Performance optimization integrated learning soil salinity prediction precision improvement method

Abstract

The invention discloses a performance optimization integrated learning soil salinity prediction precision improvement method, and belongs to the technical field of soil quality monitoring. The method comprises the following steps: acquiring a first spectral image of to-be-predicted soil; performing feature extraction on the first spectral image to obtain a target spectral feature; inputting the target spectral features into a target soil salinity prediction model to obtain a soil salinity prediction result; the target soil salinity prediction model is obtained by the following steps: testing a plurality of first soil salinity prediction models based on a first soil salinity prediction model test set corresponding to sample soil to obtain a test scoring result of each first soil salinity prediction model; determining at least one second soil salinity prediction model from the plurality of first soil salinity prediction models based on the plurality of test scoring results, and taking the at least one second soil salinity prediction model as a target soil salinity prediction model; the input feature of the first soil salinity prediction model test set is the first spectral feature of the sample soil, and the target variable is the salinity of the sample soil.

Description

Technical Field

[0001] The present application belongs to the technical field of soil quality monitoring, and in particular, relates to a method for improving soil salinity prediction accuracy through performance optimization and integrated learning. Background Art

[0002] Traditional soil salinity monitoring methods rely on field sampling and laboratory analysis. Although they can provide accurate data, they are time-consuming, labor-intensive and costly, and it is difficult to conduct high-density sampling in large areas. In contrast, optical remote sensing technology provides an effective alternative for large-scale soil salinity monitoring with high spatial resolution. At present, based on drone sensors, combined with ensemble learning methods, it has become the main means of fine-scale soil salinity monitoring. Specifically, soil salinity prediction methods based on ensemble learning can be divided into three categories: bagging, boosting and stacking. The bagging method is sensitive to noisy data and it is difficult to significantly reduce the bias of the model, so its application in actual soil salinity prediction is limited. The boosting method is prone to overfitting and may require a lot of computing resources and memory when processing very large data sets. In contrast, the stacking model can integrate the advantages of various basic learners, so it performs better in prediction efficiency and suitability.

[0003] Although Stacking ensemble learning has advantages in fine-scale soil salinity prediction, the performance differences between base learners will affect the accuracy of the model. Therefore, optimizing the prediction performance of base learners remains a difficult point in improving the accuracy of fine-scale soil salinity prediction. Generally, when the prediction accuracy of some base learners is low, their errors will weaken the overall performance of the Stacking model. Currently, cross-validation techniques are widely used to select the best combination of base learners and optimize hyperparameters. Obviously, this method does not completely solve the differences in the contribution of each base learner to the final prediction, which will lead to a decrease in the performance of the Stacking model. In addition, complex combinations and hyperparameter optimization not only increase the computational cost, but also affect the stability of the model. Therefore, it is urgent to propose a more effective model optimization strategy to comprehensively improve the performance and stability of the Stacking model, which is crucial to improving the accuracy of soil salinity prediction.

[0004] In the current soil salinity prediction based on Stacking ensemble learning, the selection of base learners often depends on the subjective judgment of researchers, and this selection fails to fully consider the performance differences between the base learners. If some poorly performing base learners are included in the model, their low accuracy will directly affect the overall prediction effect of the Stacking model. Especially when the data set features are complex or the sample size is insufficient, the prediction errors of low-performance base learners may cause significant performance degradation. In addition, the strong correlation between base learners may also introduce redundant information, thereby increasing the risk of model overfitting. In this case, the overall performance of the Stacking model may be significantly lower than that of a single high-performance base learner, reducing its application value in soil salinity prediction.

[0005] Specifically, the incorrect predictions of low-performance base learners will not only directly affect the final results, but will also further amplify these errors through the weighting mechanism of the Stacking model. When the outputs of the base learners are used as the input of the meta-learner, if these outputs contain a lot of erroneous information, the final prediction results will inevitably be affected. This situation weakens the advantages of the Stacking model and even causes it to perform poorly in soil salinity prediction, thereby affecting the overall stability and accuracy of the model, reducing the reliability of predictions, and further limiting its promotion and implementation in practical applications, and failing to fully tap the potential of ensemble learning. Summary of the invention

[0006] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a method for improving the accuracy of soil salinity prediction by performance optimization integrated learning, which can overcome the problem of subjective selection of basic learners leading to reduced accuracy of soil salinity prediction models, and the results of soil salinity prediction by the target soil salinity prediction model are more accurate.

[0007] In a first aspect, the present application provides a method for improving soil salinity prediction accuracy through performance optimization ensemble learning, the method comprising:

[0008] Acquire a first spectral image of the soil to be predicted;

[0009] Extracting features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted;

[0010] Inputting the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model;

[0011] The target soil salinity prediction model is obtained by the following steps:

[0012] Testing multiple first soil salt prediction models based on a first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salt prediction models, wherein the test score result is obtained by weighted summing the scores of the first soil salt prediction model on multiple evaluation indicators;

[0013] Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as the target soil salinity prediction model;

[0014] Among them, the first soil salt prediction model is trained by the first soil salt prediction model training set corresponding to the sample soil, the input feature of the first soil salt prediction model test set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model test set is the salt of the sample soil, the input feature of the first soil salt prediction model training set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model training set is the salt of the sample soil, and the type of the target spectral feature corresponds to the type of the first spectral feature.

[0015] According to the performance optimization integrated learning soil salinity prediction accuracy improvement method of the present application, multiple first soil salt prediction models are tested, and weighted summation is performed according to the scores of the first soil salt prediction model on multiple evaluation indicators to obtain the test score result of the first soil salt prediction model. Based on the multiple test score results, at least one second soil salt prediction model is determined from the multiple first soil salt prediction models, and the at least one second soil salt prediction model is used as the target soil salt prediction model. This can overcome the current problem of reduced accuracy of soil salt prediction models due to subjective selection of basic learners, and the results of soil salt prediction by the target soil salt prediction model are more accurate.

[0016] According to an embodiment of the present application, inputting the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model includes:

[0017] Inputting the target spectral feature into at least one of the second soil salinity prediction models to obtain a soil salinity prediction sub-result output by each of the second soil salinity prediction models;

[0018] Based on at least one of the soil salinity prediction sub-results, the soil salinity prediction result is obtained.

