最新刊期
卷 26 ， 期 8 ， 2022
-
摘要：The subtropical region of China covers a vast area with special and unique geographical characteristics. Typical geographical characteristics include complex natural landscape with a mass of mountains and forests, cloudy and rainy climate, and rich biodiversity. And the subtropical region is the major producing areas of rice in China. Moreover, the subtropical region has abundant rivers, lakes, and mineral resources, which induce the sewage from the mining area spread widely along the river basins. All these geographical characteristics lead to high ecological environment sensitivity and frequent natural disasters of subtropical region in China. The characteristics of remote sensing for large range of rapid observation make it essential for precise monitoring of natural resources and environmental disasters in the wide subtropical region. Furthermore, it is urgent to develop special remote sensing theory and technology for subtropical region in China, so as to support the sustainable development and ecological civilization construction. In recent years, many researchers have gradually paid their attention to the subtropical satellite remote sensing. Internationally, the research topics mainly include land use cover change, urban environmental monitoring, wetland mapping, earthquake damage and its secondary disaster monitoring, water quality monitoring, vegetation biomass inversion, and aerosol parameter inversion and so on. In China, scholars mainly focus on some specific applications, such as flood disaster monitoring, mangrove monitoring, and forest degradation etc. These works provide abundant cases and raw materials for the formation and development of the theoretical system of subtropical remote sensing. However, most of the current studies just concentrated on some particular objects, local areas, or specific problems. So far, the theory and technology system of subtropical remote sensing is still in its infancy, and lacks in systematic analysis on characteristics of research status, fundamental problems and future development. In this paper, we first described the basic characteristics of subtropical region of China and the related practice researches in remote sensing. In this part, we analyzed the practice of remote sensing in subtropical region from the perspective of remote sensing data sources, including optical, hyperspectral, microwave, and multi-source data. Furthermore, we discussed the common problems of the practical researches in subtropical remote sensing. In this part, we analyzed and pointed out the two fundamental problems of subtropical remote sensing, i.e. the problems of geographical objects and remote sensing information. Accordingly, we elucidated the key tasks and scientific problems derived from the two fundamental problems. Then, we analyzed the historical opportunity of subtropical remote sensing development, and put forward the core scientific problems and development approaches of subtropical remote sensing that should be focused on in the future. In conclusion, there are some basic features and common problems in subtropical remote sensing. The innovation and development of subtropical remote sensing, including the collaborative observations of multiple sensors and various platforms, are inevitable trend. This paper aims to explore the development ideas for theories and methods of subtropical remote sensing. Meanwhile, we commit to clarify the innovation direction of subtropical remote sensing technology and application. Most important of all, we hope to promote the application of remote sensing in resource and environment monitoring, disaster prevention and reduction, and ecological civilization construction in the subtropical region of China.关键词：subtropical remote sensing;geographical objects;remote sensing information;resources and environment;natural disasters- 1735
- |
- 698
- |
- 1
发布时间：2022-09-29 -
摘要：Hyperspectral imaging can provide narrow bands and continuous spectrum information. However, hyperspectral image data have the characteristics of high dimensionality, rich features, information redundancy, small samples, and significant uncertainty, which result in difficulties in hyperspectral image data processing. Dimensionality reduction of hyperspectral remote sensing is one of the important topics in hyperspectral image data processing. Hyperspectral image data have hundreds of bands and can provide rich information, but a strong correlation exists between different bands, resulting in data redundancy. Therefore, the dimensionality problem is encountered during the processing of hyperspectral data, such as the increase in time complexity and the overfitting of the prediction model due to the increase in spectral feature dimension. More importantly, the number of training samples available for hyperspectral remote sensing images is small, and the feature dimension is much larger than the training sample. The classification accuracy will increase first and then decrease with the increase of feature dimensionality, that is, the “Hughes” phenomenon. Therefore, exploiting the rich information of hyperspectral images data and solving the problem of high feature dimension through certain methods have become key issues in the research on hyperspectral imaging data processing. The dimensionality reduction of hyperspectral remote sensing image is an approach to reduce the dimensionality of hyperspectral imaging through feature extraction or band selection while retaining as much effective information or features as possible. Feature extraction methods, such as principal component analysis, linear discriminant analysis, independent component analysis, manifold learning, and deep learning-based methods, use the projection