IEEE Home | Shop IEEE | Join IEEE | myIEEE | Contact IEEE | IEEEXplore

IEEE Signal Processing Society (SP)
Baltimore Chapter

SP Home   |   IEEE Home   |   IEEE Baltimore   |   Join SP   |   Contact SP
SP Main Menu

Meeting Number:   8

March 27, 2007


Detection and Identification of Raman Signatures of Known Chemicals in the Presence of Arbitrary Noisy Backgrounds via Pedestal Estimation


Abhijit Yeshwantpur
Hughes Network Systems


Tuesday, March 27, 2007


5:45 PM:   Snacks
6:00 PM:   Talk begins


Historical Electronics Museum (HEM)
1745 W. Nursery Road, Linthicum, MD 21090

Please Respond To

Please respond to if you are planning to attend this meeting. Also, let me know if you will be joining us afterwards at Ruby Tuesdays for dinner so I can make reservations. Again, only the speaker’s dinner is paid for. The rest of us need to pay our own way.


Laser Interrogation of Surface Agents (LISA) uses Light Detection and Ranging (LIDAR) along with Raman spectroscopy to identify chemicals from a distance (on the order of tens of meters). Raman signatures that provide chemical identification information are obtained by irradiating chemical molecules with laser pulses. Raman signatures are relatively low intensity signatures that are degraded by fluorescence spectrum, additive noise, interference from non-target chemicals, and the instrument used to measure them. The highly varying broadband fluorescence combined with the instrument effects causes a non-uniform rise in the spectrum, termed as the pedestal, which is difficult to mathematically model. The pedestal is the most dominant source of signal degradation.

We developed and tested three algorithms that achieve detection via estimating the pedestal and addressing the other sources of signal degradation. The pedestal estimation technique differs in these algorithms. The first algorithm uses a technique called the linear interpolation of pedestal estimation (LIPE) to estimate the pedestal based on linear interpolation on the entire set of local minima in the data. The second algorithm addresses the shortcomings of LIPE via using a sliding-window technique of pedestal estimation (SWTPE) to estimate the pedestal with due consideration to multiplets. The third algorithm called feature-metric algorithm (FMA) achieves a computationally efficient detection algorithm that reduces the detection time by nearly an order of magnitude.

We show that high detection rates can be achieved as long as the measured Raman-shift data and the library data are processed similarly. We show that correlation, used as the measure of closeness between the measured and library data, can be improved by choosing a wavenumber range of interest (WROI) within which all chemical signatures are known to be present. Finally, we show that feature-oriented processing drastically reduces the computation load and may be used effectively if SNR of the data is moderate to high.


Abhijit Yeshwantpur received his Bachelor's degree in Engineering in Electronics and Communications from the Visveswaraiah Technological University (VTU), India in 2002 and MS in Electrical Engineering from the University of Maryland, Baltimore County in 2006. His graduate thesis work is pending patent. He is currently working with Hughes Network Systems as a DSP Engineer.

SP Home   |    IEEE Home   |    IEEE Baltimore   |    Privacy & Security   |    Terms & Conditions