Wetlands located in coastal zone have important ecological and economic development value. It is of great significance to understand spatiotemporal characteristics and influencing factors of wetland change for maintaining regional ecosystem balance and sustainable development. Taking Landsat TM/ETM+/OLI images as the basic data source， wetland information extraction of low-elevation coastal zone of Southeast Fujian in 1985， 2000， and 2015 were carried out combing with object-oriented and deep learning classification methods. The spatiotemporal evolution characteristics and driving factors of wetland change were revealed. The results showed that： using object-oriented deep learning classification method， the overall classification accuracy of wetlands was more than 93%， and the classification results were desirable. During 1985~2015， the natural wetlands showed a decreasing trend， and the human-made wetlands showed an increasing trend， with -250.31 km2 and 251.36 km2， respectively. Among the second-class wetland types， the estuary/shallow sea water and mudflats decreased the most area in 30 years， and the salt pans/aquaculture ponds increased the most area. The types of wetland change were diverse from 1985 to 2015， and the wetland changes from 2000 to 2015 were more drastic than those from 1985 to 2000. The wetland dynamics attributed to natural environment change and the influence of human activities， in which human activities were the critical causes. This study can provide technical support and decision-making references for the monitoring， conservation， and management of coastal zone wetlands.
Lake Wetlands have larger ecological function such as climate regulation and biodiversity， and economic effects-flood storage and shipping. In recent decades， the spatiotemporal variation ofLake Wetlands in Northeast China is different from the global change feature. Based on the landsat-5/8 image data with 30m spatial resolution from 2006 to 2016， thedetection method based on the Dynamic Ratio algorithm is used for extracting the change ecological information， and determining the change area of Lake Wetlands； the classification scheme based on the multidimensional-indexes is constructed to extract the change types of Lake Wetlands. In addition， the change types of Lake Wetlands are divided into the transfer-off （this is， the decrease of wetlands）， the transfer-in （the increase of wetlands）， and the conversion of wetlands （the relatively unchanged wetlands）. The finalresults showed that： from 2006 to 2016， based on the dynamic change results of the DRM method， the correct detection ratio of wetland change in Songnen Plain， Xingkai Lake and Hulun Lake are 90.48， 90.2 and 93.81%， respectively. Meanwhile， the overall accuracy and Kappa coefficient of the land-cover classification results in the experimental area reached 84.31% and 0.788 respectively. The Lake Wetlands in Northeast China have the change trend characterized by the improvement feature， which can represent the actual fluctuation of wetland types in the study area. This method also has higher detection accuracy under complex surface types， which is a beneficial supplement to the resource’s investigation of Lake Wetlands and remote sensing monitoring on the wetland change.
Researches on Net Primary Productivity （NPP） of wetland vegetation are of great significance to the study of global change and carbon cycle. Taking three typical wetland samples in Northeast China as study area， based on Landsat 8 OLI and a large number of field data， this paper contrasted the combination forms of the basic structural formula of light utilization model. Results show that the model based on structure of NPP = f（f（VI1））×f（VI2） and two vegetation indices （NDVI and MSAVI） are optimal， with an accuracy of 89%， which was higher than those ofMODIS BIO-BGC and CASA model. In 2014， mean NPP of Phragmites australis for Qixinghe， Chaganhu and Shuangtaihekou wetland was 3 001， 3 050 and 3 621 gC·m–2·yr–1， respectively. NPP is obvious different spatially in three wetland samples， which is mainly influenced by hydrological conditions and human activities. For Phragmites australis wetland with small spatial scale， remote sensing method can be used to estimate NPP conveniently and efficiently. This study can provide a reference and guide for the study of wetland vegetation NPP regionally.