transformation method to map hyperspectral data from high-dimensional space to low-dimensional space. Feature selection eliminates redundant bands without changing the original feature structure and finds representative feature band subsets, such as the selection based on information measurement and feature correlation. With the development of new technologies, evolutionary and intelligent algorithms, such as the genetic, ant colony, and firefly algorithms, have been applied in hyperspectral remote sensing dimensionality reduction.This article systematically summarizes and reviews the current advances in dimensionality reduction for hyperspectral remote sensing, especially for feature extraction and selection. For feature extraction, we review the advances of feature extraction algorithms based on index and parameters, projection and transformation, band combination, spatial algorithm, manifold learning, and deep learning. For band selection, the advances in information measurement, search strategy, optimized band number, multi-feature quality assessment, and optimization algorithms are reviewed. The challenges of dimensionality reduction for hyperspectral remote sensing are analyzed from five aspects: feature separability, feature quality evaluation, feature number determination, multi-feature optimization, and problem-oriented feature selection. Intelligent dimensionality reduction will be one of the most popular topics with the development of intelligent hyperspectral remote sensing. Meanwhile, multi-feature quality assessment, search strategy optimization and application requirements will attract special attention in the future. The dimensionality reduction of hyperspectral remote sensing will play an important and irreplaceable role in hyperspectral image data acquisition and applications.关键词：hyperspectral remote sensing;dimensionality reduction;feature extraction;feature selection;multiple features optimization- 4810
- |
- 1652
- |
- 8
发布时间：2022-09-29 -
摘要：With the rapid development of aerospace industry and remote sensing and the strong support from the government, various military and civilian commercial satellite systems have been developed. The establishment of a relatively complete satellite remote sensing data acquisition system injects new momentum for promoting high-quality development of economy and society. At the same time, the rapid development of artificial intelligence has greatly improved the intelligence and precision of data analysis and has brought new development opportunities for remote sensing big data analysis and application. In the context of the Internet era, combining advanced technologies, such as artificial intelligence, big data, Internet of Things, and 5G, is a general trend to promote the development of remote sensing applications in the direction of intelligence, popularization, and industrialization.Based on the current development status and actual needs of intelligent remote sensing technology for land observation satellites, first, this review briefly describes the development of earth observation satellite systems, such as GF and ZY satellites. Second, it classifies and introduces the development status and trends of artificial intelligence technology in the field of remote sensing. Furthermore, the application status of artificial intelligence-driven remote sensing technology in the fields of resource investigation, environmental monitoring, disaster monitoring, smart city, agriculture, forestry, and fishery automation analysis is discussed. Finally, by analyzing existing remote sensing technologies, the challenging problems and development trends of AI in remote sensing are concluded.Different from previous reviews, the present study has two major contributions. On one hand, it carefully reviewed the development status of existing AI-based remote sensing methods. Although AI has been successfully and widely applied in remote sensing, its performance is still unsatisfactory and far behind the intelligence of remote sensing experts in many domains. To address this problem, further development of a new-generation AI and wider application of AI in remote sensing is the key to success. On the other hand, this work provides five typical and key future research directions of future AI-based remote sensing technologies. First, the rapid knowledge mining technology of remote sensing big data is studied, and the comprehensive perception and intelligent analysis of remote sensing are realized with the support of AI technology. Second, the collaborative sensing technology of observation network constructed by multiple remote sensing satellites is studied to achieve more comprehensive, more accurate, and more efficient earth observations. Third, cross-modal multisource remote sensing data fusion and recognition technology are investigated. By fusing multisource remote sensing data of different types, such as visible light, multispectral, infrared, hyperspectral, and microwave, the performance of remote sensing image recognition and interpretation is expected to be dramatically improved. Fourth, the on-orbit intelligent processing technology of remote sensing data, including on-orbit processing hardware and software systems, are examined. Lastly, the human-machine hybrid enhanced intelligent remote sensing technology is studied. In the future, humans and intelligent remote sensing systems are expected to be closely coupled and work together to form a stronger remote intelligent sensing ability.关键词：satellite remote sensing;artificial intelligence;application services;remote sensing big data;data interpretation- 2824
- |
- 1296
- |
- 3
发布时间：2022-09-29
Review
-
