Rapid warehousing method and system based on remote sensing images

Abstract

The invention discloses a rapid warehousing method and system based on a remote sensing image, and relates to the technical field of remote sensing image processing, and the method comprises the steps: accurately distinguishing effective signals and interference signals of ground features through synchronously collecting multispectral data of an atmospheric scattering characteristic peak frequency band and a cloud shadow absorption characteristic peak frequency band and combining with ground feature dielectric signal preliminary screening processing; then relying on a dual-band dielectric difference-multispectral characteristic peak intensity coupling correction model, model parameters are dynamically adjusted in combination with a scene interference type identification library, various common interferences such as atmospheric interference, terrain shadow and sensor noise can be specifically eliminated, the precision deviation caused by a traditional single correction mode is avoided, and the accuracy of the system is improved. It is ensured that finally stored remote sensing image data can truly reflect ground feature features; meanwhile, manual intervention is not needed in the whole process, efficiency loss and error risks caused by manual operation are reduced, and data storage and subsequent query speed is further increased through block transmission and index construction during storage.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing image processing, and particularly relates to a remote sensing image rapid storage method and system. BACKGROUND

[0002] With the rapid development of remote sensing technology, the imaging capability of satellite, unmanned aerial vehicle and other remote sensing devices is continuously improved, and the amount of acquired remote sensing image data is explosively increasing. Remote sensing images are widely applied in the fields of land surveying and mapping, ecological monitoring, city planning, disaster emergency response and the like. As a core source of geographic information data, the data quality of remote sensing images directly determines the accuracy of subsequent application analysis results, and therefore, the images need to be stored after professional processing to eliminate external interference.

[0003] In the process of remote sensing image data acquisition and transmission, the images are easily superimposed with various interferences due to factors such as atmospheric environment, terrain condition and device performance. Atmospheric scattering may cause spectral information deviation, terrain shadow may cause local feature loss, and sensor noise may damage the integrity of image data. In order to guarantee the image quality, the traditional technology usually adopts a single correction algorithm (such as a single atmospheric correction algorithm or a cloud shadow removal algorithm) to process the interference. However, in actual scenes, various interferences often exist simultaneously and interact with each other, and the single correction method cannot comprehensively solve the coordinated interference problems of spectral deviation, feature loss and data noise, and it is difficult to meet the quality requirements of image data for high-precision applications. In order to solve the above technical problems, the present application provides a remote sensing image rapid storage method and system. SUMMARY

[0004] The present application aims to provide a remote sensing image rapid storage method and system, which synchronously collects multi-spectral data of double characteristic frequency bands, establishes a multi-spectral characteristic peak coupling correction model, and realizes synchronous elimination of atmospheric scattering and cloud shadow interference, so as to solve the problems proposed in the above background technology.

[0005] In a first aspect, the embodiments of the present application provide a remote sensing image rapid storage method, which comprises the following steps:

[0006] S1: acquiring original remote sensing image data to be stored;

[0007] S2: performing importance classification on an image area corresponding to the original remote sensing image data, marking the image area as different priorities according to a preset rule, and synchronously processing the image areas of different priorities through a GPU parallel framework;

[0008] S3: performing multi-spectral characteristic collection on the processed remote sensing image data, and synchronously acquiring multi-spectral data of the remote sensing image data in a first characteristic frequency band and a second characteristic frequency band;

[0009] S4: performing dielectric signal of ground object preliminary screening processing based on the multispectral data of the first characteristic frequency band and the second characteristic frequency band, to obtain a dielectric signal difference value of ground object;

[0010] S5: establishing a coupling correction model in combination with the dielectric signal difference value of ground object;

[0011] S6: constructing a scene interference type identification library, the identification library containing typical interference characteristic spectra of atmospheric interference, terrain shadow, and sensor noise;

[0012] S7: judging the main interference type of the current image through the scene interference type identification library, dynamically adjusting the model parameters of the coupling correction model; inputting the multispectral data and the dielectric signal difference value of ground object into the adjusted coupling correction model, and outputting high-precision remote sensing image data after interference elimination by the adjusted coupling correction model;

[0013] S8: performing a database operation on the high-precision remote sensing image data.

[0014] In some embodiments, the first characteristic frequency band is an atmospheric scattering characteristic peak frequency band, and the second characteristic frequency band is a cloud shadow absorption characteristic peak frequency band.

[0015] The obtaining of the multispectral data in S3 specifically includes:

[0016] The preset dual-band spectral acquisition parameters are obtained, the dual-band spectral acquisition parameters including a first acquisition wavelength range, a first sampling frequency of the first characteristic frequency band, and a second acquisition wavelength range, a second sampling frequency of the second characteristic frequency band, the dual-band spectral acquisition parameters being pre-stored in a storage unit of the processing device;

[0017] According to the dual-band spectral acquisition parameters, the multispectral acquisition component is controlled to simultaneously perform spectral data acquisition on the processed remote sensing image data through the dual-channel data acquisition circuit;

[0018] The collected spectral data are sequentially subjected to photoelectric conversion and analog-to-digital conversion to obtain first spectral data and second spectral data, which together constitute the multispectral data;

[0019] The multispectral data are temporarily stored in a cache unit of the processing device, and the acquisition effectiveness is verified by calculating the signal-to-noise ratio;

[0020] When the signal-to-noise ratio is greater than a preset signal-to-noise ratio threshold, it is determined that the acquisition is effective; if it is less than or equal to the preset signal-to-noise ratio threshold, the spectral data acquisition is performed again until the acquisition is effective.