[0019] According to an embodiment of the present application, before testing a plurality of first soil salinity prediction models based on the first soil salinity prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salinity prediction models, the method further includes:

[0020] Calculating a first information amount of the evaluation indicator, and calculating a first correlation between the evaluation indicator and other evaluation indicators;

[0021] Determining a weight coefficient corresponding to the evaluation index based on the first information amount and the first correlation;

[0022] The first soil salt prediction model test set based on the sample soil corresponding to the first soil salt prediction model is tested on multiple first soil salt prediction models to obtain a test score result of each of the first soil salt prediction models, including:

[0023] Testing the first soil salinity prediction model based on the first soil salinity prediction model test set to obtain scores of the first soil salinity prediction model on the plurality of evaluation indicators;

[0024] Based on the weight coefficient corresponding to each of the evaluation indicators, the scores on the multiple evaluation indicators are weighted and summed to obtain the test score result of the first soil salinity prediction model.

[0025] According to one embodiment of the present application, the first spectral feature is determined based on the following steps:

[0026] Based on the second soil salt prediction model training set of the sample soil, a feature screening model is trained to obtain a feature score for each second spectral feature output by the feature screening model, wherein the input feature of the second soil salt prediction model training set is the second spectral feature, and the target variable of the second soil salt prediction model training set is the salt content of the sample soil;

[0027] Based on the plurality of feature scores, at least one third spectral feature is determined from the plurality of second spectral features, and the at least one third spectral feature is used as the first spectral feature.

[0028] According to one embodiment of the present application, the feature screening model includes at least one of a partial least squares regression model, an extreme gradient boosting model, and a random forest model.

[0029] According to an embodiment of the present application, the types of the plurality of first soil salinity prediction models are different.

[0030] According to one embodiment of the present application, the type of the first soil salinity prediction model is a cube model, a quantile random forest model, a Bayesian regularized artificial neural network model, a multilayer perceptron model, a lightweight gradient boosting machine model, a multivariate adaptive regression spline model, a conditional inference tree model, an elastic net model, a Gaussian process model, a K nearest neighbor model, a support vector machine model or a classification and regression tree model.

[0031] According to one embodiment of the present application, the evaluation index includes at least one of a coefficient of determination, a root mean square error, a mean absolute error, a prediction-to-deviation ratio, and a linear correlation consistency coefficient.

[0032] In a second aspect, the present application provides a device for improving soil salinity prediction accuracy by performance optimization integrated learning, the device comprising:

[0033] An acquisition module, used for acquiring a first spectral image of the soil to be predicted;

[0034] A first processing module is used to extract features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted;

[0035] A second processing module is used to input the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model;

[0036] The target soil salinity prediction model is obtained by the following steps:

[0037] Testing multiple first soil salt prediction models based on a first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salt prediction models, wherein the test score result is obtained by weighted summing the scores of the first soil salt prediction model on multiple evaluation indicators;

[0038] Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as the target soil salinity prediction model;

[0039] Among them, the first soil salt prediction model is trained by the first soil salt prediction model training set corresponding to the sample soil, the input feature of the first soil salt prediction model test set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model test set is the salt of the sample soil, the input feature of the first soil salt prediction model training set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model training set is the salt of the sample soil, and the type of the target spectral feature corresponds to the type of the first spectral feature.

[0040] According to the performance optimization integrated learning soil salinity prediction accuracy improvement device of the present application, multiple first soil salinity prediction models are tested, and weighted summation is performed according to the scores of the first soil salinity prediction model on multiple evaluation indicators to obtain the test score result of the first soil salinity prediction model. Based on the multiple test score results, at least one second soil salt prediction model is determined from the multiple first soil salt prediction models, and the at least one second soil salt prediction model is used as the target soil salt prediction model. This can overcome the current problem of reduced accuracy of soil salt prediction models due to subjective selection of basic learners, and the result of soil salinity prediction by the target soil salt prediction model is more accurate.

[0041] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for improving soil salinity prediction accuracy through performance optimization integrated learning is implemented as described in the first aspect above.

[0042] In a fourth aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the performance optimized integrated learning method for improving soil salinity prediction accuracy as described in the first aspect above.

[0043] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the performance optimization integrated learning method for improving soil salinity prediction accuracy as described in the first aspect above.

[0044] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0046] Figure 1This is one of the flow diagrams of the performance optimization integrated learning soil salinity prediction accuracy improvement method provided in the embodiment of the present application;

[0047] Figure 2 This is the second flow chart of the method for improving soil salinity prediction accuracy through performance optimization integrated learning provided in the embodiment of the present application;

[0048] Figure 3 This is the third flow chart of the performance optimization integrated learning soil salinity prediction accuracy improvement method provided in the embodiment of the present application;

[0049] Figure 4 It is a structural schematic diagram of a device for improving soil salinity prediction accuracy through performance optimization integrated learning provided in an embodiment of the present application;

[0050] Figure 5 It is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0052] The terms "first", "second", etc. in the specification and claims of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of the same type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0053] In combination with the accompanying drawings, the performance optimization integrated learning soil salinity prediction accuracy improvement method, performance optimization integrated learning soil salinity prediction accuracy improvement device, electronic device and readable storage medium provided in the embodiments of the present application are described in detail through specific embodiments and their application scenarios.

[0054] Among them, the performance optimization integrated learning soil salinity prediction accuracy improvement method can be applied to the terminal, and can be specifically executed by the hardware or software in the terminal.

[0055] The terminal includes, but is not limited to, a portable communication device such as a mobile phone or tablet computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad). It should also be understood that in some embodiments, the terminal may not be a portable communication device, but a desktop computer with a touch-sensitive surface (e.g., a touch screen display and / or a touch pad).

[0056] In the following various embodiments, a terminal including a display and a touch-sensitive surface is described. However, it should be understood that the terminal may include one or more other physical user interface devices such as a physical keyboard, a mouse and a joystick.

[0057] The performance optimization integrated learning method for improving soil salinity prediction accuracy provided in the embodiments of the present application may be executed by an electronic device or a functional module or functional entity in an electronic device that can implement the performance optimization integrated learning method for improving soil salinity prediction accuracy. The electronic devices mentioned in the embodiments of the present application include but are not limited to computers, etc. The performance optimization integrated learning method for improving soil salinity prediction accuracy provided in the embodiments of the present application is described below using an electronic device as an example of the execution subject.

[0058] like Figure 1 As shown, the performance optimization integrated learning soil salinity prediction accuracy improvement method includes: step 110, step 120 and step 130.

[0059] Step 110: Acquire a first spectral image of the soil to be predicted.

