Method and system for estimating soil salinity based on feature extraction and machine learning

Abstract

The invention discloses a method and system for estimating soil salinity based on feature extraction and machine learning, and relates to the technical field of agricultural water and soil engineering.The method comprises the following steps that data information is collected, vegetation canopy reflectivity data is calculated according to the collected data information, feature parameters are extracted from the data information, and the feature parameters are evaluated; according to an evaluation result, combining the evaluated characteristic parameters into different modeling schemes, and constructing different soil salinity estimation models according to the modeling schemes; verifying the different soil salinity estimation models to obtain a qualified soil salinity estimation model; and deploying the qualified soil salinity estimation model on a hardware platform, and analyzing the sample. According to the method, the soil salinity estimation model is established by collecting the canopy reflectivity, the remote sensing image and the salinity data in different growth periods under vegetation coverage, and compared with the precision of models established based on different data sources, the investment of manpower and material resource cost is greatly saved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural water and soil engineering, and in particular to a method and system for estimating soil salinity based on feature extraction and machine learning. Background Art

[0002] At present, soil salinity is an important indicator in agricultural production, which directly affects the growth, development and yield of crops. Traditional soil salinity monitoring methods usually rely on field soil sample collection and laboratory testing. This method is not only time-consuming and labor-intensive, but also has limitations, such as uneven sample collection points and irregular sample processing. In addition, since soil salinity is affected by many factors, traditional monitoring methods often cannot fully and accurately evaluate the distribution of soil salinity. Summary of the invention

[0003] In view of the problems existing in the existing systems for estimating soil salinity based on feature extraction and machine learning, the present invention is proposed.

[0004] Therefore, the problem to be solved by the present invention is that traditional monitoring methods are often unable to comprehensively and accurately evaluate the distribution of soil salt.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0006] In a first aspect, an embodiment of the present invention provides a method for estimating soil salinity based on feature extraction and machine learning, which comprises:

[0007] Collect data information, calculate vegetation canopy reflectance data based on the collected data information, extract characteristic parameters from the data information, and evaluate the characteristic parameters;

[0008] According to the evaluation results, the characteristic parameters that have passed the evaluation are combined into different modeling schemes, and multiple groups of soil salinity estimation models are constructed according to the modeling schemes;

[0009] Verify multiple groups of soil salt estimation models and select the optimal soil salt estimation model;

[0010] The optimal soil salinity estimation model was deployed on the hardware platform and the samples were analyzed.

[0011] As a preferred embodiment of the method for estimating soil salinity based on feature extraction and machine learning described in the present invention, the step of collecting data information includes:

[0012] Obtain multi-source remote sensing data by using drones carrying different sensors, including drone multispectral images and thermal infrared data, and obtain spectral band reflectance based on multi-source remote sensing data;

[0013] Obtain soil salinity data through conductivity meter;

[0014] The raw data of vegetation canopy is obtained through the canopy analyzer.

[0015] As a preferred solution of the method for estimating soil salinity based on feature extraction and machine learning described in the present invention, the calculation of vegetation canopy reflectance data includes the calculation of canopy transmittance, the calculation of extinction coefficient and the calculation of leaf area index of black leaf model, wherein the calculation formula for canopy transmittance is:

[0016]

[0017] The calculation formula for the extinction coefficient is:

[0018]

[0019] The calculation formula for leaf area index of black leaf model is:

[0020] L a =L[1-g(1-a)]

[0021] In the formula, τ represents the canopy transmittance, L represents the canopy leaf area index, La represents the black leaf model leaf area index, K represents the extinction coefficient, a represents the absorption coefficient of the PAR band, and f b It is expressed as the ratio of direct light to incident light, x is the distribution parameter of the elliptical leaf inclination angle, θ is the incident zenith angle of sunlight, and g, A, B and C are all expressed as functions of x.

