Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears

Clapham, Melanie; Miller, Ed; Nguyen, Mary; Darimont, Chris T.

doi:10.1002/ece3.6840

Cited by 60 publications

(65 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As this knowledge is crucial for targeted management and conservation practices (Behr et al 2017), we advocate for the use of detection dogs and scat analysis or any other method that allows reliably identifying single individuals. Some very recent developments of facial recognition algorithms may soon facilitate individual recognition based on camera trapping also for species that lack unique pelage markings (Clapham et al 2020).…”

Section: Discussionmentioning

confidence: 99%

Eyes, ears, or nose? Comparison of three non-invasive methods to survey wolf recolonisation

Cozzi

Hollerbach

Suter³

et al. 2021

Mamm Biol

View full text Add to dashboard Cite

The development and use of cost-effective and appropriate survey methods to assess species distribution and to monitor range expansion and contraction of wild populations is crucial due to the limited financial resources for conservation. Of particular importance, yet little studied, is the ability to collect information before a wild population is well established, i.e. at the early stages of recolonisation. During 2018 and 2019, we used camera traps, audio recorders, and scat detection dogs simultaneously to investigate composition, detection probability, and territorial extent of a pack of wolves in the Swiss Alps. We compared the efficacy of these survey methods by assessing sampling effort, data obtained, and costs. We show that, under the presented setup, camera traps and scat detection dogs substantially outperformed audio recorders in detecting wolves, representing the packs’ territorial extent, and revealing the number of adult wolves. The detection dogs did not detect pups but, unlike the other methods, allowed the identification of single individuals. The use of four camera traps during 13 weeks, a 24-km-long transect walked with the detection dog, or the use of one audio recorder during 148 weeks were necessary to obtain a comparable wolf detection probability. Our results show that no single method was able to return all information that we hoped to collect. Comprehensive and cost-effective information was best obtained by combining data from camera traps and detection dogs. We suggest both methods to be simultaneously used to successfully investigate wolf recolonisation into historical range.

show abstract

Section: Discussionmentioning

confidence: 99%

Eyes, ears, or nose? Comparison of three non-invasive methods to survey wolf recolonisation

Cozzi

Hollerbach

Suter³

et al. 2021

Mamm Biol

View full text Add to dashboard Cite

show abstract

“…Recent studies also apply approaches such as posture identification 38 , 40 to incorporate additional information. Moreover, alignment points (landmarks) are frequently used 43 , 45 , 48 , 54 , 55 to adjust, orientate, and standardize images regarding their final alignment to receive homogeneous data samples and consequently counteract the scale and rotation invariance of CNNs. In case of killer whale individual identification, such concepts are not relevant.…”

Section: Discussionmentioning

confidence: 99%

“…Data distribution is also very important next to the mentioned data complexity. Most of the research approaches did not have uniformly distributed image data for each individual 42 , 48 , 55 , 61 , which means that some animals are observed significantly more often than others, leading to the aforementioned long-tailed distribution. Exactly the same long-tailed phenomenon can be observed in our case (see Fig.…”

Section: Discussionmentioning

confidence: 99%

“…Most recently, machine (deep) learning algorithms have been setting new standards for image processing/analysis across various research areas and fields of application 25 – 33 , due to increasing memory space and performance of central processing units (CPU) and graphics processing units (GPU) 34 – 37 . Among many other image processing problems handled by deep learning, deep neural networks have recently also been applied to the detection and classification of individual animals of several species including amur tigers ( Panthera tigris altaica ) 38 – 41 , elephants ( Proboscidea ) 42 , right whales ( Eubalaena ) 43 , 44 , humpback whales ( Megaptera novaeangliae ) 45 – 47 , brown bears ( Ursus arctos ) 48 , giraffes ( Giraffa camelopardalis ) 49 , pigs ( Sus scrofa domesticus ) 50 , manta rays ( Mobula birostris ) 51 , common dolphins ( Delphinus delphis ) 52 , chimpanzees ( Pan troglodytes verus ) 53 , red pandas ( Ailurus fulgens ) 54 , giant pandas ( Ailuropoda melanoleuca ) 55 , birds (e.g. sociable weaver ( Philetairus socius ), great tit ( Parus major ), zebra finch ( Taeniopygia guttata )) 56 , gorillas ( Gorilla ) 57 , primates (e.g.…”

