CN116229459A - Domestic satellite multispectral image pixel-by-pixel quality marking method - Google Patents

Domestic satellite multispectral image pixel-by-pixel quality marking method Download PDF

Info

Publication number
CN116229459A
CN116229459A CN202211504671.2A CN202211504671A CN116229459A CN 116229459 A CN116229459 A CN 116229459A CN 202211504671 A CN202211504671 A CN 202211504671A CN 116229459 A CN116229459 A CN 116229459A
Authority
CN
China
Prior art keywords
cloud
pixel
area
snow
data
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211504671.2A
Other languages
Chinese (zh)
Inventor
胡昌苗
赵理君
霍连志
李宏益
唐娉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Aerospace Information Research Institute of CAS
Original Assignee
Aerospace Information Research Institute of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Aerospace Information Research Institute of CAS filed Critical Aerospace Information Research Institute of CAS
Priority to CN202211504671.2A priority Critical patent/CN116229459A/en
Publication of CN116229459A publication Critical patent/CN116229459A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/24—Aligning, centring, orientation detection or correction of the image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Remote Sensing (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Astronomy & Astrophysics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)
  • Radio Relay Systems (AREA)

Abstract

The application relates to the technical field of remote sensing image processing, and provides a pixel-by-pixel quality marking method for domestic satellite multispectral images. The key steps involved in the method are realized by adopting a mature algorithm, and the method has higher stability.