[0021] In some embodiments, the dielectric signal difference value of ground object obtained in S4 specifically includes:

[0022] The first spectral data and the second spectral data are determined from the multispectral data, and are extracted from the cache unit of the processing device through a preset data identifier;

[0023] The first spectral data and the second spectral data are subjected to noise filtering processing respectively, and data denoising is completed by using a preset noise filtering mode to obtain filtered first spectral data and second spectral data;

[0024] The pre-stored dielectric signal feature templates of different ground object types in the corresponding frequency bands are called;

[0025] The similarity is calculated by using a normalized cross-correlation matching mode, the dielectric signal corresponding to the template with the highest similarity is selected as the first ground object dielectric signal and the second ground object dielectric signal respectively, and the difference between the two is calculated to generate a ground object dielectric signal difference value after point-by-point matching calculation.

[0026] In some embodiments, the establishment of the coupling correction model in S5 specifically includes:

[0027] A plurality of sets of sample data are collected, each set of sample data including sample dual-frequency band dielectric difference values and corresponding sample multi-spectral feature peak intensities;

[0028] The sample data are derived from remote sensing images under different geographical, time and weather conditions, and the sample multi-spectral feature peak intensities are extracted by a preset mode;

[0029] The plurality of sets of sample data are preprocessed to remove abnormal sample data;

[0030] The preprocessed sample dual-frequency band dielectric difference values are used as independent variables, and the preprocessed sample multi-spectral feature peak intensities are used as dependent variables, and a partial least squares regression mode is used for training;

[0031] The preprocessed sample data are divided into training data and validation data, the model fitting degree is judged by a coefficient of determination, and the model prediction error is judged by a root mean square error, and after the model fitting degree and the prediction error both meet preset requirements, the model parameters are stored in a model database of a processing device, and finally a coupling correction model is formed.

[0032] In some embodiments, the preprocessing of the plurality of sets of sample data specifically includes:

[0033] The first mean and the first standard deviation of the sample dual-frequency band dielectric difference values in each set of sample data, and the second mean and the second standard deviation of the sample multi-spectral feature peak intensities are calculated;

[0034] According to the first mean, the first standard deviation, the second mean and the second standard deviation, an abnormal data rejection threshold range is set;

[0035] The sample dual-band dielectric difference value needs to satisfy "first mean value-3 times first standard deviation ≤ sample dual-band dielectric difference value ≤ first mean value+3 times first standard deviation", and the sample multi-spectral feature peak intensity needs to satisfy "second mean value-3 times second standard deviation ≤ sample multi-spectral feature peak intensity ≤ second mean value+3 times second standard deviation";

[0036] The sample data exceeding the range is removed, and the valid sample data is retained.

[0037] In some embodiments, before S8 is performed, the method further comprises:

[0038] The high-precision remote sensing image data after interference elimination is subjected to quality detection, and the quality detection comprises spectral accuracy detection and spatial resolution detection.

[0039] If the preset storage standards are all satisfied, S8 is performed.

[0040] If not, return to S2 for reprocessing, and then sequentially perform the subsequent steps.

[0041] In some embodiments, the spectral accuracy detection of the high-precision remote sensing image data after interference elimination specifically comprises:

[0042] A preset standard spectral data set is obtained, the standard spectral data set contains standard spectral curves corresponding to different ground object types, and is stored in a standard database of the processing device.

[0043] A threshold method is used for image segmentation to divide ground object regions, the spectral data of each region is subjected to average processing, and an actual spectral curve is extracted.

[0044] A cosine similarity method is used to calculate the similarity between the actual spectral curve and the corresponding standard spectral curve.

[0045] If the similarity is greater than a preset similarity threshold, it is determined that the spectral accuracy of the high-precision remote sensing image data after interference elimination satisfies the preset storage standard; otherwise, it is determined that the preset storage standard is not satisfied.

[0046] In some embodiments, the spatial resolution detection of the high-precision remote sensing image data after interference elimination specifically comprises:

[0047] A spatial resolution parameter of the high-precision remote sensing image data after interference elimination is obtained.

[0048] The spatial resolution parameter is compared with a spatial resolution requirement in the preset storage standard.

[0049] If the spatial resolution parameter is greater than or equal to the spatial resolution requirement, it is determined that the spatial resolution of the high-precision remote sensing image data after interference elimination satisfies the preset storage standard; otherwise, it is determined that the preset storage standard is not satisfied.

[0050] In some embodiments, the importance classification and processing of the image region corresponding to the original remote sensing image data in S2 specifically include:

[0051] The application scene importance of the image region is taken as the division basis, the city core area, the disaster sensitive area and the ecological protection area are marked as high-priority areas, and the desert, the gobi and the uninhabited area are marked as low-priority areas.

[0052] The GPU parallel framework is used to allocate computing power for different priority areas.

[0053] The image is divided into fragments in a data fragmentation manner according to a preset size, and the data processing is synchronously performed on the fragment areas of different priorities to obtain the processed remote sensing image data.

[0054] In a second aspect, the embodiments of the present application provide a remote sensing image rapid warehousing system based on remote sensing images, which comprises:

[0055] A data acquisition module is configured to acquire original remote sensing image data to be warehoused.

[0056] A data processing module is configured to classify the importance of the image region corresponding to the original remote sensing image data, mark the image region as different priorities according to a preset rule, and synchronously process the image regions of different priorities through a GPU parallel framework.

[0057] A spectrum acquisition module is configured to acquire multi-spectral data of the processed remote sensing image data in a first characteristic frequency band and a second characteristic frequency band.

[0058] A dielectric preliminary screening module is configured to perform dielectric signal preliminary screening processing based on the multi-spectral data in the first characteristic frequency band and the second characteristic frequency band to obtain a dielectric signal difference.

[0059] A model construction module is configured to establish a coupling correction model in combination with the dielectric signal difference.

[0060] An identification library module is configured to construct a scene interference type identification library, and the identification library contains typical interference feature maps of atmospheric interference, terrain shadow and sensor noise.

