New edge-residual method improves prohibited item detection accuracy
DERA: Detached Edge-Residual Adaptation for Prohibited item Detection
Computer Vision and Pattern Recognition
Summary
Detecting prohibited items in X-ray images is hard because objects overlap and details can be fuzzy or cluttered. The authors proposed a new technique called DERA that learns to better recognize object edges by focusing on differences in nearby pixels combined with semantic image features. This helps the system separate object boundaries from other visual information, making the detection more precise. Their method showed measurable improvements over existing approaches on multiple X-ray security datasets.
What this means in practice
- •For airport security teams: Enhance X-ray scanner software to improve detection of hidden prohibited items with clearer object boundary recognition.
- •For industrial inspection firms: Improve detection of defects or unwanted materials in layered or cluttered object scans using edge-sensitive adaptation.
Authors
Yonathan Michael, Mohamad Alansari, Mohammed Bennamoun, Dwarikanath Mahapatra, Andreas Henschel, Naoufel Werghi
Abstract
Prohibited-item detection in X-ray imagery remains challenging due to object superposition, weak texture, and material clutter which obscure both semantic appearance and object boundaries. We propose \textbf{DERA}, a \textbf{D}etached \textbf{E}dge-\textbf{R}esidual \textbf{A}daptation framework for prohibited item detection under X-ray imagery. DERA combines hierarchical visual features with a parallel pixel-difference edge pyramid and learns an object-specific boundary prior from training-time contours of the instance masks. The detached prior gates edge-sensitive features, which are injected into the early visual stages through residual heads. This staged design preserves the foundation detector at the start of adaptation, isolates boundary supervision from semantic feature learning, and restricts the final adaptation stage to only \(14.7\)K trainable parameters. Evaluated on PIDray, CLCXray, and STCray, DERA improves the baseline by \textbf{3.1}, \textbf{1.6}, and \textbf{2.4} AP points, respectively.