Big data-based geographic mapping image classification processing system and method

Abstract

The present application relates to the technical field of image classification, in particular to a geographic surveying and mapping image classification processing system and method based on big data, the system comprising: a regional feature module, a fluctuation analysis module, a spectral anomaly module, an edge adjustment module, and a classification correction module.In the present application, the gray brightness and multi-band reflectivity are extracted by dividing the region, combined with variance contrast and spectral smoothness analysis, and the dimensionality is reduced by principal component analysis to reduce redundancy and retain key features.The interval division and fluctuation frequency extraction are combined with statistical test constraints to identify potential anomalies in advance and reduce noise interference.The reflectivity difference and spectral extreme detection are combined with smoothness threshold to capture abnormal changes.The gradient direction offset and edge closure ratio monitor boundary continuity and complement the spectral detection.The aggregation degree and sparsity ratio are calculated and dynamically corrected priority to avoid bias caused by a single feature, making the classification result more accurate and stable.

Description

Technical Field

[0001] This invention relates to the field of image classification technology, and in particular to a geographic mapping image classification and processing system and method based on big data. Background Technology

[0002] Image classification technology encompasses computer-based methods for segmenting and recognizing images. Its core function is to divide input image data into categories with specific semantics or attributes through steps such as image segmentation, feature extraction, and classification. Technically, this field includes image acquisition, image preprocessing, image segmentation, feature representation and analysis, and classification. Common implementation methods involve edge detection, region segmentation, texture feature extraction, and classifier construction based on training samples.

[0003] Among them, the big data-based geographic mapping image classification and processing system and method refers to a specific scheme for classifying large-scale image data in geographic mapping scenarios. It covers geographic mapping image acquisition and preprocessing, geospatial feature extraction, big data-based image segmentation, and the establishment of geographic image classification rules. The approach involves using a large-scale data management framework to complete image storage and scheduling, combining segmentation algorithms to divide the mapping images into regions, and then constructing classification criteria based on geographic features to assign different regions to corresponding categories, thereby forming a classification and processing system for geographic mapping images.

[0004] Existing technologies, in practical applications, primarily rely on conventional segmentation, feature extraction, and classifier discrimination processes, resulting in insufficient representation of regional features. Because feature extraction is performed only globally or in a single dimension, it often fails to capture subtle differences within a region, leading to blurred boundaries or category confusion in classification results. When processing large-scale mapping images, traditional methods rely heavily on pre-defined rules or training samples, which, limited by sample coverage and rule adaptability, are prone to biases in complex scenes. For example, among features with similar spectral reflectance, conventional methods struggle to distinguish between vegetation and artificial structures, causing classification errors. Regarding edge processing, the lack of depth constraints on boundary continuity and directional offset easily leads to edge breaks or excessive smoothing, affecting the completeness of subsequent region identification. Faced with large-scale data scheduling and storage, the processing logic of existing technologies has limited capabilities in outlier identification and dynamic correction, often directly classifying outliers into a single category without further assessment of aggregation and sparsity, ultimately resulting in insufficient precision and stability in classification results. Summary of the Invention

[0005] To address the technical problems existing in the prior art, embodiments of the present invention provide a geographic mapping image classification and processing system and method based on big data. The technical solution is as follows:

[0006] On the one hand, a geographic mapping image classification and processing system based on big data is provided, which includes:

[0007] The regional feature module divides geographic mapping image regions and collects pixel grayscale, brightness, and multi-band reflectance. It analyzes the grayscale direction variance, contrast difference, and spectral stability of each region, and inputs them into principal component analysis for feature dimensionality reduction to generate regional feature sets, which are then transferred to the fluctuation analysis module.

[0008] The fluctuation analysis module calls the regional feature set, divides the gray-scale direction variance and contrast difference intervals and extracts the fluctuation frequency, calculates the constrained fluctuation intensity through the chi-square test and transmits it to the spectral anomaly module.

[0009] The spectral anomaly module obtains the reflectance difference of multiple regions and bands based on the constrained fluctuation intensity and detects extreme values. It also determines that the spectral stability exceeds the stability threshold, generates spectral anomaly results, and transmits them to the edge adjustment module.

[0010] The edge adjustment module calculates the pixel gradient direction offset and the image edge closure ratio using the Sobel operator and detects whether the continuity is abnormal. It adjusts the priority based on the spectral anomaly results, generates a classification priority adjustment result, and passes it to the classification correction module.

[0011] The classification correction module calculates the aggregation degree and sparsity ratio of outliers based on the classification priority adjustment results, corrects the classification priority, and generates the mapping image classification results.

[0012] As a further embodiment of the present invention, the regional feature set includes gray-level direction variance, contrast difference, and spectral stability coefficient; the constraint fluctuation intensity includes gray-level variance interval frequency, contrast difference interval frequency, and fluctuation test chi-square value; the spectral anomaly results include band reflectance extreme values, stability threshold judgment values, and anomaly region coefficients; the classification priority adjustment results include edge closure ratio, gradient direction offset, and priority ranking value; and the mapping image classification results include aggregation degree parameter and sparsity ratio parameter.

[0013] As a further aspect of the present invention, the region feature module specifically comprises:

[0014] The pixel acquisition submodule divides the geographic mapping image into regions, acquires the grayscale value, brightness value and multi-band reflectance data of pixels in the region, calculates the grayscale direction distribution, and statistically analyzes the distribution of brightness distribution sequence and reflectance distribution sequence to generate the pixel distribution coefficient of the region.

[0015] The difference calculation submodule calculates the variance of the gray-scale direction distribution of multiple zones based on the distribution coefficient of the partitioned pixels, detects the contrast difference between the brightness distribution sequence and the gray-scale direction variance value, counts the mean square fluctuation of the reflectance distribution sequence and converts it into spectral stability, and combines the three types of indicators to generate the spectral difference of the zones.

[0016] The feature dimensionality reduction submodule calls the partition spectral difference degree, extracts the gray-level direction variance value, contrast difference and spectral stability value and inputs them into principal component analysis, extracts principal component coefficients, retains the partition features corresponding to the principal component coefficients, and generates a regional feature set.

[0017] The region feature set consists of the projection features of the gray-level direction variance, contrast difference, and spectral stability coefficient corresponding to the principal component coefficients extracted by principal component analysis.

[0018] As a further aspect of the present invention, the process of combining the spectral difference of the partition is specifically as follows: the gray-scale direction variance value calculated based on the gray-scale direction distribution sequence, the contrast difference value obtained by comparing the brightness distribution sequence with the gray-scale direction variance value, and the spectral stability value converted from the mean square fluctuation calculated based on the reflectance distribution sequence are weighted and combined to generate the spectral difference of the partition.

[0019] As a further aspect of the present invention, the fluctuation analysis module specifically comprises:

[0020] The interval division submodule calls the region feature set, extracts the gray-level direction variance and contrast difference and detects the value range respectively, groups the upper and lower limits of the gray-level direction variance and the upper and lower limits of the contrast difference into intervals and generates difference interval combinations.

[0021] The frequency extraction submodule, based on the combination of difference intervals, statistically analyzes the distribution of gray-scale direction variance and contrast difference in each interval, detects the cumulative number of occurrences of values ​​and converts them into the number of fluctuations, and obtains the fluctuation frequency coefficient.

[0022] The strength constraint submodule calls the fluctuation frequency coefficient to obtain the current frequency and expected distribution values ​​of multiple intervals, calculates the weighted correction chi-square value by constructing a formula that includes the frequency fluctuation adjustment coefficient and the frequency correction offset, analyzes the constraint coefficients of multiple intervals and merges them into a unified scale to generate the constraint fluctuation strength.

[0023] The constrained fluctuation intensity includes a weighted corrected chi-square value that incorporates interval weights and offset corrections.

[0024] As a further aspect of the present invention, the spectral anomaly module specifically comprises:

[0025] The band difference submodule filters key analysis areas based on the constrained wave intensity, obtains the band reflectance values ​​of the key analysis areas, subtracts the band reflectance values ​​of adjacent areas one by one and records the difference, and integrates all the difference values ​​in the order of the regions to generate the band reflectance difference.

[0026] The extreme value detection submodule calls the band reflection difference to detect the extreme values ​​in the multi-region band reflection difference, calculates the extreme value range and verifies the extreme interval position, marks the extreme value segment and records the range value to obtain the reflection extreme value range.

