With the development of convolutional neural network technology， recent research has paid more attention to the improvement of accuracy and the improvement of semantic information. Mask R-CNN network is a further improved segmentation network of Faster R-CNN. It has a good segmentation effect in high-resolution remote sensing image feature recognition. However， since the convolutional neural network can only be trained and predicted with small tile images， there is a large semantic information error in the prediction results. Faced with this problem， this paper proposed a gap-repairing algorithm based on the defect of prediction result of convolutional neural network. The approach use overlapsize algorithm to improve the matching degree between the prediction result and the ground-truth result at first. Then fill the gap through the correlation function in the PostGIS database to repair the small tile， which can make it be spliced ??into a complete picture. The research and experiment results showed that the algorithm could improve the image semantic information well and has practicability.
To address the problem that most current deep learning models can only discriminate land use types for cropped images of high-resolution remote sensing images， this paper combines VGGNet and Mask R-CNN to carry out a study on intelligent construction land target detection. On the basis of establishing remote sensing image datasets of four types of land use types in the study area， we compare the classification accuracy of three convolutional neural network models， VGGNet， ResNet and DenseNet， and select the neural network model with the best classification effect， VGGNet and Mask R-CNN， to achieve intelligent construction land target detection. The results show that： （1） the classification accuracies of the three convolutional neural network models VGGNet， ResNet and DenseNet are 97.44%， 93.75% and 95.13%， respectively， and the VGG16 model has the least number of iterations and relatively less training time； （2） the Mask R-CNN threshold setting has an important influence on the target detection accuracy， when the threshold is set to is 0.3， the joint model of VGG16 combined with Mask R-CNN has the highest calibration frame accuracy for construction land detection. Also the joint model has higher accuracy than the single use of Mask R-CNN model for construction land detection， and shows more adaptability and robustness.
The overexploitation of mineral resources will have a serious negative impact on the natural environment. The monitoring of the mine environment is of great significance to the construction of ecological civilization. Machine learning algorithms have been widely used in traditional mine monitoring and have achieved good results. In recent years， with the rapid development of the field of deep learning， relevant theoretical knowledge has gradually been applied to remote sensing image processing. In this study， the deep learning algorithm is combined with the object-oriented method， and the GF-2 image is used to extract the land occupation type by applying the conventional neural network from the mining area in Yuzhou City， Henan Province. To compare the performance of the proposed methods， the support vector machine method was used. The results show that the overall accuracy of the convolutional neural network is 91.85% and the kappa coefficient is 0.90， which is higher than the support vector machine method. This paper shows the advantages and feasibility of this method in the extraction of open-pit mining areas and provides reliable technical support for the scientific management and environmental monitoring of open-pit mining areas.
To address the problem that traditional aircraft detection methods have low detection accuracy on remote sensing images with complex backgrounds and dense targets， an improved remote sensing image aircraft target detection algorithm based on Faster-RCNN （Faster-Regions with Convolutional Neural Network） is proposed. ResNet50 is used as the basic feature extraction network of the algorithm， and the dilated bottlenecks are introduced for multi-layer feature fusion to construct a new feature extraction network， which improve the feature extraction capability of the algorithm. First， the cross-validation training method is used on the UCAS-AOD data set to verify the stability of the model on different training sets and test sets， and compare the detection performance of different algorithms. Then， comparative experiment is conducted on the NWPU VHR-10 data set to verify the generalization of the model. Experimental results showed that： The average precision of the proposed algorithm is 97.1% on the UCAS-AOD data set and 96.2% on the NWPU VHR-10 data set. The study indicated that the proposed algorithm in this paper can not only improve the detection accuracy of aircraft in remote sensing images， but also have a stronger generalization， which has certain reference significance to the rapid detection of aircraft in remote sensing images.
With the rapid development of modern remote sensing technology， remote sensing image with high quality and quantity has been significantly promoted， new technology of high resolution remote sensing images contain more abundant information， how to make full use of the means of artificial intelligence auxiliary to mine these abundant information has become one of the important researches in remote sensing image analysis and understanding. At the same time， represented by deep convolutional neural networks based Artificial Intelligence （AI） technology is brilliant in the field of image processing. Thanks to the layer-wised convolutional and pooling structures which mimces human brain retinal systems， deep convolutional neural network can achieve excellent performance in image segmentation and classification. So this paper proposed a U-Net based model to extract features from high resolution remote sensing images with 2 m spatial resolution. Different from traditional methods based on hand craft image features， the proposed model can be automatically applied on massive amounts of high resolution remote sensing image feature extraction， it can also exert complicated nonlinear characteristics of high resolution remote sensing image with the help of the spectral features and texture features. The experimental results show that the time of using the U-Net model to calculate the land use classification of Xinchang County is 55.7s， and the accuracy is 90.95%， and the kappa coefficient is 0.86. U-Net model can quickly and accurately obtain the land cover features in high-resolution remote sensing images， and can get high-precision land use classification results， which shows that the deep learning into remote sensing image land use classification extraction has a broad prospect.