[0060] The soil to be predicted is the soil for which salt prediction is to be performed, for example, the soil in farmland, nature reserve, industrial area or coastal area, etc. The salt prediction may include the prediction of the salt concentration, salt distribution pattern and salt change trend of the soil.

[0061] The first spectral image is a spectral image corresponding to the area where the soil to be predicted is located. The first spectral image may be a multi-spectral image or a hyperspectral image.

[0062] In actual implementation, the spectral image acquisition device can be mounted on a drone, and the drone can be controlled to fly above the area where the soil to be predicted is located. The spectral image acquisition device acquires images of the area where the soil to be predicted is located to obtain a first spectral image.

[0063] Step 120: extract features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted.

[0064] The target spectrum features may include spectrum value features and spectrum index features.

[0065] The spectral value feature may be the spectral value corresponding to the pixel point sampled in the first spectral image. The spectral value reflects the information of the pixel point in each spectral band, can be used to characterize the characteristics of the ground surface corresponding to the pixel point, and can be used as an effective indicator for soil salinity prediction.

[0066] The spectral index feature may include multiple spectral indices, and the types of spectral indices may include vegetation index, salt index, brightness index, intensity index, etc. Vegetation index is used to characterize vegetation growth under salt stress, salt index is used to indicate soil conditions, brightness index is used to reflect basic physical properties of the surface, and intensity index is used to indicate the impact of human activities.

[0067] In this step, the spectral value characteristics of each pixel in different bands can be obtained by sampling the pixels of the first spectral image. The values ​​of the pixels in each spectral band are calculated, that is, the values ​​of the pixels in each spectral band are substituted into the corresponding spectral index calculation formula, and the calculation results are determined as spectral index characteristics.

[0068] Step 130: input the target spectral feature into the target soil salinity prediction model to obtain the soil salinity prediction result output by the target soil salinity prediction model.

[0069] The target soil salinity prediction model is a model for predicting soil salinity based on input spectral features. The input of the target soil salinity prediction model is the spectral features, and the output of the target soil salinity prediction model is the soil salinity prediction result.

[0070] The soil salinity prediction results can be presented as a soil salinity spatial distribution map, which reflects the salt concentration, distribution pattern and change trend of different areas of the predicted soil. The soil salinity spatial distribution map can be used to identify saline-alkali land, optimize irrigation and fertilization management, and improve agricultural production efficiency.

[0071] In this embodiment, the sensitive features of the survey area, namely the target spectral features, are extracted from the drone images, and the performance optimized ensemble learning model (POEL), namely the target soil salinity prediction model, is used to predict the soil salinity in the survey area, thereby generating a predicted spatial distribution map of the soil salinity in the survey area.

[0072] The residuals between the predicted and measured values ​​of soil salinity were interpolated using the ordinary Kriging method, and the spatial distribution map of the soil salinity prediction residuals in the survey area was drawn.

[0073] The predicted spatial distribution map of soil salinity in the survey area is superimposed with the corresponding spatial distribution map of prediction residuals to generate the spatial distribution map of soil salinity in the survey area.

[0074] Among them, Figure 2 As shown in Figure 2, the target soil salinity prediction model is obtained through the following steps:

[0075] Step 210: Test multiple first soil salinity prediction models based on the first soil salinity prediction model test set corresponding to the sample soil to obtain a test score result of each first soil salinity prediction model.

[0076] In this embodiment, the first soil salinity prediction model is trained by the first soil salinity prediction model training set corresponding to the sample soil. The input feature of the first soil salinity prediction model test set is the first spectral feature of the sample soil, and the target variable of the first soil salinity prediction model test set is the salinity of the sample soil. The input feature of the first soil salinity prediction model training set is the first spectral feature of the sample soil, and the target variable of the first soil salinity prediction model training set is the salinity of the sample soil. The type of the target spectral feature corresponds to the type of the first spectral feature.

[0077] The sample soil may be soil from farmland, nature reserve, industrial area or coastal area. The first soil salinity prediction model training set is a training set for training the first soil salinity prediction model, and the first soil salinity prediction model test set is a test set for testing the first soil salinity prediction model.

[0078] In this embodiment, a soil sample may be collected and the salt content of the soil sample may be determined by a mass method.

[0079] First, soil sampling was carried out. Soil samples were collected from farmland in the survey area using a grid sampling scheme with a sampling depth of 0-10 cm, and the geographic coordinates of each soil sampling point were recorded using a handheld differential Global Positioning System (GPS) measuring device.

[0080] Then, the soil samples were pretreated. After being naturally air-dried in the laboratory, the soil samples were ground using a 2 mm pore size sieve to remove crop roots, stones and other impurities.

[0081] After pretreatment of soil samples, soil sample composition was tested and the salt content of soil samples was determined by mass method.

[0082] In this embodiment, based on drone images, a series of features are selected through pixel extraction and band calculation to predict soil salinity, and the features include spectral value features and spectral index features.

[0083] UAV image acquisition was performed. Under clear weather conditions, UAV multispectral images of the soil sample area were collected to ensure consistency with the soil sampling time. The UAV flight altitude was set to 100 meters, and the ground resolution could reach 3 cm. PIX4D software was used to pre-process the UAV multispectral images, including generating point clouds, building 3D models, and orthorectification. At the same time, the UAV images were geometrically corrected with the help of ground control points.

[0084] In this embodiment, according to the geographical coordinates of the collected soil sample sampling points, the spectral value characteristics of the image pixels at the positions corresponding to the soil sample sampling points are obtained, and the spectral index characteristics are calculated as the first spectral characteristics.

[0085] The first spectral feature of the soil sampling point is used as the input feature, and the corresponding soil salt content is used as the target variable to construct a first soil salt prediction model data set. The first soil salt prediction model data set is divided to obtain a first soil salt prediction model training set and a first soil salt prediction model test set. For example, 70% of the data from the first soil salt prediction model data set is selected as the first soil salt prediction model training set for training the first soil salt prediction model, and the remaining 30% of the data is retained as the first soil salt prediction model test set for testing and verifying the first soil salt prediction model.

[0086] It is understandable that the type of the target spectral feature corresponds to the type of the first spectral feature. For example, the first spectral feature includes a spectral value, a vegetation index, a salt index, a brightness index, and an intensity index, and then the first spectral image is feature extracted to obtain a target spectral feature including a spectral value, a vegetation index, a salt index, a brightness index, and an intensity index.

