This PDF file contains the front matter associated with SPIE Proceedings Volume 11404, including the Title Page, Copyright information, and Table of Contents.

X-ray simulation of realistic object models is relevant across all areas in which X-ray systems are employed, including medical, industrial, and security applications. A particularly exciting area of impact stems from the development of machine learning approaches to classification, detection, and data processing. The continued development of these techniques requires large labeled datasets. Traditionally, this data needed to be collected with physical machines, creating steep logistical challenges. Moreover, the testing and evaluation of such X-ray scanners present their own challenges, as machines need to be shipped to a site capable of handling certain anomalies. To help alleviate these burdens, virtual models and simulations can be used in lieu of empirical measurements. The confluence of powerful computers and advanced data processing techniques presents an opportunity to develop tools to aid in dataset creation as well as system analysis. We present efforts toward the maturity of such tools. Building on previous work to validate the performance of simulation software, we show how modeling realistic virtual objects can produce data representative of real-world measurements. Furthermore, we present the efficiency of such an approach that leverages advances in computer graphics, ray-tracing utilities, and GPU hardware.

Recently, progress has been made in the supervised training of Convolutional Object Detectors (e.g. Faster RCNN) for threat recognition in carry-on luggage using X-ray images. This is part of the Transportation Security Administration's (TSA's) mission to ensure safety for air travelers in the United States. Collecting more data reliably improves performance for this class of deep algorithm, but requires time and money to produce training data with threats staged in realistic contexts. In contrast to these hand-collected data containing threats, data from the real-world, known as the Stream-of-Commerce (SOC), can be collected quickly with minimal cost; while technically unlabeled, in this work we make a practical assumption that these are without threat objects. Because of these data constraints, we will use both labeled and unlabeled sources of data for the automatic threat recognition problem. In this paper, we present a semi-supervised approach for this problem which we call Background Adaptive Faster R-CNN. This approach is a training method for two-stage object detectors which uses Domain Adaptation methods from the field of deep learning. The data sources described earlier are considered two “domains": one a hand-collected data domain of images with threats, and the other a real-world domain of images assumed without threats. Two domain discriminators, one for discriminating object proposals and one for image features, are adversarially trained to prevent encoding domain-specific information. Penalizing this encoding is important because otherwise the Convolutional Neural Network (CNN) can learn to distinguish images from the two sources based on superficial characteristics, and minimize a purely supervised loss function without improving its ability to recognize objects. For the hand-collected data, only object proposals and image features completely outside of areas corresponding to ground truth object bounding boxes (background) are used. The losses for these domain-adaptive discriminators are added to the Faster R-CNN losses of images from both domains. This technique enables threat recognition based on examples from the labeled data, and can reduce false alarm rates by matching the statistics of extracted features on the hand-collected backgrounds to that of the real world data. Performance improvements are demonstrated on two independently-collected datasets of labeled threats.

Coherent x-ray scatter is material specific, and imaging systems utilizing information from coherently scattered x rays are promising for security and medical applications requiring material identification with high sensitivity. A persistent challenge for practical implementation of these systems has been slow image acquisition. Our approach to reducing acquisition time is to develop a multibeam projection imaging system rather than a volumetric (CT or otherwise) imaging system. Previously we implemented a synchrotron-based system with five coplanar pencil beams and continuous motion of the object. Now we present a tabletop x-ray scatter imaging system built using a rotating-anode x-ray tube and a scintillating, energy-integrating flat-panel detector. A conventional source is more challenging to use than a synchrotron beam due to polychromaticity, low intensity, beam divergence, and x-ray tube thermal considerations. Simulations were performed to determine the system layout that optimized the intensity and angular resolution of scatter signals. The tube is inclined 6.1° to reduce apparent focal spot size. The primary collimation allows for an array of up to three rows by five columns of pencil beams, 3mm diameter and 2 cm apart at the object midplane 35 cm from the source, to irradiate the object simultaneously. There is no scatter collimation and the multiplexed scatter signals are disentangled using a maximum-likelihood expectation maximization algorithm. Motorized translation stages scan the object through the beams. The system can image objects up to 10 × 10 × 10 cm³ and 1 kg. Post-object primary beam attenuators allow for the same detector to measure transmitted and scattered x rays simultaneously. Initial images acquired with the system are presented. Using 15 beams, a 6000-pixel scatter image of a 6 cm × 10 cm region was acquired in 4.6 min.

Inspecting shipping containers using X-ray imagery is critical to safekeeping our borders. One of major tasks of inspecting shipping containers is manifest verification, which has two components: 1) determine what cargos are contained in a shipping container, which can be carried out in cargo segmentation, and 2) compare the cargos in the container with the cargos declared in the manifest. We focus our study on cargo segmentation. Cargo segmentation is the process of partitioning the cargo inside the container into regions with similar appearance. Assign a cargo class label to each pixel in the X-ray images. Our contribution is the development of a deep learning neural net based cargo segmentation algorithm that significantly improves the traditional ways of performing cargo segmentation. The cargo segmentation process is implemented by first partitioning the X-ray images into image tiles of certain sizes, and then train a deep learning (DL) model-based semantic segmentation algorithms using the annotated image tiles to partition the cargo into regions of similar appearance. The DL based semantic segmentation algorithm we used is an encoderdecoder structure often used for semantic segmentation. The DL network implementation chosen for our cargo segmentation is DeepLab v3+, which includes the atrous separable convolution composed of a depthwise convolution and pointwise convolution. Our X-ray cargo images used for development is a government-provided data set (GPD).