The accidents of the tailing ponds in China are frequent， the damage caused by dam breaking is extremely serious. The current quantity and distribution of tailings pond is necessary for preventing tailings pond accidents and carrying out emergency work in tailings pond. The traditional survey method is mainly based on ground investigations， which is difficult to achieve large-scale high-frequency monitoring. A tailing pond detection method based on deep learning detection was proposed in this paper， which can quickly identify the locations of the tailing ponds and obtain their geographical distribution. The suitable training samples are produced based on the study of the characteristics of the tailing ponds on the remote sensing image. SSD （Single Shot Multibox Detector） model is adjusted according to the samples characteristics during the model training. The extraction of the tailing ponds in the Beijing-Tianjin-Hebei Region is realized based on optimized model. The experiment result shows that there are 2 696 tailing ponds which were detected in the Beijing-Tianjin-Hebei Region，the recall reaches 93.3%.This paper realized the extract the tailings pond with the method of deep learning target detection， and has achieve good results which can provides method for the national and global extraction of tailing ponds.
Recently， Convolutional Neural Network （CNN） shows great potential in various remote sensing applications Taking Imperial County of California as the study area， and calculating vegetation index NDVI， EVI， RVI and TVI form landsat-8 OLI time series remote sensing images. Then， input it into the constructed CNN model to achieve crop classification. In order to verify the superiority of the convolution model， a deep learning model based on recurrent neural networks and its variants was built. The results show that： ①Adding other time series features can effectively improve the classification accuracy of CNN. The overall accuracy and Kappa coefficient of NDVI+EVI+TVI+RVI combination features are best， respectively 89.6674% and 0.8560， which is nearly 4% and 0.6 higher than the single time series features. ②Convolutional neural networks have the highest classification accuracy in comparison with other deep learning models. It can capture crop timing feature information more accurately from time series data. Although RNN is widely used for sequential data representation， but the classification results are slightly worse than the convolutional neural network. Experiments show that the introduction of other vegetation index assistance on the basis of NDVI can effectively improve the classification accuracy. A deep learning framework based on 1D convolutional neural networks provides an effective and efficient method for Multi-Temporal classification tasks.
At present， a quantity of urban construction waste is constantly produced and seriously accumulated， and its utilization rate is low， which endanger the urban ecological environment. The recognition of construction waste is the technical basis for the segmentation， extraction and monitoring of construction waste. However， it is difficult to identify and monitor construction waste due to its complex characteristics， the scale difference and spectral difference of remote sensing image. In this paper， a method of automatic identification of construction waste based on transfer learning and retraining model is proposed. Firstly， a sample bank is constructed according to the typical remote sensing features of construction waste. Then， based on the advanced international deep learning environment Tensorflow， the Inception-V3 model is retrained by using transfer learning， and the recognition model of construction waste is obtained. After verification， the overall recognition accuracy of construction waste can reach 94.88%. Compared with the traditional manual identification methods such as aerial photo monitoring and field investigation， the method studied in this paper has higher efficiency and recognition accuracy， which can provide a technical basis for real-time monitoring and accurate management of construction waste in the whole process.
Maize is the crop with the largest planting area in the middle reaches of the Heihe River， with large water requirements and high evapotranspiration during the growing period. Accurately obtaining the maize planting area has important significances for the adjustment of crop planting structure and reasonable planning of water resources in the region. The object of this paper is to assess the value of multi-temporal Sentinel-2 data for extraction of maize planting area in the middle reaches of the Heihe River from April to September 2019. The random forest algorithm was adopted in this work. The research methods were divided into two categories： extraction directly and two-step extraction. Further discussed the impact of multi- temporal information as input on the classification accuracy， and analyzed the importance of the input feature parameters of the model in the extraction process. The results showed that the two-step extraction method based on Sentinel-2 multi-temporal images could accurately extract the maize planting area in the study area with the overall classification accuracy of 85.03%， F1_Score of 0.70， and Kappa coefficient of 0.83. Compared with single image， multi-temporal images could effectively improve the accuracy of crop classification， obtaining differently crop phenology information. The research demonstrates the value of obtaining highly heterogeneous crop information based on high-resolution optical image combined with machine learning method.
