FieldPrism: A system for creating snapshot vouchers from field images using photogrammetric markers and QR codes
This article is part of the special issue “Advances in Plant Imaging across Scales.”
Field images are important sources of information for research in the natural sciences. However, images that lack photogrammetric scale bars, including most iNaturalist observations, cannot yield accurate trait measurements. We introduce FieldPrism, a novel system of photogrammetric markers, QR codes, and software to automate the curation of snapshot vouchers.
Methods and Results
Our photogrammetric background templates (FieldSheets) increase the utility of field images by providing machine-readable scale bars and photogrammetric reference points to automatically correct image distortion and calculate a pixel-to-metric conversion ratio. Users can generate a QR code flipbook derived from a specimen identifier naming hierarchy, enabling machine-readable specimen identification for automatic file renaming. We also developed FieldStation, a Raspberry Pi–based mobile imaging apparatus that records images, GPS location, and metadata redundantly on up to four USB storage devices and can be monitored and controlled from any Wi-Fi connected device.
FieldPrism is a flexible software tool designed to standardize and improve the utility of images captured in the field. When paired with the optional FieldStation, researchers can create a self-contained mobile imaging apparatus for quantitative trait data collection.
In contemporary specimen collection, the integration of multiple digital data sources necessitates the utilization of advanced processing techniques to consolidate notes, observations, photographs, and specimens into a comprehensive and cohesive data set. Fortunately, software packages can streamline this cumbersome task. There are apps to help organize field observations (Maya-Lastra, 2016; Powell et al., 2019; Seltzer, 2019), manage ecological surveys (Aanensen et al., 2009; Anastácio et al., 2017; Esri, 2019; Nowak et al., 2020), and measure specimens (Easlon and Bloom, 2014; Schrader et al., 2017; Weaver et al., 2020). However, the curation of digital images captured during fieldwork has received little attention.
Researchers frequently photograph specimens to document an occurrence or observation, but these images can often also be rich data sources of trait measurements for the original collector or other researchers. For example, herbarium specimen vouchers are widely used in the natural sciences to chronicle climate change (Calinger, 2013; Willis et al., 2017b; Jones and Daehler, 2018; Meineke et al., 2018), ground plant systematics and taxonomy (Stuessy, 1975; Wen et al., 2015; Espinosa and Pinedo Castro, 2018), inform conservation initiatives (Greve et al., 2016; Nualart et al., 2017; Lughadha et al., 2018; Rocchetti et al., 2021), and provide a wealth of trait data for phylogenetic and phenological studies (Savolainen et al., 1995; Edwards and Smith, 2010; Bakker, 2017; Willis et al., 2017a; Mishler et al., 2020; Folk et al., 2021). Vouchers are particularly useful because of their standardization and availability. However, herbarium specimens are not without their challenges. Preservation and mounting techniques can often distort or obscure phenological and morphological traits, particularly those related to three-dimensional structure, while colors may also fade and change as specimens dry and age (Willis et al., 2017a). In addition to vouchers, images taken at the time of collection can serve as supplementary information to the extended specimen record, allowing for a more complete understanding of specimens and their conditions at the time of collection (Baskauf and Kirchoff, 2008; Gómez-Bellver et al., 2019; Hardisty et al., 2022). Platforms like iNaturalist have led to an explosion of digital-only records that are regularly used to study phenology, species distributions, conservation, and climate change, and can be linked to physical vouchers (Heberling and Isaac, 2018). If these field images were standardized and metrically validated, they could also be used for quantitative trait data collection.
Several authors have established standardized protocols for capturing images of live plants that can serve as a supplement or substitute for physical herbarium specimens. The term “photo vouchers” is defined by Funk et al. (2017) as a collection of photographs of a plant that facilitates identification and documents its presence when the collection of a physical specimen is not possible. Gómez-Bellver et al. (2019) provide a definition for “fusion vouchers” as a compilation of images attached to a standard herbarium sheet that enhances a physical specimen by providing supplementary contextual information when its size, structure, toxicity, fragility, or special status requires further representation. While these protocols provide valuable context and content that can provide a more comprehensive understanding of plant habit, color, structure, and identification, they offer limited utility for measuring morphological traits. Quantitative traits are increasingly used in a wide array of biological research, but most field images lack photogrammetric scale bars required for accurate measurements.
To enable quantitative trait data collection from field images, we introduce a new specimen documentation method called “snapshot vouchers.” This method involves capturing images of living plant organs against a photogrammetric background, enabling precise measurement of morphological traits. Snapshot vouchers can enhance physical vouchers or serve as digital-only records. By incorporating a photogrammetric background, iNaturalist photos can also qualify as snapshot vouchers.
To realize the full potential of snapshot vouchers and field images we created FieldPrism, a flexible software and hardware suite designed to streamline image processing and management. FieldPrism addresses common workflow bottlenecks by providing automatic image skew, yaw, and pitch correction; pixel-to-metric conversion; file renaming; and metadata collection tools. Optionally, FieldPrism can also address the administerial task of organizing and renaming hundreds or thousands of uninformatively named image files, an otherwise laborious and error-prone procedure (Forssman et al., 2016). Handwritten labels in images are not readily machine readable, so we employ QR codes to discriminate between collections. QR codes are a popular organizational tool in biological research and have been used to manage DNA extraction from herbarium specimens (Folk et al., 2021), manage museum collections (Sendino, 2013), link specimens with their digital metadata (Diazgranados and Funk, 2013), and link iNaturalist records to physical specimens (Heberling and Isaac, 2018).
Given the unpredictability of specimen collection in the field and the wide range of project scopes and complexities, we saw the need for a flexible and scalable system for identifying snapshot vouchers. This motivated the development of our QR code flipbook where users can assign QR codes to different species, sites, days, leaves on a plant, or other collection identifiers. For example, the QR codes for numbers 1–5 may correspond to five different species, while the codes for letters A–E may correspond to different locations. When the user takes a picture of species 1 at location A, they simply flip to those codes, take the picture with those codes in view, and continue collecting. Alternatively, QR codes can directly contain predefined collection identifiers, like the species name, collector's name, or location. FieldPrism will organize and rename the photos accordingly. The flipbook can then be reused on another trip where the numbers and letters will refer to other species and locations.
Archival images of herbarium specimen vouchers typically include a scale bar, which is a key feature often missing or incorrectly utilized in field images because they are often skewed and not rectilinear. If the camera is not positioned perpendicular to the subject or the scale bar is not on the same plane as the subject, its usefulness in obtaining accurate measurements is significantly limited. Additionally, converting pixel distance into metric distance requires a manual process, such as using ImageJ (Rueden et al., 2017) to measure the scale bar. With FieldPrism, we developed a simple photogrammetric background template that, in conjunction with our machine learning and computer vision algorithms, allows for the automated conversion of pixels to metric units. By using this template, FieldPrism can correct a variety of image distortions, including skew, yaw, and pitch, and produce a rectilinear snapshot voucher that compensates for cases in which the camera was not positioned perpendicularly to the subject.
