An intelligent classification method for saline-alkali soil based on improved measure effect

Abstract

The present application relates to the field of soil classification, and more particularly to a method for intelligent classification of saline-alkali soil based on effects of improvement measures, which comprises: obtaining a basic parameter set by collecting and adaptively calculating multi-source feature data of a land plot; obtaining a macroscopic obstacle pattern correction factor by conducting difference analysis on obstacle factor structures of the land plot; obtaining a microscopic improvement feasibility correction factor by jointly analyzing improvement demand degree and adjustment features of the land plot; obtaining an improvement effect modified distance by fusing and calculating Euclidean distance and the correction factor; and obtaining an intelligent classification result of the saline-alkali soil by conducting cluster analysis on the improvement effect modified distance, so as to solve the problems of land improvement mode recognition distortion and measure matching error caused by isotropy of distance measurement in existing cluster algorithms.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of soil classification, and particularly relates to a saline-alkali soil intelligent classification method based on the effect of improvement measures. BACKGROUND

[0002] Saline-alkali soil is one of the global land degradation problems, and has characteristics such as high soil salt content, imbalance of acid-base degree and structural degradation, which seriously restricts the improvement of agricultural productivity and the sustainable use of land resources. Saline-alkali soil improvement and utilization has become a core problem in agricultural ecological management and farmland protection. For different regions and different types of saline-alkali soil, there are significant differences in the composition of obstacle factors, for example, in some regions, high electrical conductivity (EC) induced salt stress is the main problem, and in some regions, high acid-base degree (pH) induced chemical inhibition is the dominant problem, so how to accurately identify the dominant obstacle type of different plots and develop differentiated improvement measures is the key link of saline-alkali soil improvement project. At present, the mainstream method of saline-alkali soil classification and zoning is mainly based on the clustering analysis of plot feature vectors. Specifically, researchers usually collect multi-dimensional soil feature data through remote sensing images, unmanned aerial vehicle monitoring and ground sampling, including soil electrical conductivity (EC), acid-base degree (pH), organic matter content, salt ion concentration, cation exchange capacity (CEC) and various vegetation spectral indices. Then, these indexes are taken as feature vectors to input clustering algorithms (such as K-means, DBSCAN or FCM algorithm), and automatic grouping is carried out according to the Euclidean distance measurement in multi-dimensional space, so as to obtain the similarity division and classification results of the plot. This kind of clustering method based on numerical similarity has high calculation efficiency in automatic analysis of plot big data, but the underlying measurement mechanism has inherent isotropic characteristics, that is, all feature dimensions are treated equally when calculating the similarity between plots, without distinguishing the importance of different features in the context of saline-alkali soil improvement.

[0003] Since the improvement characteristics of saline-alkali land have obvious "barrel effect" properties, that is, the difficulty and direction of land improvement are mainly determined by the most serious obstacle factor, rather than the average level of all characteristics, the clustering method based on standard Euclidean distance is prone to distortion in practical application. For example, a saline land plot with high EC as the main problem and an alkaline land plot with high pH as the main problem may be mutually compensated in other indicators, resulting in a close distance in the numerical space, and thus being misjudged as similar plots and being classified into the same class. However, the two types of plots require completely different management measures in the actual improvement process, the former is suitable for salt leaching, and the latter needs chemical conditioning or gypsum application. Such incorrect classification results will directly lead to the mismatch of improvement strategies, waste of resources, and even secondary salinization problems, making it difficult to achieve the goal of precise management. In addition, existing clustering models often only start from the geometric difference of the feature space, and fail to fully consider the sensitivity adjustment characteristics and improvement response characteristics of the plots. For example, two plots with the same high salt problem may have significant differences in improvement effect due to differences in internal properties such as soil clay content, porosity, cation exchange capacity (CEC), etc. The traditional distance measurement method cannot reflect this difference in improvement feasibility, making it difficult for the classification results to reflect the real response characteristics of the plots in the improvement process, thereby limiting the improvement measure recommendation and regional decision support based on the clustering results.

[0004] Therefore, how to overcome the limitation of distance measurement isotropy in existing clustering algorithms, and construct a distance correction mechanism that can consider the dominance of obstacle factors and the difference in improvement feasibility, so that the plot classification results can be directly related to the effect of improvement measures, has become a technical problem to be solved in the field. SUMMARY

[0005] Therefore, the present application aims to provide an intelligent classification method for saline-alkali land based on the effect of improvement measures, in order to solve the problems of plot improvement mode recognition distortion and measure matching error caused by distance measurement isotropy in existing clustering algorithms.

[0006] To achieve the above-mentioned purpose, the technical solution of the present application is as follows:

[0007] An intelligent classification method for saline-alkali land based on the effect of improvement measures, the method comprising:

[0008] Step S1: obtaining a basic parameter set by collecting and adaptively calculating multi-source feature data of the plot;

[0009] Step S2: obtaining a macro obstacle mode correction factor by analyzing the difference in obstacle factor structure of the plot;

[0010] Step S3: obtaining a micro improvement feasibility correction factor by jointly analyzing the improvement demand degree and adjustment characteristics of the plot;

[0011] Step S4: obtaining an improved effect correction distance by fusing the Euclidean distance and the correction factor;

[0012] Step S5: obtaining the intelligent classification result of the saline-alkali soil by clustering analysis on the improved effect correction distance.