[0022] As a preferred solution of the method for estimating soil salinity based on feature extraction and machine learning described in the present invention, the step of extracting feature parameters from data information includes:

[0023] Set the number of division groups X, divide the data information into X groups, and pre-process the data information;

[0024] Calculate the spectral index and vegetation index of vegetation growth through data information;

[0025] The texture features are extracted from multispectral images using gray level co-occurrence matrix method;

[0026] The spectral index, texture characteristics and spectral band reflectance are integrated into a feature parameter set;

[0027] The preprocessing includes cleaning and standardization, and the spectral indicators include salt index and vegetation index.

[0028] As a preferred embodiment of the method for estimating soil salinity based on feature extraction and machine learning described in the present invention, the method for evaluating the feature parameter set includes the steps of:

[0029] Assign importance scores to each feature parameter;

[0030] Normalize the selected characteristic parameters;

[0031] Set a scoring threshold and filter out feature parameters with high contribution based on the scoring threshold;

[0032] The screening method includes: when the score of the feature parameter is greater than the score threshold, the feature parameter is retained; when the score of the feature parameter is not greater than the score threshold, the feature parameter is deleted.

[0033] As a preferred embodiment of the method for estimating soil salinity based on feature extraction and machine learning of the present invention, the modeling step of the soil salinity estimation model includes:

[0034] The obtained feature parameters are combined to design multiple sets of modeling schemes, wherein the modeling schemes include a single feature scheme, a multi-feature combination scheme, and a full feature scheme;

[0035] The modeling scheme was trained using a machine learning algorithm, with measured soil salinity data used as the training set;

[0036] A soil salinity estimation model was constructed based on the modeling scheme.

[0037] As a preferred embodiment of the method for estimating soil salinity based on feature extraction and machine learning described in the present invention, the steps of verifying different soil salinity estimation models are as follows:

[0038] Divide the dataset into k subsets of equal size;

[0039] A subset was randomly selected as the validation set, and the remaining subsets were used as training sets. The subset used as the validation set was marked as verified, and the evaluation indicators of the soil salinity estimation model on the validation set were recorded.

[0040] Randomly select a subset that is not marked as verified as a verification set, and record the evaluation indicators of the soil salinity estimation model on the verification set until all subsets are marked;

[0041] Based on the recorded evaluation indicators, the optimal soil salinity estimation model was obtained;

[0042] Among them, the judgment method for obtaining the optimal soil salinity estimation model is:

[0043] In the validation of multiple sets of soil salinity estimation models, the predicted soil salinity content was recorded;

[0044] Calculate the coefficient of determination R for each soil salinity estimation model 2and the root mean square error RMSE, through the determination coefficient R 2 It reflects the fitting degree of the soil salinity estimation model to the data set, and the root mean square error RMSE reflects the difference between the predicted value and the measured value;

[0045] The determination coefficient R of different models at the same depth 2 The prediction accuracy and stability of all soil salinity estimation models were analyzed by comparing with the root mean square error (RMSE) value.

[0046] Among them, the coefficient of determination R 2 The calculation formula of the root mean square error RMSE is:

[0047]

[0048] In the formula, yi represents the measured value of soil salt content, Expressed as predicted value, It is expressed as the average predicted value, and n is the number of samples;

[0049] When comparing the determination coefficient R2 and the root mean square error RMSE value, a scoring formula is established to determine the soil salt estimation model with the largest value of the scoring formula, and the soil salt estimation model is output as the optimal soil salt estimation model;

[0050] The scoring formula is:

[0051] F i =(R 2 i -R 2 min )·α+(RMSE max -RMSE i )·β

[0052] In the formula, F i It is represented as the score of the i-th model, α and β are the weight values ​​of the determination coefficient and the root mean square error, respectively, and R 2 min Expressed as the minimum value of the coefficient of determination, RMSE max Expressed as the maximum value of the root mean square error.

[0053] In a second aspect, an embodiment of the present invention provides a system for estimating soil salinity based on feature extraction and machine learning, which includes an information collection module, a feature evaluation module, a model building module, and an analysis module;

[0054] The information acquisition module is used to acquire multi-source remote sensing data through sensors and perform preliminary calculations based on the acquired data;

[0055] The feature evaluation module is used to evaluate each feature parameter and establish different modeling schemes;

[0056] The model building module is used to establish a soil salinity estimation model according to the modeling scheme;

[0057] The analysis module is used to evaluate soil salinity estimation models, select qualified models, and analyze samples using qualified models.

