Recognizability bias in citizen science photographs
Citizen science and automated collection methods increasingly depend on image recognition to provide the amounts of observational data research and management needs. Recognition models, meanwhile, also require large amounts of data from these sources, creating a feedback loop between the methods and...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
The Royal Society
2023-02-01
|
| Series: | Royal Society Open Science |
| Subjects: | |
| Online Access: | https://royalsocietypublishing.org/doi/10.1098/rsos.221063 |
| _version_ | 1797858617953615872 |
|---|---|
| author | Wouter Koch Laurens Hogeweg Erlend B. Nilsen Robert B. O’Hara Anders G. Finstad |
| author_facet | Wouter Koch Laurens Hogeweg Erlend B. Nilsen Robert B. O’Hara Anders G. Finstad |
| author_sort | Wouter Koch |
| collection | DOAJ |
| description | Citizen science and automated collection methods increasingly depend on image recognition to provide the amounts of observational data research and management needs. Recognition models, meanwhile, also require large amounts of data from these sources, creating a feedback loop between the methods and tools. Species that are harder to recognize, both for humans and machine learning algorithms, are likely to be under-reported, and thus be less prevalent in the training data. As a result, the feedback loop may hamper training mostly for species that already pose the greatest challenge. In this study, we trained recognition models for various taxa, and found evidence for a ‘recognizability bias’, where species that are more readily identified by humans and recognition models alike are more prevalent in the available image data. This pattern is present across multiple taxa, and does not appear to relate to differences in picture quality, biological traits or data collection metrics other than recognizability. This has implications for the expected performance of future models trained with more data, including such challenging species. |
| first_indexed | 2024-04-09T21:16:14Z |
| format | Article |
| id | doaj.art-5a0595c2448f4d77ada09f3497e958bd |
| institution | Directory Open Access Journal |
| issn | 2054-5703 |
| language | English |
| last_indexed | 2024-04-09T21:16:14Z |
| publishDate | 2023-02-01 |
| publisher | The Royal Society |
| record_format | Article |
| series | Royal Society Open Science |
| spelling | doaj.art-5a0595c2448f4d77ada09f3497e958bd2023-03-28T08:50:59ZengThe Royal SocietyRoyal Society Open Science2054-57032023-02-0110210.1098/rsos.221063Recognizability bias in citizen science photographsWouter Koch0Laurens Hogeweg1Erlend B. Nilsen2Robert B. O’Hara3Anders G. Finstad4Department of Natural History, Norwegian University of Science and Technology, 7491 Trondheim, NorwayIntel Benelux, High Tech Campus 83, 5656 AE Eindhoven, The NetherlandsNorwegian Institute for Nature Research, Postboks 5685 Torgarden, 7485 Trondheim, NorwayDepartment of Mathematical Sciences, Norwegian University of Science and Technology, 7491 Trondheim, NorwayDepartment of Natural History, Norwegian University of Science and Technology, 7491 Trondheim, NorwayCitizen science and automated collection methods increasingly depend on image recognition to provide the amounts of observational data research and management needs. Recognition models, meanwhile, also require large amounts of data from these sources, creating a feedback loop between the methods and tools. Species that are harder to recognize, both for humans and machine learning algorithms, are likely to be under-reported, and thus be less prevalent in the training data. As a result, the feedback loop may hamper training mostly for species that already pose the greatest challenge. In this study, we trained recognition models for various taxa, and found evidence for a ‘recognizability bias’, where species that are more readily identified by humans and recognition models alike are more prevalent in the available image data. This pattern is present across multiple taxa, and does not appear to relate to differences in picture quality, biological traits or data collection metrics other than recognizability. This has implications for the expected performance of future models trained with more data, including such challenging species.https://royalsocietypublishing.org/doi/10.1098/rsos.221063citizen scienceimage recognitionmachine learningrecognizability |
| spellingShingle | Wouter Koch Laurens Hogeweg Erlend B. Nilsen Robert B. O’Hara Anders G. Finstad Recognizability bias in citizen science photographs Royal Society Open Science citizen science image recognition machine learning recognizability |
| title | Recognizability bias in citizen science photographs |
| title_full | Recognizability bias in citizen science photographs |
| title_fullStr | Recognizability bias in citizen science photographs |
| title_full_unstemmed | Recognizability bias in citizen science photographs |
| title_short | Recognizability bias in citizen science photographs |
| title_sort | recognizability bias in citizen science photographs |
| topic | citizen science image recognition machine learning recognizability |
| url | https://royalsocietypublishing.org/doi/10.1098/rsos.221063 |
| work_keys_str_mv | AT wouterkoch recognizabilitybiasincitizensciencephotographs AT laurenshogeweg recognizabilitybiasincitizensciencephotographs AT erlendbnilsen recognizabilitybiasincitizensciencephotographs AT robertbohara recognizabilitybiasincitizensciencephotographs AT andersgfinstad recognizabilitybiasincitizensciencephotographs |