Classification of multitemporal Sentinel-2 data for field-level monitoring of rice cropping practices in Taiwan

Cadastral information of rice fields is important for monitoring cropping practices in Taiwan due to official initiatives. Remote sensing based rice monitoring has been a challenge for years because the size of rice fields is small, and crop mapping requires information of crop phenology, relating to spatiotemporal resolution of satellite data. This study aims to develop an approach for mapping rice-growing areas at field level using multi-temporal Sentinel-2 data in Taiwan. The data were processed for 2018, following four main steps: (1) construct time-series Enhanced Vegetation Index (EVI) and Land Surface Water Index (LSWI), (2) noise filtering of the time-series data using wavelet transform, (3) rice crop classification using information of crop phenology, and (4) parcel-based accuracy assessment of the mapping results. The parcel-to-parcel comparisons between mapping results and ground reference data indicated satisfactory results. These findings were confirmed by close agreement between satellite-derived rice area and government’s statistics. Although some factors, including mixed-pixel issues and cloud-cover effects, lowered the mapping accuracies of townships along the coastline, this study has demonstrated the efficacy of using multitemporal Sentinel-2 data to create a reliable database of rice-growing areas over a large and heterogeneous region. Such a quantitative information was important for updating rice crop maps and monitoring cropping practices.     

Monitoring of rice cropping intensity in the upper Mekong Delta,Vietnam using time-series MODIS data

Information on rice growing areas is important for policymakers to devise agricultural plans. This research explores the monitoring of rice cropping intensity in the upper Mekong Delta, Vietnam (from 2001 to 2007) using time-series MODIS NDVI 250-m data. Data processing includes three steps: (1) noise is filtered from the time-series NDVI data using empirical mode decomposition (EMD); (2) endmembers are extracted from the filtered time-series data and trained in a linear mixture model (LMM) for classification of rice cropping systems; and (3) classification results are verified by comparing them with the ground-truth and statistical data. The results indicate that EMD is a good filter for noise removal from the time-series data. The classification results confirm the validity of LMM, giving an overall accuracy of 90.1% and a Kappa coefficient of 0.7. The lowest producer and user accuracies were associated with single crop rain-fed rice class due to the mixed pixel problems. A strong yearly correlation at the district level was revealed in the MODIS-derived areas (R² ? 0.9). Investigation of interannual changes in rice cropping intensity from 2001 to 2007 showed a remarkable conversion from double to triple crop irrigated rice from 2001 to 2003, especially in the Thoai Son and Phu Tan districts. A big conversion from triple crop rice back to double crop rice cultivation was also observed in Phu Tan from 2005 to 2006. These changes were verified by visual interpretation of Landsat images and examination of NDVI profiles.     

Land use/land cover classification using time series Landsat 8 images in a heavily urbanized area

It is of great significance to timely, accurately, and effectively monitor land use/cover in city regions for the reasonable development and utilization of urban land resources. The remotely sensed dynamic monitoring of Land use/land cover (LULC) in rapidly developing city regions has increasingly depended on remote-sensing data at high temporal and spatial resolutions. However, due to the influence of revisiting periods and weather, it is difficult to acquire enough time-series images with high quality at both high temporal and spatial resolution from the same sensor. In this paper we used the temporal-spatial fusion model ESTARFM (Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model) to blend Landsat8 and MODIS data and obtain time-series Landsat8 images. Then, land cover information is extracted using an object-based classification method. In this study, the proposed method is validated by a case study of the Changsha City. The results show that the overall accuracy and Kappa coefficient were 94.38% and 0.88, respectively, and the user/producer accuracies of vegetation types were all over 85%. Our approach provides an accurate and efficient technical method for the effective extraction of land use/cover information in the highly heterogeneous regions.     

Glacier mapping based on rough set theory in the Manas River watershed

Precise glacier information is important for assessing climate change in remote mountain areas. To obtain more accurate glacier mapping, rough set theory, which can deal with vague and uncertainty information, was introduced to obtain optimal knowledge rules for glacier mapping. Optical images, thermal infrared band data, texture information and morphometric parameters were combined to build a decision table used in our proposed rough set theory method. After discretizing the real value attributes, decision rules were calculated through the decision rule generation algorithm for glacier mapping. A decision classifier based on the generated rules classified the multispectral image into glacier and non-glacier areas. The result of maximum likelihood classification (MLC) was used to compare with the result of the classification based on the rough set theory. Confusion matrix and visual interpretation were used to evaluate the overall accuracy of the results of the two methods. The accuracies of the rough set method and maximum likelihood classification were compared, yielding overall accuracies of 94.15% and 93.88%, respectively. It showed the area difference based on rough set was smaller by comparing the glacier areas of the rough set method and MLC with visual interpreter, respectively. The high accuracy for glacier mapping and the small area difference for glacier based on rough set theory demonstrated that this method was effective and promising for glacier mapping.     

