最新刊期
卷 25 ， 期 7 ， 2021
-
摘要：Through the earth observation, the geographical phenomena, patterns, and evolutionary processes on the earth’s surface can be fully reflected by the earth observation of remotely sensed imageries. The Land Use/Cover Change (LUCC) products from the High-resolution Remote Sensing (HSRS) data can provide full coverage, quantitative, and fast-updating background information for analyzing the law of spatial distribution of geographic features and their changing mechanisms. The development of HSRS is deeply driving the accuracy of LUCC information. At present, from the implementation effect of major survey projects such as national geographic conditions survey and natural resources survey deployed at the national level, the industrial production of LUCC information can be realized through the combination of human-computer interaction, in-house interpretation, and field survey. The development of high-resolution remote sensing is deeply driving the accuracy of LUCC information. How to improve the intelligent productivity of LUCC products through the innovation of theory and technology is an important bottleneck and key challenge.In the previous research, we put forward the basic concept of geo-parcel/geo-object based on the experience and mode of LUCC production in the departments of land investigation, and determined that the high-precision-level LUCC production is a gradually deepening cognitive process from external visual understanding to internal mechanism analysis. Therefore, this paper takes the geo-parcel as the basic unit of the cognition of land information of earth's surface, and further clarifies the meaning and geographical characteristics of LUCC.Based on the above background, we proposed a new concept of Precise LUCC (P-LUCC) in this paper, which integrates (accurate) quantitative index inversion model on the spatial structure of (fine) geo-parcels. First, this concept is a derivative of our developed spatial-spectrum cognitive theory, which is achieved by coupling HSRS visual features (TU-spatial maps) with multi-source and multi-modal observation mechanism features (PU-spectrum). Moreover, based on the geographical idea of “the unity of five land features”, we further proposed a series of intelligent remote sensing information extraction methods for P-LUCC production. They are organized hierarchically in three kinds of models, i.e. stratified perception model, spatiotemporal synergistically inversion model, and multi-granular decision-making model. In this aspect, we analyzed the cooperative computing mechanism of three types of machine learning models, namely deep learning for visual perception, transfer learning of external knowledge integration, and reinforcement learning via incremental self-organizing. Multi-type learning algorithms based on these mechanism are organized and transformed organically by using the route of “zoning partition-stratified extraction-graded transfer-functional reconstruction”. Thus, we designed a P-LUCC product production line for an information system of HSRS intelligent interpretation.Experiments were performed in the Suzhou High-tech Zone, China, and the accuracy and production efficiency of P-LUCC products were analyzed comprehensively. Through this large regional experimental verification, we show that our proposed technology has obvious advantages in the production accuracy and efficiency of LUCC products. Finally, we also provide some new ideas on thematic application based on P-LUCC information products. In conclusion, this intelligent production mode is worthy of popularization and application in engineering natural resources survey.关键词：geo-parcel/TUPU;precision LUCC;the unity of five land features;machine learning;intelligent computing;spatial optimization- 1813
- |
- 637
- |
- 11
发布时间：2021-07-23
Review
-
摘要：Isidis Planitia is a potential landing area for China’s first Mars exploration mission “Tianwen-1.” Fingerprint terrain and rampart crater are widely developed on the surface of Isidis Planitia. The rampart crater has one or more fluidized ejecta, which is generally considered the product of the interaction between the subsurface ice-rich layer and the hypervelocity impactor. Considering that water is an essential nutrient that nurtures and maintains all known life forms, the water evolution history of Mars has always been a research hotspot in the planetary community. Therefore, a detailed study on the rampart craters in the Isidis Planitia can provide strong constraints for analyzing the current and past subsurface ice-rich layers in this region.Through the use of high-resolution optical images obtained by the Context Camera (CTX) onboard the Mars Reconnaissance Orbiter (MRO), a comprehensive study has been performed on the rampart craters in the Isidis Planitia using image interpretation, morphologic analysis, and crater count dating. The