摘要：Ecological protection and restoration of mountains, rivers, forests, farmlands, lakes, and grasslands is an important measure to ensure national ecological security. In this study, we developed scientific and rational monitoring and performance evaluation methods to ensure successful implementation of ecological protection and restoration projects and sustainable development of ecological restoration achievements.This study uses a combination of literature review and case analysis to discuss the specific application of remote sensing technology in the monitoring and performance evaluation of ecological restoration of mountains, rivers, forests, farmlands, lakes, and grasslands.Based on the pattern and quality of regional ecosystem, the land cover type, vegetation growth, and water quality parameters obtained from remote sensing data are selected as evaluation indices to monitor ecological elements of mountains, rivers, forests, farmlands, lakes, and grasslands in general. To maintain and improve the regional ecosystem services, we select water conservation, soil erosion, ecosystem bearing capacity, and other indicators to discuss the role of remote sensing data in assisting the performance evaluation of ecological protection and restoration projects of mountains, rivers, forests, farmlands, lakes, and grasslands.Results show that actively exploring new remote sensing data mining methods, which combine remote sensing technology and traditional investigation methods, is expected to form a more objective and faster executable monitoring and performance evaluation system for ecological protection and restoration projects of mountains, rivers, forests, farmlands, lakes, and grasslands.关键词：Ecological protection and restoration;“Mountains;Rivers;Forests- 1629
- |
- 474
- |
- 3
发布时间：2022-09-29 -
摘要：Permanganate index (CODMn) is an important water quality parameter to reflect the degree of organic pollution. At present, the retrieval of organic pollution by remote sensing technology is mostly based on empirical models and requires considerable manpower for data collection. Meanwhile, it has time and space limitations because it cannot process each image under different imaging conditions adaptively. The integrated water quality index, CDOM, and DOC of the inverted parameters are not water quality indexes. Thus, they cannot be directly used for actual water quality evaluation. Therefore, a novel quantitative remote sensing technology method for the retrieval of water permanganate index with clear understanding on mechanism is proposed.The method based on the radiation transmission process of electromagnetic waves and the characteristics of the water body in the study area consider the three major water quality factors of suspended sediment, chlorophyll, and oxygen-consuming organic, analyze the absorption and scattering coefficients of oxygen-consuming organic matter, and separate the contribution of the water column to the remote sensing signal from the effect of the bottom. The diffuse extinction coefficients (c) of water quality components are expressed as functions of in-water absorption (a) and scattering (b). Finally, the concentration of CODMn was derived with the remote-sensing reflectance below the surface (rrs).The experiment on the GF-1 Wide Field of View (WFV) imageries of the three major reservoirs in Shenzhen shows that the model method is reliable with overall accuracy of R2=0.832 and RMSE=46.4%. The spatial—temporal characteristics of the three major reservoirs in Shenzhen during 2018—2019 were investigated. The overall CODMn concentration of the three major reservoirs is low with average CODMn concentrations of less than 4 mg/L; it is affected by mild organic pollution. No pollution diffusion occurred at the junction of the reservoirs, and the peak concentration mostly appeared near the residential areas at the reservoir corner. The highest hotspot was observed in spring and autumn, whereas the lowest was in rainy summer From March 2018 to May 2019. The water quality improved, consistent with the background of Shenzhen’s special water treatment activities in 2018. The core of reservoir water quality protection is recommended to control external pollution and avoid the input of pollution sources during the flood season.A distinct advantage of the models is broadly applicable due to their physical basis, which satisfied the application requirements. The model solving method is based on the inherent optical properties of typical water bodies in Guangdong Province, and these properties have seasonal variability. The seasonal variations of inherent optical properties of water bodies can improve the stability of the model. In addition, the spectrum of shallow waters is affected by the depth and the reflection at the bottom. CODMn concentration inversion from satellite data with more spectrum bands remains underexplored. The RS scheme used in this study can not only provide support for inland water resource development and policy formulation in Shenzhen, but also a valuable reference for the evolution of inland water organic pollution in other regions.关键词：permanganate index;Organic pollution;absorption coefficient;GF-1 WFV;Shenzhen;water quality remote sensing- 1268
- |
- 563
- |
- 1
发布时间：2022-09-29 -