[0027] The stability determination submodule calculates the rate of change of the difference between multiple bands in adjacent ranges based on the range of reflection extreme values, compares it with a set stability threshold, filters out band regions that exceed the stability threshold, integrates the corresponding difference values, and generates spectral anomaly results.

[0028] The stability threshold is set by the distribution of the rate of change of the band reflection difference and the allowable deviation range;

[0029] The spectral anomaly results include the extreme values ​​of the band reflectance difference selected, the identification of anomaly regions that exceed the stability threshold, and the corresponding difference values.

[0030] As a further aspect of the present invention, the edge adjustment module specifically comprises:

[0031] The gradient calculation submodule acquires geographic mapping image metadata, calculates the horizontal and vertical gradient components of the gray values ​​of multiple pixels based on Sobel convolution kernels and introduces neighborhood statistical features, then combines the two directional components into a gradient direction angle and corresponds it to the pixel position, calculates the offset distribution and records it to obtain the gradient direction offset.

[0032] The continuity detection submodule calculates the edge closure ratio of the image based on the gradient direction offset and detects the connection between adjacent edges. It records the length of continuous segments and the length of broken segments, compares the ratio of the two to determine whether there is an anomaly, and combines the anomaly ratio with the edge closure ratio to generate an edge continuity index.

[0033] The priority adjustment submodule calls the edge continuity index and combines it with the spectral anomaly results. It performs weighted calculations based on the weight relationship between the two types of values ​​to determine the priority ranking of the multi-region classification results, and integrates the ranking results to obtain the classification priority adjustment result.

[0034] The classification priority adjustment result is a region sequence based on a weighted sort of edge continuity index and normalized spectral outliers.

[0035] As a further aspect of the present invention, the classification correction module specifically comprises:

[0036] The aggregation degree calculation submodule, based on the classification priority adjustment results, extracts outlier location data and counts the distance distribution between adjacent outliers, calculates the ratio of the average adjacent distance to the average global distance, and analyzes the aggregation degree value in combination with the proportion of outlier numbers to obtain the outlier aggregation degree.

[0037] The sparsity ratio calculation submodule calls the aggregation degree of the outliers, calculates the sum of the blank spacing between outliers and divides it by the total region length, calculates the sparsity ratio and compares it with the aggregation degree to obtain the sparsity coefficient of the outlier distribution and generate the outlier sparsity ratio.

[0038] The priority correction submodule calculates the weighted average of the outlier sparsity ratio and the classification priority adjustment result, and adjusts the priority order of the classification results to generate the mapping image classification result.

[0039] The outlier aggregation degree is a calculated value combining the ratio of the average distance between adjacent outliers to the average global distance and the proportion of the number of points. The outlier sparsity ratio is the ratio of the sum of the blank spacing between outliers to the total length of the region to the aggregation degree.

[0040] As a further aspect of the present invention, the process of generating the outlier sparsity ratio specifically involves dividing the sum of the blank spacing between outliers calculated based on the outlier aggregation degree by the total region length, limiting the total region length to a preset standard length, calculating the sparsity ratio and comparing it with the aggregation degree value, and limiting the comparison threshold range to obtain the sparsity coefficient of the outlier distribution and generate the outlier sparsity ratio.

[0041] On the other hand, a big data-based geographic mapping image classification and processing method, which is executed based on the aforementioned big data-based geographic mapping image classification and processing system, includes the following steps:

[0042] S1: Divide the geographic mapping image area and collect pixel grayscale, brightness and multi-band reflectance, analyze the grayscale direction variance, contrast difference and spectral stability of each area, and input them into principal component analysis for feature dimensionality reduction to generate regional feature sets.

[0043] S2: Call the region feature set, divide the gray-level direction variance and contrast difference intervals and extract the fluctuation frequency, and calculate the constraint fluctuation intensity through the chi-square test;

[0044] S3: Based on the constrained fluctuation intensity, obtain the reflectivity difference of multiple regions and bands and detect the extreme values, and determine that the spectral stability exceeds the stability threshold to generate spectral anomaly results;

[0045] S4: Calculate the pixel gradient direction offset and image edge closure ratio using the Sobel operator and detect whether the continuity is abnormal. Adjust the priority based on the spectral anomaly results and generate a classification priority adjustment result.

[0046] S5: Based on the classification priority adjustment results, calculate the aggregation degree and sparsity ratio of outliers and correct the classification priority to generate the mapping image classification results.

[0047] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following:

[0048] By dividing geographic mapping image regions and extracting pixel grayscale, brightness, and multi-band reflectance, subtle differences within these regions can be quantitatively represented. Further analysis of grayscale directional variance, contrast difference, and spectral stability is combined with principal component analysis for feature dimensionality reduction, reducing redundant information and retaining key discriminative features, making the classification process more focused on critical attributes. Based on this, feature parameters are divided into intervals and fluctuation frequencies are extracted. Statistical tests constrain fluctuation intensity, enabling earlier identification of potential anomalies and reducing noise impact. By utilizing reflectance difference and spectral extreme value detection combined with a stability threshold, spectral anomalies are effectively captured, improving the ability to distinguish differences in complex regions. Calculations of pixel gradient direction offset and edge closure ratio not only monitor edge continuity but also complement spectral anomaly detection results, improving the accuracy of classification boundaries. In the results output stage, the aggregation degree and sparsity ratio of anomalies are calculated to dynamically adjust classification priorities, avoiding bias caused by single features and ensuring stable and robust classification results. The overall logic significantly enhances the accuracy and reliability of image classification through multi-level synergy of feature extraction, fluctuation constraints, spectral detection, edge optimization, and result correction, providing more flexible support for the structured processing of large-scale geographic mapping data. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a system schematic diagram of the present invention;

[0051] Figure 2 This is a flowchart of the region feature module in this invention;

[0052] Figure 3 This is a flowchart of the fluctuation analysis module in this invention;

[0053] Figure 4 This is a flowchart of the spectral anomaly module in this invention;

[0054] Figure 5 This is a flowchart of the edge adjustment module in this invention;

[0055] Figure 6 This is a flowchart of the classification correction module in this invention. Detailed Implementation

[0056] 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.

[0057] 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.

[0058] Please see Figure 1 This invention provides a technical solution: a geographic mapping image classification and processing system based on big data includes:

[0059] The regional feature module divides geographic mapping image regions and collects pixel grayscale, brightness, and multi-band reflectance. It analyzes the grayscale direction variance, contrast difference, and spectral stability of each region, and inputs them into principal component analysis for feature dimensionality reduction to generate regional feature sets, which are then transferred to the fluctuation analysis module.

[0060] The fluctuation analysis module calls the regional feature set, divides the gray-scale direction variance and contrast difference intervals and extracts the fluctuation frequency, calculates the constrained fluctuation intensity through the chi-square test and passes it to the spectral anomaly module.

[0061] The spectral anomaly module obtains the reflectance difference of multiple regions and bands based on the constrained fluctuation intensity and detects extreme values. It also determines that the spectral stability exceeds the stability threshold, generates spectral anomaly results, and transmits them to the edge adjustment module.

[0062] The edge adjustment module calculates the pixel gradient direction offset and the image edge closure ratio using the Sobel operator and detects whether the continuity is abnormal. It adjusts the priority based on the spectral anomaly results, generates a classification priority adjustment result, and passes it to the classification correction module.

[0063] The classification correction module calculates the aggregation degree and sparsity ratio of outliers based on the classification priority adjustment results, corrects the classification priority, and generates the mapping image classification results.

[0064] The regional feature set includes gray-level variance, contrast difference, and spectral stability coefficient. The constraint fluctuation intensity includes gray-level variance interval frequency, contrast difference interval frequency, and fluctuation test chi-square value. The spectral anomaly results include band reflectance extreme values, stability threshold judgment values, and anomaly region coefficients. The classification priority adjustment results include edge closure ratio, gradient direction offset, and priority ranking value. The mapping image classification results include aggregation degree parameter and sparsity ratio parameter.

[0065] Please see Figure 2 The specific regional feature module is as follows:

[0066] The pixel acquisition submodule divides the geographic mapping image into regions, acquires the grayscale value, brightness value and multi-band reflectance data of pixels in the region, calculates the grayscale direction distribution, and statistically analyzes the distribution of brightness distribution sequence and reflectance distribution sequence to generate the pixel distribution coefficient of the region.

