Monitoring populations of solitary bees using image processing techniques

Abstract

Results from monitoring New Zealand's native bees based on image processing techniques are presented. Rather than identifying individual bees directly images of active ground nests are collected, processed and counted. The number of nests is a good estimate of population since the number of bees in each nest is constant for most native species; described as solitary because a single female constructs a single nest. When the numbers of insects in flight around the trees located next to nest sites are combined with these data they can give a good quantitative indication of the population of native bees within a community. Open source software FIJI was used to pre-process and classify images. Accuracies and classifier performance were verified using the data mining software WEKA. To date, the random forest classifier has returned fast effective results, classifying nests which are otherwise impossible to identify or count with the naked eye. The number of active nests also compares well to the number of insects in flight and when data of active nests across two seasons are compared a fluctuation in population is evident. It is therefore anticipated that this method could provide another practical alternative to traditional ecological surveys.