摘要：Forest fires seriously affect the environment and social economy, e.g., damaging infrastructure, causing economic losses, and endangering human health. Effective simulation and prediction of forest fire growth are greatly important. Fire behavior models can provide analytical schemes for characterizing and predicting the speeds and directions of fire spread. However, fire spread models are subject to assumptions and limitations that inherently produce compounding errors during simulations. Satellite remote sensing monitoring of forest fire can be used to analyze the spatial dynamic change process of large-scale fires. It is an economical and effective technology for obtaining fire information in a large range and a short period. It can also provide fire location information for fire spread models.This study proposes a new approach of fire spread simulations based on the assessment of simulated fire growth discrepancies by using satellite active fire data. The FARSITE fire spread simulator was used to simulate the spread of forest fires that occurred on May 17, 2017 in Chenbaerhuqi, Inner Mongolia autonomous region, China, and the S-NPP\VIIRS forest active fire data were applied into the FARSITE simulator for calibration and re-initialization. The Landsat-8 and GF-1 data were used to generate the data required by the FARSITE. The fire field for different time periods was monitored by the multisource satellite data Sentinel-2A, GF-1 and GF-4 data. 375 m VIIRS active fire monitoring data were employed for re-initializing FARSITE fire simulation. We combined the satellite fire data and fire spread model for reducing errors of simulation results caused by the condition limitation of fire model, and the Sørensen’s coefficient (SC) was employed to evaluate the accuracy of fire spread simulation results at FARSITE before and after reinitializing the simulator for VIIRS active fire data.The re-initialization results of the FARISTE simulator by VIIRS active fire data showed that the simulation accuracy in each simulation process gradually decreased along with time. The distribution of simulation results indicated that the simulation findings after re-initialization were consistent with the actual fire perimeter monitored by high or moderate resolution remote sensing data. The highest precision in the process using active fire data increased by 56.89%, and the final accuracy increased by 45.45%. The final SC value increased from 54.14% to 78.76% when the satellite data were used to re-initialize the FARSITE fire simulation system, with increment of 42.76%. The maximum SC value was 87.8% for VIIRS active fire data re-initialization during simulation. The re-initialization approach meaningfully improved the accuracy of fire simulation.The use of satellite remote sensing active fire data and the re-initialization of FARSITE limited the further expansion of the error of fire spread model and improved the reliability and accuracy of forest fire simulation. This innovative approach represents a potential scheme for reducing the error of large-scale fire simulation results that can improve the reliability of fire spread model. This method provides an effective data assimilation method for fire prediction. It also provides a basis for fire management departments to manage forests and develop fire suppressing plans. In this study, the actual ground and air fire-fighting forces change the results of fire spread. They are an important factor for the deviation between the simulation results and the actual results.关键词：remote sensing;forest fires;forest fires spread model;VIIRS;FARSITE;fire behavior simulations- 1669
- |
- 679
- |
- 1
发布时间：2022-09-29
Ecology and Environment
-
摘要：Since 2000, China has launched four sun-synchronous ocean optical remote sensing satellites, namely, HY-1A, HY-1B, HY-1C, and HY-1D. The detection of Sea Surface Temperature (SST) distribution and variation is one of the main tasks of Chinese Ocean Color and Temperature Scanner (COCTS), which is the main load of HY-1 satellite. The dynamic range of the actual water temperature detection channel is required to cover the temperature range of 200 K to 320 K, considering the detection of sea ice, typhoon, and other meteorological elements over the ocean. The variation of temperature in some ocean areas leads to severe weather disasters. Thus, SST detection channels should satisfy the requirements of detection sensitivity and quantification accuracy.This study aims to design an information acquisition circuit of infrared channel for COCTS according to the technical requirements, including a pre-amplifier circuit to amplify the weak signal of the detector, AC amplifier, which eliminates the basic level to improve the dynamic range, and the channel amplifier circuit, which can realize the DC recovery and dynamic range adjustment of the signal.Based on the study of the working mechanism of the photoconductive infrared detector used, and combined with the system composition and the characteristics of COCTS, the form and parameters of each stage amplifier circuit were determined thru theoretical analysis, calculation, and simulation to ensure that the contradictory requirements of high dynamic range and high sensitivity are met at the same time. The corresponding high-pass and low-pass filter are designed to achieve stable reference level detection and single pixel signal detection of the whole field of view. The system performance of COCTS is measured in the vacuum environment simulation laboratory to verify the reasonability of information acquisition circuit design.Results of the infrared radiometric calibration in the laboratory show that the dynamic range of the two infrared channels covers 177 to 327 K and 173 to 324 K; they satisfy the technical requirements of 200 to 320 K. The Noise Equivalent Temperature Difference (NETD) of the two infrared channels in the whole dynamic range is between 20 and 110 mK. At the appraisal position of 300 K, NETD has reached 21 to 34 mK, which is much better than the technical requirements of 0.2 mK. The space test environment is more complex than the laboratory, and the measuring accuracy has some differences. The results of in orbit test show that the dynamic range of the two infrared channels is 186 to 328 K and 185 to 326 K, and the NETD in the whole dynamic range is between 50 mK and 110 mK, according to the window size of the selected target area. The performance is better than the technical requirements.Conclusion The infrared channel can track the change in the blackbody signal on the satellite with modifications in time and the surrounding environment. Thus, the calibration coefficient of the infrared channel can be corrected in real-time. The expected goal of real-time radiometric calibration in orbit can be achieved. This lays a foundation for the quantitative inversion of SST and can obtain and develop high-quality global SST products.关键词：HY-1;information acquisition;sea surface temperature (SST);infrared photoconductive detector;DC recovery;noise equivalent temperature difference;modulation transfer function (MTF)- 2760
