Land Surveying Data Analysis System Based on Image Recognition

Abstract

This invention relates to the field of image recognition technology, specifically to a land surveying data analysis system based on image recognition. The system includes a gradient response extraction module, a boundary classification judgment module, a coupling barrier detection module, an error field construction module, and a coordinate reverse registration module. This invention constructs a multi-dimensional directional gradient distribution sequence between edge pixels of land types and extracts the average rate of change of local response amplitude, enabling quantitative judgment of boundary ambiguity. It combines the boundary position difference distribution to generate response expansion amplitude and performs classification operations, improving the accuracy of automatic identification of spatial barrier structures. It calculates the error difference between geographic standard coordinates and the boundary center point and constructs an error vector flow graph to achieve directional correction of coordinate errors. By updating the boundary point positions step-by-step through an error-guided path, the identified image better conforms to the geographic coordinate system, improving the accuracy of land feature identification and registration consistency.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a land surveying data analysis system based on image recognition. Background Technology

[0002] Image recognition technology involves the automatic identification, analysis, and interpretation of image content. Core aspects include image acquisition, image preprocessing, feature extraction, object detection, image classification, and image semantic understanding. In the image recognition process, neural network models are typically used to model pixel information in an image. The parameters obtained through training are then used to process the input image to output information such as the type, location, and quantity of objects or scenes present in the image. This technology is widely used in industrial inspection, facial recognition, security monitoring, autonomous driving, remote sensing mapping, and many other fields. Especially when combined with geographic information systems (GIS) for remote sensing image analysis, it can automatically identify and classify surface objects, thus providing support for land use, urban planning, and agricultural production.

[0003] Traditional land surveying data analysis systems refer to systems that extract, identify, and analyze land data through manual surveying, remote sensing image interpretation, and basic image processing tools. These systems typically involve surveyors annotating remote sensing images, delineating features, and dividing regions, and then judging land types, usage attributes, or changes based on experience. Some automated systems utilize image processing methods such as image grayscale distribution, edge detection, and histogram matching to segment and identify image regions, and then combine this with geographic coordinate information for statistical analysis of land data. During the analysis process, multiple judgment rules need to be set according to different land characteristics, and recognition accuracy must be ensured through manual verification. However, in cases of complex feature types, significant terrain undulations, or low image resolution, the accuracy and data processing efficiency of traditional methods are limited. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a land surveying data analysis system based on image recognition.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a land surveying data analysis system based on image recognition, the system comprising: The gradient response extraction module acquires aerial images of the surveyed area, constructs edge pixel groups, generates a multidimensional directional gradient distribution sequence, and calculates the average rate of change of local response amplitude within a unit bandwidth. The boundary classification judgment module quantifies the blurring rate of land feature boundaries based on the average change rate of the local response amplitude, generates the response expansion amplitude by combining the average change rate of the local response amplitude and the boundary position difference, and classifies the land feature boundary distribution data based on the response expansion amplitude. The coupling barrier detection module extracts a cross section along the main axis of the land use boundary data, calculates the projection width sequence, generates a first-order difference absolute value sequence of the projection width sequence, and obtains the spatial coupling barrier identification result based on the density of extreme points and the fracture rate of the first-order difference absolute value sequence. The error field construction module acquires geographic standard coordinates, calculates the distance difference between the center point of the spatial coupling barrier identification result and the geographic standard coordinates, constructs a two-dimensional coordinate error matrix, generates an error direction vector based on the two-dimensional coordinate error matrix, and forms a vector flow graph structure. The coordinate reverse registration module constructs a reverse registration path based on the vector flow graph structure, adjusts the coordinates of boundary points along the error gradient, and superimposes the error gradient patch area onto the coordinate system to update the node positions, generating a coordinate-fitted mapping and recognition image.

[0006] The present invention improves upon the following: the average rate of change of local response amplitude includes gradient amplitude increment, pixel interval weight, and response direction factor; the land cover boundary distribution data includes boundary probability matrix, geometric contour coordinates, and transition zone width index; the spatial coupling barrier identification result includes a set of blocking line segments, regional segmentation topology, and barrier semantic attributes; the vector flow graph structure includes offset vector nodes, local deformation tensor, and error flow path; and the mapping and identification image includes a corrected raster matrix, a spatial reference coordinate system, and land cover attribute metadata.

[0007] The present invention is improved in that the gradient response extraction module includes: The edge pixel group construction submodule acquires aerial images of the survey area, calculates the grayscale contrast value of each pixel in the image, compares the grayscale contrast value with the preset edge extraction benchmark value, selects pixels with grayscale contrast values ​​greater than the edge extraction benchmark value, and constructs land use edge pixel groups based on the spatial adjacency relationship of the pixels. The directional gradient sequence generation submodule calls the edge pixel group of the land type, traverses the adjacent pixels in the group, calculates the gray-level change amplitude in the horizontal, vertical and diagonal dimensions respectively, arranges the calculation results according to the spatial distribution order, and generates a multidimensional directional gradient distribution sequence. The response amplitude change calculation submodule, based on the multidimensional directional gradient distribution sequence, determines the preset unit boundary band width, extracts the gradient response values ​​within the bandwidth range, and calculates the average rate of change of local response amplitude.

[0008] The present invention is improved in that the formula for obtaining the average rate of change of the local response amplitude is specifically as follows: ; in, Represents the average rate of change of the local response amplitude. This represents the total number of pixels within the calculation unit. This represents the normalized directional gradient observation of the j-th pixel. The normalized baseline value representing the background gradient. Represents the tangential distance of the boundary curve. Represents the width of the unit boundary band. Represents the noise suppression coefficient. This represents the normalization scaling factor.

[0009] The present invention is improved in that the boundary classification judgment module includes: The blur rate quantization submodule selects a local neighborhood window of the target pixel based on the average change rate of the local response amplitude, calculates the statistical variance of the change rate value within the window, compares the statistical variance with the preset texture clarity benchmark value, analyzes the grayscale correlation of the pixels within the window, and obtains the blur coefficient of the ground feature boundary. The expansion amplitude generation submodule calls the average rate of change of the local response amplitude and the blur coefficient of the ground object boundary, measures the geometric distance of the pixel point relative to the estimated boundary center, calculates the difference between the pixel gradient and the neighborhood mean, extracts the information entropy and high-frequency energy features of the local gray-scale distribution, and calculates and obtains the response expansion amplitude. The classification judgment submodule performs threshold segmentation processing on the response expansion amplitude, identifies pixel regions that exceed the preset classification threshold, performs morphological closing operation on the pixel regions, connects the broken edges, marks the category attributes and location coordinates of the pixels, and constructs land use boundary distribution data.

