AI-based real estate surveying optimization method and system

Abstract

The present application relates to the technical field of intelligent surveying and mapping, in particular to a real estate surveying and mapping optimization method and system based on AI. In the present application, boundary error data is generated by quantifying the deviation between global navigation satellite system coordinates and cadastral vector data, the spatial features of the real estate subject boundary are deeply extracted using a convolutional neural network, and targeted boundary fitting correction is performed on the super-limit point position in combination with the error data, overcoming the passive acceptance and subjective judgment of errors in traditional data processing, realizing intelligent perception and adaptive optimization of the boundary geometric form, and subsequently using a Bayesian inference model, taking the corrected boundary as a prior distribution and fusing new observation data for probability updating, dynamically obtaining the boundary point position coordinates with minimized uncertainty, solving the problem that static data coverage cannot effectively fuse multi-source information, and then performing consistency screening on multi-temporal cadastral data based on the updated boundary.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent surveying and mapping technology, and in particular to an AI-based real estate surveying and mapping optimization method and system. BACKGROUND

[0002] The field of intelligent surveying and mapping technology mainly involves the use of modern information technology and automated equipment for the collection, processing and analysis of geographic spatial data, including but not limited to the collection of measurement data, the generation of maps, the modeling and visualization of spatial information, and the accurate determination and optimization of geographic data.

[0003] Among them, the real estate surveying and mapping optimization method refers to obtaining the spatial position, boundary and other data of real estate through measurement equipment during the real estate surveying and mapping process, and conducting surveying and mapping according to existing standards, and providing basis for real estate registration, valuation, management and other aspects through subsequent data processing and analysis.

[0004] The existing technology lacks an objective and automated conflict diagnosis and processing mechanism when the spatial position data obtained by field measurement does not conform to the existing cadastral map, and the evaluation and decision often rely on manual experience. This subjective judgment is easy to introduce new errors or cover up old mistakes. For example, a small but non-standard boundary point deviation may be simply attributed to measurement error and forced to be adjusted, resulting in a small inaccuracy in the area of the plot, which may cause ownership disputes over time. At the same time, the existing technology usually takes historical cadastral data as a static background reference, and only performs simple overlay comparison or direct replacement when updating data, without longitudinal analysis of multi-time phase data, which leads to the fact that some historical and systematic errors cannot be found and eliminated. For example, a systematic deviation of all boundary points in a certain area may occur due to a one-time coordinate conversion error many years ago. In subsequent surveying and mapping, if only the previous data is compared, the systematic error will be considered as normal and will continue to be passed on. In addition, data processing often takes a single real estate subject or isolated boundary point as the object, ignoring the spatial topological relationship between real estate units. When updating the boundary of a plot, the shared boundary of adjacent plots may not be adjusted synchronously, resulting in gaps or overlaps between plots in the database, thereby damaging the overall integrity of the cadastral map. SUMMARY

[0005] The purpose of the present application is to solve the shortcomings in the prior art and to provide an AI-based real estate surveying and mapping optimization method and system.

[0006] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme: an AI-based real estate surveying and mapping optimization method, comprising the following steps:

[0007] S1: GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary are acquired, a plane offset value and a direction angle deviation of the real estate subject boundary point position are calculated, and boundary error data are generated;

[0008] S2: The GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary are input into a convolutional neural network model, the spatial features of the real estate subject boundary are extracted, and boundary fitting correction is performed according to the boundary error data, to generate boundary correction data;

[0009] S3: Prior distribution of the position observation equation of the real estate subject boundary point position is established through the boundary correction data, observation likelihood of the real estate subject boundary point position is set through a Bayesian inference model, and real estate subject boundary point coordinate is updated according to the prior distribution and the observation likelihood, to obtain boundary position update data;

[0010] S4: Based on the boundary position update data, the multi-temporal cadastral data of the real estate subject are screened for consistency, and a consistency screening result is generated;

[0011] S5: Based on the consistency screening result, the real estate subject boundary point position and the real estate subject boundary position parameter are compared for spatial fusion, the real estate subject polygon data are updated, and real estate surveying and mapping optimization result is obtained.

[0012] As a further scheme of the present application, the boundary error data includes a plane offset of the real estate subject boundary point position, a direction angle deviation of the real estate subject boundary point position, and a coordinate residual of the real estate subject boundary point position, the boundary correction data includes a curve fitting parameter of the real estate subject boundary line segment, a corrected coordinate of the real estate subject boundary point position, and a direction difference of the real estate subject boundary line segment, the boundary position update data includes a posteriori coordinate of the real estate subject boundary point position, a posteriori probability value of the real estate subject boundary point position, and a coordinate correction amplitude of the real estate subject boundary point position, the consistency screening result includes a time series difference of the real estate subject boundary point position, a classification identification of the real estate subject boundary line segment, and a screening state of the real estate subject boundary point position, and the real estate surveying and mapping optimization result includes updated real estate subject polygon data, fused coordinate set of the real estate subject boundary, and spatial topology structure of the real estate subject.

[0013] As a further scheme of the present application, the boundary error data acquisition step specifically includes:

[0014] S111: GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary are acquired, the GNSS coordinate data are uniformly projected and converted according to the national geodetic coordinate system, the real estate subject boundary point positions in the cadastral vector boundary data are matched with the corresponding boundary point positions in the converted GNSS coordinate data point by point, and a matched point position coordinate set is established;

[0015] S112: calling the matching point position coordinate set, performing difference operation on GNSS coordinates and cadastral vector coordinates for each real estate main body boundary point position, respectively acquiring the plane offset value and the direction angle deviation of the real estate main body boundary point position, and generating point position offset quantization value;

[0016] S113: performing root mean square statistics on the plane offset value and the direction angle deviation set of all real estate main body boundary point positions, taking the statistical result as the observation noise level, and combining with the point position offset quantization value to obtain boundary error data.

[0017] As a further scheme of the present application, the boundary correction data acquisition step is specifically:

[0018] S211: inputting the GNSS coordinate data and the cadastral vector boundary data of the real estate main body boundary into a convolutional neural network model, judging the continuity of the real estate main body boundary line segment according to the boundary gradient response value output by the convolutional layer of the convolutional neural network model, and acquiring the spatial feature of the real estate main body boundary;

[0019] S212: screening the real estate main body boundary point positions whose plane offset value exceeds the set offset threshold value in the boundary error data, and performing curve fitting operation on the screened point positions in combination with the spatial feature of the real estate main body boundary, calculating the boundary fitting correction parameter according to the comparison result of the fitting curve direction difference and the original boundary direction difference;

[0020] S213: correcting the boundary fitting of the real estate main body boundary point positions whose plane offset value exceeds the set offset threshold value according to the boundary fitting correction parameter, and integrating the corrected real estate main body boundary point positions and the cadastral vector boundary data to generate boundary correction data.