Understanding the spatial and temporal evolution of coastal saltmarsh wetland distribution is the baseofscientific management of coastal wetland ecosystems. Spartina alterniflora has rapidly invaded and spread in the coastal intertidal zone of China， which has significantly changed the structure and function of the native coastal wetlands， leading to great challenges to coastal wetland protection and management. At present， the large-scale remote sensing analysis of the spatial and temporal dynamics of coastal saltmarsh vegetation is very limited， and there is still insufficient understanding of the historical evolution of saltmarsh spatial distribution and its control mechanisms. Based on the Google Earth Engine platform and Landsat imagery， this study usedcontinuous change detection and classification algorithm to obtain the spatial and temporal distribution of saltmarsh vegetation in coastal wetlands in southern China （south of Zhejiang Province） during the past three decades， and then analyzed the impact of tidal flooding on the spatial and temporal distribution of saltmarsh vegetation. The results showed that： （1） The total distribution area of saltmarsh vegetation decreased from 2000 to 2004， and then showed a continuously growing trend； （2） There were three growthmodes of saltmarsh vegetation： fluctuating， linear， and exponential growth； （3） The distribution of saltmarsh vegetation and the frequency of flooding showed a hump-like spatial pattern， and the spatial and temporal distribution of saltmarsh vegetation evolved from less to more inundated area over the intertidal zone. This study helps to understand the spatial and temporal evolution of coastal wetland vegetation and provides decision support for the scientific management of coastal wetlands.
Wetlands are usually featured by evident seasonality， and thus high temporal-resolution remote sensing monitoring of their consecutive changes would greatly benefit to more objectively and accurately detecting the characteristics of spatial-temporal changes. The Poyang Lake wetland， as the largest freshwater lake in China， which shows significant intra-annual variability， was selected as the demonstrative case in this study. By collecting all available remote sensing images of Sentinel-1 & 2 and Landsat-8 from 2017 to 2019 based on the Google Earth Engine platform， we adopted the Random Forest （RF） method to map various types of wetlands of the Poyang Lake. It aims to demonstrate the capacity of Sentinel-2 optical images integrated with Sentinel-1 SAR and Landsat-8 data applicable to monitor wetland variations at both the inter-annual and intra-annual timescales. Results show that the Sentinel-2 images enable to provide a powerful data base for monitoring the dynamics of Poyang Lake wetland， and the overall classification accuracy was higher than 90%. the areas of the classification results were statistically analyzed in the 3 years， in February of each year， mudflat and vegetation reach the maximum area， while water area is the minimum.In June and July of each year， the water area reaches the largest in the year， while the mudflat and vegetation area is the smallest. All types of wetlands in the Poyang Lake show evidently seasonal changes， and the monthly classification results can more accurately illustrate the intra-annual changes characteristics of various types. Overall， the integration of Seninel-2 data with Sentinel 1 and Landsat-8 images， can effectively monitor the wetland changes at fine timescale， which is crucial for timely and costly management of wetland resources.
Baiyangdian is an important shallow lake wetland in the North China Plain， which has important ecological value for the green development of Xiong’an New Area. Wetland mapping of the highly heterogeneous landscape pattern of Baiyangdian can provide guidance for the remote sensing monitoring of Baiyangdian Lake wetland resources. In view of the seasonal changes of wetlands， a representative Sentinel-2 image is selected for each season of Baiyangdian in 2019. Three commonly used machine learning classifiers， including Classification and Regression Tree （CART）， Support Vector Machine （SVM） and Random Forest （RF）， were used to classify 15 classification scenario. The advantages and disadvantages of different seasonal remote sensing images and their combinations for extracting Baiyangdian wetland information were analyzed. The results showed that the combination of multi-seasonal images can significantly improve the classification accuracy. The combination of spring and summer images obtained the optimal classification accuracy. Compared with the single seasonal images， the overall accuracy was improved by 10.9%~25.5% and the kappa coefficient was improved by 0.09~0.29. The classification performance of the SVM classifier was relatively stable， and the highest classification accuracy can be obtained. The ability of CART classifier in processing high-dimensional features was not as good as that of random forest and SVM. The contribution of different features to the wetland information extraction was described as follows： red-edge spectral feature > traditional spectral feature > tasselled cap transformation feature > principal component analysis feature > texture feature. The research results can provide a basis for the remote sensing mapping of Baiyangdian wetland.
