In support of the detection of explosives and threat chemicals by active infrared backscatter hyperspectral imaging, we are training algorithms to process and alert on possible threats. Surfaces are interrogated using infrared quantum cascade lasers (QCL) and the backscattered signal is collected using a cooled MCT focal plane array (FPA). The QCLs can tune across their full wavelength range, from 6 – 11 m, in less than one second. Full 128 X 128 pixel frames from the FPA are collected and compiled into a hyperspectral image (HSI) cube containing spectral and spatial information from the target. The HSI cubes are processed and the spectra from extracted pixel locations are then run through an algorithm to detect and identify traces of explosives. We train our algorithms on both synthetic and experimental data. In this presentation, we utilize machine learning algorithms to classify HSI cubes from a series of targets coupons fabricated on relevant substrates (glass, painted metal, plastics, cardboard). We explain how the algorithm training uses reference spectral measurements from our cart system as well as from a benchtop FTIR. The generation and utility of synthetic data is described regarding how we populate the algorithms’ spectral library more densely than would be possible using only measured experimental data. The performance of several ML algorithms is described.
We seek to detect and classify chemical threats based on their infrared spectra. Specifically, we are interested in utilizing spectral signatures observed with standoff technologies that interrogate analyte micro-particles on relevant substrate surfaces such as glass, metal and plastics. In this work, we have applied six Machine Learning algorithms to classify analytes based on their infrared spectra. Two synthetic datasets were used, the first one containing 40 analytes and the second one containing 55 analytes. In both datasets the analytes were synthetically placed onto 9 substrates. The 40 analytes dataset contains 18,000 spectra, 450 for each analyte with mass loading varying from 1 to 50 μg/cm2. The 55 analytes dataset consists of 49,500 spectra, 900 for each analyte and mass loadings in the range 1 to 100 μg/cm2. Two of the algorithms used in this work are coming from the statistical field; k nearest neighbors (k-NN) and Logistic Regression. The Support Vector Machine algorithm was developed by the Machine Learning community. Multilayer Perceptrons (MLP) as well as Convolutional Neural Networks are considered Deep Learning Algorithms. In addition to that, we have considered the hybrid deep learning algorithm one dimensional CNN-LSTM. Our experimental results lead us to the conclusion that k-NN and logistic regression outperform deep learning algorithms for our synthetic data sets. However, after dimensionality reduction using PCA, the accuracy of k-NN decreases and the performance of deep learning algorithms improves. We also considered the effect of mass loadings and added noise on the performance of the classifiers.
We are developing machine learning algorithms to identify chemicals of interest by their diffuse infrared (IR) reflectance signatures. For capturing the signatures themselves, we are developing a cart-based mobile system for the detection of trace explosives on surfaces by active infrared (IR) backscatter hyperspectral imaging (HSI). We refer to this technology as Infrared Backscatter Imaging Spectroscopy (IBIS). A wavelength tunable multi-chip infrared quantum cascade laser (QCL) is used to interrogate a surface while an MCT focal plane array (FPA) collects backscattered images to comprise a hyperspectral image (HSI) cube. The HSI cube is processed and the extracted spectral information is fed into an algorithm to detect and identify chemical traces. The machine learning algorithm utilizes a 1-dimensional convolutional neural network (CNN) that has been trained on augmented FTIR diffuse reflectance spectra. In this manuscript, we implement a 1-D CNN to identify chemicals within an IBIS hypercube. This demonstrates a form of active chemical imaging where the CNN identifies a chemical within each pixel of an IBIS hypercube. Chemical imaging capability goes beyond point detection and identification to indicate where each chemical is within the field of view, as well as identifying multiple target chemicals simultaneously.
We are developing algorithms to identify chemicals of interest by their diffuse infrared (IR) reflectance signatures when they are deposited as particles on surfaces. For capturing the signatures themselves, we are developing a cart-based mobile system for the detection of trace explosives on surfaces by active infrared (IR) backscatter hyperspectral imaging (HSI). We refer to this technology as Infrared Backscatter Imaging Spectroscopy (IBIS). A wavelength tunable multi-chip infrared quantum cascade laser (QCL) is used to interrogate a surface while an MCT focal plane array (FPA) collects backscattered images to comprise a hyperspectral image (HSI) cube. The HSI cube is processed and the extracted spectral information is fed into an algorithm to detect and identify chemical traces. The algorithm utilizes a convolutional neural network (CNN) that has been pre-trained on synthetic diffuse reflectance spectra. In this manuscript, we present an approach to generate large libraries of synthetic infrared reflectance spectra for use in training and testing the CNN. We demonstrate advancements in the number of analytes, a method to generate synthetic substrate spectra, and the benefits of subtracting the substrate “background” to train and test the CNN on the resulting differential spectra.
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.