In his research at the Taiwan Forestry Research Institute, Adrian Teng-Amnuay worked on an automated method of distinguishing between many families of wasps using computer vision and machine learning.
While at the Taiwan Forestry Research Institute (TFRI), I worked closely with TFRI scientists Chau Chin Lin, Sheng-Shan Lu, and Yu-Huang Wang to develop a computer vision and analysis system for automated taxonomic identification of wasps. Taxonomic identification has traditionally been carried out by individuals with years of experience and in-depth knowledge of the particular taxa. The process of identifying specimens can often be time-consuming and tedious. Working with TFRI and UCSD mentors Tony Fountain and Serge Belongie, we developed a system for automated image classification on a small subset of the Vespidae family of wasps.
This system utilizes a combination of computer vision and machine learning tools, including the OpenCV library for computer vision. I developed an analysis workflow for transforming raw images into quantitative features and then conducted experiments with various feature selection and machine learning algorithms. Our experiments produced high classification accuracy on the target problem and confirmed the feasibility of this approach. While this system still has much room for improvement, it provides a demonstration of how image classification can be automated, and it provides a foundation for further studies in this area. For more on the OpenCV library, visit opencv.willowgarage.com/wiki.
PARTICIPATING RESEARCHERS: TFRI: Chau Chin Lin, Sheng-Shan Lu, Yu-Huang Wang; Calit2/UCSD: Tony Fountain, Serge Belongie