面向对象的天然绿洲与人工绿洲区分——以民勤县湖区绿洲为例

以地处河西走廊东端、石羊河下游的民勤县湖区绿洲为例,以Landsat 8 OLI影像为数据源,从天然绿洲和人工绿洲的基本概念出发,在影像数据预处理、多尺度分割的基础上,综合考虑光谱、纹理、形状、上下文等信息,引入NDVI、最大化差异、紧致度、形状指数和空间邻接关系等多个特征,构建规则集进行天然绿洲和人工绿洲的区分,并将区分结果与基于最大似然法监督分类的绿洲区分结果进行比较分析。结果表明:使用面向对象的影像分析方法区分天然绿洲和人工绿洲的总体精度达到了91.75%,Kappa系数为0.65;较之面向像元的最大似然法监督分类结果,总体精度提高了10.40%,Kappa系数提高了0.13,其中人工绿洲条件Kappa系数提高了0.19,天然绿洲条件Kappa系数提高了0.30。面向对象的影像分析方法能够在一定程度上克服单一光谱特征分类方法的局限性,避免"异物同谱"和"同物异谱"现象带来的混淆,提高天然绿洲和人工绿洲区分的精度。     

基于高分辨遥感数据的胡杨与柽柳树冠提取

胡杨、柽柳是干旱荒漠区生境的指示种,其树冠提取是荒漠生境遥感定量监测的基础。以塔里木河下游胡杨、柽柳为研究对象,基于QuickBird数据,使用光谱单数据源SVM、光谱结合纹理SVM、面向对象分类和最大似然分类法提取树冠。结果表明:1光谱结合纹理SVM比光谱单源SVM分类精度高9.65%,冠幅估测精度高7.18%,表明高分辨影像上纹理是提高分类精度的重要因素;2面向对象分类法精度最高,分类总体精度86.47%,较光谱单源SVM提高15.67%,较光谱结合纹理SVM提高6.02%,较最大似然法提高22.58%,其冠幅估测精度达87.45%。它兼顾面向对象影像分割与支持向量机方法优点,有效利用分割对象光谱、纹理和空间等信息,较好地解决了其他方法"同物异谱、异物同谱"造成提取树冠破碎的问题,使树冠提取具有较好的稳定性和较高精度。     

基于多时相Landsat 8 OLI影像的农作物遥感分类研究

时序遥感数据及地物细微光谱特征对于提取作物分布有重要作用,基于此,利用多时相Landsat 8 OLI影像,结合光谱角填图和决策树分类提取大同市新荣区东部地区主要农作物分布情况,并与最大似然法提取的分布结果进行对比。研究发现:①研究区内春玉米、谷物、大豆和马铃薯种植面积依次减小并呈镶嵌式分布;②结合光谱角填图与决策树分类总体精度为85.34%,Kappa系数为0.76,与最大似然法结果相比,总体精度提高22.51%,Kappa系数增加0.31,分类结果与实际作物分布具有更好的一致性;③利用时序遥感影像进行作物分类的精度明显高于单时相遥感影像的分类精度,且从光谱角差异的角度分析时序数据可有效削弱中高分辨率影像物谱不一致现象的影响。研究结果验证了多时相遥感影像对农作物分类研究的积极作用,并发展了光谱角填图法结合决策树分类在中高分辨率遥感影像中进行农作物分类的用法,具有一定的应用前景。     

基于航空高光谱影像的额济纳绿洲土地覆被提取

土地覆被分类是生态环境评价、植被变化分析以及区域生态水文过程研究的基础。航空高光谱遥感具有高机动、高空间分辨率和高光谱分辨率等特点,在土地覆被提取方面极具优势。以黑河下游机载高光谱遥感数据为基础,针对额济纳旗胡杨林国家级自然保护区植被单一、景观破碎和异质性强的景观特点,以及高光谱数据量大、冗余度高等数据特点,对比分析最小噪声变换与主成分分析两种降维方法,最大似然法、支持向量机与面向对象3种监督分类方法。依据研究结果,首先利用NDVI区分高光谱遥感数据中的植被与非植被类别,然后采用最小噪声变换分别进行降维处理,最后利用最大似然法对研究区内土地覆被类型进行分类提取,提取结果聚类处理。依据随机验证点结合地面调查数据和正射影像,对土地覆被分类结果进行精度验证,总体精度和Kappa系数分别为87.95%和0.855,表明分类结果精度高,能够为生态研究等提供有效数据。     

