基于MATLAB深度学习工具箱的CNN卷积神经网络训练和测试仿真,含仿真操作录像

clc; clear; close all; warning off; addpath(genpath(pwd)); rng('default') trainD(:,:,:,1)=rand(22,5,3); trainD(:,:,:,2)=rand(22,5,3); trainD(:,:,:,3)=rand(22,5,3); trainD(:,:,:,4)=rand(22,5,3); targetD=categorical([0;0;1;1]); %% Define Network Architecture % Define the convolutional neural network architecture. layers = [ imageInputLayer([22 5 3]) % 22X5X3 refers to number of features per sample convolution2dLayer(5,16,'Padding','same') % 5x5 filtr is used, u can try 3x3 filtr also % padding -- same, means output of conv layer is same siz as input(automatic padding) % stride if not speficied means its 1; stride window movement steps os % speed over image % vist this website for easy cnn understanding, http://cs231n.github.io/convolutional-networks/ reluLayer % activation function % i have not used any pooling layer here, since small data size % if u giving big data use pooling layer % pooling layer reduces size of the matrix fullyConnectedLayer(384) % 384 refers to number of neurons in next FC hidden layer fullyConnectedLayer(384) % 384 refers to number of neurons in next FC hidden layer fullyConnectedLayer(2) % 2 refers to number of neurons in next output layer (number of output classes) softmaxLayer classificationLayer]; options = trainingOptions('sgdm',... 'MaxEpochs',500, ... 'Verbose',false,... 'Plots','training-progress'); net = trainNetwork(trainD,targetD',layers,options); predictedLabels = classify(net,trainD)'