Research On Vietnam Mainland Coastline Intelligence Interpretation With Remote Sensing Images

The coastline is the dividing line between land and sea where the average high tide could arrive.It is the basic element of a map or chart.The coastline affected by natural and human factors is changing constantly.So we need to acquire the information about coastline rapidly and accurately,which is especially essential for the basic geographic information updating,resource surveying and scientific management of coastal zones.This paper aimed to study a method of obtaining the coastline information from remote sensing images quickly and accurately.This paper is supported by 863 major project —comprehensive analysis and decision-making simulation about the South China Sea and its spatial neighborhood.The study object of the paper is Vietnamese mainland coastline,which was divided into five types.OLI images of Landsat8 were studied for the coastline obtaining work.The coastline obtaining method was studied in the basis of image segmentation,features selecting and intelligence classification.A multi-level intelligence interpretation model of coastline was established and used to guide the research work in this paper.A variety of machine learning methods for intelligence interpretation of the mainland coastline of Vietnam were used in the study.The main works and innovations are as the following:1.A knowledge-based method of segmentation scale selecting was put forward.It could get the optimal segmentation scale with following steps: selection scales were gotten by scale judgment features firstly,and then the optimal segmentation scale was obtained from selection scales by feature difference criterion.Experiments showed that the method could obtain the optimal segmentation scale of multiple objects accurately.At last,cluster analysis methods were used to analyze the separability of defined classes and a hierarchy structure was established to extract the specific categories and feature information.2.A variety of features based on OLI images were analyzed and evaluated systematically.Correlation coefficient of Pearson was used to measure the extent of correlation.Variation coefficient was used to measure the amount of information carried by features.One-way analysis of variance was used to measure the correlation between features and defined classes.By association rule mining,strong related features were obtained and the rules were interpreted.3.Feature selecting approaches were proposed to reduce the uncertainty of interpretation process.Several methods were studied to control the amount and quality of classification features.The sensitivity analysis was adopted to sort the features based on BP neural network weights.In order to reduce classification uncertainties and enhance robustness of rule expression,the influence of external factors and internal factors on the feature values were considered.The impact of thin cloud in the image was regarded as one of external factors.The segmentation scale was regarded as one of internal factors.At last,a method of factor analysis was used to extract new features to achieve the feature dimension reduction.4.Five common machine learning algorithms were used as classifiers and their results of classification were compared and analyzed.A series of experiments were made to obtain the optimum parameters for each classifier.And a suitable machine learning model was constructed for classification of coastal features.The experiments showed that: was the most sitable clasifier for the study,the classification results of SVM and random forest were highly complementary.5.A method of classifying image objects with feature relation rules was put forward.The rules could classify the image objects by comparing the properties or characteristics of classes.The relation rules are not only easy to understand but also robust.Firstly,the relation features were structured,and then the relation rules of three defined classes were gained by machine learning.These rules got from machine learning could be used to complement the knowledge of experts.6.The statistical data were analyzed in many ways based on the obtained coastline of Vietnam in 2013.The main conclusions were as follows: the total length of Vietnamese mainland coastline was about 4067 km in 2013,which the artificial coastline was the longest,while muddy coastline was the shortest.Artificial coastline distributed the most widely,sandy coastline mainly distributed in the southern center mainland,bedrock coastline located in center of the mainland,mangrove coastline distributed in the north and south ends of the country,muddy coastline scattered on the entire continent.Quang Ninh province had the longest coastline and Ninh Binh province had the shortest coastline in 28 coastal provincial administrative regions.Nam Dinh Province has the longest artificial coastline,Khanh Hoa province has the longest bedrock coastline,Binh Thuan Province has the longest sandy coastline,Quang Ninh province has the longest muddy coastline,Ca Mau Province has the longest mangrove coastline.