DrawerDissect: Whole-drawer insect imaging, segmentation, and transcription using AI

Many museums curate vast collections of insect specimens. These collections represent invaluable records of biodiversity information, ecological patterns and phenotypic variation. However, traditional methods of imaging specimens and digitizing their metadata are labor-intensive and risk damaging delicate specimens. High-throughput imaging of entire specimen drawers integrated with computer vision artificial intelligence (AI) models can provide a potential solution. 2. We present DrawerDissect, a python-based pipeline for processing high-resolution drawer photographs, and a workflow to use it in entomological collections. By using custom vision models trained in the platform Roboflow and LLM-based transcription with Claude, DrawerDissect can crop and segment specimens from images and extract metadata from specimen labels. DrawerDissect is flexible, tuneable and modular, allowing seamless integration with downstream analyses of phenotypic features (e.g. color, pattern, and size). 3. We validated Drawerdissect by digitizing the Field Museum of Natural History's (FMNH’s) entire tiger beetle (family Cicindelidae) collection, processing 13,484 specimens to generate high-resolution dorsal photographs, backgroundless specimen images, and basic body measurements. Geographic data were successfully extracted for 3,648 specimens. To demonstrate the utility of the masked images, we provide an example integration of DrawerDissect with existing image analysis methods in R and ImageJ. Finally, to show the research potential of high-quality specimen images, we trained a species identification model, Cicindel-ID, using ~7,000 images of specimens in the genus Cicindela. 4. DrawerDissect’s novel multi-model AI workflow provides an efficient and reproducible framework that meets the demands of high-throughput digitization of natural history museum collections, unlocking the potential of vast specimen collections for future analyses.