This paper explores defense methodologies based on reinforcement learning in a target defense game. The scenario involves a defending aircraft seeking to protect a target from an attacker. We assume the attacker is a fixed-wing vehicle with a speed advantage, while the defender is a slower multirotor aircraft capable of varying its flight speed and agile turns. In this context, the reinforcement learning agent develops a guidance strategy that capitalizes on the maneuverability differences between the attacker and the defender. The paper discusses strategies such as reward shaping to ensure stable convergence of the agent. Simulations, considering various performance and strategies of attacking aircraft, demonstrate the feasibility and success of the proposed reinforcement learning-based approach.