Multiple sensor fusion under unknown distributions
In a system of N sensors, the sensor {ital S{sub i}}, i = 1, 2 ..., N, outputs {ital Y}{sup (i)} {element_of} {Re}, according to an unknown probability distribution P{sub Y{sup (i)}}{vert_bar}X, corresponding to input X {element_of} {Re}. A training {ital n}-sample (X{sub 1}, Y{sub 1}), (X{sub 2},Y{sub 2}), ..., (X{sub n},Y{sub n}) is given where {ital Y{sub i}} = (Y{sub i}{sup (1)},Y{sub i}{sup (2)},...,Y{sub i}{sup (N)}) such that Y{sub i}{sup (j)} is the output of S{sub j} in response to input X{sub i}. The problem is to design a fusion rule expected square error: I({ital f}) = {integral}[X - f (Y)]{sup 2}dP{sub y{vert_bar}X}dPx, where Y=(Y{sup (1)}, Y{sup (2)},..., Y({sup N)}),is minimized over a family of functions {ital F}. Let f{sup *} minimize I(.) over {ital F}; in general, f{sup *} cannot be computed since the underlying distributions are unknown. We consider sufficient conditions based on smoothness and/or combinatorial dimensions of {ital F} to ensure that an estimator {cflx {ital f}} satisfies P[I({cflx {ital f}}) - I(f{sup *}) > {epsilon}] < {delta} for any {epsilon} > 0 and 0 < {delta} < 1. We present two methods for computing {cflx {ital f}} based on feedforward sigmoidal networks and Nadaraya-Watson estimator. Design and performance characteristics of the two methods are discussed, based both on theoretical and simulation results.
- Research Organization:
- Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
- Sponsoring Organization:
- USDOE, Washington, DC (United States)
- DOE Contract Number:
- AC05-96OR22464
- OSTI ID:
- 391704
- Report Number(s):
- CONF-9608144-1; ON: DE96014690
- Resource Relation:
- Conference: Workshop on foundations of information/decision fusion, Arlington, VA (United States), 7-9 Aug 1996; Other Information: PBD: [1996]
- Country of Publication:
- United States
- Language:
- English
Similar Records
Stochastic approximation methods for fusion-rule estimation in multiple sensor systems
Fusion rule estimation in multiple sensor systems with unknown noise distributions