[0067] The geographic mapping image is divided into regions. A 1000x1000 pixel image is divided into 100 zones, each zone being 10x10 pixels in size, totaling 100 pixels. The zones are evenly divided based on pixel coordinates. Grayscale values ​​of the pixels within each zone are collected, ranging from 0 to 255, and directly read using an image sensor. For example, the grayscale values ​​of pixels within a zone are arrays such as [100, 110, 120, 130, ...]. The brightness value is obtained by calculating the weighted sum of the RGB components, with a weight of 0.299R + 0.587G + 0.114B, where R and G... The B value ranges from 0 to 255. For example, if a pixel's RGB value is (150, 160, 170), then the brightness value is 0.299*150 + 0.587*160 + 0.114*170 = 44.85 + 93.92 + 19.38 = 158.15, approximately 158. Multi-band reflectance data includes red, green, and blue band reflectance, with values ​​between 0 and 1, obtained through remote sensing equipment measurements. For example, the red band reflectance is 0.4, 0.45, etc. The gray-scale directional distribution is calculated, which refers to the gradient change of pixel gray-scale values ​​in the spatial direction. This is achieved by calculating the gradient of each pixel's gray-scale value. This is achieved using the grayscale gradient magnitude and direction of pixels. The Sobel operator is used to calculate the gradients in the x and y directions, with an magnitude of sqrt(gx^2 + gy^2) and atan2(gy, gx). The directions are then quantized into eight intervals (0-45 degrees, 45-90 degrees, etc.). The sum of the gradient magnitudes in each interval is calculated to form a direction histogram. For example, the pixel gradient direction histogram within a partition might be [sum1, sum2, ..., sum8], where sum1 is the sum of magnitudes in the 0-45 degree interval, set to a value of 50. A brightness distribution sequence is also calculated. The brightness distribution sequence refers to... For an array of all pixel brightness values ​​within a partition, such as [158, 162, 155, ...], calculate the mean, variance, skewness, and other statistical measures of the sequence. The mean brightness value is [158, 162, 155, 160]. The mean is (158+162+155+160) / 4=158.75; the variance is (158-158.75)^2=0.5625, (162-158.75)^2=10.5625, (155-158.75)^2=14.0625, and (160-158.75)^2=1.5625.

[0068] The variance is (0.5625 + 10.5625 + 14.0625 + 1.5625) / 4 = 6.6875. The reflectance distribution sequence is calculated, which refers to an array of multi-band reflectance values ​​for all pixels within a partition, such as red band reflectance [0.4, 0.45, 0.5, 0.55]. The mean and variance of this sequence are calculated; for example, the mean red band reflectance is (0.4 + 0.45 + 0.5 + 0.55) / 4 = 0.4. 75. Variance calculation is similar to luminance sequence calculation, generating the distribution coefficients of pixels in each partition. This is achieved by weighted combination of gray-level direction distribution, luminance statistics, and reflectance statistics. The weights are set based on empirical values: gray-level direction distribution weight is 0.4, luminance statistics weight is 0.3, and reflectance statistics weight is 0.3. First, the gray-level direction histogram is normalized to the range of 0-1, for example, histogram values ​​[50, 40, 30, 20, 10, 5, 3, 2], with a total of 160. Normalized to [0.3125, 0.25, 0.1875, 0.125, 0.0625, 0.03125, 0.01875, 0.0125], the mean of 0.125 is taken as the directional distribution index. The variance of the brightness statistic is taken as 6.6875, and the variance of the red band is taken as the reflectance statistic. The variance of the red band reflectance is set to 0.0025, then the coefficient is 0.4 * 0.125 + 0.3 * 6.6875 + 0.3 * 0.0 025 = 0.05 + 2.00625 + 0.00075 = 2.057. The weighting is set according to common remote sensing data processing standards. The grayscale direction has a higher weight because directionality is sensitive to texture. The brightness has a medium weight because brightness changes are common. The reflectance has a lower weight because the correlation between bands is high. The threshold is used to determine the coefficient range. For example, a coefficient less than 1 is considered low distribution, 1-3 is medium, and greater than 3 is high. Based on actual data statistics, the coefficient 2.057 in this example belongs to the medium distribution.

[0069] Table 1: Examples of image metadata representation

[0070] ;

[0071] As shown in Table 1, this table lists example data of 4 pixels in a partition, which are used to calculate grayscale value, brightness value and reflectance. Grayscale value is directly collected, reflectance is measured by sensor, and brightness value is obtained through weighted calculation. These data are used as the basis for subsequent statistics and coefficient generation to generate the partition pixel distribution coefficient.

[0072] The difference calculation submodule calculates the variance of the gray-scale direction distribution of multiple zones based on the distribution coefficient of the pixels in the zone, detects the contrast difference between the brightness distribution sequence and the gray-scale direction variance value, counts the mean square fluctuation of the reflectance distribution sequence and converts it into spectral stability, and combines the three types of indicators to generate the spectral difference of the zones.

[0073] Based on the pixel distribution coefficients of the partitions, for example, obtaining a coefficient value of 2.057 from the aforementioned paragraph, the variance of the gray-level direction distribution in multiple partitions is calculated. Three partitions are set up, with gray-level direction distribution indices of 0.125, 0.15, and 0.13 respectively. The mean is (0.125+0.15+0.13) / 3=0.135, and the variances are (0.125-0.135) / 2=0.0001, (0.15-0.135)^2=0.000225, (0.13-0.135)^2=0.000025, and the sum is 0.00035. The variance is 0.00035 / 3≈0.0001167. The contrast difference between the brightness distribution sequence and the gray-level direction variance value is detected. The reflectance distribution sequence is calculated by taking the luminance variance of each zone. For example, the luminance variances of three zones are 6.6875, 7.0, and 6.5, and the gray-scale direction variance is 0.0001167. The contrast difference is calculated as the absolute difference between the luminance variance and the gray-scale direction variance. For example, the difference between the luminance variance of the first zone (6.6875 - 0.0001167) and the gray-scale direction variance (0.0001167) is |6.6875 - 0.0001167| = 6.6873833. The mean square fluctuation of the reflectance distribution sequence is then calculated. The reflectance distribution sequence refers to the variance sequence of multi-band reflectance. For example, the reflectance variance of the red band is 0.0025, 0.003, and 0.002. The mean square fluctuation is calculated as the root mean square of these variances: sqrt((0.0025^2 + 0.002)). .003^2+0.002^2) / 3)=sqrt((0.00000625+0.000009+0.000004) / 3)=sqrt(0.00001925 / 3)=sqrt(0.0000064167)≈0.002534, converting to spectral stability. Spectral stability is obtained by taking the reciprocal of the mean square fluctuation and scaling. For example, stability = 1 / mean square fluctuation = 1 / 0.002534≈394.5, then normalized to the 0-1 range. Setting the maximum mean square fluctuation of 0.01 corresponds to a stability of 100, then with a scaling factor of 100, the stability is 394.5 / 100=3.945. However, it is usually limited to 1, so the minimum value is taken. 1. Stability / Benchmark): The benchmark value is set to 500 based on the typical fluctuation range of remote sensing data. For example, a mean square fluctuation of 0.002 corresponds to a stability of 500. Therefore, 3.945 > 1, so stability = 1. The three indicators are combined: gray-scale direction variance 0.0001167, contrast difference 6.6873833, and spectral stability 1. The spectral difference of the region is generated by combining them through a weighted average. The weights are set based on the importance of the indicators: gray-scale direction variance weight 0.2, contrast difference weight 0.5, and spectral stability weight 0.3. The difference = 0.2 * 0.0001167 + 0.5 * 6.6873833 + 0.3 * 1 = 0.00002334 + 3.34369165 + 0.3 = 3.643715, the weighting is based on data sensitivity. Contrast difference has a high weight because it reflects differences in brightness and direction; stability has a medium weight because spectral stability is important; and directional variance has a low weight because its value is small. Thresholds determine the range of difference; for example, less than 2 is low difference, 2-5 is medium, and greater than 5 is high. In this example, 3.643715 falls into the medium difference category.

