Image Classification of Pollen-Bearing Honey Bees
Abstract
Bees are essential pollinators, about 80% of pollination around the world is taken care of by honey bees. With changes in land use, increased pesticide use, and climate change, as well as a number of other factors, bee communities have been declining. The number of worker bees in a colony is essential to its functioning; they collect pollen on their legs – the hairs on their bodies attract grains of pollen, which are then formed into “baskets” on their legs for carrying back to the hive. To understand behavior of bees and colony health, hive and honey bee activities can be observed, such as pollen collection. Machine learning has the potential to observe and classify bee activity automatically and more quickly than humans, which can allow for large scale data collection and can possibly lead to new insights. This all leads to the research question: Can we create a convolutional neural network (CNN) model that can accurately classify whether a bee is carrying pollen or not? To address this question, I used a honey bee image dataset from Kaggle that is based on data used in Recognition of Pollen-Bearing Bees from Video Using Convolutional Neural Network by Rodriguez et al. (2018). The dataset contains 714 image files of pollen-bearing bees or non-pollen-bearing bees. The images were split 80/20 into training and test sets, and a convolutional neural network model was created to attempt to classify the test images. The model performed with a training accuracy of 87.82% and a test accuracy of 87.41%. Compared to results in published literature, Rodriguez et al. developed a shallow two layer CNN with an accuracy of 96.4%. In The Application of Convolutional Neural Network for Pollen Bearing Bee Classification by Sledevič (2018), a three layer CNN performed with 94% accuracy. While my model performed quite well and produced promising results, literature review shows that there is still room for improvement.
References
Background Research
- EPA. (2018, July 19). Pollinator Health Concerns. EPA. https://www.epa.gov/pollinator-protection/pollinator-health-concerns.
- Michigan State University Department of Entolomogy. (n.d.). Pollination. Native Plants and Ecosystem Services. https://www.canr.msu.edu/nativeplants/pollination/.
- Schwartz, J., Kuenzle, M., & Pinsky, D. (2014, June 18). Save the Bees. Greenpeace USA. https://www.greenpeace.org/usa/sustainable-agriculture/save-the-bees/.
- University of Arkansas Department of Agriculture. (n.d.). The importance of pollinators. Bees as Pollinators; Arkansas Pollinators. https://www.uaex.edu/farm-ranch/special-programs/beekeeping/pollinators.aspx.
Data
- Rodriguez, I. F. (2018, November 20). Honey Bee pollen. Kaggle. https://www.kaggle.com/ivanfel/honey-bee-pollen.
Literature Review
- Rodriguez, I. F., Megret, R., Acuna, E., Agosto-Rivera, J. L., & Giray, T. (2018). Recognition of Pollen-Bearing Bees from Video Using Convolutional Neural Network. 2018 IEEE Winter Conference on Applications of Computer Vision (WACV). https://doi.org/10.1109/wacv.2018.00041
- Sledevic, T. (2018). The Application of Convolutional Neural Network for Pollen Bearing Bee Classification. 2018 IEEE 6th Workshop on Advances in Information, Electronic and Electrical Engineering (AIEEE). https://doi.org/10.1109/aieee.2018.8592464
Presentation Photos
- Brown, K. (2020, July 12). Photo by Kelsey Brown on Unsplash. Beautiful Free Images & Pictures. https://unsplash.com/photos/DUH-BOpXtAA.
- Honey Bee PNG Images: Honey Bee Transparent PNG. Vippng. (n.d.). https://www.vippng.com/ps/honey-bee/.
Code: honeybeeCNN.py