Comparison of Landsat OLI,ASTER, and Sentinel 2A data in lithological mapping : A Case study of Rich area (Central High Atlas,Morocco)

The eastern part of the Rich area consists of the massive Paleozoic and Meso-Cenozoic cover formations that present the geodynamic development of the study area, where is characterized by various carbonate facies of Jurassic age. The geographical characteristic of the study area leaves the zone difficult to map by conventional methods. The objective of this work focuses on the mapping of the constituent lithological units of the study area using multispectral data of Landsat OLI, ASTER, and Sentinel 2A MSI. The processing of these data is based on a precise methodology that distinguishs and highlights the limits of the different lithological units that have an approximate similarity of spectral signature. Three techniques were used to enhance the image including Principal Component Analysis (PCA), Minimum Noise Fraction (MNF), and Independent Component Analysis (ICA). Lithological mapping was performed using two types of supervised classification : Maximum likelihood classifier (MLC) and Support Vector Machine (SVM).The results of processing data show the effectiveness of Sentinel 2A data in mapping of lithological units than the ASTER and Landsat OLI data. The classification evaluation of two methods of the Sentinel 2A MSI image showed that the SVM method give a better classification with an overall accuracy of 93,93% and a Kappa coefficient of 0.93, while the MLC method present an overall accuracy of 82,86% and a Kappa coefficient of 0.80. The results of mapping obtained show a good correlation with the geological map of the study area as well as the efficiency of remote sensing in identification of different lithological units in the Central High Atlas.     

An ensemble deep learning based shoreline segmentation approach (WaterNet) from Landsat 8 OLI images

Shorelines constantly vary due to natural, urbanization and anthropogenic effects such as global warming, population growth, and environmental pollution. Sustainable monitoring of coastal changes is vital in terms of coastal resource management, environmental preservation and planning. Publicly available Landsat 8 OLI (Operational Land Manager) images provide accurate, reliable, temporal and up-to-date information about coastal areas. Recently, the use of machine learning and deep learning algorithms have become widespread. In this study, we used our public Landsat 8 OLI satellite image dataset to create a majority voting method which is an ensemble automatic shoreline segmentation system (WaterNet) to obtain shorelines automatically. For this purpose, different deep learning architectures have been utilized namely as Standard U-Net, Dilated U-Net, Fractal U-Net, FC-DenseNet, and Pix2Pix. Also, we have suggested a novel framework to create labeling data from OpenStreetMap service to create a unique dataset called YTU-WaterNet. According to the results, IoU and F1 scores have been calculated as 99.59% and 99.79% for the WaterNet. The results indicate that the WaterNet method outperforms other methods in terms of shoreline extraction from Landsat 8 OLI satellite images.     

Selection of classification techniques for land use/land cover change investigation

The concerns over land use/land cover (LULC) change have emerged on the global stage due to the realisation that changes occurring on the land surface also influence climate, ecosystem and its services. As a result, the importance of accurate mapping of LULC and its changes over time is on the increase. Landsat satellite is a major data source for regional to global LULC analysis. The main objective of this study focuses on the comparison of three classification tools for Landsat images, which are maximum likelihood classification (MLC), support vector machine and artificial neural network (ANN), in order to select the best method among them. The classifiers algorithms are well optimized for the gamma, penalty, degree of polynomial in case of SVM, while for ANN minimum output activation threshold and RMSE are taken into account. The overall analysis shows that the ANN is superior to the kernel based SVM (linear, radial based, sigmoid and polynomial) and MLC. The best tool (ANN) is then applied on detecting the LULC change over part of Walnut Creek, Iowa. The change analysis of the multi temporal images indicates an increase in urban areas and a major shift in the agricultural practices.     

Is the phenology-based algorithm for mapping deciduous rubber plantations applicable in an emerging region of northern Laos?

