Biologically Inspired Signal Processing for Sensor Applications
Chi Yung Fu
Thursday, December 4, 2003
1:30 PM
Building 151, Room 1209 (Stevenson Conference Room)

Abstract:
Results from non-accelerator neutrino oscillation experiments have provided Considerable research effort has been expended on developing new sensors. However, intelligent signal processing, such as extracting trace-level signals from noisy environments out in the field, advanced pattern recognition of the signatures of nuclear weapons, as well as biological and chemical agents and self-adaptive response algorithms, have not been vigorously pursued. Also the number of sensors being deployed or on the way to be deployed is increasing, and these sensors represent systems of diverse origins and technologies. We will present our work on using wavelet technology to preprocess noisy data and artificial neural network to extract relevant signals from some raw gas chromatographic and radiation detector data. The approach roughly mimics our biological signal processing system. The retina performs some multi-level processing somewhat akin to wavelet processing. Our neurological system extracts relevant information to analyze the degree of threat. The work presented here has the potential to be applicable to other types of sensors because of the adaptive nature of the algorithm. Once relevant signals are extracted, such information can be compressed and encrypted effectively with our previous work on data compression to handle data storage and transmission since the amount of data from sensors is expected to explode because of the almost exponential growth characteristics of data available from sensors.