morphometric parameters of the rampart craters, including ejecta mobility and lobateness, are calculated for all the identified rampart craters in this region. Moreover, the absolute model ages (AMAs) of representative rampart craters that have intact fluidized ejecta are determined by the Crater Size-Frequency Distribution (CSFD) measurement.This study found that 120 rampart craters are currently located in the Isidis Planitia. Their minimum diameter is 1.5 km, and most of their layered ejecta are highly irregular (sinuous) and extend to approximately 1.3 crater radii from the rim. The AMAs of 15 rampart craters reveal that they all formed in the Amazonian. According to the spatial superposition relationship between the rampart crater and the fingerprint terrain, this study infers that that the cones of the fingerprint terrain were formed in the Early Amazonian between 2.38-3.24 Ga, and they are more likely to be rootless cones/pseudo craters formed by explosive steam that break through the lava surface when a voluminous magma flows through wet or frozen ground and vaporizes the underlying (melt) water. According to the empirical formula between crater diameter and excavation depth, this study reveals that the depth of the subsurface ice-rich layer that is conducive to the formation of rampart crater in the Isidis Planitia is currently stable at least approximately 1 km and may rise or fall slightly by 0.1 km because of the effect of periodic changes of Mars' tilt (obliquity) on the climate.The results of this study are of great scientific significance for reconstructing the evolution history of the subsurface ice environment in the Isidis Planitia and are expected to be verified by the detection of the subsurface exploration radar onboard the rover and orbiter of the “Tianwen-1” probe.关键词：Mars;Isidis Planitia;Tianwen-1;rampart craters;absolute model ages;subsurface ice-rich layer;thumbprint terrain- 848
- |
- 312
- |
- 0
发布时间：2021-07-23 -
摘要：China’s first autonomous Mars exploration mission probe “Tianwen-1” was successfully launched in July 2020, and the rover of “Tianwen-1” successfully landed on the pre-selected landing area in the southern Utopia Plain of Mars on May 15. After landing on Mars, China’s first Mars rover “Zhurong” will carry out inspections and bring the raw scientific data for Chinese Mars exploration. The multispectral camera mounted on the “Zhurong” rover can be used to study the types of mineral and rock near the landing area. Identifying the types of mineral and rock on Mars is of great significance for understanding the Martian atmospheric change, environmental conditions, geological evolution, and future livability. The study in this article selected 18 kinds of common mineral and rock on the surface of Mars, and shot these mineral and rock with the same multispectral camera as on “Zhurong” rover in the earth environment. Then, this study uses the obtained multispectral image carry out the research of mineral and rock identification, hoping to provide guidance for future in-situ identification of mineral and rock on the surface of Mars based on the “Zhurong” multispectral camera data. Considering that 8-band spectral data is not enough to capture the spectral characteristics of all mineral and rock, it will result in fewer mineral and rock identified, this paper uses color features to assist in the identification of mineral and rock. This paper uses band operation based on multispectral image and HSV color feature histogram extraction based on color image to identify different types of mineral and rock. The above method can identify different types of mineral and rock from the perspective of multispectral features and color features. This study collected multispectral images of mineral and rock under three different shooting conditions, and found that the spectral characteristics of mineral and rock may change under different shooting conditions, which will have an impact on mineral and rock identification. The multispectral image data used in this paper was collected under the conditions of the solar elevation angle at about 60 °, the shooting height at 1.8 meters and the shooting angle at about 37°. Under this condition, the research method in this paper has extracted identification indexes of 12 kinds of mineral and rock. The identification indexes corresponding to specific mineral and rock are listed in the article, all results are based on the above shooting condition. The study shows that: using the band operation based on multispectral image, mineral and rock with prominent spectral characteristics under 8 specific bands can be identified; and using color feature histogram extracted from HSV color image, mineral and rock with a relatively concentrated color feature can be identified. In addition, since different shooting conditions will have an impact on the identification of mineral and rock, it is necessary to do the identification research under different conditions in the future.关键词：Tianwen-1;Zhurong;multispectral image;band operation;color feature;mineral and rock identification- 1092