基于纹理信息CART决策树的林芝县森林植被面向对象分类

以西藏自治区林芝县的Landsat-8影像、地形图为信息源,结合样地调查数据及森林资源二类调查数据,研究基于纹理信息的CART决策树的面向对象分类对研究区内的森林地物类别进行提取,分类的总体精度和Kappa系数分别为82.53%和0.768,相较于不利用纹理信息的决策树分类和基于最大似然分类法的研究区地物类别的提取总体精度均高近10%,Kappa系数分别高0.12和0.111。结果表明:基于纹理信息的CART决策树面向对象分类方法对研究区Landsat-8影像进行植被类型提取,分类结果较好,能够满足研究要求。     

基于面向对象的多源卫星遥感影像玉米倒伏面积提取

风灾引起的玉米倒伏可能导致玉米大量减产,利用遥感技术准确监测玉米倒伏面积与空间分布信息对灾情的评估非常重要。利用Planet和Sentinel-2影像分别结合面向对象与基于像元方法提取研究区玉米倒伏,同时评估了不同影像特征（光谱特征、植被指数和纹理特征）与不同分类方法（支持向量机法SVM、随机森林法RF和最大似然法MLC）对玉米倒伏提取精度的影响。结果表明：①使用高空间分辨率的Planet影像进行玉米倒伏提取的精度普遍高于Sentinel-2影像;②从分类精度和面积精度来看,Planet影像的光谱特征+植被指数+均值特征结合面向对象RF分类,总体精度和Kappa系数分别为93.77%和0.87,面积的平均误差最低为4.76%;③采用Planet和Sentinel-2影像结合面向对象分类提取玉米倒伏精度高于基于像元分类。研究不仅分析了面向对象方法的优势,还评估了使用不用影像数据结合面向对象方法的适用性,可以为遥感提取作物倒伏相关研究提供一定的借鉴。     

新疆干旱区绿洲土壤盐渍化信息提取对比研究

在遥感影像分类的过程中非光谱特征起着重要的辅助作用。纹理特征作为一种重要的非光谱特征对于遥感影像分类精度的提高也有很重要的作用。以渭干河-库车河三角洲绿洲为例,利用ETM+数据,探讨了该绿洲盐渍化土地覆盖信息的提取方法。提出了基于SVM的光谱和纹理两种信息复合的分类方法,通过此方法对该绿洲进行分类研究,并将分类结果与最小距离法、最大似然法(MLC)、神经网络法(Neural net)和单源数据(光谱)SVM分类结果进行定性和定量比较分析。研究结果表明：该方法能够有效地解决单数据源分类效果破碎、分类精度不高等问题,并对高维输入向量具有较高的推广能力。总精度达到93.179 5%,比单源信息的SVM分类法提高了3.161 8%,比最大似然法提高了4.825 2%,比神经网络法提高了7.475 6%,而与最小距离法相比,总精度甚至提高了11.102 9%,取得了良好的效果。与传统的分类方法的比较表明,文中所提出的分类方法具有明显的优越性和良好的前景,因此该方法更适合于遥感图像分类和盐渍化信息提取,是地物遥感信息提取的有效途径。     

基于支持向量机的遥感影像湿地信息提取研究*

以ETM 影像数据为例,采用基于支持向量机的方法对黄河中上游区域湿地信息进行提取,并将该方法与传统的最大似然分类提取方法以及面向对象的提取方法进行对比分析。结果表明:基于支持向量机方法的提取精度高达93.57%,Kappa系数也超过了0.9,比单纯的最大似然分类方法或者面向对象的方法提取精度高得多,而且该方法操作性和实用性也很强。     

基于多时相遥感数据的东亚飞蝗生境分类研究

东亚飞蝗的发生及成灾与其存在的生境有十分密切的关系,因此对东亚飞蝗生境进行分类是有效防治蝗灾的基础。以河北省渤海新区为研究区,利用环境减灾小卫星多时相CCD数据,采用4种组合方案,分别使用最大似然法和RuleGen决策树进行了东亚飞蝗生境的遥感分类。结果表明:最大似然法和决策树分类方法总体精度相差不大,但在类别数据较多时,决策树分类方法的执行效率变低。利用5月20日单时相的光谱数据进行分类的总体精度仅有76.43%,Kappa系数〖JP2〗0.7396;加入NDVI时间序列信息后,总体分类精度可以达到93.93%,Kappa系数0.9323。因此,使用多时相信息可以较好地解决异物同谱问题,降低混合像元带来的影响,提高生境分类的精度。〖JP〗     