Our tools can be used either independently or together depending on the project requirements. For instance, the QR code flipbook can be used to organize snapshot vouchers on a specimen collection trip, but it may not be necessary for an iNaturalist collection. FieldPrism methods are compatible with any digital camera, including smartphones. However, for additional utility, we developed a mobile imaging apparatus called FieldStation that enhances imaging ergonomics, operates without an internet connection, captures GPS coordinates, and saves all data redundantly to multiple USB storage devices. FieldStation's monitoring and control interface is accessible via any smartphone, tablet, or computer, but it can also be operated without a monitor using physical controls and audio feedback. With the goal of supporting the generation of quantitative trait data, FieldPrism uniformly curates snapshot vouchers and emulates the standardization of physical herbarium specimen vouchers.
METHODS AND RESULTS
Below, we provide an overview of each FieldPrism component and outline possible use cases. First, we describe the photogrammetric background template, which is foundational for producing snapshot vouchers. Then, we explore three possible ways to use our QR code system as part of a specimen identification hierarchy and how to organize the QR codes in a flipbook. Finally, we describe the FieldStation and the FieldPrism software and test our methods under a variety of conditions.
Photogrammetric background template
In many disciplines, scale bars and rulers routinely accompany the subject matter to provide a notion of scale or enable pixel-to-metric conversion a posteriori. We developed a photogrammetric background template (FieldSheet) to enable precise pixel-to-metric conversion and image distortion correction (Figure 1, Figure 2A). The simplest FieldPrism collection workflow involves attaching a FieldSheet to a clipboard, placing the subject onto the FieldSheet, and taking a photo with a digital camera (e.g., a smartphone; Figure 1A). The FieldSheet consists of four markers, each containing four 1-cm squares, and includes a 10-cm line for reference. Users can print copies of our FieldSheet using any standard printer onto six common page sizes: A3, A4, A5, Tabloid, Letter, and Legal. FieldSheet templates can be downloaded from https://www.fieldprism.org, or users can customize FieldSheets by editing the “FieldSheetBuilder.yaml” configuration file (see Data Availability Statement). If even larger templates are required, the four corner markers can be cut out and mounted on a larger surface, with care to maintain rectilinear placement. This custom size is supported in the FieldPrism software by updating the “custom” template dimensions. We validated a large custom FieldSheet using a 55 cm × 71 cm white hobby poster board with FieldSheet markers pasted in each corner.
When printing FieldSheets, it is critical to verify that the printer did not improperly scale the PDF. Users can manually measure the 10-cm line or compare the size of a credit card with the accompanying size validation sheet and set the printer to 100% scaling if necessary. After verification, the FieldSheet can be attached to a clipboard and is ready for use. We recommend printing FieldSheets on waterproof paper or the waterproof shipping labels (e.g., Avery 5526 Waterproof Film Shipping Labels; Avery Products Corporation, Brea, California, USA) and sticking the whole sheet onto a plastic or metal clipboard. Make sure that the subject of each photograph is positioned at the center of the FieldSheet and does not cover the markers, and take care to ensure that it is in the same plane or as close as possible, especially for subjects with three-dimensional structure. Images that include our FieldSheet can be processed with the FieldPrism software to produce curated snapshot vouchers.
To ensure that FieldPrism is broadly accessible, we validated the FieldPrism software with images taken using smartphone cameras (Android and Apple), mirrorless and digital single-lens reflex (DSLR) professional cameras with a variety of lenses, and an OAK-1 machine vision camera (Luxonis, Littleton, Colorado, USA) (Table 1). Each camera produced usable snapshot vouchers. While FieldPrism can correct skewed images, it cannot correct distortion introduced by fisheye or wide-angle lenses. Wide-angle lenses should be avoided because they produce curvilinear images that distort metric increments across the image. If the manufacturer provides a lens distortion correction profile, it should be applied to the batch of images prior to using FieldPrism to correct skew, yaw, and pitch distortions. If the specimen contains small features or if the FieldSheet is far from the camera and occupies a fraction of the frame (see Figure 3, row 4), then consider using a camera with a sensor resolution greater than 20 megapixels to retain detail.
|Camera||Megapixels||Focal distance (mm)a||Horizontal field of view (degrees)||Aperture||ISO||Shutter speed (s)|
|Samsung Galaxy S9+||12||26*||65||f/1.5-2.4||Auto||Auto|
|Panasonic Lumix G9||20.3||50*||40||f/5.6||800||1/125|
- a Focal distances with an asterisk are the full-frame equivalent focal distances, as those cameras have smaller sensors.
- b Images taken by these cameras were used to train the machine learning algorithm but were not part of validation testing.
QR codes and specimen identification hierarchy
To fully take advantage of our QR code system for generating unique specimen names for collections projects, users must define a specimen identification hierarchy before venturing into the field. A specimen identification hierarchy is a modular organizational scheme that enables users to define short identifiers that can be easily translated into QR codes (Table 2). A project's specimen identification hierarchy is prepared as a CSV file where each column corresponds to a hierarchical level and each row is an identifier component of the parent hierarchical level. We recommend using no more than six hierarchical levels because using more would make the QR code flipbook impractical, but there is no upper limit to the number of variants (rows) per level.
|QR code usage||Columns representing hierarchical levels|
|B. Indeterminate with Aliasb||Level_1||Level_1_Alias||Level_2||Level_2_Alias||Level_3||Level_3_Alias|
- a Example reconstructed determinate image identifiers: Site-2_Day-3_Collector-4_Quercus velutina_sun_adaxial, Site-3_Day-2_Collector-3_Quercus rubra_shade_adaxial.
- b Example reconstructed indeterminate image identifiers: Quercus alba_sun_adaxial, Quercus velutina_shade_.
- c Example reconstructed hybrid image identifiers: Site-2_Quercus alba_shade, Site-3_Quercus rubra_sun.
Users have three options for defining a specimen identification hierarchy: (1) predefining the names of specimen identifier components, (2) using placeholder values for specimen identifier components when the names of specimens or locations cannot be anticipated, or (3) applying a hybrid implementation with some predefined names and some placeholder names. Placeholder components allow for indeterminate imaging schemes. For example, a field expedition is unlikely to know which species or how many individuals they will need to photograph. In this case, a placeholder QR code “A” could represent “Quercus macrocarpa” (Table 2B). While in the field, write down “Quercus macrocarpa” next to the “A” QR code in the flipbook and then add this alias to the naming hierarchy CSV file as the alias to “A”; FieldPrism will then know to translate “A” accordingly. A mixture of determinate and indeterminate components can also be used in a hybrid format. Additional specimen identification hierarchy examples can be found in Table 2.
