Multistep classification problem using EVSI Bayesian preposterior framework

A network of small distributed wireless sensors is scattered over an extended geographic area that is to be monitored. A scenario of moving object entered the area and need to be classified as a Friend or Target is posed in this paper. Using Bayes theorem, sensors build beliefs around the object and classification decisions. As the object moves, sensors collaboratively choose a successor sensor and hand-off beliefs and classification decisions from the current active sensor to the successor. Sensor selection scheme is formulated as an Expected Value of Sample Information (EVSI) problem in the sequential Bayesian framework. Under the assumption that each measurement is independent and identically distributed-we extended the EVSI problem analytically to account for consecutive samples of information. The scheme presents dynamic classification-driven solution for sensor optimization problem under incomplete information and object's time varying dynamics assumptions.