Data Capture: Technology versus Manpower

Transcription

1 Data Capture: Technology versus Manpower Entomological Perspective /media/images/robot.jpg

2 Capturing specimen data from labels 01Book/images/collect/fig23.jpg

3 Manual data entry Semi-automated digitization in between Mantle et al. (2012)

4 Manual data entry: basics Databaser enters specimen data directly from insect label Taxonomic information; locality information (country, subdivisions, exact locality & georeference); collection event collector & date; specimen information Unique identifier (GUID, USI) physically associated with each specimen Standard approach during the past 10 (or so) years

5 Manual data entry

6 Challenges Blagoderov et al. (2012) Speed (or cost/specimen)

7 Challenges Speed (or cost/specimen) Training and managing of data entry personnel Space Transcription errors: quality control of data Damage to specimen Label (typically) not imaged: difficult to go back to the original data Specimen imaging treated as separate workflow

8 Solutions? Speed (or cost/specimen): ADBC experience May 2013: range $ (some higher ) Training and managing of data entry personnel: data entry guides, videos, senior databasers become mentors Space Quality control of data: data cleaning at various levels (databaser supervisor; database manager; expert colleagues; data aggregators, e.g. Discover Life)

9 Solutions? Label (typically) not imaged: difficult to go back to the original data: image labels at least for very important specimens (e.g., type specimens) Specimen imaging treated as separate workflow: imaging focuses on exemplar specimens per species rather than capture it all ; retrospective imaging possible

10 Benefits Full specimen data directly available: specimen can be mapped, data queried and downloaded etc. Staging (=specimen preparation) typically involves some collection curation (e.g., sexing): timeconsuming but beneficial for collection Each specimen is physically associated with a GUID/USI label Information from stacks of labels easily captured

11 Traditional specimen databasing approaches too slow and expensive Damage to specimens from handling

12 Semi-automated digitization: basics More recent take on specimen digitization Capture images of whole drawers High-resolution imaging Stitching of individual images into panoramas Capture specimen data from images: OCR etc. Don t physically attach USI labels to specimen (unless you move/touch them)

13 GigaPan North Carolina State University 2,700 drawers with small cost/time Images available through GigaPan webpage

14 Challenges Distortion of images (when stitched)

15 Solutions? SatScan (e.g., used at the Natural History Museum, London Blagoderov et al. (2012) Camera moves over specimens; no image distortion

16 Solutions? Hasselblad H4D-200MS (Queensland Museum)

17 Challenges Specimen covers label data Multiple labels cover each other

18

19 Solutions? Delta Robot Stitching of images: development of new algorithms 3D reconstruction of images: multi-view stereo (MVS) reconstructions Dietrich et al. (2012)

20 Solutions? Dietrich et al. (2012) Image medium-sized specimens, manually capture large and small? Micro CT scans?

21 Challenges How to capture label data? How to assign UIDs to individual specimens? Solutions? Segment images Assign UID

22 Dietrich et al. (2012) Challenges OCR for insect specimen labels: quality so far poor Solutions? Crowd sourcing

23 Benefits of semi-automated digitization (Very!) fast and low-cost Technology is advancing very quickly and a lot of the involved software is free Current drawbacks Does not work for large or small (= card or point-mounted) specimens No physically attached UID

24 In between Remove labels Image label and specimen Crowd Sourcing to transcribe labels

25 Citizen Scientists and Crowd Sourcing

26

27

28

29

30

31 How to deal with volunteer fatigue..???