QR code flipbook
After defining a specimen identification hierarchy, the FieldPrism QR code generator will create printable PDFs containing either 24 or 40 QR codes per sheet depending on the length of the label text. To create a flipbook, users cut out each QR code and attach them to a small memo-sized notebook so that each page contains only one variant per hierarchical level (Figure 2B). Users then cut the notebook page between each QR code. The result is an organized flipbook that can display any combination of variants in the defined naming hierarchy. We strongly recommend printing QR codes onto the previously mentioned waterproof package shipping labels. These labels prevent ink bleeding, are weather resistant, and conveniently attach to Rite-In-The-Rain notebooks (Rite in the Rain, Tacoma, Washington, USA). Printing QR codes with toner onto regular paper is discouraged due to smearing. We provide two predefined QR code templates that will accommodate most use cases, but users can edit the “FieldSheetBuilder.yaml” configuration file to customize the QR codes (e.g., shrinking the codes to fit 64 per sheet).
Holding a clipboard while setting the QR code flipbook, adjusting the specimen, and taking a photo with a smartphone or professional camera is an ungraceful ballet at best and may even require two people. To ameliorate poor ergonomics, we created FieldStation, a mobile imaging apparatus that can be used in conjunction with the FieldPrism software and FieldSheets (Figure 2D). FieldStation is built around a Raspberry Pi 4 Model B (https://www.raspberrypi.com; Raspberry Pi Foundation, Cambridge, United Kingdom) and an OpenCV OAK machine vision camera and can be assembled using off-the-shelf components in less than two hours for approximately $600 USD, depending on the configuration (Appendix S1; see Supporting Information with this article). We used a storage clipboard to contain the Raspberry Pi, batteries, USB storage, and GPS module. A microphone stand is attached to the top of the clipboard to hold the OAK-1 camera and GPS antenna. A mini six-key keyboard is used to control the camera and GPS. For monitor-free use, headphones can be plugged into the Raspberry Pi; different tones will sound to indicate successful operations or errors. Camera straps are attached to the sides of the clipboard, improving ergonomics by allowing the user to wear the imaging apparatus, leaving hands free for other tasks, and ensuring that the camera is aligned with the FieldSheet.
The FieldStation GUI allows the user to control and monitor the imaging process. A focus check window shows a cropped portion of the live camera feed, while the larger window displays the saved image. In the right panel, various verification statuses are displayed. We use Laplacian variance to detect blurry and out-of-focus images and give the user a chance to take a sharp photo. The suffix “_b” is optionally added to all blurry photos for quick quality control. GPS information is also displayed here and can be refreshed independently of capturing an image. FieldStation can save all images and metadata to four USB storage devices simultaneously for data redundancy.
FieldPrism is a simple-to-use Python program that can be run from the command line and customized with a configuration file. The source code, user manual, and FieldStation build guide are freely available at https://www.fieldprism.org and https://github.com/Gene-Weaver/FieldPrism (see Data Availability Statement). We also provide FieldSheets (six sizes), size-check templates, example naming hierarchy files, and example images at https://www.fieldprism.org/demo/FieldPrism_Kit.zip. The software can be installed on Windows, Linux, and OS X systems inside of a Python virtual environment. Full installation instructions are available at our GitHub repository. To process images with FieldPrism, adjust settings in the “FieldPrism.yaml” configuration file and then run the “FieldPrism.py” file. We outline the steps of the FieldPrism software below.
Step 1: Preprocessing and supported image formats
Before running FieldPrism, it may be useful to perform some preprocessing on the images. FieldPrism does not adjust color, brightness, or contrast, so if lighting conditions were poor at the time of collection, then batch-processing tools like Adobe Lightroom (Adobe, San Jose, California, USA) can be used to improve the quality of the images. Similarly, if the camera lens has a manufacturer-provided lens distortion correction profile, then it should be applied to the batch of images at this stage. All images should be exported as full-resolution JPG files. If the collection project is intended to emulate fusion or photo vouchers, or if color accuracy is important, then an archival-grade color correction card should also be included in the image, affixed to the clipboard (Funk et al., 2017; Gómez-Bellver et al., 2019).
Step 2: Identifying photogrammetric markers
After preprocessing images, FieldPrism needs to identify the photogrammetric markers. Prior to processing, FieldPrism will rotate all images into a portrait orientation, which is necessary to determine the orientation and skew of the FieldSheet in each image. FieldPrism will also check for corrupt image files and illegal characters in file names, replacing illegal characters with a dash or underscore.
We trained a custom implementation of a YOLO-v5 object detection machine learning network (Jocher et al., 2022) to locate photogrammetric markers on the FieldSheet. We compiled 1119 images taken with six different cameras (Table 1) that contained examples of all five versions of our FieldSheet, indoor and outdoor locations, very dim to very bright lighting conditions, full QR code sheets, QR code flipbooks, and examples of ideal and poor imaging scenarios. To bolster the reliability and generalizability of our network, we used transfer learning to extend the capabilities of a YOLO-v5 network that was trained on 6394 fully annotated herbarium specimen images from an unpublished data set. This data set is composed of angiosperm images sampled from all contributing herbaria in the Global Biodiversity Information Facility (GBIF; https://www.gbif.org/); the images were annotated to identify archival components including rulers, labels, maps, photographs, attached items, barcodes, color correction cards, and weights. The full list of sampled herbaria data sets is available in Appendix 1. The FieldPrism YOLO-v5 variant locates QR codes, photogrammetric markers, and (optionally) text labels.
Step 3: Distortion correction
With bounding box predictions around each of the four photogrammetric markers, FieldPrism calculates centroids and determines the orientation of the FieldSheet within the image based on the relative marker locations. It then calculates the perspective transformation required to deskew the FieldSheet and apply a geometric transformation to the entire image, creating a distortion-corrected image. If all four photogrammetric markers cannot be identified, then the image cannot be deskewed, but all identified photogrammetric markers are processed in step 4 to attempt to calculate the pixel-to-metric conversion ratio.
Step 4: Pixel-to-metric conversion
Using the distortion-corrected image, FieldPrism re-identifies the four photogrammetric markers with the trained YOLO-v5 network. Within each predicted bounding box, FieldPrism progressively implements image binarization (the process of segmenting an image into zeros and ones given a color or intensity threshold) through a range of threshold values until the binary image matches a known, ideal binary representation of the photogrammetric markers. In the user manual, we outline a progressive process that attempts metrological conversion despite poor image quality (e.g., uneven lighting).