- |
- 375
- |
- 5
发布时间：2021-07-23
Mars Exploration
-
摘要：This paper is targeted at the key and different technology appeared in system design and manufacture of high resolution stereo mapping satellite. On the basis of analyzing the development history and technical characteristic of mapping satellite both at home and abroad，and according to achievement of the high image positioning accuracy which is the key design of the mapping satellite，the paper shows the result of research about the design of high resolution stereo mapping satellite，such as design constraints，choose of the mapping system，and key and difficult points of design for both payload and platform products.The paper analyses the different special technique and engineering constraints of three line array camera system, two line array camera system, and single line array camera system. Since the satellite with single line array camera system can’t fulfilling large area’ mapping requirement, and it also have difficult on getting steady performance, the two or three line array camera systems are chosen by different contries’ special satellites with main task of mapping the earth. Considering the carrying capacity of satellite platform, to get high precision and high resolution stereo image, the two line array camera system is thought to be the best choice at this time. The paper gives out the design elements and achievement methods for high image positioning accuracy, such as orbit measurement, attitude measure and control, time measure and synchronization, camera inner elements, and it also gives out the detail about the influence of every element.Then the paper gives the design of the GF-7 satellite which is designed according to two line array camera system, and is the first stereo mapping satellite with sub-meter level resolution of China. After gives out the performance requirements to the satellite, and its major components also was shown. The paper shows the technical innovation of satellite which is to fulfil the requirement of satellite mapping task, such as new mapping system with both active and passive imaging, high resolution mapping camera, new laser altimeter, high precision attitude measure and control system, and high speed date transmission ability.At last，the image quality test results about GF-7 satellite images are given out by difficult consumers, including: Ministry of Natural Resources, Ministry of Housing and Urban-Rural Development, and National Bureau of Statistics. The test results show that due to the research results which given out by this paper, GF-7 satellite works very well in orbit, the actual image positioning accuracy is better than initial requirement. The satellite image can be used to produce 1∶10000 scale map, make investigation to city’s three dimensional transportation, measure the highway’s altitude, and check the variety of agricultural land. The satellite performance is better than the task’s requirement, and reaches the world leading level. The research results in this paper can be a strong reference to the design of larger scale stereo mapping satellite in the future.关键词：stereo mapping satellite;high resolution;key technology;image positioning accuracy;two line array camera;three line array camera;GF-7 satellite;performance in orbit- 991
- |
- 576
- |
- 10
发布时间：2021-07-23 -
摘要：Noise estimation of hyperspectral images (HSIs) is not only a crucial part of image quality evaluation but is also an important index of sensor performance. Spatial and spectral decorrelation method is a widely used approach for estimating noise in HSIs. This method is based on the high correlation of HSIs in space and spectrum, and a pixel can be predicted well using its spatial and spectral neighbors. Any prediction error can be considered noise. A series of noise estimation algorithms, such as Spatial and Spectral Decorrelation (SSDC), Residual-scaled Local Standard Deviation (RLSD), and Homogeneous Region Division and Spectral Decorrelation (HRDSDC), have been developed on this basis.The images are divided by rule or by some distances between spectrums in the general noise estimated methods. The local standard deviations or the residuals of multiple linear regression of imaging blocks are calculated as the image noise estimation. However, the sub-blocks of the images acquired by these methods are not completely uniform, and the edges of objects are still retained, thereby resulting in inaccurate outcomes of the image noise estimation. To obtain the uniform imaging blocks in the image effectively, an optimized SSDC method for estimating noise in HSIs has been used. The spectral angle and Euclidean distance are used to obtain the uniform imaging blocks, and