Object-based land cover classification using airborne LiDAR

Light Detection and Ranging (LiDAR) provides high resolution horizontal and vertical spatial point cloud data, and is increasingly being used in a number of applications and disciplines, which have concentrated on the exploit and manipulation of the data using mainly its three dimensional nature. LiDAR information potential is made even greater though, with its consideration of intensity.Elevation and intensity airborne LiDAR data are used in this study in order to classify forest and ground types quickly and efficiently without the need for manipulating multispectral image files, using a supervised object-orientated approach. LiDAR has the advantage of being able to create elevation surfaces that are in 3D, while also having information on LiDAR intensity values, thus it is a spatial and spectral segmentation tool. This classification method also uses point distribution frequency criteria to differentiate between land cover types. Classifications were performed using two methods, one that included the influence of the ground in heavily vegetated areas, and the other which eliminated the ground points before classification. The classification of three meanders of the Garonne and Allier rivers in France has demonstrated overall classification accuracies of 95% and 94% for the methods including and excluding the ground influence respectively. Five types of riparian forest were classified with accuracies between 66 and 98%. These forest types included planted and natural forest stands of different ages. Classifications of short vegetation and bare earth also produced high accuracies averaging above 90%.     

Livestock detection in aerial images using a fully convolutional network

Computational Visual Media - In order to accurately count the number of animals grazing on grassland, we present a livestock detection algorithm using modified versions of U-net and Google...     

深度卷积神经网络特征提取用于地表覆盖分类初探

目的 地表覆盖监测是生态环境变化研究、土地资源管理和可持续发展的重要基础,在全球资源监测、全球变化检测中发挥着重要作用。提高中等分辨率遥感影像地表覆盖分类的精度具有非常重要的意义。方法 近年来,深度卷积神经网络在图像分类、目标检测和图像语义分割等领域取得了一系列突破性的进展,相比于传统的机器学习方法具有更强的特征学习和特征表达能力。基于其优越的特性,本文进行了深度卷积神经网络对中分辨率遥感影像进行特征提取和分类的探索性研究。以GF-1的16 m空间分辨率多光谱影像为实验数据,利用预训练好的AlexNet深度卷积神经网络模型进行特征提取,以SVM为分类器进行分类。分析了AlexNet不同层的特征以及用于提取特征的邻域窗口尺寸对分类结果的影响,并与传统的单纯基于光谱特征和基于光谱+纹理特征的分类结果进行对比分析。结果 结果表明在用AlexNet模型提取特征进行地表覆盖分类时,Fc6全连接层是最有效的特征提取层,最佳的特征提取窗口尺寸为9×9像素,同时利用深度特征得到的总体分类精度要高于其他两种方法。结论 深度卷积神经网络可以提取更精细更准确的地表覆盖特征,得到更高的地表覆盖分类精度,为地表覆盖分类提供了参考价值。     

全卷积网络结合改进的条件随机场循环神经网络用于SAR图像场景分类

传统合成孔径雷达（SAR）图像基于粗分割像素块提取相关特征,后接支持向量机（SVM）和马尔可夫随机场（MRF）或条件随机场（CRF）进行分类,该方法存在同一像素块内部不同类别像素的误差,而且只考虑邻近区域未充分用到全局信息和结构信息。故考虑基于像素点引入全卷积网络（FCN）,以ESAR卫星图像为样本,基于像素点级别构建卷积网络进行训练,得到各像素的初始类别分类概率。为了考虑全局像素类别的影响后接CRF-循环神经网络（CRF-RNN）,利用FCN得到的初始概率,结合CRF结构得到全局像素类别转移结果,之后进行RNN的迭代进一步优化实验结果。由于基于像素点和考虑了全局信息与结构信息,克服了传统分类的部分缺点,使正确率较传统SVM或CRF方法平均提高了约6.5个百分点。由于CRF-RNN的距离权重是用高斯核人为拟合的,不能随实际训练样本来改变和确定,故存在一定误差,针对该问题提出可训练的全图距离权重卷积网络来改进CRF-RNN,最终实验结果表明改进后方法的正确率较未改进的CRF-RNN又提高了1.04个百分点。     

Investigation of the effect of the incidence angle on land cover classification using fully polarimetric SAR images