[0021] As a further scheme of the present application, the boundary fitting correction parameter is calculated according to the comparison result of the fitting curve direction difference and the original boundary direction difference, specifically:

[0022] acquiring the fitting curve direction difference;

[0023] acquiring the original boundary direction difference;

[0024] calculating the angle deviation value between the fitting curve direction difference and the original boundary direction difference;

[0025] calling the boundary error data to acquire the plane offset value corresponding to the real estate main body boundary point position;

[0026] setting an angle deviation weight coefficient;

[0027] setting a plane offset weight coefficient;

[0028] The angle deviation value is multiplied by the angle deviation weight coefficient to obtain an angle correction amount;

[0029] The plane offset value is multiplied by the plane offset weight coefficient to obtain an offset correction amount;

[0030] The angle correction amount and the offset correction amount are substituted into a preset correction function, and a weighted sum operation is performed on the angle correction amount and the offset correction amount to generate the boundary fitting correction parameter.

[0031] As a further scheme of the present application, the boundary position updating data obtaining step is specifically:

[0032] S311: The boundary correction data is set as a prior mean value of a real estate subject boundary point position observation equation, and a root mean square statistical result in the boundary error data is obtained and set as a prior variance estimation value of the real estate subject boundary point position observation equation, the prior mean value and the prior variance estimation value are combined to establish a real estate subject boundary point prior distribution;

[0033] S312: GNSS coordinate data of the real estate subject boundary is obtained, the GNSS coordinate data of the real estate subject boundary is taken as an observation quantity through a Bayesian inference model, and a square value of the root mean square statistical result in the boundary error data is set as a variance of an observation value to generate a point observation likelihood distribution;

[0034] S313: According to the real estate subject boundary point prior distribution and the point observation likelihood distribution, a posterior coordinate is calculated at each real estate subject boundary point, and a coordinate value corresponding to a maximum of the posterior coordinate is retrieved as an updated real estate subject boundary point coordinate to generate boundary position updating data.

[0035] As a further scheme of the present application, the consistency screening result obtaining step is specifically:

[0036] S411: Multi-temporal cadastral data formed by surveying and mapping of the real estate subject in multiple periods is obtained, cadastral data coordinates of the same-name real estate subject boundary points in the multi-temporal cadastral data are point-by-point executed time series difference calculation based on the boundary position updating data, and difference results of all real estate subject boundary points are summarized to generate time series difference values;

[0037] S412: The time series difference values are executed difference absolute median statistics, and the obtained national surveying and mapping accuracy standard values are compared item by item to establish a point deviation discrimination set;

[0038] S413: According to the point deviation discrimination set, binary classification is performed on all real estate main body boundary points in the multi-temporal cadastral data, the points are divided into two categories of consistent and inconsistent, and the consistent screening result is generated by integrating the boundary position update data.

[0039] As a further scheme of the present application, the acquisition of the real estate surveying and mapping optimization result specifically comprises:

[0040] S511: Based on the consistent screening result, consistent real estate main body boundary points are selected, and the real estate main body boundary position parameters in the boundary position update data are called for spatial fusion comparison and matching. For adjacent real estate main body boundary segments, the direction parameters and curvature parameters in the boundary correction data are called for line segment relocation to obtain a set of relocated boundary points.

[0041] S512: The set of relocated boundary points is called, and an updated coordinate point set in the boundary position update data is acquired. The Hausdorff distance between the two is calculated, the distance is taken as the matching degree, the matching degree is compared with the set matching threshold range, the real estate main body boundary points are screened, and a set of matched boundary points is obtained.

[0042] S513: According to the set of matched boundary points, the polygon data of the updated real estate main body is formed, and is input into the real estate surveying and mapping database to obtain the real estate surveying and mapping optimization result.

[0043] The AI-based real estate surveying and mapping optimization system is used to execute the AI-based real estate surveying and mapping optimization method, and the system comprises:

[0044] A boundary error calculation module acquires GNSS coordinate data and cadastral vector boundary data of a real estate main body boundary, calculates the planar offset value and direction angle deviation of the real estate main body boundary points, and generates boundary error data.

[0045] A boundary feature correction module inputs the GNSS coordinate data and cadastral vector boundary data of the real estate main body boundary into a convolutional neural network model, extracts the spatial features of the real estate main body boundary, and performs boundary fitting correction according to the boundary error data to generate boundary correction data.

[0046] A boundary position update module establishes a prior distribution of a position observation equation of the real estate main body boundary points through the boundary correction data, sets an observation likelihood of the real estate main body boundary points through a Bayesian inference model, updates the coordinates of the real estate main body boundary points according to the prior distribution and the observation likelihood, and obtains boundary position update data.

[0047] The cadastral consistency screening module screens the multi-time-phase cadastral data of the real estate subject based on the boundary position update data, and generates a consistency screening result;

[0048] The plot space fusion module compares the real estate subject boundary point position and the real estate subject boundary position parameter based on the consistency screening result, updates the plot data of the real estate subject, and obtains a real estate surveying and mapping optimization result.

[0049] Compared with the prior art, the advantages and positive effects of the present application are that:

[0050] In the present application, the boundary error data is generated by quantifying the deviation between the global navigation satellite system coordinates and the cadastral vector data, the spatial features of the real estate subject boundary are deeply extracted by using the convolutional neural network, and the boundary fitting correction is performed on the out-of-limit point position in combination with the error data, which overcomes the passive acceptance and subjective judgment of errors in traditional data processing, realizes the intelligent perception and adaptive optimization of the boundary geometric form, and subsequently uses the Bayesian inference model to take the corrected boundary as the prior distribution and fuse the new observation data to update the probability, dynamically obtains the boundary point position coordinates with minimized uncertainty, solves the problem that the static data coverage cannot effectively fuse multi-source information, and then performs consistency screening on the multi-time-phase cadastral data based on the updated boundary, identifies and classifies the unstable point positions in the historical data through time series difference statistics, effectively excavates and processes the long-existing but neglected surveying and mapping errors or real changes, and finally performs spatial fusion and line segment relocation on the screened stable boundary point positions, ensures the uniqueness and topological correctness of the boundary between adjacent real estate subjects, fundamentally avoids the plot gap or overlap problem caused by isolated processing point positions, and finally forms a real estate surveying and mapping optimization result with high precision, time sequence consistency and topological correctness. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 It is a workflow schematic diagram of the present application;

[0052] Figure 2 It is a flowchart of step S1 of the present application;

[0053] Figure 3 It is a flowchart of step S2 of the present application;

[0054] Figure 4 It is a flowchart of step S3 of the present application;

[0055] Figure 5 It is a flowchart of step S4 of the present application;

[0056] Figure 6 It is a flowchart of step S5 of the present application. DETAILED DESCRIPTION

[0057] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0058] Please refer to Figure 1 The present application provides a technical solution: an AI-based real estate surveying and mapping optimization method, comprising the following steps:

[0059] S1: obtaining GNSS coordinate data and cadastral vector boundary data of the real estate main body boundary, calculating the plane offset value and direction angle deviation of the real estate main body boundary point position, and generating boundary error data;

[0060] S2: inputting the GNSS coordinate data and cadastral vector boundary data of the real estate main body boundary into a convolutional neural network model, extracting the spatial features of the real estate main body boundary, and performing boundary fitting correction according to the boundary error data to generate boundary correction data;

[0061] S3: establishing the prior distribution of the position observation equation of the real estate main body boundary point position through the boundary correction data, setting the observation likelihood of the real estate main body boundary point position through a Bayesian inference model, updating the coordinates of the real estate main body boundary point position according to the prior distribution and the observation likelihood, and obtaining boundary position update data;

