Heikki Kälviäinen presents Automatic image-based re-identification of ringed seals (online at https://lut.zoom.us/j/67484764485
On 2022-06-08 10:00:00 at G205, Karlovo náměstí 13, Praha 2
zoom: https://lut.zoom.us/j/67484764485?pwd=OEhLTGZpb2N4RzhScVYrSkJLcFNOQT09
To monitor an animal population and to track individual animals in a
non-invasive way for conservation of nature purposes, the identification of
individual animals based on certain distinctive characteristics is necessary.
In this presentation, the automatic image-based individual re-identification of
the critically endangered Saimaa ringed seals (Phoca hispida saimensis) is
considered. These ringed seals living in the Lake Saimaa have a distinctive
permanent pelage pattern which is unique to each individual seal. This can be
used as a basis for the identification process.
The general pipeline for the automatic processing of camera trap and handheld
images of seals is proposed. Without this kind of the pipeline, biologists
should analyze the images manually which is laborious and time-consuming. The
pipeline consists of the following steps: image preprocessing including seal
segmentation, extraction of local pelage patterns, patch extraction of the
pattern images, and re-identification using the Siamese neural networks and
feature aggregation. For experiments, the novel dataset of the Saimaa ringed
seal images (SealID) has been collected and annotated. Moreover, Ladoga ringed
seals (Pusa hispida ladogensis), a sister species of the Saimaa ringed seals
were studied in case of individual grouping of seals. The experiments show that
the approach can assist the biologists to automate the identification and could
be applied for on-line monitoring in the Lake Saimaa, instead the current
manual
analysis with a significant delay.
Further information:
https://www.it.lut.fi/project/coexist/index.shtml
https://www.it.lut.fi/project/sealvision/index.shtml
To monitor an animal population and to track individual animals in a
non-invasive way for conservation of nature purposes, the identification of
individual animals based on certain distinctive characteristics is necessary.
In this presentation, the automatic image-based individual re-identification of
the critically endangered Saimaa ringed seals (Phoca hispida saimensis) is
considered. These ringed seals living in the Lake Saimaa have a distinctive
permanent pelage pattern which is unique to each individual seal. This can be
used as a basis for the identification process.
The general pipeline for the automatic processing of camera trap and handheld
images of seals is proposed. Without this kind of the pipeline, biologists
should analyze the images manually which is laborious and time-consuming. The
pipeline consists of the following steps: image preprocessing including seal
segmentation, extraction of local pelage patterns, patch extraction of the
pattern images, and re-identification using the Siamese neural networks and
feature aggregation. For experiments, the novel dataset of the Saimaa ringed
seal images (SealID) has been collected and annotated. Moreover, Ladoga ringed
seals (Pusa hispida ladogensis), a sister species of the Saimaa ringed seals
were studied in case of individual grouping of seals. The experiments show that
the approach can assist the biologists to automate the identification and could
be applied for on-line monitoring in the Lake Saimaa, instead the current
manual
analysis with a significant delay.
Further information:
https://www.it.lut.fi/project/coexist/index.shtml
https://www.it.lut.fi/project/sealvision/index.shtml
External www: https://www.it.lut.fi/project/sealvision/index.shtml