The growing activities in the Arctic have led to a need for accurate modeling of ice properties and interaction forces with structures to ensure safe operations of ships and offshore platforms. Traditional and machine learning-based methods are being explored for predicting ice forces, often utilizing image processing techniques to extract information from ice floe images. However, challenges remain in accurately extracting complex ice features from images due to their varied shapes and lighting conditions. This research introduces two novel Hybrid models that directly extract ice characteristics from images and train machine learning-based force predictors. The models are evaluated based on image segmentation and force prediction performance, suggesting their potential for improving ice force predictions. Future work could involve further refining these Hybrid models and exploring alternative approaches for direct ice force prediction from images.
Read more: - https://research.library.mun.ca/15936/