基于极限学习机的GF-2影像分类

遥感图像分类是提取图像有效信息过程中重要的一部分,为了探寻最优的分类方法,许多机器学习算法逐步应用于遥感分类中。极限学习机（extreme learning machine,ELM）以其高效、快速和良好的泛化性能在模式识别领域得到广泛应用。本文采用训练速度快、运算量小的极限学习机算法与支持向量机（support vector machines,SVM）算法和最大似然法进行分类对比,对高分辨率遥感图像进行分类,分析极限学习机算法对于遥感图像分类的准确度等性能。选取吉林省长春市部分区域的GF-2遥感数据,将融合后的影像设置为原始数据,利用3种方法进行分类。研究结果表明,极限学习机算法分类图像总体分类精度达到85%以上,kappa系数达到0.718,与其他分类方法相比分类准确度较高,且极限学习机运行时间比支持向量机运行时间约短2 480 s,约为支持向量机运行时间的1/8,因此具有良好的性能和实用价值。     

面向对象的喀斯特地区石漠化遥感信息提取研究——以贵州省大方地区为例

针对现有基于像素的监督和非监督分类方法在地质环境复杂、地形起伏较大、阴影明显的喀斯特石漠化地区难以满足石漠化信息提取精度要求的问题,采用基于纹理特征数据和地形数据辅助面向对象方法进行喀斯特地区石漠化信息的提取。该方法首先依据石漠化分布在TM/ETM+影像面积大小不均匀的特征,利用纹理和地形因子计算最优分割参数进行多尺度分割;然后根据植被覆盖率、岩石裸露率以及坡度因子构建石漠化分级指标;最后参照石漠化分级标准、光谱信息以及纹理特征等建立的分类规则提取喀斯特地区石漠化信息。选取贵州省石漠化严重的大方县时序TM/ETM+影像进行石漠化信息提取试验,结果表明:与基于像素的监督分类和非监督分类方法相比,基于面向对象的分类可以有效地减少因复杂地形导致石漠化信息提取结果"椒盐化"现象,提取精度明显优于基于像素的监督分类和非监督分类方法。      

An adaptive framework for spectral-spatial classification based on a combination of pixel-based and object-based scenarios

Remotely sensed image analysis using spectral-spatial information plays a key role in modern remote sensing applications. This article presents a new semi-automatic framework for spectral-spatial classification of hyperspectral images. The proposed framework benefits from a combination of pixel-based and object-based classification scenarios in which the main parameters are adaptively tuned. In order to reduce the complexity of the method, an unsupervised band selection technique is used as well. Meanwhile, the wavelet thresholding is applied in order to smooth the selected bands. The classification results after applying the proposed method to well-known standard hyperspectral datasets are better than those of the most of the other state-of-the-art approaches. As an example, the overall classification accuracy achieved by applying the proposed semi-automatic spectral-spatial classification framework to the Salinas dataset is more than 99% for 10% training samples per class. Moreover, the vital parameters are adaptively set in our approach.     

卷积神经网络和视觉注意力语义分割模型在高分辨率遥感影像分类中的性能分析

随着深度学习语义分割的快速发展,基于计算机视觉语义分割模型的高分辨率遥感影像分类方法也大量涌现。为系统定量地研究经典的和先进的视觉语义分割模型在遥感影像分类中的性能,在总结深度学习语义分割进展的基础上,选择9种基于卷积神经网络(CNN)和视觉注意力的语义分割算法,对米级和厘米级2个尺度的遥感数据集进行分析研究。在模型构建上基于计算机视觉通用的语义分割框架,训练时采用红绿蓝3波段遥感图像并基于ImageNet预训练权重进行迁移学习训练。研究结果表明：通用的语义分割模型通过常规训练设置进行训练能取得较好的遥感影像分类效果,部分地物的交并比(IoU)可以达到90%以上;基于视觉注意力的遥感影像分类模型的精度普遍高于基于CNN的模型,且MaskFormer能更有效地提取离散的地物信息;不同类别的精度最高值并不全在总体最优模型中,部分会存在于次优模型中;类似的地物在更高分辨率遥感数据集中可以获得更高的精度。     