[0061] An interference elimination module is configured to determine the main interference type of the current image through the scene interference type identification library, dynamically adjust the model parameters of the coupling correction model, input the multi-spectral data and the dielectric signal difference into the adjusted coupling correction model, and output high-precision remote sensing image data after eliminating interference from the adjusted coupling correction model.

[0062] An image warehousing module is configured to perform warehousing operation on the high-precision remote sensing image data.

[0063] The present application has the following advantages:

[0064] 1. By synchronously collecting multi-spectral data of atmospheric scattering characteristic peak frequency band and cloud shadow absorption characteristic peak frequency band, combining with preliminary screening processing of ground feature dielectric signal, the effective signal and interference signal of ground feature can be accurately distinguished; then relying on the "double-band dielectric difference-multiple spectral characteristic peak intensity" coupling correction model, combined with the scene interference type recognition library dynamic adjustment model parameter, the common interference such as atmospheric interference, terrain shadow, sensor noise and other common interference can be targetedly eliminated, the precision deviation caused by traditional single correction method is avoided, and it is ensured that the final remote sensing image data stored in the database can truly reflect the ground feature characteristics;

[0065] 2. On the one hand, by classifying the image area according to the importance of the application scene, and by means of the GPU parallel framework, the computing power is allocated to different priority areas and is processed synchronously, so that the processing efficiency of the original image data is greatly improved, and the time-consuming problem caused by overall processing is avoided; on the other hand, from the effectiveness verification of multi-spectral data acquisition, to the automatic execution of model training and interference elimination, to the quality detection before storage and the automatic return and reprocessing of unqualified data, the whole process does not need manual intervention, reduces the efficiency loss and error risk caused by manual operation, and at the same time, through block transmission and index construction during storage, the data storage and subsequent query speed are further accelerated, meeting the demand of rapid storage of massive remote sensing images;

[0066] 3. In the whole storage process, a multi-link verification and feedback mechanism is set: the data effectiveness is ensured by verifying the signal-to-noise ratio during multi-spectral data acquisition, the model reliability is verified by the determination coefficient and root mean square error during model training, and the qualified data is screened by spectral accuracy and spatial resolution detection before storage, if the detection is unqualified, it is automatically returned to the previous processing step for re-execution, effectively avoiding the storage of low-precision, damaged or invalid data; at the same time, the database sets hierarchical access permission and data index, which not only guarantees the security of the stored data and prevents data leakage or tampering caused by unauthorized operation, but also quickly locates the data through the index, providing protection for remote sensing data application. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 A flowchart of a remote sensing image rapid storage method provided by an embodiment of the present application is shown;

[0068] Figure 2 A structural diagram of a remote sensing image rapid storage system provided by an embodiment of the present application is shown;

[0069] Figure 3 A flowchart of high-precision remote sensing image quality detection of a remote sensing image rapid storage method provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0070] The application will be further described in detail below with reference to the accompanying drawings. It is necessary to point out here that the following specific embodiments are only used to further illustrate the application and cannot be understood as limiting the scope of protection of the application. Those skilled in the art can make some non-essential improvements and adjustments to the application according to the above application content.

[0071] It should be noted that remote sensing images are electromagnetic wave information carriers of ground objects obtained by imaging devices carried by remote sensing platforms such as satellites and unmanned aerial vehicles, and can intuitively reflect features such as land cover, terrain and ground object attributes, and are widely used in fields such as land surveying and mapping, ecological monitoring, disaster emergency response, and urban planning. In practical applications, the amount of remote sensing image data usually reaches GB or even TB, and is affected by atmospheric scattering, terrain shadow, sensor noise and other interference. The original image needs to be processed and corrected before it can meet the accuracy requirements of the database. The remote sensing image rapid database entry method provided in the embodiments of the application improves the data processing efficiency through regional priority grading and GPU parallel processing, eliminates interference by combining dual-band multispectral acquisition and coupling correction model, and finally realizes the rapid database entry of high-precision remote sensing images, solving the pain points of low processing efficiency, insufficient accuracy and difficult interference elimination in the traditional database entry process.

[0072] The specific embodiments of the application will be described below through examples:

[0073] As shown in Figure 1 A remote sensing image rapid database entry method is applied to a remote sensing data processing device. The device can be equipped with an NVIDIA A100 GPU chip, 1 TB cache unit and 2 TB storage unit, and supports dual-channel spectral data parallel acquisition. The method comprises the following steps:

[0074] S1: Obtain original remote sensing image data to be entered into the database.

[0075] Specifically, the original remote sensing image data is obtained from a low-orbit remote sensing satellite. The multispectral sensor carried by the satellite takes pictures of the target area at a preset period (such as 1 day / time), and the generated original remote sensing image data is transmitted to the ground receiving station through the satellite-ground communication link (such as Ka-band satellite communication). The ground receiving station performs preliminary demodulation and format conversion (converts the satellite original data format to the general TIFF image format) on the transmission data, and then stores it to the ground data server. In the embodiments of the application, a microprocessor (such as an embedded processor carrying an ARM Cortex-A9 architecture) obtains the original remote sensing image data to be entered into the database from the ground data server through Ethernet. CRC32 check algorithm is used to verify the data integrity during the acquisition process. If the verification fails, the data acquisition request is reinitiated to ensure that the original data is not lost or damaged.

[0076] S2: importance classification is performed on the image area corresponding to the original remote sensing image data, different priorities are marked for the image area according to a preset rule, and the image areas of different priorities are synchronously processed through a GPU parallel framework.

[0077] Specifically, the method of importance classification and processing of the image area in S2 includes:

[0078] S21: The application scene importance of the image area is taken as the division basis, and the priority is marked by a region identification module in the processing device.

