Saline-alkali soil identification method and system based on remote sensing decision

Abstract

The embodiment of the invention relates to the technical field of data analysis, in particular to a saline-alkali soil identification method and system based on remote sensing decision. And constructing an enhanced decision reconstruction vector to determine a saline-alkaline state attention discrimination vector so as to obtain a saline-alkaline state label. The method has the advantages that optical remote sensing and microwave remote sensing data are integrated, and saline-alkali soil recognition accuracy is improved in multiple aspects. The two vectors respectively represent different remote sensing detection data, the two vectors are mined to associate and construct a reinforced vector, and complementarity of the two kinds of data is fully utilized. A discrimination vector is determined through a dual-enhancement decision reconstruction vector, and the characteristics of the saline-alkali soil can be deeply mined by integrating multi-source information. Finally, the saline-alkali state label is accurately judged, misjudgment is reduced, and a reliable basis is provided for saline-alkali soil treatment, monitoring and the like.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of data analysis technology, and in particular, to a saline-alkali land identification method and system based on remote sensing decision-making. Background Art

[0002] The accurate identification of saline-alkali land is of great significance in land resource management, agricultural production, and ecological and environmental protection. Traditional saline-alkali land identification technology often relies on a single data source or a simple data processing method, and there are many problems. For example, when only optical remote sensing data is used, the accuracy of saline-alkali land identification is limited because optical sensors are easily affected by factors such as weather and vegetation cover; and when microwave remote sensing data is used alone, the data interpretation is difficult and lacks comprehensive consideration with other land features. Existing technologies make it difficult to fully and accurately obtain saline-alkali land characteristic information, resulting in inaccurate judgment of the status of saline-alkali land, which in turn affects the effective implementation of saline-alkali land governance and monitoring. Summary of the invention

[0003] In order to at least overcome the above-mentioned deficiencies in the prior art, one of the purposes of the embodiments of the present invention is to provide a method and system for identifying saline-alkali land based on remote sensing decision-making.

[0004] The embodiment of the present invention provides a saline-alkali land identification method based on remote sensing decision-making, which is applied to a saline-alkali land identification system. The method includes:

[0005] Acquire an optical remote sensing spectrum vector and a microwave remote sensing band vector of a target saline-alkali land area, wherein the optical remote sensing spectrum vector is used to characterize the optical remote sensing detection data corresponding to the target saline-alkali land area, and the microwave remote sensing band vector is used to characterize the microwave remote sensing detection data corresponding to the target saline-alkali land area;

[0006] For the optical remote sensing spectrum vector, obtaining a first enhanced decision reconstruction vector according to a local microwave remote sensing band vector associated with the optical remote sensing spectrum vector in the microwave remote sensing band vector, wherein the first enhanced decision reconstruction vector is used to represent the optical remote sensing spectrum vector integrated with the microwave remote sensing band vector;

[0007] For the microwave remote sensing band vector, obtaining a second enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector, wherein the second enhanced decision reconstruction vector is used to represent the microwave remote sensing band vector integrated with the optical remote sensing spectrum vector;

[0008] Determining a salinity state attention discrimination vector through the first reinforcement decision reconstruction vector and the second reinforcement decision reconstruction vector;

[0009] The salinity state label to which the target saline-alkali land area belongs is determined according to the salinity state attention discrimination vector.

[0010] Optionally, the step of obtaining the optical remote sensing spectrum vector and the microwave remote sensing band vector of the target saline-alkali land area includes:

[0011] Extracting optical remote sensing spectrum state elements of the optical remote sensing detection data, combining the optical remote sensing spectrum state elements with the optical remote sensing distribution vector to obtain the optical remote sensing spectrum vector;

[0012] The microwave remote sensing environment state element of the microwave remote sensing detection data is extracted, and the microwave remote sensing environment state element is combined with a microwave remote sensing distribution vector to obtain the microwave remote sensing band vector.

[0013] Optionally, the optical remote sensing detection data includes spectrum category data and spectrum band data; the optical remote sensing spectrum vector includes spectrum category vector and spectrum band vector;

[0014] The extracting of the optical remote sensing spectrum state element of the optical remote sensing detection data comprises: extracting the spectrum category spectrum state element of the spectrum category data and the spectrum band spectrum state element of the spectrum band data;

[0015] The combining of the optical remote sensing spectrum state element and the optical remote sensing distribution vector to obtain the optical remote sensing spectrum vector includes: combining the spectrum category spectrum state element and the spectrum category distribution vector to obtain the spectrum category vector; combining the spectrum band spectrum state element and the spectrum band distribution vector to obtain the spectrum band vector.

