Image processing method and system based on artificial intelligence

Abstract

The invention discloses an image processing method and system based on artificial intelligence, and relates to the technical field of artificial intelligence, and the image processing method based on artificial intelligence comprises the steps: S1, obtaining to-be-processed data according to an initial collection frequency; s2, preprocessing the to-be-processed data to obtain preprocessed data; s3, obtaining a target mining data set according to the collected codes, analyzing the preprocessed data through the target mining data set to obtain a current difference value, analyzing the current difference value, and if the current difference value belongs to a normal range, executing S1; if the current difference value belongs to the critical range, adjusting the initial acquisition frequency to obtain a new acquisition frequency, obtaining to-be-processed data according to the new acquisition frequency, and executing S2; and if the current difference value belongs to the abnormal range, sending alarm information. According to the method, the to-be-processed data can be corrected, so that the data quality is improved, and timely early warning can be performed on disaster hidden dangers.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to an image processing method and system based on artificial intelligence. Background Art

[0002] Oil and gas exploration includes three links: field seismic data acquisition, seismic data processing, and seismic data interpretation; among them, in the field data acquisition link, the artificial seismic method (for example: artificial blasting or using a vibroseis vehicle to generate artificial seismic waves) is mainly used to collect seismic waves returning from underground to form field observation data. The task of field acquisition in seismic exploration is to obtain original data, and the quality of the original data directly affects the quality of data digital processing and the accuracy of interpretation results. The field acquisition work of seismic exploration consists of various stages such as on-site reconnaissance, construction design, test work, and formal production, and needs to be closely coordinated by multiple types of work such as surveying, drilling, excitation, reception, and interpretation.

[0003] During the exploitation process, using multi-temporal remote sensing images to monitor the surface changes in the oil and gas exploration area, such as ground subsidence, crack expansion, etc., can timely detect potential geological hazards and environmental problems during oil and gas exploitation. However, the existing image processing methods for detecting geological hazards during oil and gas exploitation have the following problems:

[0004] (1) Due to the different geological environments and acquisition methods in the oil and gas exploitation area, the collected data may be affected by various factors such as bad weather (such as heavy rain, thick fog, etc.), poor lighting conditions (such as deep sea or underground exploitation environment), imaging methods, etc., resulting in blurred images, low contrast, and difficulty in clearly identifying key geological hazard features (such as formation cracks, signs of landslides), as well as data quality problems such as geometric deformation caused by terrain undulation and camera pose changes.

[0005] (2) Oil and gas exploitation is a dynamic process, and geological hazards may occur at any time, requiring rapid analysis of images for timely early warning. However, a large amount of image data plus complex processing algorithms will result in the processing speed not being able to keep up with the actual needs, making it difficult to detect disasters in a timely manner. Summary of the Invention

[0006] The purpose of this application is to provide an image processing method and system based on artificial intelligence, which can correct the data to be processed, thereby improving the data quality, and can timely give early warnings of potential disasters.

[0007] To achieve the above object, the present application provides an image processing method based on artificial intelligence, including the following steps: S1: Obtain data to be processed according to the initial acquisition frequency, where the data to be processed includes: acquisition coding, acquisition time, and acquisition data, and the acquisition data at least includes: optical remote sensing data and synthetic aperture radar data; S2: Preprocess the data to be processed to obtain preprocessed data, where the preprocessed data at least includes: acquisition coding, acquisition time, first correction data, and second correction data; S3: Obtain the target mining dataset according to the acquisition coding, analyze the preprocessed data through the target mining dataset to obtain the current difference value, and analyze the current difference value. If the current difference value is within the normal range, execute S1; if the current difference value is within the critical range, adjust the initial acquisition frequency to obtain a new acquisition frequency, obtain the data to be processed according to the new acquisition frequency, and execute S2; if the current difference value is within the abnormal range, send an alarm message.

[0008] As above, where the sub-steps of preprocessing the data to be processed to obtain preprocessed data are as follows: S21: Query the standard database according to the acquisition coding, and use the pre-mining data in the standard data packet with the query coding consistent with the acquisition coding as the target data; S22: Use the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data; S23: Filter the synthetic aperture radar data in the acquisition data to obtain the second correction data; S24: Use the acquisition coding, acquisition time, first correction data, and second correction data as the preprocessed data.

[0009] As above, where the sub-steps of using the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data are as follows: S221: Select multiple uniformly distributed ground control points on the original optical remote sensing data in the target data; S222: Use the ground control points to perform imaging correction processing on the optical remote sensing data in the acquisition data to obtain imaging correction data; S223: Use the ground control points to perform deformation correction processing on the imaging correction data to obtain deformation correction data, and use the deformation correction data as the first correction data.

[0010] As described above, in which, the sub-steps of performing imaging correction processing on the optical remote sensing data in the acquired data by using ground control points to obtain imaging correction data are as follows: S2221: Analyze the optical remote sensing data in the acquired data by using a first classification model based on artificial intelligence to determine the imaging category to which the optical remote sensing data in the acquired data belongs, where the imaging category at least includes: central projection imaging, multi-central projection imaging, and area array imaging; S2222: Select a corresponding imaging correction model according to the imaging category, and use the corresponding imaging correction model and ground control points to correct the optical remote sensing data in the acquired data, and output initial imaging correction data, where the imaging correction model at least includes: polynomial correction model, physical model, and rational function model; S2223: Perform accuracy evaluation on the initial imaging correction data to obtain a first accuracy evaluation result. If the first accuracy evaluation result is qualified, use the initial imaging correction data as the imaging correction data; if the first accuracy evaluation result is unqualified, re-perform imaging correction processing on the optical remote sensing data in the acquired data.