Updated information of rubber plantations is essential for assessing socioeconomic and environmental impacts, especially in the emerging region of northern tropics. Here, a phenological method was modified to detect rubber plantations using Landsat Operational Land Imager (OLI) imagery in Phongsaly Province of northern Laos, where it begun a rubber boom in the mid-2000s due to geo-economic cooperation. It highlighted the landscape and pixel differences of deciduous rubber plantations in the tri-temporal phases (i.e., pre-defoliation, defoliation, and foliation) during the dry season due to phenological changes. Six commonly used vegetation indices (VIs), including the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Land Surface Water Index (LSWI), Atmospherically Resistant Vegetation Index (ARVI), Normalized Burn Ratio (NBR and NBR2) derived from OLI imagery during 2013–2016 were compared to determine the most suitable VI for discriminating the phenological differences of rubber plantations from natural forests. Then, the Differences of Normalized Burn Ratio (DNBR) was applied to generate the 30 m map of rubber plantations in 2016, by combining two masks of Landsat-derived forest and suitable elevation for rubber trees cultivation. The resultant map of rubber plantations had a classification accuracy of 93.7% and the Kappa coefficient of 0.848. Our study demonstrated the usefulness of the Landsat-derived tri-temporal phenological DNBR approach in an emerging region of northern Laos, despite requiring more scenes compared with single- and double temporal window methods.     

Remote sensing and GIS-based landslide hazard analysis and cross-validation using multivariate logistic regression model on three test areas in Malaysia

This paper presents the results of the cross-validation of a multivariate logistic regression model using remote sensing data and GIS for landslide hazard analysis on the Penang, Cameron, and Selangor areas in Malaysia. Landslide locations in the study areas were identified by interpreting aerial photographs and satellite images, supported by field surveys. SPOT 5 and Landsat TM satellite imagery were used to map landcover and vegetation index, respectively. Maps of topography, soil type, lineaments and land cover were constructed from the spatial datasets. Ten factors which influence landslide occurrence, i.e., slope, aspect, curvature, distance from drainage, lithology, distance from lineaments, soil type, landcover, rainfall precipitation, and normalized difference vegetation index (ndvi), were extracted from the spatial database and the logistic regression coefficient of each factor was computed. Then the landslide hazard was analysed using the multivariate logistic regression coefficients derived not only from the data for the respective area but also using the logistic regression coefficients calculated from each of the other two areas (nine hazard maps in all) as a cross-validation of the model. For verification of the model, the results of the analyses were then compared with the field-verified landslide locations. Among the three cases of the application of logistic regression coefficient in the same study area, the case of Selangor based on the Selangor logistic regression coefficients showed the highest accuracy (94%), where as Penang based on the Penang coefficients showed the lowest accuracy (86%). Similarly, among the six cases from the cross application of logistic regression coefficient in other two areas, the case of Selangor based on logistic coefficient of Cameron showed highest (90%) prediction accuracy where as the case of Penang based on the Selangor logistic regression coefficients showed the lowest accuracy (79%). Qualitatively, the cross application model yields reasonable results which can be used for preliminary landslide hazard mapping.     

Landsat thematic mapper studies of land cover spatial variability related to hydrology

The Landsat 4 and 5 Thematic Mapper (TM) provides increased spatial, spectral, and radiometric capability relative to the Multispectral Scanner (MSS). Visual inspection of TM imagery confirms this. Land cover detail is evident that would be of use in watershed management and planning activities. Specific studies have been conducted in Georgia, West Virginia, Michigan and Maryland to compare MSS and TM for urbanizing watersheds, wetlands, and floodplain mapping situations. These studies show that only modest improvements in classification accuracy (Anderson Level I/II) have been achieved using existing classification approaches. An attempt to identify the visibly apparent interstate highways and secondary and residential streets in TM data via conventional approaches failed due to an inability to derive separable spectral signatures. The basis for a non-parametric approach to classification is presented in which roads are identified by locating linear local minima in the greenness transformed dimension. Preliminary results indicate that such a method provides more reliable road locations than MSS or TM used singly.     

Integration of Landsat-8 and Sentinel-1 dataset to extract geological lineaments in complex formations of Tepal mountain area,Shahrood, north Iran

Recently, the detection and extraction of geological lineaments have become an essential analytical technique to find relationships between the characteristics and occurrence of hydrogeology, and tectonic studies. The use of remote sensing, with the progressive development of image enhancement techniques, provides an opportunity to produce more reliable and comprehensive lineament maps. In this paper, semi-automatic approach based on Landsat 8 and Sentinel 1 radar data is proposed for lineaments extraction and validation. The combined method of linear filtering and automatic line module ensures a high degree of accuracy resulting in a lineament map. Based on identified lineaments, Sentinel1 is more capable of detecting edges than Landsat8, but the primary orientation lineaments extracted from Landsat8 and Sentinel1 were different. So, by combining band6 of Landsat8, and VV and VH polarization of Sentinel1, the area lineaments were extracted with high accuracy. Rose diagram showed the extracted lineaments' orientation is in good compliance with the region's existing faults. Also, the formations' lineament length density has good consistent with the density of the faults in the geological map.     