[0016] Optionally, the first enhanced decision reconstruction vector includes a spectral category enhanced decision reconstruction vector and a spectral band enhanced decision reconstruction vector, and the obtaining of the first enhanced decision reconstruction vector for the optical remote sensing spectral vector according to a local microwave remote sensing band vector associated with the optical remote sensing spectral vector in the microwave remote sensing band vector includes:

[0017] For the spectral category vector, according to at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector, obtaining the spectral category enhanced decision reconstruction vector, wherein the spectral category enhanced decision reconstruction vector is used to represent the spectral category vector after integrating the spectral band vector and the microwave remote sensing band vector;

[0018] For the spectral band vector, the spectral band enhancement decision reconstruction vector is obtained based on at least part of the vectors in the spectral category vector and the microwave remote sensing band vector that are respectively associated with the spectral band vector. The spectral band enhancement decision reconstruction vector is used to represent the spectral band vector after integrating the spectral category vector and the microwave remote sensing band vector.

[0019] Optionally, the obtaining the spectral category enhancement decision reconstruction vector according to at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector comprises:

[0020] Obtaining a first spectral category enhanced decision reconstruction vector according to at least a portion of the vectors in the spectral band vector that are associated with the spectral category vector;

[0021] Obtaining a second spectral category enhanced decision reconstruction vector according to at least a portion of vectors in the microwave remote sensing band vector that are associated with the spectral category vector;

[0022] The first spectrum category enhanced decision reconstruction vector and the second spectrum category enhanced decision reconstruction vector are collided to obtain a third spectrum category enhanced decision reconstruction vector;

[0023] Fully connect the third spectrum category enhancement decision reconstruction vector to obtain the spectrum category enhancement decision reconstruction vector.

[0024] Optionally, obtaining the spectral band enhancement decision reconstruction vector according to at least part of the vectors in the spectral category vector and the microwave remote sensing band vector respectively associated with the spectral band vector comprises:

[0025] Obtaining a first spectral band enhancement decision reconstruction vector according to at least a portion of vectors in the spectral category vector associated with the spectral band vector;

[0026] Obtaining a second spectral band enhancement decision reconstruction vector according to at least a portion of vectors in the microwave remote sensing band vector that are associated with the spectral band vector;

[0027] The first spectral band enhancement decision reconstruction vector and the second spectral band enhancement decision reconstruction vector are collided to obtain a third spectral band enhancement decision reconstruction vector;

[0028] Fully connect the third spectral band enhancement decision reconstruction vector to obtain the spectral band enhancement decision reconstruction vector.

[0029] Optionally, the obtaining of a second enhanced decision reconstruction vector for the microwave remote sensing band vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector comprises:

[0030] For the microwave remote sensing band vector, the second enhanced decision reconstruction vector is obtained according to the spectrum category vector and the local optical remote sensing spectrum vectors respectively associated with the microwave remote sensing band vector in the spectrum band vector.

[0031] Optionally, acquiring the second enhanced decision reconstruction vector according to the local optical remote sensing spectrum vectors respectively associated with the microwave remote sensing band vector in the spectrum category vector and the spectrum band vector comprises:

[0032] Obtaining a first microwave remote sensing enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the spectrum category vector;

[0033] Obtaining a second microwave remote sensing enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the spectrum band vector;

[0034] The first microwave remote sensing enhanced decision reconstruction vector and the second microwave remote sensing enhanced decision reconstruction vector are collided to obtain a third microwave remote sensing enhanced decision reconstruction vector;

[0035] The third microwave remote sensing enhanced decision reconstruction vector is fully connected to obtain the second enhanced decision reconstruction vector.

[0036] Optionally, the optical remote sensing distribution vector and the microwave remote sensing distribution vector are both relative distribution vectors between the optical remote sensing detection data and the microwave remote sensing detection data, and the relative distribution vector is used to characterize the priority and correlation between the optical remote sensing detection data and the microwave remote sensing detection data.

[0037] Optionally, the relative distribution vector is obtained by:

[0038] Determine, based on a plurality of optical remote sensings in the optical remote sensing detection data and a plurality of microwave remote sensings in the microwave remote sensing detection data, distribution information corresponding to each of the plurality of optical remote sensings and the plurality of microwave remote sensings in the target saline-alkali land area;

[0039] According to the distribution information, a global distribution feature relationship network is constructed, wherein any relationship entity in the global distribution feature relationship network is used to characterize the relative distribution information between the optical remote sensing corresponding to the column to which the entity belongs and the microwave remote sensing corresponding to the row to which the entity belongs;

[0040] A relative distribution vector between any one of the plurality of optical remote sensings and any one of the plurality of microwave remote sensings is determined based on the global distribution feature relationship network.

