Complex mountain landform intelligent classification method and system based on dual-scale TPI optimization

Abstract

The invention relates to the technical field of landform classification, and discloses a complex mountain landform intelligent classification method and system based on dual-scale TPI optimization, and the method comprises the steps: firstly obtaining ASTER GDEM data of a research area, dividing a landform analysis unit, and extracting initial landform factors (large / small scale landform position indexes, gradients and elevations); after standardization processing, calculating a Pearson correlation coefficient and a variance expansion factor to diagnose factor colinearity, and then determining an optimal terrain factor combination through screening calibration; then matching the data with historical landform classification data of a research area, judging whether to adjust input parameters of a machine learning model, if so, determining primary and target influence coefficients of the model, and optimizing hyper-parameters of a random forest, extreme gradient lifting and a deep neural network according to the primary and target influence coefficients; and finally, finishing the landform classification of the research area by utilizing the optimized model and outputting a result. By optimizing terrain factor combination and machine learning model parameters, the accuracy and efficiency of landform classification of the research area are improved.

Description

Technical Field

[0001] The present invention relates to the field of landform classification technology, and in particular to a complex mountain landform intelligent classification method and system based on dual-scale TPI optimization. Background Art

[0002] Landform classification is a key technology for demarcating different landform units based on the morphological characteristics of the Earth's surface. It is crucial for revealing the interactions between topography and surface phenomena, supporting disaster prevention, ecological protection, and land use planning. Traditional landform classification relies heavily on visual interpretation of topographic maps and aerial imagery, combined with field surveys. This is labor-intensive and resource-intensive, and highly reliant on expert experience, limiting classification efficiency and accuracy.

[0003] With the development of geographic information systems and remote sensing technology, the application of digital elevation models (DEMs) and remote sensing imagery has driven the development of automated landform classification methods. However, existing methods still have shortcomings: the single-scale Terrain Position Index (TPI) struggles to account for both macro- and micro-scale landform characteristics, resulting in low accuracy in identifying complex mountainous landforms; the screening of terrain factors lacks a systematic collinearity diagnosis mechanism, which can reduce model efficiency due to factor redundancy; and there is a lack of standardized processes for parameter optimization and classification result verification in machine learning models, making it difficult to reuse classification methods across different study areas.

[0004] Therefore, it is necessary to design an intelligent classification method and system for complex mountain landforms based on dual-scale TPI optimization to solve the problems existing in traditional landform classification, such as poor scale adaptability, unscientific factor screening, no standard for model optimization, and lack of dynamic feedback mechanism. Summary of the Invention

[0005] In view of this, the present invention proposes an intelligent classification method and system for complex mountain landforms based on dual-scale TPI optimization, aiming to solve the problems existing in traditional landform classification, such as poor scale adaptability, unscientific factor screening, no standard for model optimization, and lack of dynamic feedback mechanism.

[0006] In one aspect, the present invention proposes an intelligent classification method for complex mountainous landforms based on dual-scale TPI optimization, comprising:

[0007] The ASTER GDEM data of the study area were obtained and divided into multiple terrain analysis units. The initial terrain factors of each unit were extracted and standardized to obtain the characteristic data of each factor. The Pearson correlation coefficient and variance inflation factor between the characteristic data were calculated. The degree of collinearity of the factors was determined based on the Pearson correlation coefficient and variance inflation factor, and a collinearity diagnostic series was generated.

[0008] Screening and calibrating the factors in the collinearity diagnostic series, wherein the screening and calibration include high collinearity calibration and low collinearity calibration, counting the number of factors in the high collinearity calibration, counting the number of factors in the low collinearity calibration, and determining the optimal terrain factor combination based on the proportion of the two types of calibrations;

[0009] Obtain historical landform classification data for the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is determined to be necessary, determine the model's primary influence coefficient based on the weight of the terrain factors, and determine the model's target influence coefficient based on the number of sample points;

[0010] The hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network are determined according to the target influence coefficient of the model, the model parameters are optimized according to the adjustment factors, the optimized model is used to classify the landforms in the study area, and the classification results are output.

[0011] Furthermore, after using the optimized model to classify the landforms in the study area and output the classification results, it also includes:

[0012] Obtain spatial distribution data of each landform type in the classification results, extract boundary information and area proportion of each type of landform based on the GIS system, and determine the accuracy verification index of landform classification based on the boundary information and area proportion;

[0013] A first accuracy threshold and a second accuracy threshold are preset, and the first accuracy threshold is smaller than the second accuracy threshold;

[0014] Compare the accuracy verification index with each preset accuracy threshold:

[0015] When the accuracy verification index is greater than or equal to the second accuracy threshold, the classification result is determined to be qualified and the final landform classification map is directly output;

[0016] When the accuracy verification index is greater than or equal to the first accuracy threshold and less than the second accuracy threshold, the landform types with an area share of less than 5% in the classification results are locally corrected and the classification map is output after correction;

[0017] When the accuracy verification index is less than the first accuracy threshold, the optimal terrain factor combination is retrieved again, the hyperparameters of the machine learning model are adjusted, and classification is performed again until the accuracy verification index is greater than or equal to the first accuracy threshold;

[0018] At the same time, the final classification results were superimposed and analyzed with the historical geomorphological map of the study area to calculate the spatial consistency. When the consistency was greater than or equal to 85%, the classification process was completed;

[0019] When the degree of agreement is less than 85%, the classification results are calibrated with the field survey data.

[0020] Furthermore, when the initial terrain factors are standardized and collinearity diagnostic series are generated, the following steps are included:

[0021] Based on the collinearity diagnostic series, the variance inflation factor and tolerance of each initial terrain factor are extracted, and a collinearity determination threshold is set;

[0022] Compare the variance inflation factor and tolerance of each factor with the decision threshold:

[0023] When a factor satisfies both the variance inflation factor greater than or equal to 10 and the tolerance less than 0.1, it is marked as a high collinearity factor;

[0024] When a factor does not meet the variance inflation factor greater than or equal to 10 or the tolerance greater than or equal to 0.1, it is marked as a low collinearity factor;

[0025] The proportion of high collinearity factors and low collinearity factors is counted. When the proportion of low collinearity factors is greater than or equal to 70%, the low collinearity factor set is directly used as the optimal terrain factor combination.

