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。     

级联卷积神经网络的遥感影像飞机目标检测

传统遥感影像飞机目标检测算法依赖于人工设计特征，对大范围复杂场景和多尺度的飞机目标稳健性较差，基于深层卷积神经网络的目标检测算法通常难以有效应对大幅影像的目标搜索和弱小目标检测问题，针对上述问题，本文提出了一种基于级联卷积神经网络的遥感影像飞机目标检测算法。首先根据全卷积神经网络能够支持输入任意大小图像的特点，采用小尺度浅层全卷积神经网络对整幅影像进行遍历和搜索，快速获取疑似飞机目标作为兴趣区域，然后利用较深层的卷积神经网络对兴趣区域进行更精确的目标分类与定位。为提高卷积神经网络对地物目标的辨识能力，在卷积层中引入多层感知器，并在训练过程中采取多任务学习与离线难分样本挖掘的策略；在测试阶段，建立影像金字塔进行多级搜索，并结合非极大值抑制消除冗余窗口，从而实现由粗到精的飞机目标检测与识别。对多个数据集下多种复杂场景的遥感影像进行测试，结果表明，本文方法具有较高的准确性和较强的稳健性，可为大幅遥感影像的飞机目标检测问题提供一个快速高效的解决方案。     

栅格DEM微地形分类的BP神经网络法

栅格DEM微地形分类是数字地形精细化应用的基础，基于规则化知识的栅格DEM微地形分类方法存在自动化程度低、分类残缺等问题。本文利用BP神经网络的优势构建了栅格DEM微地形分类的人工智能方法与实现途径。以山体部位分类为微地形分类典型样例进行试验验证与分析，试验结果表明，栅格DEM微地形分类的BP神经网络法较已有的地形因子叠加分析方法存在明显优势，不仅在流程上可避免烦琐的数据叠加分析过程，而且分类结果的完整性和错分率都得到有效改善；在山体部位分出的6种微地形中，冲积地对该方法适应性最强，准确率为100%，背坡的适应性最弱准确率为89.23%。     

Techniques of classification for landuse/landcover with special reference to forest type mapping in Jaldapara Wildlife Sanctuary, Jalpaiguri District, West Bengal—a case study

The accuracy of three classification techniques namely Maximum likelihood, contextual and neural network for landuse/landcover with special emphasis on forest type mapping was evaluated in Jaldapara Wildlife Sanctuary area using IRS-1B LISS II data of Dec. 1994. The area was segregated into ten categories by using all the three classification techniques taking same set of training areas. The classification accuracy was evaluated from the error matrix of same set of training and validating pixels. The analysis showed that the neural net work achieved maximum accuracy of 95 percent, maximum likelihood algorithm with 91.06 percent and contextual classifier with 87.42 percent. It is concluded that the neural network classifier works better in heterogeneous and contextual in homogenous forestlands whereas the maximum likelihood is the best in both the conditions.     

几种具有编解码结构的深度学习模型在建筑物综合中的应用与比较

深度学习技术促使诸多领域研究取得突破性进展, 基于深度神经网络的地图综合研究备受期待。将建筑物综合过程抽象解释为编解码过程, 系统地研究基于编解码结构的深度神经网络在建筑物综合中的应用。首先, 利用空间划分与矢量-栅格数据转换相结合的方式构建样本和样本集; 然后, 利用样本集训练基于编解码结构的深度神经网络, 实现建筑物综合学习泛化并测试、评估其效果; 最后, 搭建5种代表性的基于编解码结构的深度神经网络, 分析比较各模型在建筑物综合中的应用效果。实验结果表明, 基于编解码结构的深度神经网络能够从建筑物综合样本中学习或推理出部分建筑物综合知识和综合操作, 且5种模型中Pix2Pix更适用于建筑物综合的学习模拟。     

BP神经网络的道路场景杆状地物自动分类

针对车载激光扫描数据中杆状地物分类精度不高、自动化程度低的问题，本文提出一种基于BP神经网络的分类方法。首先根据杆状地物点云特征选取10个特征值，获取杆状地物聚类单元的特征向量，构建特征矩阵；然后使用样本集训练BP神经网络模型并保存该分类模型；最后使用BP神经网络分类模型对试验区内的杆状地物进行分类。试验结果表明，该方法对杆状地物的分类精度可达95.34%，验证了文中所述方法对杆状地物分类的有效性。     

BP神经网络用于GPS高程转换的网络配置

BP神经网络的输入与输出关系是一个高度非线性映射关系,其用于GPS高程转换中有着较高的精度。但它存在不少问题,如网络的隐含层和隐含层节点个数选取尚无理论上的指导,参加学习的样本的质量如何影响仿真精度等。本文结合实例分析了上述问题,从而得出了BP神经网络用于GPS高程转换时网络配置问题的一些相关结论。     