[0010] The present invention is improved in that the coupling barrier detection module includes: The projection width calculation submodule identifies the main axis extension direction of the land feature distribution area based on the land feature boundary distribution data, sets the spatial step size according to the preset sampling density, constructs multiple equally spaced orthogonal cross-sectional segments along the main axis direction, traverses each orthogonal cross-sectional segment, counts the spatial span of pixels marked as land feature boundaries within the segment range, calculates the effective coverage width of land features corresponding to each cross-sectional position, and generates a regional projection width sequence according to the spatial arrangement order of the cross-sections on the main axis. The difference sequence generation submodule calls the region projection width sequence, extracts the projection width values ​​of two adjacent positions in the sequence according to the index order, performs the subtraction operation between the preceding and following terms to obtain the width change, performs absolute value processing on the width change to eliminate the positive and negative influence of the change direction, extracts the numerical features characterizing the degree of geometric change of the boundary, and reassembles and aligns the calculated absolute values ​​according to the original sequence index order to generate a local first-order difference absolute value sequence. The barrier determination and identification submodule identifies the local maxima positions in the sequence waveform based on the local first-order difference absolute value sequence, calculates the number of maxima within a unit length, obtains the extreme point distribution density, detects segments where the sequence value returns to zero or undergoes a step drop, counts their frequency of occurrence, calculates the edge continuity break rate, compares the extreme point distribution density and the edge continuity break rate with a preset threshold, and determines and outputs the spatial coupling barrier identification result.

[0011] The present invention is improved in that the process of comparing the extreme point distribution density and the edge continuity breakage rate with a preset barrier determination threshold, and determining and outputting the spatial coupling barrier identification result is specifically as follows: Set a sliding window of fixed length, slide it point by point along the index direction of the local first-order difference absolute value sequence, count the number of maximum points within the coverage area of ​​each sliding window, and use the ratio of the number of maximum points to the length of the sliding window as the distribution density of the extreme points. Identify continuous index segments in the local first-order difference absolute value sequence where the amplitude returns to zero or is lower than the background noise, and calculate the proportion of the cumulative length of the continuous index segments to the total length of the sequence as the edge continuity break rate. Using sample data from standard barrier-free regions, the statistical distribution of texture complexity of the sample data is calculated, and the upper limit of the confidence interval of the statistical distribution is selected as the threshold for barrier density determination. Based on the minimum geometric connectivity requirements of the ground feature boundaries, the maximum allowable percentage of fractures is set as the fracture tolerance threshold. The extreme point distribution density is compared with the barrier density judgment threshold, and the edge continuity breakage rate is compared with the breakage tolerance threshold. When the distribution density of extreme points is greater than the barrier density judgment threshold and the edge continuity breakage rate is less than the breakage tolerance threshold, it is determined that a dense coupling barrier exists. When the edge continuity failure rate is greater than the failure tolerance threshold, it is determined that there is a blocking coupling barrier. Extract the cross-sectional location indices corresponding to dense coupling barriers and blocking coupling barriers, and output the spatial coupling barrier identification results.

[0012] The present invention is improved in that the error field construction module includes: The coordinate error matrix construction submodule calls the spatial coupling barrier identification results, parses the outline of the identified independent ground feature area, calculates the geometric centroid coordinates of each area, synchronously obtains the geographic standard coordinates corresponding to the surveying area as a reference, performs pairing operation between the centroid coordinates and the standard coordinates, solves the Euclidean distance difference between the two on the two-dimensional plane projection, maps and fills the difference according to the grid index of the spatial distribution of ground features, and generates a two-dimensional coordinate error matrix. The error gradient vector generation submodule, based on the two-dimensional coordinate error matrix, traverses the error data nodes in the matrix according to the local difference window, calculates the rate of change of error values ​​in the horizontal and vertical dimensions respectively, synthesizes a gradient vector representing the local error tendency based on the rate of change of the two dimensions, calculates the magnitude and direction angle of the gradient vector, performs spatial position marking and attribute encapsulation on the gradient vectors in the whole region, and generates a set of error direction vectors. The vector flow graph structure mapping submodule extracts the direction and amplitude attributes of each position vector for the error direction vector set, establishes the flow direction transmission topology between adjacent vectors based on grid adjacency relationships, constructs a vector field network, maps the amplitude attribute to field strength parameters, maps the direction attribute to streamline tangential constraints, assigns attribute values ​​to network nodes and organizes them in a structured manner, and generates a vector flow graph structure.

[0013] The present invention is improved in that the coordinate reverse registration module includes: The registration path construction submodule, based on the vector flow graph structure, identifies the flow trend characteristics of the error vector in the spatial distribution, locates the node positions where the vector magnitude exceeds the preset mutation threshold, plans the connection order from the error peak point to the edge stable region according to the gradient descent rule, establishes the topological association between nodes, constructs the trajectory line to guide coordinate correction, and generates a self-guided reverse registration path. The gradient patch generation submodule calls the self-guided reverse registration path, traverses the boundary pixels of the ground objects one by one along the dominant direction of the error gradient, calculates the spatial displacement compensation amount at each stage based on the vector amplitude, performs geometric distortion correction on the local boundary area, eliminates positional deviation, defines the boundary of the pixel range that needs to be resampled, and generates the error gradient patch area. The image correction output submodule dynamically overlays the error gradient patch area onto the coordinate system of the original surveying data, performs real-time grid node coordinate updates, performs bilinear interpolation remapping on the image pixels within the patch coverage area, corrects the spatial position of the ground feature outline, and generates a surveying and recognition image.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing a multidimensional directional gradient distribution sequence among edge pixels of land use types and extracting the average rate of change of local response amplitude, a quantitative judgment of the degree of boundary ambiguity can be achieved. By combining the response expansion amplitude generated by the boundary position difference distribution and performing classification operations, the ability to distinguish boundary types is enhanced. By extracting the projection width sequence through the regional principal axis direction and analyzing the density of first-order difference extreme points and the edge continuity break rate, the accuracy of automatic identification of spatial barrier structures is improved. The error difference between the geographic standard coordinates and the boundary center point is calculated and an error vector flow graph is constructed to achieve directional correction of coordinate errors. By updating the boundary point position step by step through the error guidance path, the identified image is made to fit the geographic coordinate system better, improving the accuracy of land feature identification and registration consistency. Attached Figure Description