[0058] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, any step of the above-mentioned method for estimating soil salinity based on feature extraction and machine learning is implemented.

[0059] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned method for estimating soil salinity based on feature extraction and machine learning is implemented.

[0060] The beneficial effects of the present invention are as follows: a soil salinity estimation model is established by collecting canopy reflectance, remote sensing images, and salt data at different growth stages under vegetation coverage, and the accuracy of models constructed based on different data sources is compared, and the salt data under different vegetation coverage are used to verify the accuracy of the model. The results show that the accuracy of the model obtained by combining multiple data sources is higher than that of the model constructed by a single variable group. The use of remote sensing technology, feature optimization, and machine learning methods to monitor soil salinity greatly saves the investment in manpower and material costs, further promotes the development of agricultural research and the transformation of results, and timely obtains the dynamic changes in soil salinity. It can provide a decision-making basis for the formulation of irrigation and fertilization systems, improve crop yields, and provide important parameter support for farmland ecosystem simulation, which is of great significance in the context of the world's growing food demand.

[0061] Based on multi-source remote sensing data, spectral band data, spectral index and texture features are used as modeling feature parameters, different data sources are combined, and different modeling schemes are constructed. In the existing technology, band data, vegetation index and temperature index are mostly used as modeling factors. On the basis of these studies, the present invention adds texture features and combines them to establish multiple groups of models respectively, and selects the best model according to the mean value of the accuracy and stability of each model. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0063] Figure 1 Scenario diagram of the method for estimating soil salinity based on feature extraction and machine learning.

[0064] Figure 2 Structural diagram of the method for estimating soil salinity based on feature extraction and machine learning.

[0065] Figure 3 Diagram of the first support unit structure of the method for estimating soil salinity based on feature extraction and machine learning.

[0066] Figure 4 Another view of the first support unit of the method for estimating soil salinity based on feature extraction and machine learning.

[0067] Figure 5 Diagram of the second support unit structure for the method of estimating soil salinity based on feature extraction and machine learning. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0069] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0070] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0071] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0072] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0073] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0074] Example 1

[0075] Reference Figure 1 to Figure 5 , which is the first embodiment of the present invention, provides a method for estimating soil salinity based on feature extraction and machine learning, comprising the following steps:

[0076] S1. Collect data information, calculate vegetation canopy reflectance data based on the collected data information, extract characteristic parameters from the data information, and evaluate the characteristic parameters.

[0077] The steps of collecting data information include:

[0078] Obtain multi-source remote sensing data by using drones carrying different sensors, including drone multispectral images and thermal infrared data, and obtain spectral band reflectance based on multi-source remote sensing data;

[0079] Obtain soil salinity data through conductivity meter;

[0080] The raw data of vegetation canopy is obtained through the canopy analyzer.

[0081] The calculation of vegetation canopy reflectance data includes the calculation of canopy transmittance, the calculation of extinction coefficient and the calculation of leaf area index of black leaf model. The calculation formula of canopy transmittance is:

[0082]

[0083] The calculation formula for the extinction coefficient is:

[0084]

[0085] The calculation formula for leaf area index of black leaf model is:

[0086] L a =L[1-g(1-a)]

[0087] Where τ represents the canopy transmittance, L represents the canopy leaf area index, La represents the black leaf model leaf area index, K represents the extinction coefficient, a represents the absorption coefficient of the PAR band (SunSCAN defaults to 0.85), and f b It is expressed as the ratio of direct light to incident light, x is the distribution parameter of the elliptical leaf inclination angle, θ is the incident zenith angle of sunlight, and g, A, B and C are all expressed as functions of x.