[0079] Specifically, the region identification module is built-in with an "application scene-priority" mapping rule library: the urban core area (such as the area with a population density ≥1000 people per square kilometer and a building coverage rate ≥60%), the disaster sensitive area (such as the earthquake activity belt and the 5-kilometer range around the river flood embankment), and the ecological protection area (such as the national nature reserve and the core area of the wetland park) are marked as high-priority areas; the desert (vegetation coverage rate ≤5%), the gobi (rock exposure rate ≥80%), and the uninhabited area (population density ≤1 person per square kilometer and no infrastructure) are marked as low-priority areas. The region identification module automatically completes the priority marking of the region by reading the geographic coordinate information (such as latitude and longitude, UTM projection coordinates) of the original remote sensing image, combining the data such as the administrative boundary and the ecological protection red line in the GIS vector database, and generating a "region-priority" table for storage in the cache unit.

[0080] S22: The GPU parallel framework allocates computing power to different priority areas and synchronously processes the fragmented image areas.

[0081] Specifically, the GPU parallel framework adopts the NVIDIA CUDA 12.0 architecture, and the computing power allocation rule is: the high-priority area computing power proportion is set to 70%, and the low-priority area computing power proportion is set to 30%; if the area proportion of the high-priority area exceeds 60% (calculated by the region area statistics module), it is dynamically adjusted to 90% for the high-priority area and 10% for the low-priority area, to ensure that the high-priority area is processed first. The data fragmentation adopts a "fixed size + edge completion" strategy: the original remote sensing image is fragmented according to the size of 512x512 pixels, and if the remaining pixels at the edge of the image are less than 512x512, it is filled to the standard size by zero padding; after fragmentation, the GPU parallel framework allocates independent computing threads to each fragment, synchronously executes "radiometric calibration and geometric correction" processing on different priority fragments, radiometric calibration eliminates the brightness deviation caused by sensor response differences, and geometric correction corrects the image geometric distortion based on ground control points (such as GPS measured control points), and finally obtains the processed remote sensing image data.

[0082] S3: Collecting multi-spectral features of the processed remote sensing image data, and synchronously acquiring multi-spectral data thereof in the first characteristic frequency band and the second characteristic frequency band.

[0083] Specifically, the first characteristic frequency band is an atmospheric scattering characteristic peak frequency band, and the wavelength range is 490-510 nm. The atmospheric scattering coefficient of this frequency band is the highest, and the atmospheric interference intensity can be accurately reflected. The second characteristic frequency band is a cloud shadow absorption characteristic peak frequency band, and the wavelength range is 1550-1650 nm. The absorption of electromagnetic waves by cloud shadow in this frequency band is the most significant, and the cloud shadow interference can be effectively identified.

[0084] The method for acquiring multi-spectral data in S3 comprises:

[0085] S31: Acquiring preset dual-frequency band spectral acquisition parameters.

[0086] Specifically, the dual-frequency band spectral acquisition parameters are pre-stored in the storage unit of the processing device. The parameters include: the first acquisition wavelength range (490-510 nm) of the first characteristic frequency band, the first sampling frequency (100 Hz, i.e. 100 groups of spectral data are collected per second); the second acquisition wavelength range (1550-1650 nm) of the second characteristic frequency band, and the second sampling frequency (100 Hz). After the spectral control module of the processing device reads the parameters, it sends a parameter configuration instruction to the multi-spectral acquisition component (such as a HyperSIS hyperspectral camera), to ensure that the wavelength and frequency of the acquisition component match the preset requirements.

[0087] S32: Controlling the multi-spectral acquisition component to synchronously collect spectral data through a dual-channel data acquisition circuit.

[0088] Specifically, the dual-channel data acquisition circuit includes two independent signal acquisition channels, corresponding to the first and second characteristic frequency bands respectively: the first channel is connected to the 490-510 nm waveband detector of the multi-spectral acquisition component, and the second channel is connected to the 1550-1650 nm waveband detector; after the spectral control module sends a "collection start" instruction, the two channels simultaneously collect spectral data of the processed remote sensing image data. The collected analog spectral signals successively pass through a photoelectric conversion module (which converts optical signals into current signals) and an analog-digital conversion module (which converts analog current signals into 16-bit digital signals), to obtain first spectral data (corresponding to the first characteristic frequency band) and second spectral data (corresponding to the second characteristic frequency band) respectively, which together constitute the multi-spectral data.

[0089] S33: Verifying the effectiveness of multi-spectral data acquisition and temporarily storing.

[0090] Specifically, the multispectral data is temporarily stored in the cache unit of the processing device, and a signal-to-noise ratio calculation module in the cache unit calculates the signal-to-noise ratio of the data by the formula "signal-to-noise ratio (SNR) = signal mean value / noise standard deviation", wherein the signal mean value is the average value of 100 groups of spectral data in a certain frequency band, and the noise standard deviation is the square root of the sum of squares of deviations of the data in the frequency band from the mean value. If the SNR is greater than a preset threshold (such as 30 dB), it is determined that the collection is valid; if the SNR is less than or equal to 30 dB, the spectral control module sends a "re-collection" instruction until valid data is collected.

[0091] S4: performing ground object dielectric signal preliminary screening processing based on the multispectral data of the first characteristic frequency band and the second characteristic frequency band to obtain a ground object dielectric signal difference value.

[0092] Specifically, the method for obtaining the ground object dielectric signal difference value in S4 includes:

[0093] S41: extracting the first spectral data and the second spectral data.

[0094] Specifically, when the multispectral data is temporarily stored, the first spectral data is marked as "Band1_spectral data", and the second spectral data is marked as "Band2_spectral data"; the data extraction module of the processing device extracts the two types of data from the cache unit through a preset data identifier, and converts them into a CSV format through a data format conversion tool after extraction is completed, so as to facilitate subsequent noise filtering processing.

