You cannot do that in a continuous optimization problem, although in sparse optimization, it is common to approximate the number of non-zeros as norm(x,1).
If it's a binary linear programming that you're solving, the number of non-zeros is simply the linear function sum(x). If it's a more general integer linear program where 0<=x(i)<=M, you can introduce extra binary variables z with the constraint that x/M<=z<=x, and the number of non-zeros will then be sum(z).