[0026] When the proportion of low collinearity factors is less than 70%, the high collinearity factors are eliminated and the collinearity of the remaining factors is recalculated until the proportion of low collinearity factors is ≥ 70%, forming the optimal terrain factor combination.

[0027] Furthermore, when determining the optimal combination of terrain factors, it includes:

[0028] The first factor quantity threshold and the second factor quantity threshold are preset, and the first factor quantity threshold is smaller than the second factor quantity threshold. The validity of the combination is determined based on the relationship between the number of factors and the threshold of the optimal terrain factor combination:

[0029] When the number of factors is greater than or equal to the second factor number threshold, the combination is determined to contain sufficient terrain information and is directly used for model training;

[0030] When the number of factors is greater than or equal to the first factor number threshold and less than the second factor number threshold, auxiliary factors such as surface cutting depth and slope variation rate are extracted until the number of factors is greater than or equal to the second factor number threshold;

[0031] When the number of factors is less than the first factor number threshold, the initial terrain factors are re-screened and the factor extraction range is expanded until the number of factors is greater than or equal to the first factor number threshold;

[0032] The final optimal factor combination must include large-scale terrain position index, small-scale terrain position index, slope, elevation, surface cutting depth, slope variation, plane curvature and profile curvature.

[0033] Furthermore, when using the mean change point method to determine the optimal analysis window of the dual-scale TPI, it includes:

[0034] Based on the annular window, the initial inner and outer ring radii are set, and the area is gradually expanded at certain intervals. The mean and standard deviation of the terrain position index under each window are calculated.

[0035] Perform logarithmic transformation on the mean, construct a sample sequence and calculate the statistics using the mean change point method, and draw a statistical difference change curve;

[0036] When the curve has an inflection point from steep to slow, the corresponding window is the optimal analysis window:

[0037] The optimal window for the small-scale terrain position index is the ring radius at the inflection point, corresponding to a certain area;

[0038] The optimal window of the large-scale terrain position index is the ring radius expanded at a certain interval, corresponding to a certain area.

[0039] Furthermore, when matching historical landform classification data with the optimal factor combination, it includes:

[0040] Calculate the feature matching degree between historical classification data and the optimal factor combination, and preset the standard matching degree;

[0041] When the feature matching degree is consistent with the standard matching degree, the initial model parameters are directly used for training;

[0042] When the feature matching degree is inconsistent with the standard matching degree, the factor weight is adjusted according to the difference between the feature matching degree and the standard matching degree:

[0043] Presetting a first difference threshold and a second difference threshold, wherein the first difference threshold is smaller than the second difference threshold;

[0044] When the difference between the feature matching degree and the standard matching degree is less than or equal to the first difference threshold, the weights of the large-scale terrain position index and the small-scale terrain position index are increased by 10%;

[0045] When the difference between the feature matching degree and the standard matching degree is greater than the first difference threshold and less than or equal to the second difference threshold, the weight of the surface cutting depth and slope is increased by 15%;

[0046] When the difference between the feature matching degree and the standard matching degree is greater than the second difference threshold, all factor weights are standardized and redistributed to ensure that the cumulative weight sum is 1.

[0047] Furthermore, when using the optimized model to classify landforms, it includes:

[0048] The sample points in the study area were divided into training set and validation set in a ratio of 7:3, and stratified random sampling was used to ensure that the proportion of each type of landform sample was consistent with the actual situation;

[0049] The classification results of the three models were evaluated using accuracy, area under the receiver operating characteristic curve, recall, F1 score, and Kappa coefficient:

[0050] When the area under the receiver operating characteristic curve of the random forest is greater than or equal to 0.93 and the accuracy is greater than or equal to 87%, it is determined to be the optimal model and its classification result is output;

[0051] When the random forest did not meet the above conditions but the area under the receiver operating characteristic curve of the deep neural network was greater than or equal to 0.90, the deep neural network result was used;

[0052] When both models were unsatisfactory, the number of XGBoost iterations was optimized and reclassification was performed until the area under the receiver operating characteristic curve was greater than or equal to 0.87.

[0053] Furthermore, when performing feature importance analysis on the classification results, it includes:

[0054] The contribution of each factor to the optimal model is calculated based on the SHAP method, and a first contribution threshold and a second contribution threshold are preset, and the first contribution threshold is greater than the second contribution threshold;

[0055] When the factor contribution is greater than or equal to the first contribution threshold, it is determined to be a key factor, and the key factors include a large-scale terrain position index, a small-scale terrain position index, a surface cutting depth, and a slope;

[0056] When the factor contribution is greater than or equal to the second contribution threshold and less than the first contribution threshold, it is determined to be an important factor, and the important factors include elevation, plane curvature and profile curvature;

[0057] When the factor contribution is less than the second contribution threshold, it is determined to be a secondary factor, wherein the secondary factor includes the slope variation rate;

[0058] Among the key factors, the contribution of large-scale topographic position index to deep-cut canyons and U-shaped valleys is greater than 50%, and the contribution of slope to plains is greater than 60%.

[0059] Furthermore, when outputting the final landform classification results, it includes:

[0060] The classification results are divided into ten landform types, including four types of macro-landforms and six types of micro-landforms;

[0061] Statistics on the spatial distribution of various landforms:

[0062] Open slopes account for the largest proportion of land, mainly distributed in Chuxiong and central Kunming;

[0063] Alpine-canyon landforms such as deeply incised canyons and U-shaped valleys are concentrated in western Yuxi and northwestern Chuxiong;

[0064] The final output classification map must include the boundaries of each landform type, area proportion and key factor contribution heat map.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] This application takes the central Yunnan region as the study area. Based on the ASTER GDEM data, it extracts the dual-scale terrain position index (TPI) and various terrain factors, and combines machine learning models such as random forest, extreme gradient boosting and deep neural networks to carry out complex landform classification research. Finally, the random forest is determined to be the optimal model with an overall accuracy of 87% and an AUC value of 0.93.