Wheat is one of the main crops in China， which is of great significance to the economic development of China.With the continuous development of remote sensing technology， remote sensing technology has become an important means to extract wheat and growth monitoring. The identification of wheat is the premise of its planting area management， and the growth research is an important indicator of its growth evaluation and yield control. In this paper， the multi-source remote sensing data such as the hyperspectral zhuhai No.1 OHS-2A satellite， the multi-spectral Sentinel-2A satellite and MODIS were used to extract wheat by using Support Vector Machine（SVM） in Xiong'an as the research area. The accuracy of wheat was evaluated and analyzed by using the confusion matrix based on the field measurement data. Comparing the two important growth stages of wheat： the return green period and the heading period， wheat growth was divided into three grades （good growth， similar growth， worse growth） for growth monitoring and comparing. The results showed that under the same environmental conditions， the Overall accuracy of OHS-2A was 82.08%， and the Kappa coefficient was 0.76；The Overall accuracy of Sentinel-2A was 85.57% ，and the Kappa coefficient was 0.81， By contrast， Sentinel-2A is the best at identification wheat. In the process of growth monitoring， the Sentinel-2A satellite is more effective than MODIS in monitoring and analyzing the growth of Xiong'an wheat by comparing the growth conditions and the relative amplitude of the change of wheat growth.This study analyzed the status of wheat identification and growth monitoring in Xiong'an from remote sensing data of different resolutions， which is conducive to wheat planting management and the formulation of agricultural policies， which is of great significance for promoting the economic development of green Xiong’an and the city.
UAV platform with features of low cost and high flexibility has the potential to reduce weakness of the traditional remote sensing platform， and provides an effective way to collect near-surface measurements for the agricultural remote sensing community. Any UAV-based observations have the incident-view observation geometry under arbitrary scenario， while there is still a lack to understand angle-effect on UAV-based observations as well its propagation in following applications. We organized a field experiment to acquire the quadrat-level multi-angle observation over the sampled flowering paddy canopy through UAV to investigate the uncertainty induced by angles. Our results show that the maximum relative difference can reach up to 30.17%， 22.03% and 27.31% respectively at red， green and blue band， the deviation is up to 62.08% in the calculated visible vegetation indices， especially for NGRDI and VDVI index with an elevated variation. The research shows that the angel-effect is an important factor that cannot be ignored in the quantitative research based on UAV observations.
Estimating nitrogen content of apple leaves rapidly non-destructive and timely is the basis of ensuring apple yield and quality， and the inversion of leaf nitrogen content using hyperspectral technology can provide theoretical basis for reasonable fertilization. The spectral and corresponding leaf nitrogen content of apple leaves were analyzed and modeling in apple critical growing stage from 2012 to 2013 in Feicheng， Shandong Province. Based on the above data， the correlation between leaf nitrogen content and original spectrum， first order differential spectrum， three-sided spectral index was firstly analysed in order to select sensitive spectral index of leaf nitrogen content； Secondly， the spectral index NDSI and RSI was built which were sensitive to leaf nitrogen content； Finally， the prediction model of the apple leaf nitrogen content was established based on the way that was grey correlation analysis-partial least squares regression and out-of-bag data- random forest algorithm. The results showed： （1） The sensitive bands between leaf nitrogen content and original spectrum and first-order differential spectrum were 553， 711， 527， 708 and 559 nm； the spectral indices sensitive to leaf nitrogen content were NDSI（567，615）and RSI（554，615）； the best correlation between leaf nitrogen content and the three-sided spectral index was Sdy. （2） The result showed that OOB-RF estimation model had better accuracy and reliability， which can guide fruit tree variable fertilization using leaf nitrogen content. This way achieved prediction of leaf nitrogen content between regional and annual levels， and had a wide range of potential applications.