[0074] The feature dimensionality reduction submodule calls the spectral difference of the partition, extracts the gray-level direction variance, contrast difference and spectral stability values ​​and inputs them into principal component analysis, extracts the principal component coefficients, retains the partition features corresponding to the principal component coefficients, and generates a regional feature set.

[0075] The regional feature set consists of the projection features of the gray-level direction variance, contrast difference, and spectral stability coefficient corresponding to the principal component coefficients extracted by principal component analysis;

[0076] Call the partition spectral difference value, for example, obtain a difference value of 3.643715 from the previous paragraph, extract the gray-level direction variance, contrast difference, and spectral stability values. The gray-level direction variance is 0.0001167, the contrast difference is 6.6873833, and the spectral stability is 1. Input the data into principal component analysis. Principal component analysis involves calculating the covariance matrix and eigenvalues ​​of these values, but avoid directly calling the algorithm. Describe the specific actions: First, combine the three values ​​into a vector [0.00... [0.0001167, 6.6873833, 1], calculate the mean vector, mean (0.0001167+6.6873833+1) / 3≈2.5625, calculate the covariance matrix, the elements are the covariance between variables, for example, the covariance between variable 1 and variable 2 is sum(value 1-mean 1)(value 2-mean 2) / n, but here there is only one set of data, set with multiple partition data, for example, a vector group of 3 partitions: partition 1 ... , 6.6873833, 1], partition 2 [0.00012, 7.0, 0.9], partition 3 [0.00013, 6.5, 1.1], mean vector [0.0001222, 6.729461, 1.0], calculate the covariance matrix, for example, the variance of variable 1 (gray-level direction variance) is sum(value-mean2) / 3, ((0.0001167-0.0001222)2=3.025e-11, (0. (0.00012-0.0001222)^2=4.84e-12, (0.00013-0.0001222)^2=6.084e-11, sum 9.593e-11, variance 3.198e-11), similarly calculate others, the covariance matrix is ​​set as [3.2e-11, 0.0001, -0.001], [0.0001, 0.5, -0.2], [-0.001, -0.2, 0.1];

[0077] Calculate eigenvalues ​​and eigenvectors. Eigenvalues ​​are obtained by solving the characteristic equation. For example, the largest eigenvalue is approximately 0.6, corresponding to the eigenvector [0.001, 0.7, 0.714]. Extract the principal component coefficients, i.e., the eigenvector elements. Retain the partition features corresponding to the principal component coefficients. For example, select the eigenvector corresponding to the largest eigenvalue as the principal component, with coefficients [0.001, 0.7, 0.714]. Generate the region feature set by projecting the original values ​​onto the principal components. For example, the projection of partition 1 = 0.0010.0001167 + 0.76.6873833 + 0.7141 ≈ 0.0000001167 + 4.68116831 + 0.714 = 5.3951684267. Calculate other partitions similarly. The region feature set is the set of these projected values. In this example, the projected value is 5.395.

[0078] Please see Figure 3 The fluctuation analysis module specifically comprises:

[0079] The interval division submodule calls the region feature set, extracts the gray-level direction variance and contrast difference, and detects the value ranges respectively. It then groups the upper and lower limits of the gray-level direction variance and the upper and lower limits of the contrast difference into intervals and generates a combination of difference intervals.

[0080] The system retrieves the region feature set, for example, the region feature set obtained earlier, which contains the gray-level directional variance and contrast difference values ​​of 5 partitions. The gray-level directional variance values ​​are arrays [0.00012, 0.00015, 0.00018, 0.00022, 0.00025], with no unit, obtained through the original calculations before principal component analysis. For example, the gray-level directional variance of partition 1, 0.00012, comes from gradient magnitude statistics, and the contrast difference values ​​are arrays [6.2, 6.5, 6.8, 7.0, 7.3], with no unit, obtained from the absolute difference calculation of brightness variance and gray-level directional variance. That is, each element in the array is read, and the value range is detected. For the gray-level directional variance, the minimum value min(0.00012, 0.00015, 0.00018 ... 15, 0.00018, 0.00022, 0.00025) = 0.00012, maximum value max(0.00012, 0.00015, 0.00018, 0.00022, 0.00025) = 0.00025, range span = maximum value - minimum value = 0.00013. For contrast difference, minimum value min(6.2, 6.5, 6.8, 7.0, 7.3) = 6.2, maximum value max(6.2, 6.5, 6.8, 7.0, 7.3) = 7.3, range span = 1.1. The upper and lower limits of the gray-scale direction variance are grouped into intervals, with 3 groups. Based on the data distribution, the lower interval is defined as 0.0001 to 0.0001. 5. The middle range is 0.00015 to 0.0002, and the high range is 0.0002 to 0.00025. The range boundaries include the lower limit but not the upper limit. For example, a value of 0.00015 belongs to the middle range. Similarly, the upper and lower limits of the contrast difference are grouped: low range 6.0 to 6.5, middle range 6.5 to 7.0, and high range 7.0 to 7.5. The boundary processing is the same. The groups are matched to generate difference range combinations, that is, each grayscale range is paired with each contrast range to form 9 combinations. For example, combination 1: low grayscale range (0.0001-0.00015) corresponds to low contrast range (6.0-6.5), combination 2: low grayscale corresponds to medium contrast, combination 3: low grayscale corresponds to high contrast, combination 4: medium grayscale corresponds to high contrast. The data points are distributed as follows: Partition 1 (grayscale 0.00012, contrast 6.2, low contrast), Partition 2 (grayscale 0.00015, contrast 6.5, high contrast), Partition 3 (grayscale 0.00018, contrast 6.8, high contrast), Partition 4 (grayscale 0.00022, contrast 7.0, high contrast), Partition 5 (grayscale 0.00025, contrast 7.0, high contrast).3 belongs to the high interval and is classified into combination 9. A list of difference interval combinations is generated, including the combination type and the corresponding data point count.

[0081] Table 2: Examples of combinations of difference intervals

[0082] ;

[0083] As shown in Table 2, this table lists examples of difference interval combinations, including grayscale range, contrast range, and the number of data points in each combination. The number of data points comes from instance allocation, with a total of 5 data points. Combination 1 has 1 point, combination 5 has 2 points, combination 9 has 2 points, and other combinations have no points, thus generating difference interval combinations.

[0084] The frequency extraction submodule, based on the combination of difference intervals, statistically analyzes the distribution of gray-scale direction variance and contrast difference in each interval, detects the cumulative number of occurrences of values ​​and converts them into the number of fluctuations, thus obtaining the fluctuation frequency coefficient.

[0085] Based on the combination of difference intervals, for example, obtaining the combination list from Table 2, the distribution of gray-scale direction variance and contrast difference within each interval is statistically analyzed. That is, the number of data points for each combination is calculated as the distribution, and the cumulative number of occurrences of the detected values ​​is calculated. The cumulative number refers to the number of points in each combination. For example, combination 1 has 1 occurrence, combination 5 has 2 occurrences, combination 9 has 2 occurrences, and others have 0 occurrences. This is converted into the number of fluctuations, which is defined as the cumulative number itself, because the number reflects the frequency of fluctuations. The fluctuation frequency coefficient is obtained by normalizing the cumulative number to the range of 0-1. The normalization formula is coefficient = number / total number. The total number is 5. In the example, the coefficient for combination 1 is 1 / 5 = 0.2, the coefficient for combination 5 is 2 / 5 = 0.4, the coefficient for combination 9 is 2 / 5 = 0.4, and the coefficient for other combinations is 0. The coefficient is set with reference to the proportion of data. There is no threshold, but a coefficient close to 1 indicates high frequency, and a coefficient close to 0 indicates low frequency. In the example, the coefficients for combinations 5 and 9 are 0.4, which is a medium frequency.

[0086] The strength constraint submodule calls the fluctuation frequency coefficient to obtain the current frequency and expected distribution values ​​of multiple intervals. It calculates the weighted correction chi-square value by constructing a formula that includes the frequency fluctuation adjustment coefficient and the frequency correction offset. It analyzes the constraint coefficients of multiple intervals and merges them into a unified scale to generate the constraint fluctuation strength.

[0087] The constrained fluctuation intensity includes a weighted corrected chi-square value that incorporates interval weights and offset corrections.

