Analysis of Coral using Deep Learning

Tilo Burghardt, Ainsley Rutterford, Leonardo. Bertini, Erica J. Hendy, Kenneth Johnson, Rebecca Summerfield

Animal biometric systems can also be applied to the remains of living beings – such as coral skeletons. These fascinating structures can be scanned via 3D tomography and made available to computer vision scientists via resulting image stacks.

In this project we investigated the efficacy of deep learning architectures such as U-Net on such image stacks in order to find and measure important features in coral skeletons automatically. One such feature is constituted by growth bands of the colony, which are extracted/approximated by our system and superimposed on a coral slice in the image below. The project provides a first proof-of-concept that machines can, given sufficiently clear samples, perform similarly to humans in many respects when identifying associated growth and calcification rates exposed from skeletal density-banding. This is a first step towards automating banding measurements and related analysis.

Coral skeletal density-banding extracted via Deep Learning

This work was supported by NERC GW4+ Doctoral Training Partnership and is part of 4D-REEF, a Marie Sklodowska-Curie Innovative Training Network funded by European Union Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 813360.

Code Repository at https://github.com/ainsleyrutterford/deep-learning-coral-analysis

Aerial Animal Biometrics

Tilo Burghardt, Will Andrew, Colin Greatwood

Traditionally animal biometric systems represent and detect the phenotypic appearance of species, individuals, behaviors, and morphological traits via passive camera settings – be this camera traps or other fixed camera installations.

In this line of work we implemented for the first time a full biometric pipeline onboard a autonomous UAV in order to gain complete autonomous agency that can be used to adjust acquisition scenarios to important settings such as individual identification in freely moving herds of cattle.

In particular, we have built a computationally-enhanced M100 UAV platform with an on-board deep learning inference system for integrated computer vision and navigation able to autonomously find and visually identify by coat pattern individual HolsteinFriesian cattle in freely moving herds.We evaluate the performance of components offline, and also online via real-world field tests of autonomous low-altitude flight in a farm environment. The proof-of-concept system is a successful step towards autonomous biometric identification of individual animals from the air in open pasture environments and in-side farms for tagless AI support in farming and ecology.

This work was conducted in collaboration with Farscope CDT, VILab and BVS.

Related Publications

W Andrew, C Greatwood, T Burghardt. Aerial Animal Biometrics: Individual Friesian Cattle Recovery and Visual Identification via an Autonomous UAV with Onboard Deep Inference. 32nd IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 237-243, November 2019. (DOI:10.1109/IROS40897.2019.8968555), (Arxiv PDF)
W Andrew, C Greatwood, T Burghardt. Deep Learning for Exploration and Recovery of Uncharted and Dynamic Targets from UAV-like Vision. 31st IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1124-1131, October 2018. (DOI:10.1109/IROS.2018.8593751), (IEEE Version), (Dataset GTRF2018), (Video Summary)
W Andrew, C Greatwood, T Burghardt. Visual Localisation and Individual Identification of Holstein Friesian Cattle via Deep Learning. Visual Wildlife Monitoring (VWM) Workshop at IEEE International Conference of Computer Vision (ICCVW), pp. 2850-2859, October 2017. (DOI:10.1109/ICCVW.2017.336), (Dataset FriesianCattle2017), (Dataset AerialCattle2017), (CVF Version)
HS Kuehl, T Burghardt. Animal Biometrics: Quantifying and Detecting Phenotypic Appearance. Trends in Ecology and Evolution, Vol 28, No 7, pp. 432-441, July 2013.
(DOI:10.1016/j.tree.2013.02.013)

Great Ape Detection and Behaviour Recognition

Tilo Burghardt, X Yang, F Sakib, M Mirmehdi

The problem of visually identifying the presence and locations of animal species filmed in natural habitats is of central importance for automating the interpretation of large-scale camera trap imagery. This is particularly challenging in scenarios where lighting is difficult, backgrounds are non-static, and major occlusions, image noise, as well as animal camouflage effects occur: filming great apes viacamera traps in jungle environments constitutes one such setting. Finding animals under these conditions and classifying their behaviours are important tasks in order to exploit the filmed material for conservation or biological modelling.

