UCSD Undergrad Tsung-Han 'Hank' Lin: From Oceans to Outer Space
San Diego, Calif., Dec. 7, 2009 — In the five years he's been an undergraduate at the University of California, San Diego, Tsung Han "Hank" Lin's research has taken him from the deep sea to the far reaches of outer space.
While his research projects don't appear on the surface to have much in common, they are driven by a shared motivation: To develop faster, more efficient and more accurate methods for classifying data, whether those data are the coordinates for spacecraft landing sites or the effects of plankton on ocean ecosystems.
A student in UC San Diego’s Computer Science and Engineering program and an affiliate of UCSD's California Institute for Telecommunications and Information Technology (Calit2), Lin was one of nine computer scientists to be selected for this year’s 15-week NASA Undergraduate Student Research Program. Lin is working with Dr. Thomas Lu, Senior Researcher of the Pasadena-based NASA Jet Propulsion Laboratory (JPL) to develop an Automated Target Recognition (ATR) system for autonomous vehicles.
Explains Lin: "The ATR system is a ‘bioinspired technology’ that has been in development for about 20 years. Its mission is to locate targets in images via a video feed. The key component is the ultra high-speed Grayscale Optical Correlator filter that can identify many ‘regions-of-interest' (ROI) in real-time. Each ROI may contain potential targets, and current work is ongoing to reduce the false detection rate."
"There are many applications for this technology," he adds. "With ATR, we can help spacecraft land by tracking landing sites, or the Navy can use ATR to remotely monitor the coastline for ships in the area. It could also be used to provide an image of the bottom of the ocean and help with things like mine detection."
The ATR system that Lin is helping to create uses artificial intelligence and a neural network algorithm based on human brain neurons to filter out "regions-of-interest" from a video feed and extract features that provide information about those region-of-interests. The data is then classified by the neural network (when associated with coastline monitoring, for example, the system could tell the user if the region-of-interest is a ship or just a large wave).
"Because this is a neural network system, we have to train the system and build up its artificial intelligence (AI) first," says Lin. "One of the challenges is to train the system to recognize an object against a 'noisy background.' We have to determine which of the modeled networks is most appropriate, and at the same time, we want to be able to generally classify and not make things too specific. On the other hand, if we don't give the system enough training, it might not classify well, so there's a trade-off."
Lin got a taste of the challenges inherent to AI classification systems when he was sent to his native Taiwan this past summer to work with marine biologists as part of Calit2's Pacific Rim Undergraduate Experiences (PRIME) Program. Under the mentorship of Calit2 Principal Research Engineer Doug Palmer Lin helped develop a Plankton Automatic Recognition System to improve the efficiency of plankton monitoring conducted by National Taiwan University.
"Our main motivation with that project was that there really weren't any good plankton classification systems for marine biologists," recalls Lin. "Because they have to look through microscopes to be able to classify the samples, it takes the scientists a really long time to get through just a few hundred specimens per day — and for a representative sample of the plankton in the area, you really need to classify thousands of specimens.
"What we wanted to do was create an automatic plankton system to analyze how plankton affects the ecosystem," he continues. "We used Zooscan to scan images of plankton within water samples and then tried to extract features and use classification algorithms to determine species. In a way, it's kind of related to what I'm doing now. I'm thinking of eventually collaborating with the JPL to see if I can use some of their AI technologies to improve this classification system."
"The military call it 'target recognition' and we call it 'plankton recognition," adds Palmer, Lin's mentor for the ongoing plankton project. "These creatures constitute 99 percent of all the biomass in the oceans, but we know very little about them. They're very understudied, so this is very important work."
Lin's affiliation with Calit2 has spanned the course of his academic career — as a sophomore, he participated in the institute's Summer Undergraduate Research Scholar Program, where he studied flower phenotypes under the direction of former UCSD CSE Professor Eleazar Eskin (now a professor of CSE at UCLA). After graduating this year, Lin says he hopes to continue working with Palmer on the plankton classification system, and perhaps even collaborate with the Calit2-based Machine Perception Laboratory, where he can explore his greatest passion: robotics.
"I wish to develop autonomous vehicles for underwater, or space aerial vehicles that include a vision system so it can track where it's been and figure out where to go next," Lin explains. "I'd also like to explore social robotics in which a robot can interact with a human and have a conversation, or work with humans.
"At NASA, there's a robot called 'robonaut' who was designed to work with astronauts in space in places that are too dangerous for humans. To help build a robot like that is something I'd really like to do one day."
Tiffany Fox, (858) 246-0353, email@example.com