[0087] In this embodiment, the test score result is used to characterize the quality of the corresponding performance of the first soil salinity prediction model when predicting soil salinity through spectral features.

[0088] The test score result is obtained by weighted summing the scores of the first soil salinity prediction model on multiple evaluation indicators.

[0089] Among them, the evaluation indicators are indicators used to evaluate the accuracy, consistency and reliability of the first soil salinity prediction model in soil salinity prediction, and the score is the quantitative value of the first soil salinity prediction model's performance in the corresponding evaluation indicators.

[0090] In this embodiment, each evaluation indicator may correspond to a weight, and the test scoring result may be a result of weighted summation of the scores of each evaluation indicator.

[0091] Step 220: Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as a target soil salinity prediction model.

[0092] In this step, each first soil salinity prediction model corresponds to a test score result. According to the test score result, multiple first soil salinity prediction models can be compared, and one or more first soil salinity prediction models with the best performance are selected as the second soil salinity prediction model. The determined second soil salinity prediction model is used as the target soil salinity prediction model.

[0093] In actual implementation, the larger the value corresponding to the test score result, the better the performance of the corresponding first soil salinity prediction model. Multiple first soil salinity prediction models can be arranged from large to small according to the test score results, and the first-ranked first soil salinity prediction model, or multiple first soil salinity prediction models ranked high, can be determined as the second soil salinity prediction model. The determined second soil salinity prediction model is used as the target soil salinity prediction model.

[0094] According to the performance optimization integrated learning soil salinity prediction accuracy improvement method provided in the embodiment of the present application, multiple first soil salinity prediction models are tested, and weighted summation is performed according to the scores of the first soil salinity prediction models on multiple evaluation indicators to obtain the test score results of the first soil salinity prediction model. Based on the multiple test score results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models. Using at least one second soil salinity prediction model as the target soil salinity prediction model can overcome the current problem of reduced accuracy of the soil salinity prediction model due to subjective selection of the basic learner, and the result of soil salinity prediction by the target soil salinity prediction model is more accurate.

[0095] In some embodiments, inputting the target spectral feature into the target soil salinity prediction model to obtain the soil salinity prediction result output by the target soil salinity prediction model includes:

[0096] Inputting the target spectral feature into at least one second soil salinity prediction model to obtain a soil salinity prediction sub-result output by each second soil salinity prediction model;

[0097] Based on at least one soil salinity prediction sub-result, a soil salinity prediction result is obtained.

[0098] In this embodiment, the target spectral characteristics can be input into each second soil salt prediction model respectively. Each second soil salt prediction model predicts the salt of the soil to be predicted according to the target spectral characteristics, and outputs a soil salt prediction sub-result. The soil salt prediction sub-result can be used to characterize the salt concentration, distribution pattern and change trend of different areas of the soil to be predicted.

[0099] In actual implementation, when the target soil salinity prediction model includes multiple second soil salinity prediction models corresponding to multiple soil salinity prediction sub-results, the multiple soil salinity prediction sub-results can be weighted summed or averaged to obtain the soil salinity prediction result.

[0100] When the target soil salinity prediction model includes a second soil salinity prediction model, the soil salinity prediction sub-result output by the second soil salinity prediction model may be used as the soil salinity prediction result.

[0101] In some embodiments, before testing a plurality of first soil salinity prediction models based on a first soil salinity prediction model test set corresponding to the sample soil to obtain a test score result of each first soil salinity prediction model, the method further includes:

[0102] Calculate the first information amount of the evaluation index, and calculate the first correlation between the evaluation index and other evaluation indexes;

[0103] Based on the first amount of information and the first correlation, a weight coefficient corresponding to the evaluation index is determined.

[0104] The first amount of information is used to characterize the amount of information included in the evaluation indicator, and the first correlation is used to characterize the degree of correlation between the evaluation indicator and other evaluation indicators.

[0105] In this embodiment, the index correlation weight method (CRITIC) can be used to obtain the first correlation of the evaluation index according to the sensitivity and importance of each evaluation index, and the weight is allocated according to the first correlation. At the same time, the first information amount of the evaluation index is obtained by further combining the variability of the index through the entropy method (Entropy) to determine the weight.

[0106] In this embodiment, the minimum deviation method can be used to calculate the CRITIC-Entropy combined weighting, effectively allocating weights to each evaluation index. This method determines the optimal model parameters by minimizing the difference between the actual observation value and the model prediction value. The specific steps are as follows:

[0107] Let M = {1, 2, ..., m}, N = {1, 2, ..., n}.

[0108] Solution set A = {A 1 ,A 2,...,A m}, where A i Represents the i-th alternative, i∈M.

[0109] Attribute set C = {C 1 ,C 2 ,...,C n}, where C j Represents the j-th attribute, j∈N.

[0110] CRITIC weight set ω S ={ω 1 S ,ω 2 S ,...,ω n S}, where ω j S represents the jth CRITIC weight, and

[0111] Entropy weight set ω O ={ω 1 O ,ω 2 O ,...,ω n O}, where ω j O represents the jth Entropy weight, and

[0112] First, in order to avoid conflicts between CRITIC weights and Entropy weights when they are combined, a consistency test is used to analyze the results obtained by the above two weighting methods.

[0113]

[0114] If the distance between CRITIC weight and Entropy weight satisfies 0≤d(ω S ,ω O )≤1, the consistency check passes and combined weighting can be performed.

[0115] Secondly, perform a linear combination of the weights. Let the CRITIC weight coefficient and the Entropy weight coefficient be φ 1 and φ 2 , then the linear combination of weights is:

[0116] ω=φ 1 (ω S ) T +φ 2 (ω O )T

[0117] Then, φ is minimized with the goal of minimizing the deviation. 1 and φ 2 To optimize, there are:

[0118] η=1,2

[0119] in,

[0120] According to the differential properties of the matrix, the first-order derivative of the above formula is calculated and a linear equation system is established:

[0121]

[0122] Finally, solve the combined weights and get the normalized weight coefficient according to the following formula

[0123]

[0124] The optimal combination weight is:

[0125]

[0126] In this embodiment, multiple first soil salinity prediction models are tested based on the first soil salinity prediction model test set corresponding to the sample soil to obtain a test score result of each first soil salinity prediction model, including:

[0127] Testing the first soil salinity prediction model based on the first soil salinity prediction model test set to obtain scores of the first soil salinity prediction model on multiple evaluation indicators;

[0128] Based on the weight coefficient corresponding to each evaluation index, the scores on multiple evaluation indexes are weighted and summed to obtain the test score result of the first soil salinity prediction model.

