最新刊期
卷 25 ， 期 5 ， 2021
- 摘要：Global climate change is a common hot issue facing the world. For a long time, mankind believed that nature had a great decisive influence on human society, and never thought that human activities could have an impact on nature, or even a profound impact. With the accumulation of scientific knowledge and the deepening of understanding of the natural world, mankind’s concept of nature has begun to change, believing that they can transform nature, and even dominate nature. Especially after the Industrial Revolution, this process has been accelerated, causing serious damage and destruction to nature. Until all kinds of new and old natural disasters became more frequent and intense, and the spread of plagues and infectious diseases became wider, some people began to reflect on and re-examine human behavior. What kind of human behavior did harm to the earth? How did the earth react to human behavior? After the global pandemic is raging, what are the characteristics of the post-epidemic era?In the post-epidemic era, global climate change impacts and responses are showing new trends, with suddenness, globality, and relevance becoming distinctive features. At present, to focus on solving the major scientific issues of climate change, we must continue to strengthen basic research pragmatically, properly handle the relationship between the challenges of global change and sustainable development; choose our own path to respond to climate change according to my country’s national conditions; further Deal with the relationship between emission reduction and increase in sinks, mitigation and adaptation; establish a major national carbon neutral project; formulate and implement larger-scale afforestation plans; focus on the future and pay close attention to the long-term human development agenda that is highly related to climate change to achieve Sustainable human development.关键词：carbon neutrality;climate change;basic research;sustainable development
- 883
- |
- 181
- |
- 2
发布时间：2021-06-07
Scholar's View Point
-
摘要：Street view imagery is a promising and growing big geo-data that provides current and historical images in more than 200 countries for urban physical environment representation and audit. Such data not only describes the visual details of the urban physical environment but also contains information about urban functions, socioeconomics, and human dynamics. Street view imagery has the potential to complement new and traditional data, such as remote sensing imagery and social sensing data.However, traditional digital image processing techniques for street view imagery handling are limited. Extracting rich semantic information from street view imagery efficiently has always been a challenging issue. Until recently, the development of artificial intelligence has led to numerous breakthroughs in image processing and machine learning. Indeed, the last few years have witnessed the fast development of deep learning and computer vision techniques, which facilitate the understanding of scene semantics from street view imagery and the quantitative representation of the urban physical and built environments. Many new applications, novel methods, and thoughts regarding street view imagery have emerged, covering research fields, such as geography, urban planning, urban design, urban economics, public health, environmental psychology, and energy. This trend has provided new perspectives for big geo-data-driven urban environment analysis, human–land relationship study, and spatial data mining and knowledge discovery.To summarize this research trend, this paper reviews the recent works on urban physical environment analysis using street view imagery. The key supporting techniques for street view imagery analytics are discussed in terms of two dimensions: deep learning and computer vision. Deep learning has been applied recently to various computer vision tasks, such as image classification, image segmentation, and object detection. The success of deep learning techniques is attributed to their ability to learn rich high-level image representations as opposed to the hand-designed low-level features used in other image understanding methods.Additionally, this paper summarizes the street view imagery applications in three aspects: place representation, sense of place, and place semantics reasoning. “Place representation” includes works that extract visual elements that constitute the urban physical environment; “sense of place” refers to works that use street view imagery to understand how people respond to their surrounding environments regarding perceptions and emotions; and “place semantics reasoning” refers to studies that attempt to infer and estimate invisible factors, socio-economics, demographics, and human dynamics from street view imagery.More importantly, the issues of this research field, such as the spatio-temporal uniformity of street view image data, and the lack of solid workflow of data analytics, are highlighted.Finally, the development prospects of street view imagery are discussed. Crowdsourcing platforms and the field of the autonomous vehicle will increase the number of street view image sources. More issues, including how physical environments are involved and whether a universal law exists regarding the distribution of physical elements in space, are expected to be explored in future works.关键词：street view imagery;place semantics;urban physical environment;deep learning;computer vision- 1121
