Sensor/data fusion based on value of information
Shubha L. Kadambe, Cindy E. Daniell · 2003
Spatially distributed network of inexpensive, small and smart nodes with multiple onboard sensors is an important class of emerging networked systems for various applications. Since this network of sensors has to operate efflciently in adverse environments, it is important that these sensors process information efficiently and share information such that the decision accuracy is improved. One way to address this problem is to measure the value of information obtained from multiple sensors on the same sensor node as well as from the neighboring nodes and fuse that information if value is added in terms of improvement in decision accuracy. In this paper, the measures that are developed for assessing the value of information are described. These measures are then used in making a decision of fusing either features or data from multiple sensors and neighboring nodes. While making this decision whether the value is added by fusing the information, is verified by conditioning it on improving the decision accuracy. The measures and value added are verified by using real data collected at Twentynine Plams, CA, USA and in the context of target detection and classification. From the results of improvement in classification accuracy and probability of detection reported in this paper, it can be seen that the utilization of measure of value of information while fusing helps in improving the decision accuracy signijicantly.