The Research Of Typical Vegetation Spectral Characteristics And Channel In Hangzhou

The research object are the47common species in Hangzhou, included two kinds of herbaceous, fivekinds of herbaceous plants, nine kinds of coniferous woody plants,31kinds of broad-leaved woodyplants, it used ASD FieldSpecPro FR field spectroradiometer to measure the spectral reflectance of thesample data. By the spectral characteristics of different vegetation, we can analysis the difference ofdifferent vegetation spectrum. Based on the cluster analysis and neural network, this article analyzedthe spectral reflectance data of different samples to find the best method of identification. Usingcorrelation analysis divided the channel intervals, using the formula of spectral dimensionality reductionto obtain the spectral curve of dimensionality reduction, and then using the best method of identificationto classify and recognize the original spectral curves and compressed spectral curves. The results cantest the reasonableness of the divided channel. Detailed results are as follows:(1) From the47different types of spectral reflectance curves, its showed that there have the sametrends, but has a fine distinction in part.(2) In cluster analysis, the classification results of average class is reasonabler than the shortestdistance method and the center of gravity method. In neural network classification, the self-organizingcompetitive neural network is the best method. Compared with the average class and self-organizingcompetitive neural network, the class average method can distinguish the vegetations on the largecategories, but when carried further subdivision, the classification result of self-organizing competitiveneural network is better than average class. Generally speaking, the classification result ofself-organizing competitive neural network is the best.(3) By the divided channel interval, it can be seen11width of the channels are narrow, it canidentify differences between vegetation effectively. Among them,6-channel located within the red band,4-channel located in the near-infrared bands, it explained that the bands which can effectively identifyand distinguish different vegetation are mostly in the red band and near infrared bands.(4) The spectral data of23channels retained the difference with less data between differentvegetation between400-2500nm, it reduced the redundancy of data, and achieved the dimensionalityreduction of date. It provides a kind of reference for remote sensing data.