the residuals of the heterogeneous blocks are calculated by multiple linear regression as the estimation of image noise. The optimized method is validated with simulated and radiance images acquired in the same aerial experiment and is compared with several useful noise estimation methods (e.g., LMLSD, RLSD, SSDC, and HRDSDC). The LMLSD method, which is based on spatial dimension, is susceptible to image texture features and is only suitable for images with relatively uniform landcover. The RLSD method has better noise estimation results than LMLSD.However, the uncertainty of the results is large and cannot indicate the noise level of images accurately. The three methods, namely, SSDC, HRDSDC and OSSDC, are all based on the spatial and spectral dimensions, have high stability, and can be applied to various images. The results of HRDSDC are significantly better than those of SSDC, and the OSSDC method exhibits better performance than HRDSDC. The OSSDC method uses the spectral angle and the Euclidean distance to determine the heterogeneous blocks, which reduce the influence of the edge of objects and the texture features. The results of image noise estimation are also accurate. In the validation, the optimized method shows distinctly enhanced robustness compared with the common methods. The estimation of the noise is also proved to be accurate. In addition, the effect of texture features on noise estimation is discussed in this paper. Results show that larger noise estimation results yield complex texture features.关键词：hyperspectral image;noise estimation;spatial and spectral;de-correlation;image quality assessment;Sensor performance evaluation- 883
- |
- 299
- |
- 1
发布时间：2021-07-23 -
摘要：Obtaining building distribution maps accurately and quickly has very important research value. For instance, it can help governments and non-governmental organizations plan urban infrastructure construction and assist disaster relief work. The Conditional Random Field (CRF) is widely used in building extraction tasks because of its flexible context information modeling and detail extraction capabilities. However, problems, such as blurry building boundaries, still exist when the CRF is used to extract buildings from high spatial resolution remote sensing images.This study proposed a Global-Local-Aware conditional random fields framework for building extraction. Global-Local-Aware D-LinkNet (GLD-LinkNet) is proposed to solve the boundary blur problem of unary potential in this framework. GLD-LinkNet makes up for the loss of local structural information by D-LinkNet while using multi-scale building information effectively. In addition, the segmentation priors are fused, and the label cost is introduced. The pairwise potential reflects the linear combination of the spatial relationship of neighboring pixels and the cost of the local class label. It can maintain the detailed information inside the buildings effectively. In addition, to solve the problem of spectral similarity between buildings and noise, a larger range of context information is used to fuse segmentation prior to the extraction of buildings. The framework can eliminate the influence of image noise and spectral diversity within class. Moreover, it can also keep the detailed information of ground objects and determine the clear boundaries of buildings.Experiments were carried out on the aerial building dataset and the satellite building dataset of the WHU building datasets. Experimental results demonstrate that the proposed dilated and segmented conditional random field framework is superior to the state-of-the art methods in terms of accuracy and IOU. The model proposed in this study, which performs well for building extraction of complex scenes, can maintain the detailed information of buildings effectively. In addition, it removes the small building blocks mistakenly extracted by D-LinkNet, and the problem of blurred building boundaries has also been improved effectively.The use of the global and local integrated D-LinkNet to model the unary potential of CRF can realize the effective combination of building features of different scales. It makes the structure of the obtained buildings more complete. Furthermore, by adding a segmentation prior to the construction of pairwise potential, a building classification map with a clean background can be obtained. The introduction of the local class label cost term also meets the high requirements of the building extraction task for the extraction of building detail information and can capture detailed information that is difficult to identify on the network. The proposed model was tested on aerial and satellite datasets, and the IoU indicators on the two datasets reach 91.72% and 89.82%, respectively. These values imply that the framework can adapt to aerial and satellite datasets.In the future, we will further study the application of large-scale high-resolution remote sensing images in building extraction and try to combine multi-source geographic information data to extract more complete building information.关键词：high spatial resolution remote sensing image;building extraction;conditional random fields;GLD-Linknet;the class label cost- 1149