基于卷积神经网络迁移学习模型的矿岩智能识别方法

文章基于Inception-v3卷积神经网络模型,通过对采集的金矿石、铜矿石、铁矿石、铅锌矿、花岗岩、片麻岩、大理岩和页岩,8种岩石453张图像进行特征提取和迁移学习,建立了岩性分类的迁移学习模型,实现了岩性的自动识别和分类。每种岩石图像随机抽取4张作为测试集进行测试,剩余421张图像作为训练集参加训练,经测试全部图像的岩性分类结果均正确,识别正确率超过80%的岩石图像占测试集图像总数的90%以上。识别正确率未达到80%的图像经过处理后重新训练并测试,其识别正确率均超过了80%,表明了该模型具有良好的岩性识别能力且鲁棒性较好,为岩性识别和自动分类提供了一种新的智能分析方法。     

基于岩样细观图像深度学习的岩性自动分类方法

在现阶段的岩土工程中,通常采用人工识别的方法来判别岩样种类,不仅耗时长、专业性强,还易受主观因素影响,准确率不理想。随着计算机技术的发展,机器学习逐渐被应用于岩性的自动识别,开启了岩样分类的新路径。本文以重庆市主城区4种典型岩样（泥岩、砂质泥岩、泥质砂岩和砂岩）的细观图像为研究对象,基于Inception V3卷积网络模型和迁移学习算法,建立了岩样细观图像深度学习模型,并完成了训练学习。结果显示：模型在训练1 000次后,训练集中的分类准确率达到92.77%,验证集中的分类准确率为76.31%。其中,验证集中的砂岩识别准确率为97.28%,泥岩识别准确率为81.85%,泥质砂岩识别准确率为72.59%,砂质泥岩识别准确率为72.35%。与现有的机器学习方法相比,本识别模型不仅可以自动识别岩性极为相近的岩样,而且具有较好的识别准确率、鲁棒性和泛化能力。     

Assessment of the reliability of supervised classifications of Landsat-7, ASTER,and SPOT-5 multispectral data in rock unit discriminations of Jabal Daf-Wadi Fatima area,Saudi Arabia

Multispectral, multiresolution remotely sensed data were processed to emphasize geological interpretation of Jabal Daf-Wadi Fatima area. The investigated area is situated in the central western part of Saudi Arabia and geologically consists of igneous and metamorphosed rocks overlain by sedimentary sequence belonging to the Arabian-Nubian Shield. Three sets of digital satellite data, Landsat-7 ETM+, ASTER, and SPOT-5, were used in this study. The application of image processing techniques enables to identify and delineate the lithologic units and the structural features of the study area. The results of this study indicate that the confusion matrix of the three maximum likelihood supervised classifications of the three datasets shows that the Landsat ETM+ bands scored the best degree of average and overall accuracy (77 and 78%, respectively). This classification distinguishes most of the rock units for mapping in the investigated area. The supervised classification of ASTER and SPOT bands has lower degrees of accuracy than the classified Landsat data. The supervised classification of SPOT bands has a degree of average and overall accuracy of 66 and 67%, respectively, but it is the best for distinguishing the spectral signatures of the different members of Fatima Formation (lower, middle, and upper members). The statistical analyses of the confusion matrices of classifications and the interpretation of the produced classified thematic maps revealed that the classification accuracy does not necessary depend on the spatial resolution of satellite data. The data of the highest spatial resolution such as SPOT data are also very useful in emphasizing and classifying the rock units of a small outcrop area. The detailed geological map of Jabal Daf-Wadi Fatima area is interpreted in this work from supervised classified images of different resolutions as well as the structure map of this area. This study shows that it is preferable to use the supervised classifications of multiresolution data for rock unit discrimination in detailed field mapping.     

