Quality Assessment of Solar EUV Remote Sensing Images Using Multi-Feature Fusion.

Area of Science:

• Space Science
• Solar Physics
• Remote Sensing

Background:

• Accurate quality assessment of solar Extreme Ultraviolet (EUV) imagery is crucial for reliable space science data and weather forecasting.
• Existing methods may lack robustness in identifying subtle image degradations.

Purpose of the Study:

• To develop a robust hybrid framework for automated quality assessment of solar EUV remote sensing imagery.
• To enhance the reliability and accuracy of solar EUV data analysis pipelines.

Main Methods:

• A hybrid framework fusing deep semantic features (HyperNet-based) with 22 handcrafted physical/statistical quality indicators was developed.
• A 24-dimensional feature vector was created by combining these diverse feature types.
• A dataset of 47,950 degraded images simulating seven degradation types was generated from high-quality solar EUV images.

Main Results:

• An XGBoost classifier trained on the fused features achieved 97.91% accuracy and an Area Under the Curve (AUC) of 0.9992.
• The hybrid approach demonstrated superior performance compared to methods using only one type of feature.
• The framework proved robust in classifying image quality across various simulated degradations.

Conclusions:

• Combining deep learning features with handcrafted indicators significantly improves the robustness of solar EUV image quality classification.
• This hybrid approach offers a scalable and effective solution for automated quality control in solar observation data processing.
• The developed framework contributes to enhanced data reliability for space science applications and space weather forecasting.