[0015] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the gradient response extraction module of the present invention; Figure 3 This is a flowchart of the boundary classification judgment module of the present invention; Figure 4 This is a flowchart of the coupling barrier detection module of the present invention; Figure 5 This is a flowchart of the error field construction module of the present invention; Figure 6 This is a flowchart of the coordinate reverse registration module of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0018] Please see Figure 1The present invention provides a technical solution, a land surveying data analysis system based on image recognition, the system including a gradient response extraction module, a boundary classification judgment module, a coupling barrier detection module, an error field construction module and a coordinate reverse registration module; The gradient response extraction module acquires aerial images of the surveyed area, constructs land cover edge pixel groups based on pixel grayscale comparison, generates a multidimensional directional gradient distribution sequence between adjacent pixels, and calculates and extracts the average rate of change of local response amplitude within a preset unit boundary band width. The boundary classification judgment module quantifies the blurring rate of land cover boundaries based on the average change rate of local response amplitude, generates the response expansion amplitude by combining the difference distribution between the change rate and the relative position of the boundary, performs boundary classification judgment operation based on the response expansion amplitude, and constructs land cover boundary distribution data. The coupling barrier detection module extracts equally spaced orthogonal cross-sectional line segments along the main axis of the area to be analyzed based on the land use boundary distribution data, calculates the regional projection width sequence, generates a local first-order difference absolute value sequence of the regional projection width sequence, and determines and outputs the spatial coupling barrier identification result based on the extreme point distribution density and edge continuity discontinuity rate in the local first-order difference absolute value sequence. The error field construction module obtains geographic standard coordinates, calculates the Euclidean distance difference between the center point of the land feature defined by the spatial coupling barrier identification result and the geographic standard coordinates, constructs a two-dimensional coordinate error matrix, and generates a set of error direction vectors based on the error gradient changes of adjacent areas in the two-dimensional coordinate error matrix, forming a vector flow graph structure with directional and amplitude information. The coordinate reverse registration module constructs a self-guided reverse registration path based on the vector flow graph structure, adjusts the spatial coordinate position of the feature boundary points in stages along the dominant error gradient direction, and updates the coordinate system in real time by adding error gradient patch areas to generate a geographic coordinate-fitted mapping and recognition image.

[0019] The average rate of change of local response amplitude includes gradient amplitude increment, pixel interval weight and response direction factor; land cover boundary distribution data includes boundary probability matrix, geometric contour coordinates and transition zone width index; spatial coupling barrier identification results include blocking line segment set, regional segmentation topology and barrier semantic attributes; vector flow graph structure includes offset vector nodes, local deformation tensor and error flow path; and surveying and mapping identification image includes corrected raster matrix, spatial reference coordinate system and land cover attribute metadata.

[0020] Please see Figure 2 The gradient response extraction module includes: The edge pixel group construction submodule acquires aerial images of the survey area, calculates the grayscale contrast value of each pixel in the image, compares the grayscale contrast value with the preset edge extraction benchmark value, selects pixels with grayscale contrast values ​​greater than the edge extraction benchmark value, and constructs land use edge pixel groups based on the spatial adjacency relationship of the pixels. The image specifications were selected as RGB three-channel orthophotos covering a 5 square kilometer survey area, with a pixel resolution better than 0.1 meters. The original RGB image was converted into an 8-bit grayscale matrix using an image preprocessing algorithm. For each pixel in the matrix... Set a center around it The neighborhood calculation window extracts the grayscale values ​​of 9 pixels within the window. It then calculates the sum of the absolute values ​​of the differences between the grayscale value of the center pixel and the grayscale values ​​of the other 8 pixels in the neighborhood. This sum is divided by the number of neighborhood pixels (8) to obtain the grayscale contrast value at that location. For example, if the grayscale value of the center pixel is 158 and the average grayscale value of the neighborhood pixels is 45, the calculated grayscale contrast value is 113. To ensure the accuracy of edge extraction, a baseline value needs to be set. This baseline value is set by selecting a known non-edge flat road surface area in the surveyed area as a sample set, statistically analyzing the grayscale contrast values ​​of 1000 pixels within the sample set, and calculating their arithmetic mean. 25 and standard deviation It is 10, based on The criterion is to set the edge extraction baseline value as follows: The grayscale contrast values ​​of all pixels in the image are compared one by one with the edge extraction benchmark value of 55. Pixels with a value greater than 55 are selected and marked as candidate edge points. Then, an 8-neighborhood connectivity analysis algorithm is used to traverse all candidate edge points, and consecutively adjacent candidate points in spatial coordinates are merged into the same set. Isolated points or small connected regions with fewer than 5 pixels are eliminated. Finally, the land use edge pixel groups are constructed based on the spatial adjacency of the pixels.

[0021] The directional gradient sequence generation submodule calls the edge pixel group of the land type, traverses the adjacent pixels in the group, calculates the gray-level change amplitude in the horizontal, vertical and diagonal dimensions respectively, arranges the calculation results according to the spatial distribution order, and generates a multidimensional directional gradient distribution sequence. For each edge pixel recorded within a group, a local coordinate system is established. The Sobel operator is then applied to all spatially adjacent pixels within the group for convolution operations along the horizontal (X-axis), vertical (Y-axis), and 45-degree diagonal directions. For example, in the horizontal direction... The convolutional kernel calculates the gray-level difference between left and right adjacent pixels to obtain the gray-level change amplitude in the horizontal dimension. Similarly, obtain the vertical dimension. and diagonal dimension Let the grayscale change amplitude of a certain edge point be calculated in three dimensions as follows: These amplitude values ​​are then arranged according to the spatial distribution order of pixels on the edge curve (along the direction of increasing coordinate index). For edge groups... Each pixel is recorded sequentially for its three-dimensional gradient magnitude, forming an ordered dataset containing position indices and multidimensional gradient values, thus generating a multidimensional directional gradient distribution sequence.