[0088] The fluctuation frequency coefficients are retrieved. For example, the coefficient array [0.2, 0, 0, 0, 0.4, 0, 0, 0, 0.4] corresponds to 9 combinations. The current frequency and expected distribution values ​​for multiple intervals are obtained. The current frequency is the fluctuation frequency coefficient, and the expected distribution value is set to a uniform distribution. The expected frequency for each combination is approximately 1 / 9 ≈ 0.1111. Since the expected frequency is the average of the 9 combinations, a weighted corrected chi-square value is calculated using a chi-square test and substituted into the formula. ,in This represents the weighted corrected chi-square value, used to measure the difference between the observed and expected frequencies. Representing the The actual observed frequency value of the interval under the current sampling state is obtained from the fluctuation frequency coefficient, for example, when i=1. When i=2 And so on. Representing the The theoretical frequency value of the interval under the preset expected distribution is set to 0.1111. For all i, this is based on a uniform distribution. Represents the corresponding number The frequency fluctuation adjustment coefficient for the interval is used to set the importance of the reference interval. For example, based on the range of data points in the interval, a smaller gray-level variance interval has a higher weight because small changes may be significant. A weight array is then set. For the low grayscale range, the weight is 1.2; for the middle grayscale range, it's 1.0; and for the high grayscale range, it's 0.8. Similar to contrast ranges, but the weights are averaged or multiplied. A simple approach is to use the grayscale weights: for example, combination 1 has a low grayscale weight of 1.2, combination 5 has a middle grayscale weight of 1.0, and combination 9 has a high grayscale weight of 0.8. Other combination weights are based on the grayscale range: i=1 weight 1.2, i=2 weight 1.2, i=3 weight 1.2, i=4 weight 1.0, i=5 weight 1.0, i=6 weight 1.0, i=7 weight 0.8, i=8 weight 0.8, and i=9 weight 0.8. Representing the The frequency correction offset for the interval is set to adjust for small frequency errors; setting it to 0 for all i simplifies calculations. This represents the total number of intervals being tested, n=9. First, calculate the numerator for each interval. This is the adjusted difference, then squared and divided by the denominator. This is the scaling factor. The final summation is for weighted evaluation of frequency differences, avoiding amplification errors at small expected frequencies. Calculate a practical example, substituting the parameter values. For i=1, , , , ,molecular ,square denominator , item value For i=2, , , , ,molecular ,square The denominator is the same, 0.01778, and the term value is 1.0. Similarly, calculate all i, i=3, i=2, term value 1.0, i=4... , , ,molecular ,square denominator Term value 1.0, i=5 , , ,molecular ,square The denominator is 0.01234, the term value is 6.76, i=6 is similar to i=4, the term value is 1.0, i=7 is similar to i=4, the term value is 1.0, i=8 is similar to i=4, the term value is 1.0, i=9 , , ,molecular ,square denominator The term value is 6.76, and the total is 0.640 + 1.0 + 1.0 + 1.0 + 6.76 + 1.0 + 1.0 + 1.0 + 6.76 = 20.16. Therefore... The advantage of the formula lies in the introduction of weights. and offset This study adjusts the sensitivity of the traditional chi-square test to small expected frequencies, making the assessment more robust and applicable to imbalanced data distributions. It analyzes constraint coefficients across multiple intervals, where constraint coefficients may refer to the chi-square value itself or derived values. The chi-square value is 20.16. These coefficients are then merged into a unified scale, which may refer to standardized values, such as converting the chi-square value to a 0-1 range, but is used directly. This generates constraint fluctuation intensity, which is defined as the chi-square value itself (20.16). The results show a significant difference between the observed frequency and the expected frequency, indicating large fluctuation intensity.

[0089] Please see Figure 4 The spectral anomaly module is specifically:

[0090] The band difference submodule filters key analysis areas based on constrained wave intensity, obtains the band reflectance values ​​of key analysis areas, subtracts the band reflectance values ​​of adjacent areas one by one and records the difference, and integrates all the difference values ​​in the order of regions to generate the band reflectance difference.

[0091] Based on the constrained wave intensity, for example, the constrained wave intensity value obtained above is 20.16. The intensity reflecting frequency differences is used to select key analysis areas. There are five key analysis areas. The red band reflectance value of each area is directly measured by a remote sensing sensor, with values ​​between 0 and 1. Specifically, the reflectance of area 1 is 0.45, area 2 is 0.50, area 3 is 0.48, area 4 is 0.52, and area 5 is 0.47. The reflectance values ​​of adjacent areas are subtracted one by one, i.e., the reflectance difference between area 2 and area 1 is calculated as |0.50 - 0.45| = 0.0. 5. The difference between region 3 and region 2 is |0.48-0.50|=0.02, the difference between region 4 and region 3 is |0.52-0.48|=0.04, and the difference between region 5 and region 4 is |0.47-0.52|=0.05. Record the differences to form a difference array [0.05, 0.02, 0.04, 0.05]. Integrate all the differences in the order of the regions, i.e., keep the array order, to generate the band reflectance difference. The difference is obtained by subtracting each item one by one. There is no unit. The value is between 0.01 and 0.1, which is consistent with the typical remote sensing reflectance variation range.

[0092] The extreme value detection submodule calls the band reflection difference to detect extreme values ​​in the multi-region band reflection difference, calculates the extreme value range and verifies the location of extreme intervals, marks the extreme value segments and records the range values ​​to obtain the reflection extreme value range;

[0093] The function calls the band reflectance difference function, for example, the difference array [0.05, 0.02, 0.04, 0.05]. It then detects the extreme values ​​in the multi-region band reflectance difference, including the minimum and maximum values. The minimum value is min(0.05, 0.02, 0.04, 0.05) = 0.02, and the maximum value is max(0.05, 0.02, 0.04, 0.05) = 0.05. The extreme value range is calculated as: range = maximum value - minimum value = 0.05 - 0.02 = 0.03. This verifies the extreme values. The interval position refers to the location of the region pair where the extreme value is located. The minimum value of 0.02 is located in region pair 2-3, and the maximum value of 0.05 is located in region pairs 1-2 and 4-5. The extreme value segments are marked by recording the region pair index, for example, the minimum value segment index is 2 (corresponding to 2-3), and the maximum value segment indices are 1 and 4 (corresponding to 1-2 and 4-5). The range values ​​are recorded, that is, the extreme values ​​0.02 and 0.05 and the range 0.03, to obtain the reflected extreme value range. The output is the extreme value information and position.

[0094] The stability determination submodule calculates the rate of change of the difference between bands in multiple regions within adjacent ranges based on the range of extreme reflection values, compares it with the set stability threshold, filters out band regions that exceed the stability threshold, integrates the corresponding difference values, and generates spectral anomaly results.

[0095] The stability threshold is set by the distribution of the rate of change of the band reflection difference and the allowable deviation range;

[0096] The spectral anomaly results include the extreme values ​​of reflectance differences in the selected bands, the identification of anomaly regions that exceed the stability threshold, and the corresponding difference values;

[0097] Based on the extreme range of reflection values, the minimum value is 0.02, the maximum value is 0.05, and the range is 0.03. The rate of change of the band difference across multiple regions within adjacent ranges is calculated. The rate of change is defined as the relative change between adjacent differences. For example, the rate of change for region 1-2 to 2-3 is |0.05-0.02| / 0.05=0.6, for region 2-3 to 3-4 it is |0.02-0.04| / 0.02=1.0, and for region 3-4 to 4-5 it is |0.04-0.05| / 0.04=0.25. The rate of change array [0.6, 1.0, 0.25] is compared with a set stability threshold. The stability threshold is determined by the rate of change of the band reflection difference. The distribution and allowable deviation range are set. The allowable deviation range is based on the stability of typical remote sensing data. A change rate exceeding 0.3 is considered non-stationary. Therefore, the stationarity threshold is set to 0.3. Band regions exceeding the stationarity threshold are filtered, that is, regions with a change rate greater than 0.3. In the example, the change rates are 0.6>0.3, 1.0>0.3, and 0.25<0.3. Therefore, regions 1-2 to 2-3 and 2-3 to 3-4 are filtered. The corresponding difference amounts are integrated, that is, the differences between these regions. The difference between region 1-2 is 0.05, the difference between region 2-3 is 0.02, and the difference between region 3-4 is 0.04. The spectral anomaly results are generated, and the output is an anomaly difference array [0.05, 0.02, 0.04].

