Abstract
III
Abstract
Polarimetric Synthetic Aperture Radar (POLSAR) has more information than
Synthetic Aperture Radar (SAR), in the recently years, it has become the one of the
important directions of the development of SAR, so Polarimetric SAR image
interpretation also will become increasingly important. The polarimetric SAR image
classification is an important part of Polarimetric SAR image interpretation, and has
been widely applied in the agricultural, military, marine and other fields.
This paper mainly focuses on the study of the polarimetric SAR image
classification method based on Affinity Propagation Clustering algorithm. Recently,
Affinity Propagation Clustering algorithms has become a hot research field of machine
learning, which can adaptively obtain the number of clusters, and the method is rapid
and effective. In this paper, combining with cluster analysis and polarization
characteristics of SAR images, three polarimetric SAR image classification methods are
proposed based on affinity propagation clustering algorithm, which mainly include the
following three aspects:
1. We proposed a new classification method based on the improved Affinity
Propagation Clustering to classify the polarimetric SAR image. The proposed method
uses the Wishart distance between the clusters to replace the Euclidean distance in the
original algorithm as similarity measure for the improved Affinity Propagation
Clustering, and combines with yamaguchi decomposition to obtain the to the initial
segmentation, then Wishart classifier is applied to iterative the whole polarimetric SAR
data to further improve classification accuracy. The proposed method is used to classify
real polarimetric SAR image, it can obtain good classification results and be adaptive to
obtain the number of categories.
2. We presented a polarimetic SAR image classification method using Turbopixel
and Affinity Propagation Clustering. In the proposed method, the polarization
characteristics that are geted by Pauli decomposition are regarded as the input of the
Turbopixel algorithm, and the polarimetric SAR image is implemented by Turbopixel
algorithm to divid the image into non-overlapping areas, and each area as the input data
points to the algorithm to improve the efficiency and perform of the Affinity
Propagation Clustering algorithm. Based on the traditional Turbopixel algorithm, we
introduced characteristics of the pixel and the spatial relationship between points to
万方数据
