一种自解释的多视角遥感图像分类方法
By employing a self-interpretive multi-view remote sensing image classification method, which utilizes an interpreter to generate pixel-level attribution masks and combines them with a master-aid classifier, the inherent interpretability and computational efficiency issues of multi-view remote sensing image classification models are resolved. This achieves high-precision, compact pixel-level interpretation, thereby enhancing the model's reliability and generalization ability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TIANMUSHAN LABORATORY
- Filing Date
- 2026-04-07
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multi-view remote sensing image classification models lack inherent interpretability, have high computational overhead and low efficiency, and suffer from inaccurate attribution due to out-of-distribution data. They also exhibit a disconnect between interpretability and classification performance, making it difficult to meet the requirements for high reliability and real-time processing.
A self-interpretive multi-view remote sensing image classification method is adopted. The interpreter generates a pixel-level attribution mask, and the total loss function is constructed by combining the main classifier and the auxiliary classifier to achieve joint optimization of the interpreter and the classifier, generating a compact and visually reliable pixel-level interpretation.
During training, the model autonomously focuses on key regions with clear semantics and strong discriminative power, generating faithful and compact pixel-level explanations, which improves the interpretability and classification accuracy of the model and overcomes the shortcomings of existing technologies.
Smart Images
Figure CN121982432B_ABST
Abstract
Citation Information
Patent Citations
Image classification method and device
CN112446398A
Multi-modal data urban land utilization scene classification method based on street scene image matching remote sensing image
CN119693781A