[0062] S4: performing consistency screening on the multi-temporal cadastral data of the real estate main body based on the boundary position update data to generate consistency screening results;

[0063] S5: performing spatial fusion comparison on the real estate main body boundary point position and the real estate main body boundary position parameter based on the consistency screening results, updating the polygon data of the real estate main body, and obtaining real estate surveying and mapping optimization results;

[0064] The boundary error data includes the plane offset of the real estate main body boundary point position, the direction angle deviation of the real estate main body boundary point position, and the coordinate residual error of the real estate main body boundary point position. The boundary correction data includes the curve fitting parameters of the real estate main body boundary line segment, the corrected coordinates of the real estate main body boundary point position, and the direction difference of the real estate main body boundary line segment. The boundary position update data includes the posterior coordinates of the real estate main body boundary point position, the posterior probability value of the real estate main body boundary point position, and the coordinate correction amplitude of the real estate main body boundary point position. The consistency screening results include the time series difference of the real estate main body boundary point position, the classification identification of the real estate main body boundary line segment, and the screening state of the real estate main body boundary point position. The real estate surveying and mapping optimization results include the updated polygon data of the real estate main body, the fused coordinate set of the real estate main body boundary, and the spatial topology structure of the real estate main body.

[0065] Please refer to Figure 2The obtaining step of the boundary error data is specifically as follows:

[0066] S111: GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary are obtained, the GNSS coordinate data is projected and converted in the national geodetic coordinate system, and point-by-point matching is performed on the real estate subject boundary points in the cadastral vector boundary data and the corresponding boundary points in the converted GNSS coordinate data, and a matched point coordinate set is established;

[0067] The GNSS coordinate data of the real estate subject boundary is collected by using a real-time dynamic measurement (RTK) device on site, and the coordinate system thereof is the WGS-84 geodetic coordinate system; the cadastral vector boundary data is obtained from a real estate registration database of a cadastral management department, and the coordinate system thereof is the 2000 national geodetic coordinate system (CGCS2000). The GNSS coordinate data is projected and converted in the national geodetic coordinate system, and the specific execution process is as follows: a seven-parameter conversion model is called, all collected WGS-84 coordinate data is converted and projected through a series of calculations, and is unified into the CGCS2000 coordinate system to generate converted GNSS coordinate data. Point-by-point matching is performed on the real estate subject boundary points in the cadastral vector boundary data and the corresponding boundary points in the converted GNSS coordinate data, and the matching process is based on the unique identifier of the point, for example, a boundary point number. For example, for a real estate unit, the cadastral vector data thereof includes a boundary point numbered "JZD01", and the coordinates of the boundary point in the CGCS2000 coordinate system are (X: 452135.45 meters, Y: 345678.90 meters). In the field collection, an operator also uses an RTK device to measure the physically existing "JZD01" boundary point, obtains the WGS-84 coordinates thereof, and after the aforementioned projection conversion, obtains the coordinates of the boundary point in the CGCS2000 coordinate system as (X: 452135.52 meters, Y: 345678.98 meters). Through comparison of the common identifier "JZD01", the two coordinate data are confirmed as a matched point. The above identifier-based searching and pairing actions are repeated for all boundary points of the real estate unit, such as "JZD02", "JZD03" and the like, until the corresponding relationship between the cadastral coordinates and the GNSS coordinates of all points with the same identifier is established, and finally a matched point coordinate set is established.

[0068] S112: The matched point coordinate set is called, a difference operation is performed on the GNSS coordinate and the cadastral vector coordinate for each real estate subject boundary point, a planar offset value and a direction angle deviation of the real estate subject boundary point are obtained respectively, and a point offset quantization value is generated;

[0069] The GNSS coordinates and cadastral vector coordinates are subjected to difference operation for each real estate subject boundary point in the matching point coordinate set. Taking the boundary point numbered "JZD01" as an example, the cadastral vector coordinates thereof are (Xcadastral: 452135.45 meters, Ycadastral: 345678.90 meters), and the converted GNSS coordinates are (XGNSS: 452135.52 meters, YGNSS: 345678.98 meters). The specific operation process for obtaining the plan offset value of the real estate subject boundary point is as follows: first, the coordinate difference value in the X direction is calculated, that is, 452135.52 meters minus 452135.45 meters, to obtain 0.07 meters; second, the coordinate difference value in the Y direction is calculated, that is, 345678.98 meters minus 345678.90 meters, to obtain 0.08 meters; then, the square of the X direction coordinate difference value is added to the square of the Y direction coordinate difference value, that is, the square of 0.07 meters (0.0049 square meters) plus the square of 0.08 meters (0.0064 square meters), to obtain 0.0113 square meters; finally, the square root of the aforementioned sum value 0.0113 square meters is calculated, to obtain about 0.106 meters, which is the plan offset value of the "JZD01" point.

[0070] S113: The plan offset values and direction angle deviation sets of all real estate subject boundary points are subjected to root mean square statistics, the statistical result is taken as the observation noise level, and is combined with the point offset quantization value to obtain boundary error data;

[0071] The plan offset value set and the direction angle deviation set of all real estate subject boundary points are subjected to weighted root mean square statistics. The advantage of this kind of way is that, by introducing the weight reflecting the GNSS observation quality and the parameter, the model can more stably evaluate the overall observation noise level. The observation noise level is calculated according to the weighted root mean square formula with the introduced parameter : , wherein, : the observation noise level, the dimension is length (meters). : the index of the boundary point in the data set. : the total number of boundary points participating in the statistics. : the first The planar offset value of each point is calculated by step S112, representing the degree of geometric discrepancy between the cadastral coordinates and the GNSS coordinates of that point, and is measured in length (meters). : No. The weight corresponding to each point is a dimensionless coefficient. It is set based on the Point Precision Factor (PDOP) during GNSS data acquisition for that point. A smaller PDOP value indicates higher observation quality, and therefore a larger weight should be assigned. This is set here. . Regularization parameter, measured in area (square meters). It is used to prevent data loss due to variations in certain points. A value that is too small (close to 0) results in a low contribution to the overall noise level after squaring, thus improving the robustness of the model. Its value is set to half of the minimum allowable squared error within the region, based on historical experience data. Systematic error basis parameter, measured in area (square meters). It represents the square of the systematic error that cannot be eliminated, even under ideal conditions, due to factors such as coordinate transformation models and inherent equipment deviations. Its value is typically calibrated through repeated measurements and calculations of high-precision control points.

[0072] Suppose a real estate unit has acquired data from three boundary points (JZD01, JZD02, JZD03).

[0073] Planar offset value : rice, rice, Meters. PDOP value: , , Weight : , , .parameter Set to 0.001 square meters. Parameter : Set to 0.0005 square meters.

[0074] Substitute into the formula to calculate:

[0075] Calculate the numerator :

[0076] Point 1: ;

[0077] Point 2: ;

[0078] Point 3: ;

[0079] Total of numerators: ;

[0080] Compute the denominator :

[0081] Denominator total: ;

[0082] Compute the final observation noise level :

[0083] m.

