BeeMachine

BeeMachine identifies images of 100 of the most common bumble bee species from around the world. BeeMachine uses a convolutional neural network, modified from InceptionV3, and is trained on over 313,000 bumble bee images gathered mainly from Global Biodiversity Information Facility (GBIF). Image data were also provided by Bumble Bee Watch, the Wisconsin Bumble Bee Brigade, the Hanamaru Maruhana Project, and Jerry Cole.

Overall test accuracy is 93.7% (> 99% top-3) but this varies by species depending on the number of training images and their level of morphological variability.

Species were nicely separated based on learned features.

Using BeeMachine

  • 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 nearly 94% of the time and is in the top 3 predictions over 99% of the time

100 Bombus species recognized by BeeMachine

SpeciesPrecisionRecallF1-ScoreImages
Bombus affinis0.9710.9740.9724,055
B. appositus0.9230.9120.9181,650
B. ardens0.7620.8820.818635
B. argillaceus0.9710.9530.962530
B. auricomus0.9570.9530.9556,845
B. barbutellus0.9130.7500.824140
B. beaticola0.9200.9200.920125
B. bellicosus0.9510.9830.967295
B. bicoloratus0.9170.9170.917180
B. bifarius0.7310.6790.704280
B. bimaculatus0.9440.9380.94115,705
B. bohemicus0.8480.8540.851685
B. borealis0.9560.9670.9615,980
B. californicus0.8540.9040.8784,015
B. caliginosus0.5670.6670.613255
B. campestris0.9320.9550.9441,010
B. centralis0.8550.8760.8652,655
B. citrinus0.9440.9660.9554,880
B. confusus1.0000.8750.933120
B. consobrinus0.9360.9160.926475
B. crotchii0.9690.9360.9521,480
B. cryptarum0.8530.8350.8441,150
B. dahlbomii0.9521.0000.975590
B. deuteronymus0.7960.9560.869225
B. distinguendus0.9240.9010.913405
B. diversus0.9730.9540.963755
B. ephippiatus0.9450.9530.9491,170
B. eximius0.9000.8330.865270
B. fervidus0.9500.9480.9498,205
B. flavidus0.8690.8930.8812,150
B. flavifrons0.8780.8690.8736,765
B. fraternus0.9700.9600.9651,365
B. frigidus0.7860.7860.786350
B. funebris0.9640.9710.967680
B. griseocollis0.9580.9580.95815,315
B. haematurus0.9380.8330.882180
B. honshuensis0.8970.9720.933180
B. hortorum0.9060.9180.9122,320
B. hortulanus0.9320.8850.908390
B. humilis0.8860.8530.8691,225
B. huntii0.9310.9270.9297,115
B. hypnorum0.9650.9700.9689,050
B. hypocrita0.9300.8550.891310
B. ignitus0.7310.8140.770350
B. impatiens0.9590.9610.96015,810
B. insularis0.8770.8920.8842,035
B. jonellus0.8330.8330.833390
B. kirbiellus0.9170.6110.73390
B. lapidarius0.9690.9630.9669,575
B. lapponicus0.8820.7900.83395
B. lucorum0.6670.7750.7171,930
B. medius0.8220.9170.867480
B. melanopygus0.9640.9570.96014,755
B. mesomelas1.0000.8330.90990
B. mexicanus0.8240.8750.84980
B. mixtus0.9070.9130.9106,750
B. monticola0.8850.8850.885260
B. morio0.8130.7220.76590
B. morrisoni0.8560.9020.878560
B. muscorum0.8670.8440.855385
B. nevadensis0.9480.9510.9494,180
B. occidentalis0.8830.9380.9102,020
B. opifex0.9570.9170.936360
B. pascuorum0.9770.9710.97415,995
B. patagiatus1.0000.8390.912155
B. pauloensis0.8730.8850.8791,170
B. pensylvanicus0.9640.9670.96617,125
B. perplexus0.9270.9470.9377,755
B. pratorum0.9620.9580.9605,995
B. pullatus0.7370.6670.700105
B. pyrosoma1.0001.0001.000105
B. quadricolor0.9500.9050.927105
B. robustus0.9060.9260.916470
B. rubicundus0.9730.9610.967380
B. ruderarius0.6830.8200.746250
B. ruderatus0.8230.7990.811670
B. rufocinctus0.9220.8830.90211,170
B. rupestris0.9040.9200.9121,435
B. sandersoni0.7000.5490.615255
B. schrencki0.8510.8510.851470
B. semenoviellus0.9290.8970.912145
B. sichelii0.7710.8180.794165
B. sitkensis0.7550.8000.7771,350
B. soroeensis0.7960.7290.761295
B. sporadicus0.8640.8640.864110
B. subterraneus0.8890.9030.896310
B. sylvarum0.9530.9530.9531,610
B. sylvestris0.8490.7180.778195
B. sylvicola0.7210.7690.744520
B. ternarius0.9680.9580.96317,775
B. terrestris0.9530.9200.93614,445
B. terricola0.9670.9680.9676,250
B. vagans0.8590.9080.8836,530
B. vancouverensis0.9410.9410.9417,860
B. vandykei0.8140.8710.8421,435
B. vestalis0.9050.9710.9371,035
B. veteranus0.8330.9090.870330
B. vosnesenskii0.9650.9500.95815,690
B. wilmattae0.9290.7650.83985
B. wurflenii0.8570.6210.720145

What’s next?

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 (bspiesman@ksu.edu) if you would like to contribute validated images (especially of poorly sampled species!). I would also be great to hear your comments and suggestions.

Collaborators & Acknowledgements

2 thoughts on “BeeMachine

Leave a Reply to Brian Spiesman Cancel reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.