[0067] The optimal analysis windows for large and small scales were determined by the mean change point method, and the optimal combination of terrain factors was screened out by combining Pearson correlation analysis and variance inflation factor collinearity diagnosis. The SHAP method was used to identify large scale, small scale, surface cutting depth and slope as key influencing factors.

[0068] This application effectively addresses the limitations of traditional single-scale TPI in complex landform classification, realizes the coordinated identification of macro and micro landform features, and provides scientific data support and methodological reference for disaster prevention and control, ecological protection, and land use planning in central Yunnan.

[0069] On the other hand, the present application also provides a complex mountain landform intelligent classification system based on dual-scale TPI optimization, which is applied to the above-mentioned complex mountain landform intelligent classification method based on dual-scale TPI optimization, including:

[0070] A data preprocessing module is used to obtain ASTER GDEM data of the study area, divide the data into multiple terrain analysis units, extract the initial terrain factors of each unit, the terrain including large-scale terrain position index, small-scale terrain position index, slope, and elevation, standardize the initial terrain factors, obtain characteristic data of each factor, calculate the Pearson correlation coefficient and variance inflation factor between the characteristic data, determine the degree of collinearity of the factors based on the relevant Pearson correlation coefficient and variance inflation factor, and generate a collinearity diagnostic series;

[0071] A factor screening module is used to screen and calibrate the factors in the collinearity diagnostic series, wherein the screening and calibration include high collinearity calibration and low collinearity calibration, count the number of factors in the high collinearity calibration, count the number of factors in the low collinearity calibration, and determine the optimal terrain factor combination based on the proportion of the two types of calibrations;

[0072] The model parameter adjustment module is used to obtain historical landform classification data of the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is determined to be necessary, the model's primary influence coefficient is determined based on the weight ratio of the terrain factor, and the model's target influence coefficient is determined based on the number of sample points;

[0073] The classification output module is used to determine the hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network according to the target influence coefficient of the model, optimize the model parameters according to the adjustment factors, use the optimized model to classify the landforms in the study area, and output the classification results.

[0074] It is understandable that the above-mentioned complex mountain landform intelligent classification method and system based on dual-scale TPI optimization have the same beneficial effects and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0076] Figure 1 Flowchart of the intelligent classification method for complex mountainous landforms based on dual-scale TPI optimization provided by an embodiment of the present invention;

[0077] Figure 2 This is a functional block diagram of the complex mountain landform intelligent classification system based on dual-scale TPI optimization provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0078] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present disclosure and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0079] Reference Figure 1 In some embodiments of the present application, a complex mountain landform intelligent classification method based on dual-scale TPI optimization includes:

[0080] ASTER GDEM data of the study area were obtained and divided into multiple terrain analysis units. The initial terrain factors of each unit were extracted and standardized to obtain the characteristic data of each factor. The Pearson correlation coefficient and variance inflation factor between the characteristic data were calculated. The degree of collinearity of the factors was determined based on the Pearson correlation coefficient and variance inflation factor, and a collinearity diagnostic series was generated.

[0081] The factors in the collinearity diagnostic series are screened and calibrated. The screening and calibration include high collinearity calibration and low collinearity calibration. The number of factors with high collinearity calibration and low collinearity calibration are counted. The optimal terrain factor combination is determined based on the proportion of the two types of calibration.

[0082] Obtain historical landform classification data for the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is necessary, determine the model's primary influence coefficient based on the weight of the terrain factors, and determine the model's target influence coefficient based on the number of sample points.

[0083] The hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network are determined according to the model target influence coefficient. The model parameters are optimized according to the adjustment factors. The optimized model is used to classify the landforms in the study area and output the classification results.

[0084] Specifically, to comprehensively consider both macroscopic and microscopic features of the landform, a total of 11 topographic factors were selected based on previous studies. These include six macroscopic factors (topographic position index (TPI), slope (S), elevation (E), plan curvature (PLC), profile curvature (PC), and relief (RF) at both large and small scales), and four microscopic topographic factors (surface incision depth (SCD), surface roughness (SR), elevation variation coefficient (EVC), and slope variation rate (SOS). The calculation formulas for each factor are shown in Equations (1-5). To reduce data redundancy and improve model computational efficiency and classification accuracy, this study conducted correlation analysis and collinearity diagnostics on the topographic factors derived from DEM data to identify and eliminate highly correlated variables.

[0085]

[0086] In formula (1): E std 、E mean 、E min 、E maxThey are the elevation standard deviation, elevation average, elevation maximum and elevation minimum within the statistical window respectively.

[0087]

[0088] In formula (2), ΔZ, ΔD, and ΔS represent the elevation difference (i.e., height difference) between adjacent pixels, the horizontal distance (i.e., horizontal projection distance) between adjacent pixels, and the slope difference (i.e., slope change) between adjacent pixels, respectively.

[0089]

[0090] In formula (3), Z and s represent the elevation value and the curve distance along the slope direction, respectively; x and y are the horizontal coordinates.

[0091] RF=Z max -Z min (4)

[0092] In formula (4): Z max , Z min Respectively represent: the highest elevation value and the lowest elevation value in the area.

[0093] Specifically, the Topographic Position Index (TPI) is defined by calculating the difference between the elevation of a cell and the average elevation of cells within a predetermined radius around it. A positive TPI value indicates highland (ridge), a negative value indicates lowland (valley), and a zero value indicates a flat area. TPI can not only effectively distinguish the slope position of the terrain (such as ridge, valley bottom, and half-hill slope), but can also be combined with slope data to further identify the terrain morphology type (such as canyon, plain, and U-shaped valley). The extraction formula is:

[0094]

[0095] In formula (5): Z cell , N, Z i They represent the elevation value of the central cell, the number of cells in the neighborhood, and the elevation values ​​of other cells in the neighborhood, respectively.

