I don't know why I wrote an entire proof outlining why first-order PA cannot rule out infinite numbers, if you just ignore it and insist on your misconception. The set of finite numbers is undefinable in the language of first-order PA, which is why first-order induction isn't enough.
Sorry, I don't see why second-order induction is required. AFAICT all you need is a first-order induction axiom instantiated on the predicated FINITE.