联合快舟一号影像纹理信息的城市土地覆盖分类

仅依靠光谱信息无法满足高分辨率遥感分类的应用需求,辅之以纹理特征信息进行分类,可提高影像分类精度。利用KZ\|1卫星影像和Landsat\|8卫星影像数据,基于面向对象的影像分割法和灰度共生矩阵纹理分析法对新疆石河子市局部城区进行了地表覆盖分类实验,将不同空间分辨率的全色影像纹理信息、光谱信息构成多种影像特征组合进行分类比较研究,以选择最佳的分类特征集。结果表明:KZ-1影像能为城市区域的土地覆盖分类提供丰富的纹理信息,面向对象的影像分割可较好地利用高分辨率数据的几何结构信息实现优化的影像分割,从而提高多光谱影像的分类精度,总体分类精度为90.06%,Kappa系数为87.93%,比单纯利用光谱信息分类的总体精度提高了8.02%,Kappa系数提高了9.65%,表明KZ\|1数据可为光谱分类提供丰富的纹理信息,从而提高城市区域的土地覆盖分类精度。     

Object-oriented Natural and Artificial Oasis Distinguishing in Landsat Imagery: Taking Minqin Oasis as an Example

Taking Minqin Oasis in the downstream area of the Shiyang River Basin which is located in the east of Hexi Corridor as an example, the Landsat 8 OLI image was chosen as the data source. Under the consideration of the basic concept of the artificial oasis and natural oasis in this paper, combining with the information of the spectrum, texture, shape and context basing on the image data preprocessing and multi-scale segmentation, we introduce a series of indexes such as NDVI、maximum difference, compactness, shape index, the space adjacency relation and so on to construct a rule set for distinguish between natural oasis and artificial oasis. The obtained results were further compared with the results based on the maximum likelihood method. As a result, the total accuracy of using the object-oriented image analysis method to distinguishing between natural oasis and artificial oasis is 91.75%, and the Kappa coefficient is 0.65 by using the rule set established in this paper. Compared with the results based on the maximum likelihood method, the overall accuracy is improved by 10.40% and the Kappa coefficient is 0.13. The Kappa coefficient of the artificial oasis is increased by 0.19, and the Kappa coefficient of the natural oasis condition is increased by 0.30. The results showed that the object-oriented image analysis method can overcome the limitations of the classification method that only using spectral feature to a certain extent, avoid the confusion caused by the phenomenon of “same object with different spectrums” and “same spectrum with different objects”, and increase the accuracy of distinguishing between the artificial oasis and natural oasis.     

Land Cover Extraction in the Ejina Oasis by Hyperspectral Remote Sensing

The extraction of land surface coverage is the basis of ecological environment evaluation,vegetation change analysis and regional ecological and hydrological processes.Aerial hyperspectral remote sensing has great advantage in land surface coverage extraction,such as flexible,wide coverage,high spatial resolution and high spectral resolution.Research area has landscape characteristics of vegetation,landscape fragmentation and heterogeneity in Ejina Poplar Forest National Nature Reserve.Comparison and analysis of two methods of dimension reduction based on minimum noise transform and principal component analysis,three supervised classification methods based on maximum likelihood method,support vector machine and object\|oriented classification.Land surface coverage is extracted by NDVI threshold segmentation,minimum noise transform dimensionality reduction method and maximum likelihood classification method according to the characteristics of landscape fragmentation,heterogeneity and high redundancy of hyperspectral data based on the Airborne Hyperspectral Data of Ejina oasis in the lower reaches of Heihe.The land surface coverage results overall accuracy and Kappa coefficient are 87.95% and 0.885 by random sampling based on airborne remote sensing data.The results show that the classification results of high accuracy can provide effective parameters for ecological research.     

