ALICE: Angled Label Image Capture and Extraction for high throughput insect specimen digitisation

Abstract

The world’s natural history collections contain at least 2 billion specimens, representing a unique data source for answering fundamental scientific questions about ecological, evolutionary, and geological processes. Unlocking this treasure trove of data, stored in thousands of museum drawers and cabinets, is crucial to help map a sustainable future for ourselves and the natural systems on which we depend. The rate-limiting steps in the digitisation of natural history collections often involve specimen handling due to their fragile nature. Insects comprise the single largest collection type in the Natural History Museum, London (NHM), reflecting their global diversity. The NHM pinned insect collection, estimated at 25 million specimens, will take over 700 person years to digitise at current rates. In order to ramp up digitisation we have developed ALICE for Angled Label Image Capture and Extraction. This multi-camera setup and associated software processing pipeline enables primary data capture from angled images, without removal of the labels from the specimen pin. As a result ALICE enables a single user to sustainably image over 1,000 specimens per day, allowing us to digitally unlock the insect collections at an unprecedented rate.

Research applications of primary biodiversity databases in the digital age

“…Digitizing efforts for invertebrates have been particularly challenging, because many clades are so diverse, collections have much larger numbers of specimens, and the typically small specimens are difficult to digitize [96]. Automating digitization of such specimens, especially pinned insects and fluid-preserved invertebrates, faces significant obstacles [12,18,97–100].…”

Schrödinger’s phenotypes: herbarium specimens show two-dimensional images are both good and (not so) bad sources of morphological data

“…Increasing magnification and exposing hidden traits are complex and time-consuming but could be achieved if researchers fed back to institutions the images obtained while studying the specimens. Multi-camera imaging is particularly helpful for zoological specimens (Tegelberg et al, 2014; Price et al, 2018; Hereld et al, 2017; Ströbel et al, 2018; Hereld and Ferrier, 2019), as it allows proper visualization of traits that are hidden or distorted on images in a single view. Combining multi-view imaging and production lines (Tegelberg et al, 2014) is one efficient option to generate imagery that supports both 3D reconstructions (Hereld et al, 2017; Ströbel et al, 2018; Hereld and Ferrier, 2019) and accurate morphological data acquisition from the original two-dimensional images themselves, as shown here.…”

Conceptual design blueprint for the DiSSCo digitization infrastructure - DELIVERABLE D8.1

“…There is evidence that industrialising digitization of natural science collections is achievable [Beaman and Cellinese 2012, Heerlien et al 2015, Rogers 2016. The work demands and costs can be brought down at least by factors of 5-10 compared to manual methods [Oever and Gofferje 2012]. Efficient workflows can be designed when all material is first imaged and followed by the subsequent transcription of data from the images [Lehtonen et al 2011, Nelson et al 2012.…”