[0011] As described above, in which, the sub-steps of performing accuracy evaluation on the initial imaging correction data to obtain a first accuracy evaluation result are as follows: S22231: Calculate the residuals between each ground control point and the corresponding corrected coordinates in the initial imaging correction data; S22232: Perform statistical analysis on all the residuals to obtain an imaging comprehensive error value; S22233: Analyze the imaging comprehensive error value by using an imaging accuracy threshold to generate a first accuracy evaluation result; if the imaging comprehensive error value is less than the imaging accuracy threshold, the generated first accuracy evaluation result is qualified; if the imaging comprehensive error value is greater than or equal to the imaging accuracy threshold, the generated first accuracy evaluation result is unqualified.

[0012] As described above, the sub-steps of using ground control points to perform deformation correction on the imaging correction data, obtaining the deformation correction data, and using the deformation correction data as the first correction data are as follows: S2231: Analyze the imaging correction data using the second classification model based on artificial intelligence to determine the geometric deformation category to which the imaging correction data belongs; among them, the geometric deformation category at least includes: topographic geometric deformation, systematic geometric deformation, and non-systematic geometric deformation; S2232: Select the corresponding deformation correction model according to the geometric deformation category, and use the corresponding deformation correction model and ground control points to correct the imaging correction data, and output the initial deformation correction data, where the deformation correction model at least includes: orthorectification model, sensor-based physical correction model, polynomial correction model, correction model with compensation terms added on the basis of a strict physical model; S2233: Perform accuracy evaluation on the initial deformation correction data to obtain the second accuracy evaluation result. If the second accuracy evaluation result is qualified, use the initial deformation correction data as the deformation correction data and use the deformation correction data as the first correction data; if the second accuracy evaluation result is unqualified, re-perform deformation correction on the imaging correction data.

[0013] As described above, the sub-steps of performing accuracy evaluation on the initial deformation correction data to obtain the second accuracy evaluation result are as follows: S22331: Analyze each ground control point and the corresponding corrected coordinates in the initial deformation correction data to obtain the first deformation error value; S22332: Select multiple evenly distributed ground test points on the original optical remote sensing data in the target data, and analyze each ground test point and the corresponding corrected coordinates in the initial deformation correction data to obtain the second deformation error value, where the ground test points are different from the ground control points; S22333: Obtain the comprehensive deformation error value according to the first deformation error value and the second deformation error value, and analyze the comprehensive deformation error value using the deformation accuracy threshold to generate the second accuracy evaluation result; if the comprehensive deformation error value is less than the deformation accuracy threshold, the generated second accuracy evaluation result is qualified; if the comprehensive deformation error value is greater than or equal to the deformation accuracy threshold, the generated second accuracy evaluation result is unqualified.

[0014] As described above, wherein, a target mining dataset is obtained according to the acquisition code, and the preprocessed data is analyzed through the target mining dataset to obtain the sub-steps of the current difference value as follows: S31: Traverse the standard database according to the acquisition code, and use the mining dataset in the standard data packet whose query code is consistent with the acquisition code as the comparison dataset. The comparison dataset includes: data of multiple mining periods; each mining period data includes: post-mining acquisition time, post-mining optical remote sensing data, and post-mining synthetic aperture radar data; S32: Sort the mining period data and the preprocessed data in the order of multiple post-mining acquisition times and acquisition times from the earliest to the latest, and generate corresponding analysis serial numbers to obtain the data to be analyzed; wherein, each data to be analyzed includes: analysis serial number, analysis acquisition time, analysis optical remote sensing data, and analysis synthetic aperture radar data; the analysis serial numbers increase sequentially in the order of multiple post-mining acquisition times and acquisition times from the earliest to the latest; S33: Analyze the data to be analyzed to obtain the current difference value.

[0015] As described above, wherein, the expression of the current difference value is: Wherein, Dcy is the current difference value; λ1 is the pixel difference weight; ΔCxs k is the average value of the differences of all corresponding pixels between the analysis optical remote sensing data in the (k + 1)-th data to be analyzed and the analysis optical remote sensing data in the k-th data to be analyzed; ΔCxs K is the average value of the differences of all corresponding pixels between the analysis optical remote sensing data in the (K + 1)-th data to be analyzed and the analysis optical remote sensing data in the K-th data to be analyzed; K + 1 is the total number of data to be analyzed, and K is a natural number; λ2 is the disaster similarity weight; Gzh K+1 is the similarity value between the analysis synthetic aperture radar data in the (K + 1)-th data to be analyzed and the pre-constructed geological disaster feature set.

[0016] The present application also provides an artificial intelligence-based image processing system, including: at least one data acquisition node and an artificial intelligence image processing center; wherein, the data acquisition node: acquires the data to be processed according to the initial acquisition frequency or the new acquisition frequency, and sends the data to be processed to the artificial intelligence image processing center; receives the alarm information; the artificial intelligence image processing center: is used to execute the above-mentioned artificial intelligence-based image processing method.

[0017] The beneficial effects achieved by the present application are as follows:

[0018] (1) The artificial intelligence-based image processing method and system of the present application can correct the data to be processed, thereby improving the data quality.