- |
- 257
- |
- 1
发布时间：2022-09-29 -
摘要：Satellite MODIS Aerosol Optical Depth (AOD) products based on Dark Target (DT) retrieval algorithm at 3 km resolution have been widely used in the ground air pollution monitoring. However, due to the limitations of DT retrieval algorithm, these products missed a large number of pixels with low spatiotemporal coverage and limited accuracy. By contrast, MODIS 10 km DT_DB_Combined AOD products integrate products based on DT and Deep Blue (DB) retrieval algorithms. However, to some extent, DT_DB_Combined AOD products can make up for the coverage and accuracy shortcomings of MODIS 3 km DT AOD data products, the resolution of which is low. Moreover, affected by the seasonal variation of aerosol component sources and the seasonal error of surface reflectance estimation, the accuracy of MODIS AOD data products also exhibits seasonal variation. This study takes Beijing‐Tianjin‐Hebei region as the experimental area and uses MODIS 10 km DT_DB_Combined AOD products as the material. Geostatistical Inverse Model (GIM) downscaling method, which considers the spatial covariance function differences of AOD data products at different scales, is introduced. At the same time, to consider the seasonal variation characteristics of AOD, the seasonal bias correction model for the MODIS 10 km DT_DB_Combined AOD products using accurate AERONET monitoring data is developed. On this basis, the Bias-corrected GIM (BGIM) downscaling algorithm coupled with seasonal bias correction model is further proposed. The AERONET ground observation data and MODIS 3 km DT AOD products are employed as the absolute and relative evaluation reference for the BGIM downscaling results. Results show that the absolute evaluated accuracies of the downscaled MODIS 3 km DT_DB_Combined AOD data, 10 km DT_DB_Combined AOD, and 3 km DT AOD data products are relatively close; the corresponding R2 values are 0.79, 0.70, and 0.71, respectively. Compared with MODIS 3 km DT AOD products, the relative evaluated R result of the seasonal corrected MODIS 3 km DT_DB_Combined AOD data is higher than 0.93. In addition, the temporal coverage and spatial coverage are increased by 11.21% and 11.44%, respectively. The spatial coverage in spring and winter is relatively higher among four seasons. The results confirm that the BGIM downscaling algorithm can not only effectively improve the resolution and accuracy of MODIS 10 km DT_DB_Combined AOD products, but also promote the spatiotemporal coverage compared with MODIS 3 km AOD products.关键词：remote sensing;AOD;downscaling;BGIM;spatial-temporal statistics;Beijing‐Tianjin‐Hebei- 2899
- |
- 168
- |
- 0
发布时间：2022-09-29
Atmosphere and Ocean
-
摘要：Spatiotemporal fusion is a feasible way to provide synthetic satellite images with high spatial and high temporal resolution simultaneously. In recent years, some efficient STF methods based on Convolutional Neural Networks (CNNs) have been developed. However, these methods require a significant number of training image pairs, where each pair generally consists of a high spatial resolution image and a low spatial resolution image. Such a requirement limits the applicability of STF methods to actual scenarios because image pairs for training are not widely available in many cases. To overcome this important limitation, we introduce a CNN-based single image pair method for STF of remotely sensed images. Our method, called SS-CNN, uses the spatial information provided by the average image (obtained across available spectral bands) of the high spatial resolution image to perform CNN-based Super-Resolution Mapping (SRM) between the low and high spatial resolution images. The proposed SS-CNN has been tested in experiments using two simulated and one real dataset and compared with two commonly used spatiotemporal fusion methods. The experimental results show that SS-CNN can predict the phenological changes and land cover changes well. Plus, its performance in heterogeneous areas is remarkable. The disadvantage is that it will slightly blur the boundary, which needs to be further improved in the future.关键词：remote sensing;spatio-temporal fusion;remote sensing images;single image pair;Convolutional Neural Networks (CNN)- 1291
- |
- 501
- |
- 2
发布时间：2022-09-29 -