This work addresses the question whether it is possible to design a computer-vision based automatic threat recognition (ATR) system so that it can adapt to changing specifications of a threat without having to create a new ATR each time. The changes in threat specifications, which may be warranted by intelligence reports and world events, are typically regarding the physical characteristics of what constitutes a threat: its material composition, its shape, its method of concealment, etc. Here we present our design of an AATR system (Adaptive ATR) that can adapt to changing specifications in materials characterization (meaning density, as measured by its x-ray attenuation coefficient), its mass, and its thickness. Our design uses a two-stage cascaded approach, in which the first stage is characterized by a high recall rate over the entire range of possibilities for the threat parameters that are allowed to change. The purpose of the second stage is to then fine-tune the performance of the overall system for the current threat specifications. The computational effort for this fine-tuning for achieving a desired PD/PFA rate is far less than what it would take to create a new classifier with the same overall performance for the new set of threat specifications.

Porous structures are widely found in natural and engineered material systems. To study the defect initialization and damage evolution in the complex 3D network structures, we explore advanced X-ray phase tomography to provide holistic and high-resolution 3D data. A pipeline of deep learning-based phase retrieval, computer vision, and damage identification algorithms are implemented to extract various types of damage for large volumetric tomography data. We first obtain high-quality phase tomography reconstruction from noisy and insufficient CT acquisition. Based on a hybrid approach using both model-based feature filtering and data-driven machine learning, we then identifies the defects and damaged regions from the background of porous structures. This method is applied to an in-situ X-ray tomography measurement on a natural cellular material; the accurate and comprehensive defects detection reveals insight into 3D damage evolution modes for porous material systems.

In this paper, a coded aperture optimization approach based on sparse principal component analysis (SPCA) is proposed to maximize the information sensed by a set of cone-beam projections. The variables in the CT system matrix correspond to observations of the attenuation characteristics of X-ray projections. An adjusted joint variance is used to update the variables and thus the overlapping information of the kth principal component is constrained by the previous k-1 principal components. Since the coded aperture matrix is diagonal and binary, an efficient algorithm is proposed to reduce the complexity by one order of magnitude. Simulations using simulated datasets, 3D Shepp-Logan phantom, show significant gains up to 23.5dB compared with that attained by random coded apertures. Singular value decomposition (SVD) of the optimized coded apertures is used to analyze the performance of the proposed coded aperture optimization method based on SPCA.

Impacting of fast flying fragments or projectiles on a protection shield as well as high-speed rotating machineries like jet engine turbines or turbo chargers store high kinetic energies. When such impacting fragments respectively components burst caused by material fatigue failure that energy is released by local loading protection shields like ballistic armor systems or enveloping casings. The path of travel behind the armor shield and the fragmentation of the impacting objects provides engineers and designers useful information to evaluate failure risks. Ballistic testing by registration the fragments using flash X-ray technology (FXR) is a method to study the behavior of fragments or projectiles in front and behind the shield after interaction especially under harsh conditions. The path of travel as well as the residual velocity will be determined and analyzed. These results could also be used to support numerical simulation. We present a simple method to reconstruct the three-dimensional trajectory of fast-moving objects after impacting with protection shields or casings using a three-channel flash X-ray system each channel with dual remote tube heads triggered simultaneously to get three orthogonal X-ray images. These images are calibrated to reduce optical distortion. To calculate the 3D trajectory and the residual velocities the coordinates of the objects registered by the image plates of the vertical and horizontal plane will be descripted by two-dimensional vectors for each plane.

Extending our prior work, we propose an X-ray measurement model that incorporates spatial- correlated material variability. The model enables more accurate task-specific assessment of the performance of X-ray imaging and sensing systems. More specifically, the model can be used to calculate bounds on the probability of error (P_e) for threat-detection tasks. We analyze the performance of a prototypical X-ray measurement system to compare the new spatial- and energy-correlated model with the previous model, which ignores the spatial correlation.

We develop a framework to analyze the information content of X-ray measurement data for the task of material discrimination. This task-specific information (TSI) analysis provides valuable information for system design and optimization. We employ Bhattacharyya distance (BD) between measurements of different materials as the TSI metric in our analysis framework, because BD is closely related to the bounds on the probability of error (P_e). We apply this framework to compare an X-ray diffraction-based system with an X-ray attenuation-based system for several materials and different detector geometries.

Polychromatic X-ray scatter is modulated in both angle and energy in a way that encodes a material’s crystalline texture. Various texture quantification metrics have been calculated from X-ray scatter which are typically most informative for fundamental material or crystallographic research. In this work, we quantify material crystalline texture from scatter measurements made using a tabletop energy and angle dispersive diffractometer and show that these X-ray scatter-based metrics have promise as complementary metrics to the material form factor and are particularly suited for material identification applications.