Step 5: Locate and decode QR codes
Using the trained YOLO-v5 network, FieldPrism places bounding boxes around all QR codes in both the distortion-corrected images and the images that failed the correction process. The QR codes contain metadata indicating the hierarchical level to which the specimen identifier component belongs, together defining the specimen identifier contained within the set of QR codes. Users can define error handling for indecipherable QR codes, but the default setting in FieldPrism reconstructs the specimen identifier with placeholder values like “Level-2” to fill in for the unrecoverable specimen identifier component. These placeholder values aid quality control and make it simple to correct the final names. If the project utilized an indeterminate hierarchical naming structure, at this stage specimen identifiers are converted to the final informative file name using the alias table. If multiple images contain the same set of identifiers, then FieldPrism can append either a number or the original filename to the end of the specimen identifier.
Step 6: Apply user-configurable processing options
Distortion-corrected images can be processed to meet different user requirements. Step 3 outputs an image that retains most of the content and similar dimensions to the unprocessed image (Figure 3C). Users can opt for a top-left justified image, which maintains some of the original background and the same pixel density as the unprocessed image (Figure 3D). Alternatively, users can specify image dimensions for the whole data set (e.g., 3000 × 4000 pixels). All images will be scaled and cropped, resulting in a set of uniform images, regardless of the original skew or distance from the camera (Figure 3E).
We validated the performance of our FieldSheets and QR codes by photographing a full sheet of QR codes placed in the center of an A3-sized FieldSheet starting at the intended working distance and increasing up to 6 m from the FieldSheet. The intended working distance varies by camera and lens combination but is the distance where the FieldSheet photogrammetric markers nearly fill the frame (Figure 3E). In a university conference room with controlled bright lighting, we taped the FieldSheet to a whiteboard and removed most distracting items. For this test, we used four cameras including a Canon R5 (45 megapixels; mirrorless camera; Canon, Tokyo, Japan), a Panasonic Lumix G9 (20.3 megapixels; mirrorless camera; Panasonic, Osaka, Japan), a Samsung Galaxy S9+ (12 megapixels; smartphone; Panasonic, Suwon, South Korea), and the OAK-1 (12 megapixels; machine vision camera), which is the camera used by our FieldStation system. For this test, all cameras used built-in autofocus. We also printed the QR code onto three different kinds of paper using the same laser toner printer: standard copier paper, Avery 5353 Copier Shipping Labels, and Avery 5526 Waterproof Film Shipping Labels. Our performance metrics were the proportion of successfully decoded QR codes, converted photogrammetric markers, and distortion corrections at each distance.
QR code validation
FieldPrism decoded 99.5% of QR codes in images taken at the intended working distance (less than 1 m from the subject), indicating that a variety of possible equipment configurations will produce good results (Appendix S2A). At 1 m, the high-resolution Canon R5 and Lumix G9 each achieved a similar success rate of 99.1%. However, this success rate drops sharply to only 15.6% at 1 m for the pair of 12-megapixel cameras. Beyond 1 m, the high-resolution cameras achieve some success, but the low-resolution cameras fail. Image resolution and distance from the subject strongly impact the success of QR code decoding.
FieldSheet validation for distortion correction
All camera and paper combinations performed exceptionally well at the intended working distance, with successful distortion correction at 1 m from the FieldSheet (Appendix S2B). Beyond 1 m, the two high-resolution cameras performed best, with correction achieved up to 4 m from the FieldSheet. None of the cameras could correct images beyond 4 m.
With the same data set, we also calculated the proportional error between the 10-cm scale bar printed on the FieldSheet and the 10-cm scale bar predicted by FieldPrism. For this test, we used the default pixel-to-metric conversion settings in the FieldPrism configuration file, which requires all four photogrammetric markers to be located and successfully yield a conversion ratio. Four ratios are averaged per image to produce the predicted 10-cm scale bar. Image distortion correction was successful for the images in this set, so it did not include any images taken more than 4 m away from the FieldSheet (Appendix S2C). Across all combinations of paper, distance, and cameras, we found that on average FieldPrism underestimated the true conversion ratio by 0.97 mm per 10 cm or an error rate of less than 1% (Table 3).
|Variable||Error (mm)||Standard deviation||Sample size|
|Canon R5 (45 MP)||−0.96||0.77||22|
|Lumix G9 (20 MP)||−0.99||0.44||12|
|OAK-1 (12 MP)||−1.02||1.68||13|
|Samsung Galaxy S9+ (12 MP)||−0.91||0.51||12|
|Waterproof package label||−1.06||0.76||20|
|Distance from FieldSheet|
- Note: MP = megapixel.
With the help of early testers at the Morton Arboretum (Lisle, Illinois, USA), we validated a prototype implementation of our FieldSheets and QR code system by collecting images of Quercus havardii Rydb. in Texas (Figure 3, row 2; Appendix S3). The prototype system used a single QR code per specimen and had grayscale photogrammetric markers. Collected specimens were flattened on the FieldSheet using a thin piece of transparent acrylic. This test led to several improvements, including the development of the QR code flipbook, because managing individual barcodes was too cumbersome and limited flexibility while collecting specimens. The grayscale markers also made pixel-to-metric conversion inconsistent and dependent on lighting conditions, prompting the shift to solid black markers. Despite these issues, the FieldPrism software corrected image distortion for 205 of the 208 imaged specimens. The three failures were due to intense sun glare directly over the photogrammetric markers, caused by the transparent acrylic (Appendix S3C). FieldPrism identified the locations of 100% of the barcodes and successfully decoded 70.2%. The decoding failures were caused by inconsistent lighting (e.g., half of the QR code was in full sun while half was in shadow) or three-dimensional distortion of the QR code (e.g., curled paper). For the best results, we recommend printing all material on waterproof paper and attaching QR codes to a notebook to minimize three-dimensional distortion. Controlling lighting can be difficult, but FieldPrism can compensate for poor lighting so long as shadows do not bisect the QR codes.