The Orbit Hyper Spectral （OHS） sensor， with high spectral and spatial resolution， is equipped on the Zhuhai-1 satellite constellation. It exhibits considerable advantages when monitoring the environment changes of coastal waters and inland lakes. However， OHS has no on-board calibration systems， the in-orbit vicarious calibration using field measurement was conducted but the result may not suitable for low reflectance target like waters. In this paper， we propose a total radiance-based cross-calibration method for OHS by using QAA （Quasi-Analytical Algorithm） marine optical model and 6SV2.1 radiative transfer model. This method makes the multiple-spectral sensor GOCI （Geostationary Ocean Color Imager） can be used for the radiometric cross-calibration of the hyperspectral sensor OHS. The result shows that the radiance observed by GOCI and OHS are highly correlated， with the R2 higher than 0.84 at the visible bands. It also indicates the new calibration method can reduce the radiance differences between GOCI and OHS. The calibration errors are less than 9% at the visible bands. This study provides a new method for radiometric calibration of hyperspectral sensors and has important significance for quantitative application of hyperspectral sensors， particularly for the quantitative remote sensing of waters using OHS data.
Millimeter wave radar is one of the important sensors for auto driving. Its main function is to measure the distance， speed and angle of targets around the vehicle. According to the general application scenario of vehicle-mounted （automotive） millimeter-wave radar， a two-transmitter， four-receiver TDM-MIMO FMCW millimeter-wave radar scheme with a fast chirp signal as the transmission waveform and a corresponding 3D-FFT target detection algorithm are designed， which can simultaneously obtain the target distance， speed and angle. By optimizing the arrangement of antenna elements， our solution can effectively solve the problem of unambiguous speed interval reduction of conventional TDM-MIMO radar due to channel time division multiplexing when measuring the target speed. Compared with conventional triangle waveform radar， the scheme can effectively avoid the problem of multiple target speed matching. Compared with single pulse angle measurement scheme， the scheme can greatly improve the angular resolution. Finally， we validate the proposed scheme by Matlab simulation.
Due to the poor penetrability of optical remote sensing， optical images are often disturbed by weather factors such as clouds， which affect the applications of remote sensing. The existing methods based on multi- temporal or single image are difficult to recover the real features under the cloud because of the change of features and the lack of prior information. Based on the fact that the SAR images are not interfered by the factors such as cloud and illumination， an cloud removal method by fusing SAR and optical images is proposed. Firstly， the cloud area is detected by the Fractal Net Evolution Approach （FNEA） combined with the shape and spectral characteristics. Secondly， the optical and SAR images are decomposed by the Non-Subsampled Shearlet Transform （NSST）. Finally， the decomposed coefficients are dealed with the cloud area detection results， in which the low-frequency information is fused by the improved weighted energy sum， and the high-frequency information is fused by the direction information entropy and Pulse Coupled Neural Network （PCNN）. Take the GF-1/GF-2 optical and GF-3 SAR images as the experimental data source. The results show that compared with the other five algorithms， our method has higher similarity with the reference image in the cloud area and can better maintain the texture and detail features， which effectively solve the problem of cloud occlusion while realizing image enhancement and is beneficial to the following remote sensing applications such as image classification， target recognition and image discrimination.
The traditional spectralunmixing algorithm ignores the different noise levels of the image in different bands， which leads to the limited accuracy of unmixing. To overcome this problem， based on the hyperspectral imagery， an Extended linear spectral unmixing algorithm based on noise level estimation （NELMM） is proposed. First， according to the multivariate regression theory in hyperspectral applications， the noise in adjacent bands is estimated. Second， the noise weight matrix is obtained from the estimated noise. Finally，the noise weighting matrix is integrated into the linear spectral unmixing framework， which can alleviate the impact of different noise levels at different bands. In order to verify the accuracy of the algorithm， the Fully Constrained Least Squares （FCLS） and Collaborative Sparse Unmixing by variable Splitting and Augmented Lagrangian（CLSUnSAL） are used for comparative analysis， and the vegetation coverage of the TM image is inverted by this algorithm to verify its practicality on multispectral images. The final test results show that the NELMM algorithm is better than the FCLS and CLSUnSAL for the unmixing of hyperspectral images. The noise weight matrix balances the noise between the bands， and the accuracy of the NELMM algorithm for unmixing images is significantly improved. At the same time， this algorithm shows good applicability to multi-spectral image unmixing.
