An inverse model for target detection

Inverse least-squares (ILS) calibration is a well-established method in chemometrics for determining the quantity of a single constituent in a system where no explicit knowledge of the remaining constituents exists. Detection presents a very similar situation where, typically, the only precise knowledge available is that of the target signature. The traditional approach to detection involves the use of the linear mixture model, in which the contributions from all significant components must be fully specified. In this manuscript, we propose an inverse detection model as an alternative to the linear mixture model for the detection of a single target molecule in the presence of various unknown and potentially varying background components. In this inverse approach, the background constituents are implicitly modeled and, thus, no explicit knowledge or modeling of the background is required. The inverse model is applied to the automatic detection of dimethyl-methylphosphonate (DMMP) vapors from passive infrared (IR) remotely sensed hyperspectral image data.