人工神经网络遥感影像分类模型及其与知识集成方法研究

以多层感知器（MLP）为例，探讨了地学知识与ANN融合进行遥感影像分类的方法。首先对MLP网络结构、学习算法及其改进进行分析；然后总结了MLP进行遥感影像分类的一般方法和存在的缺陷；发展了基于知识的MLP神经网络遥感影像分类模型，并具体利用基于规则的MLP方法进行了遥感土地覆盖分类的实验，把获得的结果与传统统计方法与一般ANN方法进行了综合比较，获得了有意义的结果。     

Efficient collection of training data for sub-pixel land cover classification using neural networks

Artificial neural networks (ANNs) are a popular class of techniques for performing soft classifications of satellite images. They have successfully been applied for estimating crop areas through sub-pixel classification of medium to low resolution images. Before a network can be used for classification and estimation, however, it has to be trained. The collection of the reference area fractions needed to train an ANN is often both time-consuming and expensive. This study focuses on strategies for decreasing the efforts needed to collect the necessary reference data, without compromising the accuracy of the resulting area estimates. Two aspects were studied: the spatial sampling scheme (i) and the possibility for reusing trained networks in multiple consecutive seasons (ii). Belgium was chosen as the study area because of the vast amount of reference data available. Time series of monthly NDVI composites for both SPOT-VGT and MODIS were used as the network inputs. The results showed that accurate regional crop area estimation (R² > 80%) is possible using only 1% of the entire area for network training, provided that the training samples used are representative for the land use variability present in the study area. Limiting the training samples to a specific subset of the population, either geographically or thematically, significantly decreased the accuracy of the estimates. The results also indicate that the use of ANNs trained with data from one season to estimate area fractions in another season is not to be recommended. The interannual variability observed in the endmembers’ spectral signatures underlines the importance of using up-to-date training samples. It can thus be concluded that the representativeness of the training samples, both regarding the spatial and the temporal aspects, is an important issue in crop area estimation using ANNs that should not easily be ignored.     

Thematic mapping from remotely sensed data with neural networks: MLP, RBF and PNN based approaches

 Neural networks are attractive tools for the derivation of thematic maps from remotely sensed data. Most attention has focused on the multilayer perceptron (MLP) network but other network types are available and have different properties that may sometimes be more appropriate for some applications. Here a MLP, radial basis function (RBF) and probabilistic neural network (PNN) were used to classify remotely sensed data of an agricultural site. The accuracy of these classifications ranged from 86.25–91.25%. The accuracy of the PNN classification could be increased through the incorporation of prior probabilities of class membership but the accuracy of each classification could also be degraded by the presence of an untrained class. Post-classification analyses, however, could be used to identify potentially misclassified cases, including those belonging to an untrained class, to increase accuracy. The effect of the post-classification analysis on the accuracy of the classification derived from each of the three network types investigated differed and it is suggested that network type be selected carefully to meet the requirements of the application in-hand. Received: 23 March 2000 / Accepted: 9 July 2000     

Mapping salt diapirs and salt diapir-affected areas using MLP neural network model and ASTER data

Abstract

This study employs visible-near infrared and short wave infrared datasets of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) to map salt diapirs and salt diapir-affected areas using Multi-Layer Perceptron (MLP) in the Zagros Folded Belt, Iran, and introduces the role of earth observation technology and a type of digital earth processing in lithological mapping and geo-environmental impact assessment. MLP neural network model with several learning rates between 0.01 and 0.1 was carried out on ASTER L1B data, and the results were compared using confusion matrices. The most appropriate classification image for L1B input to MLP was produced by learning rate of 0.01 with Kappa coefficient of 0.90 and overall accuracy of 92.54%. The MLP result of input data set mapped lithological units of salt diapirs and demonstrated affected areas at the southern and western parts of the Konarsiah and Jahani diapirs, respectively. Field observations and X-ray diffraction analyses of field samples confirmed the dominant mineral phases identified remotely. It is concluded that MLP is an efficient approach for mapping salt diapirs and salt-affected areas.     

联合卷积神经网络与集成学习的遥感影像场景分类

针对人工设计的中、低层特征难以实现复杂场景影像的高精度分类以及卷积神经网络依赖大量训练数据等问题，结合迁移学习与集成学习，提出了一种联合卷积神经网络与集成学习的遥感影像场景分类算法。首先基于迁移学习的思想，利用在自然影像数据集上训练好的多个深层卷积神经网络模型作为特征提取器，提取图像多个高度抽象的语义特征；然后构建由Logistic回归和支持向量机组成的Stacking集成模型，对同一图像的多个特征分别训练Logistic模型，将预测概率结果融合构建概率特征；最后利用支持向量机对概率特征训练和预测，得到场景影像的分类结果。利用UCMerced_LandUse和NWPU-RESISC 45两种不同规模的遥感影像数据集进行试验，即使在只有10%的数据作为训练样本情况下，本文方法能够分别达到90.74%和87.21%的分类精度。     