[0013] Further, the acquisition of the basic parameter set by adaptive calculation on the multi-source feature data of the land block comprises:

[0014] The multi-source feature data including the obstacle factor feature and the sensitivity adjustment feature are acquired by data acquisition on all land blocks in the research area;

[0015] The obstacle factor feature includes soil conductivity data and soil pH data for reflecting the obstacle intensity of the saline-alkali soil; the sensitivity adjustment feature includes soil clay content percentage data and soil cation exchange capacity data for reflecting the response characteristics of the land block improvement; and the vegetation index calculated from the remote sensing image is taken as the comprehensive excellent index data of the land block;

[0016] For the multi-source feature data of the land block, all the land blocks are sorted from high to low according to the comprehensive excellent index data, and the land blocks with the top 3% of the comprehensive excellent index values are selected to form a reference land block subset;

[0017] The soil conductivity ideal data, the soil pH ideal data, the soil clay content percentage ideal data and the soil cation exchange capacity ideal data are acquired by mean calculation on the soil conductivity data, the soil pH data, the soil clay content percentage data and the soil cation exchange capacity data of each land block in the reference land block subset, respectively;

[0018] The soil conductivity standard deviation data and the soil pH standard deviation data of the reference land block subset are acquired by standard deviation calculation on the soil conductivity data and the soil pH data of all the land blocks in the reference land block subset;

[0019] For any target obstacle factor feature of any target land block in the reference land block subset, the absolute value of the calculation result of subtracting the target obstacle factor feature ideal data from the target obstacle factor feature data of the target land block is taken as the numerator, the standard deviation data of the target obstacle factor feature is taken as the denominator, and the formed fraction is taken as the standardization deviation of the target obstacle factor of the target land block;

[0020] For any target sensitivity adjustment feature of any target land block in the reference land block subset, the absolute value of the calculation result of subtracting the target sensitivity adjustment feature ideal data from the target sensitivity adjustment feature data of the target land block is taken as the deviation of the target sensitivity adjustment feature of the target land block.

[0021] For all plots in the reference plot subset, the Pearson correlation coefficient between the sequence formed by the standardized deviation data of the target obstacle factor of all plots and the sequence formed by the deviation data of the target sensitivity adjustment feature of all plots is taken as the correlation coefficient between the target obstacle factor and the target sensitivity adjustment feature; the absolute value of the correlation coefficient between the target obstacle factor and the target sensitivity adjustment feature is subjected to exponential mapping with a natural constant as the base, and the corresponding mapping result is taken as the influence coefficient between the target obstacle factor and the target sensitivity adjustment feature;

[0022] For any target obstacle factor feature of any target plot in the reference plot subset, the sum of squares of the standardized deviations of all obstacle factor feature dimensions is taken as the denominator, and the square of the standardized deviation of the target obstacle factor of the target plot is taken as the numerator, and the corresponding fraction is taken as the normalized quadratic deviation of the target obstacle factor feature of the target plot;

[0023] For any target obstacle factor feature and any target sensitivity adjustment feature of any target plot in the reference plot subset, the calculation result of subtracting the target sensitivity adjustment feature data of the target plot from the target sensitivity adjustment feature ideal data is taken as the first difference evaluation of the target sensitivity adjustment feature of the target plot; the calculation result of multiplying the first difference evaluation of the target sensitivity adjustment feature of the target plot and the influence coefficient between the target obstacle factor feature and the target sensitivity adjustment feature is taken as the interactive adjustment evaluation between the target obstacle factor feature and the target sensitivity adjustment feature; the interactive adjustment evaluation between the target obstacle factor feature and the target sensitivity adjustment feature is subjected to cosine function mapping, and the reciprocal of the calculation result of adding the mean value of the cosine function mapping results of all sensitivity adjustment features to the constant 1 is taken as the interactive adjustment item of the target obstacle factor feature of the target plot.

[0024] Further, the macro obstacle pattern correction factor is obtained by performing difference analysis on the plot obstacle factor structure, comprising:

[0025] The obstacle difference measurement result is obtained by performing structural difference calculation on the plot normalized quadratic deviation data; the macro obstacle pattern correction factor is obtained by performing nonlinear amplification and penalty control on the obstacle difference measurement result.

[0026] Further, the obstacle difference measurement result is obtained by performing structural difference calculation on the plot normalized quadratic deviation data, comprising:

[0027] For any target obstacle factor feature of any two target plots, square the difference between the square root of the normalized quadratic deviation of the target obstacle factor feature of the two target plots, as the first deviation difference evaluation of the target obstacle factor feature of the two target plots; sum up the first deviation difference evaluations of the target obstacle factor features of the two target plots and take the square root and divide by the square root of two, and take the corresponding calculation result as the obstacle difference measurement result of the two target plots.

[0028] Further, the macro obstacle pattern correction factor is obtained by nonlinear amplification and penalty control on the obstacle difference measurement result, including:

[0029] Set the maximum penalty coefficient and the sensitivity index; for any two target plots, take the obstacle difference measurement result of the two target plots and multiply it by the sensitivity index, and add the calculation result to the constant 1, and take the calculation result as the macro obstacle pattern correction factor of the two target plots.

[0030] Further, the micro improvement feasibility correction factor is obtained by joint analysis of plot improvement demand degree and adjustment feature, including:

[0031] The plot improvement demand degree data is obtained by normalizing the ratio of the plot standardized deviation and the interaction adjustment item data;

[0032] The improvement feasibility difference measurement result between plots is obtained by weighted difference analysis of the improvement demand degree and the obstacle factor weight data;

[0033] The micro improvement feasibility correction factor is obtained by nonlinear mapping and range control on the improvement feasibility difference measurement result.

