I am looking for an NN architecture that can perform a task of predicting sprinkler placement on a lawn. After some time of researching, I came up with possible solution that on paper suits me the best.
My solution is a CNN, where one channel is basically representing a polygon (lawn) which is basically input data, and second channel representing sprinkler placement (or prediction value from 0 to 1 which determines if sprinkler should be placed here or not) which is output data and the data I want to predict based on unique polygon that I will give to the NN. As a dataset I have decent amount of hand-made projects where sprinklers were placed manually.
The question is: Can I even separate the channels like this and does this architecture make any sense? More specifically, can CNN be designed like this, where I will be able to feed one channel (lawn) to the neural network to predict second channel (sprinkler placement)?