[0098] Please see Figure 5 The edge adjustment module specifically comprises:

[0099] The gradient calculation submodule acquires geographic mapping image metadata, calculates the horizontal and vertical gradient components of the gray values ​​of multiple pixels based on Sobel convolution kernels and introduces neighborhood statistical features, then combines the two directional components into a gradient direction angle and corresponds it to the pixel position, calculates the offset distribution and records it to obtain the gradient direction offset.

[0100] Geographic image metadata is acquired, for example, by acquiring an image from a remote sensing sensor, setting the image size to 100x100 pixels, and the grayscale value range to 0 to 255. Specific metadata is obtained by reading the sensor output; for example, the grayscale values ​​of the 3x3 neighborhood of the center pixel position (r, j) are shown in Table 3. An improved gradient calculation model is used to calculate the horizontal and vertical gradient components of the multi-pixel grayscale values. The Sobel operator's horizontal convolution kernel weight coefficients are fixed, and the formula is used... ,in The representative is located at the Line number The horizontal correction gradient component values ​​for each pixel, with no units, are used to represent gradient strength. This represents the horizontal convolution kernel in the Sobel operator at the 1st... Line number The column weight coefficients are fixed preset values, based on the Sobel kernel definition. For example, when... hour , hour , hour , hour , hour , hour , hour , hour , hour It is obtained through standard image processing libraries, requiring no computation. The representative is located at the Line number The grayscale values ​​of the pixels are collected by a remote sensing sensor, ranging from 0 to 255. When the grayscale value of the neighborhood is represented as follows, The representative is located at the Line number The horizontal gradient grayscale compensation value of the column pixel is calculated as the arithmetic mean of the grayscale values ​​of a 3x3 neighborhood. The mean of the grayscale value array [50, 60, 70, 80, 90, 100, 110, 120, 130] is 810 / 9 = 90. The representative is located at the Line number The horizontal position fluctuation correction coefficient of the column pixel is calculated as the variance of the gray values ​​in the 3x3 neighborhood. For example, the variance is calculated as (50-90)^2=1600, (60-90)^2=900, (70-90)^2=400, (80-90)^2=100, (90-90)^2=0, (100-90)^2=100, (110-90)^2=400, (120-90)^2=900, (130-90)^2=1600, totaling 6000. The variance is 6000 / 9≈666.67. =666.67, where c represents the constant coefficient used for denominator stabilization calculations, set to 0.001 to prevent the denominator from being zero, and the summation symbol in the formula... This represents a cyclic summation of p and q from -1 to 1, simulating a convolution operation. The goal is to extract horizontal gradient features. r=2, j=2, and the convolution sum is calculated. Substitute the values: p=-1, q=-1: K=-1, I=50, product -50; p=-1, q=0: K=0, I=60, product 0; p=-1, q=1: K=1, I=70, product 70; p=0, q=-1: K=-2, I=80, product -160; p=0, q=0: K=0, I=90, product 0; p=0, q=1: K=2, I=1 00, product 200, p=1, q=-1: K=-1, I=110, product -110, p=1, q=0: K=0, I=120, product 0, p=1, q=1: K=1, I=130, product 130, sum -50+0+70-160+0+200-110+0+130=80, μ{r,j}=90, numerator 80-90=-10, =666.67, c=0.001, denominator sqrt(666.67+0.001)≈25.82, =-10 / 25.82≈-0.387, similar to calculating the vertical gradient component, using a vertical convolution kernel, setting the vertical convolution sum to obtain values ​​in a similar way, for example, setting the vertical convolution sum to 60, the vertical compensation amount to the same 90, the vertical fluctuation coefficient to be similar, calculating the vertical gradient component, and then synthesizing the gradient direction angle, angle θ=atan2(vertical gradient, horizontal gradient), for example, the horizontal gradient is -0.387, the vertical gradient is set to 0.5, then θ=atan2(0.5, -0.387)≈atan2(0.5, -0.387) in the second quadrant, the calculated value is about 127 degrees, calculate the offset distribution, the offset is defined as the difference between the angle and the expected direction (such as horizontal 0 degrees), for example |127-0|=127 degrees, record the offset of all pixels, and obtain the gradient direction offset array, for example [127, 45, 80, 100, 120] degrees.

[0101] Table 3: Examples of pixel grayscale values

[0102] ;

[0103] As shown in Table 3, the table lists the gray values ​​of the pixel (r, j) and its neighborhood. The unit is unknown, and the values ​​are between 50 and 130, which is consistent with the typical gray range. It is used to calculate the gradient component and obtain the gradient direction offset.

[0104] The continuity detection submodule calculates the edge closure ratio of the image based on the gradient direction offset and detects the connection between adjacent edges. It records the length of continuous segments and the length of broken segments, compares the ratio of the two to determine whether there is an anomaly, and combines the anomaly ratio with the edge closure ratio to generate an edge continuity index.

[0105] Based on gradient direction offsets [127, 45, 80, 100, 120] degrees, corresponding to 5 pixels, the image edge closure ratio is calculated. The edge closure ratio is defined as the proportion of consecutive edge pixels to the total edge pixels. Edge pixels are obtained using an edge detection algorithm (such as Canny), but the algorithm is avoided by directly setting the edge pixel positions. For example, pixels 1 and 2 are edges, pixel 3 is not, and pixels 4 and 5 are edges, for a total of 3 edge pixels. Consecutive edges refer to adjacent edges that are connected, such as pixels 1 and 2 being connected. Pixels 4 and 5 are connected, but pixels 2 and 4 are not. Therefore, the lengths of the continuous segments are 2 (pixels 1-2) and 2 (pixels 4-5), and the length of the broken segment is 1 (the gap between pixels 2 and 4). The edge closure ratio = number of continuous edge pixels / total number of edge pixels = (2+2) / 3 ≈ 1.333. However, the ratio is usually 0-1, so it needs to be adjusted. Set the total edge pixels to 5, the continuous pixels to 4, and the ratio to 0.8. Detect the connection of adjacent edges. The connection judgment is based on the similarity of the offset. For example, if the offset difference is less than 30 degrees, then a connection is made. Pixel 1 offset 127, Pixel 2 offset 45, difference 82 > 30, not connected; Pixels 2 and 4 offset 45 and 100, difference 55 > 30, not connected; Pixels 4 and 5 offset 100 and 120, difference 20 < 30, connected. Record the length of the continuous segment and the length of the broken segment. Continuous segment: length 2 for Pixel 4-5; Broken segment: no connection between Pixel 1 and 2, length is considered 1 (interval); no connection between Pixel 2 and 4, length 1. Compare the ratio of the two: sum of continuous segment lengths / sum of broken segment lengths, total length of continuous segment. Degree 2, total length of fracture segment 2, ratio 1, determine whether there is an anomaly. An anomaly is defined as a ratio below a threshold, which is set to 0.5. Based on experience, 1>0.5, no anomaly. Combine the anomaly ratio with the edge closure ratio. The anomaly ratio refers to the ratio of abnormal edges. Here there is no anomaly, so the ratio is 0. The edge closure ratio is 0.8. Generate an edge continuity index. The index is weighted and combined. The weight is anomaly ratio 0.4, edge closure ratio 0.6, and the index is 0.48. The weight setting is based on the importance of edge integrity.

[0106] The priority adjustment submodule calls the edge continuity index and combines it with the spectral anomaly results. It performs weighted calculations based on the weight relationship between the two types of values ​​to determine the priority ranking of the multi-region classification results, and integrates the ranking results to obtain the classification priority adjustment results.

[0107] The classification priority adjustment result is a region sequence based on a weighted sort of edge continuity index and normalized spectral outliers;

[0108] The edge continuity index of 0.48 is used, combined with spectral anomaly results. For example, the spectral anomaly result array [0.05, 0.02, 0.04] (difference) obtained from the previous section is used. A weighted calculation is performed based on the weight relationship between the two types of values. The weight settings are based on the application scenario: edge continuity weight 0.7, spectral anomaly weight 0.3. Because edge information is more stable, the spectral anomaly values ​​need to be normalized. For example, the mean of the anomaly differences is taken as (0.05 + 0.02 + 0.04) / 3 = 0.0367, and then scaled to the range of 0-1. Set the maximum outlier difference to 0.1, the normalized value to 0.0367 / 0.1=0.367, and the weighted comprehensive value to 0.7*0.48+0.3*0.367=0.4461. Determine the priority ranking of the multi-region classification results. The region priority is based on the comprehensive value in descending order. There are 3 regions: region A with a comprehensive value of 0.4461, region B with a comprehensive value of 0.6, and region C with a comprehensive value of 0.3. Sort B, A, and C, integrate the ranking results, obtain the classification priority adjustment result, and output the ranking list [B, A, C].