摘要：Aircraft detection from optical imagery is a significant application in remote sensing. Traditional methods based on corner points or shape of the aircraft can only generate shallow features with limited representative ability. These methods are insufficient for detecting aircraft in remote sensing imagery under complex and diverse circumstances. Current methods based on CNNs, especially Faster R-CNN, have improved the detection performance greatly with its magnificent feature extraction ability. However, detecting aircraft on a single-scale feature map is unsuitable for multiscale aircraft in remote sensing imagery. After several pooling operations on a single-scale feature map, the feature map loses its precise details and small target that corresponds to a smaller area in the feature map. Thus, aircraft detection may result in low target positioning accuracy and target missing.An advanced Faster R-CNN is presented by constructing a multiscale feature extraction network using multistage fusion structure to detect aircraft with multiple scales. The promoted network produces features of higher resolution by upsampling deep feature maps. These features are then enhanced with shallow features at the same scale. After this modification, we end up with four feature maps F2, F3, F4, and F5, which have different scales. The structure combines the high-level semantic information with the low-level detailed information. Thus, the generated multiscale feature maps have high positioning accuracy and good distinguishability. In addition, because the original RPN anchors are extremely large to cover the range of aircraft sizes in remote sensing imagery, we select suitable RPN anchor parameters for aircraft detection, i.e., anchor size of 322 for the larger-scale feature map F2, 642 for the large-scale F3, 1282 is set for the F4, and 2562 for the small-scale F5. With these settings, the RPN can generate proposals, which can cover the aircraft of multiple scales. Finally, these proposals are assigned to their corresponding feature map, and we use the classification and regression network to obtain our final detection results.The experiment was carried out on RSOD dataset, in which only the aircraft dataset was used for training, validation, and testing. Comparison of detection performance with different anchor scales showed that anchor scales greatly affect detection accuracy, and our selection of anchor scales is suitable for the dataset. Three feature extraction networks (ZF, VGG-16, and ResNet-50) were modified based on Faster R-CNN using multistage fusion structure. The experiment showed that the modification can effectively improve the model’s ability of detecting multiscale aircraft. Compared with models without the modification, AP increased by 11.34%, 9.87%, and 1.66% for the three networks. The qualitative and quantitative results also showed that this modification can generate adaptive detection box. The experiment results on Beijing Capital International Airport GF-2 imagery showed that this method performs well in different remote sensing imagery, in which most airplanes in the airport were detected successfully.We can draw the following conclusions: (1) the proposed method is suitable for multiscale aircraft detection, and it can generate detection box consistent with the scale of multiscale aircraft targets while reducing missing targets; (2) correction of the RPN candidate region scale improves the accuracy of aircraft detection in remote sensing imagery; (3) the method has good generalization ability.关键词：remote sensing image;object detection;Faster R-CNN;multiple stages fusion structure;multi-scale- 1790
- |
- 614
- |
- 5
发布时间：2022-09-29 -
A dual-attention capsule network for building extraction from high-resolution remote sensing imagery
摘要：Automatic extraction of buildings from high-resolution remote sensing images is greatly important in disaster prevention and mitigation, disaster loss estimation, urban planning, and topographic map making. With the advancement of optical remote sensing techniques in image resolutions and qualities, remote sensing images have provided an important data source for assisting the rapid updating of building footprint database. Despite the large number of algorithms proposed with enhanced performance, fulfilling highly accurate and fully automated extraction of buildings from remote sensing images is still difficult due to the considerable challenging scenarios of buildings, such as color diversities, topology variations, occlusions, and shadow covers. Thus, exploiting advanced and high-performance techniques to further improve the accuracy and automation level of building extraction is greatly meaningful and urgently required by a large variety of applications.To overcome the issues of strong variability and weak homogeneity of traditional convolutional neural networks, we propose a novel dual-attention capsule encoder–decoder network DA-CapsNet for extracting buildings. In this network, a deep capsule encoder–decoder network, along with the channel-spatial attention blocks, is developed to enhance the capability of extracting high-level feature information from very high resolution remote sensed images. Thus, this model has the ability to extract buildings covered by shadows and discriminate buildings from non-building impervious surfaces. Specifically, we initially employ a deep capsule encoder–decoder network to extract and fuse multiscale building capsule features, resulting in a high-quality building feature representation. Moreover, spatial attention and channel attention modules are designed to further rectify and enhance the captured contextual information to obtain a competitive performance in processing buildings in the diverse challenging scenarios. The contributions include the following: (1) the deep capsule encoder–decoder network is designed to generate a high-quality feature representation; (2) the channel and spatial feature attention modules are designed to highlight channel-wise salient features and focus on class-specific spatial features.The proposed DA-CapsNet was evaluated on three datasets: one Google Building Dataset and two publicly-available datasets (Wuhan and Massachusetts). The experimental results achieved a competitive performance with an average precision, recall, and F1-score of 92.15%, 92.07%, and 92.18%, respectively, in handling buildings of varying challenging scenarios. Considering the overall accuracy of F1-score, the DA-CapsNet achieved the values of 92.70%, 94.01%, and 89.84% for Google, WUH, and MA datasets, respectively. Comparative studies also confirmed the robust applicability and superior performance of the DA-CapsNet in building extraction tasks.关键词：building extraction;deep learning;channel feature attention;spatial feature attention;encoder-decoder network;capsule network- 1362