- |
- 273
- |
- 5
发布时间：2021-07-23 -
摘要：Extraction of glacier by remote sensing has important value of scientific research and application. The threshold segmentation method has high efficiency and accuracy in remote sensing feature extraction. However, in terms of threshold selection, traditional manual threshold selection demonstrate low automation, repeated trial operation, and is easily affected by subjective factors. For example, the normalized snow cover index can separate the glacier information from the background information effectively. However, the threshold is almost always found manually, and hence is easily affected by subjective factors. Therefore, developing an effective method is necessary to calculate the glacier extraction threshold automatically.Gaussian Mixture Model (GMM) is established for the Normalized Difference Snow Index (NDSI) of the local glacier region by the expected maximum algorithm and to remove the mixed pixel meta classes in the region. Then, the GMM was used to simulate the NDSI distribution of purified glacier and non-glacier. According to the distribution of the improved Gaussian mixture model, the glacier extraction threshold in the region was calculated automatically.Three regions at different altitudes were experimented. Subsequently, the boundary extraction of Karlik mountain glacier in Hami, Xinjiang was compared with the glacier inventory data, and good extraction results were obtained. The experimental results show that the threshold values calculated by different grids in the same experimental area are slightly different, whereas the threshold values calculated by different grids in various experimental areas are largely different, thereby also indicating that the algorithm is still stable in the regions with large differences. As such, the results of glacier extraction threshold calculated by using this method are reliable, accurate, and stable in different regions.Result also shows that this method solves the problem of the unbalanced proportion of glacier and non-glacier in GMM curve simulation to some extent, and the automatic calculation results of glacier extraction threshold are reliable, accurate, and stable in regions with large differences. This GMM method is available for certain applications of delineating glacier boundary in remote sensing.关键词：remote sensing image;Gaussian Mixture Model;NDSI;extracting glacier information- 862
- |
- 310
- |
- 2
发布时间：2021-07-23
Technologies and Methodologies
-
Performance of cloud fraction of three satellite cloud climate date records over the Tibetan Plateau
摘要：Tibetan Plateau (TP) plays an important role in adjusting the large-scale atmospheric circulation in the northern hemisphere and the atmosphere–sea interaction from the equator to the middle latitude in the North Pacific. Obtaining complete observation data based on ground observations over TP is difficult. Satellite provides good observational data over the Tibetan Plateau. Considering the complex underlying surface types and geographical elevations in the Tibetan Plateau region, three kinds of long-term cloud fraction data that came from PATMOS-x/AVHRR, CLARA-A2/AVHRR, and MODIS / Aqua were analyzed from the perspective of data retrieval methods and data spatial attributes.The relationship among the three kinds of satellite cloud fraction and the ground observation cloud fraction was analyzed at first. Correlation analysis, linear trend, and accumulate bias were used to analyze the data. The analysis data were selected from instantaneous orbital observations and monthly and annual mean value.The annual mean cloud fraction of the three kinds of data are similar, but seasonal cloud fraction is different. CLARA-A2 has the smallest cloud fraction in summer and the highest cloud fraction in winter. Patmos-x agreed well with the ground observation. The correlation relationship between CLARA-A2 and ground was weak. Aqua/MODIS had good relationship in autumn and less correlation in spring and summer.The three kinds of long-term cloud fraction data showed similar spatial and temporal distribution. During daytime, CLARA-A2 has larger cloud fraction than MODIS and PATMOS-x. At nighttime, MODIS has the maximum cloud fraction value, and PATMOS-x and CLARA-A2 have similar values. All three kinds of cloud data committed a mistake with snow along the ridge of a mountain. The linear regression and accumulate bias analysis showed that the annual mean cloud fraction of PATMOS-x and CLARA-A2 displayed a decreasing trend from 1982 to 2015. The trend of the night time cloud fraction was more obvious