Gas Plume Species Identification in LWIR Hyperspectral Imagery by Regression Analyses

The goal of this research was to develop an algorithm for identifying the constituent gases in stack releases. At the heart of the algorithm is a stepwise linear regression technique that only includes a basis vector in the model if it contributes significantly to the fit. This significance is calculated by an F -statistic. Issues such as atmospheric compensation, gas absorption and emission, background modeling, and fitting a linear regression to a non-linear radiance model were addressed in order to generate the matrix of basis vectors. Synthetic imagery generated by the DIRISG model were used as test cases. Results show that the ability to correctly identify a gas diminishes as a function of decreasing concentration path-length of the plume. Results drawn from pixels near the stack are more likely to give an accurate identification of the gas present in the plume.