Description

Domestic satellite multispectral image pixel-by-pixel quality marking method
Technical Field
The application relates to the technical field of remote sensing image processing, in particular to a domestic satellite multispectral image pixel-by-pixel quality marking method.
Background
The mass multispectral image data acquired by the remote sensing satellite is used for producing standardized data products which are convenient for users in different fields through a complex customized algorithm. In recent years, the quality of the remote sensing satellite image standard data product is greatly improved, the quantification precision of the data product is improved, pixel-by-pixel quality marking of the data product is realized, the pixel-by-pixel marking of cloud, cloud shadow, land, water, snow and other types is realized, and a data user can conveniently interfere with pixels according to a using field Jing Guolv, such as filtering cloud and cloud shadow when researching surface change and classification. More data sources are currently used abroad, including MODIS, landsat (TM)/ETM+/OLI, and Sentinel, etc., data that contain quality-coded data products that are typically provided to data users as components of standard data products in separate bands or images.
The core of the quality marking algorithm in the remote sensing satellite multispectral image standard data product is cloud and cloud shadow detection, more than 50% of global satellite remote sensing data displayed by the data of the International Satellite Cloud Climatology Project (ISCCP) are covered by the cloud, the cloud/cloud shadow detection and marking are important preconditions for efficiently developing remote sensing application, and meanwhile, the marking of the categories such as water, snow and the like has high value for subsequent remote sensing application. The threshold method is the most main cloud detection method at present, the cloud is identified by utilizing the reflectivity difference of the cloud and typical objects in visible light and near infrared bands and the temperature difference in thermal infrared bands, and relatively high precision is obtained on foreign satellite data. Typical fixed threshold cloud detection algorithms at present are an ISCCP method, a CLAVR method, an APOLLO method and the like. Represented by Landsat series satellites, a standard data product of which, collection 1/2, contains QA quality-marked band data using CFmask [1] Cloud detection algorithm comprehensively utilizing visible light, near infrared, short wave infrared and thermal infraredThe outer wave band is combined with a threshold value to distinguish cloud, shadow under the cloud, highlight the ground surface, snow and the like, and compared with the original ACCA algorithm, the accuracy is improved.
China is a large country of remote sensing satellite emission at present, a multispectral sensor with medium and high resolution is a main imaging load, in recent years, a plurality of satellite data standard products begin to contain pixel-by-pixel quality marking data, and research on a quality marking algorithm aiming at the latest domestic satellite data becomes a task and a challenge of a plurality of domestic scholars. Compared with foreign satellite data, the main difference of the multi-spectrum data of the domestic satellite at present is that the quantity of wave bands is small and the quantification degree is not high. The multi-spectral images of domestic satellites are mostly four-band data from visible light to near infrared, and due to the lack of a medium-wave infrared band, the quality marking algorithm cannot utilize the low-temperature characteristic of high cloud in the infrared band, so that the cloud is difficult to effectively distinguish from the highlight earth surface spectrum approximation such as snow, desert and the like, and the quality marking algorithm becomes a main error source of the domestic satellite quality marking algorithm.
At present, three main ideas for improving the precision of a domestic multispectral satellite quality marking algorithm exist: first, the accuracy is improved by combining an image processing algorithm, such as Shen Huanfeng of university of Wuhan and the like [2] And the edge correction is carried out on the cloud by utilizing the guide filtering, so that the precision of the edge region of the domestic GF-1/2 satellite WFV multispectral image is improved. Secondly, accuracy is improved by combining multisource auxiliary data, such as Lin and the like [3] A dynamic threshold cloud detection algorithm (UDTCDA) supported by a priori surface reflectivity database is presented. Thirdly, a deep learning algorithm is utilized to lead the model to have the similar purpose of manual interpretation through learning training of a large number of data samples, and targets such as clouds, snow and the like are directly distinguished in RGB color images, for example, cloud-attU based on U-Net network [4] The cloud detection method identifies based on the unique features. The researches provide good reference for domestic multispectral satellite quality marking standard products.
Reference is made to:
[1]Zhu，Zhe，and Curtis E.Woodcock.2012."Object-based cloud and cloud shadow detection in Landsat imagery."Remote Sensing of Environment 118:83-94.doi:10.1016/j.rse.2011.10.028.
[2]Zhiwei Li，Huanfeng Shen，Huifang Li，Guisong Xia，Paolo Gamba，Liangpei Zhang.Multi feature combined cloud and cloud shadow detection in GaoFen-1wide field of view imagery[J].Remote Sensing of Environment 191(2017)342-358.
[3]Lin sun，jing wei，jian wang，et al.A Universal Dynamic Threshold Cloud Detection Algorithm(UDTCDA)supported by a prior surface reflectance database[J].Journal of Geophysical Research，D.Atmospheres:JGR，2016，121(12):7172-7196.DOI:10.1002/2015JD024722.
[4]Guo，Yanan，Xiaoqun Cao，Bainian Liu，and Mei Gao.2020."Cloud Detection for Satellite Imagery Using Attention-Based U-Net Convolutional Neural Network."Symmetry 12(6):1056.doi:10.3390/sym12061056.
disclosure of Invention
The embodiment of the application provides a domestic satellite multispectral image pixel-by-pixel quality marking method, which aims at the data preprocessing application of the domestic satellite multispectral image, in particular to the production of multispectral image standard data products, and provides an automatic pixel-by-pixel quality marking algorithm flow.
The embodiment of the application provides a domestic satellite multispectral image pixel-by-pixel quality marking method, which comprises the following steps:
preparing reference data, namely preparing a snow cover product and a DEM elevation image of a low-resolution satellite standard earth surface reflectivity product which cover a geographic range and are close to an imaging date according to an input multispectral image, and performing cutting and re-projection to obtain reference data with the same projection, resolution and pixel size as the input image;
dividing the surface into three categories of a water body area, a flat area and a mountain area by using the DEM elevation image, and adopting different quality marks for the three categories to generate a single-band quality mark mask image;
cloud detection, namely, for three categories of a water body area, a flat area and a mountain area, respectively utilizing different fixed thresholds to detect thick clouds with high confidence, and marking the clouds in a quality marking mask image;
dynamic threshold cloud region correction, for a flat region and a mountain region, constructing a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data, if the apparent reflectivity of a cloud pixel in a quality mark mask is smaller than the maximum value of clear surface apparent reflectivity calculated by the linear function in each wave band, correcting the cloud pixel into a clear surface, and correcting the edges of the cloud region by utilizing large-scale guide filtering for the rest cloud pixels;
step 500, detecting the snow mountain, judging whether the snow mountain exists or not by using low-resolution snow coverage data in reference data and a DEM elevation image, if so, performing snow mountain simulation to obtain a simulated snow mountain mask, and correcting a false detection cloud zone of the mountain in the quality mark mask into snow by using the simulated snow mountain mask;
shadow detection and cloud shadow matching, for flat areas and mountain areas, shadow pixels with high confidence coefficient are detected by using a fixed threshold value, shadow edges are corrected by using guide filtering, azimuth angles and distance ranges of cloud shadows and cloud shadows are calculated by using imaging geometric relations, cloud shadow correspondence is determined by using maximum pixel number matching of cloud shadows and cloud shadow communication areas, and shadow communication area pixels without matching relations are filtered;
detecting snowfall, namely correcting a cloud pixel communication area which is not matched with shadows in a quality mark mask into snow by utilizing low-resolution snow coverage data in reference data for a middle-high latitude winter snowfall area in a flat area;
and integrating the masks to obtain quality mark mask images containing cloud, snow, clear surface, water and filling value categories.
In one embodiment, the using DEM elevation images to divide the surface into three categories, water, flat and mountainous, includes:
the flat area and the mountain area are distinguished by combining slope and height statistics through a flood filling algorithm; different quality marks are adopted for the three types, three types of the Byte type quality mark mask images in the numerical range of 0 to 255 are marked in ten positions, and other types of the Byte type quality mark mask images are marked in units.
In one embodiment, for a flat area and a mountain area, constructing a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data, and constructing a dynamic threshold cloud detection model for obtaining visible light and near infrared 4 channels by utilizing a least square method to fit the linear function relation between the apparent reflectivity and the surface reflectivity, wherein a fitted data point set meets the condition that low-resolution surface reflectivity pixels correspond to a pixel area of a clear surface of the flat area and the mountain area of the high-resolution input image, and the pixel area is approximately homogeneous surface; if the apparent reflectivity of a cloud pixel in the mass label mask is less than the maximum value of the apparent reflectivity of the linear function calculation clear surface at each band, the pixel is corrected to the clear surface.
In one embodiment, the correcting the cloud pixel communication area of the quality mark mask, which is not matched with the shadow, to the snow by using the low-resolution snow coverage data in the reference data is to firstly determine whether the low-resolution snow coverage data corresponds to the cloud pixel of the flat area of the quality mark mask, and if the low-resolution snow coverage data exists and the cloud pixel communication area is not matched with the shadow, correcting the cloud pixel communication area to be the snow.
According to the pixel-by-pixel quality marking method for the domestic satellite multispectral image, which is provided by the embodiment of the invention, on the basis of combining a threshold method with algorithms such as guide filtering, connected region pixel matching and the like, the algorithm is used for respectively carrying out quality marking processing on the earth surface into three categories of a water body area, a flat area and a mountain area by utilizing a numerical elevation, correcting a cloud detection threshold value by utilizing a heterogeneous low-resolution earth surface reflectivity data product, correcting snow mountain false detection existing in the mountain area by utilizing a heterogeneous low-resolution snow coverage data product and a digital elevation, correcting snow false detection existing in the flat area by utilizing a heterogeneous low-resolution snow coverage data product, and finally obtaining a quality marking image containing cloud, snow, clear earth surface, water and filling value categories. The key steps involved in the invention are realized by adopting a mature algorithm, and the method has higher stability.
Compared with the prior art, the application has the following characteristics: the invention provides a solution for a domestic satellite multispectral image quality marking algorithm. The algorithm is fully automatic, the whole process does not need human-computer interaction, and a user only needs to simply check the final detection result. The key steps are realized by adopting a mature algorithm, and the stability and applicability are higher. The method provides a key technical support for automatically producing high-precision quality marked data products by using mass data of domestic satellites. The technology is a necessary data preprocessing process in domestic satellite data standard product production research, and solves the problems of unstable cloud detection precision of a fixed threshold value and indistinguishable spectrum of cloud and snow in visible light and near infrared bands by introducing multi-source reference data. Provides a feasible technical solution for producing domestic satellite data quality marking data products.
Drawings
For a clearer description of the present application or of the prior art, the drawings that are used in the description of the embodiments or of the prior art will be briefly described, it being apparent that the drawings in the description below are some embodiments of the present application, and that other drawings may be obtained from these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flowchart of a domestic satellite multispectral image quality marking algorithm of a domestic satellite multispectral image pixel-by-pixel quality marking method provided by an embodiment of the application;
fig. 2 is a schematic diagram of edge correction of a large-scale guide filter of a pixel-by-pixel quality marking method for a domestic satellite multispectral image according to an embodiment of the application;
fig. 3 is a schematic diagram of pixel matching of cloud and shadow areas under the cloud in a domestic satellite multispectral image pixel-by-pixel quality marking method according to an embodiment of the present application;
fig. 4 is a diagram of an example of processing high-resolution first satellite data by an algorithm of a method for marking quality of a domestic satellite multispectral image pixel by pixel according to an embodiment of the present application.
Detailed Description
For the purposes of making the objects, technical solutions and advantages of the present application more apparent, the technical solutions in the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
The technical idea is based on two limiting characteristics of the current domestic satellite multispectral remote sensing image data quality marking algorithm: 1) The cloud detection precision of the fixed threshold method is limited, and the accurate boundary of the cloud is more difficult to effectively divide. 2) The effective wave band of cloud detection is missing, only comprises visible light and near infrared wave bands, and cannot be detected by utilizing the low-temperature characteristic of thick cloud in the infrared wave band, so that a great amount of false detection of snow and high-brightness ground surface exists. Aiming at the two problems, combining with the existing related research results, the technology designs a set of automatic domestic satellite multispectral image quality marking algorithm flow, dynamically corrects a cloud detection threshold by means of heterologous reference data, and distinguishes snowy mountain from snowfall false detection so as to meet the requirement of automatically producing high-precision pixel-by-pixel quality marking product data for mass domestic satellite data.
The method for marking the multi-spectral image pixel-by-pixel quality of the domestic satellite provided by the invention is described in detail below by combining with the embodiment.
Fig. 1 is a flowchart of a domestic satellite multispectral image quality marking algorithm of the domestic satellite multispectral image pixel-by-pixel quality marking method provided in an embodiment of the application. Referring to fig. 1, an embodiment of the present application provides a method for marking quality of a domestic satellite multispectral image pixel by pixel, where the method may include:
preparing reference data, namely preparing a snow cover product and a DEM (Digital Elevation Model ) elevation image of a low-resolution satellite standard earth surface reflectivity product which covers a geographic range and has an imaging date close to each other according to an input multispectral image, and performing cutting and re-projection to obtain reference data with the same projection, resolution and pixel size as the input image;
dividing the surface into three categories of a water body area, a flat area and a mountain area by using the DEM elevation image, and adopting different quality marks for the three categories to generate a single-band quality mark mask image;
cloud detection, namely, for three categories of a water body area, a flat area and a mountain area, respectively utilizing different fixed thresholds to detect thick clouds with high confidence, and marking the clouds in a quality marking mask image;
dynamic threshold cloud region correction, for a flat region and a mountain region, constructing a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data, if the apparent reflectivity of a cloud pixel in a quality mark mask is smaller than the maximum value of clear surface apparent reflectivity calculated by the linear function in each wave band, correcting the cloud pixel into a clear surface, and correcting the edges of the cloud region by utilizing large-scale guide filtering for the rest cloud pixels;
the method comprises the steps of detecting snow mountains, judging whether the snow mountains exist or not by using low-resolution snow coverage data in reference data and DEM elevation images for mountain areas, if so, performing snow mountain simulation to obtain a simulated snow mountain mask, and correcting false detection cloud areas of the mountain areas in a quality mark mask into snow by using the simulated snow mountain mask;
shadow detection and cloud shadow matching, for flat areas and mountain areas, shadow pixels with high confidence coefficient are detected by using a fixed threshold value, shadow edges are corrected by using guide filtering, azimuth angles and distance ranges of cloud shadows and cloud shadows are calculated by using imaging geometric relations, cloud shadow correspondence is determined by using maximum pixel number matching of cloud shadows and cloud shadow communication areas, and shadow communication area pixels without matching relations are filtered;
