Automatic analysis of camera image data: an example of honey bee (Apis cerana) images from the Shanping wireless sensor network.

Abstract

Under an international collaborative program between the Taiwan Forestry Research Institute (TFRI) and Pacific RIM undergraduate experience (PRIME) of San Diego University, San Diego, CA, USA in 2010, we extended an image analysis package and applied it to honey bee observations. In this article, we describe the results of this collaboration. A tool suitable for routine measurements and counting tasks was developed to perform an automatic process. We applied blob-detecting of a computer vision technique to develop this package. We then tested the tool using images with different numbers of bees present collected from the Shanping wireless sensor network of TFRI. We compared the times consumed between the automatic and manual processes. Results showed that analysis of images with a low number of bees present (with an average bee number of ＜30 individuals per image) between the automatic process and manual process respectively required 9 and 315 min. A similar results showed that analysis of images with a high number of bees present (with an average bee number of ＞30 individuals per image) between the automatic process and manual process respectively require 23 and 409 min. Although the automatic process overestimated bee counts by 2~21%, the tool shows significant reductions in processing times. We concluded that the program provides a convenient way to determine the target and thus facilitate the examination of a large volume of honey bee images from a wireless sensor network in the field.