Multi-dimension SAR For Forest Recognition And Forest Type Classification

Monitoring and mapping forest distribution are of great significance as forest plays animportant role on global carbon cycle and nature disturbance. Optical remote sensing is ofgreat significance in resources monitoring such as forest extraction and recognition.Nevertheless, the difficult problem of data acquisition exists in complete coverage forest areasof cloudy, rainy and foggy and also the serious problem of same spectrum with differentobjects in forest type. Synthetic Aperture Radar became the beneficial supplement of opticalremote sensing for its imaging capabilities with strong penetration, all-time and all-weather,and sometimes the only means for data acquisition. However, there have been unable to havemore perfect SAR classification system to distinguish the detailed forest type up to now,mainly owe to the various limitations such as SAR speckle noise, classification strategy, lessdimension information and its complex characteristics of forest vegetation itself. As the newSAR sensors have been launched, microwave remote sensing data acquisition method has beendeveloped from single band, single polarization and single angle to multi-frequency,multi-polarization, multi-angle and multi-temporal, then this provides unprecedentedopportunity and development potential for SAR forest classification.This paper would study land cover classification and forest land type distinguish usingSAR data by the comprehensive utilization of the four dimensions including polarization,InSAR, multi-temporal and PolInSAR in order to solve the limitation of forest typeidentification technology in forest resources investigation.(1) The area covered by two scenesof ALOS PALSAR data in Xunke County, Heilongjiang Province were chosen as test site. Aland cover type and forest type classification technique of SVM based on multi-temporal,multi-polarization and InSAR had been proposed, using the sensitivity to vegetation structureof multi-temporal, multi-polarization SAR and InSAR measurements, and combing time seriescharacteristic of backscatter coefficient and correlation coefficient.(2) A classification methodto distinguish forest types (coniferous forest and deciduous forest) was proposed combiningwith PolSAR and PolInSAR data, using the data acquired by DLR airborne SAR system (ESAR) in the Traunstein test site in Germany. Firstly, forest cover was distinguished from theother non-forest cover types based on PolSAR segmentation algorithm. However it cannotidentify the detailed forest types. Therefore, the complex coherent T6matrix and the coherentcoefficient of it were utilized to classify coniferous forest from deciduous forest. In order toconsider the polarization intensity information of single PolSAR data and the coherentinformation of a pair of PolSAR data, the maximum likelihood iterative classification based onT6matrix and coherent optimization R matrix was proposed by using the interferenceinformation of full polarization SAR data and Co-polarization difference characteristicparameters (Co-Pd).The results show that:(1) the classification of the confusion between forest land andurban construction land can be nicely distinguished, according the correlation coefficientbetween HH and HV, and also combing the selected temporal, polarization and InSARcharacteristics. The land cover classification result with higher accuracy is gotten using theclassification algorithm proposed in this paper. On the other hand, the average correlationcoefficient of multi-temporal is very effective to distinguish forest land, shrub land and openforest land. Comprehensive utilization of the effective information such as multi-temporalInSAR and polarization ratio is able to stress the features and structures of object and toclassify forest land types in detail in the forest land classification.(2) The maximum likelihooditerative classification based on T6matrix and coherent optimization R matrix can reach thegoal to subdivide forest internal structure in forest type classification using PolInSAR data.And the classification algorithm proposed in this paper based on coherent optimization Rmatrix is the best method with the clear classification result.