To validate the accuracy of the FieldStation GPS module and active antenna (Appendix S1), we compared its accuracy to a consumer-grade handheld GPS receiver, the Garmin GPSMAP 66sr (Garmin, Olathe, Kansas, USA), in a lightly wooded, hilly park near the University of Michigan campus. Two people walked 1 m apart on a path through the park for 20 min. FieldStation recorded GPS coordinates every 20 s, and Garmin recorded every 15 s while in motion. Qualitatively, we observed virtually no drift with the Garmin points, but we did observe drifting points of up to 4 m off the path for the FieldStation GPS. Then, we placed both GPS devices in the same location and recorded coordinates concurrently for 15 min at the previously mentioned rates. While stationary, the FieldStation GPS module achieved greater precision than Garmin. The minimum bounding polygon that contained all Garmin coordinates (n = 67) was 6.94 m², while the FieldStation bounding polygon (n = 38) was 1.76 m². The circular error probability (CEP; the radius of a circle within which 50% of the coordinates are expected to fall) was 0.70 m for the Garmin and 0.57 m for the FieldStation. The root mean square accuracy at the 95% confidence level (RMS; a measure of the average deviation between a set of coordinates and their centroid for both horizontal and vertical distances) was 1.70 m for the Garmin and 0.97 m for the FieldStation. The FieldStation active external antenna seemed to aid stationary accuracy, while Garmin's GPS correction algorithms improved performance while in motion. For additional GPS performance and device comparisons, please refer to U.S. Forest Service Missoula Technology and Development Center (MTDC) test reports (USDA Forest Service, 2023).
FieldStation battery life
We tested battery life by instructing different configurations of the FieldStation to capture GPS data and three photos per minute until the battery drained. Battery testing revealed that the OAK camera and the Raspberry Pi should be powered with two separate battery banks. By powering the Raspberry Pi with a 26,000 mAh battery and the OAK-1 camera with a 20,000 mAh battery, we achieved between 17 and 20 h of continuous run time, yielding more than 3000 photos.
To enhance the quality and usefulness of field reference images and increase their potential for downstream applications, the FieldPrism workflow generates curated snapshot vouchers. FieldPrism standardizes images, emulating traditional herbarium specimens, with the goal of enabling accurate morphometric and phenological measurements using other tools like ImageJ or LeafMachine2 (Rueden et al., 2017; Weaver and Smith 2023). FieldPrism methods can greatly improve the utility of field images with minimal effort (Figure 1). For example, the immense trove of observations submitted to the iNaturalist portal is a valuable source of biological data, but citizen science initiatives like iNaturalist could greatly enhance the utility of the collected photographs (Heberling and Isaac, 2018). While there are more than 7 million plant observations in iNaturalist, a scale bar rarely makes an appearance (Seltzer, 2019). Our FieldSheet is a powerful enhancement tool for citizen science initiatives like iNaturalist. The inclusion of a photogrammetric scale bar extends the utility of crowdsourced images beyond the scoring of the presence or absence of traits and could produce a valuable data set of accurately measured traits at an unprecedented scale. FieldPrism is a valuable resource for organizing and standardizing field images, an important first step in collecting data and measurements from images.
W.N.W. planned and designed the project, wrote the software, built the FieldStation hardware, and wrote the initial manuscript. S.A.S. contributed to project planning and design, imaging, and testing. Both authors contributed to revising and editing the text and approved the final version of the manuscript.
The authors thank Dr. Charles Cannon, Samantha Panock, and Claire Henley at the University of Chicago Morton Arboretum for testing and providing feedback on an early prototype of our FieldSheets, QR codes, and FieldStation. The authors also thank Megan Kastelen and Sophie Farr (University of Michigan) for assisting with labeling machine learning training data and the members of the Smith Lab (University of Michigan) for providing insightful feedback and encouragement. Funding was provided by the National Science Foundation (DEB 2217116), Michigan Institute for Data Science, and the Michigan Institute for Computational Discovery and Engineering.
A list of the data sets that provided training images for our machine learning algorithm. This unpublished data set of annotated herbarium specimen vouchers sampled up to 50 images per herbarium to capture a diversity of specimen preparation techniques and image quality, yielding 6394 herbarium vouchers to supplement our 1119 FieldPrism training images. The data set is available at https://doi.org/10.15468/dl.bh9dem.
|Herbarium data set name||Data set DOI|
|University of Tennessee Vascular Herbarium||10.15468/ok8qvz|
|Herbarium of Yugra State University||10.15468/z8mpt5|
|San Diego Natural History Museum||10.15468/lneqwn|
|The Vascular Plant Collection at the Botanische Staatssammlung München||10.15468/vgr4kl|
|University of Florida Herbarium||10.15468/v5wjn7|
|Steiermarkisches Landesmuseum Joanneum||10.15468/dmdck6|
|Canadian Museum of Nature Herbarium||10.15468/kowta4|
|Species recordings from the Danish National portal Arter.dk||10.15468/q3yy4u|
|Texas Tech University||10.15468/uyakmh|
|Forest Herbarium Ibadan||10.15468/uhnd5n|
|University of Graz Institute of Plant Sciences||10.15468/axtkuz|
|Fresno State Herbarium||10.15468/puyrj8|