Desertification is one of the most serious ecological and environmental problems in the world， especially in the China-Pakistan Economic Corridor （CPEC）. Based on MODIS data， this paper extracted key surface feature parameters and quantitatively studies the law and relationship between desertification degree and surface feature parameters. Three remote sensing monitoring models of Albedo-Vegetation feature space and decision tree were constructed， and the desertification degree of CPEC in 2015 was analyzed. The results showed that the overall accuracy of Albedo-MSAVI， Albedo-NDVI and C5.0 methods were 88.33%， 85.83% and 89.2%， respectively. According to the analysis， the decision tree method was the most suitable to invert the desertification degree of CPEC. Based on the C5.0， calculated the distribution data of desertification degree from 2000 to 2015， and analyzed the changes in the desertification degree of the CPEC. The results show that the extreme and severe desertification land in the CPEC accounts for 50% to 60% of the entire region. Mild desertification land accounts for about 20%， and non-desertification land and water bodies account for about 20%. Since 1998~2002， Pakistan experienced the worst drought in 50 years， so extreme desertification and severe desertification in 2000 reached the total area 61.8%. From 2005 to 2015， extreme desertification land had decreased， and it had been converted into severe desertification land， and some mild desertification land had been converted into non-desertification land. Overall， extreme desertification had a downward trend.
Geological disasters caused by surface deformation pose a great threat to the natural environment and society.SBAS-InSAR technology has become an important means of surface deformation monitoring with its advantages of high monitoring accuracy， large monitoring range and non-contact，to prevent geological disaster and reducing disaster losses， achieve the surface deformation monitoring is importmant.In this paper， the SBAS-InSAR technology was used to process sentinel-1A data acquired during January 7， 2018 and November 27， 2018 in Xining city， Qinghai province. The average surface deformation velocity distribution map was obtained. The deformations obtained by SBAS-InSAR were compared with those obtained by 8 GPS observation stations which were installed to monitor the deformation of NanShan in Xining. Except for one GPS stations， the RMS errors on the other 7 GPS stations are within 3 mm， which proves the reliability of SBAS-InSAR. Based on the SBAS-InSAR monitoring results， it is pointed out that the landslide is the main form of ground deformation in Xining city， especially along the Mutual aid beishan and G6 highway. The quantitative deformation data on a landslide which is in northeast of Xining railway station were obtained for the first time. The quantitative deformation data are important to the deformation analysis of the landslide and the safe operation of Xining railway station.
Shadows are the key factors affecting the identification accuracy of mountains coniferous forests using multi-spectral remote sensing data. Taking Tianshan as the study area， a comprehensive classification scheme was proposed， which comprehensive considered three aspects： the time-phase features of shadow distribution， classification features and classifiers. Firstly， to eliminate the influences of shadows on coniferous forest identification， shadow recognition and shadow reclassification were carried out. Then the altitude， Normalized Difference Vegetation Index （NDVI）， spectral slope of red to near-infrared band， blue reflectance band， red reflectance band， short-wave infrared band and slope were selected as the important features for identifying the Tianshan mountain coniferous forest. Finally， three often used classifiers （Support Vector Machine （SVM）， Random Forest （RF） and Back Propagation Neural Network （BPNN）） were compared. The results show that the terrain correction method can not effectively eliminate the mountain shadows， and it may cause over-correction， which affects the subsequent identification of coniferous. However， using two-phase images with large differences in solar elevation and azimuth to eliminate the influence of shadows on coniferous forest identification can improve the overall accuracy of coniferous forest by 1.3% to 3.7%； The SVM， RF and BPNN classifier can all achieve better classification accuracy， but the SVM classifier got the highest classification accuracy and Kappa coefficient with a value 93.33% and 0.87， respectively. The proposed remote sensing comprehensive classification scheme is expected to be applied to other mountain coniferous forest areas in the north arid and semi-arid regions after adjusting the parameters.
