If a and b are positive, then this looks convex to me for x > -a/2, y > -b/2
Look at the solution by the wizard of conic reformulation, @Michal_Adamaszek as well as my direcltly following post in CVX wrong. How can I deal with this problem correctly? , and adapt that to your problem.
More specifically, look at the last row of Table 5.1 in https://docs.mosek.com/modeling-cookbook/expo.html#modeling-with-the-exponential-cone and modify to generalize 1/x to a/x and add the b/y term.