FEA-DETR: An Enhanced ConvNet for Detecting Prohibited Objects in X-Ray Images Using Frequency and Edge Aware Information

Abstract

As the demand for transportation safety grows, traditional X-ray detection technologies face significant challenges, particularly item occlusion in complex backgrounds, which hinders the detection of prohibited objects. To address these issues, we propose Frequency and Edge Aware DETR (FEA-DETR), a robust DETR-based framework. The FEA-DETR integrates the Frequency and Edge Aware Attention Module (FEAM), which enhances feature representation by leveraging frequency and edge-aware information to better handle occlusive prohibited object detection. Additionally, we introduce the Multi-Dimensional Feature Hybrid Module (MDFHM), specifically designed for prohibited object detection in X-ray images. Extensive experiments on public datasets demonstrate that our model achieves competitive performance compared to existing state-of-the-art methods.