[0019] (2) The image processing method and system based on artificial intelligence of the present application can improve the accuracy of the analysis results of disaster hazards and give timely warnings about disaster hazards. Description of the Drawings

[0020] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0021] Figure 1 It is a schematic structural diagram of an embodiment of an image processing system based on artificial intelligence;

[0022] Figure 2 It is a flowchart of an embodiment of an image processing method based on artificial intelligence. Detailed Embodiments

[0023] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, rather than all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0024] As Figure 1 shown, the present application provides an image processing system based on artificial intelligence, including: at least one data acquisition node 110 and an artificial intelligence image processing center 120.

[0025] Among them, the data acquisition node 110: acquires the data to be processed according to the initial acquisition frequency or the new acquisition frequency, and sends the data to be processed to the artificial intelligence image processing center 120; receives the alarm information.

[0026] The artificial intelligence image processing center 120: is used to execute the following image processing method based on artificial intelligence.

[0027] Furthermore, the artificial intelligence image processing center 120 includes: a transceiver unit, a preprocessing unit, an analysis unit, and a storage unit.

[0028] Among them, the transceiver unit: receives the data to be processed according to the initial acquisition frequency or the new acquisition frequency, and sends the data to be processed to the preprocessing unit.

[0029] The preprocessing unit: preprocesses the data to be processed to obtain preprocessed data.

[0030] Analysis unit: Obtain the target mining data set according to the acquisition code, analyze the pre-processed data through the target mining data set to obtain the current difference value, and analyze the current difference value. If the current difference value is within the normal range, the transceiver unit obtains the data to be processed according to the initial acquisition frequency; if the current difference value is within the critical range, adjust the initial acquisition frequency to obtain a new acquisition frequency, and the transceiver unit obtains the data to be processed according to the new acquisition frequency; if the current difference value is within the abnormal range, send an alarm message.

[0031] Storage unit: Used to store the standard database, imaging correction model database, and deformation correction model database.

[0032] Among them, the standard database includes: multiple standard data packets, one standard data packet corresponds to one query code, and each standard data packet includes: pre-mining data and mining data set. The pre-mining data includes: original acquisition time, original optical remote sensing data, and original synthetic aperture radar data. The mining data set includes: multiple mining period data; each mining period data includes: post-mining acquisition time, post-mining optical remote sensing data, and post-mining synthetic aperture radar data.

[0033] Specifically, the original acquisition time is the period for acquiring pre-mining data.

[0034] The original optical remote sensing data is: complete optical remote sensing image data acquired before mining and covering the acquisition area corresponding to the query code.

[0035] The original synthetic aperture radar data is: complete synthetic aperture radar data acquired before mining and covering the acquisition area corresponding to the query code.

[0036] Among them, the imaging correction model database includes: multiple standard imaging correction models, one standard imaging correction model corresponds to one standard imaging category, and each standard imaging correction model includes: one imaging correction module or a combination of multiple imaging correction models.

[0037] Specifically, the standard imaging categories at least include: central projection imaging, multi-central projection imaging, and area array imaging. Set the standard imaging correction model corresponding to each standard imaging category according to the actual situation.

[0038] Among them, the deformation correction model database includes: multiple standard deformation correction models, one standard deformation correction model corresponds to one standard geometric deformation category, and each standard deformation correction model includes: one deformation correction module or a combination of multiple deformation correction models.

[0039] Specifically, the standard geometric deformation categories at least include: topographic geometric deformation, systematic geometric deformation, and non-systematic geometric deformation. Set the corresponding standard deformation correction model for each standard geometric deformation category according to the actual situation.

[0040] As Figure 2 shown, this application provides an image processing method based on artificial intelligence, including the following steps:

[0041] S1: Obtain the data to be processed according to the initial acquisition frequency. Among them, the data to be processed includes: acquisition code, acquisition time, and acquisition data. Among them, the acquisition data at least includes: optical remote sensing data and synthetic aperture radar data.

[0042] Specifically, the acquisition code is the identification code of the acquisition area for obtaining the acquisition data. One acquisition area corresponds to one acquisition code, and the acquisition codes of different acquisition areas are all different.

[0043] The acquisition time is the time period for obtaining the acquisition data.

[0044] The acquisition data is high-resolution remote sensing satellite data covering the oil and gas extraction area, at least including: optical remote sensing data and synthetic aperture radar data.

[0045] S2: Preprocess the data to be processed to obtain preprocessed data. Among them, the preprocessed data at least includes: acquisition code, acquisition time, first correction data, and second correction data.

[0046] Furthermore, the sub-steps of preprocessing the data to be processed to obtain preprocessed data are as follows:

[0047] S21: Query the standard database according to the acquisition code, and use the pre-exploitation data in the standard data packet with the query code consistent with the acquisition code as the target data.

[0048] Specifically, the target data includes: original acquisition time, original optical remote sensing data, and original synthetic aperture radar data.

[0049] S22: Use the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data.

[0050] Furthermore, the sub-steps of using the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data are as follows:

[0051] S221: Select multiple evenly distributed ground control points on the original optical remote sensing data in the target data.

[0052] Specifically, the ground control points are ground object points that are easy to identify on the image and have accurate geographic coordinates, such as: road intersections, building corners, etc.

[0053] Further, the specific number of ground control points is set according to the resolution, range, and calibration accuracy requirements of the image. In this application, it is preferably that the number of ground control points is greater than or equal to six.