Incidence angle is one of the most important imaging parameters that affect polarimetric SAR (PolSAR) image classification. Several studies have examined the land cover classification capability of PolSAR images with different incidence angles. However, most of these studies provide limited physical insights into the mechanism how the variation of incidence angle affects PolSAR image classification. In the present study, land cover classification was conducted by using RADARSAT-2 Wide Fine Quad-Pol (FQ) images acquired at different incidence angles, namely, FQ8 (27.75°), FQ14 (34.20°), and FQ20 (39.95°). Land cover classification capability was examined for each single-incidence angle image and a multi-incidence angle image (i.e., the combination of single-incidence angle images). The multi-incidence angle image produced better classification results than any of the single-incidence angle images, and the different incidence angles exhibited different superiorities in land cover classification. The effect mechanisms of incidence angle variation on land cover classification were investigated by using the polarimetric decomposition theorem that decomposes radar backscatter into single-bounce scattering, double-bounce scattering and volume scattering. Impinging SAR easily penetrated crops to interact with the soil at a small incidence angle. Therefore, the difference in single-bounce scattering between trees and crops was evident in the FQ8 image, which was determined to be suitable for distinguishing between croplands and forests. The single-bounce scattering from bare lands increased with the decrease in incidence angles, whereas that from water changed slightly with the incidence angle variation. Consequently, the FQ8 image exhibited the largest difference in single-bounce scattering between bare lands and water and produced the fewest confusion between them among all the images. The single- and double-bounce scattering from urban areas and forests increased with the decrease in incidence angles. The increase in single- and double-bounce scattering from urban areas was more significant than that from forests because C-band SAR could not easily penetrate the crown layer of forests to interact with the trunks and ground. Therefore, the FQ8 image showed a slightly better performance than the other images in discriminating between urban areas and forests. Compared with other crops and trees, banana trees caused stronger single- and double-bounce scattering because of their large leaves. As a large incidence angle resulted in a long penetration path of radar waves in the crown layer of vegetation, the FQ20 image enhanced the single- and double-bounce scattering differences between banana trees and other vegetation. Thus, the FQ20 image outperformed the other images in identifying banana trees.     

Boosting a fast neural network for supervised land cover classification

It is demonstrated that the use of an ensemble of neural networks for routine land cover classification of multispectral satellite data can lead to a significant improvement in classification accuracy. Specifically, the AdaBoost.M1 algorithm is applied to a sequence of three-layer, feed-forward neural networks. In order to overcome the drawback of long training time for each network in the ensemble, the networks are trained with an efficient Kalman filter algorithm. On the basis of statistical hypothesis tests, classification performance on multispectral imagery is compared with that of maximum likelihood and support vector machine classifiers. Good generalization accuracies are obtained with computation times of the order of 1 h or less. The algorithms involved are described in detail and a software implementation in the ENVI/IDL image analysis environment is provided.     

An artificial immune network approach to multi-sensor land use/land cover classification

An optimized artificial immune network-based classification model, namely OPTINC, was developed for remote sensing-based land use/land cover (LULC) classification. Major improvements of OPTINC compared to a typical immune network-based classification model (aiNet) include (1) preservation of the best antibodies of each land cover class from the antibody population suppression, which ensures that each land cover class is represented by at least one antibody; (2) mutation rates being self-adaptive according to the model performance between training generations, which improves the model convergence; and (3) incorporation of both Euclidean distance and spectral angle mapping distance to measure affinity between two feature vectors using a genetic algorithm-based optimization, which helps the model to better discriminate LULC classes with similar characteristics. OPTINC was evaluated using two sites with different remote sensing data: a residential area in Denver, CO with high-spatial resolution QuickBird image and LiDAR data, and a suburban area in Monticello, UT with HyMap hyperspectral imagery. A decision tree, a multilayer feed-forward back-propagation neural network, and aiNet were also tested for comparison. Classification accuracy, local homogeneity of classified images, and model sensitivity to training sample size were examined. OPTINC outperformed the other models with higher accuracy and more spatially cohesive land cover classes with limited salt-and-pepper noise. OPTINC was relatively less sensitive to training sample size than the neural network, followed by the decision tree.     

The skin cancer classification using deep convolutional neural network

Multimedia Tools and Applications - This paper addresses the demand for an intelligent and rapid classification system of skin cancer using contemporary highly-efficient deep convolutional neural...     

