A linear combination of heuristics approach to spatial sampling hyperspectral data for target tracking

Persistent surveillance of the battlespace results in better battlespace awareness which aids in obtaining air superiority, winning battles, and saving friendly lives. Although hyperspectral imagery (HSI) data has proven useful for discriminating targets, it presents many challenges as a tool in persistent surveillance. HSI sensors gather data at a relatively slow rate yet the sheer volume of data can be overwhelming. Additionally, fusing HSI data with grayscale video is challenging due to the differences in frame rate and resolution. A new sensor under development has the potential of overcoming these challenges and transforming our persistent surveillance capability by providing HSI data for a limited number of pixels and grayscale video for the remainder.;This dissertation explores the exploitation of this new sensor for target tracking. Every pixel receives a utility function value based on nearness to a target of interest (TOI) (determined from the tracking algorithm) and components of the TOI. The components in the utility function are equal dispersion, periodic poling, missed measurements, and predictive probability of association error (PPAE) which is introduced as a statistical measure in this dissertation. Equal dispersion means each TOI receives an equal amount of the resources available. Periodic poling means resources are dispersed based on how long it has been since the TOI received an HSI update. TOIs that were recently updated will receive fewer resources, while those that have not had an update for a longer time will receive more resources. The missed measurement component gives more resources to TOIs that have missed measurements. The more measurements a TOI has missed, the more resources it will get. PPAE is the probability that a TOI will receive measurements that result in an association error. It is predictive since the probability can be calculated at any time in advance of receiving measurements. It varies with nearness to other TOIs and the nature of their covariance matrices. PPAE is theoretically derived and validated in experiments.;Experiments are conducted to validate the individual components. Experiments use a simulated urban environment and a Kalman filter multitarget tracking algorithm to compare the individual components to each other. The synergism of the utility function which uses all the components is shown to outperform all individual components and is 6.5 percentage points better than the baseline performance of equal dispersion. The new sensor is successfully exploited resulting in improved persistent surveillance.