基于计算机X射线断层术与扫描电镜图像的黄土微结构定量分析

以湿陷性黄土的电镜扫描（SEM）和三轴CT扫描试验为基础,针对CT图像分辨率较低、难以实现土微结构精确量化的缺陷,通过对不同放大倍数的SEM图像进行图像分析,并从其中选择标准训练样本,利用训练样本对CT图像进行监督分类,从而达到定量化分析土的微结构的目的。通过比较CT图像基于自身灰度分级和基于SEM训练样本两种不同方法进行监督分类,结果表明基于SEM训练样本的CT图像监督分类,可以更好地量化监测黄土大孔隙、团粒、黏土集粒和矿物颗粒在固结剪切过程中的变化规律,从而为土的微结构研究提供了新的视角。     

Multi-temporal landsat images based on eco-environmental change analysis in and around Chah Nimeh reservoir,Balochestan (Iran)

Chah Nimeh reservoirs have served as a water storage facility, especially during droughts over the last three decades. It is also an important wintering site for migratory birds. In this study, thematic mapper time-series data were derived from Landsat images for prolonged droughts that occurred in two satellite images (2002 and 2011). The data derived from these images were used for the detection of changes in land cover and water storage in the reservoirs. First, a vegetation cover map was produced using soil-adjusted vegetation index and field sampling. Subsequently, land use/cover maps were generated using supervised and hybrid image classification method. Using the spatial change detector (SCD v1.0) software extension, the layers were combined and the change map was generated. The overall accuracy of the produced thematic images was assessed in regards to quantity and allocation disagreements. A total of five classes were defined in this investigation: deep water, shallow water, vegetation, salt land and bare land. The results showed that during the period of study, water volume reduced and vegetation cover increased, especially around the reservoirs that are important as shelter for wintering migratory birds. Comparison of land use/cover maps showed the increase in total available surface of shallow water, which indicated an increase in the habitats for surface feeding and diving birds.     

基于RS与GIS技术的江苏海岸带湿地分类

通过对江苏海岸带TM影像进行计算机自动分类与人工解译相结合的分类研究,探讨了提高海岸带湿地分类精度与效率的方法与途径。先采用分区分层分类的方法依据海岸线将本研究区分为陆地和海滩两部分。对于陆地部分,对基础分类影像经过非监督分类和光谱聚类处理后,获得分类模板,利用此模板对基础分类影像进行监督分类,对于海滩部分依据平均高潮位线、中潮位线、NDVI对影像进行分层分类,在分类的过程中运用了人机互译判读方法。结果精度评价表明该方法能明显提高海岸带湿地的分类精度。最后,基于VB和MO控件开发了江苏海岸带湿地GIS系统。     

基于多源遥感影像的长春市城市建成区提取

夜光遥感影像记录的城市灯光与人类活动密切相关,已广泛应用于城市信息提取。珞珈一号作为新一代夜光遥感数据源,比以往的夜光数据具有更高的空间分辨率和光谱分辨率,可以更清晰地表达城市建成区范围和内部结构。本文利用珞珈一号夜光遥感影像,通过人类居住指数（human settlement index, HSI）、植被覆盖和建筑共同校正的城市夜光指数（vegetation and build adjusted nighttime light urban index, VBANUI）及支持向量机（support vector machine, SVM）监督分类3种方法对长春市城市建成区进行提取,并与利用NPP/VIIRS（suomi national polar-orbiting partnership/visible infrared imaging radiometer suite）夜光遥感影像、采用同样方法得到的结果对比。结果显示：本文提出的VBANUI提高了传统植被覆盖校正的城市夜光指数（vegetation adjusted nighttime light urban index, VANUI）的提取精度,使用珞珈一号夜光遥感影像通过VBANUI提取的城市建成区结果最优,其Kappa系数为0.80,总体分类精度为90.74%;使用珞珈一号和NPP/VIIRS夜光遥感影像通过HSI按最佳阈值提取城市建成区的Kappa系数分别为0.75和0.72,总体分类精度分别为88.27%和86.54%;复合数据的SVM监督分类法中Landsat-NDBI、Landsat-NDBI-VIIRS、Landsat-NDBI-LJ和Landsat-NDBI-LJlog的Kappa系数分别为0.602、0.627、0.643和0.681,总体分类精度分别为81.11%、81.52%、82.25%和84.48%。研究结果表明：3种提取方法下,均为使用珞珈一号夜光遥感影像的结果优于使用NPP/VIIRS夜光遥感影像的结果,证明相比于NPP/VIIRS夜光遥感影像,珞珈一号夜光遥感影像更适用于城市尺度的建成区范围提取。     

