Data versus decision fusion for distributed classification in sensor networks

Sensor networks provide virtual snapshots of the physical world via densely distributed wireless nodes that can sense in different modalities. Classification of objects moving through the sensor field is an important application that requires collaborative signal processing (CSP) between nodes. Given the limited resources of nodes, a key constraint is to exchange the least amount of information between them to achieve desired performance. Two main forms of CSP are possible. Data fusion - exchange of low-dimensional feature vectors - is needed between correlated nodes. Decision fusion - exchange of local likelihood values - is sufficient between independent nodes Decision fusion is generally preferable due to its lower communication burden. We study CSP classification algorithms based on a Gaussian model for sensor measurements that provides a simple abstraction of node correlation and yields a simple characterization of the optimal maximum likelihood classifier Two extreme sub-optimal classifiers are also considered: a data-averaging classifier that treats all measurements as correlated, and a decision-fusion classifier that treats them all as independent. Analytical and numerical results based on real data are provided to compare the performance of the three CSP classifiers. Our results indicate that the suboptimal decision fusion classifier that is most attractive in the context of sensor networks, is also a robust choice from a decision theoretic viewpoint.