Integrated Sensing and Processing for Statistical Pattern Recognition

This article presents a simple version of Integrated Sensing and Processing (ISP) for statistical pattern recognition wherein the sensor measurements to be taken are adaptively selected based on task-specific metrics. Thus the measurement space in which the pattern recognition task is ultimately addressed integrates adaptive sensor technology with the specific task for which the sensor is employed. This end-to-end optimization of sensor/processor/exploitation subsystems is a theme of the DARPA Defense Sciences Office Applied and Computational Mathematics Program’s ISP program. We illustrate the idea with a pedagogical example and application to the HyMap hyperspectral sensor and the Tufts University “artificial nose” chemical sensor.