[0041] Optionally, determining the salinity state label to which the target saline-alkali land area belongs according to the salinity state attention discrimination vector includes:

[0042] Performing feature mapping on the salinity state attention discrimination vector to obtain a salinity state mapping vector;

[0043] Quantizing and encoding the salinity state mapping vector to obtain multiple discrimination scores of the target salinity land area belonging to multiple salinity state labels;

[0044] The salinity state label corresponding to the discrimination score that meets the detection requirements is selected as the salinity state label to which the target saline-alkali land area belongs.

[0045] An embodiment of the present invention also provides a saline-alkali land identification system, including a processor and a memory and a bus connected to the processor; wherein the processor and the memory communicate with each other through the bus; the processor is used to call program instructions in the memory to execute the above-mentioned saline-alkali land identification method based on remote sensing decision.

[0046] The embodiment of the present invention further provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the above-mentioned saline-alkali land identification method based on remote sensing decision is implemented.

[0047] Beneficial Effects

[0048] The embodiment of the present invention obtains the optical remote sensing spectrum vector and the microwave remote sensing band vector, and constructs an enhanced decision reconstruction vector to determine the saline-alkali state attention discriminant vector to obtain the saline-alkali state label. The beneficial effect is remarkable, integrating optical remote sensing and microwave remote sensing data to improve the accuracy of saline-alkali land identification in many aspects. The two vectors respectively represent different remote sensing detection data, and the correlation between the two is mined to construct an enhanced vector, making full use of the complementarity of the two data. By determining the discriminant vector through the dual enhanced decision reconstruction vector, the characteristics of saline-alkali land can be deeply mined by integrating multi-source information. Finally, the saline-alkali state label is accurately determined to reduce misjudgment, providing a reliable basis for saline-alkali land management and monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 A flowchart of a saline-alkali land identification method based on remote sensing decision-making provided in an embodiment of the present invention.

[0051] Figure 2 A block diagram of a saline-alkali land identification system provided in an embodiment of the present invention.

[0052] icon:

[0053] 100-Salt-alkali land identification system;

[0054] 101 - processor; 102 - memory; 103 - bus. DETAILED DESCRIPTION

[0055] The exemplary embodiments disclosed in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the embodiments of the present invention are shown in the accompanying drawings, it should be understood that the embodiments of the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the embodiments of the present invention and to fully convey the scope of the embodiments of the present invention to those skilled in the art.

[0056] In order to better understand the above technical scheme, the technical scheme of the embodiment of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiment of the present invention and the specific features in the embodiments are detailed descriptions of the technical scheme of the embodiment of the present invention, rather than limitations on the technical scheme of the embodiment of the present invention. In the absence of conflict, the embodiment of the present invention and the technical features in the embodiments can be combined with each other.

[0057] Figure 1 This is a flow chart of a saline-alkali land identification method based on remote sensing decision-making provided according to an embodiment of the present invention, which is applied to a saline-alkali land identification system and includes steps 110 to 140.

[0058] Step 110, the saline-alkali land identification system obtains an optical remote sensing spectrum vector and a microwave remote sensing band vector of the target saline-alkali land area, wherein the optical remote sensing spectrum vector is used to characterize the optical remote sensing detection data corresponding to the target saline-alkali land area, and the microwave remote sensing band vector is used to characterize the microwave remote sensing detection data corresponding to the target saline-alkali land area.

[0059] Step 120: The saline-alkali land identification system obtains a first enhanced decision reconstruction vector for the optical remote sensing spectral vector based on a local microwave remote sensing band vector associated with the optical remote sensing spectral vector in the microwave remote sensing band vector. The first enhanced decision reconstruction vector is used to characterize the optical remote sensing spectral vector that integrates the microwave remote sensing band vector.

[0060] Step 130: The saline-alkali land identification system obtains a second enhanced decision reconstruction vector for the microwave remote sensing band vector based on a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector, wherein the second enhanced decision reconstruction vector is used to characterize the microwave remote sensing band vector that integrates the optical remote sensing spectrum vector.

[0061] Step 140: The saline-alkali land identification system determines a saline-alkali state attention discrimination vector through the first enhanced decision reconstruction vector and the second enhanced decision reconstruction vector; and determines a saline-alkali state label to which the target saline-alkali land area belongs based on the saline-alkali state attention discrimination vector.