[0054] S222: Use the ground control points to perform imaging calibration processing on the optical remote sensing data in the collected data to obtain imaging calibration data.

[0055] Further, the sub-steps of using the ground control points to perform imaging calibration processing on the optical remote sensing data in the collected data to obtain imaging calibration data are as follows:

[0056] S2221: Use the first classification model based on artificial intelligence to analyze the optical remote sensing data in the collected data to determine the imaging category to which the optical remote sensing data in the collected data belongs. Among them, the imaging category at least includes: central projection imaging, multi-central projection imaging, and area array imaging.

[0057] Specifically, central projection imaging (such as aerial remote sensing) is an image of central projection. For an image of central projection, image point displacement is its main source of geometric deformation and is related to the flight altitude and the distance of the image point from the principal point.

[0058] The geometric deformation characteristics of multi-central projection imaging (such as a linear array pushbroom remote sensing satellite) are different in the flight direction and the scanning direction.

[0059] Area array imaging (such as some high-resolution satellites) is mainly geometric deformation caused by systematic factors (such as satellite orbit and attitude changes).

[0060] S2222: Select the corresponding imaging calibration model according to the imaging category, and use the corresponding imaging calibration model and ground control points to calibrate the optical remote sensing data in the collected data, and output the initial imaging calibration data. Among them, the imaging calibration model at least includes: polynomial calibration model, physical model, and rational function model.

[0061] Further, as an embodiment, traverse the imaging calibration model database according to the imaging category, and determine the standard imaging calibration model whose standard imaging category is consistent with the imaging category as the corresponding imaging calibration model.

[0062] Specifically, the corresponding imaging calibration model is an imaging calibration module or a combination of multiple imaging calibration models.

[0063] For example: When the imaging category is central projection imaging, the selected imaging calibration model is the polynomial calibration model, but not limited to the polynomial calibration model. In this application, the polynomial calibration model is preferably used, which can fit complex deformation curves and can correct geometric deformations caused by terrain undulation and camera attitude changes.

[0064] When the imaging category is multi - center projection imaging, the selected imaging correction model is a physical model, but not limited to the physical model. It can also be a combination of a physical model and a polynomial correction model for correction. This application preferably combines a physical model and a polynomial correction model. The physical model is used to consider prior information such as satellite orbit parameters and attitude parameters and describe the geometric relationship in the image imaging process; the polynomial correction model is used to compensate for the remaining systematic errors and random errors.

[0065] When the imaging category is area - array imaging, the selected imaging correction model is a strict physical model or a rational function model, but not limited to a strict physical model or a rational function model. This application preferably uses a rational function model, which can construct the imaging geometric relationship according to the accurate parameters of the satellite orbit and attitude.

[0066] Furthermore, the sub - steps of using the corresponding imaging correction model and ground control points to correct the optical remote - sensing data in the collected data and output the initial imaging correction data are as follows:

[0067] S22221: Calculate the imaging correction parameters using the imaging correction model and ground control points.

[0068] Specifically, substitute the object - space coordinates of the ground control points (i.e., the accurate three - dimensional space coordinates on the actual ground) and the pixel - point coordinates corresponding to the ground control points in the optical remote - sensing data of the collected data into the imaging correction model, and calculate the imaging correction parameters by the imaging correction model.

[0069] S22222: Correct the optical remote - sensing data in the collected data according to the imaging correction parameters to obtain the initial imaging correction data.

[0070] Specifically, convert the coordinates in the optical remote - sensing data of the collected data into corrected coordinates through the imaging correction parameters, thereby obtaining the initial imaging correction data.

[0071] S2223: Conduct accuracy evaluation on the initial imaging correction data to obtain the first accuracy evaluation result. If the first accuracy evaluation result is qualified, use the initial imaging correction data as the imaging correction data; if the first accuracy evaluation result is unqualified, re - perform the imaging correction process on the optical remote - sensing data in the collected data.

[0072] Furthermore, the sub - steps of conducting accuracy evaluation on the initial imaging correction data to obtain the first accuracy evaluation result are as follows:

[0073] S22231: Calculate the residuals between each ground control point and the corresponding corrected coordinates in the initial imaging correction data.

[0074] Furthermore, the expression of the residual is:

[0075] ΔXc n = Xjz n - Xcs n ;

[0076] ΔYc n = Yjz n - Ycs n ;

[0077] where ΔXc n is the residual in the x - direction between the nth ground control point and the corresponding corrected coordinate in the initial imaging correction data; ΔYc n is the residual in the y - direction between the nth ground control point and the corresponding corrected coordinate in the initial imaging correction data; (Xjz n , Yjz n ) is the corrected coordinate corresponding to the nth ground control point in the initial imaging correction data; (Xcs n , Ycs n ) is the object - space coordinate of the nth ground control point; n ∈ [1, N], where N is the total number of ground control points and N is a natural number.

[0078] S22232: Perform statistical analysis on all the residuals to obtain the imaging comprehensive error value.

[0079] Furthermore, the expression of the imaging comprehensive error value is:

[0080]

[0081] where Wczc is the imaging comprehensive error value; ΔXc n is the residual in the x - direction between the nth ground control point and the corresponding corrected coordinate in the initial imaging correction data; ΔYc n is the residual in the y - direction between the nth ground control point and the corresponding corrected coordinate in the initial imaging correction data; n ∈ [1, N], where N is the total number of ground control points and N is a natural number.

