Neil Cobb and Robert Delph, Department of Biological Sciences, and
Phillip Mlsna, Department of Electrical Engineering
Biodiversity : Inventory and monitoring studies for biodiversity are becoming more important as we are faced with rapidly changing landscapes consisting of human-altered ecosystems. As a result of these human activities, there is a growing general concern over the loss of biodiversity and a more specific fear of losing key taxa that provide important ecosystems services. Biodiversity studies allow us to better understand how entire communities respond to changing habitats and identify important indicator taxa of habitat quality to target for long-term studies. Arthropods are by far the most diverse taxa of organisms on the planet (Stork 1995), with over one million described species and an estimated ten to 70 million undescribed species. Because of their diversity and abundance, arthropods are excellent bio-indicators of ecosystem health and habitat dynamics (Castillo 2000).
Grand Canyon Studies : In collaboration with David Lightfoot and Sandra Brantly ( University of New Mexico ) The Colorado Plateau Museum of Arthropod Biodiversity has been conducting arthropod inventory and monitoring studies along the Colorado River in Grand Canyon since 2001. This ongoing study is part of a larger terrestrial monitoring program. The arthropods we have collected to date have been excellent indicators of habitat change that has occurred since the construction of Glen Canyon Dam. However, the sheer number and diversity of arthropods in the canyon is overwhelming. For example, in the order Lepidoptera we have identified over 235 taxa from 50,000+ specimens. Overall, we have identified over 700 taxa and we expect to discriminate over 1000 taxa by next year.
Limitations in Human Expertise : The most limiting factor in conducting the type of biodiversity studies like the Grand Canyon project is the inability to identify taxa due to the dearth of trained researchers that can correctly classify large sample sizes by correct taxa. Identification costs could be greatly reduced if we could augment our ability to identify taxa that usually require human expertise. As Steve Marshall recently stated “Instead of asking what it should cost to have individual insect species identified again and again, we should be addressing the costs of developing the tools needed to make those individual identifications simple and accurate” (Marshall 2003). The development of machine aided techniques addresses this problem. The ability to distinguish insect taxa using computers provides an upcoming new and important role for insect museums. The huge inventory of specimens provides a rich untapped data source on which to develop the emerging technology of computer identification. Our ability to implement a long-term monitoring project for Grand Canyon National Park would be greatly aided by automating identification of large sample sizes. We envision a day when arthropod samples are placed on a conveyor belt and passed through a computer driven system that counts and classifies specimens down to species.
Information Technology: Automated image analysis and computer vision is a very challenging field of research. While much progress has been made in the past two or three decades, computer vision techniques fall far short of inherent human abilities in visual analysis and pattern recognition. Nevertheless, many computer vision successes have been achieved in well-defined, limited problem domains. Examples include military target recognition, automated product inspection, and optical character recognition. Insect identification of pests and beneficial insects has been utilized at New Mexico State University for the last 20 years ( http://www.cahe.nmsu.edu/news/1996/083096_bugsoftware.html ). There are a few other efforts underway that are similar to the project we are proposing. They will provide some help in directing our efforts, but i n general, the image processing and image analysis techniques must be tailored to the specific application in order to obtain good results.
The goal of this project is to explore the feasibility of developing a machine vision system capable of identifying a variety of Grand Canyon arthropods to the genus and possibly species level with high accuracy. The initial phase will focus on a common and diverse family of Lepidoptera, the Noctuidae (Owlet Moths).
This initial phase of the project will concentrate on machine classification of moths in the family Noctuidae. We will begin with some of the more conspicuous subfamilies of Noctuidae and attempt to distinguish genera and possibly species. The Colorado Plateau Museum of Arthropod Biodiversity will prepare a set of high quality images of moth wings, along with expert identification information. Dr. Mlsna's team will develop the algorithms and software to analyze these images and produce identification information. The machine classification will be compared with the true identities and the results used to stimulate refinement of the system.
Wing patterns, colors, textures, and shapes are the primary image features that will form the basis of the machine classification system. These features will be automatically extracted from the raw images in the early sections of the software. Using a variety of strategies, characteristic patterns of these features will need to be identified and used to key the overall classification and identification process. Some of these strategies come from the field of artificial intelligence and expert systems. Others come from artificial neural networks, which are capable of learning patterns by experience.
REFERENCES
Castillo, J. M. 2000. Ground beetle community structure as a bioindicator of forest health. PhD Dissertation, Northern Arizona University .
Marshall , S. 2003. The real costs of insect identification. Newsletter of the Biological Survey of Canada (Terrestrial Arthropods). 22 No. 1.
Stork, N.E. 1988. Insect diversity: facts, fiction and speculation. Biol. J Linn. Soc. 35:321-337.