[0109] Please see Figure 6 The classification correction module specifically comprises:

[0110] The aggregation degree calculation submodule extracts outlier location data and counts the distance distribution between adjacent outliers based on the classification priority adjustment results. It calculates the ratio of the average adjacent distance to the average global distance and combines the aggregation degree value with the proportion of outlier numbers to obtain the outlier aggregation degree.

[0111] Based on the classification priority adjustment results, for example, the classification priority adjustment results obtained from the previous text are a region priority sorting list [region B, region A, region C]. The corresponding outlier location data is extracted from remote sensing images. The coordinates of the outlier in region B are (30, 40), the coordinates of the outlier in region A are (10, 20), and the coordinates of the outlier in region C are (50, 60). The unit is pixels, and the image size is 100x100. The distance distribution between adjacent outliers is statistically analyzed. Adjacency is based on priority order, that is, region B is adjacent to region A, and region A is adjacent to region C. The Euclidean distance is calculated. The distance from region B to region A is sqrt(30-10)^2+(40-20)^2≈28.28 pixels, and the distance from region A to region C is sqrt(50-10)^2+(60-20)^2=sqrt(1600+1600)=sqrt(3200)≈56.57 pixels. The distance array is [28.28, 56.57]. Calculate the average adjacent distance: (28.28 + 56.57) / 2 = 42.425 pixels. Calculate the global average distance, including all outlier pairs. The distance between region B and region A is 28.28 pixels. The distance between region B and region C is sqrt(50-30)^2 + (60-40)^2 = sqrt(400 + 400) = 28.28 pixels. The distance between region A and region C is 56.57 pixels. The average distance is (28.28 + 28.28 + 56.57) / 2 = 42.425 pixels. 0.57) / 3=113.13 / 3≈37.71 pixels. Calculate the ratio of average adjacent distance / average global distance=42.425 / 37.71≈1.125. Combined with the analysis of the aggregation degree value based on the ratio of the number of outliers, there are 3 outliers and a total of 5 regions (the image is set to be divided into 5 regions). The ratio is 3 / 5=0.6. The aggregation degree value is defined as the ratio multiplied by the ratio. 1.125*0.6=0.675, so the outlier aggregation degree is 0.675.

[0112] The sparsity ratio calculation submodule calls the outlier aggregation degree, calculates the sum of the blank spacing between outliers and divides it by the total region length, calculates the sparsity ratio and compares it with the aggregation degree to obtain the sparsity coefficient of the outlier distribution and generate the outlier sparsity ratio.

[0113] The outlier aggregation degree is set to 0.675. The total spacing between outliers is calculated. The x-coordinates of the outliers are [10, 30, 50] (extracted from the location data). The total image region length is 100 pixels (x-axis range). The spacing includes 10 pixels from x=0 to the first point x=10, 20 pixels between points x=10 to x=30, 20 pixels between x=30 to x=50, and 50 pixels between x=50 to x=100. The total spacing is 10+20+20+50=100 pixels. Dividing this by the total region length, the sparsity ratio is 100 / 100=1. The sparsity ratio is calculated and compared with the aggregation degree. The comparison method is: sparsity ratio / aggregation degree = 1 / 0.675≈1.481. The sparsity coefficient of the outlier distribution is 1.481, and the outlier sparsity ratio is generated as 1.481.

[0114] The priority correction submodule calculates the weighted average of the outlier sparsity ratio and the classification priority adjustment result, and adjusts the priority order of the classification results to generate the mapping image classification result.

[0115] The outlier aggregation degree is calculated by combining the ratio of the average distance between adjacent outliers to the average global distance with the proportion of the number of outliers. The outlier sparsity ratio is the ratio of the sum of the blank spaces between outliers to the total length of the region to the aggregation degree.

[0116] Based on the outlier sparsity ratio of 1.481 and the region priority scores adjusted by classification priority (region B score set to 0.8, region A score to 0.6, and region C score to 0.4 from the context), a weighted average of the two is calculated. The weight settings are based on the application scenario, with a sparsity ratio weight of 0.3 and a priority score weight of 0.7. For each region, the new score = 0.3 sparsity ratio + 0.7 old priority score. Therefore, the new score for region B = 0.3 * 1.481 + 0.7 * 0.8 = 1.0043, and the new score for region A = 0.3. *1.481+0.7*0.6=0.8643, the new score for region C is 0.3*1.481+0.7*0.4=0.4443+0.28=0.7243. The priority order of the classification results is adjusted and sorted in descending order based on the new scores. Region B has the highest score of 1.0043, followed by Region A with 0.8643, and Region C has the lowest score of 0.7243. The sorting remains [Region B, Region A, Region C]. The mapping image classification result [Region B, Region A, Region C] is generated.

[0117] The big data-based geographic mapping image classification and processing method is implemented based on the aforementioned big data-based geographic mapping image classification and processing system, and includes the following steps:

[0118] S1: Divide the geographic mapping image area and collect pixel grayscale, brightness and multi-band reflectance, analyze the grayscale direction variance, contrast difference and spectral stability of each area, and input them into principal component analysis for feature dimensionality reduction to generate regional feature sets.

[0119] S2: Call the regional feature set, divide the gray-scale direction variance and contrast difference intervals and extract the fluctuation frequency, and calculate the constraint fluctuation intensity through the chi-square test;

[0120] S3: Based on the constrained fluctuation intensity, obtain the reflectivity difference of multiple regions and bands and detect extreme values, and determine that the spectral stability exceeds the stability threshold to generate spectral anomaly results;

[0121] S4: Calculate the pixel gradient direction offset and image edge closure ratio using the Sobel operator and detect whether the continuity is abnormal. Combine the spectral anomaly results to adjust the priority and generate classification priority adjustment results.

[0122] S5: Based on the classification priority adjustment results, calculate the aggregation degree and sparsity ratio of outliers and correct the classification priority to generate the mapping image classification results.

[0123] 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 geographic mapping image classification and processing system based on big data, characterized in that, The system includes: The regional feature module divides geographic mapping image regions and collects pixel grayscale, brightness, and multi-band reflectance. It analyzes the grayscale direction variance, contrast difference, and spectral stability of each region, and inputs them into principal component analysis for feature dimensionality reduction to generate regional feature sets, which are then transferred to the fluctuation analysis module. The fluctuation analysis module calls the regional feature set, divides the gray-scale direction variance and contrast difference intervals and extracts the fluctuation frequency, calculates the constraint fluctuation intensity through the chi-square test and passes it to the spectral anomaly module, calls the fluctuation frequency coefficient, obtains the current frequency and expected distribution values ​​of multiple intervals, calculates the weighted corrected chi-square value through the chi-square test, analyzes the constraint coefficients of multiple intervals and merges them into a unified scale to generate the constraint fluctuation intensity. The spectral anomaly module obtains the reflectance difference of multiple regions and bands based on the constrained fluctuation intensity and detects extreme values. It also determines that the spectral stability exceeds the stability threshold, generates spectral anomaly results, and transmits them to the edge adjustment module. The edge adjustment module calculates the pixel gradient direction offset and the image edge closure ratio using the Sobel operator and detects whether the continuity is abnormal. It obtains geographic mapping image metadata, calculates the horizontal and vertical gradient components of the gray values ​​of multiple pixels using the Sobel operator, and then combines the two directional components into a gradient direction angle and corresponds it to the pixel position. It calculates the offset distribution and records it to obtain the gradient direction offset. It adjusts the priority based on the spectral anomaly results, generates a classification priority adjustment result, and passes it to the classification correction module. The classification correction module, based on the classification priority adjustment results, calculates the aggregation degree and sparsity ratio of outliers and corrects the classification priority to generate the mapping image classification results. The region feature set includes gray-level variance, contrast difference, and spectral stability coefficient. The constraint fluctuation intensity includes gray-level variance interval frequency, contrast difference interval frequency, and fluctuation test chi-square value. The spectral anomaly results include band reflectance extreme values, stability threshold judgment values, and anomaly region coefficients. The classification priority adjustment results include edge closure ratio, gradient direction offset, and priority ranking value. The mapping image classification results include aggregation degree parameter and sparsity ratio parameter. The anomaly point sparsity ratio is the ratio of the sum of blank spacing between anomaly points to the total region length relative to the aggregation degree.