- |
- 633
- |
- 12
发布时间：2021-06-07 -
摘要：Discriminative Locality Alignment (DLA) is a linear feature extraction algorithm, which has exhibited effectiveness in many fields. The algorithm can deal with the nonlinearity of samples distribution, preserve discriminative information over local patches, and avoid the problem of matrix singularity in calculation. Generally, the PCA step is recommended for reducing noise. However, PCA extracts noise images with low order components, and bands with a small variance do not necessarily indicate poor image quality. To effectively reduce the influence of noise on DLA and further improve the accuracy of DLA in the feature extraction of hyperspectral remote sensing images, a linear feature extraction algorithm of MDLA and a nonlinear feature extraction algorithm of KMDLA are proposed in this paper. The key idea of MDLA is as follows. First, MNF transforms data from the original space into a new subspace. In the process, the SNR is used to improve the order of components, and the noise in the image can be effectively reduced by sorting according to the image quality. Second, DLA is performed in the new subspace to obtain the result of feature extraction. KMDLA is the kernel method of MDLA. The main idea of KMNF is as follows. First, the samples are mapped to the high-dimensional feature space by the kernel function in which MNF is conducted. Finally, DLA is performed in the subspace spanned by KMNF, further enhancing the nonlinear discriminant ability of MDLA for the samples. Three sets of hyperspectral remote sensing images were selected as the study area. A detailed comparison of the proposed algorithms and five algorithms, namely, the PCA, MNF, LDA, CCPGE, and DLA algorithms, is provided in this paper. For a more intuitive comparison, experiments were conducted to classify the images of all bands. The support vector machine classifier was used to classify the results of feature extraction. Experimental results demonstrate that both MDLA and KMDLA outperform the five algorithms of comparison. Compared with DLA, the proposed algorithms exhibited respective improvements of 2.06% and 2.74% on the Purdue Campus data, 1.91% and 2.45% on the Indian Pines data, and 1.41% and 2.04% on the Salinas data. In terms of efficiency , MDLA and KMDLA consumed less operation time for classification than DLA when the experimental data was small. KMDLA consumed the most operation time when the experimental data was large, whereas MDLA always consumed minimal operation time. The effect of image quality on the performance of different dimensionality reduction methods is discussed based on the Purdue Campus data. Results show that MDLA and KMDLA can still obtain good classification results with the increasing noise in the image. Therefore, the two proposed algorithms can effectively reduce the noise of hyperspectral remote sensing images, improve the classification accuracy, and classify hyperspectral remote sensing images accurately.关键词：hyperspectral remote sensing;feature extraction;MDLA;KMDLA;image classification- 743
- |
- 316
- |
- 0
发布时间：2021-06-07 -
摘要：As a kind of high-resolution imaging radar, Synthetic Aperture Radar (SAR) plays an important role in civil and military fields because it can realize all-weather and all-weather observation without the limitation of illumination and climate conditions. However, SAR also has limitations. For instance, the SAR image has many speckle noises, which is caused by the principle of coherent imaging that seriously affects the extraction and application of relevant information in the image. Therefore, to make better use of SAR image information, speckle noise reduction is a key step in SAR image processing. Among them, the bilateral filtering algorithm, which combines the geometric domain and the gray-scale domain information filter, is currently the best algorithm in the field of speckle noise removal. In this study, we take the bilateral filtering algorithm as the basic framework and then add corresponding improvement measures in view of the problems and shortcomings of the bilateral filtering algorithm, such as insufficient application of SAR image structure information and difficulty in effectively filtering out strong speckle noise. Finally, we propose an improved bilateral filtering algorithm based on background homogeneity (BH-IBF).This algorithm aims to effectively remove the speckle noise in the SAR image while retaining the real texture information of the image to the maximum extent. (1) The coefficient of variation is introduced into the weight kernel improvement of the bilateral filtering algorithm, compensating for the problem of the bilateral filter ignoring the structural information of the SAR image to a certain extent; (2) the sample truncation operation is introduced to filter the strong speckle noise in the background area to a certain extent, suppressing the influence of speckle noise, thus effectively solving the bilateral filtering. Strong speckle noise is difficult to filter out; (3) the half width of the background region for the homogeneous region can further improve the smoothness of the image.Taking the simulated and real SAR images intercepted from TerraSAR-X as the experimental objects, the comparison of the filtering effects and evaluation indexes of different filtering algorithms shows the improved results obtained by BH-IBF algorithm, indicating that the proposed algorithm achieves the research objective.The proposed BH-IBF has a better effect than the traditional filtering algorithm. BH-IBF can not only effectively suppress the speckle noise in the homogeneous region but also protect the edge texture information of the heterogeneous region better, that is, the algorithm can better guarantee the subsequent processing and application of SAR data.