Due to the combination of tectonic background， groundwater extraction， active faults and other factors， the ground fissures in Xi'an have been intensified in recent years， causing many uneven surface subsidence. The urban geological hazard chain that has a strong destructive effect on underground buildings has attracted widespread attention from competent authorities and relevant experts. In order to infer the correlations between surface fissures and ground deformation， we use networked permanent scatterer Interferometric Synthetic Aperture Radar （NPSI） to characterize the land subsidence in Xi'an， which has significant advantages in the accuracy and reliability of urban surface deformation monitoring. In this study， we obtain 15 Sentinel-1A SAR image acquired covering the ground fissures in Xi'an between March 2017 and March 2018. Our InSAR observations were verified with leveling measurements， with an accuracy of ±4.75 mm. The results show that the ground fissures are developing in the southwestern suburbs of Xi'an. Over-withdrawal of groundwater and the construction of above-ground and underground facilities have exacerbated the trend of settlement and ground fissure development. Besides， the Yuhuazhai zones， Electicity Mall zones， Qujiang New District zones and the subway Line 3 need to monitor the development trend of ground fissures in real time， and rationally plan groundwater mining and engineering construction. This systematic research may serve as a reference for related research and for the operational departments of road administration and urban construction of the city.
Synthetic Aperture Radar （SAR） plays an important role in building collapse assessment after earthquake with its all-weather observation capability and rich texture information in SAR images. In order to solve the problems of multi-texture features of collapsed buildings in SAR images， such as low utilization rate and redundant feature information， a multi-texture feature classification method based on Principal Component Analysis （PCA） is proposed. This method extracts 26 kinds of texture feature information based on gray-level histogram， gray level co-occurrence matrix， Local Binary Pattern （LBP） and Gabor filters， constructs principal component variable for multi-dimensional feature selection and dimension reduction fusion， and extracts collapse information of buildings through Random Forest classification algorithm. Taking the Kumamoto earthquake in Japan in 2016 as an example to verify the effectiveness of this method， the results show that the extraction accuracy is up to 79.85%， the identification efficiency of collapsed buildings is improved， and the classification results are superior to each texture feature extraction method and multi-texture feature combination extraction method， which can be used for the rapid extraction of earthquake damage information of buildings.
In order to study the monitoring and warning ability of Fengyun-4A Lightning Mapping Imager （LMI） lightning data in severe convective weather， the Pre-TC squall line of Typhoon Lekima （2019） is taken as a case study. Based on lightning data from FY-4A LMI， combining with FY-4A Temperature of Black Body （TBB） cloud top data，the cloud-to-ground lightning location data （ADTD）， radar composed reflectivity factor data from National Radar Network， and wind and rain data from the automatic stations in southeast coast， the spatial and temporal distribution characteristics of the total lightning activity in the Pre-TC squall line and its relationship with the convective evolution are studied. The results show that the temporal and spatial changes of FY-4A LMI lightning rates are consistent with the evolution of the Pre-TC squall line. LMI lightning burst has an indicating function of about 1h in advance on the intensification of the Pre-TC squall line. In the relationship between lightning activity and convective evolution， the temporal and spatial characteristics of total lightning observed by LMI have good correlations with the evolution of satellite TBB and radar echoes. The time series of lightning rates has a corresponding relationship with the strong echo top height from 30 dBZ to 55 dBZ， which is consistent with the area changes of the cold cloud （≤-72 ℃ ） and the strong radar composed reflectivity factor （≥35 dBZ）. Lightning activity is mostly located in the areas of large TBB gradient to the left and front side of the TBB low value area，which has an indication on the possible locations of thunderstorm gale and heavy precipitation. The comparisons between LMI and ADTD showed that the characteristics of lightning activity observed by the two systems in the Pre-TC squall line of Lekima are basically consistent.