[0129] In this embodiment, the CRITIC-Entropy Joint Weighting (CECW) method is used to optimize the best base learner, which can further improve the performance of the target soil salinity prediction model.

[0130] In this embodiment, the first soil salinity prediction model is tested by using the first soil salinity prediction model test set, and the prediction result of the first soil salinity prediction model is compared with the actual result to obtain the score of the first soil salinity prediction model on multiple evaluation indicators.

[0131] Multiply each score by the corresponding weight coefficient, and then add up all weighted scores to get the test score result.

[0132] In some embodiments, the first spectral feature is determined based on the following steps:

[0133] Based on the second soil salinity prediction model training set of the sample soil, a feature screening model is trained to obtain a feature score for each second spectral feature output by the feature screening model;

[0134] At least one third spectral feature is determined from the plurality of second spectral features based on the plurality of feature scores, and the at least one third spectral feature is used as the first spectral feature.

[0135] The input feature of the second soil salinity prediction model training set is the second spectral feature, and the target variable of the second soil salinity prediction model training set is the salinity of the sample soil.

[0136] Among them, the second soil salinity prediction model training set is a training set used to train the feature screening model. The feature screening model is a model that can select the most valuable features from the original data set. The feature score is used to characterize the value of the second spectral feature for salt prediction.

[0137] The feature screening model can select at least one third spectral feature with a high value ranking from multiple second spectral features, and use the at least one third spectral feature as the first spectral feature.

[0138] In this embodiment, soil samples can be collected, the salt content of the soil samples can be determined by the mass method, and under clear weather conditions, drone multispectral images of the area corresponding to the soil samples can be collected to ensure consistency with the soil sampling time. According to the geographic coordinates of the collected soil sample sampling points, the spectral value characteristics and spectral index characteristics of the image pixels at the corresponding positions of the soil sample sampling points are extracted as the second spectral characteristics.

[0139] In this embodiment, the second spectral feature may include a spectral value feature and a spectral index feature.

[0140] Among them, for spectral indices, in addition to selecting traditional spectral indices, replacing the original red light band with the red edge band can reveal new spectral features and develop a variety of new spectral indices.

[0141] Spectral indices can be divided into four categories: vegetation index, salinity index, brightness index and intensity index.

[0142] Among them, vegetation indices include normalized difference vegetation index (NDVI), soil adjusted vegetation index (SDVI), enhanced vegetation index (EVI), and difference vegetation index (DVI).

[0143] The salt index includes salt index (SI), normalized salt index (NDSI), normalized salt-red-edge index (NDSIREG), salt index 1 (SI1), salt-red-edge index 1 (SI1REG), salt index 2 (SI2), salt-red-edge index 2 (SI2REG), salt index 3 (SI3), salt-red-edge index 3 (SI3REG), salt index 4 (SI4), salt-red-edge index 4 (SI3REG4), salt index 5 (SI5), salt-red-edge index 5 (SI3REG5), salt index 6 (SI6), salt-red-edge index 6 (SI3REG6), salt index 7 (SI7), salt-red-edge index 7 (SI3REG7), soil index-T (SIT), red light-red edge salt index (SIR), red light-red edge salt index 1 (SIR1), red light-red edge salt index 2 (SIR2), canopy salt response index (CRSI), and combined spectral index (CORSI).

[0144] The brightness index is the brightness index (BI); the intensity index includes intensity index 1 (Int1) and intensity index 2 (Int2).

[0145] The specific calculation formula is as follows:

[0146] NDVI = (NIR-R) / (NIR+R)

[0147] SAVI=(1+L)(NIR-R) / (NIR+R+L)

[0148] EVI=2.5-(NIR-R) / (NIR+6R-7.5B+1)

[0149] DVI=NIR-R

[0150] SI=R×NIR / G

[0151] NDSI = (R-NIR) / (R+NIR)

[0152] NDSIREG=(REG-NIR) / (REG+NIR)

[0153]

[0154] SI4=B×R / G

[0155] SI4REG=B×REG / G

[0156] SI5=(BR) / (B+R)

[0157] SI5REG=(B-REG) / (B+REG)

[0158]

[0159] SI7=(G×R) / B

[0160] SI7REG=(G×REG) / B

[0161] SIT=100(R-NIR)

[0162]

[0163] Int1=(G+R) / 2

[0164] Int2=(G+R+NIR) / 2

[0165] Among them, B represents the blue band of the drone image, G represents the green band, R represents the red band, NIR represents the near-infrared band, and L represents the adjustment factor of soil surface reflectance, which can be taken as 0.5.

[0166] The second spectral feature of the soil sampling point is used as the input feature, and the corresponding soil salt content is used as the target variable to construct a second soil salt prediction model data set. The second soil salt prediction model data set is divided to obtain a second soil salt prediction model training set and a second soil salt prediction model test set.

[0167] In this embodiment, a feature screening model is trained based on a second soil salinity prediction model training set of sample soil to obtain a feature score for each second spectral feature output by the feature screening model.

[0168] In the process of training the feature screening model, the use of 5-fold cross validation and 10 repetitions can effectively prevent overfitting of the training set using the second soil salinity prediction model.

[0169] During the feature screening model training process, a feature score for each second spectral feature can be obtained. The larger the feature score value, the greater the value of the corresponding second spectral feature for the prediction result. The second spectral features can be sorted from large to small according to the feature scores, and the first-ranked second spectral feature or multiple second spectral features ranked high can be selected and determined as the third spectral feature, and the third spectral feature can be used as the first spectral feature.

[0170] In this embodiment, the prediction performance can be improved by identifying sensitive features through feature screening.

[0171] In some embodiments, the feature screening model includes at least one of a partial least squares regression model, an extreme gradient boosting model, and a random forest model.

[0172] In this embodiment, the feature screening model may include one of three machine learning algorithms: partial least squares regression model, extreme gradient boosting model and random forest model. The feature screening model may also be a hybrid embedded feature selection (HEFS) model, including two or three of the partial least squares regression model, extreme gradient boosting model and random forest model.

[0173] Among them, the Partial Least Squares Regression (PLSR) model can be used for feature screening and is suitable for small sample data. It has the advantages of being able to analyze multiple dependent variables at the same time and achieve feature dimensionality reduction.