关键词：SAR image speckle noise reduction;improve bilateral filtering;adaptive sample trimming;homogeneity index of the reference window;adaptive window size- 736
- |
- 199
- |
- 3
发布时间：2021-06-07 -
摘要：Geo-objects in High-resolution Remote Sensing Images (HRSIs) have clear category attributes and rich semantic information. With the support of artificial intelligence technology, the spatial relationship can be automatically recognized by a computer. At present, the semantic understanding of HRSIs mainly relies on an image caption model to generate sentences based on the global features. However, coarse-grained features can easily cause the category attribute of the object to be mispredicted during the sentence generation process. In fact, taking the geo-object as the basic unit of semantic understanding is more in line with people’s habit of cognizing geographic space. To obtain more accurate sentences, this study constructs an Object-based Geo-spatial Relation Image Understanding Dataset (OGRIUD) and proposes a dual LSTM-driven semantic understanding method.The proposed dataset is based on the object, and the sentence description includes the category and location information of the ground object, which make up the deficiency of the target category and the location information in the semantic understanding of the current remote sensing field. The proposed method uses the object detection model to identify salient objects in the image and uses the object features as input in the language model to alleviate the problem of incorrectly predicted categories in the description. Furthermore, to use HRSI scene information, we fuse the global and regional features and use dual LSTM to predict the attention distribution of each geo-object.We compare the global feature-based approach with the object feature based approach proposed in this paper. Quantitative analysis results show that the proposed method exhibits increased exact matching accuracy, from 53.5% of the original to 62.33%. The visual analysis results show that the proposed method, and the generated spatial relation description statements are also more abundant.This method enables the language model to focus on objects with actual semantics, and the matching degree between the generated description statement and the remote sensing image content is also improved. The correspondence between the visual object and description improves the interpretability of remote sensing image understanding.关键词：high resolution remote sensing image;ground objects;spatial relationships;semantic understanding;image caption- 798
- |
- 237
- |
- 1
发布时间：2021-06-07 -
摘要：Cross-view scene image matching and positioning have a wide range of applications in target search, combating crime, and positioning. With the development of deep learning, neural networks have played an important role in this issue. Given the problem of cross-view scene image matching and positioning between street view and bird’s eye images, the neural network model’s convergence is slow, and the feature correlation is weak. This paper proposes a triplet network model (Tri-NetVLAD) that combines NetVLAD and a fully connected layer and improves DBL Loss (ADBL loss). The proposed method can not only improve the convergence speed and stability of the network but also the overall positioning accuracy of the model.The proposed Tri-NetVLAD model extracts the local features of the three input images through a triplet network and inputs the local features to the fully connected and NetVLAD layers to obtain the feature vector and the global feature descriptor. The global feature descriptor can obtain the relative distribution between features, and on this basis, incorporate feature vectors, which can preserve the differences between features to improve the positioning accuracy of the model. ADBL loss improves the model’s ability to discriminate difficult samples by introducing parameters and the positioning accuracy of the model.The proposed Tri-NetVLAD is compared with several existing methods, namely, MCVPlaces, Triplet eDBL-Net, and CVM-Net, and loss functions, namely, contrastive loss, triplet loss, and DBL loss. In the US vo and hays dataset, the highest positioning accuracy of 63.5% is achieved, proving that the triplet network that combines the NetVLAD and fully connected layers can effectively improve the positioning accuracy with the ADBL Loss.Compared with existing methods, the proposed Tri-NetVLAD has the following advantages. (1) The Triplet network can increase the Euclidean distance between unmatched images while reducing the Euclidean distance between matched images. (2) The introduction of NetVLAD can aggregate the local features extracted by CNN to obtain global feature descriptors and the distribution relationship between features. (3) The fusing of the Fully Connected Layer adds the feature vector obtained through the fully connected layer to the global feature descriptor, so that the final feature vector not only represents the distribution relationship between features, but also retains the differences between features. (4) The improved loss function ADBL Loss can accelerate the gradient convergence speed and improve the overall positioning accuracy.