[0095] S42: performing noise filtering processing on the spectral data.

[0096] Specifically, the preset noise filtering method is a wavelet threshold denoising method: db4 wavelet basis is selected to perform 3-layer wavelet decomposition on the first and second spectral data, and high-frequency coefficients (corresponding to noise) and low-frequency coefficients (corresponding to valid signals) are obtained after decomposition; a soft threshold function (the threshold is set to "0.5x high-frequency coefficient standard deviation") is used to perform threshold processing on the high-frequency coefficients to suppress noise components; and finally, the processed high-frequency coefficients and low-frequency coefficients are recombined through wavelet reconstruction to obtain the filtered first spectral data and second spectral data.

[0097] S43: screening the ground object dielectric signal and calculating the difference value.

[0098] Specifically, the model database of the processing device pre-stores dielectric signal characteristic templates (template data is derived from laboratory measurement: the dielectric constant of different ground objects in the 490-510 nm and 1550-1650 nm frequency bands is measured by a dielectric constant tester to generate the characteristic templates) of different ground object types (such as cultivated land, forest land, building land, and water body) in the first and second characteristic frequency bands; the data matching module calls the templates and calculates the similarity of "filtered spectral data and templates" by using a normalized cross-correlation matching method, and the similarity calculation formula is:

[0099]

[0100] wherein, X i is the i-th sample point of filtered spectral data, is the mean value of filtered spectral data, Y i is the i-th sample point of dielectric signal template, is the mean value of template, R(x, y) is the similarity (value range [-1, 1], the closer to 1 indicates the higher matching degree). The dielectric signal corresponding to the template with the highest similarity R(x, y)≥0.85 is selected as the first ground object dielectric signal (corresponding to the first characteristic frequency band) and the second ground object dielectric signal (corresponding to the second characteristic frequency band), respectively. The difference value (difference value = first ground object dielectric signal-second ground object dielectric signal) is calculated. After matching the dielectric signal of all pixels of the image point by point, the ground object dielectric signal difference matrix, i.e. the ground object dielectric signal difference, is generated.

[0101] S5: A coupling correction model is established in combination with the ground object dielectric signal difference.

[0102] Specifically, the method for establishing the coupling correction model in S5 includes:

[0103] S51: A plurality of sets of sample data are collected.

[0104] Specifically, the sample data is derived from remote sensing images under different regions (such as North China Plain, Qinghai-Tibet Plateau, southeast coast), different times (spring, summer, autumn, and winter), and different weather conditions (sunny, cloudy, and light rain). Each set of sample data includes “sample dual-band dielectric difference” and corresponding “sample multi-spectral characteristic peak intensity”: the sample dual-band dielectric difference is extracted from the sample image by the method of S4; the sample multi-spectral characteristic peak intensity is extracted by a peak intensity extraction module, which identifies the characteristic peaks (such as the 498 nm peak of the first characteristic frequency band and the 1600 nm peak of the second characteristic frequency band) in the multi-spectral data and calculates the peak height (i.e. the signal difference between the peak top and the peak bottom) of the characteristic peak as the sample multi-spectral characteristic peak intensity.

[0105] S52: The plurality of sets of sample data are preprocessed to remove abnormal sample data.

[0106] Specifically, the preprocessing process includes:

[0107] The first mean value and the first standard deviation (σ D of the “sample dual-band dielectric difference” in each set of sample data are calculated. and the second standard deviation (σ I of the “sample multi-spectral characteristic peak intensity” are calculated.

[0108] Set the abnormal data elimination threshold range: the sample dual-band dielectric difference value needs to meet The sample multi-spectral feature peak intensity needs to meet

[0109] Eliminate sample data exceeding the threshold range through the data screening module, and retain valid sample data.

[0110] S53: Train and verify the coupling correction model.

[0111] Specifically, the preprocessed sample data is divided into training data (70%) and verification data (30%) in a ratio of 7:3; with "preprocessed sample dual-band dielectric difference value" as the independent variable (X) and "preprocessed sample multi-spectral feature peak intensity" as the dependent variable (Y), the model is trained on the MATLAB R2023a platform using partial least squares regression (PLSR): latent variables (principal components) of X and Y are extracted by PLSR, a linear regression relationship between latent variables is established, and finally the model expression Y=aX+b is obtained (where a is the regression coefficient and b is the intercept). In the verification stage: input the verification data into the model, judge the model fitting degree by the determination coefficient (R2) (R2≥0.9 is required), and judge the model prediction error by the root mean square error (RMSE) (RMSE≤0.05 is required); if both R2 and RMSE meet the preset requirements, store the model parameters (a, b and latent variable extraction rules) to the model database of the processing device, and finally form the coupling correction model; if not, adjust the number of principal components of PLSR (such as adjusting to 5-8 principal components), and train again until the requirements are met.

[0112] S6: Construct a scene interference type identification library, which contains typical interference feature spectra of atmospheric interference, terrain shadow, and sensor noise.

[0113] Specifically, the construction process of the scene interference type identification library includes:

[0114] Collect typical samples of three types of interference: atmospheric interference samples (such as remote sensing images of foggy and dusty days, extract scattering signal spectra of 490-510nm frequency band), terrain shadow samples (such as shadow area images around mountainous areas and high-rise buildings, extract shadow absorption spectra of 1550-1650nm frequency band), and sensor noise samples (such as stripe noise and random noise images caused by sensor aging, extract noise signal spectra of full frequency band);

[0115] Feature extraction of sample atlas: adopt edge detection (Canny algorithm), gray histogram analysis and other methods to extract the "frequency band-signal intensity" characteristic curve of each type of interference, such as the characteristic of atmospheric interference is "490-510nm frequency band signal intensity abnormally high (more than 30% higher than sunny day)", the characteristic of terrain shadow is "1550-1650nm frequency band signal intensity significantly reduced (more than 50% lower than non-shadow area)";

[0116] The typical feature atlas and feature description of the three types of interference are stored in the recognition library unit of the processing device, and an atlas update interface is set to support the periodic import of new interference samples (such as interference samples under extreme weather) to update the content of the recognition library.