Remote sensing is an important method to obtain regional surface heat and water vapor fluxes， however there is a mismatch of spatial scale between the observation data and remotely-sensed data when the remotely-sensed data is verified. Combined with footprint analysis， this problem can be better solved， providing a reference basis for the verification of remote sensing models. Based on the eddy covariance data from the Arou station in the upper reaches of the Heihe River basin and the Daman Superstation in the middle reaches， the sensitivity analysis of the input parameters of three commonly used flux footprint models， namely Kormann&Meixner model （hereafter referred as KM）， Kljun model and Hsieh model was performed， and the difference in the footprint results of the three models at single time and daily scales is compared and analyzed. The objective of this study is not only to provide a reference basis for the reasonable selection of footprint model， but also to serve for the discrimination of data quality and the verification of relevant remote sensing models. The results showed that： （1） The KM and Hsieh models are very sensitive to the Obukhov length （L）. When L changes， the footprint result of Hsieh model varies much more than that of KM model， while Kljun model is less sensitive to L. Observation height （zm） and standard deviation of lateral wind fluctuations （σv） are also sensitive factors of the three models. （2） On the every 30 min time scale， the footprint results between KM and Hsieh models are in good agreement in extent and shape， but there are significant differences with Kljun model. The footprint extent of Kljun model is obviously smaller， while the estimated position of maximum flux contribution is much larger as compared to KM and Hsieh models. The peak distance of the footprint to the tower is obviously smaller than that of the other two models. （3） On a daily time scale， the flux contribution source area of the three models are similar in shape， but the source region of the Kljun model is the smallest. The results of this study provide important information for the selection of proper footprint models， which is used for data quality control and remotely-sensed products evaluation.
At present， remote sensing technology has become an important method for monitoring water quality parameters， and a more accurate water quality parameter inversion model is the focus of current water quality monitoring. However， due to multiple reasons such as the complexity of the water environment and the limitations of remote sensing data， the accuracy of water quality parameter remote sensing inversion is limited， and most of them focus on the inversion of water color water quality parameters. In order to obtain a better accurate water quality parameter inversion model， taking the lower reaches of the Haihe River in Tianjin as the research area， Landsat 8 OLI remote sensing images were subjected to atmospheric correction， radiometric calibration and other pretreatments， and the total phosphorus， ammonia nitrogen， total nitrogen concentration and conductivity of the water body were determined by laboratory physical and chemical analysis. The statistical regression model and neural network model of measured water quality parameters and Landsat 8 OLI remote sensing image data are established. Coefficient of determination （R2）， Mean Absolute Error （MAE） and Root Mean Square Error （RMSE） are used to test the accuracy， and the neural network model inversion results R2 is greater than 0.85， MAE is 0.019， 0.09， 0.242， 0.411， RMSE is 0.024， 0.118， 0.286， 0.562， and the inversion accuracy is better. The results show that the water quality parameter inversion model based on neural network has high accuracy.
With the increasing requirements of facility agriculture management， it is necessary to extract the spatial distribution information of plastic greenhouses with large range and low density in high-resolution remote sensing images as the basis for agricultural management and resource allocation. This study takes Tonglu County， Zhejiang Province as the study area， and uses high-resolution remote sensing images to compare and analyze the effect of extracting plastic sheds using different machine learning methods. It was found that the ENVINet5 deep learning architecture could perform plastic shed extraction and area estimation by small-sample semantic learning， and the overall accuracy and kappa coefficient reached 97.84% and 0.81； in the extraction results of U-net deep learning network， the overall accuracy and kappa coefficient were 96.22% and 0.79， which were better than the plastic shed extraction using support vector machine results. This study shows that the extraction of sparsely distributed plastic sheds in high-resolution remote sensing images by deep learning has good results and can provide support for agricultural cash crop management， planning， and weather assurance.
Promoting regional balanced， coordinated and sustainable development is one of the most important strategies in China. Net Primary Productivity （NPP） plays a significant role in indicating the sustainable development of Chinese ecological environment. This paper divides China into East and West based on the "Hu Huanyong Line". Then we studies and analyzes the spatio-temporal changes of NPP， population， and per capita NPP， especially the regional differences between east and west at pixel level and county level respectively. The results show that Chinese population has grown rapidly from 1.005 billion in 1982 to 1.395 billion in 2017. The proportion of population of the western region bounded by the "Hu Huanyong Line" has increasedfrom 5.91% to 6.42%； NPP has shown an overall growth trend among studying years， which increased from 2.69 Pg C in 1982 to 3.24 Pg C in 2015， with a growth rate of 16.60 Tg C/yr. The growth rate of NPP （12.30 Tg C/yr） in the east was nearly three times that （4.30 Tg C/yr） in the west； The per capital NPP in the West is much larger than that in the east and the whole China. In 1982， 2000， 2010， and 2017， the per capita NPP in the west and the whole China continued to decline， but the decline rate slowed down slightly. The per capita NPP in the east increased for the first time in 2017. Based on this， it can be seen that Chinese ecological environment is general in a state of restoration， while there are large differences between different regions. Therefore， regional differences should be fully considered in the formulation of relevant policies to achieve the coordinated development of Chinese ecological environment.