关键词：cross-view;scene image matching and geolocation;Triplet Network;NetVLAD;CNN- 610
- |
- 280
- |
- 3
发布时间：2021-06-07
Frontier Progress
-
摘要：Given the influence of imaging environment and equipment limits, hyperspectral images (HSIs) are often disturbed by noise. Thus, denoising is necessary for the subsequent image processing. Noise type and level are important parameters of the denoising algorithm. Furthermore, noise estimation can help people understand the image quality objectively. Many HSI noise estimation algorithms consider images to contain additive noise and measure the level of noise by estimating the statistical characteristics, such as standard deviation and co-variance. To our knowledge, the existing HSI noise estimation methods do not consider the specific type of noise. Unlike previous works, we propose a method that combines spatial and spectral domain analyses for the separation and estimation of different types of noise.Considering the characteristics of HSI noise, the noise contained in a HSI is modeled in this work as the linear combination of stripe and Gaussian noise. According to the spatial characteristics (horizontal and vertical distribution) of stripe noise, after Fourier transform, stripe noise can be represented as a specific central cross line distribution in the Fourier spectrum map. Therefore, stripe noise can be separated and quantitatively estimated by processing the pixels on the cross line. However, the useful information in the HSI may also distribute on the central cross line in the Fourier spectrum map. To eliminate the estimation error of stripe noise, the heterogeneity function is introduced to decrease the estimation error, and local mean filtering is used to further separate stripe noise and useful signal. The level of stripe noise is estimated by the sum of the pixel values on the cross line. After removing the stripe noise, a method based on multiple regression is used to extract the Gaussian noise in an image. Using the correlation among adjacent bands and the randomness of noise, a single band image can be represented by the linear combination of the remaining bands and the residual. Given that residuals can be approximately represented as a Gaussian distribution, the mean and standard deviation of the extracted residuals are calculated to describe the distribution characteristics of Gaussian noise in different bands.In the simulation experiment, image bands with stripe noise were detected successfully; the estimation Gaussian noise standard deviation is 0.0527 (theoretical value is 0.05), the mean value is near the theoretical value of 0. Furthermore, seven HSIs captured by satellites GF-5 and airborne hyperspectral imager Nano-Hyperspec were tested. The estimation results of real-world HSIs show that the mean of Gaussian noise is very near 0 for each band, which is consistent with the assumption in most denoising algorithms. For the stripe noise, some distribution rules of stripe noise are provided.In this study, we proposed a noise estimation method based on Fourier transform and multiple linear regression. This method can separate and estimate the level of the two types of noise. Experimental results show that the proposed method is efficient, and the noise levels of a HSI vary in different bands, sensors, and scenes. More importantly, the noise properties of HSIs were analyzed in this work. Some conclusions about the characteristic of HSI noise can be obtained.关键词：hyperspectral image;noise estimation;noise separation;fourier transform;stripe noise;Gaussian noise- 825
- |
- 784
- |
- 1
发布时间：2021-06-07 -
摘要：In recent years, high-resolution remote sensing image object detection has attracted increasing interest and become an important research field of computer vision due to its wide applications in civil and military fields, such as environmental monitoring, urban planning, precision agriculture, and land mapping. The natural scene object detection frameworks based on deep learning have made a breakthrough progress. These algorithms have good detection performance on the open data sets of natural scenes. However, although these algorithms have greatly improved the accuracy and speed of remote sensing image object detection, they have not achieved the expected results. Given the large variations of object sizes and inter-class similarity, most of the conventional object detection algorithms designed for natural scene images still face some challenges when directly applied to remote sensing images. To address the above challenges, we propose an end-to-end