[0088] The steps of extracting characteristic parameters from data information include:

[0089] Set the number of division groups X, divide the data information into X groups, and pre-process the data information;

[0090] Calculate the spectral index and vegetation index of vegetation growth through data information;

[0091] The texture features are extracted from multispectral images using gray level co-occurrence matrix method;

[0092] The spectral index, texture characteristics and spectral band reflectance are integrated into a feature parameter set;

[0093] Wherein, the preprocessing includes cleaning and standardization, and the spectral index includes salt index and vegetation index;

[0094] For example, the number of groups X is set to 20, and the data is randomly divided into 20 groups based on the measured soil salinity data, spectral image data and thermal infrared data. The modeling feature parameters are selected. There is a strong correlation between vegetation growth and soil salinity. On the one hand, there are differences in leaf spectral reflectance. On the other hand, due to changes in soil salinity or leaf differences, the texture characteristics of spectral images will also have significant differences. Studies have shown that texture features have been widely used to estimate vegetation parameters. Combining texture features with spectral indicators can effectively improve the accuracy of the model.

[0095] It should be noted that some spectral indicators commonly used for soil salinization monitoring were selected, including 15 salt indexes, 15 vegetation indexes and 1 normalized relative canopy temperature; among them, the texture characteristics of spectral images reflect the changes in soil surface color and grayscale, which are closely related to the salt and water content factors of the soil itself;

[0096] When extracting texture features, the co-occurrence measurement window of ENVI 5.3 software was used, and the second-order statistical filtering tool was used to calculate the eight characteristic parameters of each band of the UAV multispectral image, including mean, variance, uniformity, contrast, difference, entropy, second-order moment and correlation. Finally, a total of 77 characteristic parameters including 5 band reflectances, 1 temperature index, 31 spectral indices and 40 texture feature data were used as independent variables and integrated into a characteristic parameter set.

[0097] S2. According to the evaluation results, the characteristic parameters that have passed the evaluation are combined into different modeling schemes, and multiple groups of soil salinity estimation models are constructed according to the modeling schemes.

[0098] The method for evaluating a feature parameter set comprises the steps of:

[0099] Assign importance scores to each feature parameter;

[0100] Normalize the selected characteristic parameters;

[0101] Set a scoring threshold and filter out feature parameters with high contribution based on the scoring threshold;

[0102] The screening method includes: when the score of the feature parameter is greater than the score threshold, the feature parameter is retained; when the score of the feature parameter is not greater than the score threshold, the feature parameter is deleted;

[0103] For example, according to the gradient boosting decision tree, 77 feature parameters of different depths are optimized, and the results are as follows Figure 2 As shown, the characteristic parameters of different groups are normalized, the threshold is set to 0.1, and the factors common to different depths are selected as modeling characteristic parameters. The following modeling characteristic parameters are selected according to the groups;

[0104] Spectral band group: blue light, red edge band reflectance and near infrared;

[0105] Spectral index group: Salinity Index 2, R-edge Salinity Index, Brightness Index, SI-T Salinity Index, Difference Vegetation Index, Enhanced Vegetation Index, Soil Adjusted Vegetation Index, Optimized Soil Adjusted Vegetation Index, Vegetation Difference Index, and Enhanced Vegetation Index;

[0106] Texture data set: blue-mean, blue-variance, blue-contrast, blue-entropy, blue-second-order moment, green-variance, red-edge band reflectance-mean, near-infrared-mean, near-infrared-uniformity, near-infrared-entropy, and near-infrared-second-order moment.

[0107] The modeling steps of the soil salinity estimation model include:

[0108] The obtained feature parameters are combined to design multiple sets of modeling schemes, wherein the modeling schemes include a single feature scheme, a multi-feature combination scheme, and a full feature scheme;

[0109] Among them, the single feature schemes are spectral band scheme, spectral index scheme and texture feature scheme;

[0110] The multi-feature combination schemes are spectral band-spectral index scheme, spectral band-texture feature scheme and spectral index-texture feature scheme;

[0111] All feature schemes are spectral band-spectral index-texture feature schemes;

[0112] The modeling scheme was trained using a machine learning algorithm, with measured soil salinity data used as the training set;

[0113] A soil salinity estimation model was constructed based on the modeling scheme;

[0114] For example, after using the boosted decision tree to screen the variables, machine learning algorithms, such as RF, SVM, GPR, and BPNN, were used to test the effects of different types of modeling feature parameter combinations on the accuracy of the soil salinity estimation model.

