越南大陸海岸線遙感智能解譯方法研究
發(fā)布時間：2019-02-19 11:57
【摘要】：海岸線是指平均大潮高潮時的水陸分界線,是地形圖和海圖的基礎(chǔ)要素。受自然因素和人為因素的影響,海岸線不斷變化,快速而準確地監(jiān)測海岸線位置和性質(zhì)的動態(tài)變化,對于基礎(chǔ)地理信息快速更新、資源調(diào)查和海岸帶的科學(xué)管理具有重要意義。本文旨在研究從遙感影像中快速、準確地獲取海岸線信息的方法,在863重大項目課題“南海及其鄰域空間情勢綜合分析與決策模擬”的支持下,以越南大陸海岸線為研究對象,將越南大陸海岸線分為五種類型,使用Landsat8的OLI影像,以影像最優(yōu)分割、影像特征優(yōu)選與影像智能分類技術(shù)為研究基礎(chǔ),構(gòu)建多層次的海岸線智能解譯模型,采用多種機器學(xué)習(xí)方法對越南大陸海岸線的智能解譯展開了深入研究。論文的主要內(nèi)容和創(chuàng)新如下:1.針對影像分割中的尺度選擇問題,提出了基于知識反饋的最優(yōu)分割尺度獲取方法,根據(jù)地物類的尺度判斷特征獲取尺度備選集,在不同類的地物特征差異最大的理論指導(dǎo)下構(gòu)建特征差異判別準則,最后利用特征差異判別準則從尺度備選集中獲取最優(yōu)分割尺度。實驗表明,該方法能夠根據(jù)解譯任務(wù)精確地獲取多個目標地物的最優(yōu)分割尺度。以獲取的最優(yōu)分割尺度為依據(jù),結(jié)合聚類方法分析地物類別的可分性,確定了具體的地物類別和信息提取的層次結(jié)構(gòu)。2.系統(tǒng)地分析和評價了基于OLI影像定義的多種特征�；赑earson樣本相關(guān)系數(shù)評價特征之間的相關(guān)性,使用變異系數(shù)來衡量特征攜帶的信息量,使用單因素方差分析的方法分析特征與目標地類之間的相關(guān)性,通過關(guān)聯(lián)規(guī)則挖掘,得到與10個地物類強相關(guān)的特征并對規(guī)則進行解譯。3.提出了以降低解譯不確定性為目標的特征選擇方法�；诙喾N方法對特征進行優(yōu)選:基于BP神經(jīng)網(wǎng)絡(luò)的權(quán)值進行敏感性分析,通過敏感性系數(shù)的大小對特征的重要性進行排序并進行特征選擇;從減少分類不確定性,增強規(guī)則表達的魯棒性等方面考慮,研究了外界因素和內(nèi)在因素對特征值的影響,外界因素中研究了影像中薄云的影響,內(nèi)在因素研究了影像分割尺度對影像對象特征值的影響;使用因子分析法對原始的高維特征進行特征抽取實現(xiàn)維度歸約。4.對比分析了五種類型常用的機器學(xué)習(xí)算法在海岸地物分類中的應(yīng)用效果,通過一系列實驗得到各分類器的較優(yōu)參數(shù)設(shè)置,構(gòu)建了適用于海岸地物分類的機器學(xué)習(xí)模型,并通過影像分類實驗分析各個分類器對地物類的分類能力,實驗表明,SVM對10個地物類別的分類精度都很高,分類效果最好,SVM和隨機森林兩種分類器的分類結(jié)果互補性強。5.提出了一種使用關(guān)系規(guī)則進行分類的方法,這種規(guī)則通過比較屬性或特征間的關(guān)系實現(xiàn)分類,具有易于理解和魯棒的特點。本文首先構(gòu)造影像對象的關(guān)系特征,然后通過機器學(xué)習(xí)獲得了植被相關(guān)、水體相關(guān)和砂石相關(guān)三個大類的多條關(guān)系規(guī)則,對影像解譯的專家知識進行補充。6.基于獲取的海岸線數(shù)據(jù)進行了多種方式的統(tǒng)計分析,得出的主要結(jié)論有:2013年越南大陸海岸線總長度約為4067km,其中人工岸線占的比例最大、淤泥質(zhì)岸線占的比例最小;人工岸線分布最廣,砂質(zhì)岸線主要分布于越南的南中部,基巖岸線分布于越南中部,紅樹林岸線分布于越南大陸的南北兩頭,淤泥質(zhì)海岸在整條大陸海岸線上零星分布;28個沿海省級行政區(qū)中海岸線最長的是廣寧省,最短的是寧平省,人工岸線最長的是南定省,基巖岸線最長的是慶和省,砂質(zhì)岸線最長的是平順省,淤泥質(zhì)岸線最長的是廣寧省,紅樹林岸線最長的是金甌省。
[Abstract]:The coastline refers to the surface boundary between the average tide and the tide, which is the basic element of the topographic map and the chart. With the influence of natural and human factors, the change of the coastline, the rapid and accurate monitoring of the dynamic changes of the position and the nature of the coastline is of great significance to the rapid updating of the basic geographic information, the resource investigation and the scientific management of the coastal zone. The purpose of this paper is to study the method of fast and accurate acquisition of the coastline information from the remote sensing image. With the support of the comprehensive analysis and decision-making simulation of the South China Sea and its neighborhood space situation of the major project of 863 project, the coast line of Vietnam is divided into five types, and the OLI image of Landsat 8 is used for optimal image segmentation. The image feature is based on the image intelligent classification technology, and a multi-level coastline intelligent interpretation model is constructed, and a variety of machine learning methods are adopted to further study the intelligent interpretation of the coast of the Vietnamese mainland. The main content and innovation of the paper are as follows: 1. aiming at the problem of scale selection in image segmentation, a method for acquiring an optimal segmentation scale based on knowledge feedback is proposed, and finally, the characteristic difference discrimination criterion is utilized to obtain the optimal segmentation scale from the scale alternative set. The experiment shows that the method can accurately acquire the optimal segmentation scale of a plurality of target objects according to the interpretation task. based on the obtained optimal segmentation scale, the classification of the object class and the hierarchical structure of information extraction are determined by combining the classification of the object class in combination