A Remote Sensing Classification Methods of Coastal Wetlands Considering the Distance to Main Water

Coastal wetlands is complex,"different objects with the same spectrum" is serious in the remote sensing image,so the classification accuracy only based on spectral information is low.For this issue,based on the coastal zone wetland's spatial distribution rule,this paper established two kinds of distance layers,distance to coastline layer and distance to river layer,which applied maximum likelihood method and decision tree method,and developed a coastal wetland remote sensing information extraction methods,taking Sheyang County,Jiangsu Province for example.The developed methods highly improved the classification accuracy with the overall classification accuracy of 81.5%,and Kappa of 0.79.The maximum likelihood supervised classification method classification accuracy was lower with overall classification accuracy of 62.3%,and Kappa of 0.60.     

基于高分二号数据的面向对象城市土地利用分类研究

It is of great significance to study the method of extracting urban features from GF-2 remote sensing data.Taking the urban area of Jixi City as the study area,and the GF-2 image is used as the data source.The image is divided into multiple scales,the classification rules of the corresponding objects are established,and the object-based classification method of the rule set is used to classify the objects.Compare with SVM supervised classification results.The results show that the overall accuracy of object-oriented classification is 92.52%,and the Kappa coefficient is 0.91,which is significantly higher than the SVM supervised classification.Using the object-oriented classification method to classify the GF-2 image is better and the precision is higher.Object-oriented classification method based on GF-2 data is an effective method for extracting urban land use classification.     

面向对象高分辨遥感影像分类研究

高空间分辨率遥感影像采用传统基于像元分类方法精度较低,本文通过分析高分辨遥感影像特征,采用面向对象的最近邻监督分类方法对QuickBird影像进行分类研究,首先对影像进行对象分割,然后将分割对象信息、形状特征与及上下文联系等特征构成特征空间进行最近邻监督分类,并与传统的基于像元最近邻分类方法分类进行比较分析,结果表明,本方法能够较好的识别高分辨率地物类型,总精度为92.19%,Kappa系数为0.8835,较好地改善分类效果,适合高分辨遥感影像分类。     

基于高分二号数据的面向对象城市土地利用分类研究

It is of great significance to study the method of extracting urban features from GF-2 remote sensing data.Taking the urban area of Jixi City as the study area，and the GF-2 image is used as the data source.The image is divided into multiple scales，the classification rules of the corresponding objects are established，and the object-based classification method of the rule set is used to classify the objects.Compare with SVM supervised classification results.The results show that the overall accuracy of object-oriented classification is 92.52%，and the Kappa coefficient is 0.91，which is significantly higher than the SVM supervised classification.Using the object-oriented classification method to classify the GF-2 image is better and the precision is higher.Object-oriented classification method based on GF-2 data is an effective method for extracting urban land use classification.     

The Dynamic Extraction of Classification Rules for Land Consolidation through Object-oriented Features Space Optimization

The high resolution remote sensing image is an important data sources for the accurate extraction of land consolidation area surface information.In this paper,a new object-based method,combining with genetic algorithm and artificial immune algorithm,is used to extract classification rules based on the characteristics of the sample image.After fuzzy classification,the results show that overall accuracy is increasing from 40% by traditional method to 86% corresponding to the genetic algorithm and 90% corresponding to the artificial immune algorithm,and the Kappa coefficient is increasing from 0.3 by traditional methods to 0.82 corresponding to the genetic algorithm and 0.89 corresponding to the artificial immune algorithm.All in all,not only this method can improve the convenience and versatility,changing the previous situation that the rule extraction requires users a large amount of priori knowledge and testing,but also the test results show the significant improvement in classification accuracy.Therefore,it has an important significance for land consolidation,especially using the high-remote sensing images for feature identifying and monitoring.     

基于WorldView-2影像和语义分割模型的小麦分类提取

为使用高分辨率遥感影像和深度学习语义分割模型实现快速准确的小麦种植空间信息提取,以WorldView-2遥感影像为数据源,制作尺度分别为128×128、256×256、512×512的样本数据集,对U-net和DeepLab3+语义分割模型的参数进行训练,建立小麦遥感分类模型;通过与极大似然和随机森林方法比较,检验深度学习分类效果。结果显示：①不同尺度样本训练得到的模型总体精度、Kappa系数分别在94%和0.82以上,模型精度稳定,样本尺度大小对小麦分类提取模型影响较小;②深度学习方法的小麦分类总精度和Kappa系数分别在94%和0.89以上,极大似然和随机森林则在92%和0.85以下,表明该研究建立的小麦遥感分类模型优于传统分类方法。研究结果可为高分辨率遥感影像作物种植信息的深度学习方法提取提供参考。