Research On Classification Methods Of Fully Polarimetric SAR Image Based On Target Decomposition Theory

Compared with the traditional optical remote sensing image, fully polarimetric SAR image contains more abundant information. Therefore, it can be effectively extract the polarimetric characteristics of various kinds ofgeographic information from the fully polarimetric SAR image and classify them according to the polarization characteristics.In this paper, Zijin Mountain National Forest Park was chosen as the case study area, fully polarimetric SAR image in 2011 and 2015 wereusedasthe data source. Sinclair, Pauli,Freeman and Cloude four kinds of target decomposition methods were used to extract the twelve polarization characteristic values of the fully polarimetric SAR image in the study area.Four classification methods namely decision tree, neural network, maximum likelihoodand random forest were used toclassify thepolarization characteristics images of the study area.Based on the classification results,the classification accuracy of the four classification methods was evaluatedandthe importance of the polarization characteristic value is analyzed.Finally, analyzed the dynamic changes of land types during 2011 to 2015, and predicted the land type in the study area in 2019. Through the analysis of land types shift driving factors, proposed the policy recommendationsto protect the ecological environment of the study area.Research results showedas follows.(1)From the polarization characteristics imagesof the four kinds of polarization target decomposition methods, it can be seen: the color characteristics of the water, construction and bare land was obvious and it was easy to distinguish, and the color characteristics of the grassland, coniferous forest and broadleaf forest was not obvious, it was difficult to distinguish the different land types.From the classification accuracy of the six types of land in 2011 and 2015, it can be seen:the land types with obvious color characteristics have higher classification accuracy. Therefore, in the fully polarization SAR image classification the water, construction land and bare land have higher classification accuracy than the grassland, coniferous forest and broadleaf forest.(2)From the classification accuracy of fully polarimetric SAR image, it can be seen:the Overall Accuracy of the four classification methods reachedabove 80%, while the Kappa coefficient reachedabove 0.7. This shows that the classification results of fully polarimetric SAR image by using the four kinds of classification methods werequite satisfactory. However, due to the random forest classification method had many advantages, making it the highest classification accuracy of these four classification methods. Therefore, random forest classification was a good method of fully polarimetric SAR image classification.By analyzing the importance of the polarization characteristic value to the classification results, it can be seen that the polarization characteristic value obtained by the Cloude decomposition plays a key role in the fully polarization SAR image classification. Cloude decomposition was a good method for polarimetric target decomposition when extracting the polarimetric characteristics of fully polarimetric SAR image.(3)From the changes ofthe six land types instudy area during 2011 to 2015,it can be seen: water and construction land area increased slightly; bare land area was greatly reduced,mainlychanged into construction land and grassland, a small part changed into coniferous forest and broadleaf forest; grassland area wasalso greatly reduced, mainly changed into construction land and coniferous forest and broadleaf forest;coniferous forest area was significantly reduced, the major changed into broadleaf forest; broadleaf forest area has increased a lot, a small partchanged into coniferous forest. And the prediction of land types result in 2019 showsthat, the construction land and broadleaf forest area continue to increase.Water area was basically unchanged.The bareland, grassland and coniferous forest area wasfurther reduced. This was consistent with the trend of land types in the study area during 2011 to 2015.(4)The mutual transformation of different land types in the study areaduring 2011 to 2015 was mainly caused by anthropogenic disturbancesand natural disturbances, especially,the influence of anthropogenic disturbances was the most obvious. In order to reduce these disturbances damage the ecological environment of the study area,and keep the ecological environment in a healthy and stable direction,some policy suggestions were put forward.The above research can provide a scientific reference for the application of fully polarimetric SAR image in land use classification of Urban Forest Park.It can also provide a scientific and rational planning for the future management of the study area and the similar Urban Forest Park.