train convolution neural network from scratch

https://in.mathworks.com/matlabcentral/fileexchange/62990-deep-learning-tutorial-series?requestedDomain=true

In the code demo_trainfromscratch in the above link:

fc1 = fullyConnectedLayer(64,'BiasLearnRateFactor',2);

how are 64 filters chosen in fc1.

Also

fc1.Weights = single(randn([64 576])*0.1);

what does 576 indicate here and how are these values obtained.