English

Abstract

The conservation and sustainable utilization of global biodiversity necessitates the mapping and assessment of the current status and the risk of loss of biodiversity as well as the continual monitoring of biodiversity. These demands in turn require the reliable identification and comparison of animal and plant species or even subspecies. We have developed theAutomated Bee Identification System, which has not only been successfully deployed in several countries, but also supports taxonomical research as part of the Entomological Data Information System. Within this framework our paper presents two contributions. Firstly, we explain how we employ a model-driven extraction of polymorphic features to derive a rich and reliable set of complementary morphological features. Thereby, we emphasize new improvements of the reliable extraction of region-induced and pointinduced features. Secondly, we present how this approach is employed to derive new important results in biodiversity research. 1 The Role of Identification in Biodiversity Research Genera and species with ecological or economical use are a very relevant topic in biodiversity research. The bees, Apoidea, are of special interest, because about 70% of the world’s food crops are pollinated by them. The annual value of this pollination service has been reported to be about 65 billion US-$, cf. [6]. Since there is an alarming extinction of crop pollinating bees due to diseases, parasites, and pests, as well as due to human interventions like clearing, pesticides, etc., bees are specially focused by biodiversity research. ? Martin Drauschke is now with the Department of Photogrammetry, Institute of Geodesy and Geoinformation, University of Bonn, Nussallee 15, 53115 Bonn, Germany ?? Artur Pogoda de la Vega is now with CA Computer Associates GmbH, Emanuel-Leutze-Str. 4, 40547 Dusseldorf, Germany ? ? ? Tiago Mauricio Francoy is now with de Genetica, Faculdade de Medicina de Ribeirao Preto, USP, Av. Bandeirantes, 3900, 14049-900 Ribeirao Preto, SP, Brazil, tfrancoy@rge.fmrp.usp.br Due to the worldwide lack of bee exerts (less than 50 experts, so-called bee taxonomists, worldwide) on the one hand and the urgency of monitoring and protecting endangered pollinating bee species, theAutomated Bee Identification SystemABIS has been developed. After training, ABIS performs a fully automated analysis of images taken from the forewings of bees. This fully automated analysis is the outanding quality of ABIS compared with all other approaches, which demand for intensive interactive processing of each specimen, e.g. [2] or [7]. Compared with these interactive approaches ABIS has been reported to be at least 50 times more effective since there is no need to adjust the wings into an exact position of the camera field. The fully automated approach provides together with the ABIS database an information system for the surveillance and management of biometrical data of one of the most important pollinating group worldwide. It has been successfully deployed for monitoring purposes in Germany, Brazil and the United States, cf. [12]. ABIS is also part of Entomological Data Information System (EDIS), a co-operative research project jointly undertaken by a number of leading German natural history museums and university institutions in the area of biodiversity informatics. Fig. 1. Left: classification on demand: Bee experts use ABIS during their excursions. Middle: input image of an Africanized bee. Due to the image aquisition, the scale of all wings is in a small range. The wing’s position and orientation does not impact the feature extraction. Right: detected cells and vein net of wing image. The success of ABIS on identifying bee species leads now to a new challenge: the identification within the next taxonmical levels, i.e. subspecies and races, which are far more similar and therefore more difficult to separate. To face this difficult task of subspecies identification, we improve the extraction algorithms of morphological features, especially of the region-induced and point-induced features, i.e. cells and vein junctions. We demonstrate our improved approach within a fascinating taxonomical (oder biodiversity) research project on the so-called killer bees. In 1957, swarms of the African honey bee Apis mellifera scutellata accidentally escaped near Rio Claro, Brazil, and initiated a series of crosses and backcrosses with the pre-introduced European honey bee races, cf. [5]. This hybridization process resulted in the Africanized honey bee, which is better adapted to the tropics and broadly more productive than the European bees. Nevertheless, many beekeepers abandoned their activities in Brazil at the beginning of the Africanization process, because these bees presented undesirable traits like a high swarming rate and a strongly defensive behaviour (which led to their popular, but actually wrong name killer bees). In the meanwhile, the beekeepers have learnt how to work with these aggressive bees. Thus, the Africanized honey bee has spread throughout the (sub)-tropic Americas. 26