[0082] S22233: Analyze the imaging comprehensive error value using the imaging accuracy threshold to generate the first accuracy evaluation result; if the imaging comprehensive error value is less than the imaging accuracy threshold, the generated first accuracy evaluation result is qualified; if the imaging comprehensive error value is greater than or equal to the imaging accuracy threshold, the generated first accuracy evaluation result is unqualified.

[0083] Specifically, the smaller the imaging comprehensive error value, the higher the overall correction accuracy.

[0084] Furthermore, the specific value of the imaging accuracy threshold is set according to the actual situation.

[0085] S223: Use ground control points to perform deformation correction processing on the imaging correction data to obtain deformed correction data, and use the deformed correction data as the first correction data.

[0086] Furthermore, the sub-steps of using ground control points to perform deformation correction processing on the imaging correction data to obtain deformed correction data and using the deformed correction data as the first correction data are as follows:

[0087] S2231: Analyze the imaging correction data using a second classification model based on artificial intelligence to determine the geometric deformation category to which the imaging correction data belongs; wherein, the geometric deformation category at least includes: topographic geometric deformation, systematic geometric deformation, and non-systematic geometric deformation.

[0088] Specifically, topographic geometric deformation is the geometric deformation caused by the projection difference of the image due to the terrain in the acquisition area with large terrain undulations.

[0089] Systematic geometric deformation is the geometric deformation caused by unstable sensor attitude (such as roll, pitch, yaw) or orbital parameter changes.

[0090] Non-systematic geometric deformation is the non-systematic deformation caused by complex factors such as atmospheric refraction and earth rotation.

[0091] S2232: Select a corresponding deformation correction model according to the geometric deformation category, and use the corresponding deformation correction model and ground control points to correct the imaging correction data, and output the initial deformed correction data, wherein the deformation correction model at least includes: orthorectification model, sensor-based physical correction model, polynomial correction model, and correction model with compensation terms added on the basis of a strict physical model.

[0092] Furthermore, as an embodiment, traverse the deformation correction model database according to the geometric deformation category, and determine the standard deformation correction model whose standard geometric deformation category is consistent with the geometric deformation category as the corresponding deformation correction model.

[0093] For example: when the geometric deformation category is topographic geometric deformation, the selected deformation correction model is the orthorectification model, but not limited to the orthorectification model. This application preferably uses the orthorectification model, which is based on the digital elevation model (DEM), and eliminates the displacement caused by terrain undulations by projecting the image points onto the reference plane according to the terrain undulations.

[0094] When the geometric deformation category is systematic geometric deformation, the selected deformation correction model is a sensor-based physical correction model, but not limited to a sensor-based physical correction model. Preferably, the present application uses a sensor-based physical correction model, which can construct a strict imaging geometric relationship based on the internal parameters of the sensor (such as focal length, pixel size) and external parameters (satellite position, attitude) to accurately correct the image.

[0095] When the geometric deformation category is non-systematic geometric deformation, the selected deformation correction model is a polynomial correction model or a correction model with a compensation term added to a strict physical model, but not limited to a polynomial correction model or a correction model with a compensation term added to a strict physical model. Preferably, the present application uses a polynomial correction model, which can fit a complex deformation curve. By selecting a sufficient number of control points on the image and determining the polynomial coefficients, the deformation can be corrected.

[0096] Specifically, select a suitable correction model according to the imaging method and geometric deformation characteristics of the remote sensing image, and use the selected correction model to correct the optical remote sensing data in the collected data multiple times to obtain first correction data, which can improve the correction accuracy and thus improve the data quality.

[0097] Further, the sub-steps of correcting the imaging correction data using the corresponding deformation correction model and ground control points and outputting the initial deformation correction data are as follows:

[0098] S22321: Calculate the deformation correction parameters using the deformation correction model and ground control points.

[0099] Specifically, substitute the object space coordinates of the ground control point (i.e., the precise three-dimensional space coordinates on the actual ground) and the pixel point coordinates corresponding to the ground control point in the imaging correction data into the deformation correction model, and calculate the deformation correction parameters by the deformation correction model.

[0100] S22322: Correct the imaging correction data according to the deformation correction parameters to obtain the initial deformation correction data.

[0101] Specifically, convert the coordinates in the imaging correction data to the corrected coordinates through the deformation correction parameters to obtain the initial deformation correction data.

[0102] S2233: Evaluate the accuracy of the initial deformation correction data to obtain a second accuracy evaluation result. If the second accuracy evaluation result is qualified, use the initial deformation correction data as the deformation correction data and use the deformation correction data as the first correction data; if the second accuracy evaluation result is unqualified, re-perform the deformation correction process on the imaging correction data.

[0103] Further, the sub-steps for performing accuracy evaluation on the initial deformation correction data to obtain the second accuracy evaluation result are as follows:

[0104] S22331: Analyze each ground control point and the corresponding corrected coordinates in the initial deformation correction data to obtain the first deformation error value.

[0105] Further, the expression for the first deformation error value is:

[0106]

[0107] where Wczb is the first deformation error value; (Xcs n , Ycs n ) is the object space coordinate of the nth ground control point; n ∈ [1, N], N is the total number of ground control points, and N is a natural number; (Xxz n , Yxz n ) is the corresponding corrected coordinate in the initial deformation correction data for the nth ground control point.

[0108] S22332: Select multiple uniformly distributed ground check points on the original optical remote sensing data in the target data, and analyze each ground check point and the corresponding corrected coordinates in the initial deformation correction data to obtain the second deformation error value, where the ground check points are different from the ground control points.