- |
- 348
- |
- 2
发布时间：2022-09-29
Remote Sensing Intelligent Interpretation
-
摘要：Light Detection And Ranging (LiDAR) has been increasingly used in forestry research. Individual tree segmentation algorithm for airborne LiDAR point cloud data is greatly important for tree growth monitoring and forest management planning. In this study, a Nystrӧm-based spectral clustering algorithm was proposed to improve the accuracy and efficiency of individual tree segmentation for airborne LiDAR point cloud data.The proposed method is based on spectral clustering algorithm, and the mean shift voxelization and Nystrӧm method was introduced to maintain the good segmentation performance while improving the computational efficiency. First, the mean shift method was used to transform the point cloud dataset into a voxel space for efficient calculation. Second, a Gaussian similarity function with voxel weights was used to construct a similarity graph in the voxel space. Third, the approximation of the eigenvectors and eigenvalues of the similarity matrix was calculated using the Nystrӧm method. Fourth, the K-means clustering method was performed in the eigenspace, and the segmentation results were mapped back to the original point cloud to obtain the clusters of individual trees. Finally, individual tree parameters were obtained directly from the point cloud of each tree cluster.Assessed with field measurements, the overall matching rate of the proposed algorithm is 65% for the segmentation of airborne LiDAR point cloud data in the study area. For different stem density plots, the matching rates increased from 61% to 72% as the density decreased. The matching rates for the height layer of 20—25 m and >25 m reached 77% and 78%, respectively. Based on segmented trees, tree heights were extracted in high accuracy with the R2 value of 0.86 and the Root Mean Square Error (RMSE) of 1.62 m. Compared with other methods, the proposed algorithm produced satisfactory results in segmentation accuracy with the matching rate slightly worse than the spectral clustering algorithm but better than the K-means algorithm. In terms of computing time, the proposed algorithm achieved the highest computational efficiency, which was about 96 times that of the spectral clustering algorithm and three times that of the K-means algorithm.The proposed Nystrӧm-based spectral clustering algorithm achieved good performance in both segmentation accuracy and computational efficiency. The voxelization method based on mean shift reasonably compressed the volume of LiDAR point cloud and effectively reduced the computational burden of subsequent processes. The Nystrӧm method optimized the eigen-decomposition efficiency of the similarity matrix. Overall, the Nystrӧm-based spectral clustering algorithm can provide feasible individual tree segmentation for airborne LiDAR point cloud data, and key tree parameters can be obtained from the segmentation results.关键词：remote sensing;airborne LiDAR;point cloud;individual tree segmentation;spectral clustering;Nystrӧm method;voxelization- 903
- |
- 221
- |
- 0
发布时间：2022-09-29 -
摘要：Remote sensing image object detection based on depth neural network model has achieved great success largely due to the support of large-scale data sets. However, from the perspective of the existing remote sensing image datasets, the number distribution of different types of ground objects is inconsistent, and the same type of ground objects is presented in different sizes. This factor leads to the scale imbalance of ground object samples.To alleviate this problem, the strategy of weighted image fusion and multiscale feature selection is adopted. First, the pixel values of the two images in the data set are weighed to obtain the fused image. Therefore, the different types of feature samples are more balanced and have higher background diversity. Second, the target category of the corresponding scale is predicted by selecting the appropriate scale feature map, and the same scale target can be predicted on the adjacent feature map. Thus, the model can be trained according to the target scale. Finally, the bounding box of the target is predicted based on the feature map of the target center area, and the result is more consistent with the scale of the target itself.The image fusion and multiscale feature selection network is experimentally compared in the paper on two datasets, RSOD and NWPUVHR10. The results show that the trained model is more accurate in the recognition of imbalanced objects in complex background and can adapt to the recognition of different scale objects in remote sensing images. In addition, the proposed method enhances the diversity of sample scenes and can be adapted to different scales of targets. Qualitative analysis shows that the proposed method has better robustness and performance advantages when applied to remote sensing images.The proposed method can cope with remote sensing images with complex backgrounds, mitigate the effects of category imbalances and better fit the scenarios in which remote sensing images are used. Mitigation of category imbalances and Scale Selection enhance the performance of remote sensing image object detection.关键词：image fusion enhancement;multi scale selection and expression;high resolution remote sensing image;object detection;convolutional neural network- 1072
- |
- 433
- |
- 2
发布时间：2022-09-29 -