基于对抗和迁移学习的灾害天气卫星云图分类

针对卫星云图中的灾害天气数据存在严重不平衡问题，提出一个结合生成对抗学习（GAN）和迁移学习（TL）的卷积神经网络（CNN）框架以解决上述问题进而提高基于卫星云图的灾害天气分类精度。该框架主要包含基于GAN的数据均衡化模块和基于迁移学习的CNN分类模块。上述2个模块分别从数据和算法层面解决数据的类间不平衡问题，分别得到一个相对均衡的数据集和一个可在不同类别数据上提取相对均衡特征的分类模型，最终实现对卫星云图的分类，提高其中灾害天气的卫星云图类别分类准确率。与此同时所提方法在自建的大规模卫星云图数据上进行了测试，消融性和综合实验结果证明了所提数据均衡方法和迁移学习方法是有效的，且所提框架模型对各个灾害天气类别的分类精度都有显著提升。      

Examining LiDAR – WorldView-3 data synergy to generate a detailed stand map in a mixed forest in the north-west of Turkey

Although stand delineation approach based on aerial photographs and field survey produces high accuracy maps, it is labour-intensive and time consuming. Furthermore, conventional forest stand maps may have some uncertainties that can hardly be verified due to the experiments and skills of photo-interpreters. Therefore, researchers have been seeking more objective and cost-effective methods for forest mapping. LiDAR (Light Detection and Ranging) data have a high potential to automatically delineate forest stands. Unlike optical sensors, LiDAR height data provides information about both the horizontal and vertical structural characteristics of forest stands. However, it deprives of spectral data that may be successfully used in separating tree species. In this study, we investigate the potential of LiDAR – WorldView-3 data synergy for the automatic generation of a detailed forest stand map which can be used for a tactical forest management plan. Firstly, image segmentation was applied to LiDAR data alone and LiDAR/WorldView-3 data set in order to obtain the most suitable image objects representing forest stands. Visual inspection of the segmentation results showed that image objects based on the LiDAR/WorldView-3 data set were more compatible with the reference forest stand boundaries. After the segmentation process, the LiDAR and LiDAR/WorldView-3 data sets were independently classified using object-based classification method. We tested two levels of classification. The first was a detailed classification with 14 classes considering reference stand types. The second was the rough classification with 9 classes where some stand types were combined. The mean, standard deviation and texture features of LiDAR metrics and spectral information were used in the classification. The accuracy assessment results of LiDAR data showed that the Overall Accuracy (OA) was calculated as 0.31 and 0.43, and the Kappa Index (KIA) was calculated as 0.26 and 0.32 for the detailed and rough classifications, respectively. For the LiDAR/WorldView-3 data set, the OA values were calculated as 0.50 and 0.61, and the KIA were calculated as 0.46 and 0.55 for the detailed and rough classifications, respectively. These results showed that the forest stand map derived from the LiDAR/WorldView-3 data synergy is more compatible with the reference forest stand map. In conclusion, it can be said that the forest stand maps produced in this study may provide strategic forest planning needs, but it is not sufficient for tactical forest management plans.     

基于深度神经网络的像素级别可见光图像配准

现有图像配准算法中,借助图像采集设备参数的方法存在硬件内参难以获得或精度不够的问题,采用匹配图像特征计算图像单应性的方法存在对场景深度信息利用不全的问题。针对这一现象,提出了结合可见光图像与其深度信息来生成更具有真实性的配准图像对数据,用以训练得到一个可以进行像素级别图像配准的深度神经网络PIR-Net。建立了一个大规模、多视角、超仿真的图像配准数据集:多视角配准(MVR)数据集,该数据集包含7 240对含有深度信息的待配准图像及其像素级别的坐标对准真值;基于编码器-解码器的深度神经网络结构,训练得到一个能以全分辨率形式对2幅输入图像之间的坐标变化矩阵进行重建的PIR-Net。通过实验验证了PIR-Net能够在未知相机内参的情况下实现不同视角的可见光图像配准,并比传统算法具有更高的配准精度。在MVR数据集上,PIR-Net的配准误差仅为通用的特征匹配对准算法(SIFT+RANSAC)的18%,同时减少了30%的时间消耗。      

