Online Semisupervised Active Classification for Multiview PolSAR Data.

Area of Science:

• Remote Sensing
• Machine Learning
• Data Science

Background:

• Polarimetric synthetic aperture radar (PolSAR) data acquisition is sequential, yielding multiple views from various feature extractors or frequency bands.
• Fast and accurate classification of PolSAR data in dynamic environments presents a significant challenge.
• Online learning offers a solution by incrementally updating classifiers from data streams.

Purpose of the Study:

• To propose an online semisupervised active learning framework (OSAM) for multiview PolSAR data classification.
• To address the challenges of dynamic, large-scale, and evolving PolSAR datasets.
• To improve the efficiency and accuracy of PolSAR data classification in changing environments.

Main Methods:

• Developed a novel online active learning strategy leveraging multiview relationships and randomized rules to query informative samples.
• Introduced an online semisupervised learning model using co-regularized multiview learning and graph regularization.
• Designed the framework to handle increasing data volume and class numbers during the learning process.

Main Results:

• The proposed OSAM framework demonstrates effectiveness in classifying multiview PolSAR data.
• Experimental results on real PolSAR data validate the algorithm's performance.
• The method successfully handles dynamic and large-scale datasets with evolving class structures.

Conclusions:

• OSAM provides an effective solution for online, semisupervised, active learning in multiview PolSAR data classification.
• The framework is robust to dynamic environments and scales with increasing data and class complexity.
• The rigorous derivation of the mistake bound supports the method's theoretical foundation.