摘要：Laser altimeter rec waveform saturation commonly exists in Polar Regions, lakes and other areas, but studies about it are few. This study summarizes the previous research results and analyzes the causes of waveform saturation and the influence of waveform saturation on measurement accuracy. A method for waveform saturation recognition and height measurement error correction is proposed to increase the number of available laser points and improve the accuracy of saturation waveform measurement.The waveform saturation recognition algorithm uses a combination of saturation threshold and waveform kurtosis. The minimum saturation threshold is used to remove waveform data whose amplitude is low and unsaturated. Then, the kurtosis of the original waveform data is compared with that of the average distribution data. If the waveform kurtosis is less than the average distribution kurtosis, then it is considered t saturated; otherwise, it is a normal waveform. The saturation correction algorithm uses the centroid position difference to calculate the height measurement error. Gaussian fitting is performed on the original waveform data, the intersection point of the Gaussian fitting curve and the original data is calculated, the intersection point is selected according to the principle of selecting points, the two selected intersection points are connected to form a line segment, and the closed area surrounded by the line segment and the Gaussian fitted waveform is computed. The difference between the centroid position of the Gaussian fitting and that of the closed area is the time deviation of the height measurement caused by the waveform saturation. The time deviation is multiplied by one-half the speed of light to obtain the saturation correction value.Data collected by Ice, Cloud, and Land Elevation Satellite/Geo-science Laser Altimeter System (ICESat/GLAS) in Qinghai Lake, Nam Co, and Selin Co are used for experiments. Experimental results show that the average error of the data after algorithm correction is 0.03 m, and the large lake area can achieve an accuracy of about 0.05 m. The proposed saturation recognition algorithm is more accurate than that in the previous studies. GLAS uses a fixed threshold to identify missing saturated waveforms. The proposed saturation correction algorithm is easier to implement and has higher accuracy than the algorithm provided by GLAS. This proposed algorithm only relies on the shape characteristics of waveform data for calculation; thus, it has more universal applicability. Different from previous studies, the experiments also found that the waveform saturation does not only cause low conditions on elevation measurement, but also high conditions. The comparison between the corrected elevation and the actual elevation proves that this situation does exist.The number and accuracy of available laser points can be greatly increased by the effective correction of saturation waveform data. The proposed method can provide some references for saturation processing of echo data of domestic laser altimeter, but it is mainly suitable for effective identification and correction of single waveform saturation data, such as lake, flat land, and polar ice sheet. How to identify and correct multiwaveform overlapping saturation data in complex areas such as forest should be studied.关键词：satellite laser altimeter;Saturation waveform;error correction;ICESat/GLAS;GF-7 satellite- 990
- |
- 282
- |
- 0
发布时间：2022-09-29 -
摘要：When processing remote sensing images with complex features, the conventional Super-Resolution Reconstruction (SRR) methods are often not ideal, especially for remote sensing images containing various non-uniform object information. A universal method to solve this problem is difficult to construct at present. A new SR reconstruction method of mixed sparse representation model (MSR-SRR) combined with the sparse representation and non-convex high-order total variational regularizer has been proposed to solve this problem. In this method, the sparse representation of remote sensing images in multiple transform domains is regarded as a prior probability model, and the SR reconstruction is completed by regularization. The obtained image not only retains the edge information of the image result by SR reconstruction, but also smoothens the “ladder effect” of the image. The efficiency of operation and the quality of SR reconstruction results are improved by an effective re-weighted l1 alternating direction method. Results show that the sharpness of the image increases by 31.74% on the average, the half-peak width of PSFs is the largest, and the Gaussian variance value reaches 1.8415. The GF-4 satellite images have been selected to carry out validation experiment to verify the feasibility and validity of MSR-SRR. The reconstruction results show that the images using the MSR-SRR method have better definition, richer details, and higher quality than those with non-uniform interpolation, the POCS method, and IBP method. The support vector machine method is used to classify and evaluate the accuracy of the images before and after SR reconstruction. The results show that the overall accuracy and Kappa coefficient of the reconstructed super-resolution image are improved more significantly than the original image classification results. The OA value increases by 5.96%, and the Kappa coefficient increases by 9.7%. The findings confirmed that the MSR-SRR method is effective and feasible and has extensive practical value.关键词：remote sensing;GF-4;Super-resolution reconstruction (SRR);Mixed sparse representation (MSR);Total variation (TV);Non-convex- 1339
- |
- 308
- |
- 1
发布时间：2022-09-29
Remote Sensing Technology and Method
- 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