Introduction
Computer vision is a sophisticated, non-destructive, efficient and robust inspection and monitoring technology with the potential to automate and integrate a wide range of processes. Nowadays, computer vision methods are successfully applied in various industries but seldom used in aquaculture. The applications found in this field rely often on manual interpretation and analysis of low quality imaging material. This is due to difficult optical and environmental underwater conditions - lighting, visibility and deployment stability are commonly not well controllable - along with highly complex and varying imaging scenes. Applied computer vision approaches use mostly color segmentation techniques, background subtraction, shape priors, moving average algorithms or pattern classification methods, but most of these techniques rely on manual labeling techniques to mark the features of interest and manual parameter adjustment in order to adapt to the change of scenery. A fundamental feasibility study of computer vision technology, as partly presented her, has to be discussed with respect to these challenges. Applications and methods relevant for the industry will be demonstrated with the potential for automation and improvement of salmon production.
Materials and method
Imaging material has been generated in full scale trials at industrial salmon farms. The data material was recorded during different experiments, which investigating different aquaculture operational settings. Sources of the image material presented are e.g. standalone camera rigs, feeding cameras, embedded camera solutions within remotely operated vehicles (ROVs) or unmanned aerial vehicles (UAVs) with mono and stereo vision, as well as sonar technology.
We evaluated established techniques from other fields of applications and adapted and improved the algorithms in order to overcome certain limitations caused by underwater and illumination conditions. Subjects related to image enhancement, non-linear operations, image segmentation, motion estimation, video tracking, objet recognition, pose estimation and machine learning will be presented.
Discussion and Conclusion
The presented imaging material and methods will be discussed with respect to potential applications of computer vision technology in salmon farming and related research. This will include evaluations on whether analysis results obtained from such data and analysis tools can be used to achieve a higher degree of automation in daily operation at fish farms. All discussions will be conducted in light of central aspects in fish farming such as welfare monitoring and analysis, operational use on underwater vehicles, cage inspection monitoring and feeding. Specific case-studies will be investigated whether the usability of computer vision can provide knowledge and technology towards automation of cage-fish-environment systems in order to develop future welfare monitoring, feeding and fishnet inspection methods.