Search

• A dual domain deep-learning-based approach is proposed for MAR in baggage CT. • The SIN is used to repair the metal erosion trace in the sinogram domain. • The PRN is proposed to refine and correct the sinogram inconsistency artifacts. During the security check, the metal in the baggage engenders serious metal artifacts on the Computed Tomography (CT) image. To reduce the effect of metal artifacts on the judgment of prohibited items during such activity, a Deep Learning (DL) method combined with dual domain information in CT images is proposed in this work. As for the methodology of work, the metal areas are segmented, in the first phase, from the metal artifact CT image and they are then projected into the sinogram domain using the Forward Projection (FP) algorithm. As the trace of the metal-corresponding projection area in the sinogram domain is considered to be missing data, the linear interpolation method is adopted to correct the metal missing trace, and the Sino-Inpainting Network (SIN) is deployed to repair the metal erosion trace. By adopting the Filtered Back Projection (FBP) algorithm to reproduce the results of the sinogram restoration, the mutual information between the sinogram domain and the image domain is completed. In the second phase, the sinogram inconsistent artifacts are repaired using the Partial Refine Network (PRN) after the corrected image restoration. The PRN only depends on the effective pixels outside the metal damaged area to restore the trace area; thus, this technique can be more effective to refine the image details. Finally, the metal mask, obtained by threshold segmentation, is inserted into the repaired reconstructed image. Using both simulated data and real data, a comparison between the proposed method, the conventional method, and the DL method is performed. Quantitative results show that the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Measure (SSIM) in the reconstructed image of the baggage, obtained through the proposed network, are 31.0722 dB and 0.9718, respectively; thus, they are better than the other tested methods. Moreover, the results of the experiments demonstrate the effectiveness of the suggested method in recovering the lost data of the metal corrosion area in the sinogram and in suppressing the secondary artifacts in the reconstructed image. [ABSTRACT FROM AUTHOR]

Several limitations in algorithms and datasets in the field of X-ray security inspection result in the low accuracy of X-ray image inspection. In the literature, there have been rare studies proposed and datasets prepared for the topic of dangerous objects segmentation. In this work, we contribute a purely manual segmentation for labeling the existing X-ray security inspection dataset namely, SIXRay, with the pixel-level semantic information of dangerous objects. We also propose a composition method for X-ray security inspection images to effectively augment the positive samples. This composition method can quickly obtain the positive sample images using affine transformation and HSV features of X-ray images. Furthermore, to improve the recognition accuracy, especially for adjacent and overlapping dangerous objects, we propose to combine the target detection algorithm (i.e., the softer-non maximum suppression, Softer-NMS) with Mask RCNN, which is named as the Softer-Mask RCNN. Compared with the original model (i.e., Mask RCNN), the Softer-Mask RCNN improves by 3.4% in accuracy (mAP), and 6.2% with adding synthetic data. The study result indicates that our proposed method in this work can effectively improve the recognition performance of dangerous objects depicting in the X-ray security inspection images. [ABSTRACT FROM AUTHOR]