Fang Shang, Assistant Professor, Graduate School of Informatics and Engineering.

Fang Shang is working on the interpretation and applications of data from Synthetic Aperture Radar (SAR). Previously, Shang proposed a method based on stokes vector. One of the topics is how to use stokes vector based parameters for supervised land classification. Usually, quaternion neural networks (QNN) are used. But recently, Shang found that the processing in a QNN is not always isotropic.

That means that if the feature points are distributed on a sphere, the data after processing from a neural network is not distributed on a sphere. It will be somewhat cubic.

So, Shang tried to make the algorithm isotropic. After isotropization, the neural network will have higher performance for land classification. In the future, Shang will continue to improve her method and try to use it for specific applications such as observation of forests and monitoring of glaciers.

Further information

Fang Shang
Assistant Professor, Graduate School of Informatics and Engineering, The University of Electro-Communications, Tokyo.

Research Highlight: Isotropization of Quaternion-Neural-Network-Based PolSAR Adaptive Land Classification in Poincare-Sphere Parameter Space