The broader impact of this Small Business Innovation Research (SBIR) Phase I project is in improving crop yields, while reducing the amount of pesticides used. By significantly improving the accuracy and timeliness of insect surveillance, we will allow more effective pest management, allowing the applications of insect interventions to be targeted in space and time. For example, rather than a blanket spraying of harsh pesticides across the entire field, our system could suggest spraying of a milder (and cheaper) pesticide in just a few "hot spots", at the optimal time of day. Reducing the volume of pesticides has further positive benefits to society at large, it will reduce pollution, and the use of pesticides has been implicated as a contributor to climate change and to colony collapse disorder. The hardware/algorithms/representations/data-models created in this project have an obvious application to mosquito surveillance, which has implications for control of insect vectored diseases of both humans and livestock. The commercial potential of this SBIR Phase I project is obvious. Insects damage or destroy about 150 billion dollars' worth of crops each year. If we prevent reduce this by just one percent, we have a billion-dollar market. The proposed project will investigate techniques to improve the state-of-the-art in flying insect classification, with the goal of producing a platform that allows insect surveillance for precision agriculture. In particular, we will take the current algorithms and representations (many of which were invented by the current PIs) and make them invariant to the wide range of conditions (temperature, pressure, humidity) encountered in the field. The company's research has shown that without creating such invariances, the variability induced by changing environment conditions will swamp the regularities in the features that are currently exploited by classification algorithms, and reduce the accuracy to random guessing. It is well understood how temperature, pressure, humidity effect air density, and how air density effects insect flight. However, the current models treat the insects as idealized objects using aerospace equations for density vs. lift and completely ignore the effects of the environment on insect physiology. The company plans to achieve this by creating a model that compensates for environmental conditions. To achieve these ambitious goals, they plan to use machine learning to learn the appropriate invariances and model corrections. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
