Deep Learning on SAR Imagery: Transfer Learning Versus Randomly Initialized Weights
This work addresses the challenge of frequent sea ice mapping for safe marine navigation, but it is incremental as it compares existing transfer learning methods rather than introducing new techniques.
The paper tackled the problem of limited expert-labeled data for deep learning on Synthetic Aperture Radar (SAR) imagery for sea ice mapping by comparing models trained from scratch with randomly initialized weights against fine-tuned pre-trained models, finding that pre-trained models performed better, particularly during the melt season.
Deploying deep learning on Synthetic Aperture Radar (SAR) data is becoming more common for mapping purposes. One such case is sea ice, which is highly dynamic and rapidly changes as a result of the combined effect of wind, temperature, and ocean currents. Therefore, frequent mapping of sea ice is necessary to ensure safe marine navigation. However, there is a general shortage of expert-labeled data to train deep learning algorithms. Fine-tuning a pre-trained model on SAR imagery is a potential solution. In this paper, we compare the performance of deep learning models trained from scratch using randomly initialized weights against pre-trained models that we fine-tune for this purpose. Our results show that pre-trained models lead to better results, especially on test samples from the melt season.