|UiT Tromsa Museum||10.15468/14epds|
|Charles University Prague||10.15468/8xrt7r|
|University of Michigan Herbarium||10.15468/nl8bvi|
|Rio de Janeiro Botanical Garden Herbarium Collection||10.15468/bbsqoa|
|CRSN herbarim from Kahuzi-Biega National Park||10.15468/bhvwem|
|University of Lethbridge Herbarium||10.5886/wrt547hq|
|San Diego State University Herbarium||10.15468/8sx2ag|
|UTEP Plants (Arctos)||10.15468/yhb6ky|
|TRH, NTNU University Museum||10.15468/zrlqok|
|Flora of the Korean Peninsula||10.15468/0vcvsq|
|Botanical Museum Berlin-Dahlem||10.15468/ed17cn|
|University of Tartu Natural History Museum||10.15468/5hqb2z|
|Museum d'Histoire Naturelle of Aix-en-Provence||10.15468/fqykeb|
|Institut Scientifique Mohamed V University||10.15468/48pwft|
|Rhoen and Vogelsberg||10.15468/hbhfi3|
|George Mason University||10.15468/t8ar55|
|The Exsiccatal Series||10.15468/qxmief|
|CSBG SB RAS Digital Herbarium||10.15468/6f3ybc|
|Instituto do Meio Ambiente do Estado de Alagoas||10.15468/mu8w57|
|Forest Herbarium Ibadan Nigeria||10.15468/rhbyxz|
|Cal State LA Herbarium||10.15468/36qz6p|
|E. C. Smith Herbarium||10.15468/zc4csq|
|Clemson University Herbarium||10.15468/srjd22|
|California Botanic Garden Herbarium||10.15468/0yosx9|
|University of California, Los Angeles Herbarium||10.15468/33k42a|
|Cal Poly State University||10.15468/mypdjd|
|Harvard University Herbarium||10.15468/29fhdy|
|Field Museum of Natural History||10.15468/pyjtoc|
|Carnegie Museum of Natural History Herbarium||10.15468/d51v1f|
|Botanische Staatssammlung München||10.15468/ni5yho|
|Royal Ontario Museum Green Plant Herbarium||10.5886/g7j6gct1|
|San Jose State University||10.15468/t3a60p|
|University of New Mexico Herbarium||10.15468/dlvoyt|
|Artportalen (Swedish Species Observation System)||10.15468/kllkyl|
|Royal Botanic Gardens, Kew||10.15468/ly60bx|
|Norwegian Species Observation Service||10.15468/zjbzel|
|Universidade Federal do Ceara||10.15468/s8xuen|
|National Academy of Sciences of Republic of Armenia||10.15468/xn64eb|
|Humboldt State University||10.15468/qguk7r|
|Tallinn Botanic Garden||10.15468/hfs8d4|
|Naturalis Biodiversity Center||10.15468/ib5ypt|
|Brauckmann at the Botanische Staatssammlung München||10.15468/onfqgb|
|Komarov Botanical Institute||10.15468/udzn9d|
|Genus Medicago in CSBG Herbarium||10.15468/jvrxeh|
|Qarshi Botanical Garden||10.15468/pjxa84|
|Rio de Janeiro Botanical Garden Herbarium||10.15468/7ep9i2|
|Kenai National Wildlife Refuge (Arctos)||10.15468/ycpd7y|
|Tropicos Specimen Data||10.15468/hja69f|
|Herbarium Horti Botanici Pisani||10.15468/soyil7|
|University of Sargodha Herbarium||10.15468/n4k5s9|
|CSBG SB RAS||10.15468/67ouin|
|Jardins botaniques and Conservatoire Botanique of Nancy||10.15468/g1zohr|
|Field Museum of Natural History||10.15468/4nodxs|
|Asociacian Jardan Botanico La Laguna||10.15468/gfwydn|
|National Museum of Natural History Luxembourg||10.15468/s2iu7d|
|Botanical Collections of the Abo Akademi||10.15468/mpsjrk|
|Natural History Museum, Vienna||10.15468/5sl7sh|
|Quaid-i-Azam University Herbarium||10.15468/bp6jy3|
|Universidade Federal de Parana||10.15468/fpf5j6|
|Dr. Sultan Ahmad Herbarium||10.15468/xaju4z|
|Ministerio del Medio Ambiente de Chile||10.15468/ezyu58|
|NSW AVH data||10.15468/jf3yae|
|Arizona State University||10.15468/a2o8vy|
|Moscow University Herbarium||10.15468/cpnhcc|
|The New York Botanical Garden Herbarium||10.15468/6e8nje|
|San Francisco State University||10.15468/6zdzvc|
|University of South Florida Herbarium||10.15468/mdnmzb|
|Institut Botanic de Barcelona||10.15468/pff0t6|
|Universidade Federal do Oeste do Para||10.15468/ztzkde|
|South-Siberian Botanical Garden||10.15468/y6xmme|
|Universidad del Valle de Guatemala||10.15468/u339qt|
|Universidade Federal de Sergipe||10.15468/9xujh5|
|California State University, Long Beach||10.15468/3y25yl|
|University of Tennessee Fungal Herbarium||10.15468/da30il|
|Botanische Staatssammlung München||10.15468/sookye|
|Herbarium Willing at Herbarium Berolinense, Berlin||10.15468/abcz8i|
|University of California, Riverside||10.15468/ai1kou|
|South Australian Museum Australia||10.15468/wz4rrh|
|Central Siberian Botanical Garden Herbarium||10.15468/5wcerp|
|Universidade Tecnolagica Federal do Parana||10.15468/4b74v2|
|Newhaven Sanctuary Observations||10.15468/mwgsdh|
|Georgian Academy of Sciences||10.15468/6tbhmd|
|Arkansas Natural Heritage Commission Herbarium||10.15468/v94jsu|
|Royal Botanic Garden Edinburgh Herbarium||10.15468/ypoair|
|University of Jena, Herbarium Haussknecht||10.15468/8arhjc|
|University of North Carolina at Chapel Hill||10.15468/63vxjd|
|Real Jardan Botanico||10.15468/mug7kr|
|Estonian University of Life Sciences||10.15468/m3x9uu|
|UC Davis Herbarium||10.15468/on4axg|
|Herbier du Quabec||10.5886/jd11sg3p|
|Eastern Kentucky University||10.15468/fy8dsi|
|Sociata des Sciences Naturelles et Mathamatiques de Cherbourg||10.15468/lmznjw|
|Natural History Museum||10.5519/0002965|
|California State University, Northridge||10.15468/nrcdx7|
|The New York Botanical Garden||10.15468/5y84ye|
|CHAS Botany Collection (Arctos)||10.15468/ji4vbl|
|Herbarium of Numto Nature Park||10.15468/g4gcrg|
|Staten Island Museum||10.15468/ctqpb5|
|University of Tartu Natural History Museum and Botanical Garden||10.15468/d59dmk|
|University of Kentucky||10.15468/fi4vfu|
|Coleccian Herbario Federico Medem Bogota||10.15472/ighftu|
|Herbario Joao de Carvalho e Vasconcellos||10.15468/olfpjv|
|B.M. Kozo-Polyansky VSU||10.15468/xyqng3|
|University of Gothenburg||10.15468/asgd85|
|NHMD Vascular Plants Collection||10.15468/4zygkn|
|Naturhistorisches Museum Mainz||10.15468/l0wmu8|
|Universidade Federal do Rio Grande do Norte||10.15468/gtxawd|
|Botanische Staatssammlung München||10.15468/zinzhd|