detecting snowfall, namely correcting a cloud pixel communication area which is not matched with shadows in a quality mark mask into snow by utilizing low-resolution snow coverage data in reference data for a middle-high latitude winter snowfall area in a flat area;
and integrating the masks to obtain quality mark mask images containing cloud, snow, clear surface, water and filling value categories.
In one embodiment, the using DEM elevation images to divide the surface into three categories, water, flat and mountainous, includes:
the flat area and the mountain area are distinguished by combining slope and height statistics through a flood filling algorithm; different quality marks are adopted for the three types, three types of the Byte type quality mark mask images in the numerical range of 0 to 255 are marked in ten positions, and other types of the Byte type quality mark mask images are marked in units.
In one embodiment, for a flat area and a mountain area, constructing a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data, and constructing a dynamic threshold cloud detection model for obtaining visible light and near infrared 4 channels by utilizing a least square method to fit the linear function relation between the apparent reflectivity and the surface reflectivity, wherein a fitted data point set meets the condition that low-resolution surface reflectivity pixels correspond to a pixel area of a clear surface of the flat area and the mountain area of the high-resolution input image, and the pixel area is approximately homogeneous surface; if the apparent reflectivity of a cloud pixel in the mass label mask is less than the maximum value of the apparent reflectivity of the linear function calculation clear surface at each band, the pixel is corrected to the clear surface.
In one embodiment, the correcting the cloud pixel communication area of the quality mark mask, which is not matched with the shadow, to the snow by using the low-resolution snow coverage data in the reference data is to firstly determine whether the low-resolution snow coverage data corresponds to the cloud pixel of the flat area of the quality mark mask, and if the low-resolution snow coverage data exists and the cloud pixel communication area is not matched with the shadow, correcting the cloud pixel communication area to be the snow.
The following are included in the present application:
input data: a domestic multispectral image relates to image data acquired by a medium-high resolution multispectral sensor mounted on a domestic military/civil satellite, such as satellites of GF series, HJ series, ZY series, JB series and the like. The spatial resolution of the acquired multispectral image is not lower than 30 meters, and only comprises four wave bands from visible light to near infrared. The input data has been converted from the original DN (Digital Number) to the apparent reflectivity by a scaling parameter.
The processing steps are as follows: and automatically inquiring and downloading by referring to the data, automatically downloading a low-resolution satellite standard earth surface reflectivity and snow cover product which cover the geographical range and are close to the imaging date on line by an algorithm through a network according to the geographical range and the imaging time of the input multispectral image, extracting a corresponding earth surface reflectivity wave band and a corresponding snow cover wave band, and cutting and reprojecting to obtain Byte images with the same projection, resolution and pixel size as the input image.
The standard earth surface reflectivity product of the low-resolution satellite can use a 500-meter resolution MOD09A1 product of an MODIS series satellite at present, the data source can realize automatic network online query downloading by using the geographic range and imaging time code of an input image after registering an account, the product performs accurate positioning and geometric correction based on high observation coverage, low view angle, no cloud or cloud shadow and aerosol concentration, and selects the optimal Level 2Grid of each product pixel within 8 days for observation, has higher overall accuracy, and uses a blue light wave band (0.459-0.479 mu m), a green light wave band (0.545-0.565 mu m), a red light wave band (0.62-0.67 mu m) and a near infrared wave band (0.841-0.876 mu m) of the MOD09A1 product.
The standard snow cover product of the low-resolution satellite can be used at present, and the standard snow cover product of the 500 m resolution of the MODIS series satellite can be also used for automatically inquiring and downloading on line, downloading MOD10A2 data synthesized in 8 days or MOD10A1 data day by day, and the snow coverage rate wave band (Fractional Snow Cover, FSC) is adopted in the snow wave band. The snow cover product MOD10A1 is generated based on a multiband sensor image mounted on a MODIS-series satellite. The automatic mapping of snow coverage is a normalized snow index (NSDI) method calculated using satellite reflectance in MODIS band 4 (0.545-0.565 μm) and band 6 (1.628-1.652 μm), using a binary scale snow product ("Snow Cover Daily Tile") in combination with threshold test and MODIS cloud mask data generation. MOD10A2 is an 8-day synthesized data product obtained by further integration processing, partial cloud coverage is filtered out by utilizing the motion characteristics of the cloud, and the accuracy and quality are higher, but partial product data cannot be produced and retrieved.
The spatial resolution of the DEM digital elevation proposal is not lower than 90 m, global SRTM data with 30 m spatial resolution can be used at present, the SRTM data products are not updated daily like MODIS data products, and the online query and download are needed when in use, and all the SRTM data can be downloaded in advance to be directly queried and searched when in use on a local hard disk. And cutting and re-projecting the digital elevation of the DEM according to the geographic range of the input multispectral image to obtain a single-band DEM elevation image with the same projection, resolution and pixel size as the input image.
The processing steps are as follows: and (3) classifying the elevation, namely classifying the earth surface into three categories of a water body area, a flat area and a mountain area according to the elevation image of the DEM. The method aims to improve the overall quality marking precision by adopting different quality marking algorithms according to different earth surface categories. It should be noted that the above three categories are a simple rough classification, as opposed to a precise remote sensing classification. The three categories of partitioning serve only the intrinsic quantity marking algorithm flow and are not embodied in the final quality marking data product.
The classification of the water body is determined according to the data products of the DEM elevation image, for example, -32768 filling values in the SRTM data are marked as the water body, the water body only comprises large-area deep sea and lakes, generally SAR cannot measure the terrain area, but does not comprise small-area lakes and rivers, and the method is simple and applicable to threshold value method cloud and shadow detection. It should be noted that the final quality-marking data product contains water body marking using other reference data, and is completed by combining a special fusion algorithm with a water body detection algorithm, the water body marking reference data using global surface water resource amount data (Global Surface Water Occurrence, GSWO), which can also be downloaded through an automatic network.
The flat area is distinguished from the mountain area by a Flood Fill algorithm (Flood Fill). The flood fill algorithm is a classical image segmentation algorithm whose idea is named like the diffusion of a flood from one region to all reachable regions. In GNU Go and mine sweeping, a flood fill algorithm is used to calculate the area that needs to be cleared. The flooding fill algorithm may be constructed in a number of ways, with many algorithms implementing code resources, most algorithms using either explicitly or implicitly a queue or stack data structure. The flood filling algorithm has two main parameters: a starting point and a target height. The method comprises the steps of determining a starting point and a target height through statistics of gradient and height of an elevation image of the DEM, wherein the gradient is calculated by height difference of a sliding window and pixel size of the sliding window, a fixed gradient threshold is given as a boundary for distinguishing flatness and steepness, a starting point maximum value can be determined by a statistical histogram of all pixel values below the threshold, and a target height maximum value can be determined by a statistical histogram of all pixel values above the threshold.
The three categories adopt different quality marks in the quality mark mask image, the three categories in the Byte type quality mark mask image with the numerical value ranging from 0 to 255 are marked in ten positions, such as a water body mark 10, a flat area mark 20, a mountain area mark 30, the other categories are marked in units, such as a filling value of 0, a clear surface of 1, a snow mark of 4 and a cloud mark of 5, such as a numerical value of 15 represents a water body area cloud.