[0117] S7: judging the main interference type of the current image through the scene interference type recognition library, dynamically adjusting the model parameters of the coupling correction model; inputting the difference between the multispectral data and the ground object dielectric signal into the adjusted coupling correction model, and outputting the high-precision remote sensing image data after eliminating interference by the adjusted coupling correction model.

[0118] Specifically, the execution process of S7 includes:

[0119] S71: judging the main interference type of the current image.

[0120] Specifically, the interference recognition module of the processing device calls the scene interference type recognition library to perform similarity matching (using the normalized cross-correlation matching method in S4) between the multispectral data atlas of the current image and the typical feature atlas of the three types of interference in the library; if the similarity with the atmospheric interference atlas is the highest (R(x, y) ≥ 0.8), it is determined that the main interference type is atmospheric interference; if the similarity with the terrain shadow atlas is the highest (R(x, y) ≥ 0.8), it is determined that the main interference type is terrain shadow; if the similarity with the sensor noise atlas is the highest (R(x, y) ≥ 0.8), it is determined that the main interference type is sensor noise; if all are lower than 0.8, it is determined that there is no significant interference.

[0121] S72: dynamically adjusting the parameters of the coupling correction model.

[0122] Specifically, the parameter adjustment rules are preset in the model database:

[0123] For atmospheric interference: adjust the scattering coefficient weight in the model (increase the original coefficient a by 20% to enhance the suppression of atmospheric scattering signal);

[0124] For terrain shadow: correct the light angle related parameters (add a shadow compensation term c in the model, the model expression becomes Y=aX+b+c, c takes a value of 0.1-0.3, and is dynamically adjusted according to the shadow intensity);

[0125] For sensor noise: increase the smoothing filter coefficient (add Gaussian filter before model output, filter window size is set to 3x3, reduce noise impact);

[0126] When there is no significant interference, keep the model parameters unchanged. The adjusted model parameters are stored in the cache unit in real time, covering the original parameters.

[0127] S73: input data and output high-precision image.

[0128] Specifically, the multispectral data (first and second spectral data) obtained in S3 and the difference value of the ground object dielectric signal obtained in S4 are input into the adjusted coupling correction model; the model eliminates interference through the following steps:

[0129] According to the frequency band distribution of the interference signal identified from the multispectral data;

[0130] Combine the ground object dielectric signal difference value to distinguish "interference signal" and "ground object effective signal";

[0131] Adjust the parameters to suppress the interference signal (such as weakening the abnormal signal of atmospheric scattering, compensating for the signal loss in the shadow area);

[0132] Finally, output the high-precision remote sensing image data after eliminating interference, and uniformly convert the data format to TIFF format, and keep the highest resolution of the original image (such as 0.5 meters / pixel).

[0133] S8: After quality detection of the high-precision remote sensing image data after interference elimination, perform the warehousing operation, such as Figure 3 The specific process includes:

[0134] S81: Quality detection is performed on the high-precision remote sensing image data after interference elimination, and the quality detection includes spectral accuracy detection and spatial resolution detection.

[0135] Specifically, the spectral accuracy detection method includes:

[0136] Obtain the preset standard spectral data set: the data set is stored in the standard database of the processing device, and contains the standard spectral curves corresponding to 12 common ground objects such as cultivated land, forest land, building land and water body (such as the standard reflectivity of cultivated land in the frequency band of 490-510nm is 15%-20%, and in the frequency band of 1550-1650nm is 5%-8%), and the standard curve data is derived from the ground spectral measurement report certified by the National Institute of Metrology;

[0137] Dividing ground object area and extracting actual spectral curve: threshold method is used for image segmentation, different ground object areas are divided according to the gray threshold of multi-spectral data (such as water gray value ≤ 50, building land gray value ≥ 180), and the spectral data of each area is processed by "pixel by pixel average" method to obtain the actual spectral curve of the area;

[0138] Calculate similarity and determine: the cosine similarity formula is used to calculate the similarity of the actual spectral curve and the corresponding standard spectral curve, the formula is:

[0139]

[0140] Where, A i is the reflectivity of the i-th sampling point of the actual spectral curve, B i is the reflectivity of the i-th sampling point of the standard spectral curve, and cosθ is the similarity (value range [0, 1]). If cosθ > 0.9 (preset similarity threshold), it is determined that the spectral accuracy meets the preset storage standard; otherwise, it is determined that it does not meet the standard.

[0141] The spatial resolution detection method comprises:

[0142] Obtaining spatial resolution parameter: the metadata information of high-precision remote sensing image data is extracted through the resolution reading module of the processing device, and the spatial resolution parameter (such as 0.5 meters / pixel, 1 meter / pixel) is obtained;

[0143] Parameter comparison and determination: the obtained spatial resolution parameter is compared with the preset storage standard (such as the storage standard in the field of land surveying and mapping is "spatial resolution ≥ 1 meter / pixel", and the storage standard in the field of ecological monitoring is "≥ 0.5 meter / pixel"); if the spatial resolution parameter is greater than or equal to the preset standard, it is determined that the spatial resolution meets the requirement; otherwise, it is determined that it does not meet the requirement.

[0144] S82: performing storage or reprocessing operation according to the quality detection result.

[0145] Specifically, if the spectral accuracy and the spatial resolution both meet the preset storage standard, the storage control module of the processing device transmits the high-precision remote sensing image data and its metadata (including collection time, geographic coordinates, resolution, ground object type statistics) to the target database (such as PostgreSQL + PostGIS spatial database), adopts block transmission strategy (each block is set to 100MB in size) in the transmission process, and ensures data integrity through CRC32 check; if any detection item does not meet the standard, the storage control module triggers the "reprocessing" instruction, and returns the data to the S2 step to re-execute the "region priority classification-GPU parallel processing-multi-spectral acquisition-interference elimination" whole process until the quality detection is qualified.

[0146] S83: data management and index construction after warehousing.

[0147] Specifically, after the data warehousing is completed, the database management module automatically constructs a spatial index (such as an R-tree index) and an attribute index (such as a collection time index and a ground object type index) for the image data, so as to quickly locate the data during subsequent queries. For example, when a user queries "remote sensing images of a certain city core area in summer 2024", the system can filter the time and region through the attribute index, and then locate the specific image through the spatial index, and the query response time is controlled within 3 seconds. At the same time, the database sets access permission control, which is divided into three levels of "administrator", "ordinary user" and "visitor". The administrator can modify and delete data, the ordinary user can only read and download data, and the visitor can only preview the thumbnail, so as to ensure the data security.

[0148] In a second aspect, referring to the drawings Figure 2 Based on the same inventive concept, the embodiment further provides a rapid warehousing system based on remote sensing images, which is applied to execute the rapid warehousing method based on remote sensing images and comprises:

[0149] A data acquisition module is configured to acquire original remote sensing image data to be warehoused.

[0150] A data processing module is configured to perform importance classification on an image region corresponding to the original remote sensing image data, mark the image region as different priorities according to a preset rule, and synchronously process the image regions of different priorities through a GPU parallel framework.

[0151] A spectrum acquisition module is configured to perform multispectral feature acquisition on the processed remote sensing image data, and synchronously acquire multispectral data of the remote sensing image data in a first feature frequency band and a second feature frequency band.

[0152] A dielectric preliminary screening module is configured to perform dielectric signal preliminary screening processing on the multispectral data in the first feature frequency band and the second feature frequency band, and obtain a dielectric signal difference value.

[0153] A model construction module is configured to establish a coupling correction model in combination with the dielectric signal difference value.

[0154] An identification library module is configured to construct a scene interference type identification library, and the identification library comprises typical interference feature spectra of atmospheric interference, terrain shadow and sensor noise.

[0155] An interference elimination module is configured to determine a main interference type of a current image through the scene interference type identification library, dynamically adjust model parameters of the coupling correction model, input the multispectral data and the dielectric signal difference value into the adjusted coupling correction model, and output high-precision remote sensing image data after interference elimination from the adjusted coupling correction model.

[0156] An image warehousing module is used for warehousing high-precision remote sensing image data.

[0157] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as limiting the scope of the patent of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, several modifications and improvements can be made, which belong to the protection scope of the present application.

Claims

1. A rapid warehousing method based on remote sensing images, characterized in that, The method comprises: S1: obtaining original remote sensing image data to be stored in a warehouse; S2: classifying the importance of an image area corresponding to the original remote sensing image data, marking the image area as different priorities according to a preset rule, and synchronously processing image areas of different priorities through a GPU parallel framework; S3: collecting multispectral features of the processed remote sensing image data, and synchronously acquiring multispectral data thereof in a first feature frequency band and a second feature frequency band; S4: performing preliminary screening processing on the dielectric signal of ground objects based on the multispectral data in the first feature frequency band and the second feature frequency band, to obtain a dielectric signal difference value of ground objects; S5: combining the dielectric signal difference value of ground objects, and establishing a coupling correction model; S6: constructing a scene interference type recognition library, the recognition library containing typical interference characteristic spectra of atmospheric interference, terrain shadow, and sensor noise; S7: determining the main interference type of the current image through the scene interference type recognition library, dynamically adjusting the model parameters of the coupling correction model, inputting the multispectral data and the dielectric signal difference value of ground objects into the adjusted coupling correction model, and outputting high-precision remote sensing image data after interference elimination by the adjusted coupling correction model; S8: storing the high-precision remote sensing image data in the warehouse.

2. The method of claim 1, wherein, The first feature frequency band is an atmospheric scattering characteristic peak frequency band, and the second feature frequency band is a cloud shadow absorption characteristic peak frequency band; In S3, the multispectral data is specifically obtained as follows: acquiring preset dual-band spectral acquisition parameters, the dual-band spectral acquisition parameters including a first acquisition wavelength range, a first sampling frequency of the first feature frequency band, and a second acquisition wavelength range, a second sampling frequency of the second feature frequency band, the dual-band spectral acquisition parameters being pre-stored in a storage unit of a processing device; controlling a multispectral acquisition component to simultaneously perform spectral data acquisition on the processed remote sensing image data through a dual-channel data acquisition circuit according to the dual-band spectral acquisition parameters; sequentially performing photoelectric conversion and analog-digital conversion on the acquired spectral data to obtain first spectral data and second spectral data, which together constitute the multispectral data; temporarily storing the multispectral data in a cache unit of the processing device, and verifying the acquisition effectiveness by calculating the signal-to-noise ratio; when the signal-to-noise ratio is greater than a preset signal-to-noise ratio threshold, it is determined that the acquisition is valid; if it is less than or equal to the preset signal-to-noise ratio threshold, spectral data acquisition is performed again until the acquisition is valid.