|Capture of Primary Biodiversity Data for West African Plants||10.15468/9czcig|
|Flora of Sumatra: ANDA Herbarium||10.15468/ue7xyn|
|Universidade Federal da Bahia||10.15468/tbtrr3|
|Universidade Federal de Goias||10.15468/fw6hdt|
|Black Hills State University Herbarium||10.15468/ptcrqx|
|Desert Botanical Garden Herbarium||10.15468/abe1lg|
|NSW South Coast||10.15468/px2xfi|
|Musaum National da Histoire Naturelle, Paris||10.15468/kw8pex|
|Flora Sumatra: (ANDA)-Part 2||10.15468/55evew|
|Nova Scotia Museum of Natural History||10.15468/tl3cde|
|Universita de Montpellier||10.15468/gyvkrn|
|Meise Botanic Garden Herbarium||10.15468/wrthhx|
|Fairchild Tropical Botanic Garden||10.15468/hdpruf|
|R. L. McGregor Herbarium||10.15468/htptzr|
|Institute of Biological Problems of the North, Far East Branch RAS||10.15468/ms9q2t|
|Universidade Federal do Esparito Santo||10.15468/kasze8|
|Central Siberian Botanical Garden||10.15468/qdfdqq|
|Centro de Pesquisas do Cacau||10.15468/vg8rjh|
|University of Vienna, Institute for Botany||10.15468/tnj8wm|
|California State University San Bernardino||10.15468/t885ps|
|Estonian Museum of Natural History||10.15468/bquqpv|
|Flora of the Korean Peninsula||10.15468/fyxnsd|
|University of Balochistan Herbarium||10.15468/qrau0v|
|Royal Botanic Gardens, Kew||10.15468/rvrsru|
|The James C. Parks Herbarium at Millersville University||10.15468/qdatdf|
|Earth Sciences Collection (Arctos)||10.15468/4n2ev3|
|Brown University Herbarium||10.15468/njgg1a|
|Botanische Staatssammlung München||10.15468/zdcclb|
|Bush Heritage - Carnarvon Station Reserve||10.15468/q0dhpr|
|Herbarium Berolinense, Berlin||10.15468/dlwwhz|
|University of Manitoba Herbarium||10.5886/2fva5p4r|
|University of Colorado Museum of Natural History||10.15468/wyofjv|
|Museu Paraense Emalio Goeldi||10.15468/rdq4nx|
|Herbiers Universitaires de Clermont-Ferrand||10.15468/9axq0b|
|Vascular Plant Herbarium, Oslo||10.15468/wtlymk|
|Kathryn Kalmbach Herbarium||10.15468/axrelr|
|Museu Botanico Municipal Curitiba||10.15468/v52pmc|
|North Carolina State University||10.15468/9ufthy|
|Botanische Staatssammlung München||10.15468/lqetda|
|Universidade Estadual do Norte Fluminense||10.15468/qsaagd|
|Herbario Museo de La Salle Bogota||10.15472/ppzpea|
|Universidade Regional de Blumenau||10.15468/vse5f3|
|Appalachian State University||10.15468/ivsxey|
|Clamence Lortet herbarium||10.15468/e64fbk|
|Yale Peabody Museum||10.15468/hrztgn|
|Cape Breton University Collection||10.15468/7dtqgn|
|UAM Herbarium (Arctos)||10.15468/iawody|
|Harvard University Herbaria||10.15468/o3pvnh|
|Colorado Plateau Museum of Arthropod Biodiversity||10.15468/du1hci|
|Coleccian de plantas vasculares del herbario de la Universitat de Valancia||10.15468/xmki52|
|NCSM Herbarium Collection||10.36102/dwc.12|
|Botanischer Garten und Botanisches Museum Berlin-Dahlem Herbarium||10.15468/tgwryf|
|University of Vermont||10.15468/crnsua|
|Universidade de Sao Paulo||10.15468/nt6dng|
|Universidade Estadual de Feira de Santana||10.15468/gsy3jn|
|Natural History Museum Rotterdam||10.15468/kwqaay|
|University of South Carolina||10.15468/fmj4at|
|Northern Arizona University||10.15468/b7tfpa|
|CSBG SB RAS Digital Herbarium||10.15468/7anvyu|
|CSBG SB RAS Herbarium Collections||10.15468/sunx5n|
|McGill University Herbarium||10.5886/srzbj7|
|Universidade Federal de Uberlandia||10.15468/cshs8n|
|Musaum National d'Histoire Naturelle||10.15468/nc6rxy|
|Universidade Federal do Rio Grande Do Sul||10.15468/suhqjx|
|Lord Fairfax Community College Herbarium||10.15468/c2gj2t|
|University of California Santa Barbara Herbarium||10.15468/qpxmw0|
|Vascular Plant Herbarium, UiB||10.15468/ofn0lf|
|Botanische Staatssammlung München||10.15468/dixlft|
|Field Museum of Natural History||10.15468/nxnqzf|
|University of Cincinnati||10.15468/bhgpmq|
|Universidade Federal Rural do Rio de Janeiro||10.15468/0svt7m|
|Santa Barbara Botanic Garden||10.15468/adb2bb|
|Auckland Museum Botany Collection||10.15468/mnjkvv|
|University of California Santa Cruz||10.15468/uavt0t|
|CSBG SB RAS||10.15468/be6owh|
|Central Michigan University Herbarium||10.15468/iykbez|
|University of Hargeisa Herbarium||10.15468/qvbvdp|
|Yale Peabody Museum||10.15468/0lkr3w|
|University of Tennessee||10.15468/64w2b1|
|Universidade Federal de Rondania||10.15468/5cjyj6|
|University of Cincinnati||10.15468/xkca3p|
|Herbarium of the University of Granada||10.15470/k97bjm|
|Missouri Botanical Garden||10.15468/mmbcpb|
|California State University Fullerton||10.15468/1uvzxh|
|Universidade Estadual do Oeste do Parana||10.15468/eqp1dr|
|Herbarium of Andalas University||10.15468/sncpxn|
|Chico State Herbarium||10.15468/ckxw7v|
|Western Carolina University Herbarium||10.15468/sk26v2|
|Museu de Biologia Mello Leitao||10.15468/dmkg7b|
|Plant Resources Center||10.15468/g85t8z|
|University of Vermont||10.15468/zsgiog|
|University of British Columbia Herbarium||10.5886/rtt57cc9|
|CSBG SB RAS||10.15468/zw7jnn|
Open Research Badges
DATA AVAILABILITY STATEMENT
The FieldPrism source code, example data sets, FieldSheets, FieldStation build guide, and user manual are available at https://github.com/Gene-Weaver/FieldPrism and https://www.FieldPrism.org. Video tutorials for using the FieldStation are available at https://www.FieldPrism.org.
|aps311545-sup-0001-APPS_D_23_00011_SupportingInformation_AppS1.docx26.7 KB||Appendix S1. FieldStation build guide and alternatives.|
|aps311545-sup-0002-APPS_D_23_00011_SupportingInformation_AppS2.docx441 KB||Appendix S2. Validating QR code and FieldSheet performance at increasing distances.|
|aps311545-sup-0003-APPS_D_23_00011_SupportingInformation_AppS3.docx712.4 KB||Appendix S3. Three example images from our field test processed with FieldPrism.|
Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.