[0084] This 0.112 meters is the weighted root mean square statistical result of the plane offset value. The same statistical method is used for the direction angle deviation set to obtain the statistical result of the direction angle deviation. The two statistical results are jointly used as the observation noise level, and are combined with the quantitative values of each point position offset generated in step S112 to obtain boundary error data.

[0085] See Figure 3 , the acquisition step of the boundary correction data is specifically:

[0086] S211: input the GNSS coordinate data of the real estate subject boundary and the cadastral vector boundary data into the convolutional neural network model, judge the continuity of the real estate subject boundary line segment according to the boundary gradient response value output by the convolutional layer of the convolutional neural network model, and obtain the spatial features of the real estate subject boundary;

[0087] The GNSS coordinate data of the real estate subject boundary and the cadastral vector boundary data are input into the convolutional neural network model. Before input, the discrete coordinate data needs to be converted into a two-dimensional raster image format. The specific process is as follows: a blank two-dimensional raster matrix covering the entire real estate subject area is created, and the size of the raster unit is set according to the required accuracy, for example, set to 5 cm x 5 cm. Then, the boundary line segments formed by the GNSS coordinate data and the boundary line segments formed by the cadastral vector data are respectively "drawn" onto two independent raster layers. The drawing process is as follows: for a line segment, traverse all the raster units on its path, and assign the pixel values of these units to 255, and the pixel values of other areas to 0. In this way, two binary images representing the GNSS boundary and the cadastral boundary are generated. These two images are stacked into a multi-channel input image as two channels, and then input into the convolutional neural network model. The structure of the convolutional neural network model is as follows: a first convolutional layer containing 16 3x3 pixel convolutional kernels for extracting low-level features of the boundary line segment, such as edges and directions; a rectified linear unit (ReLU) activation layer for introducing nonlinearity; a maximum pooling layer with a size of 2x2 pixels for reducing feature dimension and preserving the most significant features; a second convolutional layer containing 32 3x3 pixel convolutional kernels for combining low-level features to identify complex spatial patterns such as breakpoints and intersection angles. The continuity of the real estate subject boundary line segment is judged according to the boundary gradient response value output by the convolutional layer of the convolutional neural network model. The boundary gradient response value is the sum of the product of the weights of the convolution kernel and the corresponding pixel value when the convolution kernel slides on the image. In a continuous and clear boundary line segment area, the output value of the convolution kernel (such as an edge detection operator) will continuously be high (for example, the response value is greater than 200); if the boundary has breakpoints or is blurred, the pixel value at the corresponding position will not change sharply, and the response value of the convolution kernel will decrease significantly (for example, the response value is less than 50). By checking the response value sequence along the boundary path, if a continuous segment of response values is below a pre-set continuity judgment threshold (the threshold is set by statistical analysis of a large number of clear and interrupted sample images, and is the limit that can distinguish the response value distribution of the two types of samples, for example, set to 100), it is judged that there is spatial discontinuity at that line segment position. Finally, the position information of all the line segments judged to be discontinuous is collected to obtain the spatial features of the real estate subject boundary.

[0088] S212: Screen the real estate subject boundary points with a plane offset value exceeding a set offset threshold value in the boundary error data, and perform curve fitting operation on the screened points in combination with the spatial features of the real estate subject boundary, calculate the boundary fitting correction parameter according to the comparison result of the fitting curve direction difference and the original boundary direction difference;

[0089] Among them, the boundary fitting correction parameter is calculated according to the comparison result of the fitting curve direction difference and the original boundary direction difference, specifically:

[0090] Obtaining a fitting curve direction difference;

[0091] Obtaining an original boundary direction difference;

[0092] Calculating an angle deviation value between the fitting curve direction difference and the original boundary direction difference;

[0093] Calling boundary error data to obtain a plane offset value corresponding to a real estate subject boundary point;

[0094] Setting an angle deviation weight coefficient;

[0095] Setting a plane offset weight coefficient;

[0096] Performing a multiplication operation on the angle deviation value and the angle deviation weight coefficient to obtain an angle correction amount;

[0097] Performing a multiplication operation on the plane offset value and the plane offset weight coefficient to obtain an offset correction amount;

[0098] Substituting the angle correction amount and the offset correction amount into a preset correction function to perform a weighted sum operation on the angle correction amount and the offset correction amount, and generating a boundary fitting correction parameter;

[0099] The real estate subject boundary points with a plan offset value exceeding a set offset threshold value in the boundary error data are screened. The offset threshold value is set according to the limit difference of the point position error of the boundary point in the relevant national specifications. For example, for a third-level urban cadastral district, the limit difference of the point position error of the boundary point is 0.10 meters. To ensure the effectiveness of the screening, the following verification experiment is carried out: 100 known sample points containing measurement errors and 100 correct sample points checked with high precision are selected, and the plan offset values thereof are calculated. The experimental data show that the plan offset values of 96% of the error sample points are greater than 0.10 meters, and the plan offset values of 99% of the correct sample points are less than 0.10 meters. Based on the experimental results and the specification requirements, the set offset threshold value is determined as 0.10 meters. The screening process is as follows: each point in the boundary error data is traversed, the plan offset value thereof is read, and the plan offset value is compared with 0.10 meters. If the plan offset value of a point is 0.12 meters, the point is screened out because the plan offset value is greater than 0.10 meters. If the plan offset value of another point is 0.08 meters, the point is not screened out. All the screened points are collected, and curve fitting operations are performed on the screened points and the adjacent point sequences thereof in combination with the spatial features of the real estate subject boundary indicating the boundary breakpoints or the abnormal regions of the shape of the real estate subject boundary obtained in step S211. The specific process of the curve fitting operation is as follows: for a boundary point sequence containing abnormal points, a cubic B-spline curve is used for fitting, the positions of the control points are adjusted through iteration, so that the generated curve can smoothly pass through or approach the points in the sequence, while keeping the shape consistent with the continuous part shown in the spatial features. According to the comparison result of the fitting curve direction difference and the original boundary direction difference, the boundary fitting correction parameter is calculated. Specifically, at the point “JZD01”, the tangent direction of the fitting curve is obtained, and the fitting curve direction difference is calculated as 46.50 degrees. The direction of the original boundary line segment composed of “JZD01” and its adjacent points in the cadastral data is obtained, and the original boundary direction difference is calculated as 49.00 degrees. The angular deviation value between the two is calculated, that is, 49.00 degrees minus 46.50 degrees, and the result is 2.5 degrees. The boundary error data of step S113 is called to obtain the plan offset value of the “JZD01” point, which is 0.12 meters. The angular deviation weight coefficient and the plan offset weight coefficient are set. The setting of the two coefficients is obtained through a correction experiment on 50 different shape boundary error samples. In the experiment, different weight combinations from 0.1 to 0.9 are tested, and the minimum residual error between the corrected coordinates and the true coordinates is taken as the target, and finally the angular deviation weight coefficient is determined as 0.7 and the plan offset weight coefficient is determined as 0.3. This setting reflects the priority of ensuring the correctness of the boundary trend in the correction. The angular correction amount is obtained by multiplying the angular deviation value 2.5 degrees by the angular deviation weight coefficient 0.7, which is 1.75 degrees. The offset correction amount is obtained by multiplying the plan offset value 0.12 meters by the plan offset weight coefficient 0.3, which is 0.036 meters.The angle correction amount and the offset correction amount are substituted into a preset correction function, and a weighted sum operation is performed on the two to generate a boundary fitting correction parameter. The parameter is a composite parameter containing rotation and translation information, which specifically indicates that the original point position needs to be adjusted by 1.75 degrees of angle and 0.036 meters of displacement.

