ARS | Publication request: Detection of Fire Ant Mounds in Airborne Digital Images: Hierarchical Learning for Automatic Feature Extration.

Submitted to: Imported Fire Ants Conference Proceedings
Publication Acceptance Date: May 1, 2004
Publication Date: September 14, 2004
Citation: Vogt, J.T. 2004. Detection Of Fire Ant Mounds In Airborne Digital Images: Hierarchical Learning For Automatic Feature Extration.. Imported Fire Ants Conference Proceedings.

Interpretive Summary: Researchers with the USDA, ARS Biological Control of Pests Research Unit tested commercially available software for automatic detection of fire ant mounds in airborne digital images. The software performed almost as well as a human photointerpreter, detecting more than 60% of mounds with no false positives (detection of a mound where no mound is present). Adaptation of this software, and development of additional automation techniques for examining images of large areas, will allow researchers to more efficiently map fire ant populations to track success of control efforts and guide research efforts in regional management programs.

Technical Abstract: Quantifying imported fire ant mounds over large areas is expensive and time-consuming. New methods of detecting and quantifying mounds will be useful for researchers engaged in regional management projects, assessment of biological control agents following release, and examination of landscape effects on fire ant populations. Regulatory personnel tracking new introductions and spread of fire ants will also benefit. Airborne digital imagery can be used to detect >70% of imported fire ant mounds in pasture areas, but photointerpretation of imagery is time-consuming. Use of commercially available software for supervised classification of images resulted in average detection of >60% of mounds in airborne imagery, with no commission errors.