[0022] The response amplitude change calculation submodule, based on a multidimensional directional gradient distribution sequence, determines a preset unit boundary band width, extracts the gradient response values ​​within the bandwidth range, and uses the following formula: ; The average rate of change of the local response amplitude is obtained through calculation. in, Represents the average rate of change of the local response amplitude. This represents the total number of pixels within a calculation unit, obtained by counting the number of pixels within the selected unit's boundary band. This represents the current index of the summation operation, corresponding to each pixel in the traversal process. The normalized directional gradient observation value of the j-th pixel is obtained by calculating the gradient magnitude of that pixel and dividing it by the maximum theoretical value of the image grayscale gradient. The normalized baseline value representing the background gradient is obtained by extracting the average gradient intensity of the non-edge background region and then normalizing it. The tangential distance representing the boundary curve is obtained by calculating the Euclidean distance from the pixel along the local edge tangent direction to the calculation center point. The width of the unit boundary band is obtained by reading the pixel width parameter of the edge detection region preset by the system. The noise suppression coefficient is obtained by analyzing the statistical variance of the high-frequency components of the image and calculating its reciprocal. This represents the normalization scaling factor, which is obtained by calculating the proportional relationship between the gradient strength dimension and the noise figure at this calculation scale. For example, setting the unit boundary band width The unit is 3.0 pixels. The gradient response values ​​of all pixels within this bandwidth are extracted using the formula: ; The average rate of change of the local response amplitude is obtained through calculation. The parameters and their calculation logic in the formula are as follows: It represents the average rate of change of the local response amplitude and is used to characterize the net fluctuation of the gradient intensity in the edge region; The summation symbol represents the summation of the calculation results for all pixels within the calculation unit; Used to calculate the arithmetic mean of the accumulated results to eliminate the influence of differences in pixel count; Indicates the first The absolute deviation of the gradient observation value of each pixel relative to the background reference value is used to extract the effective edge signal; It is a gain coefficient based on geometric relationships, where and The ratio of the hypotenuse to the legs of the right triangle is used to weight and amplify the signal based on the degree of pixel deviation from the center; the farther away from the center ( The larger the value, the greater the weight, in order to compensate for the edge diffusion effect; This is a noise suppression term; the scaled noise component is subtracted from the total signal to ensure the purity of the result. In the specific calculation, the parameter assignment process is as follows: It is obtained by counting the actual number of pixels contained within the unit boundary band, for example, by taking a value of 5; The value is obtained by normalizing the pixel gradient magnitude (e.g., 180) by the maximum theoretical gradient (255), and the result is taken as a value. ; The value is obtained by dividing the average gradient of the background region (e.g., 20) by 255. ; Calculate the Euclidean distance from the pixel to the center, for example, take a value of 1.5 pixels; Read the system preset value of 3.0 pixels; High-frequency wavelet decomposition was performed on a flat region of the image, and the variance of its coefficients (e.g., 0.04) was calculated, along with their reciprocals. Here, the units need to be adjusted by multiplying by a proportionality factor. ,final (Value is 0.025). By calculating the ratio of the gradient strength dimension (1.0) to the noise figure (0.025) and introducing a scaling constant, set to 1.2, the above values ​​are substituted into the formula to calculate the individual terms: ; For all Calculate the points and average them, then round the final result to three decimal places. This generates the average rate of change of the local response amplitude. Specific parameter examples are shown in Table 1.

[0023] Table 1. Parameters for calculating the rate of change of local response amplitude: ; The results show that within the current boundary zone, the effective edge gradient intensity, after geometric weighting correction and noise suppression, remains at a normalized level of 0.687. This value directly reflects the salience of the ground feature edge and generates the average rate of change of the local response amplitude.

[0024] Please see Figure 3 The boundary classification judgment module includes: The blur rate quantization submodule selects a local neighborhood window of the target pixel based on the average change rate of the local response amplitude, calculates the statistical variance of the change rate values ​​within the window, compares the statistical variance with the preset texture sharpness benchmark value, analyzes the grayscale correlation of pixels within the window, and obtains the blur coefficient of the ground feature boundary. Establish with the target pixel as the center The local neighborhood window is used to extract the average rate of change of local response amplitude for all pixels within the window. The statistical variance of these 25 values ​​is calculated; for example, the variance is 0.015. To determine the degree of blur represented by this variance, it needs to be compared with a preset texture sharpness benchmark. This benchmark is obtained by collecting high-resolution ground texture samples, calculating the variance distribution of their response rate of change, and taking the lower quartile of the distribution as the benchmark value; for example, a benchmark value of 0.005 is set. The ratio of the current variance 0.015 to the benchmark value 0.005 is calculated. Simultaneously, the gray-level correlation of pixels within the window is analyzed. The energy characteristics of the Gray-Level Co-occurrence Matrix (GLCM) are used as the correlation weight (e.g., 0.8). The ratio is multiplied by the weight, i.e., The calculation result quantifies the degree of texture divergence in the boundary region and obtains the fuzziness coefficient of the ground feature boundary.