[0100] S213: According to the boundary fitting correction parameter, the real estate main boundary point position whose plane offset value exceeds the set offset threshold is corrected by boundary fitting, and the corrected real estate main boundary point position is integrated with the cadastral vector boundary data to generate boundary correction data.

[0101] According to the boundary fitting correction parameter, each real estate main boundary point position whose plane offset value exceeds the set offset threshold is corrected by boundary fitting. Taking the point position "JZD01" as an example, the boundary fitting correction parameter specifies an angle correction amount of 1.75 degrees and an offset correction amount of 0.036 meters. The correction process is as follows: first, taking the original coordinates (X: 452135.45 meters, Y: 345678.90 meters) of the point position "JZD01" in the cadastral vector boundary data as the basis. The point position is regarded as a vector, and according to the angle correction amount of 1.75 degrees, the reference point of the point position relative to the line segment is rotated and transformed. Then, according to the offset correction amount of 0.036 meters, the coordinates of the rotated point position are translated along the normal direction of the fitting curve at the point, or according to the offset vector direction. After performing the two geometric transformations, a corrected coordinate of "JZD01" is obtained, for example, (X': 452135.54 meters, Y': 345678.96 meters). For all the point positions selected in step S212 whose plane offset value exceeds the limit, the correction process is repeated to update the coordinates using their respective boundary fitting correction parameters. All the corrected real estate main boundary point position coordinates are integrated with the point position coordinates in the cadastral vector boundary data that are not selected (i.e. whose plane offset value does not exceed the limit), original, and do not need to be corrected. The integration process is to create a new point position coordinate dataset that contains the final coordinates of all boundary point positions, part of which are new corrected coordinates and part of which are original old coordinates. By connecting these point positions in the original order, a complete and locally optimized boundary line is generated, which is the boundary correction data.

[0102] Please refer to Figure 4 , the acquisition step of the boundary position update data is:

[0103] S311: Set the boundary correction data as the prior mean value of the real estate main boundary point position observation equation, and set the root mean square statistical result in the boundary error data as the prior variance estimate value of the real estate main boundary point position observation equation, combine the prior mean value and the prior variance estimate value, and establish the prior distribution of the real estate main boundary point position.

[0104] The boundary correction data is set as the prior mean value of the position observation equation of the real estate subject boundary point. Specifically, for the boundary point "JZD01", its coordinate in the boundary correction data is (X': 452135.54 meters, Y': 345678.96 meters), and this coordinate value is taken as the expected value or mean value of the prior distribution of the point. At the same time, the observation noise level calculated in step S113 is obtained, which is about 0.112 meters. The square of this value is set as the prior variance estimate value of the position observation equation of the real estate subject boundary point, that is, square meters. The prior mean value (452135.54, 345678.96) and the prior variance estimate value (0.012544, 0.012544) are combined to establish the prior distribution of the real estate subject boundary point. This prior distribution is a two-dimensional Gaussian distribution, the mathematical expression of which is centered at the corrected coordinate point, and the dispersion degree of the distribution is determined by the observation noise level. The same operation is performed for each boundary point of the real estate to establish an independent prior distribution model describing the probability distribution of the position of each point.

[0105] S312: Obtain the GNSS coordinate data of the real estate subject boundary, take the GNSS coordinate data of the real estate subject boundary as the observation through the Bayesian inference model, and set the square of the root mean square statistical result in the boundary error data as the variance of the observation to generate the point observation likelihood distribution;

[0106] The latest GNSS coordinate data of the real estate subject boundary is obtained, which is the result of a new round of field measurement and is taken as an independent observation. For example, the RTK measurement of the "JZD01" point is performed again and the coordinate is converted to obtain a new GNSS coordinate, that is, the observation, whose coordinate is (X observation: 452135.58 meters, Y observation: 345678.99 meters). The Bayesian inference model is used to take the GNSS coordinate data of the real estate subject boundary as the observation. The square of the observation noise level calculated in step S113, that is, about 0.012544 square meters, is set as the variance of the normal distribution of this observation. This variance represents the random error level existing in the GNSS measurement process itself. Based on this observation and variance, the point observation likelihood distribution is generated. This likelihood distribution is also a two-dimensional Gaussian distribution, the center of which is located at the coordinate (452135.58, 345678.99) of the current GNSS observation, and the variance is (0.012544, 0.012544). This likelihood distribution describes the probability of observing the current GNSS coordinate given a true position. For all boundary points, their latest GNSS observation is used, and the same variance is set to generate their respective point observation likelihood distributions.

[0107] S313: Calculate the posterior coordinates at each cadastral subject boundary point according to the cadastral subject boundary point prior distribution and the point position observation likelihood distribution, and retrieve the coordinate value corresponding to the maximum posterior coordinate as the updated cadastral subject boundary point coordinate to generate boundary position update data;

[0108] According to the cadastral subject boundary point prior distribution and the point position observation likelihood distribution generated in step S312, the posterior coordinates at each cadastral subject boundary point are calculated. This calculation uses an improved Kalman gain form, which realizes the dynamic weighted fusion of prior information and new observation information by introducing model parameters. The calculation formula of the posterior coordinates is: wherein the gain factor is determined by the following formula: , : represents the posterior coordinates after the fusion of the prior distribution and the observation likelihood distribution, i.e. the final updated point position coordinates, with the dimension of length (m). : the coordinate of the prior distribution, from the boundary correction result in step S213, which is the best estimate of the point position before this update, with the dimension of length (m). : the coordinate of the observation likelihood distribution, from the latest GNSS measurement result in step S312, which is the new information introduced in this update, with the dimension of length (m). : gain factor, a dimensionless number that determines the extent to which the new observation value can correct the prior estimate. Its value is between 0 and 1. : prior variance estimate, from step S311, representing the uncertainty degree of the coordinate of the prior distribution, with the dimension of area (square meters). : observation value variance, from step S312, representing the uncertainty degree of the GNSS observation, with the dimension of area (square meters). : prior confidence parameter, with the dimension of area (square meters). Used to adjust the confidence in the prior model. When the boundary correction in S2 is significant and the model reliability is high, the value of should be appropriately increased, thereby increasing the numerator and increasing the weight of prior information in the fusion. : model uncertainty parameter, with the dimension of area (square meters). Used to express the uncertainty in the entire fusion model that has not been quantified. When the observation environment is complex or the quality of historical data is questionable, the value of should be appropriately increased, thereby increasing the denominator and reducing the gain , making the update process more conservative.

[0109] Take point "JZD01" as an example for calculation.

[0110] the coordinate of the prior distribution​ Coordinate of observation likelihood distribution Prior variance 0.0125 square meters. Observation variance 0.0125 square meters. Parameter 0.001 square meters according to the quality assessment of this revision. Parameter 0.002 square meters considering the observation environment of medium complexity.

