Graph Embedding Interclass Relation-Aware Adaptive Network for Cross-Scene Classification of Multisource Remote Sensing Data | JoVE Visualize

Area of Science:

Computer Science
Remote Sensing
Artificial Intelligence

Background:

Unsupervised domain adaptation (UDA) is crucial for cross-scene remote sensing image classification, enabling the use of labeled data from one scene for another.
Existing UDA methods face challenges with multisource data heterogeneity, incomplete feature representation (neglecting global information), and inaccurate distribution alignment.
Simple classifiers in UDA are susceptible to domain shifts, leading to suboptimal performance.

Purpose of the Study:

To propose a novel network, GeIraA-Net, for unsupervised classification of multi-source remote sensing data.
To facilitate knowledge transfer at the class level by perceiving inter-class relationships through aligned features.
To enhance the accuracy and robustness of cross-scene classification in the presence of domain shifts.

Main Methods:

A graph-based progressive hierarchical feature extraction network is employed to capture both local and global features, consolidating domain information into a unified space.
A joint de-scrambling alignment strategy is utilized with a three-step pseudo-label generation module for precise domain calibration.
An adaptive inter-class topology-based classifier is developed to make the classification process domain-adaptive at the category level.

Main Results:

GeIraA-Net demonstrates significant advantages over current state-of-the-art cross-scene classification methods.
The proposed methods effectively address challenges in multisource data heterogeneity and feature distribution alignment.
The network successfully leverages aligned features to perceive and utilize inter-class relationships for improved classification.

Conclusions:

GeIraA-Net offers a robust solution for unsupervised domain adaptation in multi-source remote sensing image classification.
The approach effectively mitigates domain shifts and enhances classification performance by considering inter-class relationships.
This work advances the field by providing a more accurate and adaptable method for cross-scene remote sensing analysis.