[0025] The extended amplitude generation submodule calls the average rate of change of local response amplitude and the blur coefficient of ground feature boundary to determine the geometric distance of a pixel relative to the estimated boundary center, calculates the difference between the pixel gradient and the neighborhood mean, and extracts the information entropy and high-frequency energy features of the local gray-level distribution using the following formula: ; The calculation yields the response expansion magnitude; in, Represents the magnitude of response expansion. This represents the total number of local sampling points, obtained by setting the number of pixels in the calculation window. Represents the current index of the summation operation, corresponding to each sampling point within the window; Represents the average rate of change of the local response amplitude. The normalized value representing the relative positional deviation of the boundary is obtained by calculating the Euclidean distance from the sampling point to the center of the predicted boundary and dividing it by the maximum search radius. The normalized value representing the difference in gradient distribution is obtained by calculating the absolute value of the difference between the gradient magnitude of the sampling point and the average gradient of the neighborhood, and then performing normalization. The normalized value representing the spatial continuity factor is obtained by calculating the gradient direction consistency coefficient along the boundary extension direction. The normalized value representing the ambiguity coefficient of the ground feature boundary is obtained by mapping the calculation results from the previous steps. The normalized value representing the local information entropy is obtained by statistically analyzing the gray-level histogram within the local window, calculating the Shannon entropy, and then normalizing it. The normalized value representing the proportion of high-frequency components is obtained by performing wavelet decomposition on local image patches and calculating the proportion of high-frequency subband energy. The classification threshold is obtained by statistically analyzing the numerical distribution characteristics of the response expansion amplitude of all pixels in the test area. Specifically, a frequency histogram of the response expansion amplitude is constructed, the maximum variance point between the foreground boundary pixel set and the background noise pixel set is calculated using the maximum inter-class variance principle, and the amplitude value that maximizes the inter-class variance is determined as the classification threshold. The average rate of change of local response amplitude (e.g., 0.687) and the blurring coefficient of the ground feature boundary (e.g., 2.4, after normalization) are used to determine the geometric distance of the pixel relative to the estimated boundary center, and the gradient value of the pixel (e.g., 0.75) and its... The difference is calculated using the neighborhood mean (e.g., 0.65), and the Shannon information entropy (e.g., 4.5) and wavelet high-frequency energy proportion (e.g., 0.3) are calculated for the local gray-level distribution, using the following formula: ; The calculation yields the response expansion magnitude. The advantage of this formula lies in its ability to... The term geometrically couples edge strength with spatial location deviation and gradient consistency, and enhances the amplitude characteristics under the combined effect of the two by using square root operations. The introduction of ambiguity, information entropy, and frequency feature modulation ensures that the calculation of the expansion amplitude fully considers texture complexity and frequency distribution. The meanings and values ​​of each parameter are as follows: Represents the magnitude of response expansion. Set the number of local sampling points (e.g., 9). Take the previously calculated value of 0.687, To normalize the positional deviation, it is obtained by calculating the distance of the sampling point from the center (e.g., 2 pixels) and dividing it by the maximum search radius (e.g., 10 pixels), with a value of 0.2. To calculate the normalized value of the gradient distribution difference, ; The spatial continuity factor is obtained by calculating the cosine consistency of the gradient direction along the edge tangent, and its value is 0.05. The normalized value of the ambiguity coefficient for the ground feature boundary is obtained by mapping the aforementioned coefficient 2.4 to the Sigmoid function. The range is 0.91; The local information entropy normalization value is obtained by dividing the entropy value of 4.5 by the maximum entropy of 8, which yields a value of 0.56. The normalized value for the proportion of high-frequency components is 0.3. Substitute this value into the formula to calculate the individual term: ; right The final result after averaging the points is approximately The results indicate that, after comprehensively considering location, texture, and frequency features, the edge response exhibits a significant expansion trend. The classification threshold is obtained by statistically analyzing the numerical distribution characteristics of the response expansion amplitude of all pixels within the test area. Specifically, a frequency histogram of the response expansion amplitude is constructed, the maximum variance point between the foreground boundary pixel set and the background noise pixel set is calculated using the maximum inter-class variance principle, and the amplitude value that maximizes the inter-class variance is determined as the classification threshold (e.g., the calculated threshold is 0.45). The response expansion amplitude is then calculated.

[0026] The classification judgment submodule performs threshold segmentation processing on the response expansion amplitude, identifies pixel regions that exceed the preset classification threshold, performs morphological closing operation on the pixel regions, connects the broken edges, marks the category attributes and location coordinates of the pixels, and constructs land use boundary distribution data. The response expansion magnitude (e.g., 0.599) is numerically compared with the calculated classification threshold (0.45). Because... The pixel is determined to belong to a potential land feature boundary region. After comparing all pixels in the entire image, a binary mask is formed. For small holes and breaks in the mask, a morphological closing operation is performed using a circular structuring element with a radius of 2 pixels, i.e., first dilation and then erosion, to connect the break edges with a spacing of less than 4 pixels. Then, based on the connected component features of the pixel, its land feature category attributes (such as building outlines, road edges) and precise location coordinates are labeled to construct land feature boundary distribution data.

[0027] Please see Figure 4 The coupling barrier detection module includes: The projection width calculation submodule identifies the main axis extension direction of the land feature distribution area based on the land feature boundary distribution data, sets the spatial step size according to the preset sampling density, constructs multiple equally spaced orthogonal cross-sectional segments along the main axis direction, traverses each orthogonal cross-sectional segment, counts the spatial span of the pixels marked as land feature boundaries within the segment range, calculates the effective coverage width of land features corresponding to each cross-sectional position, and combines them according to the spatial arrangement order of the cross-sections on the main axis to generate a regional projection width sequence. Principal component analysis (PCA) is used to calculate the covariance matrix of pixel coordinates in the feature distribution area, and the direction of the first principal component is extracted as the principal axis extension direction. Based on a preset sampling density, a spatial step size of 10 pixels is set, and multiple equally spaced orthogonal cross-sectional segments perpendicular to the principal axis are constructed along this direction. Each orthogonal cross-sectional segment is traversed, and the starting and ending pixels marked as feature boundaries are identified. The Euclidean distance between these pixels is calculated as the feature's spatial span. For example, if the calculated span for a cross-section is 15.6 meters (after conversion), this value is used as the effective feature coverage width at that cross-section location. These width values ​​are combined according to the spatial arrangement order of the cross-sections on the principal axis (e.g., indexed from south to north, 1 to N) to generate a regional projection width sequence.

[0028] The difference sequence generation submodule calls the region projection width sequence, extracts the projection width values ​​of two adjacent positions in the sequence according to the index order, performs the subtraction operation between the preceding and following terms to obtain the width change, performs absolute value processing on the width change to eliminate the positive and negative influence of the change direction, extracts the numerical features characterizing the degree of geometric change of the boundary, and reassembles and aligns the calculated absolute values ​​according to the original sequence index order to generate a local first-order difference absolute value sequence. Call the region projection width sequence (e.g.) Extract the projected width values ​​of two adjacent positions in the sequence according to the index order, for example, extract... and The change is obtained by performing the subtraction operation between the preceding and following terms. The absolute value of the width change is calculated to obtain 3.2, which eliminates directional differences in width narrowing or expansion. This value of 3.2 characterizes the degree of geometric abrupt change in the boundary within the interval. This operation is performed on all adjacent pairs within the sequence, and the calculated absolute values ​​are then rearranged and aligned according to the original sequence index order, for example, to form a sequence. This generates a local first-order difference absolute value sequence.