[0111] Plug into the formula to calculate:

[0112] Calculate the gain factor

[0113]

[0114] Calculate the posterior coordinate

[0115] X coordinate: meters.

[0116] Y coordinate:

[0117] meters.

[0118] Retrieve the coordinate value corresponding to the maximum of the posterior coordinate, which is the posterior mean under Gaussian distribution. Therefore, the coordinate (452135.56, 345678.975) is taken as the updated coordinate of the real estate subject boundary point of "JZD01". Repeat this process for all boundary points to generate boundary position update data.

[0119] See Figure 5 , the steps for obtaining the consistency screening result are as follows:

[0120] S411: Obtain multi-temporal cadastral data of the real estate subject formed in multiple periods, based on the boundary position update data, perform time series difference calculation on the cadastral data coordinates of the same name real estate subject boundary points in the multi-temporal cadastral data and the corresponding boundary point coordinates stored in the boundary position update data point by point, and collect the difference results of all real estate subject boundary points to generate time series difference values;

[0121] ​​​​​​​The real estate subject in multiple historical periods, such as 2010, 2015, 2020, is obtained. The multi-temporal cadastral data stored in the database is formed by the surveying and mapping work. Based on the boundary position update data generated in step S313, the cadastral data coordinates of the boundary point of the same real estate subject in the multi-temporal cadastral data are calculated by time difference calculation with the corresponding boundary point coordinates stored in the boundary position update data. Taking point JZD01 as an example, its coordinate in the boundary position update data is (452135.56, 345678.975). The historical data is called, and its coordinate in 2010 is (452135.40, 345678.85), in 2015 is (452135.45, 345678.90), and in 2020 is (452135.48, 345678.92). The specific process of time difference calculation is: the plane distance between the coordinate of each historical period and the updated coordinate is calculated respectively. For example, the difference value in 2010 is the Euclidean distance between (452135.56, 345678.975) and (452135.40, 345678.85), which is about 0.20 meters. Similarly, the difference value in 2015 is 0.128 meters, and the difference value in 2020 is 0.095 meters. For all the boundary points of the real estate, this operation is performed, and the difference value of each point in each historical period relative to the current updated position is calculated. The difference results of all the boundary points of the real estate subject in all historical periods are collected to generate a large set of time difference values.

[0122] S412: Perform difference absolute median statistics on the time difference values, and compare them with the obtained national surveying and mapping accuracy standard values to establish a point position deviation discrimination set;

[0123] For the set of time series difference values, the differential absolute median statistics is performed. The statistical process is as follows: first, for a point "JZD01", its time series difference value sequence is [0.20 meters, 0.128 meters, 0.095 meters]. The median of this sequence is calculated, which is 0.128 meters. Then, the absolute deviation between each difference value and the median is calculated, resulting in [|0.20-0.128|, |0.128-0.128|, |0.095-0.128|], i.e. [0.072, 0, 0.033]. Finally, the median of these absolute deviations is calculated, which is 0.033 meters. This 0.033 meters is the differential absolute median of point "JZD01". The time series difference sequence of all points is executed. Then, it is compared with the obtained national surveying and mapping accuracy standard value item by item. The national surveying and mapping accuracy standard value is derived from "Digital Line Graph (DLG) Quality Inspection Technical Regulations", which specifies the point accuracy requirements of different ground features under different scales. For example, for a house corner point under the scale of 1:500, the allowed point error is 0.15 meters. This 0.15 meters is used as the reference value for discrimination. The process of establishing the point deviation discrimination set is as follows: compare the differential absolute median of each point with 0.15 meters. If the differential absolute median of point "JZD01" is 0.033 meters, which is less than 0.15 meters, the point is determined to be "stable". If the differential absolute median of another point "JZD05" is calculated to be 0.25 meters, which is greater than 0.15 meters, the point is determined to be "unstable". The identifier of each point and its determination result (stable / unstable) are stored in a set, which is the point deviation discrimination set.

[0124] S413: According to the point deviation discrimination set, binary classification is performed on all real estate subject boundary point positions in the multi-temporal cadastral data, and the point positions are divided into consistent or inconsistent two categories, integrated with the boundary position update data, and a consistency screening result is generated.

[0125] According to the point deviation discrimination set, binary classification is performed on all real estate subject boundary point positions in multi-temporal cadastral data. The classification process is: traverse each entry in the discrimination set. If a point position, such as "JZD01", is recorded as "stable" in the discrimination set, the point position is classified into the "consistent" category. If another point position, such as "JZD05", is recorded as "unstable", it is classified into the "inconsistent" category. "Consistent" here means that the point position has maintained relative position stability in a long time series, and its historical coordinate change is within the allowed surveying and mapping error range; "inconsistent" indicates that the point position may have undergone real position changes, such as boundary marker relocation, building demolition and reconstruction, etc., or there may be historical gross surveying errors. The classification results (consistent / inconsistent) of all point positions are integrated with the boundary position update data generated in step S313. The integration method is: on the basis of the boundary position update data, an attribute field is added for each point position to store its classification result. For example, for "JZD01", in addition to the updated coordinates, a "consistency" field with a value of "consistent" is also attached. Through this operation, a comprehensive data set containing not only the most accurate point position coordinates but also the historical stability of each point position is generated, which is the consistency screening result.

[0126] Referring to Figure 6 The acquisition steps of the real estate surveying and mapping optimization result are as follows:

[0127] S511: Based on the consistency screening result, consistent real estate subject boundary point positions are selected, and the real estate subject boundary position parameters in the boundary position update data are called to perform spatial fusion comparison and matching. For adjacent real estate subject boundary line segments, the direction parameters and curvature parameters in the boundary correction data are called to perform line segment relocation to obtain a set of relocated boundary points;

[0128] Based on the consistency screening result, all real estate subject boundary point positions classified as "consistent" are selected. The real estate subject boundary position parameters of these consistent point positions in the boundary position update data of step S313 are called to perform spatial fusion comparison and matching. For adjacent real estate subject boundary segments, the direction parameters and curvature parameters in the boundary correction data of step S213 are called to perform segment relocation. The specific execution process is: considering two adjacent real estate units, they share a common boundary. From the consistency screening result, the point position sequences of the two units marked as "consistent" on the common boundary are extracted respectively. Ideally, after the latest observation is fused, the two point position sequences should be exactly coincident. However, due to small residual errors, there may be slight misalignment. The segment relocation operation aims to eliminate this misalignment. The direction and curvature information of the segment boundary in the correction process recorded in the boundary correction data are called, which reflects the geometric form of the boundary. With these form parameters as constraints, the two point position sequences are considered as a whole, and joint adjustment calculation is performed to solve a new point position sequence that can best fit the two sequences while satisfying the form constraints. This new sequence is the relocated boundary point set, which represents the only common boundary between the two real estate subjects after negotiation and optimization.