[0115] S3. Verify multiple groups of soil salinity estimation models and select the optimal soil salinity estimation model.

[0116] The steps to validate different soil salinity estimation models are as follows:

[0117] Divide the dataset into k subsets of equal size;

[0118] A subset was randomly selected as the validation set, and the remaining subsets were used as training sets. The subset used as the validation set was marked as verified, and the evaluation indicators of the soil salinity estimation model on the validation set were recorded.

[0119] Randomly select a subset that is not marked as verified as a verification set, and record the evaluation indicators of the soil salinity estimation model on the verification set until all subsets are marked;

[0120] Based on the recorded evaluation indicators, the optimal soil salinity estimation model was obtained;

[0121] For example, the data set is divided into 10 subsets of equal size, and the soil salinity estimation model is trained in turn. 20 modeling analyses are performed for different modeling schemes. The results are as follows: Figure 3 As shown in the figure, the remaining 30% validation set data in the 20 groups of data are used to verify the models selected from the four depths of 0-10cm, 10-20cm, 20-30cm and 30-40cm. The statistic R 2 The soil salinity estimation model was evaluated by 20 validations. Figure 4 As shown;

[0122] The modeling results of 20 different modeling schemes were summarized. Figure 3 The results of the analysis and comparison of the modeling methods showed that: at a depth of 0 to 10 cm, the modeling accuracy obtained by the full variable combination of RF and GPR methods as modeling factors was higher than that of the single variable group and the pairwise combination; the model accuracy of the SVM and BPNN methods based on the spectral index and texture feature data set was the highest; at a depth of 10 to 20 cm, the model accuracy obtained by the combination of spectral index and texture features of RF and GPR methods was higher than that of other schemes, and the results obtained by the other two modeling methods were poor; at a depth of 20 to 30 cm, the RF model constructed based on spectral index and texture features showed better results, which was better than the model established under a single variable; at a depth of 30 to 40 cm, the GPR model constructed based on the full variable combination had the best accuracy, and the model accuracy based on the multivariate combination under other modeling methods was also higher than that of the single variable combination. Comprehensive comparative analysis showed that the texture data group in the single variable performed poorly, but the combination of texture features of multispectral images with other variables could improve the accuracy of the soil salinity inversion model under vegetation cover at different depths. Compared with the modeling using spectral index or texture features alone, the accuracy of the model after the combination of the two was improved.

[0123] Among them, the judgment method for obtaining the optimal soil salinity estimation model is:

[0124] In the validation of multiple sets of soil salinity estimation models, the predicted soil salinity content was recorded;

[0125] Calculate the coefficient of determination R for each soil salinity estimation model 2 and the root mean square error RMSE, through the determination coefficient R 2 It reflects the fitting degree of the soil salinity estimation model to the data set, and the root mean square error RMSE reflects the difference between the predicted value and the measured value;

[0126] The determination coefficient R of different models at the same depth 2 The prediction accuracy and stability of all soil salinity estimation models were analyzed by comparing with the root mean square error (RMSE) value.

[0127] Among them, the coefficient of determination R 2 The calculation formula of the root mean square error RMSE is:

[0128]

[0129] In the formula, yi represents the measured value of soil salt content, Expressed as predicted value, It is expressed as the average predicted value, and n is the number of samples;

[0130] Coefficient of determination R 2 When compared with the root mean square error RMSE value, a scoring formula is established to determine the soil salt estimation model with the largest value of the scoring formula, and the soil salt estimation model is output as the optimal soil salt estimation model;

[0131] The scoring formula is:

[0132] F i =(R 2 i -R 2 min )·α+(RMSE max -RMSE i )·β

[0133] In the formula, F i It is represented as the score of the i-th model, α and β are the weight values ​​of the determination coefficient and the root mean square error, respectively, and R 2 min Expressed as the minimum value of the coefficient of determination, RMSE max Expressed as the maximum value of the root mean square error, for example, when the coefficient of determination R 2 When the importance is twice the root mean square error RMSE, set α to 2 and β to 1;

