,3,3,3,4,4,5,5,6,6,7,7]';[m,n]      = size(TestData0);TestData   = (TestData0-repmat(meanVec, m, 1))*V; % 经过pca变换降维
endif TEST_DATA == 2%第二组样本jaffeTestData0  = readfaceFeature2(nExpressionPerPerson, nPerson,1,70,flag);TestLabel = [1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7,...1,2,3,4,5,6,7]';[m,n]      = size(TestData0);TestData   = (TestData0-repmat(meanVec, m, 1))*V; % 经过pca变换降维
endclasses    = multiGRNNClassify(TestData',net);     
nError     = sum(classes ~= TestLabel);
accuracy   = 1 - nError/length(TestLabel);N1=0;
N2=0;
N3=0;
N4=0;
N5=0;
N6=0;
N7=0;for i = 1:length(TestLabel)if TestLabel(i) == 1 & classes(i) == 1N1 = N1 + 1; endif TestLabel(i) == 2 & classes(i) == 2N2 = N2 + 1;  endif TestLabel(i) == 3 & classes(i) == 3N3 = N3 + 1;  end    if TestLabel(i) == 4 & classes(i) == 4N4 = N4 + 1;  end    if TestLabel(i) == 5 & classes(i) == 5N5 = N5 + 1;  endif TestLabel(i) == 6 & classes(i) == 6N6 = N6 + 1;  endif TestLabel(i) == 7 & classes(i) == 7N7 = N7 + 1;  end
endSET = [SET;[N1,N2,N3,N4,N5,N6,N7]/10];end

4.操作步骤与仿真结论