[0029] The barrier identification submodule identifies the location of local maxima in the waveform based on the local first-order difference absolute value sequence, calculates the number of maxima per unit length, obtains the distribution density of extreme points, detects segments where the sequence value returns to zero or undergoes a step drop, counts their frequency, calculates the edge continuity break rate, compares the distribution density of extreme points and the edge continuity break rate with a preset threshold, and determines and outputs the spatial coupling barrier identification result. The process of comparing the extreme point distribution density and edge continuity breakage rate with a preset barrier determination threshold to determine and output the spatial coupling barrier identification result is as follows: Set a sliding window of fixed length, slide it point by point along the index direction of the local first-order difference absolute value sequence, count the number of maximum points within the coverage area of ​​each sliding window, and take the ratio of the number of maximum points to the length of the sliding window as the distribution density of the extreme points. Identify continuous index segments in local first-order difference absolute value sequences where the amplitude returns to zero or is lower than the background noise, and calculate the proportion of the cumulative length of the continuous index segments to the total length of the sequence as the edge continuity break rate. Using sample data from standard barrier-free regions, the statistical distribution of texture complexity of the sample data is calculated, and the upper limit of the confidence interval of the statistical distribution is selected as the threshold for barrier density determination. Based on the minimum geometric connectivity requirements of the ground feature boundaries, the maximum allowable percentage of fractures is set as the fracture tolerance threshold. The extreme point distribution density is compared with the barrier density judgment threshold, and the edge continuity breakage rate is compared with the breakage tolerance threshold. When the distribution density of extreme points is greater than the barrier density judgment threshold and the edge continuity breakage rate is less than the breakage tolerance threshold, it is determined that a dense coupling barrier exists. When the edge continuity failure rate is greater than the failure tolerance threshold, it is determined that there is a blocking coupling barrier. Extract the cross-sectional location indices corresponding to dense coupling barriers and blocking coupling barriers, and output the spatial coupling barrier identification results.

[0030] Set a fixed-length sliding window (e.g., 50 sequence points), slide it point by point along the index direction of the local first-order difference absolute value sequence, and count the number of maxima within the coverage area of ​​each sliding window (e.g., count 8 maxima). Calculate the ratio of the number of maxima (8) to the sliding window length (50). This serves as the distribution density of extreme points. It identifies continuous index segments in the local first-order difference absolute value sequence where the amplitude returns to zero or is below the background noise (e.g., 0.1), and calculates the proportion of the cumulative length of these continuous index segments (e.g., 120 points) to the total sequence length (e.g., 1000 points). The edge continuity breakage rate is used as the threshold. Using sample data from standard barrier-free areas, the texture complexity statistical distribution of the sample data is calculated, and the upper limit of the 95% confidence interval of the statistical distribution (e.g., 0.10) is selected as the barrier density determination threshold. Based on the minimum geometric connectivity requirement of the feature boundaries, the maximum allowable breakage percentage (e.g., 0.05) is set as the breakage tolerance threshold. The extreme point distribution density of 0.16 is compared with the barrier density determination threshold of 0.10, and the edge continuity breakage rate of 0.12 is compared with the breakage tolerance threshold of 0.05. Due to the edge continuity breakage rate... The presence of blocking coupling barriers is determined. The cross-sectional location indices corresponding to dense coupling barriers and blocking coupling barriers are extracted, and the spatial coupling barrier identification results are output.

[0031] Please see Figure 5 The error field construction module includes: The coordinate error matrix construction submodule calls the spatial coupling barrier identification results, parses the outline of the identified independent ground feature areas, calculates the geometric centroid coordinates of each area, synchronously obtains the corresponding geographic standard coordinates of the surveying area as a reference, performs pairing operations between the centroid coordinates and the standard coordinates, solves the Euclidean distance difference between the two on the two-dimensional plane projection, maps and fills the difference according to the grid index of the spatial distribution of ground features, and generates a two-dimensional coordinate error matrix. For example, to obtain the centroid of a certain building area is... The geographic standard coordinates corresponding to the surveyed area are acquired synchronously as a reference. This reference comes from high-precision satellite positioning data, and the corresponding centroid standard coordinates are... Perform paired operations between the centroid coordinates and standard coordinates, and calculate the difference in Euclidean distance between their projections onto the two-dimensional plane. The difference vector The raster index of the spatial distribution of ground features is mapped and filled, and the Kriging interpolation method is used to estimate the uncovered areas to generate a two-dimensional coordinate error matrix.

[0032] The error gradient vector generation submodule, based on a two-dimensional coordinate error matrix, traverses the error data nodes within the matrix using a local difference window, calculates the rate of change of error values ​​in the horizontal and vertical dimensions respectively, synthesizes a gradient vector representing the local error tendency based on the rate of change in the two dimensions, calculates the magnitude and direction angle of the gradient vector, marks the spatial position and encapsulates the attributes of the gradient vectors in the entire region, and generates a set of error direction vectors. according to The local difference window iterates through the error data nodes within the matrix, calculating the rate of change of error values ​​in both the horizontal and vertical dimensions. For example, at a certain node, the horizontal rate of change is 0.05, and the vertical rate of change is 0.03. Based on the rates of change in these two dimensions, a gradient vector representing the local error tendency is synthesized. Calculate the magnitude of the gradient vector. With direction angle Spatial location marking and attribute encapsulation are performed on the gradient vectors across the entire region to generate a set of error direction vectors.

[0033] The vector flow graph structure mapping submodule extracts the direction and amplitude attributes of each position vector for the error direction vector set, establishes the flow direction transmission topology between adjacent vectors based on grid adjacency, constructs a vector field network describing the overall deformation trend, maps the amplitude attribute to field strength parameters, maps the direction attribute to streamline tangential constraints, assigns attribute values ​​to network nodes and organizes them in a structured manner, and generates the vector flow graph structure. For the current node vector (direction angle 31°), search its 8 neighboring nodes. If the direction angle of a neighboring node vector differs from that of the current node by less than a preset threshold (e.g., 15°), then a connection edge is established. For example, if a neighboring node has a direction angle of 35°, and the difference is 4° (less than 15°), then a connection is established. Construct a vector field network describing the overall deformation trend, mapping the amplitude attribute (e.g., 0.058) to the field strength parameters of the network nodes, mapping the direction attribute to streamline tangential constraints, and calculating the edge weight based on the reciprocal of the angle difference of the connection edges. Assign attribute values ​​to the network nodes and organize them in a structured manner to generate a vector flow graph structure.