Appraisal of land use/land cover of mangrove forest ecosystem using support vector machine

Human activities in many parts of the world have greatly changed the natural land cover. This study has been conducted on Pichavaram forest, south east coast of India, famous for its unique mangrove bio-diversity. The main objectives of this study were focused on monitoring land cover changes particularly for the mangrove forest in the Pichavaram area using multi-temporal Landsat images captured in the 1991, 2000, and 2009. The land use/land cover (LULC) estimation was done by a unique hybrid classification approach consisting of unsupervised and support vector machine (SVM)-based supervised classification. Once the vegetation and non-vegetation classes were separated, training site-based classification technology i.e., SVM-based supervised classification technique was used. The agricultural area, forest/plantation, degraded mangrove and mangrove forest layers were separated from the vegetation layer. Mud flat, sand/beach, swamp, sea water/sea, aquaculture pond, and fallow land were separated from non-vegetation layer. Water logged areas were delineated from the area initially considered under swamp and sea water-drowned areas. In this study, the object-based post-classification comparison method was employed for detecting changes. In order to evaluate the performance, an accuracy assessment was carried out using the randomly stratified sampling method, assuring distribution in a rational pattern so that a specific number of observations were assigned to each category on the classified image. The Kappa accuracy of SVM classified image was highest (94.53 %) for the 2000 image and about 94.14 and 89.45 % for the 2009 and 1991 images, respectively. The results indicated that the increased anthropogenic activities in Pichavaram have caused an irreversible loss of forest vegetation. These findings can be used both as a strategic planning tool to address the broad-scale mangrove ecosystem conservation projects and also as a tactical guide to help managers in designing effective restoration measures.     

集成极限学习机的中小河流洪水预报方法研究

为利用水文现象相似性和极限学习机(ELM)集成学习提高洪水预报精度,提出了一种基于相似度匹配的集成ELM洪水预报方法(SM-ELM)。方法首先从多个ELM模型中,为每一个训练样本找到最优的ELM模型,然后从训练集中,为测试样本匹配出最相似的前k个训练样本,最后利用这k个训练样本分别对应的最优ELM模型,对测试样本采用加权平均法进行集成预报。为证明提出方法的可行性和有效性,以昌化流域的历史洪水为例进行了验证。结果表明,相对于单个ELM,集成ELM模型能有效地提高预测精度。从均方根误差上看,集成ELM模型性能比单个ELM模型提升了10%~15%。在三种集成方法中,SM-ELM能够以较少的模型数量获得较高且稳定的预报精度。     

Forecasting future changes in Manzala Lake surface area by considering variations in land use and land cover using remote sensing approach

This study assesses the changes in surface area of Manzala Lake, the largest coastal lake in Egypt, with respect to changes in land use and land cover based on a multi-temporal classification process. A regression model is provided to predict the temporal changes in the different detected classes and to assess the sustainability of the lake waterbody. Remote sensing is an effective method for detecting the impact of anthropogenic activities on the surface area of a lagoon such as Manzala Lake. The techniques used in this study include unsupervised classification, Mahalanobis distance supervised classification, minimum distance supervised classification, maximum likelihood supervised classification, and normalized difference water index. Data extracted from satellite images are used to predict the future temporal change in each class, using a statistical regression model and considering calibration, validation, and prediction phases. It was found that the maximum likelihood classification technique has the highest overall accuracy of 93.33%. This technique is selected to observe the changes in the surface area of the lake for the period from 1984 to 2015. Study results show that the waterbody surface area of the lake declined by 46% and the area of floating vegetation, islands, and land agriculture increased by 153.52, 42.86, and 42.35% respectively during the study period. Linear regression model prediction indicates that the waterbody surface area of the lake will decrease by 25.24% during the period from 2015 to 2030, which reflects the negative impact of human activities on lake sustainability represented by a severe reduction of the waterbody area.     

