Determination of trace-gas amounts in plumes by the use of orthogonal digital filtering of thermal-emission spectra

Abstract

The thermal emission of gases in a plume can be measured by a Fourier-transform spectrometer that is located some distance from the plume. In order to measure quantitatively the amount of a particular gas of interest, in general a large spectrally structured background must be removed. Differencing techniques, in which a measured background spectrum is subtracted from a measured spectrum believed to contain a target, often do not remove background spectral features adequately. The inadequacy of two-spectrum differencing techniques is due to the spatial and the temporal variations in a scene. We present a method by which to reduce spatial and temporal spectral clutter to instrument random noise, allowing the measurement of gas amounts in an effluent plume. The method is applied to simulated data and field data to show its effectiveness.

Scene analysis and detection in thermal infrared remote sensing using independent component analysis

“…The chemical signature of the plume is crucial information in this regime, and is typically used in matchedfilter detection 13 or a multiple regression of the chemical along with some set of basis vectors that account for the background clutter. 11 A plume that is weak and present in only a small number of pixels is a problem similar to subpixel target detection for solid objects in a scene. 16 Weak plumes are expected to be associated with radiance changes of a few percent or less from on-plume to off-plume pixels.…”

“…A similar chemical detection procedure using PCA to generate basis vectors has been described previously. 11 This is effectively equivalent to orthogonal subspace projection. 12 We tried PCA on our simulation and obtained a very similar result to Fig.…”

Chemometrics in Process Analytical Technology (PAT)

“…While many analytical and PAT practitioners avoided the use of the rigid and resource -intensive CLS method in favor of more fl exible and user -friendly inverse methods (such as MLR and PLS), the possibility of such extended mixture models had already been discussed some time ago [1] , and such methods were being utilized extensively in other fi elds such as remote sensing [60] . In more recent years, resurgence in these types of models is starting to make its way back to the PAT community [47 -51] .…”

Geostatistical and local cluster analysis of high resolution hyperspectral imagery for detection of anomalies

“…This phenomenon was observed and demonstrated in Chang and Heinz (2000). This weighting of the inverse of the PCs is also shared by signal identification methods, which aim to divide the at-sensor radiance received from a pixel into signal and noise or clutter components: orthogonal subspace projection (Harsanyi & Chang, 1994), orthogonal background suppression (Hayden et al, 1996), and matched filters (Funk et al, 2001). The risk, however, is to give too much importance to minor noisy components; hence, in practice, the RXD statistic incorporates only a smaller subset of the first t components:…”