Imaging Science Ph.D. Defense: Cara Murphy
Ph.D. Dissertation Defense
Methods for Generating High-Fidelity Trace Chemical Residue Reflectance Signatures for Active Spectroscopy Classification Applications
Imaging Science Ph.D. Candidate
Chester F. Carlson Center for Imaging Science, RIT
Advisor: Dr. John Kerekes
Standoff detection and identification of trace chemicals in hyperspectral infrared images are enabling capabilities in a variety of applications relevant to defense, law enforcement, and intelligence communities. Though multiple classes of algorithms exist for the detection and classification of these signatures, they are limited by the availability of relevant reference datasets. This research demonstrates and compares three methods for generating more accurate spectral libraries used for chemical classification applications.
We first address the lack of physics-based models that can accurately predict trace chemical spectra. Most available models assume that the chemical takes the form of spherical particles or uniform thin films. This research presents an improved signature model for residual chemicals called sparse transfer matrix (STM), which includes a log-normal distribution of film thicknesses.
There remain limitations in the STM model which prevent the predicted spectra from being well-matched to the measured data in some cases. To overcome this, we propose the first one-dimensional (1D) conditional generative adversarial network (GAN) to perform spectrum-to-spectrum translation of reflectance signatures. The method demonstrates an increase in overall classification accuracy, though the performance improvement is biased towards data included in the GAN training set.
The final method focuses on learning a model that is more robust to parameter combinations for which we do not have measured data. We develop a physics-guided neural network (PGNN) for predicting chemical reflectance for a set of parameterized inputs. The outputs of the PGNN provide a higher classification accuracy on real data than the spectral libraries generated by the STM and GAN models, including on chemicals excluded from the model training set.
Undergraduates, graduates, and experts. Those with interest in the topic.
When and Where
Open to the Public