Import Keras Tensorflow Model into Simulink

I've got a trained model from Python, which I need to run on Simulink for my application.

Is it possible to import trained Tensorflow Keras models and implement them in Simulink?

Cheers,

Lucas

Answer 1

[sruthi gundeti]

Hi,

You can import your keras network using

https://in.mathworks.com/help/deeplearning/ref/importkerasnetwork.html#d120e51332

net = importKerasNetwork(modelfile)
net = importKerasNetwork(modelfile,Name,Value)

model file can .h5 format ot json format with weights

Then write a entry level function for the imported network

https://in.mathworks.com/help/gpucoder/ref/coder.loaddeeplearningnetwork.html?s_tid=answers_rc2-1_p4_MLT

function out = resnetFun(in)

persistent mynet;
if isempty(mynet)
    mynet = coder.loadDeepLearningNetwork('resnet50', 'myresnet');
end

out = predict(mynet,in);

You can integrate MATLAB function into your Simulink model using the MATLAB Function Block.

You can execute your MATLAB code from within the model to read the input data from your sensor and then pass this data into your Simulink block.

Refer to this example on how to use a MATLAB function block in your model.

[Lucas Ferreira-Correia]

Thank you for your help! I've imported my regression model and accordingly a placeholder layer is inserted on top of the model. I need a sequence input layer to replace the placeholder, is that possible? From the documentation I only see an image input layer available.

[sruthi gundeti]

VxNet=importKerasNetwork('Network.json','WeightFile','Network.h5','OutputLayerType','regression')

This can import a regression network

Answer 2

[Fernando Angel]

Hi,

Is it possible to use this method to generate code for a microcontroller in c / c ++? Thank you.

[Fernando Angel]

Thanks, yes, is possible, my keras NN is working perfectly in a STM32!

[Jie Li]

Hi, Fernando, Could you post how you put it in the simulink? My code says the importKerasNetwork cannot be used when complie.Thanks