with the clustering method. Multiple features based on OLI image definition are systematically analyzed and evaluated. based on the correlation between the characteristics of the Pearson sample correlation coefficient evaluation feature, the coefficient of variation is used to measure the information quantity carried by the characteristic, the correlation between the characteristic and the target land class is analyzed by using a single-factor variance analysis method, and the rules are interpreted according to the characteristics that are strongly related to the 10 ground objects. A feature selection method based on the reduction of the interpretation uncertainty is presented. The characteristics are preferably selected based on a plurality of methods: sensitivity analysis is carried out based on the weight value of the BP neural network, the importance of the characteristics is sorted through the size of the sensitivity coefficient, and the feature selection is performed; and from the aspects of reducing the classification uncertainty, enhancing the robustness of the rule expression, and the like, The influence of external factors and internal factors on the characteristic value is studied, the influence of the thin cloud in the image is studied in the external factors, and the influence of the image segmentation scale on the characteristic value of the image object is studied. The feature extraction of the original high-dimensional features is carried out using the factor analysis method to realize the dimension reduction. In this paper, the application effect of five types of commonly used machine learning algorithms in the classification of the coast features is analyzed, and the optimal parameter setting of each classifier is obtained through a series of experiments, and a machine learning model suitable for the classification of the coast features is constructed. The classification ability of each classifier to the ground object class is analyzed by the image classification experiment. The experiment shows that the classification accuracy of the SVM is very high for the 10 ground objects, and the classification effect is the best. The classification results of the two classifiers of the SVM and the random forest are complementary to each other. This paper presents a method of classification using the relation rules, which can be classified by comparing the relation between the attributes or the features, and it has the characteristics of easy to understand and stick. In this paper, the relationship between the image objects is constructed, and then the relationship rules of the three large classes of vegetation-related, water-related and sandstone are obtained through the machine learning, and the expert knowledge of the image interpretation is supplemented. The main conclusions are as follows: the total length of the coast of Vietnam in 2013 is about 4067km, the proportion of the artificial coastline is the largest, the proportion of the sludge line is the smallest, the distribution of the artificial coastline is the most, the sandy shore line is mainly distributed in the middle of the south part of the Vietnam, the shoreline of the bedrock is distributed in the middle of the Vietnam, the coastline of the mangrove is distributed on the two ends of the north and the south of the Vietnamese mainland, the muddy coast is distributed sporadically on the whole continental coastline, and the longest coastline of the 28 coastal provincial administrative regions is the Guangning province, The shortest is in the province of Nanding, the longest in the artificial line is the province of Nanding, the longest of the bedrock is the Qing and the province, the longest of the sandy shore is the smooth province, the longest is the Guangning province, and the longest of the mangroves is the province of Jinyi.