As a product of the development of modern industry and mining， soil heavy metal lead pollution has gradually invaded agricultural production and agricultural products. Hyperspectral technology has become an important method for monitoring heavy metals in soil due to its macroscopic， rapid and efficient characteristics. This study takes the Pb element of vineyard soil in Xinjiang Turpan Basin as the research object， analyzes the relationship between soil spectral reflectance data and soil Pb content under 15 spectral transformations including the original soil spectrum， and constructs a partial least square regression of soil Pb content （ PLSR） model and geographic weighted re-regression （GWR） model， comparative analysis and discussion of the feasibility of using soil hyperspectral to estimate the vineyard soil Pb content. The results show that the original spectral reflectance of the soil can effectively enhance the spectral characteristics of the vineyard soil Pb element and the estimation accuracy of the model through spectral transformation. Among them， the SRSD transformation PLSR model and GWR model have the best estimation capabilities. The GWR model is better than the PLSR model to explain the hyperspectral estimation of the heavy metal Pb content in vineyard soil. From the perspective of model stability and accuracy， in the SRSD differential transformation， the GWR model R2 is increased from 0.262 of the PLSR model to 0.866， and the RMSE is reduced by 1.009. Using GWR model can effectively improve the accuracy of estimating the Pb content of vineyard soil. This study provides a useful reference for the study of soil heavy metal pollution and soil environmental safety in Chinese vineyard bases.
Taking Nanning County of Jilin Province as the research area， using Sentinel-1B dual polarization data as data source， multiple texture eigenvalues of typical crops such as corn， soybean and rice were extracted， and the best crop identification parameters were selected. Combined with eCognition software The rule base in the model fully mines the attribute information contained in the texture features of crops in SAR data， constructs a decision tree， extracts typical crops based on object-oriented classification methods， and obtains the optimal classification phase of crops in the study area through the analysis of SAR crop extraction results. And the best crop identification texture information combination， classify and map each typical crop， and explore the feasibility of improving the accuracy of crop identification based on the back-scattering characteristics of SAR images. The results show that SAR data can provide richer crop texture information than optical data. Selecting suitable texture information as auxiliary data for classification can effectively solve the phenomenon of "foreign matter homology" in optical data and improve the accuracy of crop identification. The three SAR texture features that contribute the most to crop extraction are： mean， variance， and dissimilarity.
Irrigation is an effective regulation mechanism for crops to response to extreme climatic conditions such as drought. Due to global climate change， the frequency and severity of extreme weather events such as drought are expected to increase in the future， quantitative analysis of the impact of drought on crop growth of farmland ecosystem under irrigation and rain-fed conditions will help to better assess the ability of human beings to cope with the negative impact of extreme climate events on the ecosystem， and provide a basis for formulating reasonable and effective ecosystem protection measures. The dry lands on northern China is taken as the study area. Based on Standardized Precipitation Evapotranspiration Index （SPEI） products and Enhanced Vegetation Index （EVI）， Gross Primary Productivity （GPP） products provided by MODIS， this paper analyzes the trends of drought and EVI， GPP of farmland ecosystems in the study area from 2000 to 2014 by using MK trend analysis and explores the lag time of crop productivity response to drought by using Pearson correlation coefficient. Then， the effects of drought on EVI and GPP of farmland ecosystem under the corresponding time lag are analyzed by using linear regression analysis and the differences in the effects of drought on EVI and GPP of irrigated farmland and rain-fed farmland are further compared. Study results indicate during 2000~2014， 64.10% of the study area showed a trend of drought alleviation， and 75.78% and 81.87% of the study area showed a trend of increased EVI and GPP， of which 64.82%， 68.34% of the areas with an increase in EVI， GPP were accompanied by drought alleviation. Expect for the lag time of rain-fed crop EVI in semiarid dry land response to drought was 2 months， all the rest lag time was 1 month. Based on the lag time， the SPEI and EVI， GPP showed a significant positive correlation. Compared to rain-fed farmlands， irrigation alleviated the negative effects of drought on EVI and GPP by 32.22% and 29.42%. The degree of mitigation in arid area is overall higher than that in semi-arid area. This study quantifies the differences of the impact of drought on the GPP and EVI of irrigated and rain fed farmland ecosystems， which provides a reference for the study of the impact of irrigation resistance on vegetation ecosystems.
Crop biomass is a vital fundamental substance for predicting yield. Remote sensing is an important technology to monitor crop above-ground biomass efficiently and objectively， which is of great significance for agricultural production and management. Taking Longkang farm in Anhui province as the research area， this paper analyzes the relationship between aboveground biomass of winter wheat and 4 LAI-VIs， 2 DMIs and 8 combined vegetation indices by PROSAIL simulation spectrum， and builds retrieval models. The results show that the correlation between DMIs and crop biomass is higher than LAI-VIs， and the combined vegetation index enhances the crop biomass detection ability of commonly used vegetation indices. The biomass retrieval models are validated with the measured biomass of winter wheat， and the results show that the combined vegetation index generally improves the above-ground biomass retrieval accuracy of single vegetation index， among which MTVI2×NDMI has the highest accuracy （RMSE=606.8 kg/hm2）.This paper provides a new technique for high precision retrieval of crop above-ground biomass.
