Skip to Main Content
A cost effective image based approach is proposed for monitoring New Zealand native bees, they are difficult to study, and require expert taxonomic identification due to minimal morphological differences between species. They have seasonal life-cycles which require long-term studies. Rather than identifying individual bees directly, as is done in most traditional ecological methods, the ground nests are identified and counted. The number of active nests can then be used to estimate the population of bees. This is possible because the number of bees in each nest is constant for most solitary mining species. A thorough field study has been conducted and a range of rich image data has been collected. Open source programs, Fiji and WEKA were used to implement computer vision techniques for pre-processing images, classification, accuracy verification and comparisons between random forest and support vector machine classifiers. The randon forest classifier in Fiji provided fast effective results classifying nests which were otherwise difficult to identify with the naked eye. This method is shown to be robust and simple with a potential to provide ecologists with repeatable and reliable estimations of the population status of New Zealand native bees.