Extension and implementation of a model-based approach to hyperspectral change detection

Abstract

A new method for hyperspectral change detection derived from a parametric radiative transfer model was recently developed. The model-based approach explicitly accounts for local illumination variations, such as shadows, which act as a constant source of false alarms in traditional change detection techniques. Here we formally derive the model-based approach as a generalized likelihood ratio test (GLRT) developed from the data model. Additionally, we discuss variations on implementation techniques for the algorithm and provide results using tower-based data and HYDICE data.

Application of Model-Based Change Detection to Airborne VNIR/SWIR Hyperspectral Imagery

“…The change in state estimates beyond 20 iterations is minimal. Additionally, the RT basis matrices S, D, and P are generated using singular value decomposition on several thousand MOD-TRAN realizations generated using various combinations of atmospheric parameters as described in previous work [10], [24].…”

“…The initial estimates are relatively accurate for the time-2 data but over-estimated for the time-1 data. The estimation and detection results are compared for the MB algorithm run in the three different reflectance estimation modalities discussed in previous work [24]. The three modalities include no spectral filtering (NF) of the reflectance estimates, spectral filtering (SF) of the reflectance estimates, and use of a reflectance subspace (RS).…”

“…By defining the at-sensor radiance from a vegetation pixel as (24) and the at-sensor radiance from a shade pixel (i.e., path radiance) as…”

“…. , K) associated with spatial position (m, n) of a hyperspectral data cube has been previously described by [10], [23], [24] …”

“…Recent work has developed an alternative model-based (MB) approach to HSCD [10], [23], [24]. Shadow coefficients are included in the data model to account for the local illumination differences that tend to introduce false alarms.…”