Research On Water Extraction Technology Of Domestic Surveying And Mapping Satellite Imagery Based On Multispectral Information

In the production process of aerospace digital photogrammetry,the dense matching accuracy of water pixels in the images is low and the reliability is not strong.Pre-extracting the water body in the image will improve the reliability and efficiency in fully automatic production.Panchromatic images are the main input data in the production of digital photogrammetry,but it does not contain multispectral information of water body.The multispectral images with the same source and the same scene(usually low spectral resolution,that is,the number of imaging spectral bands are few)contain the spectral information of the water body,but they are generally not directly used in the production process.Using the spectral information of the water body in the multispectral image to extract the water body,the water body in the panchromatic image registered with it can be extracted indirectly.Based on this background,the water extraction algorithm for low-spectral resolution images is studied.Mainly,discrete particle swarm water extraction algorithm based on digital number matching,improved digital number matching based on support vector machine,the relative accuracy and absolute accuracy of the improved algorithm and the comparison of different water extraction methods were discussed.The main work and innovations of the thesis are as follows:1.It summarizes the existing methods for extracting water bodies with high maturity and extensive use,and analyzes the deficiencies in extracting water bodies with low spectral resolution images.2.Combining the characteristics of multispectral imagery imaging bands and using the digital number matching algorithm to measure the spectral similarity,a discrete particle swarm water extraction algorithm is designed.The algorithm quantitatively describes the spectral similarity by calculating the Euclidean distance and the angle of the cosine.Then the objective function is constructed,and the optimal solution of the objective function value is iteratively solved through discrete particle swarm optimization.The experimental results show that the discrete particle swarm optimization algorithm based on digital number matching can effectively extract water in some multispectral images.3.Due to the digital number matching algorithm is not sensitive to the shape of the spectral curve,an algorithm based on support vector machine is used for improvement.According to the grid search method and cross-validation method,the optimized parameters are used to train the support vector machine.The experimental results show that the classification results based on support vector machines have higher absolute accuracy and are more sensitive to the shape of the spectral curve.4.A discrete particle swarm water extraction algorithm based on joint support vector machine is proposed.The smoothing function is used to map the results of the support vector machine to the probability interval,and the water body probability of the pixel is obtained,and the classification result is obtained by the discrete particle swarm optimization algorithm.The comparison experiment results of water body extraction algorithm show that for various types of water body images,the discrete particle swarm optimization algorithm based on support vector machine is superior to the normalized water body index method and its stability is stronger than the HSV color transformation method.This method has wide applicability.