[0096] The above embodiment takes the central Yunnan region as the study area, and based on the ASTER GDEM data, extracts the dual-scale terrain position index (TPI) and a variety of terrain factors, combines machine learning models such as random forest, extreme gradient boosting and deep neural network to carry out complex landform classification research, and finally determines that random forest is the optimal model with an overall accuracy of 87% and an AUC value of 0.93. The optimal analysis window for large and small scales is determined by the mean change point method, and the optimal combination of terrain factors is screened out by combining Pearson correlation analysis and variance inflation factor collinearity diagnosis, and the SHAP method is used to identify large scale, small scale, surface cutting depth and slope as key influencing factors. This application effectively solves the limitations of traditional single-scale TPI in complex landform classification, realizes the coordinated identification of macro and micro landform features, and provides scientific data support and method reference for disaster prevention and control, ecological protection and land use planning in the central Yunnan region.

[0097] Specifically, after using the optimized model to classify the landforms in the study area and output the classification results, it also includes:

[0098] Obtain the spatial distribution data of each landform type in the classification results, extract the boundary information and area proportion of each type of landform based on the GIS system, and determine the accuracy verification index of the landform classification based on the boundary information and area proportion;

[0099] A first accuracy threshold and a second accuracy threshold are preset, and the first accuracy threshold is smaller than the second accuracy threshold;

[0100] Compare the accuracy verification indicators with the preset accuracy thresholds:

[0101] When the accuracy verification index is greater than or equal to the second accuracy threshold, the classification result is determined to be qualified and the final landform classification map is directly output;

[0102] When the accuracy verification index is greater than or equal to the first accuracy threshold and less than the second accuracy threshold, the landform types with an area share of less than 5% in the classification results are locally corrected and the classification map is output after correction;

[0103] When the accuracy verification index is less than the first accuracy threshold, the optimal terrain factor combination is retrieved again, the hyperparameters of the machine learning model are adjusted, and classification is performed again until the accuracy verification index is greater than or equal to the first accuracy threshold;

[0104] At the same time, the final classification results were superimposed and analyzed with the historical geomorphological map of the study area to calculate the spatial consistency. When the consistency was greater than or equal to 85%, the classification process was completed;

[0105] When the degree of agreement is less than 85%, the classification results are calibrated with the field survey data.

[0106] It can be understood that the above embodiment takes the central Yunnan region as the study area, extracts the dual-scale terrain position index (TPI) and multiple terrain factors based on ASTER GDEM data, screens the optimal factor combination through Pearson correlation analysis and variance inflation factor collinearity diagnosis, uses random forest, extreme gradient boosting and deep neural network to perform landform classification, determines random forest as the optimal model (overall accuracy 87%, AUC 0.93), and uses the SHAP method to identify large-scale TPI, small-scale TPI, surface cutting depth and slope as key factors, and finally forms a classification result that takes into account both macro and micro landform characteristics, providing data support for disaster prevention and control in the region.

[0107] Specifically, when standardizing the initial terrain factors and generating collinearity diagnostic series, it includes:

[0108] Based on the collinearity diagnostic series, the variance inflation factor and tolerance of each initial terrain factor are extracted, and the collinearity judgment threshold is set;

[0109] Compare the variance inflation factor and tolerance of each factor with the decision threshold:

[0110] When a factor satisfies both the variance inflation factor greater than or equal to 10 and the tolerance less than 0.1, it is marked as a high collinearity factor;

[0111] When a factor does not meet the variance inflation factor greater than or equal to 10 or the tolerance greater than or equal to 0.1, it is marked as a low collinearity factor;

[0112] The proportion of high collinearity factors and low collinearity factors is counted. When the proportion of low collinearity factors is greater than or equal to 70%, the low collinearity factor set is directly used as the optimal terrain factor combination.

[0113] When the proportion of low collinearity factors is less than 70%, the high collinearity factors are eliminated and the collinearity of the remaining factors is recalculated until the proportion of low collinearity factors is ≥ 70%, forming the optimal terrain factor combination.

[0114] Specifically, when the initial terrain factors are standardized and collinearity diagnostic series are generated, the variance inflation factor and tolerance of each factor are extracted. The collinearity judgment threshold is set as variance inflation factor ≥ 10 and tolerance < 0.1. Factors that meet this condition are high collinearity factors, and factors that do not meet this condition are low collinearity factors. When the proportion of low collinearity factors is ≥ 70%, it is directly used as the optimal combination. When the proportion is < 70%, the high collinearity factors are eliminated and the collinearity is recalculated until the proportion of low collinearity factors is ≥ 70%. The optimal factor combination finally determined includes terrain position index (large and small scales), surface incision depth, slope, elevation, slope variability, plane curvature and profile curvature.

[0115] The above embodiment achieves accurate classification of complex landforms in central Yunnan by combining the optimal dual-scale TPI and machine learning model. It takes into account both macro and micro landform characteristics, solves the shortcomings of traditional single-scale methods, and improves classification accuracy through factor screening and model optimization (random forest overall accuracy 87%, AUC 0.93), identifies key influencing factors, and provides reliable data support and scientific basis for regional disaster prevention and control, ecological protection and land use planning.

[0116] Specifically, when determining the optimal combination of terrain factors, it includes:

[0117] The first factor quantity threshold and the second factor quantity threshold are preset, and the first factor quantity threshold is smaller than the second factor quantity threshold. The validity of the combination is determined based on the relationship between the number of factors and the threshold of the optimal terrain factor combination:

[0118] When the number of factors is greater than or equal to the second factor number threshold, the combination is determined to contain sufficient terrain information and is directly used for model training;

[0119] When the number of factors is greater than or equal to the first factor number threshold and less than the second factor number threshold, auxiliary factors such as surface cutting depth and slope variation rate are extracted until the number of factors is greater than or equal to the second factor number threshold;

[0120] When the number of factors is less than the first factor number threshold, the initial terrain factors are re-screened and the factor extraction range is expanded until the number of factors is greater than or equal to the first factor number threshold;

[0121] The final optimal factor combination must include large-scale terrain position index, small-scale terrain position index, slope, elevation, surface cutting depth, slope variation, plane curvature and profile curvature.

[0122] Specifically, when determining the optimal combination of terrain factors, the threshold of the first factor number is preset to 5 and the threshold of the second factor number is preset to 8 (first threshold < second threshold). When the number of factors is ≥8, they are directly used for model training; when 5≤number of factors<8, auxiliary factors such as surface incision depth and slope variability are extracted to a number ≥8; when the number of factors is <5, the initial terrain factors are rescreened and the extraction range is expanded to a number ≥5. The optimal factor combination finally determined includes large-scale terrain position index, small-scale terrain position index, slope, elevation, surface incision depth, slope variability, plane curvature and profile curvature, a total of 8 factors.