multi-resolution feature fusion framework for object detection in remote sensing images, which can effectively improve the object detection accuracy. Specifically, we use a Feature Pyramid Network (FPN) to extract multi-scale feature maps. Then, a Multi-resolution Feature Extract (MFE) module, which can promote the network to learn the feature representations of the objects at different resolutions and narrow the semantic gap between different scales, is inserted into the feature layers of different scales. Next, to achieve an effective fusion of multi-resolution features, we use an Adaptive Feature Fusion (AFF) module to obtain more discriminative multi-resolution feature representations. Finally, we use a Dual-scale Feature Deep Fusion (DFDF) module to fuse two adjacent-scale features, which are the output of the adaptive feature fusion module. In the experiments, to demonstrate the effectiveness of each module of our proposed method, including the MFE, AFF, and DFDF modules, we first conducted extensive ablation studies on the large-scale remote sensing image data set DIOR, and the experimental results show that our proposed MFE, AFF, and DFDF modules could improve the average detection accuracy by 1.4%, 0.5%, and 1.3%, respectively, compared with the baseline method. Furthermore, we evaluate our method on two publicly available remote sensing image object detection data sets, namely, DIOR and DOTA, and obtain improvements of 2.5% and 2.2%, respectively, which are measured in terms of mAP comparison with Faster R-CNN with FPN. The detection results of the ablation studies and the comparison experiments indicate that our method can extract more discriminative and powerful feature representations than Faster R-CNN with FPN, which can significantly boost the detection accuracy. Moreover, our method works well for densely arranged and multi-scale objects. Although many improvements have been achieved in this work, some aspects still require improvement. For example, our method performs poorly in terms of detecting objects with big aspect-ratios, such as bridges, possibly because most anchor-based methods have difficulty ensuring a sufficiently high intersection over union rate with the ground-truth objects with big aspect-ratios. Our future work will focus on addressing these problems by exploring the advantages of anchor-free based methods.关键词：convolutional neural networks;multi-resolution feature fusion;remote sensing images;object detection- 1230
- |
- 746
- |
- 12
发布时间：2021-06-07 -
摘要：At present, Most vegetation radiative transfer models were developed on the basis of a simplified canopy structure when simulating the interaction between solar radiation and vegetation. They retain the structure and spatial distribution characteristics of leaves but ignore the influences of wood elements (such as branches) on the reflection characteristics of a canopy. LESS, as one of the computer simulation models, can fully consider the spectral and structural characteristics of various components (leaves and branches) of vegetation and accurately simulate the process of light scattering and radiation in the canopy. Thus, it can be applied to analyze the effects of wood elements on the reflectance of a forest canopy on the basis of a reconstructed realistic three dimensional (3D) forest scene.On the basis of field data, we developed a basic framework to reconstruct a 3D scene of a complex forest with single tree as basic unit. Diameter at Breast Height (DBH) was selected as the main variable to divide trees into six levels (T1—T6). The mean DBH, mean tree height, mean crown width, and mean height of branches at level were used as typical parameters to build a tree model by using OnyxTREE. When a near-real 3D forest scene was constructed, the appropriate model in the constructed single-tree library was selected with the DBH level as the standard. The computer simulation model LESS was used to simulate the reflectance of 3D scenes of forests with and without wood elements. The effects of forest wood elements on canopy reflectance were analyzed quantitatively.Ignoring the wood elements will lead to the deviation of vegetation canopy reflectance, especially in the NIR band. The relative deviation of reflectance in the NIR band is more than 40% for all scenes with different LAIs. High spatial resolution is another important factor highlighting the influences of wood elements. As the spatial resolution increases, the deviation increases. Different grades of woody structure affect canopy reflectance; even ignoring a twig will cause an estimation error of 17.7% (NIR band). The use of wooden area instead of leaf area can partially alleviate the difference in canopy reflectance caused by completely ignoring wooden elements, but it still leads to overestimation (NIR) or underestimation (visible light) of canopy reflectance.The vegetation radiative transfer models that use statistical features to replace 3D structure distribution can no longer meet the accuracy requirements of quantitative remote sensing. Hence, the deviation caused by ignoring wood elements should be considered, specially for high-resolution remote sensing images.关键词：radiative transfer model;forest;reflectance;wood elements;3D reconstruction;LESS (LargE-Scale remote sensing data and image Simulation framework)- 705