[0034] Further, the plot improvement demand degree data is obtained by normalizing the ratio of the plot standardized deviation and the interaction adjustment item data, including:

[0035] For any target obstacle factor feature of any target plot, take the standardized deviation of the target plot in the target obstacle factor feature as the numerator, and take the interaction adjustment item of the target plot in the target obstacle factor feature as the denominator, and take the calculation result of adding a minimum positive number to the corresponding fraction as the improvement demand degree data of the target plot in the target obstacle factor feature.

[0036] Further, the improvement feasibility difference measurement result between plots is obtained by weighted difference analysis of the improvement demand degree and the obstacle factor weight data, including:

[0037] For any two target plots, the absolute value of the calculation result of subtracting the improvement demand degree data of the two target plots in the same obstacle factor feature from each other is taken as the improvement demand degree difference evaluation of the two plots in the same obstacle factor feature.

[0038] For any two target plots, the absolute value of the calculation result of subtracting the improvement demand degree data of the two target plots in the same obstacle factor feature from each other is taken as the improvement demand degree difference evaluation of the two plots in the same obstacle factor feature.

[0039] The weighted sum of the improvement demand degree difference evaluations of the two target plots in all obstacle factor feature dimensions according to the corresponding feature weights is obtained, and the calculation result of dividing the weighted sum result by the sum of all obstacle factor feature weight data is taken as the improvement feasibility difference degree measurement result of the two target plots.

[0040] Further, the microcosmic improvement feasibility correction factor is obtained by performing nonlinear mapping and range control on the improvement feasibility difference degree measurement result, including:

[0041] A global scaling parameter is set, for any two target plots, the calculation result of multiplying the improvement feasibility difference measurement result of the two target plots by the global scaling parameter is mapped by a hyperbolic tangent function, and the calculation result of adding the corresponding mapping result to a constant 1 is taken as the microcosmic improvement feasibility correction factor of the two target plots.

[0042] Further, the improvement effect correction distance is obtained by fusing the Euclidean distance and the correction factor, including:

[0043] For any two target plots, the calculation result of multiplying the macroscopic obstacle mode correction factor of the two target plots by the microcosmic improvement feasibility correction factor is taken as the distance optimization weight of the two target plots, and the calculation result of multiplying the distance optimization weight of the two target plots by the Euclidean distance of the two target plots is taken as the improvement effect correction distance of the two target plots.

[0044] Compared with the prior art, the present application has the following advantages:

[0045] The intelligent saline-alkali soil classification method based on the effect of improvement measures provided by the embodiment of the present application realizes deep semantic recognition of differences between saline-alkali soil plots by constructing a data-driven improvement effect correction distance model. The model takes obstacle factor characteristics and sensitivity adjustment characteristics as core inputs, no longer relies on isotropic numerical similarity, but comprehensively discriminates according to the problem composition, improvement difficulty and response potential of plots on key obstacle factors. By introducing a macro obstacle pattern correction mechanism, the algorithm can automatically identify the essential differences in obstacle types and problem structures between plots, avoiding the false similarity misjudgment of the traditional Euclidean distance when facing different obstacle dominant patterns, so that the clustering results are more in line with the actual soil improvement rules, and the scientificity and interpretability of plot classification are significantly improved. In addition, the micro-improvement feasibility correction mechanism is further integrated in the plot similarity evaluation, so that the classification results not only reflect the differences in plot representation, but also reflect the influence of different soil internal structure characteristics on the response efficiency of improvement measures. Through this technical means, the algorithm can distinguish the improvement difficulty of different plots with the same obstacle problem, and realize consistent partitioning based on the effect of improvement measures. Therefore, the classification results form a direct corresponding relationship with the actual improvement strategy, providing quantifiable basis for the agricultural department to formulate differentiated strategies and accurately allocate resources in saline-alkali soil improvement, and significantly improving the pertinence of saline-alkali soil governance and the utilization efficiency of improvement investment. BRIEF DESCRIPTION OF DRAWINGS

[0046] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, and are incorporated in and constitute a part of this application. The embodiments of the present application and their

[0047] Figure 1 The method flowchart of the intelligent saline-alkali soil classification method based on the effect of improvement measures provided by the embodiment of the present application. DETAILED DESCRIPTION

[0048] The present application will be described in detail below with reference to the accompanying drawings and embodiments.

[0049] Reference Figure 1 The method flowchart of the intelligent saline-alkali soil classification method based on the effect of improvement measures provided by the embodiment of the present application is shown in Figure 1 The intelligent saline-alkali soil classification method based on the effect of improvement measures can include:

[0050] Step S1, acquiring a basic parameter set by collecting and adaptively calculating multi-source feature data of plots.

[0051] By collecting data of all plots in the study area, multi-source feature data including obstacle factor characteristics and sensitivity adjustment characteristics are obtained;

[0052] The obstacle factor characteristics include soil conductivity data and soil pH data for reflecting the obstacle intensity of saline-alkali land; the sensitivity adjustment characteristics include soil clay content percentage data and soil cation exchange capacity data for reflecting the response characteristics of land improvement; and the vegetation index calculated from the remote sensing image is taken as the comprehensive excellent index data of the land.