[0109] Specifically, the ground check points are the points in the target data other than the ground control points. The corresponding positions of the ground check points in the initial deformation correction data are uniformly distributed, which can comprehensively evaluate the calibration accuracy of different positions of the image.

[0110] Further, the expression for the second deformation error value is:

[0111]

[0112] where Wmae is the second deformation error value; (Xjy m , Yjy m ) is the object space coordinate of the mth ground check point; m ∈ [1, M], M is the total number of ground check points, and M is a natural number; (Xxz m , Yxz m ) is the corresponding corrected coordinate in the initial deformation correction data for the mth ground check point.

[0113] S22333: Obtain the comprehensive deformation error value based on the first deformation error value and the second deformation error value, and analyze the comprehensive deformation error value using the deformation accuracy threshold to generate the second accuracy evaluation result; if the comprehensive deformation error value is less than the deformation accuracy threshold, the generated second accuracy evaluation result is qualified; if the comprehensive deformation error value is greater than or equal to the deformation accuracy threshold, the generated second accuracy evaluation result is unqualified.

[0114] Specifically, the smaller the comprehensive deformation error value, the higher the overall calibration accuracy. The specific value of the deformation accuracy threshold is set according to the actual situation.

[0115] Furthermore, the expression of the comprehensive deformation error value is:

[0116]

[0117] Among them, Zhc is the comprehensive deformation error value; Wczb is the first deformation error value; Wmae is the second deformation error value; η1 is the weight of the first deformation error value; η2 is the weight of the second deformation error value.

[0118] Specifically, the specific values of η1 and η2 are set according to the actual situation.

[0119] S23: Perform filtering processing on the synthetic aperture radar data in the collected data to obtain the second calibration data.

[0120] Specifically, use the existing technology to perform filtering processing on the synthetic aperture radar data in the collected data to remove noise and obtain the second calibration data. For example, use multi-look processing to reduce the speckle noise of the synthetic aperture radar data in the collected data. The second calibration data is a high-quality remote sensing image after calibration. The radiation value of the image of the second calibration data can truly reflect the reflection characteristics of the ground objects, the geometric position is accurate, and the image noise of the second calibration data is suppressed to a preset degree, which is convenient for subsequent accurate analysis.

[0121] S24: Use the collection code, collection time, the first calibration data, and the second calibration data as the preprocessed data.

[0122] Specifically, the artificial intelligence-based image processing method and system of the present application can calibrate the data to be processed, thereby improving the data quality. Due to the high data quality, the accuracy of the disaster hazard analysis result can be further improved.

[0123] S3: Obtain the target mining dataset according to the acquisition code, analyze the preprocessed data through the target mining dataset to obtain the current difference value, and analyze the current difference value. If the current difference value is within the normal range, execute S1; if the current difference value is within the critical range, adjust the initial acquisition frequency to obtain a new acquisition frequency, obtain the data to be processed according to the new acquisition frequency, and execute S2; if the current difference value is within the abnormal range, send an alarm message.

[0124] Specifically, the specific value of the normal range is set according to the actual situation. If the current difference value is within the normal range, it means there is no hidden danger of disaster.

[0125] The specific value of the critical range is set according to the actual situation. If the current difference value is within the critical range, it means that there is no hidden danger of disaster yet, but the probability of an impending disaster is very high.

[0126] The specific value of the abnormal range is set according to the actual situation. If the current difference value is within the abnormal range, it means that there is already a hidden danger of disaster. When the current difference value is within the abnormal range and an alarm message is sent, relevant departments and personnel can timely handle the hidden danger of disaster.

[0127] Furthermore, the alarm message at least includes: the alarm time, the category of the hidden danger of disaster, and the degree of the hidden danger of disaster.

[0128] Furthermore, the new acquisition frequency is greater than the initial acquisition frequency, that is: within the same time period, the number of data to be processed obtained by the new acquisition frequency is greater than the number of data to be processed obtained by the initial acquisition frequency.

[0129] Specifically, the specific value of the new acquisition frequency is set according to the actual situation. Within the same time period, the greater the acquisition frequency, the more acquisition times, and the more data to be processed obtained. When the current difference value is within the critical range, adjust the initial acquisition frequency to a new acquisition frequency with a greater frequency, and obtain more data to be processed according to the new acquisition frequency, which can timely analyze and warn of the hidden danger of disaster.

[0130] Furthermore, the sub-steps of obtaining the current difference value by analyzing the preprocessed data through the target mining dataset according to the acquisition code are as follows:

[0131] S31: Traverse the standard database according to the acquisition code, and use the mining dataset in the standard data packet whose query code is consistent with the acquisition code as the comparison dataset. Among them, the comparison dataset includes: data of multiple mining periods; each mining period data includes: the acquisition time after mining, the optical remote sensing data after mining, and the synthetic aperture radar data after mining.

[0132] S32: Sort the data during the mining period and the preprocessed data according to the multiple post-mining collection times and the order from the earliest to the latest collection time, and generate corresponding analysis serial numbers, thereby obtaining the data to be analyzed. Among them, each piece of data to be analyzed includes: analysis serial number, analysis collection time, analyzed optical remote sensing data, and analyzed synthetic aperture radar data. The analysis serial numbers increase sequentially according to the multiple post-mining collection times and the order from the earliest to the latest collection time.

