Contour awareness for few-shot X-ray prohibited item detection with optimal distribution transfer

The X-ray prohibited item detection (XPID) model is designed to automatically identify and locate prohibited items in X-ray images of passengers’ luggage. Due to strict government regulations, the rarity of prohibited items makes it challenging to collect sufficient training data, which increases the risk of overfitting. Recently, few-shot XPID (FSXPID) has garnered significant attention in the computer vision community by enabling XPID models to train effectively with limited X-ray images. To further enhance FSXPID performance, we propose a method that integrates a contour awareness module (CAM) and an optimal distribution transfer module (ODTM). Specifically, given the inherent visual characteristics of X-ray images, the CAM emphasises key contour details that describe the spatial geometry of foreground regions. This focus enhances the discrimination of prohibited item features during feature extraction. The ODTM leverages an optimal transport-based strategy to accurately model feature distributions for novel categories in few-shot scenarios. Consequently, the ODTM facilitates the sampling of additional features from novel distributions, which effectively mitigates data limitations. Extensive experiments conducted on a customised FSXPID dataset demonstrate the effectiveness and robustness of the proposed method. The source code and dataset are publicly available at https://github.com/DonkryJie/XrayFSOD.