[0053] For the multi-source characteristic data of the land, all lands are sorted according to the comprehensive excellent index data from high to low, and lands with the top 3 comprehensive excellent index values are selected to form a reference land subset;

[0054] After the reference land subset is obtained, the mean values of the soil conductivity data, the soil pH data, the soil clay content percentage data and the soil cation exchange capacity data of each land in the reference land subset are calculated to obtain ideal soil conductivity data, ideal soil pH data, ideal soil clay content percentage data and ideal soil cation exchange capacity data;

[0055] The standard deviations of the soil conductivity data and the soil pH data of all lands in the reference land subset are calculated to obtain the soil conductivity standard deviation data and the soil pH standard deviation data of the reference land subset;

[0056] For any target obstacle factor characteristic of any target land in the reference land subset, the absolute value of the calculation result of subtracting the target obstacle factor characteristic ideal data from the target obstacle factor characteristic data of the target land is taken as the numerator, the standard deviation data of the target obstacle factor characteristic is taken as the denominator, and the corresponding fraction is taken as the standardized deviation of the target obstacle factor of the target land;

[0057] For any target sensitivity adjustment characteristic of any target land in the reference land subset, the absolute value of the calculation result of subtracting the target sensitivity adjustment characteristic ideal data from the target sensitivity adjustment characteristic data of the target land is taken as the deviation of the target sensitivity adjustment characteristic of the target land;

[0058] For all lands in the reference land subset, the Pearson correlation coefficient between the sequence formed by the standardized deviation data of the target obstacle factor of all lands and the sequence formed by the deviation data of the target sensitivity adjustment characteristic of all lands is taken as the correlation coefficient between the target obstacle factor and the target sensitivity adjustment characteristic; the absolute value of the correlation coefficient between the target obstacle factor and the target sensitivity adjustment characteristic is exponentially mapped with a natural constant as the base, and the corresponding mapping result is taken as the influence coefficient between the target obstacle factor and the target sensitivity adjustment characteristic;

[0059] For any target obstacle factor feature of any target plot in the reference plot subset, the sum of the squares of the standardized deviations of all obstacle factor feature dimensions is used as the denominator, the square of the standardized deviation of the target obstacle factor of the target plot is used as the numerator, and the corresponding fraction is used as the normalized quadratic deviation of the target obstacle factor feature of the target plot.

[0060] For any target obstacle factor feature and any target sensitivity adjustment feature of any target plot in the reference plot subset, the result of subtracting the target sensitivity adjustment feature data from the ideal data of the target sensitivity adjustment feature is used as the first difference assessment of the target sensitivity adjustment feature of the target plot; the result of multiplying the first difference assessment of the target sensitivity adjustment feature of the target plot and the influence coefficient between the target obstacle factor feature and the target sensitivity adjustment feature is used as the interaction adjustment assessment between the target obstacle factor feature and the target sensitivity adjustment feature; the interaction adjustment assessment between the target obstacle factor feature and the target sensitivity adjustment feature is mapped by a cosine function, and the reciprocal of the result of adding the mean of the cosine function mapping results of all sensitivity adjustment features to a constant 1 is used as the interaction adjustment term of the target obstacle factor feature of the target plot.

[0061] In one implementation, assume the first The first plot of land The sensitivity modulation feature data are as follows: ;No. Ideal data for each sensitivity modulation feature are: ;No. The characteristics of the obstacle factor and the first The influence coefficient between the individual sensitivity modulation features is The number of dimensions of the sensitivity modulation feature is Then the first The first plot of land The calculation expression for the interaction moderating term of the target obstacle factor features is as follows:

[0062]

[0063] in, Indicates the first The first plot of land Interactive moderating terms of the characteristics of each target obstacle factor; Indicates the first The first plot of land Sensitivity modulation feature data; Indicates the first Ideal data for each sensitivity modulation feature; Indicates the first The characteristics of the obstacle factor and the first an impact coefficient between the sensitivity adjustment features; a number of dimensions representing the sensitivity adjustment features.

[0064] Up to now, the basic parameter set is obtained by collecting and adaptively calculating the multi-source feature data of the land plot.

[0065] In step S2, the macro obstacle mode correction factor is obtained by difference analysis on the obstacle factor structure of the land plot.

[0066] The existing technology has technical defects in using the standard Euclidean distance measurement in the clustering algorithm for classifying saline-alkali land. The defect is derived from the isotropy of the Euclidean distance, that is, it treats the differences of all feature dimensions equally, and cannot reflect the importance difference of different feature dimensions in the specific application scenario of saline-alkali land improvement.

[0067] In order to solve the above technical problems, the distance measurement mechanism needs to be modified so that it can compare the distribution and composition of key obstacle factors between different land plots. First, based on the degree of deviation of each feature of the land plot from its ideal value, a mathematical description is constructed for each land plot, which can objectively reflect the distribution of the problem severity in each feature dimension. Further, when measuring the similarity of any two land plots, the structural difference in the problem distribution mathematical description is calculated. When the problem distribution structure of two land plots is highly similar, that is, the composition of the problem severity contribution share in each feature dimension is basically consistent, it can be considered that they are truly similar land plots in the sense of improvement. On the contrary, if the calculation result shows that the problem distribution structure of two land plots is significantly different, it means that they face different key obstacle factors. At this time, even if the Euclidean distance of their original feature vectors is very close in value, a multiplication correction factor must be used to apply a large enough penalty gain to their distance calculation to ensure that they are effectively distinguished in the subsequent clustering process.

[0068] In summary, first, the obstacle difference measurement result is obtained by calculating the structural difference of the normalized quadratic deviation data of the land plot. Specifically, for any target obstacle factor feature of any two target land plots, the square of the difference between the square roots of the normalized quadratic deviation of the target obstacle factor feature of the two target land plots is taken as the first deviation difference evaluation of the target obstacle factor feature of the two target land plots. The sum of the first deviation difference evaluations of all obstacle factor features of the two target land plots is calculated and the square root is calculated and divided by the square root of two. The corresponding calculation result is taken as the obstacle difference measurement result of the two target land plots.