[0123] The above embodiment determines the optimal large-scale and small-scale TPI analysis windows through the mean change point method, and combines Pearson correlation and variance inflation factor to screen out the optimal combination of terrain factors including dual-scale TPI, slope, elevation, etc., and uses models such as random forest to achieve classification. Among them, the overall accuracy of random forest reaches 87% and the AUC value is 0.93. It also identifies dual-scale TPI, surface cutting depth and slope as key factors, effectively taking into account macro and micro landform characteristics, and solving the shortcomings of traditional single-scale methods in complex landform division, providing accurate landform data support for disaster prevention and control, ecological protection and land use planning in the region.

[0124] Specifically, when using the mean change point method to determine the optimal analysis window of the dual-scale TPI, it includes:

[0125] Based on the annular window, the initial inner and outer ring radii are set, and the area is gradually expanded at certain intervals. The mean and standard deviation of the terrain position index under each window are calculated.

[0126] Perform logarithmic transformation on the mean, construct a sample sequence and calculate the statistics using the mean change point method, and draw a statistical difference change curve;

[0127] When the curve has an inflection point from steep to slow, the corresponding window is the optimal analysis window:

[0128] The optimal window for the small-scale terrain position index is the ring radius at the inflection point, corresponding to a certain area;

[0129] The optimal window of the large-scale terrain position index is the ring radius expanded at a certain interval, corresponding to a certain area.

[0130] Specifically, the TPI analysis window includes circular, annular, fan-shaped and rectangular windows. Due to the complex terrain characteristics of the central Yunnan region, the annular window can better capture this radial or curved terrain structure, especially for the analysis of terrain types such as ridges, valleys, and slopes. Therefore, this study uses the annular window for analysis.

[0131] Specifically, in order to determine the optimal TPI value of a certain point in a specific geomorphological area, this study uses a circular window as the basis and gradually expands the radius for analysis. As the analysis area increases, the TPI at the raised surface gradually increases, while that at the depressed surface gradually decreases. When the analysis area increases to a certain threshold, the change in the average elevation of the neighborhood tends to be stable, and the inflection point where the average elevation growth rate changes from fast to slow is defined, and its corresponding area is the statistical unit. The interpretation methods of inflection points include the standard deviation method, the fitting curve method, and the change point method. Among them, the mean change point analysis method is widely used because it can scientifically and concisely calculate the turning point position on the fitting curve from steep to slow. The average elevation within this statistical unit can reasonably characterize the relative terrain position of the research point.

[0132] The TPI analysis window includes circular, annular, fan-shaped and rectangular windows. Due to the complex terrain characteristics of the central Yunnan region, the annular window can better capture the radial or curved terrain structure, especially for the analysis of terrain types such as ridges, valleys, and slopes. Therefore, this study uses the annular window for analysis.

[0133] In order to determine the optimal TPI value of a certain point in a specific geomorphological area, this study uses a circular window as the basis and gradually expands the radius for analysis. As the analysis area increases, the TPI at the convex part of the surface gradually increases, while that at the concave part gradually decreases. When the analysis area increases to a certain threshold, the change in the average elevation of the neighborhood tends to be stable. The inflection point where the average elevation growth rate changes from fast to slow is defined, and its corresponding area is the statistical unit. The interpretation methods of the inflection point include the standard deviation method, the fitting curve method, and the change point method. Among them, the mean change point analysis method

[42] is widely used because it can scientifically and concisely calculate the turning point position on the fitting curve from steep to slow. The average elevation within this statistical unit can reasonably characterize the relative topographic position of the research point.

[0134] The initial analysis window was set with an inner ring radius of 1 and an outer ring radius of 3. The inner and outer ring radii were gradually expanded at intervals of 2 pixels until the inner ring radius reached 55 and the outer ring radius reached 57. The mean TPIm, statistical unit area, and standard deviation (SD) of the terrain position index for analysis windows of different scales were statistically analyzed (Table 1).

[0135] Table 1. Statistical characteristics of TPI under different analysis windows

[0136]

[0137] This study used the ArcMap 10.8 toolbox module, Focal Statistics, and Neighborhood tools, to extract TPI data from DEM data in central Yunnan. Centered on the target grid, a circular analysis window was selected, starting with an inner and outer ring radius of (1,3) and gradually increasing until the inner and outer ring radius reached (55,57).

[0138] Since the above analysis of inflection points is based on human judgment and may be affected by subjective factors, there are certain errors in the results. The logarithm of the TPI mean in Table 1 is used to construct the sample sequence {x i}, where i = 1, 2, ..., N, N is the number of samples. The sample is x i The point is divided into two segments, and the statistics are calculated according to the following formula:

[0139]

[0140] In formulas (6) and (7), and are the arithmetic means of each sample, represents the average value of the entire sample, S i and S represent the intermediate calculated values ​​of the mean change point method.

[0141] 2.2.3 Construction of landform type classification system

[0142] After determining the optimal TPI analysis window, this window is selected as the small-scale TPI (FTPI). Starting from this window, the inner and outer ring radii are gradually expanded with an interval of 5 pixels, and the large-scale TPI (BTPI) is determined according to the same method. Combining the small-scale TPI with the large-scale TPI helps to distinguish different types of nested terrain features. Referring to the research of Weiss

[32] , the obtained FTPI and BTPI are standardized according to formula (8):

[0143]

[0144] In formula (6), μ and σ represent the mean and standard deviation of TPI respectively. That is, the TPI is standardized so that its mean is 0 and its standard deviation is 1. ×100 means scaling the standardized value to an integer range, and the scale +0.5 and int(...) indicate rounding to an integer. Referring to the classification scheme of previous related studies

[43] , the landforms in central Yunnan are classified according to 1 standard deviation unit (=100 grid value units) as shown in Table (2):

[0145] Table 2. Landform classification standards

[0146]

[0147] Specifically, when matching historical landform classification data with the optimal factor combination, it includes:

[0148] Calculate the feature matching degree between historical classification data and the optimal factor combination, and preset the standard matching degree;

[0149] When the feature matching degree is consistent with the standard matching degree, the initial model parameters are directly used for training;

[0150] When the feature matching degree is inconsistent with the standard matching degree, the factor weight is adjusted according to the difference between the feature matching degree and the standard matching degree:

[0151] Presetting a first difference threshold and a second difference threshold, wherein the first difference threshold is smaller than the second difference threshold;

[0152] When the difference between the feature matching degree and the standard matching degree is less than or equal to the first difference threshold, the weights of the large-scale terrain position index and the small-scale terrain position index are increased by 10%;

[0153] When the difference between the feature matching degree and the standard matching degree is greater than the first difference threshold and less than or equal to the second difference threshold, the weight of the surface cutting depth and slope is increased by 15%;

[0154] When the difference between the feature matching degree and the standard matching degree is greater than the second difference threshold, all factor weights are standardized and redistributed to ensure that the cumulative weight sum is 1.

[0155] The above embodiment calculates the feature matching degree between historical classification data and the optimal factor combination and pre-sets a standard matching degree. If the feature matching degree is consistent with the standard matching degree, it means that the current data and the factor combination have a good fit. The initial model parameters can be directly used for training to ensure that the model starts smoothly based on the existing data. If the two do not match, the factor weights are adjusted based on the size of the difference. The first difference threshold is pre-set to 0.1 and the second difference threshold is pre-set to 0.2 (the first difference threshold is less than the second difference threshold). When the difference between the feature matching degree and the standard matching degree is less than or equal to 0.1, the weights of the large-scale terrain location index and the small-scale terrain location index are increased by 10% to enhance their influence in the model and better adapt the data. When the difference is greater than 0.1 and less than or equal to 0.2, the surface incision depth and slope play a key role in the classification of terrain morphology and landform type. In this case, their weights are increased by 15%, allowing the model to pay more attention to the role of these key terrain factors. When the difference is greater than 0.2, it indicates that the overall data deviation is large. The weights of all factors are standardized and redistributed, and mathematical methods are used to ensure that the cumulative weight sum is always 1, so that each factor can find a balance in the model again, and the model's fitting effect on the data is optimized, thereby improving the accuracy of complex mountain landform classification.

[0156] Specifically, when using the optimized model for landform classification, it includes:

[0157] The sample points in the study area were divided into training set and validation set in a ratio of 7:3, and stratified random sampling was used to ensure that the proportion of each type of landform sample was consistent with the actual situation;

[0158] The classification results of the three models were evaluated using accuracy, area under the receiver operating characteristic curve, recall, F1 score, and Kappa coefficient:

[0159] When the area under the receiver operating characteristic curve of the random forest is greater than or equal to 0.93 and the accuracy is greater than or equal to 87%, it is determined to be the optimal model and its classification result is output;

[0160] When the random forest did not meet the above conditions but the area under the receiver operating characteristic curve of the deep neural network was greater than or equal to 0.90, the deep neural network result was used;

[0161] When both models were unsatisfactory, the number of XGBoost iterations was optimized and reclassification was performed until the area under the receiver operating characteristic curve was greater than or equal to 0.87.

[0162] It is understandable that the random forest (RF) classifier achieves classification by integrating the prediction results of multiple decision trees. Its advantage lies in its ability to effectively handle the nonlinear and complex interactions between variables in high-dimensional landform data. Therefore, it is considered a reliable and robust classification method. XGBoost is a machine learning algorithm based on gradient boosting trees. This algorithm has demonstrated excellent performance in classification and regression tasks, combining flexibility and efficient scalability, and has now become one of the most influential and widely used algorithms in the field of machine learning. With the significant improvement in computing power and the continuous expansion of data scale, deep learning methods have achieved rapid development. Among them, the application of deep neural networks (DNN) in landform classification can significantly improve classification accuracy and efficiency, especially when combined with high-resolution DEM data.

[0163] Specifically, when performing feature importance analysis on classification results, it includes:

[0164] The contribution of each factor to the optimal model is calculated based on the SHAP method, and a first contribution threshold and a second contribution threshold are preset, and the first contribution threshold is greater than the second contribution threshold;

[0165] When the factor contribution is greater than or equal to the first contribution threshold, it is determined to be a key factor. The key factors include large-scale terrain position index, small-scale terrain position index, surface cutting depth and slope;

[0166] When the factor contribution is greater than or equal to the second contribution threshold and less than the first contribution threshold, it is determined to be an important factor. Important factors include elevation, plane curvature and profile curvature;

[0167] When the factor contribution is less than the second contribution threshold, it is determined to be a secondary factor, and the secondary factor includes the slope variation rate;

[0168] Among the key factors, the contribution of large-scale topographic position index to deep-cut canyons and U-shaped valleys is greater than 50%, and the contribution of slope to plains is greater than 60%.

[0169] It can be understood that the above embodiment uses the SHAP method to perform feature importance analysis on the classification results, clarifying the key factors (large scale, small scale, surface incision depth, slope), important factors and secondary factors and their contribution to different landform types (such as large-scale TPI contributes more than 50% to deep-cut canyons, and slope contributes more than 60% to plains). It not only reveals the inherent mechanism of how terrain factors affect landform classification, but also provides a scientific basis for the selection and application of factors in the classification of complex mountain landforms, thereby improving the interpretability and reliability of the classification results.

[0170] Specifically, when outputting the final landform classification results, it includes:

[0171] The classification results are divided into ten landform types, including four types of macro-landforms and six types of micro-landforms;

[0172] Statistics on the spatial distribution of various landforms:

[0173] Open slopes account for the largest proportion of land, mainly distributed in Chuxiong and central Kunming;

[0174] Alpine-canyon landforms such as deeply incised canyons and U-shaped valleys are concentrated in western Yuxi and northwestern Chuxiong;

[0175] The final output classification map must include the boundaries of each landform type, area proportion and key factor contribution heat map.