[0034] Please see Figure 6 The coordinate reverse registration module includes: The registration path construction submodule, based on the vector flow graph structure, identifies the flow trend characteristics of error vectors in spatial distribution, locates the node positions where the vector magnitude exceeds the preset mutation threshold, plans the connection order from the error peak point to the edge stable region according to the gradient descent rule, establishes the topological association between nodes, constructs the trajectory line to guide coordinate correction, and generates a self-guided reverse registration path. For example, the magnitude at coordinates (500, 500) is 0.15. Following the gradient descent rules, the connection order is planned, starting from the error peak point. The error magnitudes of neighboring nodes are compared, and the node with the largest magnitude decrease is moved to the neighboring node, for example, moving to (500, 501) where the magnitude is 0.14. This process is iterated until a stable edge region (magnitude less than 0.01) is reached. Topological relationships between nodes are established, the coordinate sequences of all traversed nodes are recorded, a trajectory line guiding coordinate correction is constructed, and a self-guided reverse registration path is generated.

[0035] The gradient patch generation submodule calls the self-guided reverse registration path, traverses the boundary pixels of the ground objects one by one along the dominant direction of the error gradient, calculates the spatial displacement compensation amount at each stage based on the vector amplitude, performs geometric distortion correction on the local boundary area, eliminates positional deviation, defines the boundary of the pixel range that needs to be resampled, and generates the error gradient patch area. The spatial displacement compensation for each stage is calculated based on the vector amplitude. The calculation formula is as follows: ,in This is the error vector for the current point. This is the correction factor (e.g., 10.0). For example, for the vector (0.05, 0.03), the calculated displacement compensation is... Pixels. Geometric distortion correction is performed on local boundary regions, eliminating positional deviations through inverse coordinate transformation, i.e. , Define the boundaries of the pixel range that needs texture resampling (e.g., centered on the correction point). (Region), generating error gradient patch regions.

[0036] The image correction output submodule dynamically overlays the error gradient patch area onto the coordinate system of the original surveying data, performs real-time grid node coordinate updates, performs bilinear interpolation remapping on the image pixels within the patch coverage area, corrects the spatial position of the ground feature outline, assigns spatial reference attributes to the image data that conform to geographic standards, and generates a surveying and mapping recognition image. For the calculated non-integer new coordinates Extract its integer part and decimal part Utilizing the four neighboring pixels , , , The grayscale value is obtained through the formula. Calculate the corrected pixel values. Correct the spatial position of ground feature outlines, assign spatial reference attributes (such as WGS84 coordinate system parameters) to the image data to conform to geographic standards, and generate a mapping and recognition image.

[0037] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A land surveying data analysis system based on image recognition, characterized in that, The system includes: The gradient response extraction module acquires aerial images of the surveyed area, constructs edge pixel groups, generates a multidimensional directional gradient distribution sequence, and calculates the average rate of change of local response amplitude within a unit bandwidth. The boundary classification judgment module quantifies the blurring rate of land feature boundaries based on the average change rate of the local response amplitude, generates the response expansion amplitude by combining the average change rate of the local response amplitude and the boundary position difference, and classifies the land feature boundary distribution data based on the response expansion amplitude. The coupling barrier detection module extracts a cross section along the main axis of the land use boundary data, calculates the projection width sequence, generates a first-order difference absolute value sequence of the projection width sequence, and obtains the spatial coupling barrier identification result based on the density of extreme points and the fracture rate of the first-order difference absolute value sequence. The error field construction module acquires geographic standard coordinates, calculates the distance difference between the center point of the spatial coupling barrier identification result and the geographic standard coordinates, constructs a two-dimensional coordinate error matrix, generates an error direction vector based on the two-dimensional coordinate error matrix, and forms a vector flow graph structure. The coordinate reverse registration module constructs a reverse registration path based on the vector flow graph structure, adjusts the coordinates of boundary points along the error gradient, and superimposes the error gradient patch area onto the coordinate system to update the node positions, generating a coordinate-fitted mapping and recognition image.

2. The land surveying data analysis system based on image recognition according to claim 1, characterized in that, The average rate of change of local response amplitude includes gradient amplitude increment, pixel interval weight, and response direction factor; the land cover boundary distribution data includes boundary probability matrix, geometric contour coordinates, and transition zone width index; the spatial coupling barrier identification result includes a set of blocking line segments, regional segmentation topology, and barrier semantic attributes; the vector flow graph structure includes offset vector nodes, local deformation tensor, and error flow path; and the surveying and identification image includes a corrected raster matrix, spatial reference coordinate system, and land cover attribute metadata.

3. The land surveying data analysis system based on image recognition according to claim 2, characterized in that, The gradient response extraction module includes: The edge pixel group construction submodule acquires aerial images of the survey area, calculates the grayscale contrast value of each pixel in the image, compares the grayscale contrast value with the preset edge extraction benchmark value, selects pixels with grayscale contrast values ​​greater than the edge extraction benchmark value, and constructs land use edge pixel groups based on the spatial adjacency relationship of the pixels. The directional gradient sequence generation submodule calls the edge pixel group of the land type, traverses the adjacent pixels in the group, calculates the gray-level change amplitude in the horizontal, vertical and diagonal dimensions respectively, arranges the calculation results according to the spatial distribution order, and generates a multidimensional directional gradient distribution sequence. The response amplitude change calculation submodule, based on the multidimensional directional gradient distribution sequence, determines the preset unit boundary band width, extracts the gradient response values ​​within the bandwidth range, and calculates the average rate of change of local response amplitude.

4. The land surveying data analysis system based on image recognition according to claim 3, characterized in that, The formula for obtaining the average rate of change of the local response amplitude is as follows: ； in, Represents the average rate of change of the local response amplitude. This represents the total number of pixels within the calculation unit. This represents the normalized directional gradient observation of the j-th pixel. The normalized baseline value representing the background gradient. Represents the tangential distance of the boundary curve. Represents the width of the unit boundary band. Represents the noise suppression coefficient. This represents the normalization scaling factor.