Improving Thematic Mapper land cover classification using filtered data

In an attempt to alleviate the classification problems introduced by the higher spatial resolution of the Thematic Mapper in comparison to the Muitispectral Scanner, classifications were performed on two to six band combinations, first using Thematic Mapper bands only, and subsequently replacing band 5 by its mean-filtered and median-filtered counterpart. The combination of filtered data with non-filtered data smooths out scene noise while retaining some of the boundary detail.     

全卷积神经网络下的多光谱遥感影像分割

目的 传统的遥感影像分割方法需要大量人工参与特征选取以及参数选择,同时浅层的机器学习算法无法取得高精度的分割结果。因此,利用卷积神经网络能够自动学习特征的特性,借鉴处理自然图像语义分割的优秀网络结构,针对遥感数据集的特点提出新的基于全卷积神经网络的遥感影像分割方法。方法 针对遥感影像中目标排列紧凑、尺寸变化大的特点,提出基于金字塔池化和DUC（dense upsampling convolution）结构的全卷积神经网络。该网络结构使用改进的DenseNet作为基础网络提取影像特征,使用空间金字塔池化结构获取上下文信息,使用DUC结构进行上采样以恢复细节信息。在数据处理阶段,结合遥感知识将波段融合生成多源数据,生成植被指数和归一化水指数,增加特征。针对遥感影像尺寸较大、采用普通预测方法会出现拼接痕迹的问题,提出基于集成学习的滑动步长预测方法,对每个像素预测14次,每次预测像素都位于不同图像块的不同位置,对多次预测得到的结果进行投票。在预测结束后,使用全连接条件随机场（CRFs）对预测结果进行后处理,细化地物边界,优化分割结果。结果 结合遥感知识将波段融合生成多源数据可使分割精度提高3.19%;采用基于集成学习的滑动步长预测方法可使分割精度较不使用该方法时提高1.44%;使用全连接CRFs对预测结果进行后处理可使分割精度提高1.03%。结论 针对宁夏特殊地形的遥感影像语义分割问题,提出基于全卷积神经网络的新的网络结构,在此基础上采用集成学习的滑动步长预测方法,使用全连接条件随机场进行影像后处理可优化分割结果,提高遥感影像语义分割精度。     

Evidential fully convolutional network for semantic segmentation

Applied Intelligence - We propose a hybrid architecture composed of a fully convolutional network (FCN) and a Dempster-Shafer layer for image semantic segmentation. In the so-called evidential FCN...     

Crop and land cover classification in Iran using Landsat 7 imagery

Remote sensing provides one way of obtaining more accurate information on total cropped area and crop types in irrigated areas. The technique is particularly well suited to arid and semi‐arid areas where almost all vegetative growth is associated with irrigation. In order to obtain more information with regard to crop patterns in the irrigated areas in the Zayandeh Rud basin, a classification analysis was made of the Landsat 7 image of 2 July 2000. The target of the classification was to primarily focus on the agricultural land use. The date of the image fell in the transition period where the first crops were harvested and many fields were being prepared for the second crop. The image has therefore captured an instantaneous picture of a system generally in transition from the first to the second crop, but with significant differences from system to system, both with respect to crop types and agricultural cycles. The overall accuracy of image registration was about 30 m (one pixel). Fieldwork was conducted on various occasions in August–October 2000 and May–October 2001. Farmers were interviewed to determine the situation on 2 July 2000. Fields were mapped in detail with the GPS instruments, and data compiled for 112 fields. Using a supervised classification system, training areas were selected and initial classifications were made to determine the validity of the classes. After merging several classes and testing several new classes a final classification system was made. All seven Landsat bands were used in the determination of the feature statistics. The final classification was made with the minimum distance algorithm. The statistics with respect to areas and crop type for the districts was obtained by crossing the raster map with the irrigation district raster map. The results with respect to crop type and total irrigated area per district were compared with those of previous studies. This included both NOAA/AVHRR and conventional agricultural district statistics.     

