Honeybees are the primary pollinators of mankind's food supply, and hence an understanding of the distribution of different species of bees is highly important to agriculture in Australia. One obstacle to undertanding bee population dynamics is the difficulty of identifying bees down to the species level. This project aims to provide a user interface to enable non-experts to apply machine learning for classifying bee species from their photos. The project is done in collaboration with Dr Tanya Latty of the University of Sydney.
Wing venation is a key characteristic for identifying bee species [1,2]. However traditional approaches obtaining images of wings are labour intensive, then standard machine learning approaches can be applied for classification. This project proposes to partially automate the process of bee species indentification from general images of bees. Recent advances in deep learning in computer vision has been successful in automating the feature construction process from general images . Using these approaches, the student's task is to build a software system that reduces the amount of manual work needed. If sufficient time is available at the end of the project, the system can be extended to allow for a "don't know" response when a previously unseen species is observed .
An ideal candidate would have a background in software engineering with an understanding of computer vision and machine learning. Programming skills in Matlab, Python or C/C++, and the knowledge of a browser based app framework such as node.js or Flask, would be necessary to implement the system. The project provides an opportunity for the student to combine machine learning and computer vision techniques to solve a pressing entomological problem.
 C.J. Hall An Automated Approach to Bee Identification from Wing Venation M.Sc Thesis, University of Wisconsin - Madison, 2011  Fabiana S. Santana, Anna H. Reali Costa, Flavio S. Truzzi, Felipe L. Silva, Sheila L. Santos, Tiago M. Francoy, Antonio M. Saraiva A reference process for automating bee species identification based on wing images and digital image processing Ecological Informatics 24 (2014), 248-260  Yangqing Jia, Evan Shelhamer, Jeff Donahue, Sergey Karayev, Jonathan Long, Ross Girshick, Sergio Guadarrama, Trevor Darrell Caffe: Convolutional Architecture for Fast Feature Embedding arXiv:1408.5093, 2014  W.J. Scheirer, A. de Rezende Rocha, A. Sapkota, T.E. Boult Toward Open Set Recognition IEEE Transactions on Pattern Analysis and Machine Intelligence, vol.35, no.7, pp.1757-1772, 2013
The project provides an opportunity for the student to combine machine learning and computer vision techniques to solve a pressing entomological problem.