An Approach for Estimation of Ice Thickness Region Using TensorFlow
- Dong-Ham Kim (Ocean University) | Jong-Ho Nam (Ocean University)
- Document ID
- International Society of Offshore and Polar Engineers
- The 28th International Ocean and Polar Engineering Conference, 10-15 June, Sapporo, Japan
- Publication Date
- Document Type
- Conference Paper
- 2018. International Society of Offshore and Polar Engineers
- Ice thickness, image convolution, spatial gradient filter, lower boundary detection
- 0 in the last 30 days
- 13 since 2007
- Show more detail
When an icebreaker navigates the Arctic sea route, the sea ice is broken by the icebreaker weight. Broken pieces of ice tend to turn towards the hull and the thickness of turning ice is frequently observed. By applying the image processing technique to the recorded scenes, it is possible to estimate the ice thickness during or after voyage. When the icebreaker is in a certain environment, however, the area possessing the ice thickness is not well recognized. This paper introduces an approach to detect the area showing the ice thickness by identifying the lower boundary of turning ice section.
The ice thickness is an important factor for safety when an icebreaker navigates the Arctic sea route. Various methods, such as direct observation, the laser from satellite (Sandven and Johannesen, 2006), equipment based on electromagnetic induction (Tateyama, Shirasawa, Uto, Kawamura, Toyota and Enomoto, 2006), sonar and so forth, have been widely used to estimate the ice thickness. An icebreaker normally glides smoothly over the sea ice because of its bow shape and then uses its own weight to break the sea ice, which is called the ramming. Due to this ramming action, it is often observed at the side of an icebreaker that a patch of broken ice turns towards the hull. A human onboard can estimate the ice thickness by observing the turning ice, but it is subjective and requires training.
Researchers have taken advantage of estimating the ice thickness using image processing techniques, as the breaking scenes can be recorded and stored in a film or digital format. Kulovesi and Lehtiranta (2014) suggested a semi-automatic method that the user interactively assigns the ice thickness to a processed image. Nam, et al. (2013) introduced an automatic method that ice thickness was automatically calculated using the image processing technique. Their method worked well for the most images that clearly show the turning ice section, but occasionally failed because it did not detect an ice thickness region. This is very natural as the normal image taken in the field contains lots of ambiguities due to illumination, irregularities of broken ices, and fuzzy environment.
|File Size||1 MB||Number of Pages||4|