Master's Research
The objective for my master's research is to forecast ice-loading on arctic marine vessels and offshore structures using Machine Learning for proactive control systems.
There are two main steps:
Computer vision to detect ice
Regression model to forecast ice loading predictions.
Computer Vision
For computer vision, I am using the Mask-RCNN machine learning model. The value here is that detection edges are defined rather than a box around the detection. Some results detecting large ice floe:
Dividing up ice detections: To divide up ice data, a polar coordinate system is overlayed on the ship's center. Ice beyond 2 ship lengths fore, and 0.5 ship lengths aft have no effect on ice loading, so this ice is not considered further. Overlayed coordinate system and ice detection centers.
Regression Model
To forecast ice load predictions, a Long-Short Term Memory (LSTM) model is used to forecast ice loading predictions.
Something I would like to try, is a physics-based approach using Popov ice loading model.
A short, high level overview presentation can be accessed using this button -->