Advances in classification for land cover mapping using SPOT HRV imagery

Abstract

High-resolution data from the HRV (High Resolution Visible) sensors onboard the SPOT-1 satellite have been utilized for mapping semi-natural and agricultural land cover using automated digital image classification algorithms. Two methods for improving classification performance are discussed. The first technique involves the use of digital terrain information to reduce the effects of topography on spectral information while the second technique involves the classification of land-cover types using training data derived from spectral feature space. Test areas in Snowdonia and the Somerset Levels were used to evaluate the methodology and promising results were achieved. However, the low classification accuracies obtained suggest that spectral classification alone is not a suitable tool to use in the mapping of semi-natural cover types.     

Land cover classification in rugged areas using simulated moderateresolution remote sensor data and an artificial neural network

Rugged land cover classification accuracies produced by an artificial neural network (ANN) using simulated moderate-resolution remote sensor data exceed overall accuracies produced using the maximum likelihood rule (MLR). Land cover in spatially-complex areas and at broad spatial scales may be difficult to monitor due to ambiguities in spectral reflectance information produced from cloud-related and topographic effects, or from sampling constraints. Such ambiguities may produce inconsistent estimates of changes in vegetation status, surface energy balance, run-off yields, or other land cover characteristics. By use of a 'back-classification' protocol, which uses the same pixels for testing as for training the classifier, tests of ANN versus MLR-based classifiers demonstrated the ANNbased classifier equalled or exceeded classification accuracies produced by the MLR-based classifier in five of six land cover classes evaluated.     

Examining region-based methods for land cover classification using stochastic distances

A recent alternative to standard pixel-based classification of remote-sensing data is region-based classification, which has proved to be particularly useful when analysing high-resolution imagery of complex environments, such as urban areas, or when addressing noisy data, such as synthetic aperture radar (SAR) images. First, following certain criteria, the imagery is decomposed into homogeneous regions, and then each region is classified into a class of interest. The usual method for region-based classification involves using stochastic distances, which measure the distances between the pixel distributions inside an unknown region and the representative distributions of each class. The class, which is at the minimum distance from the unknown region distribution, is assigned to the region and this procedure is termed stochastic minimum distance classification (SMDC). This study reports the use of methods derived from the original SMDC, Support Vector Machine (SVM), and graph theory, with the objective of identifying the most robust and accurate classification methods. The equivalent pixel-based versions of region-based analysed methods were included for comparison. A case study near the Tapajós National Forest, in Pará state, Brazil, was investigated using ALOS PALSAR data. This study showed that methods based on the nearest neighbour, derived from SMDC, and SVM, with a specific kernel function, are more accurate and robust than the other analysed methods for region-based classification. Furthermore, pixel-based methods are not indicated to perform the classification of images with a strong presence of noise, such as SAR images.     

A study of land cover classification using polarimetric SAR parameters

Classification of the Earth's surface constitutes an important application of polarimetric synthetic aperture radar (SAR) data; in turn, it may represent an efficient way for investigating their different representations. The polarimetric parameters most frequently taken into account for classification have been the incoherent ones. A similar use of coherent methods appears to have been scarcely considered and remained neglected. In this contribution, we wish to address this issue, testing and comparing a wide range of polarimetric SAR parameters, coherent and incoherent. Another original aspect of this work is the study of the dependence of the classification results on the varying size of averaging windows of pixels. Such an analysis will permit us to evaluate the importance of speckle reduction and to prove if the chosen polarimetric parameters describe only point‐like physical properties of the targets or if they also contain ‘extended’, local information. The goal is to provide an objective estimate of the quality of the classification of polarimetric parameters and afford their comparison, an exercise hitherto unavailable in the literature in common knowledge.     

全卷积网络电线识别方法

目的 电线预警对于直升机和无人飞行器的低空飞行安全至关重要,利用可见光和红外图像识别电线是一个有效途径。传统识别方法需要人工设计的滤波器提取电线的局部特征,再使用Hough变换等方法找出直线,支持向量机和随机森林等机器学习方法仅给出图像中有无电线的识别结果。本文提出一种基于全卷积网络的电线识别方法,能在自动学习特征提取器的同时得到电线的具体位置等信息。方法 首先利用复杂背景生成大量包含电线图像和像素标签的成对仿真数据;然后改进U-Net网络结构以适应电线识别任务,使用仿真数据进行网络训练。由于图像中电线所占的像素很少,因此采用聚焦损失函数以平衡大量负样本的影响。结果 在一个同时包含红外图像和可见光图像各4 000幅的电力巡线数据集上,与VGG（visual geometry group）16等多种特征的随机森林方法相比,本文方法的电线识别率达到了99%以上,而虚警率不到2%;同时,本文方法输出的像素分割结果中,电线基本都能被识别出来。结论 本文提出的全卷积网络电线识别方法能够提取电线的光学图像特征,而且与传统机器学习方法相比能将电线从场景中精确提取出来,使得识别结果更加有判断的依据。     