[0133] Specifically, if the total number of the data during the mining period and the preprocessed data is K + 1, then the analysis serial number of the preprocessed data is K + 1, where K is a natural number.

[0134] The collection time of the preprocessed data after generating the corresponding analysis serial number is the analysis collection time, the first corrected data is the analyzed optical remote sensing data, and the second corrected data is the analyzed synthetic aperture radar data.

[0135] The post-mining collection time of the data during the mining period after generating the corresponding analysis serial number is the analysis collection time, the post-mining optical remote sensing data is the analyzed optical remote sensing data, and the post-mining synthetic aperture radar data is the analyzed synthetic aperture radar data.

[0136] S33: Analyze the data to be analyzed to obtain the current difference value.

[0137] Furthermore, the expression of the current difference value is:

[0138]

[0139] Among them, Dcy is the current difference value; λ1 is the pixel difference weight; ΔCxs k is the average value of the differences of all corresponding pixels between the analyzed optical remote sensing data in the (k + 1)-th data to be analyzed and the analyzed optical remote sensing data in the k-th data to be analyzed; ΔCxs K is the average value of the differences of all corresponding pixels between the analyzed optical remote sensing data in the (K + 1)-th data to be analyzed and the analyzed optical remote sensing data in the K-th data to be analyzed; K + 1 is the total number of the data to be analyzed, and K is a natural number; λ2 is the disaster similarity weight; Gzh K+1 is the similarity value between the analyzed synthetic aperture radar data in the (K + 1)-th data to be analyzed and the pre-constructed geological disaster feature set.

[0140] Specifically, λ1 and λ2 are set according to the actual situation. The similarity value between the synthetic aperture radar data in the (K + 1)-th data to be analyzed and the pre-constructed geological disaster feature set can be obtained through an artificial intelligence-based analysis model. The greater the similarity value, the greater the probability of the occurrence of such geological disasters. The geological disaster feature set is a set composed of features that can clearly express geological disasters extracted from actual collected or studied data, etc. The average value of the differences between all corresponding pixels in the analyzed optical remote sensing data in two adjacent data to be analyzed in terms of acquisition time can be achieved according to the prior art, so it will not be elaborated here.

[0141] The beneficial effects achieved by this application are as follows:

[0142] (1) The artificial intelligence-based image processing method and system of this application can correct the data to be processed, thereby improving the data quality.

[0143] (2) The artificial intelligence-based image processing method and system of this application can improve the accuracy of the analysis results of disaster hidden dangers and give timely warnings of disaster hidden dangers.

[0144] Although the preferred embodiments of this application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concept. Therefore, the protection scope of this application is intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of this application. Obviously, those skilled in the art can make various changes and variations to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application belong to the scope of the protection of this application and its equivalent technologies, this application also intends to include these modifications and variations.

Claims

1. An image processing method based on artificial intelligence, characterized in that, It includes the following steps: S1: Obtain the data to be processed according to the initial acquisition frequency. The data to be processed includes: acquisition code, acquisition time, and acquisition data. The acquisition data includes at least: optical remote sensing data and synthetic aperture radar data; S2: Preprocess the data to be processed to obtain preprocessed data. The preprocessed data includes at least: acquisition code, acquisition time, first correction data, and second correction data; S3: Obtain the target mining dataset according to the acquisition code, analyze the preprocessed data through the target mining dataset to obtain the current difference value, and analyze the current difference value. If the current difference value is within the normal range, execute S1; if the current difference value is within the critical range, adjust the initial acquisition frequency to obtain a new acquisition frequency, obtain the data to be processed according to the new acquisition frequency, and execute S2; if the current difference value is within the abnormal range, send an alarm message.

2. The image processing method based on artificial intelligence according to claim 1, wherein The sub-steps of preprocessing the data to be processed to obtain preprocessed data are as follows: S21: Query the standard database according to the acquisition code, and use the pre-mining data in the standard data packet with the query code consistent with the acquisition code as the target data; S22: Use the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data; S23: Filter the synthetic aperture radar data in the acquisition data to obtain the second correction data; S24: Use the acquisition code, acquisition time, first correction data, and second correction data as the preprocessed data.

3. The image processing method based on artificial intelligence according to claim 2, characterized in that, The sub-steps of using the target data to correct the optical remote sensing data in the acquisition data to obtain the first correction data are as follows: S221: Select multiple evenly distributed ground control points on the original optical remote sensing data in the target data; S222: Use the ground control points to perform imaging correction processing on the optical remote sensing data in the acquisition data to obtain imaging correction data; S223: Use the ground control points to perform deformation correction processing on the imaging correction data to obtain deformation correction data, and use the deformation correction data as the first correction data.

4. The image processing method based on artificial intelligence according to claim 3, wherein The sub-steps of using the ground control points to perform imaging correction processing on the optical remote sensing data in the acquisition data to obtain imaging correction data are as follows: S2221: Use the first classification model based on artificial intelligence to analyze the optical remote sensing data in the acquisition data to determine the imaging category to which the optical remote sensing data in the acquisition data belongs. The imaging categories include at least: central projection imaging, multi-central projection imaging, and area array imaging; S2222: Select the corresponding imaging correction model according to the imaging category, and use the corresponding imaging correction model and ground control points to correct the optical remote sensing data in the acquisition data, and output the initial imaging correction data. The imaging correction models include at least: polynomial correction model, physical model, and rational function model; S2223: Evaluate the accuracy of the initial imaging correction data to obtain the first accuracy evaluation result. If the first accuracy evaluation result is qualified, use the initial imaging correction data as the imaging correction data; if the first accuracy evaluation result is unqualified, re-perform imaging correction processing on the optical remote sensing data in the acquisition data.

