Fusion And Segmentation Of Sparse Data And Its Research In Land Use Classification

Land use plays an important guiding role in our daily life.We can analyze land use to plan land resources and protect the ecological environment.As the situation of global land use is constantly changing and the development of remote sensing technology is incomparable,it is necessary to obtain more accurate land use information of multispectral remote sensing images.So far,most sensors still have difficulty in meeting the requirements of land use classification research in terms of resolution,so improving the accuracy of land use classification is still a major trend and direction for scholars at home and abroad.This article takes Shizuishan City as a research area,and establishes a land use classification system that conforms to the basic situation of Shizuishan according to the land use classification system of the US Geological Survey and the Chinese Academy of Sciences and the natural and human characteristics of Shizuishan City.The atmospheric correction and principal component analysis algorithms are used to reduce the dimension of the remote sensing image,so that the dimension-reduced image can better reflect the terrain features such as forest land and grassland.Using a discrete wavelet transform(DWT)algorithm for gradient sharpening and a K-means model segmentation algorithm based on the optimal weighting method to process the image of the experimental area,providing high-quality data sets for land use classification,making The classification accuracy has been improved.Firstly,to solve the problem of missing edge information in the fusion process of multispectral images and panchromatic images,a discrete wavelet transform(DWT)algorithm based on gradient sharpening is proposed.By fusing panchromatic and multi-spectral images,the algorithm makes the spectral information more complete and provides a good foundation for the creation of data sets for land use classification below.Through the analysis of the results and comparison with other algorithms,it is found that the method proposed in this paper retains the edge details more completely and achieves better results.Secondly,aiming at the problems that traditional K-means algorithm is not easy to obtain the best quality center and tends to local optimization,a method of K-means model based on optimal weight method for remote sensing image segmentation is proposed.This method effectively solves the problems of local optimality and initial clustering center,improves the segmentation accuracy of the image,and lays a good foundation for creating label data sets below.Comparing the proposed algorithm with the traditional K-means segmentation algorithm and GA segmentation algorithm,it is found that the proposed algorithm has a more obvious effect on remote sensing image segmentation.Finally,the gradient-sharp fusion algorithm proposed in this paper and the k-means algorithm based on adaptive weights are used to select the remote sensing images to make a neural network training data set.The U-net network,maximum likelihood method,and support vector machine were used to compare the dataset processed by the improved algorithm with the original dataset.After comparison,it was found that compared with the traditional supervised classification method,the deep learning method has higher classification accuracy.The classification accuracy is improved by 20%and 7%respectively.Compared with the original data,the image classification accuracy processed by the improved algorithm is higher,and the accuracy of the three classification models is improved by 2.5%,3.6%,and 8%,respectively.