[0174] The eXtreme Gradient Boosting (XGBoost) model can be used for feature screening, which is efficient, stable and easy to interpret.

[0175] The Random Forests (RF) model can be used for feature screening and has the advantages of strong generalization ability and reduced overfitting risk.

[0176] In this embodiment, the least squares regression model, the extreme gradient boosting model and the random forest model all have an embedded feature selection mechanism.

[0177] In this embodiment, the feature screening model is a Hybrid Ensemble Feature Selection (HEFS) model. During the training process, a feature importance score (VIS) matrix can be generated to quantify the contribution of each second spectral feature to the prediction result.

[0178] Each feature score obtained by each model is normalized by the maximum-minimum value to obtain the normalized score (VIS′), and the calculation formula is as follows:

[0179]

[0180] The score vector (|VIS|) for each feature is calculated using the following formula:

[0181]

[0182] Among them, VIS PLSR ' is the normalized feature score corresponding to the partial least squares regression model, VIS XGBoost ' is the normalized feature score corresponding to the extreme gradient boosting model, VIS RF' is the normalized feature score corresponding to the random forest model.

[0183] In this embodiment, the features may be arranged in descending order according to the score vector, and a plurality of features with a high ranking may be selected as input features of the target soil salinity prediction model.

[0184] In some embodiments, the multiple first soil salinity prediction models are of different types.

[0185] In this embodiment, each first soil salinity prediction model is of a different type, and a second soil salinity prediction model is selected from different types of first soil salinity prediction models as a target soil salinity prediction model.

[0186] In some embodiments, the type of the first soil salinity prediction model is a cube model, a quantile random forest model, a Bayesian regularized artificial neural network model, a multilayer perceptron model, a lightweight gradient boosting machine model, a multivariate adaptive regression spline model, a conditional inference tree model, an elastic net model, a Gaussian process model, a K nearest neighbor model, a support vector machine model, or a classification and regression tree model.

[0187] In this embodiment, 12 different machine learning algorithms are introduced as candidate basic learners in the performance optimized ensemble learning model (POEL), including a cube model, a quantile random forest (QRF) model, a Bayesian regularized artificial neural network (BRANN) model, a multilayer perceptron (MLP) model, a lightweight gradient boosting machine (LightGBM, LGBM) model, a multivariate adaptive regression splines (MARS) model, a conditional inference tree (CTT) model, an elastic net (ENET) model, a Gaussian process (GP) model, a K-nearest neighbors (KNN) model, a support vector machine (SVM) model, and a classification and regression tree (CLST) model. At the same time, the generalized linear model (GLM) is introduced as a meta-learner.

[0188] In this embodiment, hyperparameter optimization is performed on each base learner, i.e., the first soil salinity prediction model. Each base learner is trained for 10 iterations through 5-fold cross validation, and the best hyperparameter combination is determined through automated grid search.

[0189] In some embodiments, the evaluation index includes at least one of the coefficient of determination, root mean square error, mean absolute error, prediction-to-deviation ratio, and linear correlation consistency coefficient.

[0190] In this example, the coefficient of determination (R 2 ) can be used to measure how well the model fits the data, the root mean square error (RMSE) can be used to measure the prediction accuracy of the model, the mean absolute error (MAE) can be used to measure the difference between the predicted value and the actual measured value, the prediction-to-deviation ratio (RPD) can be used to measure the degree to which the predicted value is systematically higher or lower than the actual value, and the linear correlation coefficient of consistency (LCCC) can be used to measure the strength and direction of the linear relationship between the predictor variable and the target variable.

[0191] Coefficient of determination (R 2 ) is calculated as:

[0192]

[0193] The root mean square error (RMSE) is calculated as:

[0194]

[0195] The mean absolute error (MAE) is calculated as:

[0196]

[0197] The formula for calculating the prediction-to-deviation ratio (RPD) is:

[0198]

[0199] The calculation formula of linear correlation consistency coefficient (LCCC) is:

[0200]

[0201] Where n is the number of samples, y i is the measured value of the ith sample, is the predicted value of the ith sample, is the mean value of the measured sample, is the mean of the predicted samples.

[0202] The following is a specific implementation example of a method for improving soil salinity prediction accuracy using performance optimization ensemble learning.

[0203] like Figure 3 As shown, step 1, obtaining a first soil salinity prediction model data set. Dividing the first soil salinity prediction model data set into a first soil salinity prediction model training set and a first soil salinity prediction model test set.

[0204] Step 2: Perform hybrid embedding feature selection to identify sensitive feature variables, that is, determine the first spectral feature.

[0205] Step 3: Optimize the hyperparameters of the basic learners. The 12 basic learners, i.e., the first soil salinity prediction model, were trained 10 times through 5-fold cross validation, and the best hyperparameter combination was determined through automated grid search.

[0206] Step 4: Use five evaluation indicators to evaluate the performance of the 12 candidate base learners. The five evaluation indicators are determination coefficient (R 2), root mean square error (RMSE), mean absolute error (MAE), prediction to deviation ratio (RPD) and linear correlation coefficient of consistency (LCCC).

[0207] Step 5: Combine and weight the evaluation indicators. First, the indicator correlation weight method (CRITIC) is used to assign weights according to the sensitivity and importance of each evaluation indicator. At the same time, the entropy method (Entropy) is used to further combine the variability of the evaluation indicators to determine the weights. Finally, the minimum deviation method is used to calculate the CRITIC-Entropy combined weighting, which effectively assigns the weights to the five evaluation indicators.

[0208] Step 6: Select the best basic learner. First, multiply each evaluation index score by its corresponding weight to calculate the weighted comprehensive score, i.e., the test score result. Based on multiple test score results, the top three basic learners with the best performance are selected. The obtained learners are QRF, BRANN, and ENET, which are used as the basic learners of the performance optimization integrated learning model, i.e., QRF, BRANN, and ENET are the second soil salinity prediction model.

[0209] Step 7: POEL model construction: Using the selected sensitive features and the best basic learner, the generalized linear model (GLM) is introduced as a meta-learner to construct the POEL model for soil salinity prediction, i.e., the target soil salinity prediction model.