The quality marking is respectively carried out according to three classification dividing areas of a water body area, a flat area and a mountain area, which is the core for improving the precision of the algorithm, and the three classification dividing by only using the DEM elevation data is a simple and effective technical means in engineering. Some cloud detection algorithms which appear before divide land and water through a simple spectrum threshold, such as an SFM algorithm designed for a high-resolution first satellite, but because precision is unstable due to factors of multi-spectrum image wave band setting of domestic satellites, a lot of comprehensive multi-source data appear later, a complex processing step is designed to construct a reference data set, the quantitative precision is high, but the cloud detection algorithm is difficult to be suitable for an engineering global data automatic processing flow, meanwhile, experiments show that the precision requirement of a quality marking algorithm taking cloud detection as a core is higher than precision results of classification, and more features such as integral brightness, texture, topography and the like of an underlying land surface are distinguished from cloud threshold and shadow features, and the algorithm can generally obtain better precision by using DEM elevation classification.
The processing steps are as follows: cloud detection, namely, for three categories of a water body area, a flat area and a mountain area, respectively utilizing different fixed thresholds to detect thick clouds with high confidence, correcting the edges of the cloud areas by utilizing large-scale guide filtering for the water body area, and marking the clouds in a quality marking mask.
The input multispectral image is converted from raw DN values to apparent reflectivity. The fixed threshold method of cloud detection is to set a threshold for each band, and when the pixel values of all bands of a certain pixel position exceed the threshold, the pixels of the position are marked as a cloud. Different thresholds are adopted for three categories of the water body area, the flat area and the mountain area to detect thick clouds with high confidence, and specific thresholds are determined according to satellite data and experimental accumulation and are set as algorithm configuration items which can be manually modified at any time. In order to obtain reasonable accuracy for data in different time and geographic ranges, a strict threshold is usually set by a fixed threshold, so that a detection result is mainly thick cloud.
Cloud detection uses HOT index and VBR index simultaneously, and the detection result is mainly thick cloud.
The HOT index calculation formula is:
HOT＝B1-0.5×B3
the VBR index calculation formula is:
Figure BDA0003967739000000131
wherein B1, B2 and B3 are respectively blue band, green band and red band data. The cloud detection index threshold value also adopts different threshold values for three categories of a water body area, a flat area and a mountain area, for example, the flat area threshold value is HOT >0.2& VBR >0.7, and the mountain area threshold value and the water body area threshold value are in sequence. The cloud is marked in the quality marking mask, the water body cloud area is marked as 15, and the surface cloud area is marked as 25 or 35.
The processing steps are as follows: and (3) correcting a dynamic threshold cloud area, wherein a flat area and a mountain area are used as an integral area, cloud accuracy detected through a fixed threshold in a quality marking mask in the area is not high, and a large number of false detection exists, so that in order to more finely find a cloud detection threshold suitable for current data, low-resolution surface reflectivity data close to an imaging date are used as reference data, and a more accurate cloud detection threshold is found. Firstly, constructing a linear function of each wave band between the apparent reflectivity of an input image and the clear earth surface of low-resolution earth surface reflectivity data in reference data, then calculating the maximum value of the apparent reflectivity of the clear earth surface according to whether the apparent reflectivity of cloud area pixels in a quality mark mask is larger than the maximum value of the apparent reflectivity of the linear function in each wave band, judging that the pixels are corrected clouds, and finally correcting the edges of the cloud area by utilizing large-scale guide filtering.
Constructing a linear function yi=ax of each wave band between the apparent reflectivity of the input image and the clear surface of the low-resolution surface reflectivity data in the reference data i +b, where x i Corresponding to the low-resolution earth surface reflectivity, y i The apparent reflectivity of the corresponding input image, i=1, 2 and 3 … correspond to the band numbers, a and b are linear change coefficients to be solved, and each band is different and needs to be calculated respectively. If the reference data point set { x } is determined i And the corresponding input image point set { y }, respectively i And the linear function relation between the apparent reflectivity and the surface reflectivity can be fitted by using a least square method, so that a dynamic threshold cloud detection model of the visible light and near infrared 4 channels is constructed.
Data point set { x for fitting i ，y i The low resolution surface reflectivity pixel corresponds to a pixel area of the clear surface of the flat area and the mountain area of the high resolution input image, and the pixel area is the condition of approximate homogeneous surface. Point set { x i ，y i Specific methods of obtaining are exemplified by the 16 m resolution of homemade high resolution one-number (GF-1) and the 500 m resolution of MODIS MOD09A 1. Firstly, selecting a possible point set in the MOD09A1, selecting all pixel points contained in a flat area and a mountain area by using a quality marking mask, filtering cloud pixels and low-quality pixels according to quality marking wave bands contained in the MOD09A1, and filtering out points with large pixel data difference, generally boundary points and the like according to eight neighborhood pixel value differences, wherein the rest is the possible point set in the MOD09A 1. Then, for all possible point sets in MOD09A1, pixel points with the resolution of 500 meters are respectively corresponding to a plurality of pixels of a rectangular frame with the resolution of 16 meters of GF-1 according to geographic coordinates, and the statistical moment is calculatedIf all pixels of the frame do not include cloud pixels and the pixel difference value is lower than the threshold value, the point is selected into the point set { x } i Yi }, and taking the rectangular box pixel median as y i Values. The GF-1 multispectral image comprises four wave bands, and the selection point set only uses a near infrared fourth wave band and corresponds to the wave band of the similar center wave band in the MOD09A 1.
Fitting a set of points { x } using least squares i ，y i If a band linear function fit failure occurs, or a point set { x } i ，y i And if the mean square error of the fitted linear function is larger than the threshold value, discarding dynamic threshold value correction, and directly using the detection result of the fixed threshold value to conduct guide filtering correction on the cloud zone edge. If the fitting is successful, calculating the apparent reflectance cloud threshold of GF-1 by using the fitted linear function and the earth surface reflectance wave band of MOD09A1, wherein the specific steps are that firstly, all pixels (including cloud pixels) of the earth surface reflectance wave band of MOD09A1 are calculated to the apparent reflectance by using the fitted linear function, and then, the threshold of earth surface and cloud in the obtained apparent reflectance is counted, wherein the threshold is used as a dynamic threshold of cloud detection correction. The cloud detection dynamic threshold is calculated by using all four wave bands contained in the GF-1 multispectral image, and the four wave bands correspond to four wave bands with similar center wave band wavelengths in MOD09A 1.
And correcting the quality mark mask by using the dynamic threshold of cloud detection, and if the apparent reflectivity of the cloud area pixels in the quality mark mask is smaller than the maximum value of the apparent reflectivity of the clear ground surface calculated by the linear function in each wave band, correcting the pixels into the clear ground surface. And the corrected result is used for correcting the cloud zone edge by using large-scale guide filtering, specifically, a quality mark mask image is used as input, a blue light wave band is used as a guide image, a thin cloud which is missed in the cloud zone edge is added into the cloud mask by using a guide filtering algorithm, and the large-scale is that the size of a filtering radius pixel is larger than 500 pixels, so that the correction of the large-area cloud zone missed is adapted. In the guide filtering algorithm, an average value wave band formed by three wave bands of visible light is taken as a guide image L (x, y), a cloud mask is taken as a filtering input image V (x, y), and an output image is recorded as
Figure BDA0003967739000000151
The filtering result at pixel (x, y) is expressed as a weighted average:
Figure BDA0003967739000000152
assuming that the steering filter is at the steering image L (x, y) and the filtered output
Figure BDA0003967739000000153
Between which is a local linear model:
Figure BDA0003967739000000154
/>
by minimizing the window ω below k Is a cost function of (1):
Figure BDA0003967739000000155
obtaining local linear coefficient a k ,b k Is a value of (2).