In this study， Sentinel-2 data combined with Random Forest method （RF） and Support Vector Machine method （SVM） were used to extract crop information in the Shandian River Basin in Inner Mongolia. Three schemes are proposed： pixel-based classification algorithm， object-based classification algorithm and improved integration algorithm based on pixel-based classification and object-based segmentation. Results are as follows： （1） pixel-based classification with RF gets the best extraction accuracy， with an overall accuracy up to 97.8% and Kappa coefficient of 0.974. This result shows that RF has evident advantages in crop extraction. （2） The improved integration algorithm also has good extraction accuracy. The overall accuracy is 96.4%， and kappa coefficient is 0.957. This method fully combines the advantages of pixel-based and object-based classification methods， which effectively improves the crop classification effect in Shandian River region.
In order to improve the stability of the small sample data model， a remote sensing detection model of wheat stripe rust with higher accuracy and better robustness was constructed. Firstly， the data of canopy solar-Induced chlorophyll Fluorescence （SIF） were extracted based on radiance and reflectance fluorescence index method， and then combined with reflectance spectral index sensitive to severity of wheat stripe rust， the Gradient Boost Regression Tree （GBRT） was used to detect wheat stripe rust. By comparing GBRT model with CART and Multiple Linear Regression （MLR） model， the results showed that： （1） Reflectivity derivative fluorescence index D705/D722， short-wave infrared Valley reflectance and reflectance ratio fluorescence index R740/R800 were the main factors affecting the accuracy of remote sensing detection of wheat stripe rust. The importance of chlorophyll fluorescence data was higher than that of reflectance spectrum data， and canopy SIF could reflect wheat stripe rust information more sensitively than reflectance spectrum. （2） Compared with CART model and MLR model， the Root Mean Square Error （RMSE） of GBRT model was reduced by 15.50% and 13.49%， and the determination coefficient （R2） was increased by 6.16% and 11.57% respectively. The estimated DI value of GBRT model is closer to the measured value， and the fluctuation of the estimated result is low， and the robustness of CART model is high. In small sample data， it is easy to divide data sets with different features into subsets of the same feature， and the prediction results fluctuate greatly. The prediction results of MLR model are relatively stable， but its prediction accuracy is low.
Spatial pattern of forest gap mitigates the diversity of understory species， and is also an effective structural indicator to quantify the characteristics of forest types. Recently， the rapid development of photography technology combined with flexible and convenient small Unmanned Aerial Vehicle（UAV） provides a low-cost way to obtain high-resolution and 3D structural information of forest canopies. Traditionally， the pattern metrics describing the pattern of the forest gaps was calculated based on the gap mapping by using aerial imagery . However， it is difficult to extract small forest gaps， especially when it is necessary to process a large amount of aerial data， which will greatly increase the time cost. Based on the RGB images acquired from UAV aerial photography， the relative height model （DSMr） was established on the basis of high-resolution Digital Surface Model （DSM） constructed by 3D modeling. We put forward a method that can quickly reflect the spatial distribution pattern of forest gaps based on the parameters including information entropy （H）， standard deviation （STD）， skewness coefficient （SK）， kurtosis coefficient （EK） and texture parameter （GLCM，GLDV） of the DSMr. Then the validity of metrics are tested by using the forest gap data of a well restored natural forest site a in the central Loess Plateau and a forest site shaped by plantations （Robinia pseudoacacia） in the north Loess Plateau. The results showed that the SK， EK， GLCM and GLDV have a positive correlation with the traditional pattern indices of forest gap. Both the SK and EK are significantly negatively correlated with the edge density index （ED）； the SK is negatively correlated with the patch density （PD）. The texture parameters are positively correlated with landscape shape index （LSI）. In summary， DSMr parameters can effectively indicate the structure and distribution pattern of forest gaps， and the quantification of three-dimensional structure of forest gaps provides more readily available variables for researches of forest gaps.
