Research On Remote Sensing Image Processing Method For Change Detection And Classification Of Land Use

The widely application of remote sensing technology promotes rapid development of agricultural production and research. Researching on remote sensing image processing method for change detection and classification of land use, can provide technical support for basic work and research, such as the optimization of agricultural production decision and regional planning land management. Based on the previous research, the remote sensing image processing method is deeply studied in this thesis for change detection and classification of land use, which includes image enhancement, image registration, image fusion, land use change detection and classification. The main work is as follows:Fisrtly, a remote sensing image enhancement method based on non-subsampled shearlet transform and parameterized logarithmic image processing(PLIP) model is proposed. A remote sensing image of land use is decomposed into a low frequency component and high frequency components by non-subsampled shearlet transform. Then the low frequency component is enhanced according to PLIP model, while the improved fuzzy enhancement method is used to enhance the high frequency components. Experimental results show that, compared with five kinds of image enhancement methods such as the method based on stationary wavelet transform and the method based on shearlet transform, the proposed method can more effectively improve the contrast of remote sensing image and enhance edges and texture details with better visual effects.Then, a remote sensing image registration algorithm based on dual tree complex wavelet transform and speeded up robust features(SURF) is dicussed. The reference image and the image to be registered are decomposed into the low and high frequency parts by dual tree complex wavelet transform. The selected corresponding low frequency parts serve as the input image of SURF algorithm, the coarse matching results can be obtained. Then the coarse matching point pairs are purified and mismatching point pairs are eliminated by random sample consensus algorithm. Thus the problem of more mismatching point pairs caused by SURF algorithm is solved. Meanwhile, the transform model parameters of optimal matching are calculated. Image registration is completed. Experimental results show that the proposed algorithm has higher speed, correct matching rate and registration accuracy than scale invariant feature transform(SIFT) algorithm and SURF algorithm. It also performs better in resisting to noise, rotation and brightness change.And then, a land use remote sensing image fusion method of multispectral and panchromatic images based on improved projected gradient non-negative matrix factorization(NMF) and improved pulse coupled neural network(PCNN) in non-subsampled shearlet transform(NSST) domain was proposed. Intensity hue saturation(IHS) transform was performed for multispectral image. Then, the intensity component of multispectral image and panchromatic image were decomposed by non-subsampled shearlet transform, respectively. The low frequency image was obtained by fusion of two low frequency coefficients using improved projected gradient non-negative matrix factorization. For the fusion of high frequency sub-band coefficients, improved pulse coupled neural network was adopted. Experimental results show that the proposed method enhance the spatial details in the fused image, while preserving spectral information of the multispectral image. And it is superior to the three fusion methods such as the method based on IHS transform, the method of non-subsampled contourlet transform(NSCT) combined with PCNN, and the method of NSCT combined with NMF in the objective quantitative evaluation indexes such as average gradient and spectral distortions.Subsequently, a change detection algorithm based on shearlet transform and kernel principal component analysis(KPCA) is given. Multi-scale decompositions of remote sensing images are performed by using shearlet transform. Then kernel principal component analysis is carried out on the decomposed data and the image with change information is obtained by inverse shearlet transform. Finally the image is segmented by using fuzzy local information c-means(FLICM) clustering algorithm, thus change detection of remote sensing images is completed. Experimental results show that, compared with the three change detection algorithms such as the algorithm based on principal component analysis(PCA), the algorithm based on KPCA, and the algorithm based on wavelet transform and PCA, the proposed algorithm can effectively separate change information and get the change detection image with higher change detection accuracy.Finally, a classification algorithm of remote sensing image based on Log_gabor wavelet and the spectral information is proposed. Firstly, multi-direction and multi-resolution filtering is performed on a remote sensing image by Log_gabor filter to extract texture features of the remote sensing image. Meanwhile the spectral information of the remote sensing image is got. The remote sensing image is classified according to the extracted feature vectors by random forests. Experimental results show that, compared with recent classification algorithms of remote sensing image such as the algorithm based on Gabor wavelet, the algorithm based on Log_gabor wavelet and the algorithm based on support vector machine, the proposed algorithm has a significant improvement in the subjective visual effect and objective quantitative evaluation index such as classification accuracy.