Panchromatic Image Chip Classifier

NOVASOL is pleased to propose development of an algorithm for enhancing hyperspectral target detection performance in reconnaissance imagery using spectral detection combined with spatial processing of auxiliary panchromatic high-resolution imagery. The exploitation of both spectral and spatial data from multiple bore-sighted sensors represents a heretofore untapped opportunity to improve detection performance and mitigate false alarms in reconnaissance systems. By combining spectral signatures with multiple extracted spatial features in an operator supervised classifier, a knowledge base is quickly built up that can serve to filter and prioritize future detection image chips to more effectively use the analyst’s time and attention. Spatial features can be geared towards either target detection or false alarm mitigation. Since there are many more examples of false alarms, it is more practical to focus on improving performance by filtering unwanted cues. In addition, the proposed panchromatic image chip classifier (PICC) involves relatively simple computations, making it highly amenable to real-time operation on existing processor platforms. Furthermore, the inherent parallelism of the architecture lends itself to implementation on field-programmable gate array (FPGA) hardware which would produce significant economies in weight, size, power and cost.