Font Size: a A A

Remote Sensing Methods On Vegetation Identification And Leaf Area Index Inversion

Posted on:2016-08-14Degree:MasterType:Thesis
Country:ChinaCandidate:Z DongFull Text:PDF
GTID:2180330461492149Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
As a new type of synthetic detection technology characterized by objectivity, instantaneity and non-destrctiveness, remote sensing provides data for acquiring the vegetation distribution of large areas, giving its advantage to full play in classifiying vegetation type and monitoring growth vigor. With the rapid development of high spectral, space and time resolution satellite remote sensing technology as well as its combination with geographic information system technology, the accuracy of vegatation monitoring using remote sensing method is steadily increasing. This study first analyzed the identification ability of vegetation remote sensing method based on supervised classification; then inverted an important physical parameter of vegation-leaf area index(LAI). A set of LAI inverstion system was developed at the end of the study so that the processing procedure was simplified. The content and major result is as follows:1) Pixel-based supervised classification method (support vector machine, maximum likelihood and mahalanobis distance) as well as object-oriented method (support vector machine classification based on attributes of the object) was adopted to identify vegatation type. The analysis of the result showed that:the identification accuracy of support vector machine algorithm is the highest among the three pixel-based supervised classification algorithms.The producer’s accuracy and user’s accuracy of the support vector machine are greater than the others’, and its overall classification accuracy is the highest (81.15%), much better than the maximum likelihood method (73.33%) and mahalanobis distance (61.77%). Moreover, adding spatial information, the overall identification accuracy of object-oriented method is even higher (89.24%), and the commission and omission errors are decreasing significantly with the increase of the object’s merger scale. This indicates that the identification ability of the support vector machine algorithm is the best among pixel-based classification method and the noise can be further weakened by adding object’s spatial texture attributes to improve the identification accuracy.2) Multi-spectral LAI inversion models (NDVI-LAI, RVI-LAI and SVM-LAI) were built based on vegetation index (NDVI and RVI) as well as support vector machine (SVM), using physical model (PROSAIL) data. Then we use ground measured data to conduct the inversion of LAI. The comparison with ground measured LAI reveals:NDVI-LAI and RVI-LAI model yield poor inversion results with saturation effect (R2<0.61, RMSE>1.1); inversion accuracy of SVM-LAI model is higher than that of vegetation index models with higher R2 for ASD data and TM5 data (0.7858 and 0.7447 respectively) and good anti-saturation ability. Later, LLE and PC A algorithm were used for dimension reduction of the hyperspectral data and two hyperspectral LAI inversion method were built (LLE-SVM and PCA-SVM). The verification shows that the unsatisfactory inversion result of reduced-dimension transformation data is due to the difference between model data and ground measured data (R2<0.48, RMSE>1.5)3) An LAI inversion system was developed using C# in.NET platform and external algorithm (GDAL and LIBSVM). The function of this system include: retrieval of vegetation fraction cover, inversion of LAI, batch processing, visualized analysis of output data and etc.. The algorithm of this system integrate LAI inversion algorithms based on vegetation index (NDVI and RVI) as well as SVM-based classification method and inversion method. It simplifies the processing procedure of inversion and enhance the efficiency of LAI inversion and data analysis.
Keywords/Search Tags:Remote sensing identification, Leaf area index, PROSAIL, Inversion system
PDF Full Text Request
Related items