The main calculation amount of the guide filter is a median filter, and in order to adapt to a transition area between a large-area thick cloud area and a ground surface area, namely a cloud area boundary or a thin cloud area, a large filtering radius of 500 pixels is used. As the filter radius increases, the operation efficiency of the general median filter increases sharply, and a Boxfilter rapid filtering algorithm which is irrelevant to the radius is adopted. The operations of summation, mean, variance, etc. with complexity O (MN) are reduced to a complexity that approximates O (1). Guiding filtered output image
Figure BDA0003967739000000156
And (5) masking the corrected cloud area.
Fig. 2 is a schematic diagram of edge correction of large-scale guided filtering in cloud detection, showing that the obtained cloud mask (fig. 2. D) conforms more to the boundary of the actual cloud area in the original image (fig. 2. A) than before (fig. 2. B) by guiding the filtered image (fig. 2. C).
The processing steps are as follows: the snow mountain detection is very easy to be detected as cloud by mistake due to the fact that the domestic satellite multispectral data lack of effective wave bands for distinguishing the cloud from the snow, and the snow mountain is used as a coverage category of the snow and can be distinguished in a snow mountain simulation mode due to the fact that the snow mountain is connected with the terrain elevation. And judging whether a snow mountain exists in the mountain area by using low-resolution snow coverage data in the reference data and the DEM elevation image, if so, performing snow mountain simulation to obtain a simulated snow mountain mask, and correcting a false detection cloud area of the mountain area in the quality mark mask into snow by using the simulated snow mountain mask.
And judging whether the snow mountain exists or not by utilizing the low-resolution snow coverage data in the reference data and the DEM elevation image, and if the mountain area in the DEM elevation image comprises low-resolution snow coverage pixels and the number of snow mountain pixels exceeds 2 percent, performing snow mountain simulation.
The snow mountain simulation step is to firstly count the average elevation value of the mountain area pixel values of the DEM image corresponding to all the snow pixels in the snow mask as a snow line value, wherein the pixels larger than the snow line value in the DEM image are marked as snow mountain in the simulated snow mountain mask, and the simulated snow mountain mask is a Byte type single-band image with the same projection, resolution and pixel size as the input image.
And correcting the false detection cloud area of the mountain area in the quality mark mask into snow by using the simulated snow mountain mask, wherein morphological matching is carried out on all cloud pixel communication areas in the mountain area and the simulated snow mountain pixel communication areas in the simulated snow mountain mask, and if the similarity matching exceeds a threshold value, the cloud pixel communication areas are corrected into snow.
The processing steps are as follows: shadow detection is matched with cloud shadows, the cloud shadows generally contain no or only weak surface information, and a quality marking algorithm correctly recognizes that the cloud shadows are meaningful for marking subsequent applications of data products.
Shadow detection is to mark all possible shadow areas in flat areas and mountain areas by combining threshold values with an image processing algorithm, including cloud shadows and other dark ground surfaces. Shadow pixels with high confidence are first detected using a fixed threshold, and then the shadow edges are corrected using guided filtering. Fixed threshold detection shadow pixels use the near infrared band threshold, NDVI index, and VBR index to set a fixed threshold. The NDVI index calculation formula is:
Figure BDA0003967739000000161
wherein B3 and B4 are respectively red band data and near infrared band data. For example, the threshold for coarse detection of shadows is NDVI <0.15& & VBR >0.7& B4<0.15.
When the guiding filtering corrects the shadow edge, the shadow edge correction is different from the cloud area edge correction, and the missed detection of a large-radius area cannot occur, so that the maximum radius setting 50 pixels of the guiding filtering can meet the requirement. In addition, a flood fill method (flood fill) is also a common algorithm for correcting edges of a shadow area, but the algorithm is not limited by a range and expands before a seed point reaches the highest water level, so that guide filtering correction edge omission is weighed and selected under the condition that the shadow threshold detection precision is normal.
Cloud shadow matching is to screen out shadows under the cloud from cloud shadow detection results by using a morphological image processing algorithm. The imaging geometrical relationship is utilized to calculate the azimuth angle of the shadow under the cloud relative to the cloud, the solar zenith angle, the solar azimuth angle, the observed zenith angle and the observed azimuth angle can be obtained through recording in the satellite data auxiliary file, and the azimuth angle formula of the shadow under the cloud relative to the cloud area is calculated as follows:
Figure BDA0003967739000000171
wherein alpha, beta, omega and mu are sun zenith angle, sun azimuth angle, observation zenith angle and observation azimuth angle respectively.
The shadow under the cloud is determined empirically with respect to the cloud distance range, for example, the cloud shadow distance range of a GF-1 satellite 16 meter resolution multispectral image is set to 0-300 pixels. For each cloud pixel communication area, moving the shadow under cloud by 0-300 pixels in the azimuth direction relative to the cloud, moving one pixel at a time, calculating the number of current matched pixels, and marking the shadow communication area with the maximum matching number as the shadow under cloud. Fig. 3 is a schematic diagram of cloud-to-cloud shadow region pixel matching. The actual operation algorithm is optimized for improving the calculation efficiency of cloud shadow matching, and comprises resampling a quality mark mask to be one half of the original image, matching only a communication area with the number of cloud pixels being more than 50, matching 4 pixels in a single step length, and the like.
For a large-area water body area, the algorithm does not detect shadows under the cloud. Considering that the overall brightness of the water body is relatively low compared with the earth surface, the shadow detection precision based on the threshold value is unstable, cloud-to-cloud shadow matching is easily affected by the texture of the water body, and the application scene of the water body shadow in the quality mark data is limited, so that the water body region in the quality mark mask does not contain the cloud shadow mark.
The processing steps are as follows: the method is characterized by mainly aiming at data of a middle-high latitude winter snowfall area, judging whether large-area snowfall exists in a flat area or not, wherein the judging method is to utilize low-resolution snow coverage data in reference data to judge whether the large-area snowfall exists in the flat area or not, and the judging criterion is that the proportion of snow pixels in an image after cloud pixels are removed is larger than that of a clear ground surface. If the flat area has large-area snowfall, the cloud pixel communication area which is not matched with the shadow is corrected to snow, and the cloud pixel communication area which is matched with the shadow under the cloud is regarded as the cloud above the snowfall area.
The processing steps are as follows: the mask integration is to integrate the quality mark mask data to meet the specification of quality mark data products, and after integrating three classification areas of a water body area, a flat area and a mountain area, the final quality mark mask comprises cloud, snow, clear surface, water body and filling value categories.
Outputting data: the quality marking mask is Byte type single-band compressed tiff format image data consistent with the pixel size of an input image, and comprises cloud, snow, clear surface, water body and filling value categories, wherein different categories are marked with different numerical values, the filling value is marked with 0, the clear surface is marked with 1, the water body is marked with 2, the snow is marked with 4, and the cloud is marked with 5.
It should be noted that, the core of the domestic multispectral satellite quality marking algorithm is to solve the problem of distinguishing the dynamic threshold value from cloud snow, the algorithm belongs to a part of the production algorithm flow of domestic multispectral satellite standard data products, and the quality marking mask outputted by the algorithm is stored in a hard disk as a component of the domestic multispectral satellite standard data products. FIG. 4 is an example graph of algorithmically processed high-resolution satellite data, specifically a 16-meter resolution WFV multispectral image, with quality-marking masks produced using a color table to display differences for distinguishing between different classes.
The C++ algorithm example is realized on a PC platform, and the effectiveness and the robustness of the algorithm are verified through experimental data in the early stage.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present application, and are not limiting thereof; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the corresponding technical solutions.

