The image capture setup requires five key pieces of equipment: a copy stand, lightbox, foam sheet, tempered glass, and a camera. A detailed list of supplies is provided in Suppl. material 2. We used a Beseler CS-14 copy stand with a custom-built lightbox, based on the design by Ortiz-Acevedo et al. (2020), placed on the copy stand as an imaging platform (Fig. 1). The lightbox features two strips of LED lighting along the inner walls to provide backlighting and minimize shadows. Its inner base was lined with a removable neutral gray surface, made from either craft foam or a metal plate. For enhanced contrast, white or black foam backgrounds were used for specimens with gray wing margins.

Photographs were taken with a Canon EOS7D camera, with a 60-mm lens for most specimens. Larger specimens, such as Antheraea polyphemus (Cramer), were imaged using a 18–55 mm lens. Image quality was optimized using camera features such as white balance, aperture, shutter speed, and ISO sensitivity. Midway through the project, we incorporated the Datacolor Spyder X Elite monitor calibration software (Datacolor, Lawrenceville, New Jersey, purchased 2021) to further standardize image lighting.

Our protocol includes a color card, silicone putty, and gray foam board placed on top of tempered glass. The gray foam board was constructed by gluing a layer of gray foam over plank Styrofoam. Each component of the setup was placed in a standardized location. The color card was positioned on the glass so that it would appear in the top left corner of the image. The silicone putty used to hold the specimen was placed adjacent to the foam board, ensuring the foam board did not interfere with the specimen (Fig. 2). The data labels, barcode, ruler, and scientific name were arranged on the foam board.

For pinned-specimen digitization, we either used a green colored paper barcode label, measuring 19 mm x 13 mm, printed at MGCL with BarTender 10.1 (Seagull Scientific, Bellevue, WA, USA) with code 128 or a 14 mm x 10 mm plastic laminated data matrix barcode from AlphaSystems (a Hunkar Technology Company, Midlothian, VA, USA). Each label included a human- and machine-readable unique identifier, to provide essential specimen tracking and standardize image file names.

The program EOS Utility version 2.11 (Canon, USA) was used to take .cr2 and .jpg images. Settings on the camera were F-stop: F/16, sensitivity to light (ISO-200), and exposure time (shutter speed 1/15 sec). For image processing, multiple custom scripts (Leopold and O’Connell 2023) were written in Python3 (Van Rossum and Drake 2009). These scripts can be used by other collections for the same tasks and/or have data matrix or code 128 barcodes. Scripts were for renaming images (“Datamatrix.py” script), reducing size of images (“Rescale.py” script), generating .csv file of images in current working directory to make a skeletal data file (“wls.py” script), and zipping images into folders for long term storage (“Zipper.py” script). The “Datamatrix.py” script (Leopold 2021) used the libraries “dmtx” (Misbakh-Soloviov 2016) and “zbar” (Brown 2012) to extract the barcode and rename the file. The “Rescale.py” script used the PIL module, specifically the submodules “Image” and “ExifTags”, to reduce the size and rotate the images (Clark 2021). The correct image orientation was determined by the value extracted from ExifTags, which was an integer between one and eight with one indicating zero degrees and eight indicating 90 degrees. The “wls.py” script was a skeletal file upload, which took the images in a file and generated a .csv file listing all file names. The scientific name was entered manually into the .csv file and uploaded onto SCAN. The “Zipper.py” script zipped specified images into a new folder of a specified size for long term archiving, sharing images with collaborators, or uploading onto SCAN.

Records were transcribed on NfN, a digital platform aimed at engaging citizen scientists from around the world (Hill et al. 2012). Each project is called an “Expedition.” Every specimen in the expedition must be transcribed by three people to be considered complete and the transcriptions were compiled into one .csv file. Transcriptions from NfN were cleaned then uploaded onto SCAN. Select records were also transcribed directly into SCAN.

Our workflow consisted of four modules: 1) Pre-Imaging preparation; 2) Specimen imaging; 3) Image processing; 4) Transcription. Detailed documentation for pinned-specimen digitization is provided in Suppl. material 3. Pinned-specimen digitization workflow required 1–2 people to complete the following steps: 1. Removing the specimen from the drawer and preparing the specimen and label data for imaging. 2. Imaging. 3. Repinning data labels and data matrix barcode. 4. Returning the specimen to the drawer.