[0176] It is understandable that the classification results accurately divide the landforms in central Yunnan into 10 types, clarifying the spatial distribution characteristics of macro and micro landforms (such as open slopes are mainly distributed in Chuxiong and central Kunming, and high mountain-canyon landforms are concentrated in western Yuxi, etc.), and the output classification map includes boundaries, area proportions and key factor contribution heat maps, which not only refines the spatial pattern of complex landforms, but also provides intuitive and accurate landform data support for regional disaster prevention and control, ecological protection and land use planning, thereby enhancing the application value of the classification results.

[0177] In another preferred embodiment based on the above embodiment, refer to Figure 2 As shown, this embodiment provides a complex mountain landform intelligent classification system based on dual-scale TPI optimization, including:

[0178] The data preprocessing module is used to obtain the ASTER GDEM data of the study area, divide the data into multiple terrain analysis units, extract the initial terrain factors of each unit, including the large-scale terrain position index, small-scale terrain position index, slope, and elevation, standardize the initial terrain factors, obtain the characteristic data of each factor, calculate the Pearson correlation coefficient and variance inflation factor between the characteristic data, determine the degree of collinearity of the factors based on the Pearson correlation coefficient and variance inflation factor, and generate a collinearity diagnostic series;

[0179] The factor screening module is used to screen and calibrate the factors in the collinearity diagnostic series. The screening calibration includes high collinearity calibration and low collinearity calibration. The number of factors with high collinearity calibration and low collinearity calibration are counted, and the optimal terrain factor combination is determined based on the proportion of the two types of calibration.

[0180] The model parameter adjustment module is used to obtain the historical landform classification data of the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is determined to be necessary, the model's primary influence coefficient is determined based on the weight ratio of the terrain factor, and the model's target influence coefficient is determined based on the number of sample points;

[0181] The classification output module is used to determine the hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network according to the model target influence coefficient, optimize the model parameters according to the adjustment factors, use the optimized model to classify the landforms in the study area, and output the classification results.

[0182] It is understandable that the above-mentioned complex mountain landform intelligent classification method and system based on dual-scale TPI optimization have the same beneficial effects and will not be repeated here.

[0183] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or a combination of software and hardware embodiments. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0184] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0185] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0186] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. An intelligent classification method for complex mountainous landforms based on dual-scale TPI optimization, characterized by: include: The ASTER GDEM data of the study area were obtained and divided into multiple terrain analysis units. The initial terrain factors of each unit were extracted and standardized to obtain the characteristic data of each factor. The Pearson correlation coefficient and variance inflation factor between the characteristic data were calculated. The degree of collinearity of the factors was determined based on the Pearson correlation coefficient and variance inflation factor, and a collinearity diagnostic series was generated. Screening and calibrating the factors in the collinearity diagnostic series, wherein the screening and calibration include high collinearity calibration and low collinearity calibration, counting the number of factors in the high collinearity calibration, counting the number of factors in the low collinearity calibration, and determining the optimal terrain factor combination based on the proportion of the two types of calibrations; Obtain historical landform classification data for the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is determined to be necessary, determine the model's primary influence coefficient based on the weight of the terrain factors, and determine the model's target influence coefficient based on the number of sample points; The hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network are determined according to the target influence coefficient of the model, the model parameters are optimized according to the adjustment factors, the optimized model is used to classify the landforms in the study area, and the classification results are output.

2. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 1 is characterized in that: After using the optimized model to classify the landforms in the study area and output the classification results, it also includes: Obtain spatial distribution data of each landform type in the classification results, extract boundary information and area proportion of each type of landform based on the GIS system, and determine the accuracy verification index of landform classification based on the boundary information and area proportion; A first accuracy threshold and a second accuracy threshold are preset, and the first accuracy threshold is smaller than the second accuracy threshold; Compare the accuracy verification index with each preset accuracy threshold: When the accuracy verification index is greater than or equal to the second accuracy threshold, the classification result is determined to be qualified and the final landform classification map is directly output; When the accuracy verification index is greater than or equal to the first accuracy threshold and less than the second accuracy threshold, the landform types with an area share of less than 5% in the classification results are locally corrected and the classification map is output after correction; When the accuracy verification index is less than the first accuracy threshold, the optimal terrain factor combination is retrieved again, the hyperparameters of the machine learning model are adjusted, and classification is performed again until the accuracy verification index is greater than or equal to the first accuracy threshold; At the same time, the final classification results were superimposed and analyzed with the historical geomorphological map of the study area to calculate the spatial consistency. When the consistency was greater than or equal to 85%, the classification process was completed; When the degree of agreement is less than 85%, the classification results are calibrated with the field survey data.

3. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 2 is characterized in that: When standardizing the initial terrain factors and generating collinearity diagnostic series, including: Based on the collinearity diagnostic series, the variance inflation factor and tolerance of each initial terrain factor are extracted, and a collinearity determination threshold is set; Compare the variance inflation factor and tolerance of each factor with the decision threshold: When a factor satisfies both the variance inflation factor greater than or equal to 10 and the tolerance less than 0.1, it is marked as a high collinearity factor; When a factor does not meet the variance inflation factor greater than or equal to 10 or the tolerance greater than or equal to 0.1, it is marked as a low collinearity factor; The proportion of high collinearity factors and low collinearity factors is counted. When the proportion of low collinearity factors is greater than or equal to 70%, the low collinearity factor set is directly used as the optimal terrain factor combination. When the proportion of low collinearity factors is less than 70%, the high collinearity factors are eliminated and the collinearity of the remaining factors is recalculated until the proportion of low collinearity factors is ≥ 70%, forming the optimal terrain factor combination.

4. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 3 is characterized in that: When determining the optimal combination of terrain factors, include: The first factor quantity threshold and the second factor quantity threshold are preset, and the first factor quantity threshold is smaller than the second factor quantity threshold. The validity of the combination is determined based on the relationship between the number of factors and the threshold of the optimal terrain factor combination: When the number of factors is greater than or equal to the second factor number threshold, the combination is determined to contain sufficient terrain information and is directly used for model training; When the number of factors is greater than or equal to the first factor number threshold and less than the second factor number threshold, auxiliary factors such as surface cutting depth and slope variation rate are extracted until the number of factors is greater than or equal to the second factor number threshold; When the number of factors is less than the first factor number threshold, the initial terrain factors are re-screened and the factor extraction range is expanded until the number of factors is greater than or equal to the first factor number threshold; The final optimal factor combination must include large-scale terrain position index, small-scale terrain position index, slope, elevation, surface cutting depth, slope variation, plane curvature and profile curvature.

5. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 4 is characterized in that: When using the mean change point method to determine the optimal analysis window for dual-scale TPI, the following are included: Based on the annular window, the initial inner and outer ring radii are set, and the area is gradually expanded at certain intervals. The mean and standard deviation of the terrain position index under each window are calculated. Perform logarithmic transformation on the mean, construct a sample sequence and calculate the statistics using the mean change point method, and draw a statistical difference change curve; When the curve has an inflection point from steep to slow, the corresponding window is the optimal analysis window: The optimal window for the small-scale terrain position index is the ring radius at the inflection point, corresponding to a certain area; The optimal window of the large-scale terrain position index is the ring radius expanded at a certain interval, corresponding to a certain area.

6. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 5 is characterized in that: When matching historical landform classification data with the optimal factor combination, including: Calculate the feature matching degree between historical classification data and the optimal factor combination, and preset the standard matching degree; When the feature matching degree is consistent with the standard matching degree, the initial model parameters are directly used for training; When the feature matching degree is inconsistent with the standard matching degree, the factor weight is adjusted according to the difference between the feature matching degree and the standard matching degree: Presetting a first difference threshold and a second difference threshold, wherein the first difference threshold is smaller than the second difference threshold; When the difference between the feature matching degree and the standard matching degree is less than or equal to the first difference threshold, the weights of the large-scale terrain position index and the small-scale terrain position index are increased by 10%; When the difference between the feature matching degree and the standard matching degree is greater than the first difference threshold and less than or equal to the second difference threshold, the weight of the surface cutting depth and slope is increased by 15%; When the difference between the feature matching degree and the standard matching degree is greater than the second difference threshold, all factor weights are standardized and redistributed to ensure that the cumulative weight sum is 1.

7. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 6 is characterized in that: When using the optimized model for landform classification, it includes: The sample points in the study area were divided into training set and validation set in a ratio of 7:3, and stratified random sampling was used to ensure that the proportion of each type of landform sample was consistent with the actual situation; The classification results of the three models were evaluated using accuracy, area under the receiver operating characteristic curve, recall, F1 score, and Kappa coefficient: When the area under the receiver operating characteristic curve of the random forest is greater than or equal to 0.93 and the accuracy is greater than or equal to 87%, it is determined to be the optimal model and its classification result is output; When the random forest did not meet the conditions but the area under the receiver operating characteristic curve of the deep neural network was greater than or equal to 0.90, the deep neural network result was used; When both models were unsatisfactory, the number of XGBoost iterations was optimized and reclassification was performed until the area under the receiver operating characteristic curve was greater than or equal to 0.

87.

8. The complex mountain landform intelligent classification method based on dual-scale TPI optimization according to claim 7 is characterized in that: When performing feature importance analysis on classification results, include: Calculate the contribution of each factor to the optimal model based on the SHAP method, preset a first contribution threshold and a second contribution threshold, and the first contribution threshold is greater than the second contribution threshold; When the factor contribution is greater than or equal to the first contribution threshold, it is determined to be a key factor, wherein the key factors include a large-scale terrain position index, a small-scale terrain position index, a surface cutting depth, and a slope; When the factor contribution is greater than or equal to the second contribution threshold and less than the first contribution threshold, it is determined to be an important factor, and the important factors include elevation, plane curvature and profile curvature; When the factor contribution is less than the second contribution threshold, it is determined to be a secondary factor, wherein the secondary factor includes the slope variation rate; Among the key factors, the contribution of large-scale topographic position index to deep-cut canyons and U-shaped valleys is greater than 50%, and the contribution of slope to plains is greater than 60%.

9. The intelligent classification method for complex mountainous landforms based on dual-scale TPI optimization according to claim 8 is characterized in that: When outputting the final landform classification results, it includes: The classification results are divided into ten landform types, including four types of macro-landforms and six types of micro-landforms; Statistics on the spatial distribution of various landforms: Open slopes account for the largest proportion of land, mainly distributed in Chuxiong and central Kunming; Alpine-canyon landforms such as deeply incised canyons and U-shaped valleys are concentrated in western Yuxi and northwestern Chuxiong; The final output classification map must include the boundaries of each landform type, area proportion and key factor contribution heat map.

10. An intelligent classification system for complex mountainous landforms based on dual-scale TPI optimization, characterized by: The complex mountain landform intelligent classification method based on dual-scale TPI optimization applied to any one of claims 1 to 9 comprises: A data preprocessing module is used to obtain ASTER GDEM data of the study area, divide the data into multiple terrain analysis units, extract the initial terrain factors of each unit, the terrain including large-scale terrain position index, small-scale terrain position index, slope, and elevation, standardize the initial terrain factors, obtain characteristic data of each factor, calculate the Pearson correlation coefficient and variance inflation factor between the characteristic data, determine the degree of collinearity of the factors based on the relevant Pearson correlation coefficient and variance inflation factor, and generate a collinearity diagnostic series; A factor screening module is used to screen and calibrate the factors in the collinearity diagnostic series, wherein the screening and calibration include high collinearity calibration and low collinearity calibration, count the number of factors in the high collinearity calibration, count the number of factors in the low collinearity calibration, and determine the optimal terrain factor combination based on the proportion of the two types of calibrations; The model parameter adjustment module is used to obtain historical landform classification data of the study area, match the historical data with the optimal factor combination, and determine whether to adjust the input parameters of the machine learning model based on the matching results. If adjustment is determined to be necessary, the model's primary influence coefficient is determined based on the weight ratio of the terrain factor, and the model's target influence coefficient is determined based on the number of sample points; The classification output module is used to determine the hyperparameter adjustment factors of random forest, extreme gradient boosting and deep neural network according to the target influence coefficient of the model, optimize the model parameters according to the adjustment factors, use the optimized model to classify the landforms in the study area, and output the classification results.

Images