【學(xué)位授予單位】：解放軍信息工程大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2014
【分類號】：P715.7
，
本文編號：2426479
[Abstract]:The coastline refers to the surface boundary between the average tide and the tide, which is the basic element of the topographic map and the chart. With the influence of natural and human factors, the change of the coastline, the rapid and accurate monitoring of the dynamic changes of the position and the nature of the coastline is of great significance to the rapid updating of the basic geographic information, the resource investigation and the scientific management of the coastal zone. The purpose of this paper is to study the method of fast and accurate acquisition of the coastline information from the remote sensing image. With the support of the comprehensive analysis and decision-making simulation of the South China Sea and its neighborhood space situation of the major project of 863 project, the coast line of Vietnam is divided into five types, and the OLI image of Landsat 8 is used for optimal image segmentation. The image feature is based on the image intelligent classification technology, and a multi-level coastline intelligent interpretation model is constructed, and a variety of machine learning methods are adopted to further study the intelligent interpretation of the coast of the Vietnamese mainland. The main content and innovation of the paper are as follows: 1. aiming at the problem of scale selection in image segmentation, a method for acquiring an optimal segmentation scale based on knowledge feedback is proposed, and finally, the characteristic difference discrimination criterion is utilized to obtain the optimal segmentation scale from the scale alternative set. The experiment shows that the method can accurately acquire the optimal segmentation scale of a plurality of target objects according to the interpretation task. based on the obtained optimal segmentation scale, the classification of the object class and the hierarchical structure of information extraction are determined by combining the classification of the object class in combination with the clustering method. Multiple features based on OLI image definition are systematically analyzed and evaluated. based on the correlation between the characteristics of the Pearson sample correlation coefficient evaluation feature, the coefficient of variation is used to measure the information quantity carried by the characteristic, the correlation between the characteristic and the target land class is analyzed by using a single-factor variance analysis method, and the rules are interpreted according to the characteristics that are strongly related to the 10 ground objects. A feature selection method based on the reduction of the interpretation uncertainty is presented. The characteristics are preferably selected based on a plurality of methods: sensitivity analysis is carried out based on the weight value of the BP neural network, the importance of the characteristics is sorted through the size of the sensitivity coefficient, and the feature selection is performed; and from the aspects of reducing the classification uncertainty, enhancing the robustness of the rule expression, and the like, The influence of external factors and internal factors on the characteristic value is studied, the influence of the thin cloud in the image is studied in the external factors, and the influence of the image segmentation scale on the characteristic value of the image object is studied. The feature extraction of the original high-dimensional features is carried out using the factor analysis method to realize the dimension reduction. In this paper, the application effect of five types of commonly used machine learning algorithms in the classification of the coast features is analyzed, and the optimal parameter setting of each classifier is obtained through a series of experiments, and a machine learning model suitable for the classification of the coast features is constructed. The classification ability of each classifier to the ground object class is analyzed by the image classification experiment. The experiment shows that the classification accuracy of the SVM is very high for the 10 ground objects, and the classification effect is the best. The classification results of the two classifiers of the SVM and the random forest are complementary to each other. This paper presents a method of classification using the relation rules, which can be classified by comparing the relation between the attributes or the features, and it has the characteristics of easy to understand and stick. In this paper, the relationship between the image objects is constructed, and then the relationship rules of the three large classes of vegetation-related, water-related and sandstone are obtained through the machine learning, and the expert knowledge of the image interpretation is supplemented. The main conclusions are as follows: the total length of the coast of Vietnam in 2013 is about 4067km, the proportion of the artificial coastline is the largest, the proportion of the sludge line is the smallest, the distribution of the artificial coastline is the most, the sandy shore line is mainly distributed in the middle of the south part of the Vietnam, the shoreline of the bedrock is distributed in the middle of the Vietnam, the coastline of the mangrove is distributed on the two ends of the north and the south of the Vietnamese mainland, the muddy coast is distributed sporadically on the whole continental coastline, and the longest coastline of the 28 coastal provincial administrative regions is the Guangning province, The shortest is in the province of Nanding, the longest in the artificial line is the province of Nanding, the longest of the bedrock is the Qing and the province, the longest of the sandy shore is the smooth province, the longest is the Guangning province, and the longest of the mangroves is the province of Jinyi.
【學(xué)位授予單位】：解放軍信息工程大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2014
【分類號】：P715.7
，
本文編號：2426479
本文鏈接：http://sikaile.net/kejilunwen/haiyang/2426479.html
最近更新
教材專著