[0069] After obtaining the obstacle difference measurement results of the two target plots, the macro obstacle pattern correction factor is obtained by nonlinear amplification and penalty control of the obstacle difference measurement results. Specifically, a maximum penalty coefficient and a sensitivity index are set. In this embodiment of the invention, the maximum penalty coefficient used to control the maximum value of the macro obstacle pattern correction factor is set to 8, and the sensitivity index used to adjust the sensitivity of the macro obstacle pattern correction factor to the difference response is set to 2. The parameter settings can be adjusted according to the actual scenario and are not required. For any two target plots, the obstacle difference measurement results of the two target plots are calculated by using the sensitivity index and multiplied by the maximum penalty coefficient. The calculation result is added to the constant 1 and the result is used as the macro obstacle pattern correction factor of the two target plots.

[0070] In one implementation, assume the first The first plot of land The normalized quadratic deviation of each obstacle factor feature is: ;No. The first plot of land The normalized quadratic deviation of each obstacle factor feature is: The number of dimensions of the barrier factor features is The maximum penalty coefficient is Sensitivity index is Then the first The plot of land and the first The calculation expression for the macroscopic obstacle pattern correction factor for each plot is as follows:

[0071]

[0072] in, Indicates the first The plot of land and the first Macro-level obstacle pattern correction factor for individual plots; Indicates the maximum penalty coefficient; The number of dimensions representing the features of the barrier factor; Indicates the first The first plot of land Normalized quadratic deviation of each obstacle factor characteristic; Indicates the first The first plot of land Normalized quadratic deviation of each obstacle factor characteristic; This indicates the sensitivity index.

[0073] It should be noted that the core part of the formula That is, the first The plot of land and the first The Hellinger distance between the normalized second-order deviations of the two plots is used as the basic measure to quantify the difference in their problem distribution structure. By using the Hellinger distance as an unbiased basic measure to reflect the true difference in the problem distribution of the two plots. For example, when comparing a plot with high soil conductivity as the only key obstacle and a plot with a relatively uniform problem distribution, there is a huge difference in the shape of their S spectra, and the calculated Hellinger distance will be a larger value; while comparing two plots with high soil conductivity as the key obstacle, their S spectra are highly similar in shape, and the calculated Hellinger distance will be very close to 0. However, directly using the distance value will be overly sensitive to minor differences caused by data noise and the like. In order to solve this potential problem, the present application introduces a sensitivity index, which is powered to the Hellinger distance. Through nonlinear transformation, the correction factor is made to respond slowly to minor S spectrum differences (i.e. the Hellinger distance value is small), and the punishment effect is suppressed; while for significant structural differences (i.e. the Hellinger distance value is larger), it maintains sufficient sensitivity. This ensures that the algorithm can effectively distinguish between true structural mismatches and meaningless minor fluctuations. Finally, in order to ensure that the correction factor can play a role while not destroying the overall numerical stability of the clustering algorithm, the present application strictly controls the output of the structure . As a preset maximum penalty coefficient, the macro obstacle pattern correction factor is limited to the range of , avoiding the risk of numerical explosion of the penalty factor due to extreme differences. When the key obstacle factors of the two plots are consistent, the Hellinger distance is 0, and the macro obstacle pattern correction factor is exactly 1, which does not affect the original distance at all, correctly identifying them as similar plots. Conversely, when the key obstacle factors of the two plots are significantly misaligned, i.e. facing the pseudo-similarity problem, the macro obstacle pattern correction factor will smoothly increase to the upper limit of the penalty value, effectively separating the plots in the final distance calculation.

[0074] At this point, the macro obstacle pattern correction factor is obtained by analyzing the differences in the obstacle factor structure of the plots.

[0075] Step S3, the micro-improvement feasibility correction factor is obtained by jointly analyzing the plot improvement demand degree and the adjustment characteristics.

[0076] By the macro obstacle pattern correction factor in step S2, the distance measurement mechanism has been able to identify and distinguish the plots with different macro obstacle patterns. The calculation basis of the correction factor is the distribution of EC, pH and other obstacle factor characteristics, and its function is to ensure that the plots with the same problem type are preliminarily gathered. However, this only solves the first level of the problem. In the practice of saline-alkali land improvement, there are generally two plots that face the same key obstacle factor (for example, high salt stress), but due to the difference in the sensitivity adjustment characteristics (such as soil texture, porosity, cation exchange capacity, etc.) not used in the macro obstacle pattern correction factor, their response efficiency and final effect to the same improvement measure (such as leaching salt) can be completely different. For example, a sandy high-salt plot and a clay high-salt plot have the same obstacle factor characteristics, so the calculated macro obstacle pattern correction factor will determine that they are highly similar; but in reality, the former has good permeability and significant leaching salt effect, while the latter has poor permeability and salt is difficult to be washed out, and even the secondary salinization may be aggravated due to poor drainage.

[0077] The prior art usually handles these internal attributes affecting the improvement process together with other obstacle factor characteristics when classifying, and cannot reveal the deep coupling relationship between them and the specific improvement measures. Therefore, in order to realize intelligent classification based on the effect of the improvement measures, it is necessary to further analyze on the basis of solving the problem of identifying the macro obstacle pattern. It is necessary to focus on the key obstacles that the two plots face in common and further quantify the difference in improvement feasibility determined by their respective internal attributes. If two plots have the same key obstacle factor, but the difficulty of their improvement is different, a correction factor should also be used to impose a corresponding penalty on the calculation of their distance.