[0134] Assume three models A, B and C, and their R 2 And the RMSE values ​​are as follows:

[0135] Model A: R 2 A =0.85, RMSE A =0.2;

[0136] Model A: R 2 B =0.8, RMSE B =0.25;

[0137] Model A: R 2 C =0.9, RMSE C =0.15;

[0138] And R 2 min =0.8, RMSE max =0.25, and the two sets of numerical weights are equal;

[0139] Therefore, by calculation we obtain:

[0140] Model A score: (0.85-0.8)*1+(0.25-0.2)*1=0.1;

[0141] Model B score: (0.8-0.8)*1+(0.25-0.25)*1=0;

[0142] Model C score: (0.9-0.8)*1+(0.25-0.15)*1=0.2;

[0143] From the scores, we can see that model C is better than model A which is better than model B.

[0144] S4. Deploy the optimal soil salinity estimation model on the hardware platform and analyze the samples.

[0145] right Figure 4 For analysis, it should be noted that: Figure 4 For attachment Figure 3 A summary of the 20 validation results of the best model selected at each depth;

[0146] At a depth of 0 to 10 cm, the R2 values ​​of the validation set statistics mostly fluctuate between 0.6 and 0.9, with an average R2 of 0.77, and the root mean square error (RMSE) values ​​are low, usually between 0.1 and 0.3, with an average RMSE of 0.185;

[0147] At a depth of 10 to 20 cm, the R2 value of the statistic fluctuates between 0.4 and 0.9, with an average R2 of 0.67, and the root mean square error (RMSE) value ranges from 0.2 to 0.4, with an average RMSE of 0.325;

[0148] At a depth of 20 to 30 cm, the R2 value of the validation set fluctuated between 0.5 and 0.9, with an average R2 of 0.71, and the RMSE value was usually between 0.2 and 0.5, with an average RMSE of 0.36;

[0149] At a depth of 30 to 40 cm, the R2 value of the validation set ranged from 0.4 to 0.9, the average R2 was 0.71, the RMSE value ranged from 0.2 to 0.5, and the average RMSE was 0.309.

[0150] A comprehensive comparison showed that the models screened out under different schemes all obtained good verification results. The accuracy of the soil salinity estimation model based on a single variable combination was lower than that of the model obtained by combining all variables. In the single variable group, the accuracy of the texture feature parameters was low, but the combination of texture information with other variables resulted in higher model accuracy. The model obtained by combining multiple variables was verified at different coverage levels and based on the remaining 30% of the data, and the model accuracy was higher.

[0151] In summary, a soil salinity estimation model was established by collecting canopy reflectance, remote sensing images, and salt data at different growth stages under vegetation coverage. The accuracy of the models built based on different data sources was compared, and the salt data under different vegetation coverage was used to verify the accuracy of the model. The results showed that the accuracy of the model obtained by combining multiple data sources was higher than that of the model built by a single variable group. The use of remote sensing technology, feature optimization, and machine learning methods to monitor soil salinity has greatly saved the investment of manpower and material costs, further promoted the development of agricultural research and the transformation of results, and timely acquisition of the dynamic changes in soil salinity can provide a decision-making basis for the formulation of irrigation and fertilization systems, increase crop yields, and provide important parameter support for farmland ecosystem simulation, which is of great significance in the context of the world's growing food demand.

[0152] Example 2

[0153] On the basis of the first embodiment, this embodiment further provides a system for estimating soil salinity based on feature extraction and machine learning, including an information collection module, a feature evaluation module, a model building module, and an analysis module;

[0154] The information acquisition module is used to acquire multi-source remote sensing data through sensors and perform preliminary calculations based on the acquired data;

[0155] The feature evaluation module is used to evaluate each feature parameter and establish different modeling schemes;

[0156] The model building module is used to establish a soil salinity estimation model according to the modeling scheme;

[0157] The analysis module is used to evaluate soil salinity estimation models, select qualified models, and analyze samples using qualified models.

