BUG COUNTS, CLOSE-UP
Engineers & Entomologists Innovate Arthropod ID System

The presence – or lack – of bugs in water and soil can tell humans a great deal: the water quality of a stream or watershed, the health of soil in a certain place, the impact of global warming on the ocean food chain.

Larval stoneflies, for example, are sensitive to drops in water quality caused by sedimentation, thermal pollution, eutrophication, and chemical pollution. They are an insect version of the canary in the coal mine when it comes to water quality; if their numbers drop, something’s amiss.

In order to glean this kind of knowledge, however, specimens must be gathered, identified, counted, and catalogued – specialized tasks that require highly educated, highly trained experts who are often in such high demand that supply is short. There may be fewer than 10 taxonomic specialists in the world with the expertise to perform all these analyses.

The process is also extremely time consuming (and therefore costly), especially when the arthropods being collected range from soil mites the size of the period at the end of this sentence to stoneflies an inch or more in length.

“Using current methods, it could take an entire week to catalog the specimens found in a cubic foot of soil,” says Bob Paasch, an OSU mechanical engineering professor who is collaborating with OSU computer scientists Tom Dietterich and Eric Mortensen. “Our goal is to automate this process so a computer-driven machine can do the work in a night.”

The engineers, who are part of the College’s Information Usability Research Cluster, are collaborating with OSU botanist Andrew Moldenke, OSU zoologist David Lytle, University of Washington computer scientist Linda Shapiro, and a small army of graduate and undergraduate students. The long-term goal is to develop an insect identification device that can be commercially marketed to agencies and companies that need the technology for environmental monitoring, agricultural studies, and biodiversity surveys.

But developing such a device is complicated and challenging; it involves complex algorithms and computers, high-power scanning microscopes and robotics, and of course taxonomy and entomology. “These arthropods are very tiny, very fragile objects,” says Dietterich, who is the project’s principal investigator and lead. “You can’t just tell them to assume a standard position and smile for the camera, or manipulate them like you can nuts and bolts. The computer vision algorithms have to be flexible enough to recognize them in many different positions.”

Another major challenge is developing computer vision algorithms that can distinguish thousands of different species from one another. “That’s a huge technical challenge, particularly when the species are difficult for expert entomologists to tell apart,” Dietterich says. “This requires major advances in object recognition methods, which will have many applications beyond recognizing insects.”

The computer vision, machine learning, and pattern recognition aspects of the project must mesh perfectly with the mechanical components, including a super-precise, six-axis robotic arm that can pluck a microbe as small as 50 microns from a Petri dish once it’s been identified.

No wonder nobody else is working on what the OSU team calls an “end-to-end” solution, where the system automatically scans, photographs, identifies, and catalogs specific specimens. It’s a monster project, but the potential end results inspire the research team to keep slogging.

“There are thousands of water district technicians who monitor water supplies who could use this tomorrow,” Paasch says. “We’re trying to fill a big-picture, societal need.” The researchers have just completed work funded by a 4-year, $1.7 million grant from the National Science Foundation, and are continuing the research under a new $800,000 NSF grant. Eventually, the team will work with a business partner to commercialize the system so the technology can go to work helping humanity better maintain healthier ecosystems everywhere.

Photos: (above) Principal investigator Tom Dietterich (left) and Bob Paasch look closely at stone flies in a prototype of their computer-aided insect identification device. (below) Using a mirror system, the apparatus photographs specimens, such as this stone fly, as image pairs to obtain two simultaneous views separated by 90 degrees.