High Performance and Portable Electronic Dog-Nose with Fluctuation-Enhanced Sensing and Prompt Bacterium-Identifier Option

We propose a high performance and portable system that incorporates a newly developed technique known as fluctuation enhanced sensing (FES) to enhance commercial off the shelf (COTS) sensors. First, besides the mean (steady state) sensor value, we will use a low noise amplifier to enlarge the small stochastic fluctuations in the sensor. Features such as mean-square fluctuations, skewness, kurtosis, power spectrum, bispectrum images of the fluctuations will be extracted. Second, the scalar features such as mean value, mean-square fluctuations, skewness, and kurtosis will be fed into a robust classifier called Support Vector Machine (SVM) to enhance the overall detection performance. SVM has several advantages, including no over training problem, global optimal solution, and computational efficiency. We have applied SVM to classify toxic vapors for Kennedy Space Center and Jet Propulsion Lab. Third, we will apply novel techniques from hyperspectral signal processing to classify agents based on power spectrum and amplitude density functions. Fourth, we will apply a proven image recognition technique to classify agents based on bispectrum images. Fifth, we will feed the decisions from different classification approaches into a Dempster Shafer fusion algorithm. A single decision will be drawn, which is robust and optimal.