Abstract: A technique is described for recovering positional andradiometric information on unresolved objects that areso closely spaced that their individual blur functionsoverlap. Emphasis is on point sources. A BayesianSpectral Analysis method has been modified to twodimensions and applied to resolving 'clumps' of objectsfor both simulated and real data. The method is able tojudge the amount of noise in the data and provide errorbars in the individual pulse positions and amplitudesfrom a single data set rather than from the deviationsobserved after measuring many independent sets of data.The Bayesian technique can also estimate the number ofdiscrete objects in a given clump. Noisy simulated datacontaining three sources was fitted by a one-, two-,three-, and four- source model. By the way itformulates the model, the Bayesian approach naturallyincludes a factor which reflects the reduction in thenumber of degrees of freedom for a model with a greaternumber of sources. As a result, the algorithm gives ahigher probability for the three-source model than forthe four-source model while resoundingly rejecting theone- and two-source models. The estimated centroids andamplitudes are shown to agree with the truth within thederived error bars to the degree expected by gaussianerrors. Studies of data taken during a flight test by asensor that measured a scene simultaneously in thevisible and long-wavelength regions show thatpositional information derived from visible-wavelengthdata can be 'fused' with infrared images to derive theLWIR intensities of individual objects in a unresolvedclump. The estimated LWIR intensities using the visibleassist are shown to be an improvement over working withthe LWIR data alone.!5
展开▼
