Study On Forest Vegetation Information Extraction Method Based On ZY-3 Satellite Images In Hilly Area Of East Sichuan

The high-resolution remote sensing satellite(ZY-3)made in China,has a broad application prospects in China's forest resources survey because of its prominent priceadvantage.In this paper,we mainly use ENVI5.3 remote sensing image processing software,based on the ZY-3 winter image of Yuechi County,Sichuan Province,using the CART decision tree method based on ISODATA algorithm and the objectoriented K-Nearest Neighbor method,extract the forest Vegetation information,and compared the results of the two methods of extraction,analysis of the two methods in the hilly area of eastern Sichuan forest resources survey work applicability.The main conclusions are as follows:(1)CART decision tree based on ISODATA algorithm tree forest vegetation information extraction method,shrub vegetation information extraction producer accuracy is 99.46%,the user accuracy is about 49.69%,the precision of orchard information extraction is 98.20%and the user precision is 72.30%.The K-nearest neighbor object oriented method method for forest vegetation information extraction,the producer accuracy of the extraction information of shrub vegetation information is 64.01%,and the user accuracy is about 11.77%,and the producer precision and user precision of the economic forest vegetation information were all 0.The results showed that the CART decision tree method based on ISODATA algorithm was superior to the object-oriented K-nearest neighbor method in extracting the information of shrub and economic forest vegetation,which could solve the problem of shrub and economic forest in ZY-3 winter And more difficult to distinguish between the problem.(2)CART decision tree method based on ISODATA algorithm,the total accuracy of vegetation information is 92.70%,Kappa coefficient is 0.9073;object oriented K-nearest neighbor method,the total accuracy of vegetation information is 67.02%,Kappa coefficient is 0.6072.The results show that the CART decision tree method based on ISODATA algorithm is superior to the object-oriented K-nearest neighbor method,and has strong applicability in the hilly area similar to the terrain and vegetation type in Yuechi County.(3)In this paper,Gram-Schmidt transform,principal component(PC)transform,Color normalized(Brovery)transformation and NNDiffus method are used to fuse the ZY-3 full-color image and multi-spectral image in the study area.The results show that the NNDiffus method,The better retain the original image of the spectral information and texture information,the fusion of the image is rich in color,clear contours,higher recognition,suitable for ZY-3 image processing.(4)In the object-oriented K-Nearest Neighbor Forest Vegetation Information Extraction Test,the superior scale was obtained by screening the scale and the merging scale.The results show that the segmentation method of ZY-3 image in hilly area is 40,and the scale of merging is about 80,which can distinguish the different regions and meet the needs of forestry production.