[0078] In summary, the present application first obtains plot improvement demand degree data by normalizing the ratio of plot standardized deviation and interactive adjustment item data, specifically, for any target plot and any target obstacle factor characteristic, taking the standardized deviation of the target plot in the target obstacle factor characteristic as the numerator, taking the interactive adjustment item of the target plot in the target obstacle factor characteristic as the denominator, taking the calculation result of adding the corresponding fraction to a minimum positive number, and taking the mapping result as the improvement demand degree data of the target plot in the target obstacle factor characteristic.

[0079] After obtaining the improvement demand degree data of the target plot in the target obstacle factor characteristic, the improvement feasibility difference between plots is measured by further analyzing the weighted difference between the improvement demand degree and the obstacle factor weight data, specifically, for any two target plots, the calculation result of multiplying the square roots of the normalized quadratic deviations of the two target plots in the same obstacle factor characteristic is taken as the feature weight of the two target plots in the same obstacle factor characteristic.

[0080] The absolute value of the difference between the improvement demand data of the two target plots for the same obstacle factor feature is used as the assessment of the difference in improvement demand between the two plots for the same obstacle factor feature.

[0081] The assessment of the difference in improvement needs between the two target plots across all obstacle factor feature dimensions is weighted and summed according to the corresponding feature weights. The weighted sum is then divided by the sum of all obstacle factor feature weights to determine the difference in improvement feasibility between the two target plots.

[0082] After obtaining the improvement feasibility difference measurement results of two target plots, a micro-improvement feasibility correction factor is obtained by performing nonlinear mapping and range control on the improvement feasibility difference measurement results. Specifically, a global scaling parameter is set. In this embodiment of the invention, the global scaling parameter used to adjust the sensitivity of the micro-improvement feasibility correction factor is set to 1. For any two target plots, the calculation result of multiplying the improvement feasibility difference measurement results of the two target plots by the global scaling parameter is mapped by a hyperbolic tangent function. The calculation result of adding the corresponding mapping result to the constant 1 is used as the micro-improvement feasibility correction factor for the two target plots.

[0083] In one implementation, assume the first The first plot of land The standardized deviation of each barrier factor feature is ;No. The first plot of land The standardized deviation of each barrier factor feature is The global scaling parameter is: Then the first The plot of land and the first The formula for calculating the micro-improvement feasibility correction factor for each plot of land is as follows:

[0084]

[0085] in, Indicates the first The plot of land and the first Micro-level improvement feasibility correction factor for each plot of land; Indicates the first The first plot of land Normalized quadratic deviation of each obstacle factor characteristic; Indicates the first The first plot of land Normalized quadratic deviation of each obstacle factor characteristic; Indicates the first The first plot of land Standardized deviation of each barrier factor characteristic; Indicates the first The first plot of land Standardized deviation of each barrier factor characteristic; Indicates the first The first plot of land Interactive moderating terms of the characteristics of each target obstacle factor; Indicates the first The first plot of land Interactive moderating terms of the characteristics of each target obstacle factor; Representing extremely small positive numbers, in the embodiments of the present invention ; This represents the global scaling parameter.

[0086] It should be noted that the weighting terms in the formula The normalized quadratic deviation calculated in step S2 was used. Because... and They represent the first The first obstacle factor feature is related to the first The plot of land and the first The importance of each plot of land is precisely quantified by their geometric mean. Each obstacle factor characteristic represents the importance of a common key obstacle for both plots. This weighting term ensures that the calculation of the micro-improvement feasibility correction factor automatically focuses on obstacle dimensions important to both parties, while ignoring issues not shared by both. Based on this, the formula compares the improvement feasibility of the two plots, and this invention obtains... This item. (Among them) This demonstrates the necessity of improvement (how serious the problem is), and This represents the difficulty of improvement (how much of an obstacle there is due to inherent attributes). The ratio of these two values ​​can be understood as the degree of improvement demand after adjusting for the difficulty of improvement. A plot of land with a serious problem but easy improvement will have a high value for this value, indicating that it has high improvement potential and value. The logarithmic difference between the two plots in terms of the degree of improvement demand was calculated. Combining the above two parts, The calculation involves weighting the differences in the degree of improvement required across all dimensions based on the importance of common obstacles. The denominator is... The total difference is then normalized to obtain a relative difference value. For two plots with the same key obstacles, this overall structure focuses on comparing the difficulty of improvement determined by their respective intrinsic properties and quantifies the resulting differences in improvement feasibility. If two plots both face high salinity, but one is easy to improve (e.g., sandy soil) and the other is difficult to improve (e.g., clay soil), the difference value calculated by the micro-improvement feasibility correction factor will be large. Finally, through... The structure transforms this normalized difference measure into a range within When the improvement feasibilities of two plots on a common key obstacle are similar, the micro-improvement feasibility correction factor tends to 1, and no additional penalty is generated. When the improvement feasibilities are quite different, the micro-improvement feasibility correction factor will tend to 2, and a significant gain is imposed on the calculation of the distance, so that the final classification result can not only reflect the similarity of the problem representations, but also reflect the similarity of the improvement measures.

[0087] At this point, the micro-improvement feasibility correction factor is obtained by jointly analyzing the improvement demand degree and the adjustment characteristics of the plot.

[0088] Step S4: An improvement effect modified distance is obtained by fusing the Euclidean distance and the correction factor.