自组织网络在遥感土地覆盖分类中应用研究

设计完成和比较了自组织网络的几种算法在遥感土地覆盖分类中的应用,结果表明非监督和监督学习结合方法进行遥感土地覆盖分类,各算法在分类性能上无显著差异,因此可采用算法和较简单的单竞争学习网络,根据最邻近原则进行非参数分类。     

多结构卷积神经网络特征级联的高分影像土地利用分类

高分辨率遥感影像包含丰富的土地利用类型信息，针对单一卷积神经网络提取图像特征信息不足的问题，提出了一种多结构卷积神经网络（convolutional neural network，CNN）特征级联的分类方法。首先，选择CaffeNet（convolutional architecture for fast feature embedding）、VGG-S（visual geometry group-slow）、VGG-F（visual geometry group-fast）为实验初始模型，对网络全连接层进行参数微调，采用随机梯度下降法（stochasticgradient descent，SGD）更新网络的权重；然后以微调后的网络分别作为特征提取器对图像提取特征，级联上述3种网络的第二个全连接层输出特征作为图像表达；最后，以多类最优边界分配机（multi-class optimal margindistribution machine，mcODM）获得最终分类结果。实验采用UC Merced land-use数据集进行分类效果检验，结果表明，多结构卷积神经网络级联的方法能够达到97.55%的总体分类精度，相较于CaffeNet、VGG-S和VGG-F等，分类精度分别提升了5.71%、2.72%和5.1%。因此多结构卷积神经网络特征级联的方法能够有效提取目标特征信息，提升土地利用分类精度。     

Development of a 1-km landcover dataset of China using AVHRR data

This paper describes the development of a 1-km landcover dataset of China by using monthly NDVI data spanning April 1992 through March 1993. The method used combined unsupervised and supervised classification of NDVI data from AVHRR. It is composed of five steps: (a) unsupervised clustering of monthly AVHRR NDVI maximum value composites is performed using the ISOCLASS algorithm; (b) preliminary identification is carried out with the addition of digital elevation models, eco-region data and a collection of other landcover/vegetation reference data to identify the clusters with single landcover classes; (c) re-clustering is performed of clusters with size greater than a given threshold value and containing two or more disparate landcover classes; (d) cluster combining is performed to combine all clusters with a single landcover class in one cluster, and all other clusters into one mixed cluster; and (e) supervised classification is used to carry out post-classification of the mixed cluster generated in the previous step by using the maximum likelihood algorithm and the identified single landcover classes of the previous step as training data. The classification is based on extensive use of computer-assisted image processing and tools, as well as the skills of the human interpreter to take the final decisions regarding the relationship between spectral classes defined using unsupervised methods and landscape characteristics that are used to define landcover classes.     

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.     

基于ETM＋制作土地利用覆盖图——以制作北京1：5万土地利用覆盖图为例

简要介绍了基于LANDSAT7 ETM＋影像,采用计算机非监督分类、监督分类与人工解译相结合的方法制作土地利用覆盖图的过程和所采用的关键技术,给出了适用于规模化生产土地利用覆盖数据的工艺流程图。使用该方法制作的十一种分类要素的北京地区1：5万土地利用覆盖图,平均分类精度为84．85％,可以满足一般用户对土地利用覆盖图的要求。     

Neural Networks: A General Framework for Non-Linear Function Approximation

The focus of this paper is on the neural network modelling approach that has gained increasing recognition in GIScience in recent years. The novelty about neural networks lies in their ability to model non‐linear processes with few, if any, a priori assumptions about the nature of the data‐generating process. The paper discusses some important issues that are central for successful application development. The scope is limited to feedforward neural networks, the leading example of neural networks. It is argued that failures in applications can usually be attributed to inadequate learning and/or inadequate complexity of the network model. Parameter estimation and a suitably chosen number of hidden units are, thus, of crucial importance for the success of real world neural network applications. The paper views network learning as an optimization problem, reviews two alternative approaches to network learning, and provides insights into current best practice to optimize complexity so to perform well on generalization tasks.     