The problem of “same object with different spectrum” and “different objects with same spectrum” makes that it difficult to obtain high classification accuracy for hyperspectral images using the single spectral information. Texture feature is the important structural information of spatial distribution of ground objects， which can compensate for the deficiency of spectral features in the classification to some extent. Many texture feature extraction methods have been developed in hyperspectral image classification， but they are lacking of a comprehensive comparative analysis. Therefore， this paper aim to explore the classification performance of different texture feature extraction methods. The 8 selected methods include rotational invariant local binary mode （riLBP）， Simple Linear Iteration （SLIC）， Extended Morphological Profile （EMP）， Differential Morphological Profile （DMP）， Attribute Profile （AP）， 3D-Gabor， Joint Bilateral Filtering （JBF） and Guided Filtering （GF） design classification experiments. Experimental results on Indiana Pines， Pavia University and Xiong'an datasets show that EMP behaves better than other methods both in overall classification accuracy and computational speeds.
Global climate change has substantially induced significant changes in vegetation growth events， such as timing of spring， autumn and magnitude of growth peak. Growth season length and its control parameter spring onset and autumn offset have been widely studied. Vegetation growth peak represents the maximum capability of photosynthesis and responds sensitively to climatic change. Currently， studies on the dynamic of vegetation peak growth and the underlying mechanism remain limited and should be analyzed deeply at different regions. This paper took Northeast China （NEC） as an example， then used long-term satellite-based NDVI and logistic method to derive key vegetation phenological parameters. The spatiotemporal dynamics of vegetation peak phenology and its impact on vegetation production were estimated. The results indicate that： position of peak （POP） and spring onset （SOS） show a delayed trend with an amplified PEAK. These parameters also present a period cycle of 11 year. From spatial range and magnitude， temperature and precipitation prior POP have a weak impact on PEAK， which mainly distribute in grassland. Temperature and precipitation has a significant impact on forests in the northern part of NEC. The impact from SOS on PEAK and POP is more obvious than that from temperature and precipitation， and mainly affect vegetation PEAK in forest and grassland. SOS plays a key role in regulating POP in cropland. Vegetation peak growth has an important and positive effect on vegetation production. The varying spatiotemporal patterns of POP， PEAK reflect the difference in their capability for adapting to climatic change and shape the pattern of carbon sink. This study can improve our understanding spatiotemporal changes of photosynthesis and carbon cycle in the context of climate change. Besides， these findings are helpful for ecological system management and estimation in Northeast China.
Mangrove forests are characterized as high productivity and high carbon storage coastal vegetation inhabited in the intertidal zone of tropical and subtropical region. They play a significant role in global carbon balance. Previous studies have made achievements in estimating and analyzing mangrove primary production on eddy flux tower site scale， however， rare experiments were conducted on the estimation of mangrove Gross Primary Production（GPP） on regional scale due to the limited remote sensing image resolution and patchy distribution of mangrove. In this study， we first combined high spatiotemporal resolution vegetation index datasets produced by data fusion technique and eddy flux data to calibrate and validate light use efficiency model， and then applied the model to estimate mangrove GPP in our study region. Based on our method， a high spatiotemporal resolution dataset of mangrove GPP in Gaoqiao， Guangdong province in 2012 was established. The overall accuracy of our dataset （R2=0.64） outperformed MOD17A2 and GLASS GPP product with the increase of 48.9%. Experiments results showed that the maximum light use efficiency of mangrove in Gaoqiao is 3.07 g C MJ-1， and annually average GPP is 1 915.4 g C m-2 a -1 in our study site. Besides， seasonally average GPP of Gaoqiao mangrove is higher in summer and autumn than spring and winter. Our method and dataset can be served for the regional-scale mangrove production research， as well as are effective support for mangrove protection.
Aiming at the limitations of statistical data on the spatial distribution information of GDP， a method of GDP spatialization was proposed by coupling GDP statistics， urban Point Of Interest（POI）， nighttime lighting data and land use data. First， taking Beijing as an experimental area， using the relationship between land use type and GDP， the 100m grid of the first industry GDP will be spatialized. Secondly， establishing the spatialized expression model of GDP in the second and third industries by determining coupling the POI kernel density， the nighttime lighting index and land use data. Spatially integrate the results of sub-industry and complete the GDP distribution results of Beijing with 100 m resolution finally. The results show that the multi-source coupling model with the POI is higher than the general nighttime light index single factor model. R2 is increased from 0.84 to 0.92. The generated results can better reflect the spatial difference of regional GDP. The results show that multi-source coupling model can improve the spatialization accuracy of GDP. POI can effectively reflect the spatial distribution of GDP and provide data support for GDP spatialization.