[0129] S512: Call the relocated boundary point set and obtain the updated coordinate point set in the boundary position update data, calculate the Hausdorff distance between the two, take the distance as the matching degree, compare the matching degree with the set matching threshold range, screen the real estate subject boundary point positions, and obtain the matched boundary point position set;

[0130] The set of relocation boundary points is called, and the set of updated coordinate points in the boundary position update data of step S313 is obtained. The Hausdorff distance of the two is calculated. The calculation process is as follows: for each point in the set of relocation boundary points, find the nearest point in the set of updated coordinate points, and record the distance; after traversing all the relocation points, find the maximum value in the distances, denoted as h1. Conversely, for each point in the set of updated coordinate points, find the nearest point in the set of relocation boundary points, and record the distance; after traversing all the points, find the maximum value in the distances, denoted as h2. The larger one of h1 and h2 is the Hausdorff distance between the two sets of points. For example, h1 is calculated to be 0.03 meters, and h2 is calculated to be 0.04 meters, and the Hausdorff distance is 0.04 meters. Take this distance as the matching degree, and compare the matching degree with the set matching threshold range. The setting of the matching threshold is based on the adjacency boundary point distance accuracy requirement of cadastral surveying and mapping, which is usually the real distance corresponding to 0.1 millimeters on the map. For a 1:500 map, the value is 0.05 meters. To ensure the reliability of the matching, a threshold range is set, for example, [0, 0.05 meters]. Compare the calculated Hausdorff distance 0.04 meters with the range. Because 0.04 meters is located in the interval [0, 0.05 meters], it indicates that the shape of the relocated boundary is highly consistent with the updated boundary, and the matching is successful. The real estate subject boundary points are screened, and all points that participate in this matching and whose matching degree meets the threshold range are screened out to obtain the set of matched boundary points. If the calculated Hausdorff distance is 0.08 meters, which is out of range, it indicates that the matching fails, and the related points are not included in the set of matched boundary points, and are marked as needing manual intervention.

[0131] S513: Forming the polygon data of the updated real estate subject according to the set of matched boundary points, and inputting the polygon data into the real estate surveying and mapping database to obtain the optimization result of the real estate surveying and mapping.

[0132] According to the matching boundary point set, the polygon data of the updated real estate subject is formed. The formation process is: all points belonging to the same real estate unit in the matching boundary point set are extracted, and are sequentially connected into a closed polygon according to the original topological connection order. The closed polygon is the updated polygon data of the real estate subject. The polygon data contains the geometric information (coordinates, area, perimeter) and attribute information (right holder, location, etc.) of the real estate. The polygon data of the updated real estate subject is input into the real estate surveying and mapping database. The input process is to replace the old polygon data of the corresponding real estate unit in the database with the new polygon data by performing the update operation of the database. This operation follows a strict transaction management process, including data backup, update execution, log recording and quality verification, etc., to ensure the accuracy and safety of data replacement. After this operation, the cadastral map in the real estate surveying and mapping database is substantially updated and optimized, and the real estate surveying and mapping optimization result is obtained.

[0133] The AI-based real estate surveying and mapping optimization system is used to execute the AI-based real estate surveying and mapping optimization method described above, and the system comprises:

[0134] The boundary error calculation module obtains GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary, calculates the plane offset value and direction angle deviation of the real estate subject boundary point, and generates boundary error data;

[0135] The boundary feature correction module inputs the GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary into a convolutional neural network model, extracts the spatial features of the real estate subject boundary, and performs boundary fitting correction according to the boundary error data to generate boundary correction data;

[0136] The boundary position updating module establishes the prior distribution of the boundary position observation equation of the real estate subject boundary point based on the boundary correction data, sets the observation likelihood of the real estate subject boundary point based on the Bayesian inference model, updates the coordinates of the real estate subject boundary point based on the prior distribution and the observation likelihood, and obtains boundary position updating data;

[0137] The cadastral consistency screening module screens the multi-temporal cadastral data of the real estate subject based on the boundary position updating data to generate a consistency screening result;

[0138] The polygon spatial fusion module compares the spatial fusion of the real estate subject boundary point and the real estate subject boundary position parameter based on the consistency screening result, updates the polygon data of the real estate subject, and obtains the real estate surveying and mapping optimization result.

[0139] The above merely describes the preferred embodiments of the present application, and is not intended to limit the present application in other forms. Any skilled person in the art can modify or change the disclosed technical content into equivalent embodiments with equivalent changes, and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present application, without departing from the technical solution content of the present application, still falls within the protection scope of the present application.

