研究目的
To propose an unsupervised PolSAR land classification system that can discover new and more detailed land categories by employing quaternion auto-encoder and quaternion SOM for feature extraction and classification, respectively.
研究成果
The proposed system successfully extracts features necessary for land classification while filtering noise and discovers new detailed land categories such as residential areas and factory areas. This expands the applicability of PolSAR land classification.
研究不足
The system's performance is dependent on the quality of the PolSAR data and the predefined scattering models. The discovery of new land categories is limited by the resolution and coverage of the satellite data.
1：Experimental Design and Method Selection:
The system consists of three functions: generation of six Poincare parameters, feature extraction based on quaternion auto-encoder, and unsupervised classification based on the quaternion SOM.
2：Sample Selection and Data Sources:
PALSAR (Phased Array type L-band Synthetic Aperture Radar) level
3：1 data of ALOS (Advanced Land Observing Satellite) obtained for Fujisusono, Japan. List of Experimental Equipment and Materials:
Quaternion auto-encoder and quaternion SOM.
4：Experimental Procedures and Operational Workflow:
Generation of six Poincare parameters from each pixel of PolSAR data, feature extraction using quaternion auto-encoder, and classification using quaternion SOM.
5：Data Analysis Methods:
Comparison of classification results with and without the quaternion auto-encoder.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