- |
- 242
- |
- 0
发布时间：2021-06-07
Doctor's Voice
-
摘要：Sun-Induced chlorophyll Fluorescence (SIF) is a by-product of plant photosynthesis and is closely related to plant photosynthesis. The study on SIF and its relationship with Gross Primary Productivity (GPP) is of great significance in understanding the mechanism of photosynthesis. Recent instrumental and methodological developments of the tower-based SIF observation system provide a complementary capacity for measuring and interpreting chlorophyll fluorescence in the context of physiological processes. In addition, a tower-based system can also support satellite-based measurements through validation, interpretation, and data inputs provision for models. Recently, the tower-based SIF observation system has developed rapidly with varied observation methods and system characteristics. In this paper, we discuss and summarize the recent developments of tower-based SIF observation methods and propose technical specifications by comparing different tower-based SIF observation systems.Tower-based SIF observation systema can be built with either two spectrometers or one spectrometer combined with an optical path switching trigger. A two-spectrometer SIF system measures the solar incident radiance and the radiance reflected by the canopy independently to realize synchronous measurement. This system can obtain high frequency spectral data, and nearly no time gap exists between the solar incident spectrum and the spectrum reflected by the canopy, reducing the uncertainty of the retrieved SIF caused by the mismatch between the two optical channels under varied weather conditions. However, the spectral response characteristics of the two spectrometers are not completely consistent. The spectral drift between the two optical channels is difficult to correct, which may lead to the increase of the Sif retrieval uncertainty. A single-spectrometer SIF system realizes the sequential switching between the two optical channels by using an optical path switch, which allows the measurement of the solar incident radiance and the canopy reflected radiance with reliable data quality. Although a certain time gap exists between the solar incident spectrum and the reflected spectrum, it can be used for SIF retrieval because of the second disparity. In cloudy and other rapidly changing light conditions, the acquisition time gap between the spectra from the two optical channels may increase the SIF retrieval uncertainty. Compared with the dual spectrometer system, the single spectrometer system is simpler, has lower cost, and avoids the risk of spectral drift, which is the mainstream tower-based SIF system.The tower-based SIF system can be employed with bi-hemispherical and hemispherical-conical observation configurations for field installation. The bi-hemispherical observation mode refers to the configuration in which both downwelling and upwelling bare fibers are equipped with cosine correctors, while the hemispherical-conical observation mode refers to the configuration in which only the upwelling bare fiber is equipped with a cosine corrector. The bi-hemispherical observation mode has a larger field of view, which is suitable for canopy measurements with high canopy heterogeneity or height with a limited installation height. The hemispherical-conical observation mode is suitable for low canopy, homogeneous canopy, and multi angle observation. In addition, if the canopy area is limited or the experimental observations have control factors, hemispherical-conical observation is more appropriate.The SIFprism system is a novel optical-prism-based SIF automatic observation system. This article introduces the software and hardware components and the flow of spectral data collection of the SIFprism system. Taking the SIFprism system as an example, the spectral data processing process is expounded, and the potential uncertainty of SIF retrieval is analyzed.The tower-based SIF observation system has experienced rapid development in recent years. Despite the essential and incremental research on near-surface SIF, further development of hardware and mechanistic theory is still urgently required. Several prospective areas for future work include improving the signal-to-noise ratio and radiation stability of the spectrometer and appraising the capabilities and efficacy of different retrieval algorithms in varied light conditions. Finally, research should strengthen the cooperation with industry to jointly develop a more efficient and stable field tower-based SIF system and formulate corresponding field observation technical specifications.关键词：Solar-Induced chlorophyll Fluorescence;tower-based SIF measurements;SIFprism system;measurement protocol;field data collection- 947
- |
- 294
- |
- 2
发布时间：2021-06-07 -