2. The geographic mapping image classification and processing system based on big data according to claim 1, characterized in that, The specific regional feature module is as follows: The pixel acquisition submodule divides the geographic mapping image into regions, acquires the grayscale value, brightness value and multi-band reflectance data of pixels in the region, calculates the grayscale direction distribution, and statistically analyzes the distribution of brightness distribution sequence and reflectance distribution sequence to generate the pixel distribution coefficient of the region. The difference calculation submodule calculates the variance of the gray-scale direction distribution of multiple zones based on the distribution coefficient of the partitioned pixels, detects the contrast difference between the brightness distribution sequence and the gray-scale direction variance value, counts the mean square fluctuation of the reflectance distribution sequence and converts it into spectral stability, and combines the three types of indicators to generate the spectral difference of the zones. The feature dimensionality reduction submodule calls the partition spectral difference degree, extracts the gray-level direction variance value, contrast difference and spectral stability value and inputs them into principal component analysis, extracts principal component coefficients, retains the partition features corresponding to the principal component coefficients, and generates a regional feature set. The region feature set consists of the projection features of the gray-level direction variance, contrast difference, and spectral stability coefficient corresponding to the principal component coefficients extracted by principal component analysis.

3. The geographic mapping image classification and processing system based on big data according to claim 2, characterized in that, The process of combining the spectral differences of the partitions is as follows: the gray-scale variance value calculated based on the gray-scale distribution sequence, the contrast difference value obtained by comparing the brightness distribution sequence with the gray-scale variance value, and the spectral stability value converted from the mean square fluctuation calculated based on the reflectance distribution sequence are weighted and combined to generate the spectral differences of the partitions.

4. The geographic mapping image classification and processing system based on big data according to claim 3, characterized in that, The fluctuation analysis module is specifically as follows: The interval division submodule calls the region feature set, extracts the gray-level direction variance and contrast difference and detects the value range respectively, groups the upper and lower limits of the gray-level direction variance and the upper and lower limits of the contrast difference into intervals and generates difference interval combinations. The frequency extraction submodule, based on the combination of difference intervals, statistically analyzes the distribution of gray-scale direction variance and contrast difference in each interval, detects the cumulative number of occurrences of values ​​and converts them into the number of fluctuations, and obtains the fluctuation frequency coefficient. The strength constraint submodule calls the fluctuation frequency coefficient to obtain the current frequency and expected distribution values ​​of multiple intervals. It calculates the weighted correction chi-square value by constructing a formula that includes the frequency fluctuation adjustment coefficient and the frequency correction offset. It analyzes the constraint coefficients of multiple intervals and merges them into a unified scale to generate the constraint fluctuation strength. The constrained fluctuation intensity includes a weighted corrected chi-square value that incorporates interval weights and offset corrections.

5. The geographic mapping image classification and processing system based on big data according to claim 4, characterized in that, The spectral anomaly module is specifically: The band difference submodule filters key analysis areas based on the constrained wave intensity, obtains the band reflectance values ​​of the key analysis areas, subtracts the band reflectance values ​​of adjacent areas one by one and records the difference, and integrates all the difference values ​​in the order of the regions to generate the band reflectance difference. The extreme value detection submodule calls the band reflection difference to detect the extreme values ​​in the multi-region band reflection difference, calculates the extreme value range and verifies the extreme interval position, marks the extreme value segment and records the range value to obtain the reflection extreme value range. The stability determination submodule calculates the rate of change of the difference between multiple bands in adjacent ranges based on the range of reflection extreme values, compares it with a set stability threshold, filters out band regions that exceed the stability threshold, integrates the corresponding difference values, and generates spectral anomaly results. The stability threshold is set by the distribution of the rate of change of the band reflection difference and the allowable deviation range; The spectral anomaly results include the extreme values ​​of the band reflectance difference selected, the identification of anomaly regions that exceed the stability threshold, and the corresponding difference values.

6. The geographic mapping image classification and processing system based on big data according to claim 5, characterized in that, The edge adjustment module is specifically: The gradient calculation submodule acquires geographic mapping image metadata, calculates the horizontal and vertical gradient components of the gray values ​​of multiple pixels based on Sobel convolution kernels and introduces neighborhood statistical features, then combines the two directional components into a gradient direction angle and corresponds it to the pixel position, calculates the offset distribution and records it to obtain the gradient direction offset. The continuity detection submodule calculates the edge closure ratio of the image based on the gradient direction offset and detects the connection between adjacent edges. It records the length of continuous segments and the length of broken segments, compares the ratio of the two to determine whether there is an anomaly, and combines the anomaly ratio with the edge closure ratio to generate an edge continuity index. The priority adjustment submodule calls the edge continuity index and combines it with the spectral anomaly results. It performs weighted calculations based on the weight relationship between the two types of values ​​to determine the priority ranking of the multi-region classification results, and integrates the ranking results to obtain the classification priority adjustment result. The classification priority adjustment result is a region sequence based on a weighted sort of edge continuity index and normalized spectral outliers.

7. The geographic mapping image classification and processing system based on big data according to claim 6, characterized in that, The classification correction module specifically includes: The aggregation degree calculation submodule, based on the classification priority adjustment results, extracts outlier location data and counts the distance distribution between adjacent outliers, calculates the ratio of the average adjacent distance to the average global distance, and analyzes the aggregation degree value in combination with the proportion of outlier numbers to obtain the outlier aggregation degree. The sparsity ratio calculation submodule calls the aggregation degree of the outliers, calculates the sum of the blank spacing between outliers and divides it by the total region length, calculates the sparsity ratio and compares it with the aggregation degree to obtain the sparsity coefficient of the outlier distribution and generate the outlier sparsity ratio. The priority correction submodule calculates the weighted average of the outlier sparsity ratio and the classification priority adjustment result, and adjusts the priority order of the classification results to generate the mapping image classification result. The outlier aggregation degree is a calculated value combining the ratio of the average distance between adjacent outliers to the average global distance and the proportion of the number of points. The outlier sparsity ratio is the ratio of the sum of the blank spacing between outliers to the total length of the region to the aggregation degree.

8. The geographic mapping image classification and processing system based on big data according to claim 7, characterized in that: The process of generating the anomaly sparsity ratio is as follows: when dividing the sum of the blank spacing between anomalies calculated based on the anomaly aggregation degree by the total region length, the total region length is limited to a preset standard length. The sparsity ratio is calculated and compared with the aggregation degree value, and a comparison threshold range is limited to obtain the sparsity coefficient of the anomaly distribution and generate the anomaly sparsity ratio.

9. A geographic mapping image classification and processing method based on big data, characterized in that, The implementation of the big data-based geographic mapping image classification and processing system according to any one of claims 1-8 includes the following steps: S1: Divide the geographic mapping image area and collect pixel grayscale, brightness and multi-band reflectance, analyze the grayscale direction variance, contrast difference and spectral stability of each area, and input them into principal component analysis for feature dimensionality reduction to generate regional feature sets. S2: Call the region feature set, divide the gray-level direction variance and contrast difference intervals and extract the fluctuation frequency, and calculate the constraint fluctuation intensity through the chi-square test; S3: Based on the constrained fluctuation intensity, obtain the reflectivity difference of multiple regions and bands and detect the extreme values, and determine that the spectral stability exceeds the stability threshold to generate spectral anomaly results; S4: Calculate the pixel gradient direction offset and image edge closure ratio using the Sobel operator and detect whether the continuity is abnormal. Adjust the priority based on the spectral anomaly results and generate a classification priority adjustment result. S5: Based on the classification priority adjustment results, calculate the aggregation degree and sparsity ratio of outliers and correct the classification priority to generate the mapping image classification results.

Images