[0210] The performance optimization integrated learning soil salinity prediction accuracy improvement method provided in the embodiment of the present application establishes a new performance optimization integrated learning model (POEL) to combine drone images with field measured soil salinity to achieve high-precision prediction of soil salinity. The POEL model can reasonably select the optimal basic learner by adopting the CRITIC-Entropy joint weighting (CECW) method, comprehensively evaluate the performance of multiple basic learners, and dynamically assign weights according to the prediction accuracy, thereby optimizing the contribution of each basic learner to the final prediction, minimizing the number of model parameters, and improving the overall prediction ability. This dynamic weighting mechanism not only enhances the adaptability of the model, but also effectively copes with the complexity of multi-source data, and can show stronger robustness and flexibility when processing soil salinity data with high spatial heterogeneity.

[0211] In addition, the POEL model embeds a hybrid embedded feature selection (HEFS) mechanism, which can more effectively identify sensitive features. HEFS can combine multiple machine learning algorithms with feature selection techniques to accurately identify features that have a significant impact on soil salinity prediction through in-depth analysis of feature data. Through the comprehensive evaluation of feature variables, HEFS can not only improve the efficiency of feature selection, but also reduce the interference of redundant features on the model, thereby further improving the performance of the overall model, significantly improving the prediction accuracy of the model under different environmental conditions, and better adapting to various soil characteristics.

[0212] The performance optimization ensemble learning soil salinity prediction accuracy improvement method provided in the embodiment of the present application can integrate the advantages of multiple models, especially when dealing with noise and spatial variation in soil salinity, the performance optimization ensemble learning (POEL) model can provide more accurate and robust predictions. In addition, the performance weighting mechanism can further optimize the model's adaptability to soil characteristics in different regions, thereby significantly improving the performance of the overall model.

[0213] The method for improving soil salinity prediction accuracy through integrated learning based on performance optimization provided in the embodiment of the present application may be executed by a device for improving soil salinity prediction accuracy through integrated learning based on performance optimization. In the embodiment of the present application, the device for improving soil salinity prediction accuracy through integrated learning based on performance optimization is taken as an example to illustrate the device for improving soil salinity prediction accuracy through integrated learning based on performance optimization provided in the embodiment of the present application.

[0214] The embodiment of the present application also provides a device for improving soil salinity prediction accuracy through performance optimization integrated learning.

[0215] like Figure 4 As shown, the performance optimization integrated learning soil salinity prediction accuracy improvement device includes:

[0216] An acquisition module 410 is used to acquire a first spectral image of the soil to be predicted;

[0217] The first processing module 420 is used to extract features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted;

[0218] The second processing module 430 is used to input the target spectral feature into the target soil salinity prediction model to obtain the soil salinity prediction result output by the target soil salinity prediction model;

[0219] In some embodiments, the target soil salinity prediction model is obtained by the following steps:

[0220] Testing multiple first soil salt prediction models based on the first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each first soil salt prediction model. The test score result is obtained by weighted summing the scores of the first soil salt prediction model on multiple evaluation indicators;

[0221] Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as a target soil salinity prediction model;

[0222] The first soil salinity prediction model is trained by the first soil salinity prediction model training set corresponding to the sample soil. The input feature of the first soil salinity prediction model test set is the first spectral feature of the sample soil, and the target variable of the first soil salinity prediction model test set is the salinity of the sample soil. The input feature of the first soil salinity prediction model training set is the first spectral feature of the sample soil, and the target variable of the first soil salinity prediction model training set is the salinity of the sample soil. The type of the target spectral feature corresponds to the type of the first spectral feature.

[0223] According to the performance optimization integrated learning soil salinity prediction accuracy improvement device provided in the embodiment of the present application, by testing multiple first soil salinity prediction models, a weighted sum is performed based on the scores of the first soil salinity prediction model on multiple evaluation indicators to obtain the test score result of the first soil salinity prediction model. Based on the multiple test score results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models. Using at least one second soil salinity prediction model as the target soil salinity prediction model can overcome the current problem of reduced accuracy of the soil salinity prediction model due to subjective selection of the basic learner, and the result of soil salinity prediction by the target soil salinity prediction model is more accurate.

[0224] In some embodiments, the second processing module 430 is used to input the target spectral feature into at least one second soil salinity prediction model to obtain a soil salinity prediction sub-result output by each second soil salinity prediction model;

[0225] Based on at least one soil salinity prediction sub-result, a soil salinity prediction result is obtained.

[0226] In some embodiments, the second processing module 430 is further used to calculate a first amount of information of the evaluation indicator, and calculate a first correlation between the evaluation indicator and other evaluation indicators;

[0227] Determine a weight coefficient corresponding to the evaluation index based on the first information amount and the first correlation;

[0228] Testing the first soil salinity prediction model based on the first soil salinity prediction model test set to obtain scores of the first soil salinity prediction model on multiple evaluation indicators;

[0229] Based on the weight coefficient corresponding to each evaluation index, the scores on multiple evaluation indexes are weighted and summed to obtain the test score result of the first soil salinity prediction model.

[0230] In some embodiments, the first spectral feature is determined based on the following steps:

[0231] Based on the second soil salt prediction model training set of the sample soil, a feature screening model is trained to obtain a feature score for each second spectral feature output by the feature screening model. The input feature of the second soil salt prediction model training set is the second spectral feature, and the target variable of the second soil salt prediction model training set is the salt content of the sample soil;

[0232] At least one third spectral feature is determined from the plurality of second spectral features based on the plurality of feature scores, and the at least one third spectral feature is used as the first spectral feature.

[0233] In some embodiments, the feature screening model includes at least one of a partial least squares regression model, an extreme gradient boosting model, and a random forest model.

[0234] In some embodiments, the multiple first soil salinity prediction models are of different types.

[0235] In some embodiments, the type of the first soil salinity prediction model is a cube model, a quantile random forest model, a Bayesian regularized artificial neural network model, a multilayer perceptron model, a lightweight gradient boosting machine model, a multivariate adaptive regression spline model, a conditional inference tree model, an elastic net model, a Gaussian process model, a K nearest neighbor model, a support vector machine model, or a classification and regression tree model.

[0236] In some embodiments, the evaluation index includes at least one of the coefficient of determination, root mean square error, mean absolute error, prediction-to-deviation ratio, and linear correlation consistency coefficient.

[0237] The performance optimization integrated learning soil salinity prediction accuracy improvement device in the embodiment of the present application can be an electronic device, or a component in an electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or other devices other than a terminal. Exemplarily, the electronic device can be a notebook computer, an ultra-mobile personal computer (UMPC), etc., and can also be a server, a personal computer (PC), etc., which is not specifically limited in the embodiment of the present application.