一种深度学习分割遥感影像道路的损失函数

传统的根据光谱特征或形态学算法来分割道路，存在精度低、阈值难确定等缺点，而深度学习中已有的方法并未考虑道路的特性，只是利用通用方法来分割道路。针对上述不足，提出了一种针对道路特有形态的深度学习损失函数——形态损失函数。首先使用连通性算法将预测结果划分为若干个相互分离的连通区域，分别计算这些区域的面积与外接圆面积的比值，然后取平均值作为此批训练数据的形态损失函数，最后将形态损失函数按一定的比例与交叉熵损失函数求和，得到最终的损失函数。通过在公开的遥感数据集上使用深度学习网络进行对比试验，附加了形态损失函数后平均交并比（MIoU）、准确度（ACC）及F1 Score均有提高。从预测的图形来看，附加了形态损失函数后，预测的道路更为连续。所提出的形态损失函数可用于提高光学遥感影像道路分割的精度。     

Non-linear methods in remotely sensed multispectral data classification

The aim of this research is to develop an effective approach being able to deal with the stochastic nature of remote sensing data. In order to achieve this objective it is necessary to structure the methodological knowledge in the area of data mining and reveal the most suitable methods for the prediction and decision support based on large amounts of multispectral data. The idea is to establish a framework by decomposing the task into functionality objectives and to allow the end-user to experiment with a set of classification methods and select the best methods for specific applications. As a first step, we compare our results from Bayesian classification based on non-parametric probability density estimates of the data to the results obtained from other classification methods. Tree scenarios are considered, making use of a small benchmark dataset, a larger dataset from Corine land cover project for Bulgaria and analyzing different features and feature selection methods. We show that the theoretically optimal Bayesian classification can also achieve optimal classification in practice and provides a realistic interpretation of the world where land cover classes intergrade gradually.     

LPV模型的动态压缩测量辨识算法

在解决线性参变（LPV）模型的辨识问题上,最小二乘算法以结构简单、计算复杂度低等优点被大量使用。但最小二乘算法辨识结果受制于计算精度和模型近似精度,而这两者在同一个系统中是互斥的。因此,该算法的辨识结果与真值总是存在一定的误差。另外,在高阶LPV模型辨识或采样成本高的情况下,一般模型参数要多于辨识数据,而此时最小二乘算法很难得到稳定的辨识结果。本文提出的动态压缩测量辨识（DCMI）算法从两个方面提高在该情况下的系统辨识精度。其一,利用“匀速变化”及“非匀速变化”模型表示参变函数,以提高模型近似精度。其二,利用压缩感知理论的欠采样能力,在同等数据量的情况下提高参数的计算精度、扩大模型的计算规模。仿真结果表明,基于“匀速变化”模型DCMI算法可以准确地辨识出LPV函数,而且该算法在辨识数据不足的情况下仍然能够获得稳定的辨识结果。      

Built-up land mapping capabilities of the ASTER and Landsat ETM+ sensors in coastal areas of southeastern China

Worldwide urbanization has accelerated expansion of urban built-up lands and resulted in substantial negative impacts on the global environments. Precisely measuring the urban sprawl is becoming an increasing need. Among the satellite-based earth observation systems, the Landsat and ASTER data are most suitable for mesoscale measurements of urban changes. Nevertheless, to date the difference in the capability of mapping built-up land between the two sensors is not clear. Therefore, this study compared the performances of the Landsat-7 ETM+ and ASTER sensors for built-up land mapping in the coastal areas of southeastern China. The comparison was implemented on three date-coincident image pairs and achieved by using three approaches, including per-band-based, index-based, and classification-based comparisons. The index used is the Index-based Built-up Index (IBI), while the classification algorithm employed is the Support Vector Machine (SVM). Results show that in the study areas, ETM+ and ASTER have an overall similar performance in built-up land mapping but also differ in several aspects. The IBI values determined from ASTER were consistently higher than from ETM+ by up to 45.54% according to percentage difference. The ASTER also estimates more built-up land area than ETM+ by 5.9–6.3% estimated with the IBI-based approach or 3.9–6.1% with the SVM classification. The differences in the spectral response functions and spatial resolution between relative spectral bands of the two sensors are attributed to these different performances.