A pilot computer-vision model trained to identify slide-mounted scale insect pests from the family Rhizoecidae

Conerly, A.; Rosen, B.; Schneider, S. A.

doi:10.64898/2026.10.16.759574

Home/Archive/bioRxiv

bioRxiv · 10.64898/2026.10.16.759574

A pilot computer-vision model trained to identify slide-mounted scale insect pests from the family Rhizoecidae

Conerly, A.·Rosen, B.·Schneider, S. A.

View original source ↗Export metadata ↓

Abstract

The emerging field of image classification using computer vision has yet to address a rather challenging group of insect pests--soft-bodied insects that are preserved on microscope slides. This includes many notoriously invasive herbivores that have a significant impact on agriculture, such as members of the hemipteran suborder Sternorrhyncha. Using root mealybugs (Hemiptera: Rhizoecidae) as a test case, we gathered high-resolution extended depth of field images from museum specimens and trained a convolutional neural network (CNN) to identify and distinguish among 16 root mealybug species that are encountered as agricultural pests in plant quarantine or greenhouses. A pretrained EfficientNetV2-M model was trained using BioEncoder. Our dataset contained 1,607 images, with 41 to 239 images per category. Performance was improved through the adoption of augmentation techniques. Upon completion of training, the model achieved macro precision, recall, and F1 scores of 80.1%, 79.3%, and 79.1%, respectively. The model demonstrated classification metrics >= 80% for seven species. Gradient-weighted Class Activation Mapping (Grad-CAM) was used to visualize the morphological features associated with the test-image classification. Grad-CAM images highlighted meaningful diagnostic features to distinguish among species rather than focusing on uninformative visual artifacts or specimen damage, demonstrating the models utility for identification purposes. The results suggest that further development of the model should incorporate at least 200 training images per species with a minimum resolution of 600x600 pixels. The model was designed to be adaptable; images can be added to improve its performance, and more species can be incorporated as specimens become available, expanding on the taxonomic breadth of its application. Computer-vision identification tools have the potential to enhance rapid response capabilities in plant quarantine at ports-of-entry in the United States and globally.

Explore related subjects

zoology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Conerly, A., Rosen, B., Schneider, S. A.. 2026-10-22. A pilot computer-vision model trained to identify slide-mounted scale insect pests from the family Rhizoecidae. https://doi.org/10.64898/2026.10.16.759574

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Sea slugs unplugged: Review of subpopulation sampling in phylogeographic studies of nudibranchs and their allies

Nudibranchs and other sea slugs include many cryptic species groups with unknown phenotypic and ecological diversity. How broadly and deeply should researchers sample those cryptic species pairs to capture the genetic variation within each species and potential adaptive differences between species? Fast and inexpensive surveys of genetic diversity that sample a single well-known locus across the geographical range ("phylogeography unplugged") have long been used to identify structured populations and guide choice of sampling localities for subsequent comparative genomic or phenotypic studies of the same species. Results from such studies might also be used to better understand the expected relationships between sampling and population structure in future comparative studies in other species (including many marine gastropods) for which an unplugged phylogeographic survey is not available or impractical. We reviewed analyses of molecular variance that used mitochondrial barcode sequences from marine sea slugs, and we asked whether there are characteristic patterns of subpopulation sampling that are associated with the magnitude and geographical scale of differentiation. We found wide variation in sampling designs: studies with both many and few samples discovered significant population differentiation or broadly similar distribution of haplotypes on both large and small spatial scales, in species from different higher taxonomic groups and oceans. We found two significant associations between sampling variables and population genetic variables, and we suggest using those associations to help guide (or understand the consequences of) sampling design in future studies.

zoology↗

bioRxiv2026

A novel methodology for assessing reinforcement choice in primates under human care

Positive Reinforcement Training is a technique commonly employed in zoological collections to elicit a desired behaviour from an animal under human care. Typically, a food item is offered as a reward for performing that behaviour. Three non-human primate species (Mandrillus sphinx, Pan troglodytes, and Hylobates lar) were offered four different foods (banana, grape, apple, and sweet potato) in a series of choice experiments. Latent variable models were employed to investigate which foods were the best selection to use as reinforcement items. Results show that food preference varies between the three primate species. Pan troglodytes and H. lar showed strong preferences for grape and banana, whereas M. sphinx showed a very strong preference for sweet potato. This study indicates that foraging ecology might play an important part in fruit selection and preference in primates. The findings highlight that understanding a species' natural behaviour and dietary preferences is key to informing reinforcement and training decisions.

zoology↗

bioRxiv2026

A comprehensive identification protocol for postcranial bones of Southeast Asian Ruminantia. Part 2: Morphological criteria for small and medium-sized species

The taxonomic identification of postcranial elements from small and medium-sized ruminants represents a significant challenge in zooarchaeological and Late Quaternary paleontological studies, particularly in biodiversity-rich regions such as continental Southeast Asia. This study presents a comprehensive osteological reference framework for the small and medium-sized Ruminantia of this region, based on a comparative sample of 228 specimens representing 14 species across three families (Bovidae, Cervidae, and Moschidae). Osteological characters were first described on a small initial specimen set and subsequently scored across the full sample. Although a small number of character states were found to be unique to specific taxa, most proved highly variable, limiting their individual identification potential. To address this, a Random Forest and a Naive Bayes classifier were applied to assess the discriminatory power of character combinations across four taxonomic ranks (Family, Subfamily, Genus, and Species) on independent bones. Both approaches performed similarly, achieving mean identification accuracies of 87-86% at the family level, 69% at the subfamily level, 50% at the genus level, and 35-33% at the species level. The Naive Bayes models were further deployed as an interactive web application, allowing users to perform bone identification based on any combination of observable character states from a single anatomical element. Our results highlight the limits of morphological identification of postcranial elements in this taxonomic and biogeographic context, while demonstrating that a transparent, reproducible, and statistically grounded approach can still represent a significant step forward for zooarchaeological and Late Quaternary paleontological studies in Southeast Asia and beyond.

zoology↗