Prohibited Item Detection Technology Based on Meta-DETR

Prohibited item detection in data-scarce environments poses significant challenges, especially in security applications. Building upon the Meta-DETR framework, this study incorporates adversarial learning to improve model robustness by generating synthetic samples that simulate difficult-to-detect prohibited items. Additionally, a dynamic few-shot adaptation mechanism allows the model to adjust its parameters during inference, enhancing detection accuracy across tasks of varying complexity. Together, these advancements address the limitations of data scarcity and improve the model’s ability to generalize in real-world applications. Experimental results on public benchmarks show marked improvements in detection accuracy and adaptability, particularly in X-ray security screening systems where precise identification is critical.