Object-oriented classification of land use/cover using digital aerial orthophotography

In automatic/semiautomatic mapping of land use/cover using very high resolution remote-sensing imagery, the major challenge is that a single class of land use contains ground targets with varied spectral values, textures, geometries and spatial features. Here we present an object-oriented strategy for automatic/semiautomatic classifications of land use/cover using very high resolution remote-sensing data. The strategy consists of character detecting, object positioning and coarse classification, then refining the classification result step by step. The strategy combines the form classification of the objects located on the same level by using spectral values, textures and geometric features with function classification by using spatial logic relationships existing among the objects on the same level or between different levels. Furthermore, it overcomes the problem of transformation from form classification to function classification and unifies land use classification and land cover classification organically. Such an approach not only achieves high classification accuracy, but also avoids the salt-and-pepper effect found in conventional pixel-based procedures. The borderlines of the classification result are clear, the patches are pure, and the classification objects exactly match the ground targets distributed across the study site. A feasible technical strategy for the large-scale application is discussed in this article.     

Global land cover classification at 1 km spatial resolution using a classification tree approach

This paper on reports the production of a 1 km spatial resolution land cover classification using data for 1992-1993 from the Advanced Very High Resolution Radiometer (AVHRR). This map will be included as an at-launch product of the Moderate Resolution Imaging Spectroradiometer (MODIS) to serve as an input for several algorithms requiring knowledge of land cover type. The methodology was derived from a similar effort to create a product at 8 km spatial resolution, where high resolution data sets were interpreted in order to derive a coarse-resolution training data set. A set of 37 294 x 1 km pixels was used within a hierarchical tree structure to classify the AVHRR data into 12 classes. The approach taken involved a hierarchy of pair-wise class trees where a logic based on vegetation form was applied until all classes were depicted. Multitemporal AVHRR metrics were used to predict class memberships. Minimum annual red reflectance, peak annual Normalized Difference Vegetation Index (NDVI), and minimum channel three brightness temperature were among the most used metrics. Depictions of forests and woodlands, and areas of mechanized agriculture are in general agreement with other sources of information, while classes such as low biomass agriculture and high-latitude broadleaf forest are not. Comparisons of the final product with regional digital land cover maps derived from high-resolution remotely sensed data reveal general agreement, except for apparently poor depictions of temperate pastures within areas of agriculture. Distinguishing between forest and non-forest was achieved with agreements ranging from 81 to 92% for these regional subsets. The agreements for all classes varied from an average of 65% when viewing all pixels to an average of 82% when viewing only those 1 km pixels consisting of greater than 90% one class within the high-resolution data sets.     

Object-based land cover classification of shaded areas in high spatial resolution imagery of urban areas: A comparison study

A significant proportion of high spatial resolution imagery in urban areas can be affected by shadows. Considerable research has been conducted to investigate shadow detection and removal in remotely sensed imagery. Few studies, however, have evaluated how applications of these shadow detection and restoration methods can help eliminate the shadow problem in land cover classification of high spatial resolution images in urban settings. This paper presents a comparison study of three methods for land cover classification of shaded areas from high spatial resolution imagery in an urban environment. Method 1 combines spectral information in shaded areas with spatial information for shadow classification. Method 2 applies a shadow restoration technique, the linear-correlation correction method to create a “shadow-free” image before the classification. Method 3 uses multisource data fusion to aid in classification of shadows. The results indicated that Method 3 achieved the best accuracy, with overall accuracy of 88%. It provides a significantly better means for shadow classification than the other two methods. The overall accuracy for Method 1 was 81.5%, slightly but not significantly higher than the 80.5% from Method 2. All of the three methods applied an object-based classification procedure, which was critical as it provides an effective way to address the problems of radiometric difference and spatial misregistration associated with multisource data fusion (Method 3), and to incorporate thematic spatial information (Method 1).