喀斯特地区遥感影像解译新算法——支持向量机算法

现行的遥感影像解译方法有监督分类和非监督分类。在监督分类中有平行算法,最小距离算法、最大似然算法等,而支持向量机是监督分类中的一种新的算法。本研究选择贵阳市花溪区小碧乡局部地区为研究对象,采用SPOT数据,分别运用最大似然算法和支持向量机算法对研究区遥感影像进行解译。通过建立混淆矩阵,来计算分类精度和Kappa系数。结果表明:支持向量机具有分类精度高,分类图斑完整等优点;但在时间的消耗上,支持向量机算法要比最大似然算法长。对于这两种算法而言,都存在地物光谱特征明显相异的地物易于区别,光谱相似的地物容易造成错分的现象,然而支持向量机分类精度要比最大似然分类精度高一些。支持向量机对样本数量具有敏感性,样本数量过多将导致运算时间过长。因此在实际运用中应根据实际情况,选择适合的算法。      

Automatic arable land detection with supervised machine learning

In Precision Agriculture one of the basic tasks is the classification of land zones in either arable or non-arable land. Several studies have been conducted using data obtained from soil analysis or local exploration of the parcels. However, sometimes only data from satellite images are available and then the problem not only becomes more challenging but also more interesting to solve because it is much more cost-effective. In this paper, we consider different spectral and thermal bands from the Landsat 8 satellite images corresponding to a vineyard located in Galicia, a region in Northeastern Spain, and apply a range of supervised Machine Learning methods to classify the different land zones. We conclude that an adequate choice of the algorithm parameters together with feature selection techniques can yield a classification that is both highly effective and efficient.     

Derivation of vegetation density and land-use type pattern in mountain regions of Jordan using multi-seasonal SPOT images

The mountainous region represents the most important agricultural and biodiversity haven in Jordan. The objective of this study is to characterize the seasonal pattern of land use and vegetation using multi-temporal SPOT images. Multi-temporal SPOT images were analyzed to characterize the land use and cropping pattern in the mountain regions of Jordan. The images were radiometrically corrected using invariant objects located on the image, and a linear inter-calibration method was used to calibrate the other images. A hybrid classification approach was used in the classification; the spectral signatures of the land-use classes were derived in an iterative procedure using the ISODATA and field survey data. Then, the maximum likelihood classification was applied on all images to classify the class signatures into thematic land-use types. The hybrid classification approach gives more accurate classification accuracy especially for the multi-seasonal image classification. The overall accuracy of the multi-temporal data set was achieved with 87.9%, while classification accuracy for single-date classifications were 61.3, 76.8, 72.2, and 65.5 for months of October, February, April, and June, respectively. In addition, the scene combinations that were derived from February and April were classified the land-use types almost as well as those combinations including more scenes. Regarding the classification details, the multi-temporal images enable higher level of classification for land-use types such as Anderson level 2, and produce accurate boundaries for the different cropping and farming systems.     

遥感影像的分类与识别技术在土地资源调查中的应用

遥感图像记录了地物在空间域、时间域、光谱域的变化信息。利用图像的分类技术 ,能够识别土地利用类型。计算机遥感分类识别原理 ,是利用地物的光谱能量特征差异性和结构特征差异性来识别地物信息。根据北京某地遥感图像实例资料 ,将土地利用类型分为监督分类、非监督分类、最大似然法分类、神经网络分类等。不同的分类方法有各自的特点且分类结果也有一定的差别 ,其中神经网络分类与真实情况最为接近     

一种基于多特征波段岩土层次分类方法

岩土分类与一般地表的地物分类相比难度大得多,针对已有的分类方法（监督分类和非监督分类）对于岩土分类精度不高、分类效果欠佳问题提出一种基于多特征波段岩土层次分类方法。它是一种自顶向下、逐步求精的层次分类方法,该方法结合无监督分类和监督分类两种分类方法的优势,利用多个特征波段组合,有层次地将不同类型的岩土体逐步分开,实现对岩土的精确分类。对北京市怀柔山区附近的ASTER影像数据进行的岩土分类实验结果表明,基于多特征波段岩土层次分类识别方法能显著提高岩土分类精度,总体精度提高10%,Kappa系数提高了0.1,并且能识别以往分类识别方法难以区分的岩石阴影和水体等地物,能够有效地克服“同物异谱”现象。