[0158] This embodiment also provides a computer device, which is suitable for the method of estimating soil salinity based on feature extraction and machine learning, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the method of estimating soil salinity based on feature extraction and machine learning proposed in the above embodiment.

[0159] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0160] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for estimating soil salinity based on feature extraction and machine learning proposed in the above embodiment is implemented.

[0161] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0162] Example 3

[0163] Based on the previous two embodiments, this embodiment provides a method for estimating soil salinity based on feature extraction and machine learning.

[0164] The LAI value was used as the vegetation coverage and divided into three different coverages: low, medium and high. The soil salinity data was divided according to the three different coverages to obtain three data sets with different coverages, of which 24 samples were low coverage, 27 samples were medium coverage and 34 samples were high coverage. The samples with different coverages were used to verify the estimation accuracy of the best model obtained at different depths. The results are as follows: Figure 5 shown.

[0165] It should be noted that: Figure 5 For attachment Figure 3 Summary of soil salinity inversion results based on different coverages of the inversion models selected at each depth.

[0166] Analyze the results: The inversion accuracy R of all models 2 All above 0.5, with the highest R 2The value is 0.92. Overall, the inversion accuracy is good. For the model established at a depth of 0 to 10 cm, the inversion result obtained at medium coverage is the best, which is also the highest accuracy among all models. For the inversion models established at depths of 10 to 20 cm and 20 to 30 cm, the model inversion accuracy is the best at high coverage, R 2 0.8 and 0.85; in the model established at a depth of 30-40 cm, the inversion accuracy obtained at high coverage is the highest, R 2 The inversion accuracy of the surface soil at 0-10 cm is the highest under low and medium coverage, followed by the deep soil at 30-40 cm, and the worst is the soil at 10-20 cm; under high coverage, the best inversion accuracy is the deep soil at 30-40 cm; the comparison shows that the inversion accuracy of the model at each depth under high coverage is good, which shows that the selected model has high accuracy and can estimate soil salinity more accurately.

[0167] In summary, based on multi-source remote sensing data, spectral band data, spectral index and texture features are used as modeling feature parameters, different data sources are combined, and different modeling schemes are constructed. In the existing research, most of the band data, vegetation index and temperature index are used as modeling factors. On the basis of these studies, the present invention adds texture features and combines them to establish multiple groups of models respectively, and selects the best model according to the mean value of the accuracy and stability of each model.

[0168] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the present invention.

Claims

1. A method for estimating soil salinity based on feature extraction and machine learning, characterized in that: The following steps are included: Collect data information, calculate vegetation canopy reflectance data based on the collected data information, extract characteristic parameters from the data information, and evaluate the characteristic parameters; According to the evaluation results, the characteristic parameters that have passed the evaluation are combined into different modeling schemes, and multiple groups of soil salinity estimation models are constructed according to the modeling schemes; Verify multiple groups of soil salt estimation models and select the optimal soil salt estimation model; The optimal soil salinity estimation model was deployed on the hardware platform and the samples were analyzed.

2. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 1, characterized in that: The steps of collecting data information include: Acquire multi-source remote sensing data by using drones carrying different sensors, including drone multispectral images, thermal infrared data, and use a canopy analyzer to measure canopy reflectance; Obtain soil salinity data through conductivity meter; The raw data of vegetation canopy is obtained through the canopy analyzer.

3. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 2, characterized in that: The calculation of vegetation canopy reflectance data includes the calculation of canopy transmittance, the calculation of extinction coefficient and the calculation of leaf area index of black leaf model. The calculation formula of canopy transmittance is: The calculation formula for the extinction coefficient is: The calculation formula for leaf area index of black leaf model is: L a =L[1-g(1-a)] In the formula, τ represents the canopy transmittance, L represents the canopy leaf area index, La represents the black leaf model leaf area index, K represents the extinction coefficient, a represents the absorption coefficient of the PAR band, and f b It is expressed as the ratio of direct light to incident light, x is the distribution parameter of the elliptical leaf inclination angle, θ is the incident zenith angle of sunlight, and g, A, B and C are all expressed as functions of x.

4. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 3, characterized in that: The steps of extracting characteristic parameters from data information include: Set the number of division groups X, divide the data information into X groups, and pre-process the data information; Calculate the spectral index and vegetation index of vegetation growth through data information; The texture features are extracted from multispectral images using gray level co-occurrence matrix method; The spectral index, texture characteristics and spectral band reflectance are integrated into a feature parameter set; The preprocessing includes cleaning and standardization, and the spectral indicators include salt index and vegetation index.

5. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 4, characterized in that: The method for evaluating a feature parameter set comprises the steps of: Assign importance scores to each feature parameter; Normalize the selected characteristic parameters; Set a scoring threshold and filter out feature parameters with high contribution based on the scoring threshold; The screening method includes: when the score of the feature parameter is greater than the score threshold, the feature parameter is retained; when the score of the feature parameter is not greater than the score threshold, the feature parameter is deleted.

6. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 5, characterized in that: The modeling steps of the soil salinity estimation model include: The obtained feature parameters are combined to design multiple sets of modeling schemes, wherein the modeling schemes include a single feature scheme, a multi-feature combination scheme, and a full feature scheme; The modeling scheme was trained using a machine learning algorithm, with measured soil salinity data used as the training set; A soil salinity estimation model was constructed based on the modeling scheme.

7. The method for estimating soil salinity based on feature extraction and machine learning as claimed in claim 6, characterized in that: The steps to validate multiple soil salinity estimation models are as follows: Divide the dataset into k subsets of equal size; A subset was randomly selected as the validation set, and the remaining subsets were used as training sets. The subset used as the validation set was marked as verified, and the evaluation indicators of the soil salinity estimation model on the validation set were recorded. Randomly select a subset that is not marked as verified as a verification set, and record the evaluation indicators of the soil salinity estimation model on the verification set until all subsets are marked; Based on the recorded evaluation indicators, the optimal soil salinity estimation model was obtained; Among them, the judgment method for obtaining the optimal soil salinity estimation model is: In the validation of multiple sets of soil salinity estimation models, the predicted soil salinity content was recorded; Calculate the coefficient of determination R for each soil salinity estimation model 2 and the root mean square error RMSE, through the determination coefficient R 2 It reflects the fitting degree of the soil salinity estimation model to the data set, and the root mean square error RMSE reflects the difference between the predicted value and the measured value; The determination coefficient R of different models at the same depth 2 The prediction accuracy and stability of all soil salinity estimation models were analyzed by comparing with the root mean square error (RMSE) value. Among them, the coefficient of determination R 2 The calculation formula of the root mean square error RMSE is: In the formula, yi represents the measured value of soil salt content, Expressed as predicted value, It is expressed as the average predicted value, and n is the number of samples; When comparing the determination coefficient R2 and the root mean square error RMSE value, a scoring formula is established to determine the soil salt estimation model with the largest value of the scoring formula, and the soil salt estimation model is output as the optimal soil salt estimation model; The scoring formula is: F i =(R 2 i -R 2 min )·α+(RMSE max -RMSE i )·β In the formula, F i It is represented as the score of the i-th model, α and β are the weight values ​​of the determination coefficient and the root mean square error, respectively, and R 2 min Expressed as the minimum value of the coefficient of determination, RMSE max Expressed as the maximum value of the root mean square error.

8. A system for estimating soil salinity based on feature extraction and machine learning, based on the method for estimating soil salinity based on feature extraction and machine learning according to any one of claims 1 to 7, characterized in that: It includes information collection module, feature evaluation module, model building module, and analysis module; The information acquisition module is used to acquire multi-source remote sensing data through sensors and perform preliminary calculations based on the acquired data; The feature evaluation module is used to evaluate each feature parameter and establish different modeling schemes; The model building module is used to establish a soil salinity estimation model according to the modeling scheme; The analysis module is used to evaluate soil salinity estimation models, select qualified models, and analyze samples using qualified models.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for estimating soil salinity based on feature extraction and machine learning according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for estimating soil salinity based on feature extraction and machine learning according to any one of claims 1 to 7 are implemented.

Images