Synthetic data are commonly used to train machine learning models in domains where real data are sparse. In this work, we describe a method to generate synthetic x-ray imaging data by inserting objects into a dual-energy computed tomography scan while simultaneously inserting the beam-hardening and noise artifacts that corrupt real data. This type of data augmentation is useful for training classifiers, for example, by artificially increasing the prevalence of objects of interest in a dataset. This work extends existing 3D Threat Image Projection methods by using dual-energy decomposition to model the energy-dependence of attenuation values in the sinogram data. By summing linear attenuation coefficient functions, objects can be inserted directly into a sinogram while accounting for beam-hardening in the insertion region. In addition, we introduce a calibration method to model the change in noise levels resulting from the insertion of attenuating objects. The performance of the method is demonstrated on a simple phantom scanned with a benchtop microCT system.
The Transportation Security Laboratory (TSL) performs testing of explosives detection systems using explosives and other hazardous materials. Inert simulants are also used as substitutes in potentially dangerous testing situations or at testing locations where explosives are prohibited. Each simulant must first be verified that it accurately represents the material on the specific detection platform it was designed for. In addition to the simulant-threat matching, lot-to-lot quality control testing is performed for simulants and threats to ensure that their physical properties remain consistent. Historically, x-ray verification has been limited to using features such as electron density and effective atomic number. While efficient, these features are limited in their application, as they do not provide information related to the material’s structural properties. In this study, four classification methods were tested using imagery-derived texture features to characterize materials and distinguish them from one another. The first three approaches (k-nearest neighbors, support-vector machine, and artificial neural network) were tested using 22 first- and second-order texture features derived from computed tomography images. The fourth method (convolutional neural network) used internally derived features. Based on the test results, a determination was made that the CNN and k-NN were the best algorithms to use to characterize materials based on their texture features.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.