Learning Based Image Segmentation of Pigs in a Pen
Författare
Summary, in English
As farms are getting bigger with more animals,
less manual supervision and attention can be given the animals
on both group and individual level. In order not to jeopardize
animal welfare, automated supervision is in some way already
in use. Function and control of ventilation is already in use in
modern pig stables, e.g. by the use of sensors for temperature,
relative humidity and malfunction connected to alarm. However,
by measuring continuously directly on the pigs, more information
and more possibilities to adjust production inputs would be
possible. In this work, the focus is on a key image processing
algorithm aiding such a continuous system - segmentation of pigs
in images from video. The proposed solution utilizes extended
state-of-the-art features in combination with a structured prediction
framework based on a logistic regression solver using elastic
net regularization. Objective results on manually segmented
images indicate that the proposed solution, based on learning,
performs better than approaches suggested in recent publications
addressing pig segmentation in video.
less manual supervision and attention can be given the animals
on both group and individual level. In order not to jeopardize
animal welfare, automated supervision is in some way already
in use. Function and control of ventilation is already in use in
modern pig stables, e.g. by the use of sensors for temperature,
relative humidity and malfunction connected to alarm. However,
by measuring continuously directly on the pigs, more information
and more possibilities to adjust production inputs would be
possible. In this work, the focus is on a key image processing
algorithm aiding such a continuous system - segmentation of pigs
in images from video. The proposed solution utilizes extended
state-of-the-art features in combination with a structured prediction
framework based on a logistic regression solver using elastic
net regularization. Objective results on manually segmented
images indicate that the proposed solution, based on learning,
performs better than approaches suggested in recent publications
addressing pig segmentation in video.
Avdelning/ar
- Matematik LTH
- Mathematical Imaging Group
- ELLIIT: the Linköping-Lund initiative on IT and mobile communication
Publiceringsår
2014
Språk
Engelska
Sidor
1-4
Fulltext
Länkar
Dokumenttyp
Konferensbidrag
Ämne
- Mathematics
Nyckelord
- Precision Livestock Farming
- Machine Learning
- Computer Vision
Conference name
Visual observation and analysis of Vertebrate And Insect Behavior 2014
Conference date
2014-08-24
Conference place
Stockholm, Sweden
Status
Published
Forskningsgrupp
- Mathematical Imaging Group