Together with researchers from various institutions including the Max Planck Institute for Evolutionary Anthropology we developed deep learning systems for detecting great apes in challenging imagery in the first place and for identifying animal behaviours exhibited in these camera trap clips once apes have been recognised.


Related Publications

F Sakib, T Burghardt. Visual Recognition of Great Ape Behaviours in the Wild. In press. Proc. 25th International Conference on Pattern Recognition (ICPR) Workshop on Visual Observation and Analysis of Vertebrate And Insect Behavior (VAIB), January 2021. (Arxiv PDF)
X Yang, M Mirmehdi, T Burghardt. Great Ape Detection in Challenging Jungle Camera Trap Footage via Attention-Based Spatial and Temporal Feature Blending. Computer Vision for Wildlife Conservation (CVWC) Workshop at IEEE International Conference of Computer Vision (ICCVW), pp. 255-262, October 2019. (DOI:10.1109/ICCVW.2019.00034), (CVF Version), (Arxiv PDF), (Dataset PanAfrican2019 Video), (Dataset PanAfrican2019 Annotations and Code)

Fin Identification of Great White Sharks

Tilo Burghardt, Ben Hughes

Recognising individuals repeatedly over time is a basic requirement for field-based ecology and related marine sciences. In scenarios where photographic capture is feasible and animals are visually unique, biometric computer vision offers a non-invasive identification paradigm.

In this line of work we developed the first fully automated biometric ID system for individual animals based on visual body contours. We applied the techniques to great white shark identification. The work was selected as one of the top 10 BMVC’15 papers and subsequently published in IJCV. The work was collaborative with NGO SaveOurSeas Foundation (SoSF) who employed Ben Hughes to extend and apply this work. The system is now being exploited at large scale by SoSF.

Related Publications

B Hughes, T Burghardt. Automated Visual Fin Identification of Individual Great White Sharks. International Journal of Computer Vision (IJCV), Vol 122, No 3, pp. 542-557, May 2017. (DOI:10.1007/s11263-016-0961-y), (Dataset FinsScholl2456)
B Hughes, T Burghardt. Automated Identification of Individual Great White Sharks from Unrestricted Fin Imagery. 26th British Machine Vision Conference (BMVC), pp. 92.1-92.14, ISBN 1-901725-53-7, BMVA Press, September 2015. (DOI:10.5244/C.29.92), (Dataset FinsScholl2456)
B Hughes, T Burghardt. Affinity Matting for Pixel-accurate Fin Shape Recovery from Great White Shark Imagery. Machine Vision of Animals and their Behaviour (MVAB), Workshop at BMVC, pp. 8.1-8.8. BMVA Press, September 2015. (DOI:10.5244/CW.29.MVAB.8), (Dataset FinsScholl2456)

Friesian Cattle Identification

Tilo Burghardt, Will Andrew, Jing Gao, Neill Campbell, Andrew Dowsey, S Hannuna, Colin Greatwood

Holstein Friesian cattle are the highest milk-yielding bovine type; they are economically important and especially prevalent within the UK. Identification and traceability of these cattle is not only required by exportand consumer demands, but in fact many countries have introduced legally mandatory frameworks.

This line of work has shown that robust individual Holstein Friesian cattle identification can be implemented automatically and non-intrusively using computer vision pipelines fuelled by architectures utilising deep neural networks. In essence, the systems biometrically interpret the unique black-and-white coat markings to identify individual animals robustly; identification can for instance happen via fixed in-barn cameras or via drones in the field.

This work is being conducted with the Farscope CDT, VILab and BVS.

