BeeMachine identifies images of 41 of the most common North American bumble bee species. BeeMachine uses a convolutional neural network, modified from InceptionV3, and is trained on over 120,000 bumble bee images gathered from Bumble Bee Watch, iNaturalist, Wisconsin Bumble Bee Brigade, and BugGuide. Bombus taxonomy follows Williams et al. 2016.
Overall test accuracy is 92.1% (> 97% top-3) but this varies by species depending on the number of training images and their level of morphological variability. For example, Bombus fraternus has a consistent appearance across its range and among castes so it can be reliably classified even though it was trained on few images. On the other hand, B. sandersoni is more variable so it has lower accuracy. More training images should help improve its classification accuracy.
Species were nicely separated based on learned features. More details on model performance here.
- For the most accurate results, crop your image as close to the bee as possible to reduce extraneous pixels
- Image quality helps with accuracy. Motion, poor focus, or if the bee is represented by too few pixels can reduce accuracy
- As with human-based IDs, it is always good to compare predictions from multiple images of the same individual
- The model outputs the 3 most likely species and associated probabilities. Based on the test data, the correct ID is given by the top prediction over 91% of the time and is in the top 3 predictions over 97% of the time
- The current version was trained to predict 36 bumble bee species – and these species only. So uploading a picture of a different bumble species, a honey bee, a shoe, or random noise will give the wrong prediction. A “not-a-bee” class (or something more informative) will be implemented in the future
BeeMachine is in the early stages of development. It will be frequently updated for greater accuracy, to include more bee species (not just bumble bees!) and to broaden the geographic coverage. The user interface will be updated to automatically locate the bee within your image so you won’t have to manually crop before uploading. BeeMachine can be used on a mobile browser but a dedicated mobile app is in the works.
Help improve BeeMachine by contributing images!
Contact me (email@example.com) or leave a note below if you would like to contribute validated images (especially of poorly sampled species!). I would also love to hear your comments and suggestions.
Collaborators & Acknowledgements
- Claudio Gratton, University of Wisconsin – Madison
- Rich Hatfield, The Xerces Society for Invertebrate Conservation
- William Hsu, Kansas State University
- Sarena Jepsen, The Xerces Society for Invertebrate Conservation
- Brian McCornack, Kansas State University
- Krushi Patel, University of Kansas
- Richard Wang, University of Kansas
This project was developed with data from the Xerces Society for Invertebrate Conservation and Bumble Bee Watch (credit), iNaturalist (credit), BugGuide (credit), and my personal image collection. I am grateful for the volunteer participants in these programs that share their images and taxonomic expertise.