The Recognition Of Green Tree In Remote Sensing Image Of City Based On Support Vector Machine And Texture

Urban remote sensing research dated from the1980s in our country. There are some integrated investigations on the remote sensing image at different big cities, so as to promote the ecological development of cities. The city of Guilin famous for its landscape all around the world.Its ecological function of the urban green area is the conservation of water and soil, to keep the Lijiang River alive. It not only has relation to the pillar industry of tourism, but also provides a comfortable environment of life to local people in culture. As the most effective method of statistical learning theory, support vector machine (SVM) has the advantages of studying with small sample, nonlinearity and solving the problem of dimension disaster etc. So it becomes a hot topic in the field of research of remote sensing image recognition. This paper, bases on the statistics texture of remote sensing image with the statistical learning theory, gets research on the mean of texture extraction, the kernel function of SVM, the parameters of kernel function and so on, and then applies them on the remote sensing image to estimate the area of green tree on image of target.In the condition of mobile window, the paper uses gray symbiotic matrix to get the texture information of green band of remote sensing image, which is combined with spectrum information of grey value of RGB as the input vector of the SVM. By training through the genetic algorithm, the paper works out the kernel function parameters c and g of SVM, so as to make SVM training with the optimal parameters. Finally, it estimates the area of green tree on remote sensing image. Between the four commonly used kernel function of SVM, gaussian kernel function and polynomial kernel function have better effect to estimate the area of green tree, and in some degree distinguish with green tree and artificial turf, natural grassland and shadow, which do not belong to green tree.In the condition of mobile window, the paper uses gray symbiotic matrix to get the texture information of three band of RGB on remote sensing image, and then combines it with spectrum information as the input vector of the SVM. By training through the genetic algorithm, the paper works out the kernel function parameters c and g of SVM, so as to make SVM training with the optimal parameters. Finally, it estimates the area of green tree on remote sensing image. Between the four kernel function, linear kernel function and sigdom kernel function have not obvious effect to make the difference between artificial turf, natural green grass and shadow. Gaussian kernel function and polynomial kernel function can distinguish with green tree and artificial turf, natural grassland, which do not belong to green tree. However, because of the rugged terrain, the texture of natural lawn has a small amount of confusion with the texture of the green tree. Polynomial kernel function has better effect than gaussian kernel function to recognize artificial turf and natural green grass. Gaussian kernel function has better effect than polynomial kernel function to deal with the tree and not-tree under shadow. It also has better effect to deal with the tree and not-tree under shadow than gaussian kernel function and polynomial kernel function with the texture feature of green band.