than that of daytime. CLARA-A2 displayed more obvious trend than PATMOS-x, especially at night. The year of 2000 is a turning point for the change in cloud cover over the plateau area from high to low. In January, April, and October, the decrease in cloud amount is the main change trend. Meanwhile, in July, the weak increase is the main change characteristic.Three kinds of satellite cloud data have good comparability. Three kinds of data obtained different correlations when compared with the ground observation. The reasons may come from matched data with different spatial and temporal characteristics, different payloads with various observation abilities and different data set with different cloud detection algorithms.The stability of satellite orbit and high quality of instrument calibration are the baselines of long-term climate data. MODIS has stable instrument orbit and calibration. Thus, its long term cloud data have good homogeneity.关键词：cloud fraction;satellite;climate data records;Tibetan Plateau;Patmos-x;CLARA-A2;MODIS- 971
- |
- 310
- |
- 4
发布时间：2021-07-23 -
摘要：Tailings reservoir is a necessary facility for mining activity, and it also causes danger to surrounding environment. . Watershed risk of the tailings reservoir in Chicheng was monitored and analyzed using GF-1 high-resolution remotely sensed data with the help of multiscale fusion and deep learning method, as well as the support of Remote Sensing (RS) and Geographic Information System (GIS) technology for a comprehensive and detailed identification and extraction of the risk information of the tailings reservoir and to study the dam-break path of the tailings reservoir in watershed and the risk to land surface over mining area.In this research, a sample set library for target detection was constructed by analyzing texture, hue, shape, and size of the tailings reservoir on the remotely sensed data. Subsequently, an improved multi-scale fusion algorithm (e.g., Multi_Scale Feature Map_SSD (MSF_SSD)) was constructed by adding a deconvolution module and a connection module to the original single shot multiBox detector (SSD). Next, the Pyramid Scene Parsing network (PSPnet) algorithm was selected to achieve the structure of the tailings reservoir on the basis of the target detection results. The internal structure of the tailings reservoir was obtained. With the help of RS and GIS technology, the surface of upstream catchment and the possible danger runoff were extracted, and dam-break path of the tailings reservoir is simulated on the basis of the arc hydro model. Finally, the range area affected by the dam-break were obtained by constructing the buffer zone of the dam-break path.The research results shown that the dam-break path of the tailings reservoir in Chicheng is generally from west to east and from north to south, and the total area affected by the dam-break was 480 km2. The combination analysis with land use/ cover classification indicated that forest land was 176.52 km2, farm land was 175.52 km2, urban land was 43.74 km2, rural construction land was 2.47 km2, water body was 17.72 km2, grassland was 3.60 km2, and pasture was 1.22 km2.The sample library constructed using GF-1 remotely sensed data and Google Earth 16 level image can provide the basis for the automatic recognition of tailings reservoir with deep learning framework. With the help of improved MSF_SSD and PSPnet algorithm, the semantic segmentation accuracy of the test area for pixel accuracy, mean IoU, F1 score, and mean F1 score is 0.98, 0.97, 0.99, 0.98, respectively. A comprehensive analysis of tailings reservoir dam-break range and possible damage to land surface types are performed with the help of hydrological analysis method and random forest classification results. Outcomes of this research can be used to analyze the impact area caused by dam-break, promote the capabilities of risk management and emergency response of tailings reservoir, and provide fundamental theories for decisions making in relevant departments.关键词：tailings reservoir;SSD multi-scale fusion;PSPnet deep network;Arc Hydro model;risk analysis- 894
- |
- 273
- |
- 0
发布时间：2021-07-23 -
摘要：The Paracel Islands is the largest island with the largest number of islands in the South China Sea Islands. They have a unique natural environment and particular natural flora. The vegetation on the islands has always been a key concern of botanists and geographers. In order to quickly obtain the continuous distribution and long-term dynamic changes of vegetation in a large area of the island, the study integrated multi-source high resolution remote sensing data and measured GPS sampling data, spectral data and other auxiliary information. The vegetation types of typical islands in Paracel Islands were identified based on spectral classification, which fusing Support Vector Machine (SVM) and Spectral Information Divergence (SID) two