[0062] In step 110, the optical remote sensing detection data contains information of multiple bands, which cover different spectral ranges from visible light to infrared. For example, the common Landsat satellite optical remote sensing data contains multiple bands, such as blue band (central wavelength is about 0.48 μm), green band (central wavelength is about 0.56 μm), red band (central wavelength is about 0.66 μm), near infrared band (central wavelength is about 0.86 μm), etc. For the target saline-alkali land area, the detection data of these different bands are combined in a certain order to form an optical remote sensing spectral vector. In a simplified case, only these four bands are considered, and the optical remote sensing spectrum vector can be expressed as (X = [x_1, x_2, x_3, x_4]), where (x_1) represents the detection data of the blue band (for example, the reflectance value is 0.2), (x_2) represents the detection data of the green band (the reflectance value is 0.3), (x_3) represents the detection data of the red band (the reflectance value is 0.4), and (x_4) represents the detection data of the near-infrared band (the reflectance value is 0.5). These values ​​are obtained based on the detection of the target saline-alkali land area by satellite sensors. Different saline-alkali land types, soil moisture, vegetation coverage and other factors will cause different reflectance values ​​of these bands.

[0063] Microwave remote sensing bands also have their own unique detection capabilities. For example, a certain microwave remote sensing sensor has several specific bands, which have different responses to the dielectric properties and roughness of the soil. Set the microwave remote sensing band vector to (Y = [y_1, y_2, y_3]), where (y_1) is the backscattering coefficient in a certain low-frequency microwave band (e.g. -10dB), (y_2) is the backscattering coefficient in the medium-frequency microwave band (e.g. -5dB), and (y_3) is the backscattering coefficient in the high-frequency microwave band (e.g. 0dB). These values ​​reflect the characteristics of the soil in the target saline-alkali land area in the microwave band. Due to the influence of factors such as the salt content and humidity of the saline-alkali land soil, different saline-alkali lands will show different backscattering characteristics in the microwave band, resulting in numerical differences in these band vectors.

[0064] In step 120, it is first necessary to determine the local microwave remote sensing band vector associated with the optical remote sensing spectrum vector in the microwave remote sensing band vector. This association may be based on prior knowledge or obtained through data analysis. For example, through a large number of experiments and data analysis, it is found that in the target saline-alkali land area, there is a strong correlation between (y_2) (backscattering coefficient of the medium-frequency microwave band) in the microwave remote sensing band vector and the red band (x_3) (reflectivity value of 0.4) in the optical remote sensing spectrum vector. This may be due to certain physical and chemical properties of the soil (such as the effect of salt on soil structure, which in turn affects the reflectivity of the red band and the backscattering of the medium-frequency microwave). It is then found that (y_3) (backscattering coefficient of the high-frequency microwave band) is also associated with the near-infrared band (x_4) (reflectivity value of 0.5). Then the local microwave remote sensing band vector can be determined as (Y_{local} = [y_2, y_3]).

[0065] Next, a certain method is used to integrate the local microwave remote sensing band vector into the optical remote sensing spectrum vector to obtain the first enhanced decision reconstruction vector. One possible method is to use weighted fusion. Set the weight assigned to (y_2) to (w_1=0.3) and the weight assigned to (y_3) to (w_2=0.7). For each element in the optical remote sensing spectrum vector, fusion is performed through a certain functional relationship. For example, for the red band (x_3), the new value (x_{3}^{new}=x_3+w_1*y_2), and for the near-infrared band (x_4), the new value (x_{4}^{new}=x_4+w_2*y_3). The first enhanced decision reconstruction vector (X_{recon1}=[x_1, x_2, x_{3}^{new}, x_{4}^{new}]) obtained in this way represents the optical remote sensing spectrum vector that integrates the microwave remote sensing band vector information. This integration can better utilize the complementarity of microwave remote sensing and optical remote sensing data and improve the ability to characterize saline-alkali land characteristics.

[0066] Similar to step 120, step 130 is to determine the local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector. It is assumed that through analysis, it is found that the blue band (x_1) (reflectivity value is 0.2) in the optical remote sensing spectrum vector is associated with (y_1) in the microwave remote sensing band vector (the backscattering coefficient of the low-frequency microwave band is -10dB), and the green band (x_2) (reflectivity value is 0.3) is associated with (y_2) (the backscattering coefficient of the medium-frequency microwave band is -5dB). Then the local optical remote sensing spectrum vector is determined as (X_{local}=[x_1,x_2]).

[0067] A similar weighted fusion method is used to obtain the second enhanced decision reconstruction vector. Set the weight (w_3=0.4) to (x_1) and the weight (w_4=0.6) to (x_2). For each element in the microwave remote sensing band vector, fusion is performed. For example, for (y_1), the new value (y_{1}^{new}=y_1+w_3*x_1), for (y_2), the new value (y_{2}^{new}=y_2+w_4*x_2). The obtained second enhanced decision reconstruction vector (Y_{recon2}=[y_{1}^{new}, y_{2}^{new}, y_3]) represents the microwave remote sensing band vector that integrates the optical remote sensing spectral vector information.

