Sea Ice Extraction via Remote Sensed Imagery: Algorithms, Datasets, Applications and Challenges (2306.00303v1)

Abstract: The deep learning, which is a dominating technique in artificial intelligence, has completely changed the image understanding over the past decade. As a consequence, the sea ice extraction (SIE) problem has reached a new era. We present a comprehensive review of four important aspects of SIE, including algorithms, datasets, applications, and the future trends. Our review focuses on researches published from 2016 to the present, with a specific focus on deep learning-based approaches in the last five years. We divided all relegated algorithms into 3 categories, including classical image segmentation approach, machine learning-based approach and deep learning-based methods. We reviewed the accessible ice datasets including SAR-based datasets, the optical-based datasets and others. The applications are presented in 4 aspects including climate research, navigation, geographic information systems (GIS) production and others. It also provides insightful observations and inspiring future research directions.

• The paper compares classical, machine learning, and deep learning approaches for sea ice extraction, demonstrating deep learning’s superior precision in complex scenes.
• It highlights the critical role of open-source datasets like Sentinel-1 SAR and Sentinel-2 imagery in training robust models for climate forecasting and maritime navigation.
• The study outlines major challenges such as data scarcity and underwater ice detection, paving the way for future advancements in polar environmental monitoring.

Remote Sensing-Based Sea Ice Extraction: Techniques and Applications

The paper "Sea Ice Extraction via Remote Sensed Imagery: Algorithms, Datasets, Applications and Challenges" offers a comprehensive examination of the methodologies, data sources, applications, and future challenges associated with Sea Ice Extraction (SIE) through remote sensing technologies. The authors have extensively reviewed the literature published since 2016, emphasizing the growing role of deep learning techniques in recent years.

Summary of Methodologies

The paper categorizes SIE methodologies into three primary classes: classical image segmentation methods, machine learning-based approaches, and deep learning-based techniques.

1. Classical Methods: Initial SIE techniques like Bayesian inference, Maximum Likelihood Estimation, and various thresholding methods form the backbone of early remote sensing approaches. While computationally efficient, these methods struggle with complex scenes due to reliance on pre-determined thresholds, which lack adaptability across varying image conditions.
2. Machine Learning-Based Methods: Enhancements have been seen with the introduction of machine learning models like Random Forest (RF), Support Vector Machines (SVM), and ensemble learning techniques. They leverage features such as texture and polarization to classify sea ice, offering improved robustness and accuracy over classical methods. However, feature extraction remains a manual and laborious process.
3. Deep Learning-Based Methods: Recent advancements have shifted focus towards deep learning, which automates feature extraction, significantly improving SIE's precision across complex scenarios. Techniques such as CNNs, U-Nets, and GANs have been extensively applied, with architecture modifications tailored for specific remote sensing tasks. These methods outperform traditional approaches, leveraging higher computational capabilities and vast datasets for enhanced generalization.

Datasets and Data Sources

The paper identifies several open-source datasets crucial for advancing SIE studies. Notably, Sentinel-1 SAR imagery, along with optical datasets such as from Sentinel-2 and UAV-derived imagery, provide diverse data sources instrumental for training robust SIE models. SAR data, owing to its all-weather day-and-night capabilities, remains pivotal in acquiring comprehensive ice coverage.

Applications in Climate Research and Navigation

Accurate SIE is indispensable for multiple applications:

• Meteorological and Climate Forecasting: Improved sea ice models contribute to more accurate climate predictions by reflecting the intricate interactions between ice extent and climate variables.
• Maritime Navigation: Sea ice maps derived from SIE facilitate safer navigation by enabling real-time route planning, crucial for Arctic shipping lanes experiencing rapid sea ice depletion.
• Geospatial Information Products: SIE provides foundational data for generating maps and geographic information systems that support ecological monitoring and natural resource management in polar environments.

Challenges and Future Directions

The authors underscore several ongoing challenges and prospective areas for development:

• Data Scarcity and Multi-Sensor Integration: The limited availability of accurate, labeled sea ice datasets remains a hindrance. Integrating data from multiple sensors (optical, SAR, and acoustic) can mitigate this issue by providing complementary insights into sea ice characteristics.
• Underwater Ice Detection: Current methodologies predominantly focus on surface ice. Development of technologies for assessing underwater ice features remains nascent yet crucial.
• Polar GIS Improvements: Enhancing geographic information systems to better depict dynamic and multidimensional polar data is needed for comprehensive modeling and visualization.

Conclusion

This paper presents an in-depth review of SIE methodologies, emphasizing the evolution from simpler statistical models to advanced deep learning frameworks. The prevailing challenges and potential directions outlined, such as large-scale model development and improved multi-source data fusion, set a clear trajectory for future SIE research. Addressing these will be vital for advancing the accuracy and applicability of sea ice monitoring systems essential in the context of global climate dynamics and maritime safety.