Claims (4)

1. A domestic satellite multispectral image pixel-by-pixel quality marking method is characterized by comprising the following steps:
preparing reference data, namely preparing a snow cover product and a DEM elevation image of a low-resolution satellite standard earth surface reflectivity product which cover a geographic range and are close to an imaging date according to an input multispectral image, and performing cutting and re-projection to obtain reference data with the same projection, resolution and pixel size as the input image;
dividing the surface into three categories of a water body area, a flat area and a mountain area by using the DEM elevation image, and adopting different quality marks for the three categories to generate a single-band quality mark mask image;
cloud detection, namely, for three categories of a water body area, a flat area and a mountain area, respectively utilizing different fixed thresholds to detect thick clouds with high confidence, and marking the clouds in a quality marking mask image;
dynamic threshold cloud region correction, for a flat region and a mountain region, constructing a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data, if the apparent reflectivity of a cloud pixel in a quality mark mask is smaller than the maximum value of clear surface apparent reflectivity calculated by the linear function in each wave band, correcting the cloud pixel into a clear surface, and correcting the edges of the cloud region by utilizing large-scale guide filtering for the rest cloud pixels;
the method comprises the steps of detecting snow mountains, judging whether the snow mountains exist or not by using low-resolution snow coverage data in reference data and DEM elevation images for mountain areas, if so, performing snow mountain simulation to obtain a simulated snow mountain mask, and correcting false detection cloud areas of the mountain areas in a quality mark mask into snow by using the simulated snow mountain mask;
shadow detection and cloud shadow matching, for flat areas and mountain areas, shadow pixels with high confidence coefficient are detected by using a fixed threshold value, shadow edges are corrected by using guide filtering, azimuth angles and distance ranges of cloud shadows and cloud shadows are calculated by using imaging geometric relations, cloud shadow correspondence is determined by using maximum pixel number matching of cloud shadows and cloud shadow communication areas, and shadow communication area pixels without matching relations are filtered;
detecting snowfall, namely correcting a cloud pixel communication area which is not matched with shadows in a quality mark mask into snow by utilizing low-resolution snow coverage data in reference data for a middle-high latitude winter snowfall area in a flat area;
and integrating the masks to obtain quality mark mask images containing cloud, snow, clear surface, water and filling value categories.
2. The method for marking the pixel-by-pixel quality of the domestic satellite multispectral image according to claim 1, wherein the step of using the DEM elevation image to divide the earth surface into three categories of a water body area, a flat area and a mountain area comprises the steps of:
the flat area and the mountain area are distinguished by combining slope and height statistics through a flood filling algorithm; different quality marks are adopted for the three types, three types of the Byte type quality mark mask images in the numerical range of 0 to 255 are marked in ten positions, and other types of the Byte type quality mark mask images are marked in units.
3. The method for marking the pixel-by-pixel quality of the domestic satellite multispectral image according to claim 1, wherein for a flat area and a mountain area, a linear function between each wave band of low-resolution surface reflectivity data in input image apparent reflectivity and reference data is constructed, a linear function relation between the apparent reflectivity and surface reflectivity is fitted by a least square method, a dynamic threshold cloud detection model of visible light and near infrared 4 channels is constructed, a fitted data point set is used for meeting the condition that low-resolution surface reflectivity pixels correspond to a pixel area of a clear surface of the flat area and the mountain area of the input image with high resolution, and the pixel area is approximately homogeneous surface; if the apparent reflectivity of a cloud pixel in the mass label mask is less than the maximum value of the apparent reflectivity of the linear function calculation clear surface at each band, the pixel is corrected to the clear surface.
4. The method for marking quality of domestic satellite multispectral image pixel by pixel according to claim 2, wherein the step of correcting the cloud pixel communication area which is not matched with the shadow in the quality marking mask into snow by using the low-resolution snow coverage data in the reference data is to firstly judge whether the condition that the low-resolution snow coverage data snow pixel corresponds to the cloud pixel in the flat area of the quality marking mask exists, and if the condition exists and the communication area where the cloud pixel is not matched with the shadow, correcting the cloud pixel communication area into snow.
CN202211504671.2A 2022-11-28 2022-11-28 Domestic satellite multispectral image pixel-by-pixel quality marking method Pending CN116229459A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211504671.2A CN116229459A (en) 2022-11-28 2022-11-28 Domestic satellite multispectral image pixel-by-pixel quality marking method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211504671.2A CN116229459A (en) 2022-11-28 2022-11-28 Domestic satellite multispectral image pixel-by-pixel quality marking method