[0068] Finally, the salinity state attention discriminant vector is determined by the first reinforcement decision reconstruction vector (X_{recon1}) and the second reinforcement decision reconstruction vector (Y_{recon2}). A vector concatenation method can be used, for example, the salinity state attention discriminant vector (Z = [X_{recon1}, Y_{recon2}]), that is, the two reconstruction vectors are combined in order to form a new vector. Other more complex fusion methods can also be used, such as fusion based on neural networks. For example, an exemplary multi-layer perceptron (MLP) model is constructed, the number of input layer nodes is the sum of the number of elements of (X_{recon1}) and (Y_{recon2}), the number of hidden layer nodes can be set to 10, and the number of output layer nodes is the predetermined number of features representing salinity state attention, which is set to 5. The elements of (X_{recon1}) and (Y_{recon2}) are used as input, and the salinity state attention discriminant vector (Z) is obtained after forward propagation calculation of the MLP model.

[0069] The salinity state label of the target salinity land area is determined according to the salinity state attention discriminant vector (Z). A classification algorithm, such as a support vector machine (SVM) classifier, can be used. First, a training set is required, which contains the salinity state attention discriminant vectors corresponding to samples of known salinity state. The training set is set to have 100 samples, including 30 samples of mild salinity land, 40 samples of moderate salinity land, and 30 samples of severe salinity land. The salinity state attention discriminant vector corresponding to each sample is used as the input feature. The training set is trained using the SVM algorithm to find the optimal classification hyperplane. Then the salinity state attention discriminant vector (Z) corresponding to the target salinity land area is input into the trained SVM classifier. The classifier will determine the category to which it belongs according to the position of (Z) in the feature space, that is, the salinity state label of the target salinity land area, for example, the output is different labels such as mild salinity land, moderate salinity land, or severe salinity land. Other classification algorithms can also be used, such as a decision tree classification algorithm. When building a decision tree, the elements in the salinity state attention discriminant vector (Z) are used as features, and decision rules are constructed based on the sample data of the training set. For example, if the value of an element in (Z) is greater than a certain threshold, it is judged as moderate salinity land, etc., thereby determining the salinity state label of the target salinity land area.

[0070] In an optional embodiment, the method of obtaining the optical remote sensing spectrum vector and microwave remote sensing band vector of the target saline-alkali land area includes: extracting the optical remote sensing spectrum state element of the optical remote sensing detection data, combining the optical remote sensing spectrum state element with the optical remote sensing distribution vector to obtain the optical remote sensing spectrum vector; extracting the microwave remote sensing environment state element of the microwave remote sensing detection data, combining the microwave remote sensing environment state element with the microwave remote sensing distribution vector to obtain the microwave remote sensing band vector.

[0071] In this embodiment, the optical remote sensing detection data contains rich information. For the target saline-alkali land area, the optical remote sensing spectral state elements must first be extracted. The optical remote sensing spectral state elements are key data that can reflect the unique state of the target saline-alkali land area under optical remote sensing. For example, in the optical remote sensing data obtained from the Landsat satellite, for each pixel, the reflectivity values ​​in the visible light band (400-700nm) and the near-infrared band (700-1300nm) are of great significance. It is set to subdivide the visible light band at intervals of 100nm, and obtain the reflectivity values ​​of the three sub-bands, which are (r_1=0.2) (400-500nm band), (r_2=0.3) (500-600nm band) and (r_3=0.4) (600-700nm band); take an average reflectivity value (r_4=0.5) in the near-infrared band (700-1300nm). These reflectance values ​​are part of the optical remote sensing spectral state elements. In addition, some values ​​based on band calculations can also be included, such as the Normalized Difference Vegetation Index (NDVI), which is calculated as (NDVI = (NIR-R) / (NIR + R)), where (NIR) is the near-infrared band reflectance and (R) is the red band reflectance. Set (NDVI = 0.3), which is also part of the optical remote sensing spectral state elements.

[0072] The optical remote sensing distribution vector is a vector used to describe the distribution characteristics of optical remote sensing data in space or other dimensions. For example, the statistical characteristics of the pixel distribution of optical remote sensing data in the target saline-alkali land area can be considered. Taking a small area of ​​(5*5) pixels as an example, the standard deviation (sigma_1=0.05) (for a certain band), skewness (s_1=0.1) and kurtosis (k_1=3.2) of the optical remote sensing data of these pixels are calculated. These statistics constitute the optical remote sensing distribution vector (D_1=[sigma_1, s_1, k_1]).

[0073] The extracted optical remote sensing spectrum state element (E_1 = [r_1, r_2, r_3, r_4, NDVI]) is combined with the optical remote sensing distribution vector (D_1). A common combination method is to directly splice to obtain the optical remote sensing spectrum vector (X = [r_1, r_2, r_3, r_4, NDVI, sigma_1, s_1, k_1]). This combination method can comprehensively describe the characteristics of the target saline-alkali land area under optical remote sensing, including both the state information of the spectrum itself and its distribution characteristics in a certain area.