5. The image processing method based on artificial intelligence according to claim 4, wherein The sub-steps for performing accuracy evaluation on the initial imaging correction data to obtain the first accuracy evaluation result are as follows: S22231: Calculate the residuals between each ground control point and the corresponding corrected coordinates in the initial imaging correction data; S22232: Conduct statistical analysis on all the residuals to obtain the imaging comprehensive error value; S22233: Analyze the imaging comprehensive error value using the imaging accuracy threshold to generate the first accuracy evaluation result; if the imaging comprehensive error value is less than the imaging accuracy threshold, the generated first accuracy evaluation result is qualified; if the imaging comprehensive error value is greater than or equal to the imaging accuracy threshold, the generated first accuracy evaluation result is unqualified.

6. The image processing method based on artificial intelligence according to claim 5, wherein, The sub-steps for performing deformation correction processing on the imaging correction data using the ground control points to obtain the deformation correction data and taking the deformation correction data as the first correction data are as follows: S2231: Analyze the imaging correction data using the second classification model based on artificial intelligence to determine the geometric deformation category to which the imaging correction data belongs; among them, the geometric deformation category at least includes: topographic geometric deformation, systematic geometric deformation, and non-systematic geometric deformation; S2232: Select the corresponding deformation correction model according to the geometric deformation category, and use the corresponding deformation correction model and the ground control points to correct the imaging correction data, and output the initial deformation correction data, where the deformation correction model at least includes: orthorectification model, physical correction model based on the sensor, polynomial correction model, correction model with a compensation term added on the basis of the strict physical model; S2233: Perform accuracy evaluation on the initial deformation correction data to obtain the second accuracy evaluation result. If the second accuracy evaluation result is qualified, take the initial deformation correction data as the deformation correction data and take the deformation correction data as the first correction data; if the second accuracy evaluation result is unqualified, re-perform deformation correction processing on the imaging correction data.

7. The artificial intelligence-based image processing method according to claim 6, wherein The sub-steps for performing accuracy evaluation on the initial deformation correction data to obtain the second accuracy evaluation result are as follows: S22331: Analyze each ground control point and the corresponding corrected coordinates in the initial deformation correction data to obtain the first deformation error value; S22332: Select multiple uniformly distributed ground inspection points on the original optical remote sensing data in the target data, and analyze each ground inspection point and the corresponding corrected coordinates in the initial deformation correction data to obtain the second deformation error value, where the ground inspection points are different from the ground control points; S22333: Obtain the deformation comprehensive error value based on the first deformation error value and the second deformation error value, and analyze the deformation comprehensive error value using the deformation accuracy threshold to generate the second accuracy evaluation result; if the deformation comprehensive error value is less than the deformation accuracy threshold, the generated second accuracy evaluation result is qualified; if the deformation comprehensive error value is greater than or equal to the deformation accuracy threshold, the generated second accuracy evaluation result is unqualified.

8. The image processing method based on artificial intelligence according to claim 7, wherein The sub-steps for obtaining the target mining dataset according to the acquisition code and analyzing the preprocessed data through the target mining dataset to obtain the current difference value are as follows: S31: Traverse the standard database according to the acquisition code, and use the mining dataset in the standard data packet whose query code is consistent with the acquisition code as the comparison dataset. The comparison dataset includes: data of multiple mining periods; each mining period data includes: post-mining acquisition time, post-mining optical remote sensing data, and post-mining synthetic aperture radar data. S32: Sort the mining period data and the preprocessed data according to the order of multiple post-mining acquisition times and acquisition times from first to last, and generate corresponding analysis serial numbers to obtain the data to be analyzed. Each data to be analyzed includes: analysis serial number, analysis acquisition time, analysis optical remote sensing data, and analysis synthetic aperture radar data; the analysis serial numbers increase sequentially according to the order of multiple post-mining acquisition times and acquisition times from first to last. S33: Analyze the data to be analyzed to obtain the current difference value.

9. The image processing method based on artificial intelligence according to claim 8, wherein, The expression of the current difference value is: Among them, Dcy is the current difference value; λ1 is the pixel difference weight; ΔCxs k is the average value of the differences of all corresponding pixels between the analyzed optical remote sensing data in the (k + 1)-th data to be analyzed and the analyzed optical remote sensing data in the k-th data to be analyzed; ΔCxs K is the average value of the differences of all corresponding pixels between the analyzed optical remote sensing data in the (K + 1)-th data to be analyzed and the analyzed optical remote sensing data in the K-th data to be analyzed; K + 1 is the total number of data to be analyzed, and K is a natural number; λ2 is the disaster similarity weight; Gzh K+1 is the similarity value between the analyzed synthetic aperture radar data in the (K + 1)-th data to be analyzed and the pre-constructed geological disaster feature set.

10. An image processing system based on artificial intelligence, characterized in that, Including: At least one data acquisition node and an artificial intelligence image processing center; Among them, the data acquisition node: acquires the data to be processed according to the initial acquisition frequency or the new acquisition frequency, and sends the data to be processed to the artificial intelligence image processing center; receives the alarm information. The artificial intelligence image processing center: is used to execute the artificial intelligence-based image processing method described in any one of claims 1-9.

Images