Multisensor integration using neuron computing for land-vehicle navigation

Most of the present navigation sensor integration techniques are based on Kalman-filtering estimation procedures. Although Kalman filtering represents one of the best solutions for multisensor integration, it still has some drawbacks in terms of stability, computation load, immunity to noise effects and observability. Furthermore, Kalman filters perform adequately only under certain predefined dynamic models. Neuron computing, a technology of artificial neural network (ANN), is a powerful tool for solving nonlinear problems that involve mapping input data to output data without having any prior knowledge about the mathematical process involved. This article suggests a multisensor integration approach for fusing data from an inertial navigation system (INS) and differential global positioning system (DGPS) hardware utilizing multilayer feed-forward neural networks with a back propagation learning algorithm. In addition, it addresses the impact of neural network (NN) parameters and random noise on positioning accuracy. Electronic Publication     

General Regression Neural Network and Radial Basis Neural Network for the Estimation of Crop Variables of Lady Finger

Estimation of crop variables is necessary for crop type monitoring as well as crop yield forecast. At the present era artificial neural network methodology are widely used to the remote sensing domain for numerous applications like crop yield forecasting and crop type classification. In the present work, two neural network models namely general regression neural network (GRNN) and radial basis function neural network (RBFNN) are used to estimate crop variables: leaf area index (LAI), biomass (BM), plant height (PH) and soil moisture (SM) by using ground based X-band scatterometer data. The both networks are trained and tested with X-band scatterometer data. The performance of the GRNN and RBFNN networks are found that the radial basis approach is more suitable for crop variable estimation in comparison to the GRNN approach. This work presents the applicability of neural network as an estimator and method employed could be useful to estimate the crop variables of other crops.     

融合图像和时空信息的社交媒体用户活动分类方法

社交媒体签到数据中蕴含着大量的用户活动信息。理解社交媒体用户的活动和行为类型,对探索人类的移动性和行为模式等有着重要意义。提出了一种针对新浪微博（简称为微博）的用户活动分类方法,结合图像表达和时空数据分类技术,识别微博签到数据所代表的用户活动类型。首先,根据兴趣点属性信息将微博签到数据所代表的用户活动分为餐饮、生活服务、校园、户外、娱乐、出行6大类;然后,基于卷积神经网络和K近邻分类方法,融合签到数据中的图像场景信息与时空信息,对微博用户的活动行为进行分类。实验结果表明,所提方法能够显著提高微博用户活动类型识别的准确性,为精确探索人类行为活动提供更加有效的数据支持。     

关于BP神经网络转换GPS高程的若干问题

BP神经网络用于GPS高程转换有着较高的精度,但也存在不少问题,如网络的隐含层和隐含层单元个数选取、参加学习的样本的质量如何影响仿真精度等。结合实例分析上述问题,得出一些结论。     

Development of a modified neural network-based land cover classification system using automated data selector and multiresolution remotely sensed data

Integrating multiple images with artificial neural networks (ANN) improves classification accuracy. ANN performance is sensitive to training datasets. Complexity and errors compound when merging multiple data, pointing to needs for new techniques. Kohonen's self-organizing mapping (KSOM) neural network was adapted as an automated data selector (ADS) to replace manual training data processes. The multilayer perceptron (MLP) network was then trained using automatically extracted datasets and used for classification. Two hypotheses were tested: ADS adapted from the KSOM network provides adequate and reliable training datasets, improving MLP classification performance; and fusion of Landsat thematic mapper (TM) and SPOT images using the modified ANN approach increases accuracy. ADS adapted from the KSOM network improved training data quality and increased classification accuracy and efficiency. Fusion of compatible multiple data can improve performance if appropriate training datasets are collected. This proved to be a viable classification scheme particularly where acquiring sufficient and reliable training datasets is difficult.     

一种适用于多类别遥感图象分类的新方法——复合神经网络分类方法

该文提出一种由多层神经网络与自组织神经网络相结合进行类别遥感图象分类的复合神经网络分类方法。第１步半训练样本按其统计特征分成若干组，用不同级别的训练样本分别训练ＢＰ网络。第２步将这些训练好的ＢＰ网络并联构成有监督分类器，对遥感图象进行有监督分类。第３步用ＢＰ网络的分类结果对Ｋｏｈｏｎｅｎ网络进行自组织训练，用训练好的Ｋｏｈｏｎｅｎ网络构造无监督分类器，对遥感图象进行细分。通过对ＳＰＯＴ遥感图象的分     

基于卷积神经网络的城市水体提取方法研究

针对传统方法在城市水体提取中容易受到建筑物阴影影响和难以精确提取细小水体等问题,提出了一种基于逐像元分类和多尺度分割技术的卷积神经网络遥感水体提取方法。该方法利用像元的光谱特征向量构建光谱特征矩阵,作为卷积神经网络输入特征训练水体提取模型,以多尺度分割结果抑制分类离散点与水体边缘误分现象,进一步提高提取精度。试验结果表明,该方法在细小水体的提取精度和细节上比改进的归一化水体指数算法表现更好,不仅能有效抑制建筑物阴影的影响,还能够有效区分一些相对细小的建筑对象如桥梁等,提取结果边缘也更光滑。