3. The method of claim 2, wherein, In S4, the dielectric signal difference value of ground objects is specifically obtained as follows: determining the first spectral data and the second spectral data from the multispectral data, and extracting them from the cache unit of the processing device through a preset data identifier; respectively performing noise filtering processing on the first spectral data and the second spectral data, and completing data denoising by using a preset noise filtering method to obtain filtered first spectral data and second spectral data; calling pre-stored dielectric signal feature templates of different ground object types in corresponding frequency bands; The similarity is calculated by using a normalized cross-correlation matching method, the dielectric signals corresponding to the templates with the highest similarity are screened as the first and second ground object dielectric signals, and a difference value between the two is calculated to generate the ground object dielectric signal difference value by point-by-point matching calculation.

4. The method of claim 1, wherein, The establishing of the coupling correction model in S5 specifically includes: a plurality of groups of sample data are collected, each group of the sample data including a sample dual-frequency band dielectric difference value and a corresponding sample multi-spectral feature peak intensity; the sample data are derived from remote sensing images under different regions, times and weather conditions, and the sample multi-spectral feature peak intensity is extracted by a preset method; the plurality of groups of sample data are preprocessed to remove abnormal sample data; the sample dual-frequency band dielectric difference value after preprocessing is taken as an independent variable, and the sample multi-spectral feature peak intensity after preprocessing is taken as a dependent variable, and a partial least squares regression method is used for training; the sample data after preprocessing are divided into training data and verification data, the fitting degree of the model is judged by a coefficient of determination, and the prediction error of the model is judged by a root mean square error, and after the fitting degree and the prediction error of the model both meet preset requirements, the model parameters are stored in a model database of a processing device, and finally the coupling correction model is formed.

5. The method of claim 4, wherein, the preprocessing of the plurality of groups of sample data specifically includes: a first mean value and a first standard deviation of the sample dual-frequency band dielectric difference value and a second mean value and a second standard deviation of the sample multi-spectral feature peak intensity in each group of the sample data are calculated; an abnormal data elimination threshold range is set according to the first mean value, the first standard deviation, the second mean value and the second standard deviation; the sample dual-frequency band dielectric difference value needs to meet "first mean value-3 times first standard deviation≤sample dual-frequency band dielectric difference value≤first mean value+3 times first standard deviation", and the sample multi-spectral feature peak intensity needs to meet "second mean value-3 times second standard deviation≤sample multi-spectral feature peak intensity≤second mean value+3 times second standard deviation"; sample data beyond the range are eliminated, and effective sample data are retained.

6. The method of claim 1, wherein, Before S8 is performed, the following steps are further included: the high-precision remote sensing image data after the interference is eliminated is subjected to quality detection, and the quality detection includes spectral accuracy detection and spatial resolution detection; if both meet preset storage standards, S8 is performed; if not, return to S2 for reprocessing, and then execute subsequent steps in turn.

7. The method of claim 6, wherein, the spectral accuracy detection of the high-precision remote sensing image data after the interference is eliminated specifically includes: a preset standard spectral data set containing standard spectral curves corresponding to different ground object types is obtained and stored in a standard database of a processing device; threshold method is used for image segmentation to divide ground object regions, and the spectral data of each region is averaged to extract an actual spectral curve; a cosine similarity method is used to calculate the similarity between the actual spectral curve and the corresponding standard spectral curve; if the similarity is greater than a preset similarity threshold, it is determined that the spectral accuracy of the high-precision remote sensing image data after the interference is eliminated meets the preset storage standard; otherwise, it is determined that the preset storage standard is not met.

8. The method of claim 7, wherein, the spatial resolution detection of the high-precision remote sensing image data after the interference is eliminated specifically includes: acquire a spatial resolution parameter of the high-precision remote sensing image data after interference elimination; compare the spatial resolution parameter with a spatial resolution requirement in a preset storage standard; if the spatial resolution parameter is greater than or equal to the spatial resolution requirement, it is determined that the spatial resolution of the high-precision remote sensing image data after interference elimination meets the preset storage standard; otherwise, it is determined that the high-precision remote sensing image data after interference elimination does not meet the preset storage standard.

9. The method of claim 1, wherein, The importance classification and processing of the image region corresponding to the original remote sensing image data in S2 specifically include: taking the importance of the application scene of the image region as the division basis, marking the city core area, the disaster sensitive area and the ecological protection area as high-priority areas, and marking the desert, the gobi and the uninhabited area as low-priority areas; allocating computing power for different priority areas through a GPU parallel framework; adopting a data slicing method to slice the image according to a preset size, synchronously performing data processing on different priority sliced areas, and obtaining the processed remote sensing image data.

10. A remote sensing image-based rapid warehousing system, characterized in that, It includes: a data acquisition module configured to acquire original remote sensing image data to be stored; a data processing module configured to classify the importance of an image region corresponding to the original remote sensing image data, mark the image region as different priorities according to a preset rule, and synchronously process image regions of different priorities through a GPU parallel framework; a spectrum acquisition module configured to acquire multi-spectral data of the processed remote sensing image data in a first characteristic frequency band and a second characteristic frequency band; a dielectric preliminary screening module configured to perform dielectric signal preliminary screening processing on the multi-spectral data in the first characteristic frequency band and the second characteristic frequency band to obtain a dielectric signal difference value; a model construction module configured to establish a coupling correction model in combination with the dielectric signal difference value; an identification library module configured to construct a scene interference type identification library, the identification library including typical interference feature maps of atmospheric interference, terrain shadow and sensor noise; an interference elimination module configured to determine the main interference type of the current image through the scene interference type identification library, dynamically adjust the model parameters of the coupling correction model, input the multi-spectral data and the dielectric signal difference value into the adjusted coupling correction model, and output high-precision remote sensing image data after interference elimination from the adjusted coupling correction model; an image storage module configured to store the high-precision remote sensing image data.

Images