Research On Effective Band Selection For Main Fruit Trees Species Discrimination Based On Canopy Hyperspectral Data In Southern Xinjiang Basin

Xinjiang fruit industry developed rapidly in the past ten years, so far, the characteristics of fruit areahas more than about20000000mu, fruit industry has become the dominant industry in Xinjiang to solvethe "three rural issues". However, in the fruit industry scale, the process of industrialization, informatizationconstruction of fruit industry lags. Fast, accurately grasp the characteristics of fruit resources layout, scaleand updating basic information has become an urgent requirement for rapid and healthy development ofXinjiang Featured Forestry and fruit industry, while the traditional survey method based on multi spectralremote sensing and ground assisted investigation, not only time-consuming, labor-intensive, and the cycleis long. Extraction of fruit information in hyperspectral remote sensing has the characteristcs of a rapid,accurate and wide range of application. The spectral identification of fruit tree species is the core content ofthis work, has important practical significance in the sustainable management of forest fruit industry in.Research and treatment through the canopy spectral data indicated that, the spectral data transform indifferent ways can improve classification precision of South Xinjiang basin in a certain extent, planted5fruit trees, especially the logarithmic differential transform under5kinds of fruit tree classificationprecision is high, up to84.00%. And in species identification process, spectral data of canopy shaded oradret determination has little influence on the species identification accuracy, can basically reflect theinherent characteristics of different bands of fruit species.Based on the treatment of canopy spectrum dimension reduction, our research showed that in theprocess of smoothing filter value, reasonable step interval selection is a key parameter of fruit speciesclassification accuracy. For the5kinds of fruit tree spectral data classification, the original spectral datamean filter processing step interval is too large or too small are not conducive to the overall classificationaccuracy, smoothing filter value step interval of10nm can be regarded as the best choice of step size, thisspecies the highest classification accuracy, up to97.33%. The effective band of5kinds of fruit treeidentification selected mainly concentrated in the red light, green light and near infrared band.Through the analysis of canopy spectra data, the studies show that the accuracy of speciesidentification can be improved to a great extent by the participation of feature points and featureparameters. Based on the study of fruit tree canopy spectra identification data, blue edge position, yellowedge position, red edge position and near infrared platform4feature points and the blue edge position,yellow edge position, red edge position, near infrared platform4characteristic parameters are the keysensitive wave section of tree species classification, classification accuracy up to86.67%, significantlyhigher than classification accuracy of the species that involved in the whole band, of which the totalaccuracy is72.00%.