[0074] For microwave remote sensing detection data, the extraction of microwave remote sensing environmental state elements is also critical. Microwave remote sensing is sensitive to the dielectric properties and roughness of soil. For example, the estimated value of the soil dielectric constant (epsilon = 5) (relative dielectric constant), the quantified value of soil roughness (h = 0.1) (roughness height is set in centimeters), and the penetration depth of microwaves in the soil (d = 10) (cm) and other values ​​in the target saline-alkali land area are extracted from microwave remote sensing data. These values ​​reflect the microwave remote sensing environmental state elements in the target saline-alkali land area. In addition, it can also include the values ​​of the microwave backscattering coefficient under different polarization modes, such as the backscattering coefficient under horizontal polarization (beta_h = -8) (dB) and the backscattering coefficient under vertical polarization (beta_v = -6) (dB).

[0075] Similar to optical remote sensing, microwave remote sensing distribution vectors describe the distribution characteristics of microwave remote sensing data. For example, consider the changes in the backscatter coefficient of microwave remote sensing data at different locations in the target saline-alkali land area. Taking a (3*3) area as an example, the standard deviation (sigma_2=1) (dB) and coefficient of variation (cv=0.2) of the microwave backscatter coefficient are calculated to form a microwave remote sensing distribution vector (D_2=[sigma_2, cv]).

[0076] Combine the microwave remote sensing environmental state element (E_2 = [epsilon, h, d, beta_h, beta_v]) with the microwave remote sensing distribution vector (D_2). For example, the microwave remote sensing band vector (Y = [epsilon, h, d, beta_h, beta_v, sigma_2, cv]) is obtained by vector splicing. This combination method can fully reflect the characteristics of the target saline-alkali land area under microwave remote sensing, including the distribution characteristics of the environment-related state elements and data.

[0077] With this design, first, in the construction of optical remote sensing spectral vectors, the spectral state elements are combined with the distribution vectors, which can describe the optical characteristics of the target saline-alkali land area from multiple angles. For example, the reflectance value in the spectral state element reflects the basic optical properties of soil and vegetation, while the statistics in the distribution vector reflect the changes in these characteristics in the region, making the description of the target area more detailed and comprehensive, avoiding the limitations of relying solely on a single reflectance value or a single statistic description. The same is true for microwave remote sensing band vectors. The information such as soil dielectric constant and roughness in the environmental state element is directly related to the detection of soil characteristics by microwave remote sensing, while the different polarization values ​​of the backscattering coefficient further enrich the microwave characteristic description of the soil, and the statistics in the distribution vector reflect the distribution differences of these characteristics in the region. This comprehensive vector construction method provides richer and more accurate basic data for the subsequent saline-alkali land identification steps, which helps to improve the accuracy of saline-alkali land identification and reduce the possibility of misjudgment. It has important application value in saline-alkali land monitoring, governance and other related fields.

[0078] In a replaceable embodiment, the optical remote sensing detection data includes spectral category data and spectral band data; the optical remote sensing spectrum vector includes a spectral category vector and a spectral band vector; the extraction of optical remote sensing spectral state elements of the optical remote sensing detection data includes: extracting the spectral category spectral state elements of the spectral category data and the spectral band spectral state elements of the spectral band data; the combination of the optical remote sensing spectral state elements with the optical remote sensing distribution vector to obtain the optical remote sensing spectral vector includes: combining the spectral category spectral state elements with the spectral category distribution vector to obtain the spectral category vector; combining the spectral band spectral state elements with the spectral band distribution vector to obtain the spectral band vector.

[0079] Further, the first enhanced decision reconstruction vector includes a spectral category enhanced decision reconstruction vector and a spectral band enhanced decision reconstruction vector, and the obtaining of the first enhanced decision reconstruction vector for the optical remote sensing spectral vector according to the local microwave remote sensing band vector associated with the optical remote sensing spectral vector in the microwave remote sensing band vector includes: for the spectral category vector, according to at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector, obtaining the spectral category enhanced decision reconstruction vector, the spectral category enhanced decision reconstruction vector is used to represent the spectral category vector after the spectral band vector and the microwave remote sensing band vector are integrated; for the spectral band vector, according to at least part of the vectors in the spectral category vector and the microwave remote sensing band vector respectively associated with the spectral band vector, obtaining the spectral band enhanced decision reconstruction vector, the spectral band enhanced decision reconstruction vector is used to represent the spectral band vector after the spectral category vector and the microwave remote sensing band vector are integrated.