A project plan and specimen selection were crucial to the success of digitization. A project plan communicates to collection users all details of what specimens will be digitized, project length, and outcomes, and where specimens are temporarily stored. The MGCL is a large facility, making coordination essential to avoid disruptions to other users. Specimen selection was guided by factors such as curation quality, researcher’s data needs, and institutional objectives. We moved drawers to a part of the collection near the digitization station to minimize traffic and use of the compactors. Labels with the scientific names of taxa, which were placed in the image of the corresponding species, were printed in advance.

The imaging interface, EOS Utility, was opened and a destination folder for images was created. The imaging surface of the copy stand was cleaned. The camera was adjusted to the optimal height and the camera settings, including the shutter speed, ISO, and aperture were checked.

The workflow below is described as if one person is digitizing (Fig. 3). If two people work together, one person can manage specimen related tasks and the second person can take photographs. The imaging process began with setting up the specimen. The specimen was taken out of its drawer, its labels removed from the pin, and placed on the gray foam board along with the new barcode label on the tempered glass. The specimen pin was inserted into a small piece of moldable silicone putty on the glass and the digitizer checked that the specimen’s wings were level. Labels were organized on the foam board in the following manner: scientific name was placed on top with locality labels directly below and miscellaneous labels either below or to the right of locality labels. Accession and catalog number labels were placed at the very bottom. While this is a general template on how labels are arranged, the organization differed when there were very large labels associated with the specimen. In such cases, labels were arranged so that all can fit within the area allocated for photography. The dorsal side was always imaged first because the “Datamatrix.py” script appended “_D” for the first image of a barcode. The camera was then adjusted so the specimen was in focus and a dorsal image was taken simultaneously in a .cr2 and .jpg. The specimen was turned over and the head of the pin was placed in the putty, so that the ventral side was facing the camera. Labels were checked to confirm that information was not written or printed on the back of the labels. If information on labels was present on the backside it was imaged in the ventral photo. The camera was focused again and an image was taken. In some rare cases, a third image was taken if there were large data labels. If the specimen was pinned laterally, only one image was taken. Image capture of laterally pinned was not standardized (e.g. consistently imaging either right or left side) and is an opportunity for further standardization. Data labels were placed back on the pin. The barcode label was pinned last and faced downward. The specimen was placed back into its original drawer or unit tray. At MGCL, we created a system so that at the end of the imaging session, the digitizer placed a colored pin in the insect drawer to denote where the image session was terminated. The scientific name, number of specimens imaged, and name of the digitizer were typed in a separate Excel spreadsheet to keep track of image metadata. Imaging equipment was cleaned and stored until next use.

Four Python scripts were utilized to process images (Fig. 4). The “Datamatrix.py” script processed an image to extract a specimen catalog number from a barcode. It renamed the image file to match the barcode data and used a directional indicator that rotated the image to a standardized vertical position. The script also appended the file name to note if the image was dorsal (“_D”), ventral (“_V”), or lateral (“_L”). An image of the dorsal side was taken before the ventral side, and the script used this consistency to rename images. Lateral images were renamed by hand because the default was for the first specimen of a unique barcode to be appended with “_D”. Less than one percent of specimens was renamed by hand because the barcode was unreadable. The “Rescale.py” script created a low resolution .jpg image from the original image, and the script, “Zipper.py”, zipped the downscaled images into a folder. The zipped images were uploaded onto SCAN. A fourth script, “wls.py”, generated a .csv file that listed all catalog numbers in each folder. The numbers were used to create a skeletal file which captured the scientific name and catalog number which was uploaded onto SCAN and linked to images. Scientific names were entered manually in the .csv file. Records with images and skeletal data on SCAN were then used for label transcription.