classifiers on decision-making level and generating the typical vegetation distribution maps. A typical vegetation spectrum library of the Paracel Islands was also established using to analyze the characteristics of the measured spectrum and its first derivative of the typical vegetation in the Paracel Islands, enriching the basic information of vegetation in Paracel Islands. The study compared the classification method of SVM+SID and the general Spectral Angle Mapper (SAM) method，then further obtained the accuracy assessment results of each island. Based on the vegetation distribution maps of typical islands in different periods after the accuracy assessment, the statistical change of the area occupied by each vegetation and the correlation analysis of the vegetation diversity of different islands were conducted. The results demonstrating that: (1) The average production accuracy and user accuracy of the spectral classification method (combining SVM and SID classifiers on decision-making level) were 83.49% and 85.54% and Kappa coefficient was 0.8728 of the most whole islands in Paracel Islands. Therefore, the study achieved good performances in identifying different vegetation types on typical islands. (2) From 2002 to 2018, the vegetation types and its areas increased and tended to be stable. Scaevola was prone to form single-superior vegetation community in islands and Sandbank Grass on sands account for a large area; In recent years, vegetation on many islands such as Yagong Island, West Sand, Tree Island, etc., had been artificially interfered, and it had been regularly distributed along buildings, roads in spatial, and the diversity of vegetation species had greatly increased. (3) Through correlation analysis, this study found that the number of vegetation types on each typical island was basically positively correlated with the area of the island, that is, the larger the island area, the richer the plant habitat and the greater the number of species. And the larger the area of the island, the faster the vegetation types increase over time. In addition, the distance between adjacent islands was positively correlated with the similarity of vegetation on the two islands, that is, the closer the distance between the two islands, the more conducive to the mutual penetration of flora and the higher the similarity of vegetation types. But on a long-term scale, the dynamic changes of vegetation in Paracel Islands had been affected by human activities more than other natural factors.关键词：Paracel Islands;spectral classification;vegetation identification and change;Scaevola taccada;number of vegetation types- 921
- |
- 452
- |
- 2
发布时间：2021-07-23
Remote Sensing Applications
-
摘要：Hyperspectral images (HSIs) have abundant spectral characteristics. However, their spatial resolution is relatively low. Some remote sensing data have complementary advantages with HSIs, such as the LiDAR, which can provide elevation information, and the high spatial resolution data, which have precise spatial information. The combination of HSIs with the multi-source remote sensing data for fusion classification can make up the deficiency of tits relatively low spatial resolution. In recent years, deep learning-based methods have been investigated for hyperspectral remote sensing classification and have made breakthroughs. Meanwhile, the feature extraction process of the deep network is an independent process. Therefore, it may not obtain the most beneficial features for classification accurately and may influence the classification accuracy. At the same time, the techniques may not perform well when using limited training samples in HSIs because of massive parameters and complex network structure.Aiming at this problem, the frequently used traditional features of remote sensing data for classification are discussed in this paper. A new deep learning-based feature level data fusion classification framework that integrates traditional textural features into Convolutional Neural Network (CNN) approach (T-F-CNN) is proposed for the accurate fusion classification of HSIs and multi-source remote sensing data. The proposed method can be implemented in three steps. First, the traditional features are extracted from the HSIs or the multi-source remote sensing data. Second, CNN are built. The original HSIs, the original multi-source remote sensing data, and the traditional features, which are obtained in the first step, are inputted into the CNN of the deep feature extraction. Finally, the deep features obtained in the second step are concatenated in a concatenate layer of CNN, and SoftMax is used to generate classification maps at the end of the framework.Result The proposed classification scheme is tested on two data sets, namely, Houston and Thetford Mines Area data sets. The proposed T-F-CNN is compared with the pixel-level methods, such as Support Vector Machines (SVM) with the Radial Basis Function (RBF) (T-P-SVM), the CNN fusion method (P-CNN), and the CNN with traditional features (T-P-CNN); and the feature-level methods, such as CNN fusion method (F-CNN) and CNN method combined with original traditional features (T’-F-CNN). On both data sets, the proposed method shows a higher classification accuracy than other methods. Meanwhile, when the training samples reach a minimum number, the proposed method could provide the highest overall classification accuracy.The results obtained by the proposed method on the two real hyperspectral data sets demonstrate that the classification accuracy can be improved. Furthermore, the proposed T-F-CNN method outperforms some traditional deep learning methods and exhibits higher computing efficiency than a few advanced deep learning techniques.关键词：Convolutional Neural Network (CNN);hyperspectral image;Light Detection and Ranging (LiDAR);high spatial resolution remote sensing data;data fusion;Gray Level Co-occurrence Matrix (GLCM);classification- 1713
- |
- 648
- |
- 10
发布时间：2021-07-23 -
摘要：Radar imaging of objects obscured by random media is an important issue because of its wide application in the fields of geography, medicine, and the military. However, the echo signals from the observed target(s) may be severely distorted because of the presence of random media (e.g., vegetation, atmospheric turbulence, biological tissues, or walls), thereby eventually degrading imaging quality. To obtain higher resolution, imaging technology that works in the millimeter wave or even a higher frequency band is desirable. However, the electromagnetic wave in this frequency band is more heavily affected by random media and is more susceptible to attenuation, which hinders the application of millimeter wave radar remote sensing. This limitation further highlights the urgency of research on the imaging of objects obscured by random media. Therefore, evaluating and improving the imaging performance fully and ultimately are especially important.Synthetic Aperture Radar (SAR) technology has been widely used in many fields, especially for remote sensing, since its introduction in the 1950s. Scholars have proposed different imaging algorithms and used the obtained data to analyze the dielectric properties and geometric characteristics of the observed target (e.g., INSAR, POL-SAR, POL-INSAR, and TOMO-SAR). Alternatively, some imaging technology employs the time symmetry of the field (electromagnetic or acoustics) wave equation and the reciprocity of the Green’s function to locate and imaging the targets. In particular, the Time Reversal (TR) method allows us to selectively focus on different targets separately, whereas the Time Reversal-Multiple Signal Classification (TR-MUSIC) method improves the imaging resolution greatly. However, current studies are often limited to a specific field, and research on the comparison of the performance of different methods is relatively rare. Therefore, in this paper, these typical radar imaging methods are selected to evaluate their performance toward imaging the target obscured by random media. Given that the target is obscured by random media, describing the effects caused by random media on the propagation of the electromagnetic wave is necessary. According to the radiation transfer equation, the attenuation of electromagnetic waves caused by random medium is related to optical thickness, which is equal to the sum of the scattering and absorption thickness. The model will be used to describe the interaction of electromagnetic waves with random media. For quantitative evaluation, 3 dB beam-width and the geometric location of a point target response are used.Although the results in the three methods are all degraded by the presence of random media, TR-MUSIC performs the best followed by SAR and TR. The effects of scattering thickness is the main factor that causes imaging degradation, whereas the degradation caused by absorption thickness is very weak. In summary, this phenomenon is due to the clutter enhancement from random media when the scattering thickness increases, while the effects of absorption thickness correspond to the energy of electromagnetic waves being absorbed. Among the three techniques, TR and TR-MUSIC can suppress the grating lobes better than SAR does under a sparse array, and TR-MUSIC delivers the best imaging performance.Considering the advantages of TR-MUSIC in the performance and the side lobe suppression, we focus on improving its performance further. Based on theoretical analysis, some centrally located array elements are removed to undermine clutters, and better imaging results are obtained for TR-MUSIC.关键词：radar imaging;Random Medium;Comparative Analysis;synthetic aperture radar (SAR);Time Reversal Imaging- 658
- |
- 151
- |
- 0
发布时间：2021-07-23
Doctor's Voice
- 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