Hyper-spectral data transformation and identiifcation of wetland vegetation in east Dongting lake region
In order to get better observation effect of the target, the objective of hyper-spectral identiifcation is to simultaneous observe the target by using lots of narrow bands. By taking east Dongting lake area as the studied object, the ifeld observations of ifve typical wetland vegetation such as moss grass, reeds, selengensis red-knees herb and willow were conducted with method of hyper-spectral remote sensing, then, the measured data were transformed, classified and identified. After culling, filtering and re-sampling of the data, the hyper-spectral data obtained were treated with six kinds transformation operations (including d(R)(b), log(R), d(log(R)), N(R), d(N(R)) and log(N(R)) in order to highlight the differences of spectral characteristics for various wetland vegetation. By using principal component analysis method, the dimensionality reduction of hyper-spectral data was carried out. Then six classification methods including back propagation, mahalanobis, bayes, ifsher, spectral angle mapper (SAM) and support vector machine (SVM) were employed to identify different wetland vegetation based on the principal component analysis. The results show that for the methods of data transformation, log (N(R)) had the best effect;while for the methods of vegetation identiifcation, SAM had the highest accuracy.
hyper-spectral remote sensinghyper-spectral data transformation and identificationwetland vegetationspectral characteristicseast Dongting lake region