Preventing Adversarial AI Attacks Against Autonomous Situational Awareness: A Maritime Case Study