Urban Tree Species Discrimination Based On Leaf And Canopy Level Hyperspectral Data

we selected 12 representative tree species as experimental subjects in central China, including one kind of herbage ground cover, three kinds of coniferous species, and eight kinds of deciduous species, obtaind their leaf and canopy hyperspectral data measuring by the field spectrometer.we analyzed the differences and similarities between the various vegetation hyperspectral data, from the spectral characteristic reflected in the spectral information of various tree species. According to the characteristic of vegetation spectrum, which is different from other ground object, we extracted the spectral characteristic positions and parameters, then classified the tree species based on them. Detailed Results are as follows:(1) Discovered from the leaf and canopy spectral curves of different tree species, it showed that there have the same trend, but still a fine distinction in part.(2) The distribution of extracted characteristic positions is extremely concentrated, respectively, blue-violet band absorption peak M:407nm; blue edge B:525nm; green peak G:560nm; yellow edge Y:575nm; red valley R:674nm; red edge V:721nm; near-infrared platform I:750nm.(3) Although the system cluster analysis could discriminate the different tree species well in large categories, there will esist varying degrees of confusion in further subdivision of tree species.(4) The comparison result of different classification distance algorithms tell us:the best method is Mahalanobis distance classification, Fisher classification second best, classification based on Bayesian theory the third. All these three classifications can discriminate the ground clover, however, there are some varying degrees of confusion to the other tree species.(5) The classification effect of system cluster analysis with SPSS software is better when we discriminate the large categories, but the workload will much more if we require higher accuracy and more samples in the further subdivision; when we using confusion matrix datas calculation with MATLAB software, the effect of Mahalanobis distance classification which mainly aimed at fine classification of tree species is perfect. That is to say, using system cluster analysis is better when discriminating the large categories; While there are many different tree species, and many different samples of the same tree species, using Mahalanobis distance classification is the best choice.