The main steps of an NfN expedition were specimen selection, creating a spreadsheet of data to be captured, and writing a paragraph containing information for volunteers about the expedition (Fig. 5). Expeditions had a particular theme, usually a taxonomic group to align with the needs of the institution or researcher. Once images were selected, they were downscaled using the “Rescale.py” script and zipped so that images were easier to upload to NfN. Before zipping, we made a Manifest: a .csv file that contained the catalog number and scientific name of specimens included in the NfN transcription expedition. To make the Manifest, we ran “wls.py” to capture the image file name in the folders. We manually added the scientific name and the catalog number to the .csv file, the latter of which was obtained easily from the image file name. If it was the first expedition in a series, we wrote a welcome blog, a paragraph explaining why these organisms are interesting, how data will be used and some eye-catching images. We included an instruction guide for transcribers with protocols of how to parse data and transcribe in a standardized method. The Manifest, images, welcome blog, and transcription instructions were sent to NfN via email. Once the expedition was launched, we announced the expedition on the NfN blog. We engaged transcribers throughout the expedition by answering questions and engaging in conversation. Transcribers were thanked at the conclusion of the expedition. Data from all three transcriptions per record were compiled by NfN staff, then sent to MGCL. Data in the reconciled .csv file were manually cleaned column by column to standardize data then uploaded onto SCAN.

We tested our approach through the LepNet digitization efforts. Approximately 162,000 raw images of 81,000 specimens, representing ~2500 species of Lepidoptera, were digitized at the MGCL from September 2016 to May 2021. The number of images taken increased year over year from 9,872 in 2017, to 23,438 in 2018 to 37,209 in 2019. Most specimens were imaged twice (once dorsally and once ventrally). Although not a LepNet priority, on occasion, we imaged a laterally pinned specimen and its label data to enable data capture of the record. Specimens were imaged across 23 families with the majority in Erebidae (61,015 images), Saturniidae (37,301 images), Geometridae (18,903 images), and Sphingidae (17,761 images). Approximately 54,000 transcriptions (64% of LepNet specimens) were fully or partially completed. Of these 54,000 transcriptions, 21,500 were skeletal transcriptions (only genus, species, and country and/or state/province transcribed) and 32,500 were full transcriptions. The MGCL had 23 expeditions through Notes from Nature, totaling ~18,400 specimens. Of these 18,400 specimens, 10,400 were uploaded to SCAN. The remaining 8,000 needed data cleaning and were not included in the transcription total.

From February 2019 to May 2021 (27 months), MGCL personnel involved in the project collectively worked for 70,296 minutes (1,171.6 hrs) to obtain 54,099 images (includes dorsal and ventral), an average of 23 specimens per hour (sp/hr). The rate of imaging with one digitizer ranged from 11 to 23 per hour and was ~18 sp/hr while with two digitizers, the rate ranged from 13 to 44 per hour and was ~25–26 sp/hr. Forty-one community members and/or students volunteered. There were eight staff members, four of which were students who received hands-on training in a setting that usually has few opportunities for employment. Additionally, ~30 middle- and high-school students imaged and transcribed specimens for a week-long summer camp as part of the Florida Museum of Natural History’s Junior Volunteer program.

The combination of a lightbox and a copy stand for digitization creates a user-friendly and flexible method to image Lepidoptera specimens. The setup can accommodate most butterflies and large moths; increased magnification is required for imaging micromoths. Our digitization set up can image large specimens and specimens with unique morphologies such as those with long tails on their hindwings (e.g., some Saturniidae) by imaging on the glass surface or in the interior of the lightbox. This is due to the fact that the height of the camera mount on the copy stand can be adjusted. A lightbox and copy stand can be set up at minimal cost and can be used to generate images quickly. Once set up, the equipment is easy to maintain and use. The benefit of designing a customized lightbox to use in-house is that it can be made to desired optimal specifications, usually at a lower cost point. Lightboxes at MGCL were made of readily available supplies, including a wood frame coated in white paint, with LED strips glued and soldered on the inside for illuminating specimens during digitization. The main drawbacks with a copy stand in the MGCL digitization workflow centered around lighting. It is preferable to have all digitization stations in a single workspace, but in practice, it is often necessary to have digitization stations in different workspaces that inherently have variable lighting conditions. There are multiple sources of lighting to control, including copy stand lightbulbs, LED strips, and ceiling lighting. The MGCL had three digitization stations, each in a different room and with different ambient lighting, which required additional methods to ensure the accuracy of specimen coloration. For example, wax paper was put over the bulb fixtures in order to remove excessive glare of the light from the copy stand fixtures. Light- and color-calibration was performed with DataColor Spyder X Elite for each digitization computer monitor to standardize lighting between imaging stations. Another issue was that specimens of different sizes required different lighting angles and camera settings. Digitizers kept track of camera settings and the angles and heights of lights to standardize image capture throughout each project.