Section: Introductionmentioning

confidence: 99%

“…Others address only parts of an individual identification pipeline, such as only detection 60 or classification 59 , 68 , 69 . Some approaches present a combination of modern deep learning techniques together with traditional machine learning algorithms 39 , 42 , 48 – 51 , 57 , 69 . FIN-PRINT provides a modular, transferable, and state-of-the-art identification pipeline for killer whales, exclusively applying well-established deep learning concepts, to facilitate robust and task-specific feature learning at each stage.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

FIN-PRINT a fully-automated multi-stage deep-learning-based framework for the individual recognition of killer whales

Bergler

Gebhard

Towers

et al. 2021

Sci Rep

View full text Add to dashboard Cite

Biometric identification techniques such as photo-identification require an array of unique natural markings to identify individuals. From 1975 to present, Bigg’s killer whales have been photo-identified along the west coast of North America, resulting in one of the largest and longest-running cetacean photo-identification datasets. However, data maintenance and analysis are extremely time and resource consuming. This study transfers the procedure of killer whale image identification into a fully automated, multi-stage, deep learning framework, entitled FIN-PRINT. It is composed of multiple sequentially ordered sub-components. FIN-PRINT is trained and evaluated on a dataset collected over an 8-year period (2011–2018) in the coastal waters off western North America, including 121,000 human-annotated identification images of Bigg’s killer whales. At first, object detection is performed to identify unique killer whale markings, resulting in 94.4% recall, 94.1% precision, and 93.4% mean-average-precision (mAP). Second, all previously identified natural killer whale markings are extracted. The third step introduces a data enhancement mechanism by filtering between valid and invalid markings from previous processing levels, achieving 92.8% recall, 97.5%, precision, and 95.2% accuracy. The fourth and final step involves multi-class individual recognition. When evaluated on the network test set, it achieved an accuracy of 92.5% with 97.2% top-3 unweighted accuracy (TUA) for the 100 most commonly photo-identified killer whales. Additionally, the method achieved an accuracy of 84.5% and a TUA of 92.9% when applied to the entire 2018 image collection of the 100 most common killer whales. The source code of FIN-PRINT can be adapted to other species and will be publicly available.

show abstract

Integrating video and genetic data to estimate annual age‐structured apparent survival of American black bears

Reynolds‐Hogland¹,

Ramsey²,

Muench³

et al. 2022

Population Ecology

View full text Add to dashboard Cite

Camera or genetic data are increasingly used to estimate wildlife abundance and density. We integrated video camera data with genetic data over 7 years to estimate annual age‐structured apparent survival of American black bears (Ursus americanus). We identified 70 individuals through meticulous scrutiny of 7531 video captures, cross‐referenced with 721 genetic captures from hair samples concurrently collected from stations in view of cameras. We used the Cormack–Jolly–Seber model in Program Mark to estimate annual age‐structured apparent survival for yearling males, yearling females, 2+ year‐old males, and 2+ year‐old females. We manually calculated cub survival. We compared parameter estimates based on combined video and genetic data with those based on only genetic data. Combining video and genetic data provided a means to test video‐based identification accuracy, which was highest for females (97%–100%). Annual apparent survival was highest for yearling females (φ = 0.92, SE = 0.07), followed by 2+ year‐old females (φ = 0.88, SE = 0.05), 2+ year‐old males (φ = 0.84, SE = 0.06), and yearling males (φ = 0.80, SE = 0.14). Annual cub survival (φ = 0.86, SE = 0.07) was likely biased because we could not account for mortality that occurred in‐den through early spring. Annual apparent survival and recapture probabilities derived from only genetic data were lower than those derived from combined video and genetic data. Our finding that noninvasive data can be used to estimate annual age‐structured apparent survival of a species with relatively indistinct traits is broadly relevant to wildlife research and conservation.

show abstract

Automated facial recognition for wildlife that lack unique markings: A deep learning approach for brown bears

Abstract: This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

Cited by 60 publications

References 36 publications

Eyes, ears, or nose? Comparison of three non-invasive methods to survey wolf recolonisation

Eyes, ears, or nose? Comparison of three non-invasive methods to survey wolf recolonisation

FIN-PRINT a fully-automated multi-stage deep-learning-based framework for the individual recognition of killer whales

Integrating video and genetic data to estimate annual age‐structured apparent survival of American black bears

Contact Info

Product

Resources

About