The Simple Scatterers Decomposition with Polarimetric SAR Data and its Implication for Forest Classification

Abstract

This paper proposes a new approach of decomposing Mueller matrix of full-polarimetric synthetic aperture radar (SAR) data. The large conducting sphere, the dihedral corner reflector, the short thin cylinder (wire) and the helix are known as simple scatterers. The new approach presents an algorithm of decomposing Mueller matrix into the above four simple scatterers. The components of the above simple scatterers have been applied to classify the planted forest. Two full-polarimetric imagery of Tomakomai in Japan generated by near crossed flights JPL/AIRSAR are used for classification. Four categories in the study area, Broadleaf, Larch, Fir and Spruce, have been classified. The average accuracy of classification is 83.28% by using C, L and P band data, while the average accuracy of classification is 72.25% by only using L band data. Through the new decomposing Mueller matrix method proposed in this paper, it becomes obvious that the wire scatterer is closely correlative with branch. In L-band, the wire angles of Broadleaf show as a uniform distribution and the wire power is weak. On the contrary, the wire angles of Conifer show as a Gaussian distribution with high kurtosis near a horizontal direction and the wire power is strong. As a result of classification, the components of wire scatterer are important factors to classify Broadleaf and Conifer.