From transparency to trust: Explainable AI for object detection in safety-critical defence applications

The increasing deployment of AI systems in safety-critical defence applications creates risks regarding reliability, transparency, and responsibility. Explainable AI (xAI) is often proposed as a solution to model opacity, but its practical contribution to trustworthy AI still has limitations. This PhD project investigates how xAI for object detection can be designed, evaluated, and deployed so that it meaningfully supports epistemic trust in military contexts. To do so, first, a systematic review assesses the current state-of-the-art in xAI for object detection and how this contributes to trustworthy AI. Second, an explanation method for multimodal object detection (IR+RGB) is developed to analyse modal dependencies and reliability under varying conditions. Third, a participatory design study with divergent defence stakeholders examines what is expected from explanations and how this can be unified. Finally, an experiment investigates whether xAI supports epistemic trust or risks overreliance under changing reliability conditions. In doing so, the proposed dissertation aims to connect explainability, robustness, and trust as interdependent dimensions for responsible human-AI partnerships in high-stakes environments.