Evaluation of diverse classification approaches for land use/cover mapping in a Mediterranean region utilizing Hyperion data

Information on Earth's land surface cover is commonly obtained through digital image analysis of data acquired from remote sensing sensors. In this study, we evaluated the use of diverse classification techniques in discriminating land use/cover types in a typical Mediterranean setting using Hyperion imagery. For this purpose, the spectral angle mapper (SAM), the object-based and the non-linear spectral unmixing based on artificial neural networks (ANNs) techniques were applied. A further objective had been to investigate the effect of two approaches for training sites selection in the SAM classification, namely of the pixel purity index (PPI) and of the direct selection of training points from the Hyperion imagery assisted by a QuickBird imagery and field-based training sites. Object-based classification outperformed the other techniques with an overall accuracy of 83%. Sub-pixel classification based on the ANN showed an overall accuracy of 52%, very close to that of SAM (48%). SAM applied using the training sites selected directly from the Hyperion imagery supported by the QuickBird image and the field visits returned an increase accuracy by 16%. Yet, all techniques appeared to suffer from the relatively low spatial resolution of the Hyperion imagery, which affected the spectral separation among the land use/cover classes.     

Land-use classification using ASTER data and self-organized neutral networks

Operationally AVHRR and TM/TM+ data were used and a supervised maximum likelihood classification (MLH) was applied to depict land use changes in Beijing, providing basic maps for planning and development. With rapid growth of the city these are helpful to deal with higher resolution data, whereas new classification algorithms produce land use maps more accurate. In the paper, new sensor ASTER data and the Kohonen self-organized neural network feature map (KSOM) were tested.The TSOM classified 7% more accurately than the maximum likelihood algorithm in general, and 50% more accurately for the classes ‘residential area’ and ‘roads’. The results suggest that ASTER data and the Kohonen self-organized neural network classification can be used as an alternative data and method in a land use update operational system.     

基于深度残差网络的机载LiDAR点云分类

机载LiDAR点云的分类是利用其进行城市场景三维重建的关键步骤之一。为充分利用现有的图像领域性能较好的深度学习网络模型,提高点云分类精度,并降低训练时间和对训练样本数量的要求,本文提出一种基于深度残差网络的机载LiDAR点云分类方法。首先提取归一化高程、表面变化率、强度和归一化植被指数4种具有较高区分度的点云低层次特征;然后通过设置不同的邻域大小和视角,利用所提出的点云特征图生成策略,得到多尺度和多视角点云特征图;再将点云特征图输入到预训练的深度残差网络,提取多尺度和多视角深层次特征;最后构建并训练神经网络分类器,利用训练的模型对待分类点云进行预测,经后处理得到分类结果。利用ISPRS三维语义标记竞赛的公开标准数据集进行试验,结果表明,本文方法可有效区分建筑物、地面、车辆等8类地物,分类结果的总体精度为87.1%,可为城市场景三维重建提供可靠的信息。     

A novel confidence estimation method for neural networks in multispectral image classification

Abstract

The Digital Earth concept has attracted much attention recently and this approach uses a variety of earth observation data from the global to the local scale. Imaging techniques have made much progress technically and the methods used for automatic extraction of geo-ralated information are of importance in Digital Earth science. One of these methods, artificial neural networks (ANN) techniques, have been effectively used in classification of remotely sensed images. Generally image classification with ANN has been producing higher or equal mapping accuracies than parametric methods. Comparative studies have, in fact, shown that there is no discernible difference in classification accuracies between neural and conventional statistical approaches. Only well designed and trained neural networks can present a better performance than the standard statistical approaches. There are, as yet, no widely recognised standard methods to implement an optimum network. From this point of view it might be beneficial to quantify ANN's reliability in classification problems. To measure the reliability of the neural network might be a way of developing to determine suitable network structures. To date, the problem of confidence estimation of ANN has not been studied in remote sensing studies. A statistical method for quantifying the reliability of a neural network that can be used in image classification is investigated in this paper. For this purpose the method is to be based on a binomial experimentation concept to establish confidence intervals. This novel method can also be used for the selection of an appropriate network structure for the classification of multispectral imagery. Although the main focus of the research is to estimate confidence in ANN, the approach might also be applicable and relevant to Digital Earth technologies.     

