|
|
|
University of
Pennsylvania SUNFEST Projects Summer 1997
For more detailed information click on the title below. ANALYSIS AND DESIGN OF ELECTRO-THERMAL-COMPLIANT MICRO DEVICES
IMPLEMENTING
LEARNING ALGORITHMS FOR PATTERN RECOGNITION ON A NEUROCOMPUTER Joseph Murray - (EE) University of Oklahoma Advisors: Dr. Paul Mueller and Dr. Jan Van der Spiegel Neural network learning algorithms were implemented for the University of Pennsylvania/Corticon NP-4 Neurocomputer. The NP-4 is a large-scale, programmable analog computer with a fully parallel architecture. The algorithms were implemented in software and include backpropagation, correlation matrix memories, and unsupervised Hebbian learning. The system was successfully tested on two real-time image processing tasks: detecting oriented edges and identifying the digits '0' through '9'. The speed, accuracy and performance in the presence of noise were measured. STUDY TO IMPROVE FLUORESCENT DETECTION METHODS
Juan Carlos Sáez (Chemistry) - University of Puerto Rico, Cayey Fluorescence microscopy is rapidly becoming a powerful tool because this method allows identification with high sensitivity of specific fluorescently-labeled molecules in biological material [5]. Fluorescein is one of the fluorescent dyes most commonly coupled to biological protein probes; other examples are rhodamine and Cy5ª. A major problem accompanying the use of these dyes in microscopy is light-induced bleaching, apparent as fading of the emitted fluorescent light [5]. One can see in our results that anti-fade agents do not help with the fluorescein bleaching problem very much. However, we found a new dye, Cy5ª, that is far more resistant to the photobleaching process than fluorescein. Cy5ª appears to be an excellent dye for fluorescence studies and shows very good characteristics such as resistance to photobleaching, ability to be analyzed dry and wet, available in a chemical form that can easily be attached to protein and DNA, easily measured to very low concentration (attomole level 10-18) with the CCD camera, excited by an inexpensive HeNe laser or laser diode and having a good quantum yield (> 28%). Now, many different types of analyses can be carried out in this way including rapid detection of DNA mutational events, disease analysis, forensic identifications and sequencing by hybridization. A MICROLENS ARRAY TO INCREASE THE FILL FACTOR OF A CMOS CAMERA Ali Husain (Electrical Engineering) University of Pennsylvania Advisor: Dr. J. Van der Spiegel This report describes the implementation of microlenses including
design, fabrication and preliminary test results. The lenses will be used to increase the
fill factor of a CMOS camera. Microlenses have been fabricated out of photoresist. The
photoresist was patterned on a silicon substrate into cylinders and melted into spherical
lenses. The fabrication of the lenses has several problems so far: mask patterns that do
not closely approximate circles, buckling of the lens surface due to surface tension
effects, and optical properties of the lenses. Future directions will include further
fabrication attempts with thicker resist layers, measurement of focal lengths, monolithic
fabrication of the arrays, and modeling of the lens surface.
LANDMARK BASED LOCALIZATION AND LANDMARK FEATURE TRACKING: METHODOLOGY AND APPLICATIONS
SPEECH SYNTHESIS FROM A SPECTROGRAM Gavin Haentjens (EE) - University of Pennsylvania Advisors: Mr. Ahmed Abdelatty Ali, Dr. Paul Mueller, and Dr. Jan Van der Spiegel A Matlab program was developed to quickly generate high quality speech from
spectrograms with linear and nonlinear distributions of frequency channels. The program
adds sine waves together to synthesize speech, but can also produce speech via cepstral
analysis for spectrograms with linearly distributed frequency channels. The program also
allows the user to generate spectrograms from speech samples (in .wav, .mat, or text
format) and adjust the scaling of the spectrograms by specifying a frequency scale. The
program is intended to be used in conjunction with a graphical interface developed by my
partners that allows spectrograms to be manipulated in a vast number of ways. The highest
quality speech was produced using the sine wave method on spectrograms that were scaled
using the bark scale, but the cepstral method, which produced lower quality (noisier)
speech, was significantly faster. SAM: A GRAPHICAL USER INTERFACE FOR THE ANALYSIS AND MANIPULATION OF SPECTROGRAMS O'Neil Palmer (EE) and Kelum Pinnaduwage (EE) - University of Pennsylvania Advisors: Mr. Ahmed Abdelatty Ali, Dr. Paul Mueller, and Dr. Jan Van der Spiegel The analysis and manipulation of a spectrogram can be very useful in the field of
speech recognition. This prompted the idea of creating a user-friendly Graphical User
Interface (GUI) that would enable the user to better understand speech patterns. The
"Spectrogram Analyzer and Manipulator" (S.A.M.) interface, along with the
"Speech Synthesis" program, can be used as a very powerful tool in this regard. IMPLEMENTATION AND TESTING OF A DIGITAL CMOS CAMERA Francis Chew (EE) - University of Pennsylvania Advisor: Prof. Jan Van der Spiegel A Complementary Metal Oxide Semiconductor (CMOS) digital camera was developed by L.P. Ang, Raymond Tong and D.J. Yoon as their senior design project - completed in May, 1997. Though preliminary testing has been done on the camera to ensure that the CMOS chip and board circuitry are working in a desirable manner, no testing was done on the sensitivity and sharpness of the camera as a complete unit/device. Improvements were made on chip-to-camera mounting, imaging/focusing and optimization of wiring methods. Viable image sensor applications are also introduced for the implementation of the CMOS digital camera.
|