CNN Image Recognition with Regression Output

I have realized that I have been going about my problem totally wrong; my advisor told me I must use regression to score my images - not classification.

AutoML does not all for Images to be used as input with a 2nd Numeric Lable Input, and then a Numeric Output.

I’m a novice to CNN development; how can I make an Image recognition scoring model based on regression?

image

If you want to do regression using CNN then just change the loss function of the last layer. Instead of using sigmoid or softmax use Linear activation function.
RMSE as a loss function.
Thanks