The goal of the pinned-specimen imaging workflow was to capture specimen images and label data. For spread specimens, dorsal and ventral images were taken. For field-pinned specimens with unspread wings, a single lateral image was taken. We decided to image the dorsal and ventral sides of specimens because the ventral side of lepidopterans often has species-specific characteristics such as eye spots and other markings that are absent on the dorsal side. The main workflow decisions centered around specimen handling, data gathering, and speed and imaging efficiency.

We prioritized minimizing contact with specimens to ensure their long-term their preservation by handling specimens only once. To maximize the information obtained, we captured dorsal and ventral images along with associated label data. Although imaging both sides of specimens required more time initially, our approach reduced overall handling and proved more efficient in the long term.

We also had to decide whether to image label data separately or simultaneously with the specimen. Ultimately, we chose to capture label data simultaneously in images in both dorsal and ventral images. This approach was based on the idea that the images should aid in identification and analyses, rather than serve as publication-quality outputs, while also facilitating transcription. Another benefit of capturing the label data and specimen in the same image is that the photo and transcription can be checked against each other by future researchers. However, a drawback of this method is that the specimen occupies a smaller percentage of the image due to the inclusion of label data. One advantage of our setup is that information on data labels can be easily captured and does not require additional images to be taken. By standardizing the placement of label data, we aim to simplify the re-pinning process and reduce the need for searching for the desired label during transcription.

Data labels were placed on a platform ~0.75 inches tall, constructed from plank Styrofoam. This material was chosen for its durability – it resists breaking or denting while remaining malleable enough to accommodate pins. The elevated platform ensures that the label data and the barcode are in the same focal plane as the specimen.

Another consideration was whether imaging was more efficient solo or in tandem. A single digitizer averaged 18 sp/hr, which could be as high as 23 sp/hr if the digitizer was highly experienced. In contrast, two-person teams averaged 23–24 to upwards of 44 specimens per hour. The average difference in efficiency is minimal because, in tandem setups, one person may work faster than the other, leading to brief idle periods. However if two experienced imagers are working together the average could be significantly greater than normal. Tandem imaging offers benefits such as task specialization, allowing each person to focus on fewer responsibilities, and reducing feelings of isolation—an important factor for volunteers. Additionally, imaging efficiency varies with specimen size; smaller specimens are quicker to image than large specimens.

Images were processed every few weeks, with editing taking a few hours, primarily due to the time required for scripts to run. The “Datamatrix.py” script failed to rename some barcodes because there was debris on the barcode, a pinhole through the barcode, or part of the machine-readable barcode being either out of focus or missing from the image. In these instances, image files were renamed manually. For large Lepidoptera, such as Attacus L. spp. or Antheraea polyphemus, the camera had to be mounted higher, resulting in barcodes appearing too small for successful automated renaming.

Transcription in NfN had its advantages and disadvantages, with the primary considerations being time and quality. The NfN is a more time-intensive process for the expedition coordinator, requiring significant effort to prepare images for the expedition and to clean and upload the data afterward. Smaller projects may not justify the time investment, while larger projects risk prolonged completion times.

Transcriptions from NfN exhibited greater variability, requiring significant effort to standardize. This was because transcribers often entered label data exactly as it appeared, with fields like locality frequently abbreviated in inconsistent ways. To streamline formatting, we applied batch changes in Excel to standardize fields such as names, dates, and localities. For efficiency, we prioritized cleaning frequently repeated data, such as locality and collector names. Additionally, we added scientific names before the expedition to ensure accuracy and minimize transcription error. The focus on our project was developing the imaging infrastructure, and there are opportunities to improve the transcription workflow and better understand the efficacy of transcriptions from NfN. A future direction is to incorporate technologies and software to expedite cleaning. Options include OpenRefine software to clean and reconcile messy data (Delpeuch et al. 2024) or to incorporate Label Babel, recently developed machine learning optical character recognition methods integrated with NfN (Guralnick et al. 2024). Another emerging approach is to use artificial intelligence and large language models, to do the initial transcription then check these data by humans (Herbst et al. 2025).

We conducted preliminary tests to assess whether a copy stand could be used to image an individual fluid specimen. In our tests, a specimen and data labels were removed from the jar and partially dried, then imaged on the copy stand (Suppl. material 1). We believe that this method may be useful to image individual specimens but requires further testing to optimize image quality and scalability of the method.