Estimation of chlorophyll-<Emphasis Type="Italic">A</Emphasis> concentration using an artificial neural network (ANN)-based algorithm with oceansat-I OCM data

An artificial neural network (ANN) based chlorophyll-a algorithm was developed to estimate chlorophyll-a concentration using OCEANSAT-I Ocean Colour Monitor (OCM) satellite-data. A multi-layer perceptron (MLP) type neural network was trained using simulated reflectances (~60,000 spectra) with known chlorophyll-a concentration, corresponding to the first five spectral bands of OCM. The correlation coefficient(r ²) andRMSE for the log transformed training data was found to be 0.99 and 0.07, respectively. The performance of the developed ANN-based algorithm was tested with the global SeaWiFS Bio-optical Algorithm Mini Workshop (SeaBAM) data (~919 spectra), 0.86 and 0.13 were observed asr ² andRMSE for the test data set. The algorithm was further validated with thein-situ bio-optical data collected in the northeastern Arabian Sea (~215 spectra), ther ² andRMSE were observed as 0.87 and 0.12 for this regional data set. Chlorophyll-a images were generated by applying the weight and bias matrices obtained during the training, on the normalized water leaving radiances (nL ^W) obtained from the OCM data after atmospheric correction. The chlorophyll-a image generated using ANN based algorithm and global Ocean Chlorophyll-4 (OC4) algorithm was compared. Chlorophyll-a estimated using both the algorithms showed a good correlation for the open ocean regions. However, in the coastal waters the ANN algorithm estimated relatively smaller concentrations, when compared to OC4 estimated chlorophyll-a.     

遥感图像土地覆盖分类中多源特征数据选择研究

多源特征数据可以提高遥感图像的分类精度,选择合适的特征数据十分重要。利用基尼指数对多尺度纹理信息、主成分变换前三分量、地形数据等特征进行选择,选出最佳特征子集。利用支持向量机、神经网络分类法、最大似然法分别对全部特征数据和最佳特征子集结合多光谱数据进行分类。实验结果表明:基尼指数可以有效地对多源特征数据进行选择,特征选择可以提高分类器效率,提高分类精度。     

Evaluation of multisource data for glacier terrain mapping: a neural net approach

Spectrally similar nature of land covers in a glacierized terrain hampers their automated mapping from multispectral satellite data, which may be overcome by using multisource data. In the present study, an artificial neural network (ANN)-based information extraction approach was applied for mapping the Kolahoi glacier and adjoining areas, using Landsat TM (Thematic Mapper) data and several ancillary layers such as image transformations and topographic attributes. Results reveal that ANN (highest overall accuracy (OA): 83.74%) outperforms maximum likelihood classifier (highest OA: 66.90%) and the incorporation of ancillary data into the classification process significantly enhances the mapping accuracy (>9%), particularly the addition of Near Infrared Red/Short Wave Infrared (NIR/SWIR) data to the spectral data. A nine-band combination dataset (spectral data, slope, Red/NIR and decorrelation stretch) was found to be the best multisource dataset. Results of the Z-tests (at 95% confidence level) also corroborate and statistically validate the above findings.     

通过训练样本采样处理改善小宗作物遥感识别精度

训练样本质量是决定农作物遥感识别精度的关键因素,虽然高空间分辨率卫星的发展有效地解决了农作物遥感识别过程中的混合像元问题,但是当区域内不同作物种植面积差异较大时,训练集中不同类别样本数量往往相差较大,这样的不均衡数据集影响分类器的训练,导致少数类别的识别精度不理想。为研究作物遥感识别过程中的不均衡样本问题,本文基于GF-2号卫星数据,首先挖掘了地物的光谱信息、纹理信息,用特征递归消除RFE (Recursive Feature Elimination)方法进行特征优选,然后从数据处理的角度采用了5种采样算法对不均衡训练集进行处理,最后使用采样后的均衡数据集训练分类器,对比数据采样前后决策树与Adaboost(Adaptive Boosting)两种分类器的识别结果,发现:(1)经过采样处理后两种分类算法明显提升了小宗作物的分类精度;(2)经过ADASYS (Adaptive synthetic sampling)采样处理后,分类器性能提升最多,决策树的Kappa系数提高了14.32%,Adaboost的Kappa系数提高了10.23%,达到最高值0.9336;(3)过采样的处理效果优于欠采样,过采样对分类器的性能提升更多。综上所述,选择合适的采样方法和分类方法是提高不均衡数据集遥感分类精度的有效途径。     

利用胶囊网络实现高光谱影像空谱联合分类

卷积神经网络等深度学习模型已经在高光谱影像分类任务中取得了理想的结果.然而,由于传统神经元只能进行标量计算,现有的深度学习模型无法对高光谱影像特征的实例化参数进行建模,因此无法在邻域范围受限的条件下获得令人满意的分类效果.通过引入胶囊网络结构设计了一种新型网络模型,该模型利用胶囊神经元进行向量计算,并利用权重矩阵编码特...     

