Linear output instead of classification from CNN

I would like to get a linear output from an image CNN,
instead of getting “dogs vs cats” classification, i would like to get a number lets say 0-100
as output. (I am trying to build an age determination).

I believe I should remove the softmax layer, and replace it with a sigmoid instead,
is this correct?

Any pointers how to change the network like this with the framework, using “dogs vs cats” code as a starting point?