Example Training Set of a Small Herd

Related Publications

W Andrew, C Greatwood, T Burghardt. Aerial Animal Biometrics: Individual Friesian Cattle Recovery and Visual Identification via an Autonomous UAV with Onboard Deep Inference. 32nd IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 237-243, November 2019. (DOI:10.1109/IROS40897.2019.8968555), (Arxiv PDF), (CVF Extended Abstract at WACVW2020)
W Andrew, C Greatwood, T Burghardt. Deep Learning for Exploration and Recovery of Uncharted and Dynamic Targets from UAV-like Vision. 31st IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1124-1131, October 2018. (DOI:10.1109/IROS.2018.8593751), (IEEE Version), (Dataset GTRF2018), (Video Summary)
W Andrew, C Greatwood, T Burghardt. Visual Localisation and Individual Identification of Holstein Friesian Cattle via Deep Learning. Visual Wildlife Monitoring (VWM) Workshop at IEEE International Conference of Computer Vision (ICCVW), pp. 2850-2859, October 2017. (DOI:10.1109/ICCVW.2017.336), (Dataset FriesianCattle2017), (Dataset AerialCattle2017), (CVF Version)
W Andrew, S Hannuna, N Campbell, T Burghardt. Automatic Individual Holstein Friesian Cattle Identification via Selective Local Coat Pattern Matching in RGB-D Imagery. IEEE International Conference on Image Processing (ICIP), pp. 484-488, ISBN: 978-1-4673-9961-6, September 2016. (DOI:10.1109/ICIP.2016.7532404), (Dataset FriesianCattle2015)

Great Ape Facial Recognition and Identification

Tilo Burghardt, O Brookes, CA Brust, M Groenenberg, C Kaeding, HS Kuehl, M Manguette, J Denzler, AS Crunchant, M Egerer, A Loos, K Zuberbuehler, K Corogenes, V Leinert, L Kulik

In order to evaluate the status of great ape populations and the effectiveness of conservation interventions accurate monitoring tools are needed. The utilisation and interpretation of field photography and inexpensive autonomous cameras can often provide detailed information about species presence, abundance, behaviour, welfare or population dynamics.

Together with researchers from various institutions including the Max Planck Institute for Evolutionary Anthropology, the University of Jena, and Bristol Zoo Gardens, we co-developed various computer vision and deep learning systems for detecting great ape faces in imagery and for identifying individual animals based on their unique facial features. These techniques can be applied in the wild using camera traps or manual photography, or in captive setting for studying welfare and behaviours.

Related Publications

O Brookes, T Burghardt. A Dataset and Application for Facial Recognition of Individual Gorillas in Zoo Environments. In press. Proc. 25th International Conference on Pattern Recognition (ICPR) Workshop on Visual Observation and Analysis of Vertebrate And Insect Behavior (VAIB), January 2021. (Arxiv PDF)
CA Brust, T Burghardt, M Groenenberg, C Kaeding, HS Kuehl, M Manguette, J Denzler. Towards Automated Visual Monitoring of Individual Gorillas in the Wild. Visual Wildlife Monitoring (VWM) Workshop at IEEE International Conference of Computer Vision (ICCVW), pp. 2820-2830, October 2017. (DOI:10.1109/ICCVW.2017.333), (Dataset Gorilla2017), (CVF Version)
AS Crunchant, M Egerer, A Loos, T Burghardt, K Zuberbuehler, K Corogenes, V Leinert, L Kulik, HS Kuehl. Automated Face Detection for Occurrence and Occupancy Estimation in Chimpanzees. American Journal of Primatology. Vol 79, Issue 3, ISSN: 1098-2345. March 2017. (DOI 10.1002/ajp.22627)
HS Kuehl, T Burghardt. Animal Biometrics: Quantifying and Detecting Phenotypic Appearance. Trends in Ecology and Evolution, Vol 28, No 7, pp. 432-441, July 2013.
(DOI:10.1016/j.tree.2013.02.013)