[0089] After the completion of step S2 and step S3 and the acquisition of the macro-obstacle pattern correction factor and the micro-improvement feasibility correction factor, for any two target plots, the calculation result of multiplying the macro-obstacle pattern correction factor and the micro-improvement feasibility correction factor of the two target plots is taken as the distance optimization weight of the two target plots, and the calculation result of multiplying the distance optimization weight of the two target plots and the Euclidean distance of the two target plots is taken as the improvement effect modified distance of the two target plots.

[0090] At this point, the improvement effect modified distance is obtained by fusing the Euclidean distance and the correction factor.

[0091] Step S5: An intelligent classification result of saline-alkali soil is obtained by clustering analysis on the improvement effect modified distance.

[0092] This step uses the improvement effect modified distance calculated in step S4 to perform clustering analysis on all plots in the study area, thereby realizing intelligent classification. In this embodiment, the K-means algorithm is selected to process all plot samples. Before executing the K-means algorithm, the optimal clustering number K value needs to be determined. The determination of the K value is realized by the elbow method. This method calculates the total intra-cluster sum of squares of clustering under different K values, and plots the curve of the total intra-cluster sum of squares of clustering with respect to the K value, and selects the K value corresponding to the elbow point at which the change in the slope of the curve is most significant as the optimal clustering number.

[0093] After the optimal K value is determined, the K-means algorithm is executed. In this process, the distance measurement function inside the algorithm by default is replaced by the improvement effect modified distance constructed in step S4. The algorithm iteratively assigns each plot to the cluster center with the closest modified distance and updates the position of the cluster center according to the improvement effect modified distance, until the convergence condition is met, and finally divides the entire plot set into K categories.

[0094] The classification result generated by the present application can provide clear basis for subsequent decision and precise improvement. After obtaining the final K clusters, the characteristics of each category are represented by performing statistical analysis (e.g., calculating mean, variance, etc.) on each parameter of all plots in the cluster.

[0095] For example, the mean of the normalized quadratic deviation of all plots in the category presents a very high peak in the dimension corresponding to soil conductivity, and the mean of the interaction adjustment term is significantly lower in the soil conductivity dimension. Therefore, the core problem faced by the plots in this category is high salinity, and the inherent properties of the soil are very conducive to leaching improvement.

[0096] If the mean of the normalized quadratic deviation also presents a very high peak in the soil conductivity dimension, but the mean of the interaction adjustment term is significantly higher in the soil conductivity dimension, then the core problem of the plots in this category is also high salinity, but the inherent properties of the soil pose a greater obstacle to leaching improvement.

[0097] For the first type of plot, a conventional leaching and salt removal scheme can be developed and implemented. For the second type of plot, additional measures such as deep plowing and sand mixing should be taken to improve soil permeability before leaching and salt removal. In this way, the classification result is directly related to the effect of specific improvement measures, providing a solid data foundation for intelligent zoning management of saline-alkali soil.

[0098] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A smart classification method for saline-alkali land based on the effects of improvement measures, characterized in that, The method includes: Step S1: Obtain the basic parameter set by collecting and adaptively calculating multi-source feature data of the land parcel; Step S2: Obtain macroscopic obstacle pattern correction factors by performing differential analysis on the structure of land parcel obstacle factors; Step S3: Obtain micro-level improvement feasibility correction factors by jointly analyzing the land improvement demand and adjustment characteristics; Step S4: Obtain the improved effect correction distance by fusing the Euclidean distance and the correction factor; Step S5: Obtain the intelligent classification results of saline-alkali land by performing cluster analysis on the distance corrected for the improvement effect; The step of obtaining macro-obstacle pattern correction factors by performing difference analysis on the structure of land parcel obstacle factors includes: calculating structural differences by performing normalized quadratic deviation data of land parcels to obtain obstacle difference measurement results; and obtaining macro-obstacle pattern correction factors by performing nonlinear amplification and penalty control on the obstacle difference measurement results. The method of obtaining micro-level improvement feasibility correction factors by jointly analyzing the land improvement demand and adjustment characteristics includes: obtaining land improvement demand data by normalizing the ratio of standardized deviation and interaction adjustment term data of land parcels; obtaining the improvement feasibility difference measurement results between land parcels by performing weighted difference analysis on improvement demand and obstacle factor weight data; and obtaining micro-level improvement feasibility correction factors by performing nonlinear mapping and range control on the improvement feasibility difference measurement results.

2. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The process of acquiring a basic parameter set through the collection and adaptive calculation of multi-source feature data of land parcels includes: By collecting data from all plots within the study area, multi-source feature data, including barrier factor characteristics and sensitivity modulation characteristics, were obtained. The barrier factor features include soil electrical conductivity data and soil pH data to reflect the barrier strength of saline-alkali land; the sensitivity adjustment features include soil clay content percentage data and soil cation exchange capacity data to reflect the improvement response characteristics of the plot; and the vegetation index calculated from remote sensing images is used as the comprehensive quality index data of the plot. For the multi-source feature data of land parcels, all land parcels are sorted from high to low according to the comprehensive quality index data of the land parcels, and the top 3% of land parcels with the comprehensive quality index value are selected to form a reference land parcel subset. By averaging the soil electrical conductivity, soil pH, soil clay content percentage, and soil cation exchange capacity data of each plot within the reference plot subset, ideal data for soil electrical conductivity, soil pH, soil clay content percentage, and soil cation exchange capacity are obtained. By calculating the standard deviation of soil electrical conductivity and soil pH data for all plots within the reference plot subset, the standard deviation of soil electrical conductivity and soil pH data for the reference plot subset are obtained. For any target obstacle factor feature of any target plot in the reference plot subset, the absolute value of the result of subtracting the target obstacle factor feature data from the ideal data of the target obstacle factor feature is used as the numerator, the standard deviation data of the target obstacle factor feature is used as the denominator, and the corresponding fraction is used as the standardized deviation of the target obstacle factor of the target plot. For any target sensitivity adjustment feature of any target plot in the reference plot subset, the absolute value of the result of subtracting the target sensitivity adjustment feature data of the target plot from the ideal data of the target sensitivity adjustment feature is taken as the deviation of the target sensitivity adjustment feature of the target plot. For all plots in the reference plot subset, the Pearson correlation coefficient between the standardized deviation data of the target obstacle factors of all plots and the deviation data of the target sensitivity moderating features of all plots is used as the correlation coefficient between the target obstacle factors and the target sensitivity moderating features. The absolute value of the correlation coefficient between the target obstacle factors and the target sensitivity moderating features is subjected to an exponential mapping with the natural constant as the base, and the corresponding mapping result is used as the influence coefficient between the target obstacle factors and the target sensitivity moderating features. For any target obstacle factor feature of any target plot in the reference plot subset, the sum of the squares of the standardized deviations of all obstacle factor feature dimensions is used as the denominator, the square of the standardized deviation of the target obstacle factor of the target plot is used as the numerator, and the corresponding fraction is used as the normalized quadratic deviation of the target obstacle factor feature of the target plot. For any target obstacle factor feature and any target sensitivity adjustment feature of any target plot in the reference plot subset, the result of subtracting the target sensitivity adjustment feature data from the ideal data of the target sensitivity adjustment feature is used as the first difference assessment of the target sensitivity adjustment feature of the target plot; the result of multiplying the first difference assessment of the target sensitivity adjustment feature of the target plot and the influence coefficient between the target obstacle factor feature and the target sensitivity adjustment feature is used as the interaction adjustment assessment between the target obstacle factor feature and the target sensitivity adjustment feature; the interaction adjustment assessment between the target obstacle factor feature and the target sensitivity adjustment feature is mapped by a cosine function, and the reciprocal of the result of adding the mean of the cosine function mapping results of all sensitivity adjustment features to a constant 1 is used as the interaction adjustment term of the target obstacle factor of the target plot.

3. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The process of calculating structural differences from normalized quadratic deviation data of land parcels to obtain obstacle difference measurement results includes: For any two target plots and any target obstacle factor feature, the square of the difference between the square roots of the normalized quadratic deviations of the two target plots in the target obstacle factor feature is used as the first deviation difference assessment of the two target plots in the target obstacle factor feature; the first deviation difference assessments of the two target plots in all obstacle factor features are summed and the square root is calculated and divided by the square root of 2, and the corresponding calculation result is used as the obstacle difference measurement result of the two target plots.

4. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The process of obtaining a macroscopic obstacle pattern correction factor by nonlinearly amplifying and penalizing the obstacle difference measurement results includes: Set the maximum penalty coefficient and sensitivity index; for any two target plots, calculate the obstacle difference measurement results of the two target plots through the sensitivity index and multiply it by the maximum penalty coefficient. Add the calculation result to the constant 1 and use the result as the macroscopic obstacle pattern correction factor for the two target plots.

5. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The process of obtaining land improvement demand data by normalizing the ratio of standardized deviation and interaction adjustment term data of land parcels includes: For any target obstacle factor feature of any target plot, the standardized deviation of the target plot in the target obstacle factor feature is used as the numerator, and the interaction adjustment term of the target plot in the target obstacle factor feature is used as the denominator. The calculation result of adding the corresponding fraction with the smallest positive number is logarithmically mapped, and the mapping result is used as the improvement demand data of the target plot in the target obstacle factor feature.

6. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The method involves performing a weighted difference analysis on the improvement demand and obstacle factor weights to obtain the measurement results of the differences in improvement feasibility among land parcels, including: For any two target plots, the result of multiplying the square roots of the normalized quadratic deviations of the two target plots on the same obstacle factor feature is used as the feature weight of the two target plots on the same obstacle factor feature. The absolute value of the difference between the improvement demand data of the two target plots for the same obstacle factor feature is used as the assessment of the difference in improvement demand between the two plots for the same obstacle factor feature. The difference in improvement needs between the two target plots under all obstacle factor feature dimensions is assessed by weighted summation based on the corresponding feature weights. The weighted summation result is then divided by the sum of all obstacle factor feature weight data to measure the difference in improvement feasibility between the two target plots.

7. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The process of obtaining a micro-level improvement feasibility correction factor by performing nonlinear mapping and range control on the improvement feasibility difference measurement results includes: Set global scaling parameters; for any two target plots, multiply the result of the difference in improvement feasibility between the two target plots by the global scaling parameters and map it using the hyperbolic tangent function. The result of the mapping is then added to the constant 1 and used as the micro-improvement feasibility correction factor for the two target plots.

8. The intelligent classification method for saline-alkali land based on the effect of improvement measures according to claim 1, characterized in that, The step of obtaining the improved effect correction distance by fusing Euclidean distance and correction factor includes: For any two target plots, the result of multiplying the macro-obstacle pattern correction factor and the micro-improvement feasibility correction factor of the two target plots is used as the distance optimization weight of the two target plots; the result of multiplying the distance optimization weight of the two target plots and the Euclidean distance of the two target plots is used as the improvement effect correction distance of the two target plots.

Images