- 2009. EpiCollect: Linking smartphones to web applications for epidemiology, ecology and community data collection. PLoS ONE 4(9): e6968. https://doi.org/10.1371/journal.pone.0006968
- 2017. Software for improved field surveys of nesting marine turtles. Scientific Reports 7: 10796. https://doi.org/10.1038/s41598-017-11245-6
- 2017. Herbarium genomics: Skimming and plastomics from archival specimens. Webbia 72(1): 35–45. https://doi.org/10.1080/00837792.2017.1313383
- 2008. Digital plant images as specimens: Toward standards for photographing living plants. Vulpia 7: 16–30.
- 2013. Herbarium specimens reveal the footprint of climate change on flowering trends across north-central North America. Ecology Letters 16: 1037–1044. https://doi.org/10.1111/ele.12135
- 2013. Utility of QR codes in biological collections. PhytoKeys 25: 21–34. https://doi.org/10.3897/phytokeys.25.5175
- 2014. Easy Leaf Area: Automated digital image analysis for rapid and accurate measurement of leaf area. Applications in Plant Sciences 2(7): 1400033. https://doi.org/10.3732/apps.1400033
- 2010. Phylogenetic analyses reveal the shady history of C4 grasses. Proceedings of the National Academy of Sciences, USA 107(6): 2532–2537. https://doi.org/10.1073/pnas.0909672107
- 2018. On the use of herbarium specimens for morphological and anatomical research. Botany Letters 165(3–4): 361–367. https://doi.org/10.1080/23818107.2018.1451775
- Esri. 2019. Field data collection app for mobile workers ArcGIS Field Maps. Website: https://www.esri.com/en-us/arcgis/products/arcgis-field-maps/overview [accessed 25 August 2023].
- 2021. High-throughput methods for efficiently building massive phylogenies from natural history collections. Applications in Plant Sciences 9(2): e11410. https://doi.org/10.1002/aps3.11410
- 2016. The application of quick response (QR) codes in archaeology: A case study at Telperion Shelter, South Africa. Antiquity 90(353): 1363–1372. https://doi.org/10.15184/aqy.2016.135
- 2017. Guidelines for collecting vouchers and tissues intended for genomic work (Smithsonian Institution): Botany Best Practices. Biodiversity Data Journal 5: e11625. https://doi.org/10.3897/BDJ.5.e11625
- 2019. How photographs can be a complement of herbarium vouchers: A proposal of standardization. TAXON 68(6): 1321–1326. https://doi.org/10.1002/tax.12162
- 2016. Realising the potential of herbarium records for conservation biology. South African Journal of Botany 105: 317–323. https://doi.org/10.1016/j.sajb.2016.03.017
- 2022. Digital extended specimens: Enabling an extensible network of biodiversity data records as integrated digital objects on the internet. BioScience 72: 978–987. https://doi.org/10.1093/biosci/biac060
- 2018. iNaturalist as a tool to expand the research value of museum specimens. Applications in Plant Sciences 6(11): e01193. https://doi.org/10.1002/aps3.1193
- 2022. YOLOv5 Realtime Instance Segmentation [posted 22 November 2022]. Available at Zenodo repository: https://doi.org/10.5281/zenodo.7347926 [accessed 23 August 2023].
- 2018. Herbarium specimens can reveal impacts of climate change on plant phenology; a review of methods and applications. PeerJ 6: e4576. https://doi.org/10.7717/peerj.4576
- 2018. The use and misuse of herbarium specimens in evaluating plant extinction risks. Philosophical Transactions of the Royal Society B: Biological Sciences 374(1763): 20170402. https://doi.org/10.1098/rstb.2017.0402
- 2016. ColectoR, a digital field notebook for voucher specimen collection for smartphones. Applications in Plant Sciences 4(7): 1600035. https://doi.org/10.3732/apps.1600035
- 2018. The unrealized potential of herbaria for global change biology. Ecological Monographs 88(4): 505–525. https://doi.org/10.1002/ecm.1307
- 2020. Spatial phylogenetics of the North American flora. Journal of Systematics and Evolution 58(4): 393–405. https://doi.org/10.1111/jse.12590
- 2020. Mobile GIS applications for environmental field surveys: A state of the art. Global Ecology and Conservation 23: e01089. https://doi.org/10.1016/j.gecco.2020.e01089
- 2017. Assessing the relevance of herbarium collections as tools for conservation biology. The Botanical Review 83(3): 303–325. https://doi.org/10.1007/s12229-017-9188-z
- 2019. A born-digital field-to-database solution for collections-based research using collNotes and collBook. Applications in Plant Sciences 7(8): e11284. https://doi.org/10.1002/aps3.11284
- 2021. Reversing extinction trends: New uses of (old) herbarium specimens to accelerate conservation action on threatened species. New Phytologist 230(2): 433–450. https://doi.org/10.1111/nph.17133
- 2017. ImageJ2: ImageJ for the next generation of scientific image data. BMC Bioinformatics 18(1): 529. https://doi.org/10.1186/s12859-017-1934-z
- 1995. The use of herbarium specimens in DNA phylogenetics: Evaluation and improvement. Plant Systematics and Evolution 197(1): 87–98. https://doi.org/10.1007/BF00984634
- 2017. Leaf-IT: An Android application for measuring leaf area. Ecology and Evolution 7(22): 9731–9738. https://doi.org/10.1002/ece3.3485
- 2019. Making biodiversity data social, shareable, and scalable: Reflections on iNaturalist & citizen science. Biodiversity Information Science and Standards 3: e10197. https://doi.org/10.3897/biss.3.46670
- 2013. Use of QR code labels in museum collection management. Collections 9(3): 239–254. https://doi.org/10.1177/155019061300900302
- 1975. The importance of revisionary studies in plant systematics. SIDA, Contributions to Botany 6(2): 104–113.
- USDA Forest Service. 2023. MTDC test reports. Website: https://www.fs.usda.gov/database/gps/mtdcrept/index.htm [accessed 25 August 2023].
- 2023. From leaves to labels: Building modular machine learning networks for rapid herbarium specimen analysis with LeafMachine2. Applications in Plant Sciences 11(5): e11548.
- 2020. LeafMachine: Using machine learning to automate leaf trait extraction from digitized herbarium specimens. Applications in Plant Sciences 8(6): e11367. https://doi.org/10.1002/aps3.11367
- 2015. Collections-based systematics: Opportunities and outlook for 2050. Journal of Systematics and Evolution 53(6): 477–488. https://doi.org/10.1111/jse.12181
- 2017a. Old plants, new tricks: Phenological research using herbarium specimens. Trends in Ecology & Evolution 32(7): 531–546. https://doi.org/10.1016/j.tree.2017.03.015
- 2017b. CrowdCurio: An online crowdsourcing platform to facilitate climate change studies using herbarium specimens. New Phytologist 215(1): 479–488. https://doi.org/10.1111/nph.14535