[0080] On the one hand, the method of obtaining the spectral category enhancement decision reconstruction vector based on at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector includes: obtaining a first spectral category enhancement decision reconstruction vector based on at least part of the vectors in the spectral band vector associated with the spectral category vector; obtaining a second spectral category enhancement decision reconstruction vector based on at least part of the vectors in the microwave remote sensing band vector associated with the spectral category vector; colliding the first spectral category enhancement decision reconstruction vector and the second spectral category enhancement decision reconstruction vector to obtain a third spectral category enhancement decision reconstruction vector; and performing full connection processing on the third spectral category enhancement decision reconstruction vector to obtain the spectral category enhancement decision reconstruction vector.

[0081] On the other hand, obtaining the spectral band enhancement decision reconstruction vector based on at least part of the vectors in the spectral category vector and the microwave remote sensing band vector respectively associated with the spectral band vector includes: obtaining a first spectral band enhancement decision reconstruction vector based on at least part of the vectors in the spectral category vector associated with the spectral band vector; obtaining a second spectral band enhancement decision reconstruction vector based on at least part of the vectors in the microwave remote sensing band vector associated with the spectral band vector; colliding the first spectral band enhancement decision reconstruction vector and the second spectral band enhancement decision reconstruction vector to obtain a third spectral band enhancement decision reconstruction vector; and performing full connection processing on the third spectral band enhancement decision reconstruction vector to obtain the spectral band enhancement decision reconstruction vector.

[0082] It can be understood that the optical remote sensing detection data includes spectral category data and spectral band data. Spectral category data can be data divided according to a certain classification attribute of optical remote sensing data. For example, the category data can be obtained by classifying optical remote sensing data according to the type of land object (such as vegetation, soil, water body, etc.). Spectral band data directly corresponds to the numerical data of different bands obtained by the optical remote sensing sensor, such as the reflectivity values ​​of the blue band (central wavelength of about 0.48μm), green band (central wavelength of about 0.56μm), red band (central wavelength of about 0.66μm), near infrared band (central wavelength of about 0.86μm) and other bands of the Landsat satellite mentioned above.

[0083] For spectral category data, spectral category spectral state elements are extracted. For example, for the spectral category of vegetation, vegetation coverage (C = 0.6) (indicating the proportion of vegetation coverage), vegetation health index (H = 0.8) (a numerical value reflecting the health of vegetation calculated by a certain algorithm), etc. may be extracted as spectral category spectral state elements. Then, a spectral category distribution vector is constructed, such as considering the distribution density of vegetation categories in the target saline-alkali land area (rho = 0.05) (the proportion of the number of pixels of vegetation categories per unit area), the quantitative value of the adjacent relationship between vegetation categories and other categories (A = 0.3) (a numerical value indicating the adjacent relationship calculated by a certain algorithm), etc., to form a spectral category distribution vector (D_{category} = [rho, A]). The spectral category spectral state elements are combined with the spectral category distribution vector, for example, by using vector splicing to obtain a spectral category vector (V_{category} = [C, H, rho, A]).

[0084] For spectral band data, extract the spectral band spectral state elements. Taking the red band as an example, its reflectance (r = 0.4) and the ratio of the red band reflectance to the green band reflectance (R_{ratio} = 1.3) (the green band reflectance is set to 0.3) can be used as spectral band spectral state elements. Construct a spectral band distribution vector, such as the standard deviation of the red band reflectance in the target area (sigma_{r} = 0.03), and calculate the spectral band distribution vector (D_{band} = [sigma_{r}]). Combine the spectral band spectral state elements with the spectral band distribution vector to obtain the spectral band vector (V_{band} = [r, R_{ratio}, sigma_{r}]).

[0085] The first spectral category enhancement decision reconstruction vector is obtained based on at least part of the vectors in the spectral band vector that are associated with the spectral category vector. For example, if it is found that the red band reflectance (r=0.4) in the spectral band vector is associated with the vegetation coverage (C=0.6) in the spectral category vector (which may be based on prior knowledge or data analysis). The first spectral category enhancement decision reconstruction vector can be constructed by weighted summation. Assume that the red band reflectance (r) is assigned a weight (w_1=0.5), then the new vegetation coverage (C_{new}=C+w_1*r=0.6+0.5*0.4=0.8), and the other elements in the spectral category vector are processed in this way to obtain the first spectral category enhancement decision reconstruction vector (V_{category1}).

[0086] The second spectral category enhanced decision reconstruction vector is obtained based on at least part of the vectors associated with the spectral category vector in the microwave remote sensing band vector. The backscatter coefficient (beta = -5) (dB) of a certain microwave band in the microwave remote sensing band vector is set to be associated with the vegetation health index (H = 0.8) in the spectral category vector. In a similar weighted summation method, a weight (w_2 = 0.3) is assigned to (beta), and the new vegetation health index (H_{new} = H + w_2*beta = 0.8 + 0.3*(-5) = 0.8-1.5 = -0.7) is obtained (this is just an example calculation, and the actual operation may require numerical range adjustment). In this way, other related elements in the spectral category vector are processed to obtain the second spectral category enhanced decision reconstruction vector (V_{category2}).