[0238] The device for improving soil salinity prediction accuracy through integrated learning for performance optimization in the embodiment of the present application may be a device having an operating system. The operating system may be an Android operating system, an IOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0239] The performance optimization integrated learning soil salinity prediction accuracy improvement device provided in the embodiment of the present application can achieve Figures 1 to 3 To avoid repetition, the various processes implemented by the method embodiment are not described here.

[0240] In some embodiments, Figure 5 As shown, the embodiment of the present application also provides an electronic device 500, including a processor 501, a memory 502, and a computer program stored in the memory 502 and executable on the processor 501. When the program is executed by the processor 501, each process of the above-mentioned performance optimization integrated learning soil salinity prediction accuracy improvement method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0241] It should be noted that the electronic devices in the embodiments of the present application include the above-mentioned mobile electronic devices and non-mobile electronic devices.

[0242] The embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above-mentioned performance optimization integrated learning soil salinity prediction accuracy improvement method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0243] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0244] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned performance optimization integrated learning soil salinity prediction accuracy improvement method.

[0245] The processor is the processor in the electronic device in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0246] The embodiment of the present application further provides a chip, which includes a processor and a communication interface, the communication interface and the processor are coupled, and the processor is used to run programs or instructions to implement the various processes of the above-mentioned performance optimization integrated learning soil salinity prediction accuracy improvement method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be repeated here.

[0247] It should be understood that the chip mentioned in the embodiments of the present application can also be called a system-level chip, a system chip, a chip system or a system-on-chip chip, etc.

[0248] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0249] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of each embodiment of the present application.

[0250] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

[0251] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0252] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

Claims

1. A performance optimization ensemble learning soil salinity prediction accuracy improvement method, characterized in that: include: Acquire a first spectral image of the soil to be predicted; Extracting features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted; Inputting the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model; The target soil salinity prediction model is obtained by the following steps: Testing multiple first soil salt prediction models based on a first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salt prediction models, wherein the test score result is obtained by weighted summing the scores of the first soil salt prediction model on multiple evaluation indicators; Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as the target soil salinity prediction model; Among them, the first soil salt prediction model is trained by the first soil salt prediction model training set corresponding to the sample soil, the input feature of the first soil salt prediction model test set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model test set is the salt of the sample soil, the input feature of the first soil salt prediction model training set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model training set is the salt of the sample soil, and the type of the target spectral feature corresponds to the type of the first spectral feature.

2. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to claim 1 is characterized in that: The step of inputting the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model includes: Inputting the target spectral feature into at least one of the second soil salinity prediction models to obtain a soil salinity prediction sub-result output by each of the second soil salinity prediction models; Based on at least one of the soil salinity prediction sub-results, the soil salinity prediction result is obtained.

3. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to claim 1 is characterized in that: Before testing the plurality of first soil salt prediction models based on the first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salt prediction models, the method further includes: Calculating a first information amount of the evaluation indicator, and calculating a first correlation between the evaluation indicator and other evaluation indicators; Determining a weight coefficient corresponding to the evaluation index based on the first information amount and the first correlation; The first soil salt prediction model test set based on the sample soil corresponding to the first soil salt prediction model is tested on multiple first soil salt prediction models to obtain a test score result of each of the first soil salt prediction models, including: Testing the first soil salinity prediction model based on the first soil salinity prediction model test set to obtain scores of the first soil salinity prediction model on the plurality of evaluation indicators; Based on the weight coefficient corresponding to each of the evaluation indicators, the scores on the multiple evaluation indicators are weighted and summed to obtain the test score result of the first soil salinity prediction model.

4. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to claim 1, characterized in that: The first spectral feature is determined based on the following steps: Based on the second soil salt prediction model training set of the sample soil, a feature screening model is trained to obtain a feature score for each second spectral feature output by the feature screening model, wherein the input feature of the second soil salt prediction model training set is the second spectral feature, and the target variable of the second soil salt prediction model training set is the salt content of the sample soil; Based on the plurality of feature scores, at least one third spectral feature is determined from the plurality of second spectral features, and the at least one third spectral feature is used as the first spectral feature.

5. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to claim 4 is characterized in that: The feature screening model includes at least one of a partial least squares regression model, an extreme gradient boosting model, and a random forest model.

6. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to any one of claims 1 to 5, characterized in that: The types of the plurality of first soil salinity prediction models are different.

7. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to claim 6 is characterized in that: The type of the first soil salinity prediction model is a cube model, a quantile random forest model, a Bayesian regularized artificial neural network model, a multilayer perceptron model, a lightweight gradient boosting machine model, a multivariate adaptive regression spline model, a conditional inference tree model, an elastic net model, a Gaussian process model, a K nearest neighbor model, a support vector machine model or a classification and regression tree model.

8. The method for improving soil salinity prediction accuracy by performance optimization ensemble learning according to any one of claims 1 to 5, characterized in that: The evaluation index includes at least one of a coefficient of determination, a root mean square error, a mean absolute error, a prediction to deviation ratio, and a linear correlation consistency coefficient.

9. A performance optimization integrated learning soil salinity prediction accuracy improvement device, characterized in that: include: An acquisition module, used for acquiring a first spectral image of the soil to be predicted; A first processing module is used to extract features from the first spectral image to obtain target spectral features corresponding to the soil to be predicted; A second processing module is used to input the target spectral feature into a target soil salinity prediction model to obtain a soil salinity prediction result output by the target soil salinity prediction model; The target soil salinity prediction model is obtained by the following steps: Testing multiple first soil salt prediction models based on a first soil salt prediction model test set corresponding to the sample soil to obtain a test score result of each of the first soil salt prediction models, wherein the test score result is obtained by weighted summing the scores of the first soil salt prediction model on multiple evaluation indicators; Based on the multiple test scoring results, at least one second soil salinity prediction model is determined from the multiple first soil salinity prediction models, and the at least one second soil salinity prediction model is used as the target soil salinity prediction model; Among them, the first soil salt prediction model is trained by the first soil salt prediction model training set corresponding to the sample soil, the input feature of the first soil salt prediction model test set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model test set is the salt of the sample soil, the input feature of the first soil salt prediction model training set is the first spectral feature of the sample soil, the target variable of the first soil salt prediction model training set is the salt of the sample soil, and the type of the target spectral feature corresponds to the type of the first spectral feature.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the performance optimization integrated learning method for improving soil salinity prediction accuracy as described in any one of claims 1 to 8 is implemented.

Images