联合显著性和多层卷积神经网络的高分影像场景分类

高分辨率遥感影像中的场景信息,对于影像解译和现实世界的理解具有重要意义。传统的场景分类方法多利用中、低层人工特征,但是高分辨率遥感影像的信息丰富,场景构成复杂,需要高层次的特征来表达。本文提出了一种联合显著性和多层卷积神经网络的方法,首先利用显著性采样获取包含影像主要信息的有意义的块,将这些块作为样本集输入卷积神经网络中进行训练,获得不同层次的特征表达,最后联合多层特征利用支持向量机进行分类。两组高分影像场景数据UC Merced 21类和Wuhan 7类试验表明,显著性采样能够有效地获取主要目标,减弱其他无关目标的影响,降低数据冗余;卷积神经网络能够自动学习高层次的特征,相比已有方法,本文方法能够有效提高分类精度。     

SAR image despeckling with a multilayer perceptron neural network

Speckle noise in synthetic-aperture radar (SAR) images severely hinders remote sensing applications; therefore, the appropriate removal of speckle noise is crucial. This paper elaborates on the multilayer perceptron (MLP) neural-network model for SAR image despeckling by using a time series of SAR images. Unlike other filtering methods that use only a single radar intensity image to derive their parameters and filter that single image, this method can be trained using archived images over an area of interest to self-learn the intensity characteristics of image patches and then adaptively determine the weights and thresholds by using a neural network for image despeckling. Several hidden layers are designed for feedforward network training, and back-propagation stochastic gradient descent is adopted to reduce the error between the target output and neural-network output. The parameters in the network are automatically updated in the training process. The greatest advantage of MLP is that once the despeckling parameters are determined, they can be used to process not only new images in the same area but also images in completely different locations. Tests with images from TerraSAR-X in selected areas indicated that MLP shows satisfactory performance with respect to noise reduction and edge preservation. The overall image quality obtained using MLP was markedly higher than that obtained using numerous other filters. In comparison with other recently developed filters, this method yields a slightly higher image quality, and it demonstrates the powerful capabilities of computer learning using SAR images, which indicate the promising prospect of applying MLP to SAR image despeckling.     

基于深度学习理论的无人机影像分类

针对无人机影像深度学习分类方法缺乏现状,本文利用深度学习理论卷积神经网络方法对无人机影像进行了分类。该法首先抽取无人机影像作为训练集和检验集,然后建立一个2个卷积层-池化层的卷积神经网络模型进行深度学习,通过设定参数并运行模型实现无人机影像分类。实验表明,本文提出的方法可完成较复杂地区无人机影像分类,其分类精度与支持向量机方法相当,为无人机遥感影像分类提供了一个崭新的技术视点。     

利用SVM的全极化、双极化与单极化SAR图像分类性能的比较

支持向量机(SVM)以其在小训练样本时良好的分类性能,目前已广泛应用于多个领域.本文在极化SAR图像特征提取基础上,将SVM应用于极化SAR图像分类,定性和定量地比较了全极化、双极化和单极化SAR图像的分类性能,分析了不同的极化组合对分类结果的影响,并根据地物极化散射特性分析了分类精度差异的成因.实测极化SAR数据的实验结果表明,全极化数据能获得最好的分类性能,双极化次之,单极化最低,且在某些情况下,双极化与全极化分类性能接近.     

地表覆盖分类数据对区域森林叶面积指数反演的影响

以江西省吉安市为研究区,将5种全球地表覆盖分类数据(包括美国地质调查局(USGS)、马里兰大学(UMD)和波士顿大学(BU)生成的3套数据和欧洲生成的2套数据)以及由TM影像生成的区域地表覆盖分类数据,分别与MODIS1km反射率资料结合,利用基于4尺度几何光学模型的LAI反演方法生成研究区的LAI。在1km和4km两种尺度上将反演的LAI与TM资料生成的LAI进行比较,评价地表覆盖分类数据对LAI反演结果的影响。结果表明,TM和欧洲太空局的GLOBCOVER地表覆盖分类数据用于反演LAI的结果较好,在1km尺度上,反演的LAI与统计模型估算的TMLAI相关的R2分别为0.44和0.40,在4km尺度上的R2分别为0.57和0.54;其次为波士顿大学的MODIS地表覆盖分类数据,据其反演的LAI与TMLAI相关的R2在1km和4km尺度上分别为0.38和0.51;而马里兰大学的UMD和欧洲的GLC2000地表覆盖分类数据会导致反演的LAI存在较大误差,据其反演的LAI与TMLAI之间的一致性较差,在1km和4km两种尺度上平均偏低20%左右;LAI的反演结果对聚集度系数具有强的敏感性。该研究表明,为了提高区域/全球LAI反演精度,需要有高质量的地表覆盖分类数据。