Claims

1. An AI-based real estate mapping optimization method, characterized by, The method comprises the following steps: S1: obtaining GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary, calculating the plane offset value and direction angle deviation of the real estate subject boundary point, and generating boundary error data; S2: inputting the GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary into a convolutional neural network model, extracting the spatial features of the real estate subject boundary, and performing boundary fitting correction according to the boundary error data to generate boundary correction data; S3: establishing the prior distribution of the position observation equation of the real estate subject boundary point through the boundary correction data, setting the observation likelihood of the real estate subject boundary point through a Bayesian inference model, updating the coordinates of the real estate subject boundary point according to the prior distribution and the observation likelihood, and obtaining boundary position update data; S4: performing consistency screening on the multi-temporal cadastral data of the real estate subject based on the boundary position update data, and generating consistency screening results; S5: performing spatial fusion comparison on the real estate subject boundary point and the real estate subject boundary position parameter based on the consistency screening results, updating the polygon data of the real estate subject, and obtaining real estate surveying and mapping optimization results; The step of obtaining the boundary position update data is specifically: S311: setting the boundary correction data as the prior mean value of the position observation equation of the real estate subject boundary point, obtaining the root mean square statistical result in the boundary error data, setting it as the prior variance estimation value of the position observation equation of the real estate subject boundary point, combining the prior mean value and the prior variance estimation value, and establishing the prior distribution of the real estate subject boundary point; S312: obtaining the GNSS coordinate data of the real estate subject boundary, taking the GNSS coordinate data of the real estate subject boundary as the observation quantity through a Bayesian inference model, and setting the square value of the root mean square statistical result in the boundary error data as the variance of the observation value to generate the point observation likelihood distribution; S313: calculating the posterior coordinates at each real estate subject boundary point according to the prior distribution of the real estate subject boundary point and the point observation likelihood distribution, retrieving the coordinate value corresponding to the maximum posterior coordinates, and taking it as the updated real estate subject boundary point coordinates to generate the boundary position update data; For calculating the posterior coordinates, the formula is: ； wherein, is a gain factor determined by the following equation: , is a prior variance estimate, is a variance of the observation, is a prior confidence parameter, is a model uncertainty parameter, represents a posterior coordinate after fusing the prior distribution and the observation likelihood distribution, is a coordinate of the prior distribution, is a coordinate of the observation likelihood distribution. 2.The AI-based real estate mapping optimization method of claim 1, wherein, The boundary error data includes a planar offset of a real estate subject boundary point, a direction angle deviation of the real estate subject boundary point, and a coordinate residual of the real estate subject boundary point, the boundary correction data includes a curve fitting parameter of a real estate subject boundary line segment, a corrected coordinate of the real estate subject boundary point, and a direction difference of the real estate subject boundary line segment, the boundary position update data includes a posteriori coordinate of the real estate subject boundary point, a posteriori probability value of the real estate subject boundary point, and a coordinate correction amplitude of the real estate subject boundary point, the consistency screening result includes a time sequence difference of the real estate subject boundary point, a classification identifier of the real estate subject boundary line segment, and a screening state of the real estate subject boundary point, and the real estate surveying and mapping optimization result includes updated real estate subject polygon data, fused coordinate set of the real estate subject boundary, and spatial topology structure of the real estate subject. 3.The AI-based real estate mapping optimization method of claim 1, wherein, The acquisition step of the boundary error data is specifically as follows: S111: GNSS coordinate data and cadastral vector boundary data of the real estate subject boundary are acquired, the GNSS coordinate data is projected and converted in a national geodetic coordinate system, and point-by-point matching is performed on the real estate subject boundary points in the cadastral vector boundary data and the corresponding boundary points in the converted GNSS coordinate data, and a matching point coordinate set is established; S112: The matching point coordinate set is called, difference operation is performed on GNSS coordinates and cadastral vector coordinates for each real estate subject boundary point, planar offset value and direction angle deviation of the real estate subject boundary point are acquired respectively, and point offset quantization value is generated; S113: The planar offset value and the direction angle deviation of all real estate subject boundary points are subjected to root mean square statistics, the statistical result is taken as an observation noise level, and the point offset quantization value is combined to acquire the boundary error data. 4.The AI-based real estate mapping optimization method of claim 3, wherein, The acquisition step of the boundary correction data is specifically as follows: S211: The GNSS coordinate data and the cadastral vector boundary data of the real estate subject boundary are input into a convolutional neural network model, the continuity of the real estate subject boundary line segment is judged according to the boundary gradient response value output by the convolutional layer of the convolutional neural network model, and the spatial feature of the real estate subject boundary is acquired; S212: The real estate subject boundary points with planar offset value exceeding a set offset threshold value in the boundary error data are screened, curve fitting operation is performed on the screened points in combination with the spatial feature of the real estate subject boundary, boundary fitting correction parameters are calculated according to the comparison result of the fitting curve direction difference and the original boundary direction difference; S213: The real estate subject boundary points with planar offset value exceeding the set offset threshold value are subjected to boundary fitting correction according to the boundary fitting correction parameters, and the corrected real estate subject boundary points are integrated with the cadastral vector boundary data to generate the boundary correction data. 5.The AI-based real estate mapping optimization method of claim 4, wherein, The boundary fitting correction parameters are calculated according to the comparison result of the fitting curve direction difference and the original boundary direction difference, specifically as follows: The fitting curve direction difference is acquired; The original boundary direction difference is acquired; The angle deviation value between the fitting curve direction difference and the original boundary direction difference is calculated; Call the boundary error data, get the plane offset value corresponding to the real estate subject boundary point position; Set the angle deviation weight coefficient; Set the plane offset weight coefficient; Multiply the angle deviation value and the angle deviation weight coefficient to obtain the angle correction amount; Multiply the plane offset value and the plane offset weight coefficient to obtain the offset correction amount; Substitute the angle correction amount and the offset correction amount into the preset correction function, perform weighted sum operation on the angle correction amount and the offset correction amount, and generate the boundary fitting correction parameter. 6.The AI-based real estate mapping optimization method of claim 1, wherein, The consistency screening result obtaining step is specifically: S411: Obtain multi-temporal cadastral data formed by surveying and mapping of real estate subjects in multiple periods, perform time series difference calculation on cadastral data coordinates of the same name real estate subject boundary point positions in the multi-temporal cadastral data and the corresponding boundary point coordinates stored in the boundary position update data point by point based on the boundary position update data, and collect the difference results of all real estate subject boundary point positions to generate time series difference values; S412: Perform difference absolute median statistics on the time series difference values, and compare each item with the obtained national surveying and mapping accuracy standard value to establish a point position deviation discrimination set; S413: According to the point position deviation discrimination set, perform binary classification on all real estate subject boundary point positions in the multi-temporal cadastral data, divide the point positions into two categories of consistent and inconsistent, and integrate with the boundary position update data to generate a consistency screening result. 7.The AI-based real estate mapping optimization method of claim 6, wherein, The real estate surveying and mapping optimization result obtaining step is specifically: S511: Based on the consistency screening result, select the consistent real estate subject boundary point positions, and call the real estate subject boundary position parameters in the boundary position update data for spatial fusion comparison and matching, call the direction parameters and curvature parameters in the boundary correction data to perform line segment relocation for adjacent real estate subject boundary line segments, and obtain a set of relocated boundary points; S512: Call the set of relocated boundary points and obtain an updated coordinate point set in the boundary position update data, calculate the Hausdorff distance between the two, take the distance as the matching degree, compare the matching degree with the set matching threshold range, screen the real estate subject boundary point positions, and obtain a set of matching boundary point positions; S513: Form the polygon data of the updated real estate subject according to the set of matching boundary point positions, and input the polygon data into the real estate surveying and mapping database to obtain a real estate surveying and mapping optimization result.

8. An AI-based real estate mapping optimization system, characterized by, The AI-based real estate surveying and mapping optimization method according to any one of claims 1-7, the system comprises: A boundary error calculation module obtains GNSS coordinate data and cadastral vector boundary data of a real estate subject boundary, calculates plane offset values and direction angle deviations of real estate subject boundary point positions, and generates boundary error data; A boundary feature correction module inputs GNSS coordinate data and cadastral vector boundary data of a real estate subject boundary into a convolutional neural network model, extracts spatial features of the real estate subject boundary, and performs boundary fitting correction according to the boundary error data to generate boundary correction data; The boundary position updating module establishes a prior distribution of a position observation equation of a cadastral subject boundary point through the boundary correction data, sets an observation likelihood of the cadastral subject boundary point through a Bayesian inference model, updates a coordinate of the cadastral subject boundary point according to the prior distribution and the observation likelihood, and obtains boundary position updating data; The cadastral consistency screening module screens multi-temporal cadastral data of the cadastral subject based on the boundary position updating data, and generates a consistency screening result; The plot space fusion module compares and updates plot data of the cadastral subject based on the consistency screening result, and obtains cadastral surveying and mapping optimization results. The boundary position updating data is obtained by: The boundary correction data is set as a prior mean value of a cadastral subject boundary point position observation equation, and a root mean square statistical result in the boundary error data is obtained as a prior variance estimation value of the cadastral subject boundary point position observation equation, the prior mean value and the prior variance estimation value are combined to establish a prior distribution of the cadastral subject boundary point; GNSS coordinate data of the cadastral subject boundary is obtained, the GNSS coordinate data of the cadastral subject boundary is taken as an observation value through a Bayesian inference model, and a square value of the root mean square statistical result in the boundary error data is set as a variance of the observation value to generate a point observation likelihood distribution; According to the prior distribution of the cadastral subject boundary point and the point observation likelihood distribution, a posterior coordinate is calculated at each cadastral subject boundary point, and a coordinate value corresponding to a maximum of the posterior coordinate is searched as an updated cadastral subject boundary point coordinate to generate boundary position updating data; For calculating the posterior coordinate, the formula is: ； wherein, is a gain factor determined by the following equation: , is a prior variance estimate, is a variance of the observation, is a prior confidence parameter, is a model uncertainty parameter, represents a posterior coordinate after fusing the prior distribution and the observation likelihood distribution, is a coordinate of the prior distribution, is a coordinate of the observation likelihood distribution.

Images