Fine identification of grass species is of great significance for grassland ecosystem degradation monitoring in the Three Rivers Source Region. Based on the UAV hyperspectral remote sensing system， the hyperspectral image of the typical grassland degradation area of Three-River Source Region was obtained. Firstly， using the obtained UAV hyperspectral image， the optimal bands combination were selected using XGBoost， the extended morphological attribute profile features were extracted and were combined with the selected spectral features. Secondly， sparse multinomial logistic regression and adaptive sparse representation methods were adopted to identify different grass species. Finally the shape adaptive based post-processing method was proposed to smooth the identification results. The results showed that： （1） Using the XGBoost method to select important spectral features can improve the identification result and save running time； （2） the spatial-spectral feature based method can effectively improve the identification result of grass species and the overall accuracy were improved by 4%~5% compared with the method of using only spectral features； （3） using two sparse representation methods，the overall accuracy of fine identification of grass species in the case of limited samples was 94.07% and 93.15% respectively， and the identification accuracy of various poison weed species was improved effectively by using shape adaptive post-processing method， which improved the overall accuracy by about 1.64% and 1.12%， respectively. The feature mining based sparse representation classification methods can achieve high-precision grass species fine identification of UAV hyperspectral images， and provide technical support for a wider range of grassland species fine identification.
Accurate crop identification and classification is an important part of agricultural remote sensing detection， which is very important for crop growth monitoring and yield estimation. In this paper， based on the Sentinel-2 time series images of the United States mixed agricultural belt as the research area， the Universal Normalized Vegetation Index （UNVI） for Sentinel-2 is calculated according to its sensor response function， and two comparisons are made. Experiment to analyze the performance of UNVI and other six indexes in the accurate classification of crops. Experiment 1 uses the JM （Jeffries-Matusita） distance as an indicator to analyze the separability between different crop categories. The results show that UNVI is better than NDVI， EVI， WDRVI， NDre1 and NDWI index. In corn and cotton， corn and rice， In terms of distinguishing between corn and rice， UNVI is better than other indexes in distinguishing ability， but in other crop combinations such as cotton and rice， NDVI and other indexes cannot distinguish them well. At this time， UNVI index can still perform better Distinguishing ability of experiment； Experiment 6 uses random forests and support vector machines to classify crops of the six time series index features. The results show that the UNVI index has the highest overall accuracy and Kappa coefficient， followed by the NDre1 index and the WDRVI index， and the EVI overall accuracy and The Kappa coefficient is the lowest， which indicates that UNVI distinguishes the four main crops of soybean， corn， cotton and rice in the study area better than the other five indexes. In summary， the UNVI index based on the Sentinel-2 time series has greater advantages in crop classification than other remote sensing vegetation indexes studied in this paper. UNVI can be used for agricultural research and application such as crop growth analysis and crop yield research Optional vegetation index.
Urban disaster prevention and mitigation has become an important part of urban security research. As an important place and effective way， emergency shelters are playing an increasingly important role in reducing urban disaster risk and improving urban disaster resilience. First of all， based on the survey of emergency evacuation sites in Lanzhou， this study uses GIS as the data storage and processing platform， and establishes a database of the study area based on the information of geology， hospital， firefighting， population， road and other information of Lanzhou. Secondly， established a set of evaluation index system for disaster mitigation ability of emergency shelters with security， accessibility， effectiveness， and indemnificatory by application of AHP. Lastly， according to the evaluation results of all aspects， among the 15 emergency shelters in Lanzhou， 4 have better disaster reduction ability， 8 have ordinary disaster reduction ability， and 3 have slightly less disaster reduction ability. Among them， the emergency shelter in Chengguan district has the best comprehensive disaster reduction capability.