摘要：Deep learning algorithms have been developed and applied in detecting clouds for satellite imagery in recent years. However, deep neural network models consist of thousands or millions of parameters, thus usually requiring large amounts of training data. Therefore, understanding the generalization ability of deep learning techniques is vital in their application to cloud detection of different types of satellite imagery. Taking DeepLabv3+, a typical deep semantic segmentation algorithm, as an example, this study explored the generalization ability of the algorithm on the cloud detection of satellite imagery with different landscapes, spatial resolutions, and spectral band combinations based on the cloud labeled dataset “L8 Biome.” The “L8 Biome” dataset consists of 96 typical Landsat 8 OLI images and the corresponding manual cloud mask, which has been widely used for evaluating the performance of cloud and cloud shadow detection methods. First, the cloud labeled dataset “L8 Biome” was used to generate different training and test samples with different landscapes, spatial resolutions, and band combinations. Then, the performance of DeepLabv3+ was evaluated based on different training and test sets and compared with that of the typical Function of Mask (Fmask) algorithm. Results show the following: (1) the DeepLabv3+ trained by a fully mixed training set (consisting of imagery captured over all landscapes) has higher overall cloud accuracy (92.81%) and stability (standard deviation 12.08%) than that trained by the training set of imagery captured over a single landscape and performed better than the Fmask algorithm with an overall cloud accuracy of 88.75% and stability of 17.34%, indicating that the training set of the deep learning algorithm should include images captured over various landscape types; (2) the DeepLabv3+ trained by “mixed-1” training sets, which were built by removing the images captured over a single landscape type (except of snow/ice) from the fully mixed training set, achieved comparable accuracies with that trained by a fully mixed training set, indicating that the available training set is diverse enough and has a good generalization ability on imagery captured over different landscapes; (3) the DeepLabv3+ trained by a fully mixed training set with a 30-m resolution achieved similar cloud detection accuracies on the satellite images with different spatial resolutions (i.e., 30, 60, 120, and 240 m), indicating that the trained DeepLabv3+ could be directly applied on satellite imagery with different spatial resolution, whereas the Fmask algorithm performed poorly on the images with coarse resolutions; (4) DeepLabv3+ could explore the effective information from different spectral bands for cloud detection, and more spectral band input can generally improve the overall cloud accuracy and stability of DeepLabv3+. Among all input bands, shortwave infrared bands are greatly helpful for distinguishing snow/ice from clouds, whereas the thermal infrared band marginally improves the cloud detection accuracy for DeepLabv3+. Results indicate that the DeepLabv3+ cloud detection network trained by the “L8 Biome” dataset can be applied to various types of satellite imagery and outperforms the Fmask algorithm.关键词：deep learning;cloud detection;DeepLabv3+;generalization ability;landscape;spectral band combination;spatial resolution- 743
- |
- 319
- |
- 5
发布时间：2021-06-07 -
摘要：Urban development assessment is helpful for urban planning and urban development policies. Census, survey, and nighttime light remote sensing data have been widely used to measure the urban development in previous research. However, most studies only focused on the size of urban development in a specific period and few of them have simultaneously considered the size and speed of urban development. A model that considers both the size and speed of urban development is necessary for evaluating the level of urban development, which is a dynamic process.On the basis of the nighttime light data of Suomi NPP-VIIRS from 2012 to 2019, the nighttime light kinetic energy index is proposed to measure the kinetic energy of urban development by considering the size and speed of urban development. Then, the Dynamic Time Warping (DTW) algorithm was utilized to measure the DTW distance using the nighttime light kinetic energy index. Finally, 328 cities in China were classified according to the DTW distance.Numerically, the nighttime light kinetic energy index in most cities increased significantly from 2013 to 2019, especially in the southeast coastal areas and central regions, and that in the northwest area has also increased greatly. In terms of spatial distribution, the original urban agglomerations composed of cities with high night-time lighting kinetic energy index values expanded from 2013 to 2019. The 328 cities were divided into five levels. The classified levels are more comprehensive and reasonable than the city rank released by First Finance in 2019. The Yangtze River Delta urban agglomeration was taken an example to analyze the synergy of urban agglomeration development, revealing that Tongling, Chizhou, and other cities in Anhui province still has a weak connection with the Yangtze River Delta urban agglomeration. The development of a hinterland city (such as Wuhu, Ma'anshan) is recommended to link these cities.关键词：remote sensing;urban development level;NPP-VIIRS;nighttime light kinetic energy;DTW distance- 973
- |
- 252
- |
- 6
发布时间：2021-06-07
Technologies and Methodologies
- Postal code：100190
- Tel：010-58887052 Email：nrsb@aircas.ac.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备20021838号-9
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
0