5. The land surveying data analysis system based on image recognition according to claim 4, characterized in that, The boundary classification judgment module includes: The blur rate quantization submodule selects a local neighborhood window of the target pixel based on the average change rate of the local response amplitude, calculates the statistical variance of the change rate value within the window, compares the statistical variance with the preset texture clarity benchmark value, analyzes the grayscale correlation of the pixels within the window, and obtains the blur coefficient of the ground feature boundary. The expansion amplitude generation submodule calls the average change rate of the local response amplitude and the blur coefficient of the ground object boundary to determine the geometric distance of the pixel relative to the estimated boundary center, calculate the difference between the pixel gradient and the neighborhood mean, extract the information entropy and high-frequency energy features of the local gray-scale distribution, and calculate to obtain the response expansion amplitude. The classification judgment submodule performs threshold segmentation processing on the response expansion amplitude, identifies pixel regions that exceed the preset classification threshold, performs morphological closing operation on the pixel regions, connects the broken edges, marks the category attributes and location coordinates of the pixels, and constructs land use boundary distribution data.

6. The land surveying data analysis system based on image recognition according to claim 5, characterized in that, The coupling barrier detection module includes: The projection width calculation submodule identifies the main axis extension direction of the land feature distribution area based on the land feature boundary distribution data, sets the spatial step size according to the preset sampling density, constructs multiple equally spaced orthogonal cross-sectional segments along the main axis direction, traverses each orthogonal cross-sectional segment, counts the spatial span of pixels marked as land feature boundaries within the segment range, calculates the effective coverage width of land features corresponding to each cross-sectional position, and generates a regional projection width sequence according to the spatial arrangement order of the cross-sections on the main axis. The difference sequence generation submodule calls the region projection width sequence, extracts the projection width values ​​of two adjacent positions in the sequence according to the index order, performs the subtraction operation between the preceding and following terms to obtain the width change, performs absolute value processing on the width change to eliminate the positive and negative influence of the change direction, extracts the numerical features characterizing the degree of geometric change of the boundary, and reassembles and aligns the calculated absolute values ​​according to the original sequence index order to generate a local first-order difference absolute value sequence. The barrier determination and identification submodule identifies the local maxima positions in the sequence waveform based on the local first-order difference absolute value sequence, calculates the number of maxima within a unit length, obtains the extreme point distribution density, detects segments where the sequence value returns to zero or undergoes a step drop, counts their frequency of occurrence, calculates the edge continuity break rate, compares the extreme point distribution density and the edge continuity break rate with a preset threshold, and determines and outputs the spatial coupling barrier identification result.

7. The land surveying data analysis system based on image recognition according to claim 6, characterized in that, The process of comparing the extreme point distribution density and edge continuity breakage rate with a preset barrier determination threshold, and determining and outputting the spatial coupling barrier identification result, is as follows: Set a sliding window of fixed length, slide it point by point along the index direction of the local first-order difference absolute value sequence, count the number of maximum points within the coverage area of ​​each sliding window, and use the ratio of the number of maximum points to the length of the sliding window as the extreme point distribution density. Identify continuous index segments in the local first-order difference absolute value sequence where the amplitude returns to zero or is lower than the background noise, and calculate the proportion of the cumulative length of the continuous index segments to the total length of the sequence as the edge continuity break rate. Using sample data from standard barrier-free regions, the statistical distribution of texture complexity of the sample data is calculated, and the upper limit of the confidence interval of the statistical distribution is selected as the threshold for barrier density determination. Based on the minimum geometric connectivity requirements of the ground feature boundaries, the maximum allowable percentage of fractures is set as the fracture tolerance threshold. The extreme point distribution density is compared with the barrier density judgment threshold, and the edge continuity breakage rate is compared with the breakage tolerance threshold. When the distribution density of extreme points is greater than the barrier density judgment threshold and the edge continuity breakage rate is less than the breakage tolerance threshold, it is determined that a dense coupling barrier exists. When the edge continuity failure rate is greater than the failure tolerance threshold, it is determined that there is a blocking coupling barrier. Extract the cross-sectional location indices corresponding to dense coupling barriers and blocking coupling barriers, and output the spatial coupling barrier identification results.

8. The land surveying data analysis system based on image recognition according to claim 7, characterized in that, The error field construction module includes: The coordinate error matrix construction submodule calls the spatial coupling barrier identification results, parses the outline of the identified independent ground feature area, calculates the geometric centroid coordinates of each area, synchronously obtains the geographic standard coordinates corresponding to the surveying area as a reference, performs pairing operation between the centroid coordinates and the standard coordinates, solves the Euclidean distance difference between the two on the two-dimensional plane projection, maps and fills the difference according to the grid index of the spatial distribution of ground features, and generates a two-dimensional coordinate error matrix. The error gradient vector generation submodule, based on the two-dimensional coordinate error matrix, traverses the error data nodes in the matrix according to the local difference window, calculates the rate of change of error values ​​in the horizontal and vertical dimensions respectively, synthesizes a gradient vector representing the local error tendency based on the rate of change of the two dimensions, calculates the magnitude and direction angle of the gradient vector, performs spatial position marking and attribute encapsulation on the gradient vectors in the whole region, and generates a set of error direction vectors. The vector flow graph structure mapping submodule extracts the direction and amplitude attributes of each position vector for the error direction vector set, establishes the flow direction transmission topology between adjacent vectors based on grid adjacency relationships, constructs a vector field network, maps the amplitude attribute to field strength parameters, maps the direction attribute to streamline tangential constraints, assigns attribute values ​​to network nodes and organizes them in a structured manner, and generates a vector flow graph structure.

9. The land surveying data analysis system based on image recognition according to claim 8, characterized in that, The coordinate reverse registration module includes: The registration path construction submodule, based on the vector flow graph structure, identifies the flow trend characteristics of the error vector in the spatial distribution, locates the node positions where the vector magnitude exceeds the preset mutation threshold, plans the connection order from the error peak point to the edge stable region according to the gradient descent rule, establishes the topological association between nodes, constructs the trajectory line to guide coordinate correction, and generates a self-guided reverse registration path. The gradient patch generation submodule calls the self-guided reverse registration path, traverses the boundary pixels of the ground objects one by one along the dominant direction of the error gradient, calculates the spatial displacement compensation amount at each stage based on the vector amplitude, performs geometric distortion correction on the local boundary area, eliminates positional deviation, defines the boundary of the pixel range that needs to be resampled, and generates the error gradient patch area. The image correction output submodule dynamically overlays the error gradient patch area onto the coordinate system of the original surveying data, performs real-time grid node coordinate updates, performs bilinear interpolation remapping on the image pixels within the patch coverage area, corrects the spatial position of the ground feature outline, and generates a surveying and recognition image.

Images