Texture Characteristic Extraction Of Remote Sensing Image And Its Application In Image Classfication

Remote sensing image is rich in texture information. remote sensing image texture analysis has become an important means of improving the accuracy of image classification. Accurate texture characteristic extraction is the key to image classification.At first this paper introduces the various methods of texture expression, in accordance with previous studies and the combination of remote sensing image texture features ,which is the randomness of part and statistics of the overall, the gray co-occurrence matrix method is adopted to describe the image texture.On the other hand, in allusion to related and nonlinear characteristics of remote sensing images texture parameters, based on the strongpoint of artificial neural networks which has the capacity of the human brain to dispose complex non-linear information, the paper introduces BP neural network as the classifiers.The difficulty of Gray co-occurrence matrix lies in that the different parameters produces different matrix. This paper analyzes how to choose parameters when calculating Gray co-occurrence matrix, and determines the pixel spacing of this paper. In addition, multi-scale texture is also discussed.Another feature of remote sensing images is multi-band. At present, the texture analysis is basically of the single-band.In this paper, multi-band Texture feature extraction was discussed.A method based on independent component analysis to reduce the dimension of texture is proposed on dropping dimensions of the multi-band Texture eigenvalue. so that not only the eigenvectors associated properties is removed, but also independent features is gained . On this basis a comparison with the traditional means and a principal component analysis is shown.Finally, this paper combines the gray co-occurrence matrix with wavelet decomposition to the extracted image texture and texture classification. Evaluation is given out.