Publications (1)

Publication Number Publication Date
CN116229459A true CN116229459A (en) 2023-06-06

Family

ID=86586100

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211504671.2A Pending CN116229459A (en) 2022-11-28 2022-11-28 Domestic satellite multispectral image pixel-by-pixel quality marking method

Country Status (1)

Country Link
CN (1) CN116229459A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117218452A (en) * 2023-11-02 2023-12-12 临沂市兰山区自然资源开发服务中心 Automatic classification management system for land images

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117218452A (en) * 2023-11-02 2023-12-12 临沂市兰山区自然资源开发服务中心 Automatic classification management system for land images
CN117218452B (en) * 2023-11-02 2024-02-06 临沂市兰山区自然资源开发服务中心 Automatic classification management system for land images

Similar Documents

Publication Publication Date Title
CN109581372B (en) Ecological environment remote sensing monitoring method
CN109993237B (en) Water body rapid extraction method and system based on high-resolution satellite optical remote sensing data
CN108596103B (en) High-resolution remote sensing image building extraction method based on optimal spectral index selection
CN111832518B (en) Space-time fusion-based TSA remote sensing image land utilization method
CN111242224B (en) Multi-source remote sensing data classification method based on unmanned aerial vehicle extraction classification sample points
Wu et al. Mapping invasive plant with UAV-derived 3D mesh model in mountain area—A case study in Shenzhen Coast, China
Huo et al. Supervised spatial classification of multispectral LiDAR data in urban areas
CN113537018A (en) Water and soil conservation monitoring method based on multi-temporal satellite remote sensing and unmanned aerial vehicle technology
CN111337434A (en) Mining area reclamation vegetation biomass estimation method and system
CN110988909A (en) TLS-based vegetation coverage determination method for sandy land vegetation in alpine and fragile areas
CN115271217A (en) Wheat yield prediction method based on multi-source remote sensing data of unmanned aerial vehicle
Hu et al. Seasonal change of land-use/land-cover (LULC) detection using MODIS data in rapid urbanization regions: A case study of the pearl river delta region (China)
CN114022783A (en) Satellite image-based water and soil conservation ecological function remote sensing monitoring method and device
CN116229459A (en) Domestic satellite multispectral image pixel-by-pixel quality marking method
Li et al. A new method for surface water extraction using multi-temporal Landsat 8 images based on maximum entropy model
CN117409168B (en) Flood forecasting and flood simulation method and system for real-time dynamic rendering
Bektas Balcik et al. Determination of magnitude and direction of land use/land cover changes in Terkos Water Basin, Istanbul
CN116246272A (en) Cloud and snow distinguishing method for domestic satellite multispectral image quality marks
CN115546658B (en) Night cloud detection method combining quality improvement and CNN improvement of data set
CN115294183A (en) Disc-shaped sub-lake water body time sequence extraction method based on multi-source remote sensing data
Bryceson Digitally processed satellite data as a tool in detecting potential Australian plague locust outbreak areas
CN114792322A (en) Method for detecting cloud and cloud shadow of mountain domestic high-resolution satellite image
CN114241333A (en) Method for accurately identifying new landslide region based on multi-source time sequence remote sensing image
CN115205682B (en) NDVI maximum value remote sensing data product seamless production processing method
Alexandridis et al. LAI measurement with hemispherical photographs at variable conditions for assessment of remotely sensed estimations

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination