光谱可形变卷积驱动的高光谱图像分类

基于卷积神经网络的高光谱图像分类是当前的研究热点,先后发展了空洞卷积、可形变卷积等先进模型。然而,现有可形变卷积只在空间维偏移,忽略了高光谱图像光谱之间的差异信息。为此,本文将可形变卷积从空间维扩展到光谱维,设计了光谱可形变卷积,提出了光谱可形变卷积网络SDCNN (Spectral Deformable Convolutional Neural Network)。首先,利用全连接层学习光谱可形变卷积的偏移量,采用线性差值对图像光谱维进行特征校准;其次,采用多层1×1卷积进行光谱维特征聚合;最后,使用三维卷积层提取光谱—空间联合特征。不同于空间可形变卷积,光谱可形变卷积只在光谱维上进行偏移,可以为不同类别选择更合适的特征波段,提升模型的判别性。在国际通用测试数据Indian Pines、University of Pavia以及University of Houston上进行了实验,结果表明：本文提出的SDCNN方法优于其他深度学习方法,在相同样本条件下取得了更高的分类精度,总体精度达到了98.86%(Indian Pines,10%/类)、99.81%(University of P...     

加权全极化SAR图像分类下的功能区土地时空变化特征提取

为准确分析功能区土地时空演变情况,需准确区分土地时空变化的相关特征,本文设计了功能区土地时空变化特征提取模型。首先采用全极化分解和灰度共生矩阵,对SAR图像中反映功能区土地时空变化的、不同地物的各类散射特征和纹理特征进行分类。然后确定最佳加权全极化特征组合,将该组合输入随机森林模型,完成最终图像中地物分类。最后以湖南省某市生态功能区土地时空变化特征为例,实现功能区土地时空变化特征分类提取。测试结果表明,该模型采用加权全极化特征组合,可准确描述地物分布情况,保证地物的可靠分类,能实现较好的提取效果。     

面向对象的三亚珊瑚礁底质信息提取

珊瑚礁对于海洋生态环境研究具有重要意义,通过分析珊瑚礁底栖物质的分布及健康状况,可以对珊瑚礁生态环境进行评估。本文提出了一种基于面向对象的图像分类方法,通过试验确定不同地貌的最优分割尺度,其中陆地和深海的最优分割尺度为150,各类底栖物质的最优分割尺度为30。以Sentinel-2A卫星遥感影像为例,提取海南三亚珊瑚礁自然保护区的珊瑚礁底栖物质,并使用混淆矩阵对提取结果进行精度评估。结果表明,底栖物质提取总体分类精度为87.91%,Kappa系数为0.83。面向对象分类方法可有效结合珊瑚礁底栖物质的纹理特征和光谱特征,并充分利用遥感影像不同波段的组合特性,可为三亚珊瑚礁保护管理提供方法支撑。     

Sentinel-3 OLCI数据的内陆湖泊有机悬浮物浓度遥感估算

有机悬浮物OSM (Organic Suspended Matter)是湖泊有机碳库的重要组成成分,对研究湖泊生态环境和初级生产力具有重要意义.本研究以太湖、巢湖、涌湖、小兴凯湖、滇池、洪泽湖、呼伦湖和南漪湖8个内陆湖泊为研究区,发展了适合内陆水体的有机悬浮物浓度遥感估算方法.基于B6、B7波段的斜率和B10、B11波...     

构建地块二维表征及CNN模型的作物遥感分类

本文旨在研究基于地块数据约束的深度学习模型的分类特征表示方法,以识别不同作物在不同时相上光谱差异从而对作物类型进行分类。通过Google Earth Engine平台获取作物生育期内全部Landsat 8影像,利用其质量评定波段完成研究区无云时相及区域上的地块统计,提取地块级别的各波段反射率均值按照时相顺序及波长进行排列,构建波谱、时相二维特征图作为该地块的抽象表示。通过构建相对最优的卷积神经网络CNN(Convolutional Neural Network)结构完成对特征图的分类,从而完成对地块的分类。构建CNN模型并不需要手工特征和预定义功能的需求,可完成提取特征并遵循端到端原则进行分类。将该模型的分类结果与其他最为常用机器学习分类器进行了比较,获得了优于常用遥感分类算法的分类精度。结果表明地块数据的加入可以有效的缩减计算规模并提供了准确的分类边界。所提出得方法在地块特征表示及作物分类中具有突出的应用潜力,应视为基于地块的多时相影像分类任务的优选方法。     

基于多特征融合与典型降维方法的高光谱影像分类

高光谱影像的冗余信息给影像的分类效果带来一定的负面影响。本文利用CB法(CfsSubsetEval评估器结合Best-First搜索策略)与PCA变换两种降维方法,分别结合随机森林分类器对4种多特征融合方案(共8种组合)进行高光谱影像分类对比,基于分类的总体精度、Kappa系数探究提高高光谱影像分类的最佳组合方法。结果表明:①多特征融合可提升高光谱影像的分类效果,两种降维方法的分类精度均随地理特征、纹理特征、指数特征的加入而逐渐提高。②两种降维方法中,经CB法降维后的分类精度均比通过PCA变换降维的分类精度高。在构造的8种组合中,基于所有特征信息(光谱特征、地理特征、纹理特征、指数特征)的CB法分类精度最高,其总体精度为98.01%;Kappa系数为0.969 9。     

二滩、官地库区土地覆盖类型图制作方法研究

利用航空彩红外正射影像镶嵌图,用光谱角分类与目视解译相结合的方法,对二滩、官地库区进行土地利用调查工作,取得了较好的效果。     

An improved algorithm for supervised fuzzyC-means clustering of remotely sensed data

This paper describes an improved algorithm for fuzzyc-means clustering of remotely sensed data, by which the degree of fuzziness of the resultant classification is decreased as comparing with that by a conventional algorithm: that is, the classification accuracy is increased. This is achieved by incorporating covariance matrices at the level of individual classes rather than assuming a global one. Empirical results from a fuzzy classification of an Edinburgh suburban land cover confirmed the improved performance of the new algorithm for fuzzyc-means clustering, in particular when fuzziness is also accommodated in the assumed reference data.     

Topographic and Satellite Information for Preliminary Route Location

Topographic information from maps and geographical information systems (GIS) has been combined with satellite data (SPOT Panchromatic, SPOT Multispectral and Landsat Thematic Mapper) to derive a product that may be valuable in preliminary route location studies. The objectives of this study were to evaluate the classification accuracy of this combined product, to compare levels of ground detail obtainable from different types of satellite imagery against aerial photography, and to present an example application on the use of the combined product.
The classification accuracy of the combined product was dependent on the type of land cover and was 83 to 100 per cent successful, with accuracy exceeding 95 per cent for most land cover types. The overall accuracy of the product was almost 95 per cent, with accuracy based on KHAT statistics of 92 per cent. Varying levels of ground detail were attainable from different types of satellite imagery. This detail may be adequate for preliminary route selection, especially in the absence of aerial photographs and GIS. The combined product presented in this study was applied successfully in selecting the optimal route for the Greater Amman ring road.     

The decision of the optimal parameters in Markov random fields of images by genetic algorithm

This paper introduces the principle of genetic algorithm and the basic method of solving Markov random field parameters.Focusing on the shortcomings in present methods,a new method based on genetic algorithms is proposed to solve the parameters in the Markov random field.The detailed procedure is discussed.On the basis of the parameters solved by genetic algorithms,some experim ents on classification of aerial images are given.Experimental results show that the proposed method is effective and the classification results are satisfactory.