Applications of deep convolutional neural networks to predict length, circumference, and weight from mostly dewatered images of fish

Nicholas Bravata; Dylan Kelly; Jesse Eickholt; Janine Bryan; Scott M. Miehls; Daniel Zielinski

doi:10.1002/ece3.6618

Home/Archive/USGS

USGS · 70216944

Applications of deep convolutional neural networks to predict length, circumference, and weight from mostly dewatered images of fish

Nicholas Bravata·Dylan Kelly·Jesse Eickholt·Janine Bryan·Scott M. Miehls·Daniel Zielinski

View original source ↗Export metadata ↓

Abstract

Simple biometric data of fish aid fishery management tasks such as monitoring the structure of fish populations and regulating recreational harvest. While these data are foundational to fishery research and management, the collection of length and weight data through physical handling of the fish is challenging as it is time consuming for personnel and can be stressful for the fish. Recent advances in imaging technology and machine learning now offer alternatives for capturing biometric data. To investigate the potential of deep convolutional neural networks to predict biometric data, several regressors were trained and evaluated on data stemming from the FishL™ Recognition System and manual measurements of length, girth, and weight. The dataset consisted of 694 fish from 22 different species common to Laurentian Great Lakes. Even with such a diverse dataset and variety of presentations by the fish, the regressors proved to be robust and achieved competitive mean percent errors in the range of 5.5 to 7.6% for length and girth on an evaluation dataset. Potential applications of this work could increase the efficiency and accuracy of routine survey work by fishery professionals and provide a means for longer‐term automated collection of fish biometric data.

Explore related subjects

Ecology and Evolution Article

Source-reported bounding extent: 40.588928169693745° to 46.210249600187225° latitude; -88.428955078125° to -82.529296875° longitude. This indicates report coverage, not an exact sampling location. View area on OpenStreetMap.

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Nicholas Bravata, Dylan Kelly, Jesse Eickholt, Janine Bryan, Scott M. Miehls, Daniel Zielinski. 2020-08-04. Applications of deep convolutional neural networks to predict length, circumference, and weight from mostly dewatered images of fish. https://doi.org/10.1002/ece3.6618

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

KEEP EXPLORING

Related USGS reports

USGS2026

A waterfowl reproduction model to predict carryover effects of oil spills

For wintering and migrating birds, contamination from oil and other substances may delay arrival on the breeding grounds and lead to lower body mass upon arrival. These delays and losses of body mass could lead to smaller clutch sizes and lower propensity to renest. To determine potential effects of these consequences on breeding, we used an augmented version of an existing mallard ( Anas platyrhynchos ) reproduction model. By combining that reproduction model with the outputs of a previously developed migration and energetics model, we assessed carryover effects of simulated wintering ground oil spills on duckling production and other breeding season-related population parameters. For simulated oil spills that occurred in January, reduced adult population and sublethal effects of migration delays both contributed to decreases in duckling production. For a simulated oil spill that occurred in March, sublethal effects from reduced body mass were the primary driver of decreased duckling production. By combining migration and energetics modeling with a reproduction model, this framework could provide the basis for a tool to generate quantitative estimates of lost reproduction resulting from carryover effects. Furthermore, this modeling framework could also be adapted for other avian species, facilitating assessments of reproductive loss in the context of Natural Resource Damage Assessments following contamination events.

Ecology and Evolution↗

USGS2026

The further you are, the closer you get: Environmental similarity connects aquatic invertebrate communities across the terrestrial landscape in a large river network

Biodiversity is structured on the landscape through a combination of spatial, environmental, and stochastic factors. In freshwater river networks, the interaction of the linear river network and the three-dimensional terrestrial landscape creates multiple pathways for ecological connectivity that affect aquatic community structure through habitat filtering and dispersal dynamics. Here, we use the high throughput capacity of environmental DNA (eDNA) metabarcoding to explore the influence of spatial relationships and environmental heterogeneity across an expansive and diverse landscape. We collected invertebrate eDNA samples from 70 aquatic habitats in the lower Colorado River Basin (LCRB), ranging from small, free-flowing headwater tributaries to anthropogenically-altered high order rivers, spanning five major drainage regions. Communities displayed high alpha and beta diversity among the drainage regions, structured by a combination of spatial, environmental, and stochastic factors. A novel distance-decay framework adapted from population genetics revealed the influence of terrestrial pathways of ecological connectivity among communities throughout the network. Hierarchical clustering identified patterns of convergent similarity in headwater tributaries despite extreme river network distances separating sites. These results reveal the dual influence of spatial and environmental metacommunity dynamics in aquatic invertebrate communities in the LCRB and the specific influence of interbasin terrestrial connectivity across large spatial scales. Overall, this highlights how terrestrial landscapes can shape convergent patterns of biodiversity in river networks.

Arizona, California, Nevada↗

USGS2026

Conflicts and collisions with an endangered carnivore: Landscape drivers and spatial risk pattern

Human–wildlife conflicts and anthropogenic mortality are expected to increase with the continuous encroachment on natural habitats within and around protected areas, yet the processes governing these risks remain poorly understood. We investigated the spatiotemporal dynamics and drivers of two distinct processes involving the endangered Florida panther ( Puma concolor coryi ; hereafter, panther): (1) direct human–panther interactions and livestock depredation (hereafter, conflicts), and (2) vehicle collisions (hereafter, collisions). Using long-term data collected between 2006 and 2022 within the panther's breeding range in Southwest Florida, USA, we applied dynamic occupancy models to estimate probabilities of occurrence, colonization, and extinction of conflicts and collisions while accounting for imperfect detection. A total of 277 conflicts and 239 collisions were recorded during the study period. Occurrence probabilities for both conflicts and collisions increased during the early years of the study and subsequently stabilized. Landscape configuration consistently showed strong associations with both processes. Conflicts were more likely in areas with fragmented but closely spaced habitat patches, whereas collisions were more likely along road segments characterized by connected landscapes with low resistance to panther movement. In contrast, areas that showed greater distances between habitat patches and increased proportions of protected area were associated with lower probabilities of conflict or collision persistence over time. A predictive map was created to highlight areas with high probability of conflict and road segments with high probability of collision in Southwest Florida, USA. Identifying areas with elevated risk can inform targeted mitigation strategies, including habitat management, wildlife crossings, and outreach efforts, and further support long-term coexistence between humans and large carnivore populations.

Florida↗