Save a trained neural network for Raspberry

Question

David Duran 2022-1-18

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1630890-save-a-trained-neural-network-for-raspberry

回答： Yazan 2022-9-28

I have loaded AlexNet and retrained the neural network using my own images. When I run the script it it does everything properly and trains the NN properly. The new trained NN is saved to the workspace as myNet as as a .mat file. I was wondering if it's possible to save it as an .M file so I can deploy it to a Raspberry PI 4.

clc;
clear all;
close all;
alex = alexnet; % Load Neural Network AlexNet
layers = alex.Layers; %Identify Layers in Alexnet 
layers(23) = fullyConnectedLayer(6); %Six Unique Images will be identified at Layer 23
layers(25) = classificationLayer %Layer 25 is the Classification Layer and properly identifies the object 
allImages = imageDatastore('robot','IncludeSubfolders', true, 'LabelSource', 'foldernames'); % Identifies the images that are stored are the computer and the labels associated with them
[trainingImages, testImages] = splitEachLabel(allImages, 0.8, 'randomized'); % seperates the files into different groups for training and validation. (80% for training and 20% for validation)
opts = trainingOptions("sgdm", 'InitialLearnRate', 0.001, 'MaxEpochs', 5, 'MiniBatchSize', 64, 'Plots','training-progress'); % Different options for training the Neural Network
myNet = trainNetwork(trainingImages, layers,opts); %Will save the trained network based off AlexNet as myNet
predictedLabels = classify(myNet, testImages); % This will identify images using the trained network by using the validation images
accuracy = mean(predictedLabels ==testImages.Labels) % This will display a percentage of how accurate the model is at identifying the images.
camera = webcam; % Connect to the camera
while true   
    picture = camera.snapshot;              % Take a picture using connected Camera
    picture = imresize(picture,[227,227]);  % Resize the picture to 277x277 Pixel size
    label = classify(myNet, picture);        % Classify the picture
       
    image(picture);     % Displays the picture
    title(char(label)); % Shows the label above the picture
    drawnow; % Continuously updates image from Webcam
end

2 个评论
显示无隐藏无

yanqi liu 2022-1-18

yes，sir，i think mat file include weight data，if use m file，may be only the layers structure

David Duran 2022-1-18

在 MATLAB Online 中打开

So what would be the best way to tranfer the trained NN to a Raspberry PI? I have watched the videos provided by MATLAB, but they are pulling the untrained neural networks from the Program files.

i.e. the code

edit_raspi_webcam_resnet

请先登录，再进行评论。

请先登录，再回答此问题。