[0087] The first spectral category reinforcement decision reconstruction vector (V_{category1}) and the second spectral category reinforcement decision reconstruction vector (V_{category2}) are collided. The collision operation can be a custom vector element combination operation, such as multiplying the corresponding elements of two vectors to obtain a new vector (V_{category3}) (this is just a possible collision operation method). Then (V_{category3}) is fully connected, and the fully connected processing can adopt the concept of a fully connected layer in a neural network. Set to build an exemplary fully connected layer, the number of input nodes is the number of elements of (V_{category3}), and the number of output nodes is the number of elements of the predetermined spectral category reinforcement decision reconstruction vector, such as 3. The spectral category reinforcement decision reconstruction vector (V_{category-recon}) is obtained by performing a matrix multiplication operation on the weight matrix of the fully connected layer and (V_{category3}), and adding a bias vector.

[0088] The first spectral band enhancement decision reconstruction vector is obtained based on at least part of the vectors associated with the spectral band vector in the spectral category vector. For example, if the vegetation coverage (C = 0.6) in the spectral category vector is associated with the red band reflectance (r = 0.4) in the spectral band vector. By using the weighted summation method, a weight (w_3 = 0.4) is assigned to (C), then the new red band reflectance (r_{new} = r + w_3*C = 0.4 + 0.4*0.6 = 0.4 + 0.24 = 0.64) is obtained. In this way, other elements in the spectral band vector are processed to obtain the first spectral band enhancement decision reconstruction vector (V_{band1}).

[0089] The second spectral band enhancement decision reconstruction vector is obtained based on at least part of the vectors associated with the spectral band vector in the microwave remote sensing band vector. The backscatter coefficient (beta'=-3) (dB) of a certain microwave band in the microwave remote sensing band vector is set to be associated with the red band reflectivity (r=0.4) in the spectral band vector. Assign a weight (w_4=0.2) to (beta'), then the new red band reflectivity (r_{new}'=r+w_4*beta'=0.4+0.2*(-3)=0.4-0.6=-0.2) (may also need numerical adjustment), and other related elements in the spectral band vector are processed in this way to obtain the second spectral band enhancement decision reconstruction vector (V_{band2}).

[0090] The first spectral band reinforcement decision reconstruction vector (V_{band1}) and the second spectral band reinforcement decision reconstruction vector (V_{band2}) are collided, for example, by vector addition to obtain a third spectral band reinforcement decision reconstruction vector (V_{band3}). Then (V_{band3}) is fully connected to construct a fully connected layer, the number of input nodes is the number of elements of (V_{band3}), and the number of output nodes is the number of elements of the predetermined spectral band reinforcement decision reconstruction vector, such as 4. The spectral band reinforcement decision reconstruction vector (V_{band-recon}) is obtained by calculation of the fully connected layer.

[0091] In this way, first, the optical remote sensing detection data is subdivided into spectral category data and spectral band data, and the corresponding vectors are constructed respectively. This subdivision method makes the description of optical remote sensing data more detailed and accurate. For example, the spectral category vector can describe the target saline-alkali land area from the perspective of attributes related to the type of land object, while the spectral band vector describes it from the perspective of band values ​​and their distribution characteristics. In the process of obtaining the first enhanced decision reconstruction vector, the spectral category enhanced decision reconstruction vector and the spectral band enhanced decision reconstruction vector are obtained through multi-step association and processing. This multi-step processing can fully integrate the association information between different types of data, such as the association information between spectral categories and spectral bands, between spectral categories and microwave remote sensing bands, and between spectral bands and microwave remote sensing bands. For example, through association processing, the intrinsic relationship between vegetation coverage and spectral band reflectivity and microwave backscattering coefficient can be better reflected, thereby improving the feature extraction and characterization capabilities of the target saline-alkali land area, which is helpful for more accurate determination of saline-alkali state labels in the future, reducing recognition errors, and improving the overall performance of the saline-alkali land recognition system.

[0092] In an optional embodiment, obtaining a second enhanced decision reconstruction vector for the microwave remote sensing band vector based on a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector includes: obtaining the second enhanced decision reconstruction vector for the microwave remote sensing band vector based on the spectrum category vector and the local optical remote sensing spectrum vectors associated with the microwave remote sensing band vector in the spectrum band vector.

[0093] In detail, the obtaining of the second enhanced decision reconstruction vector based on the local optical remote sensing spectral vectors associated with the microwave remote sensing band vector in the spectral category vector and the spectral band vector includes: obtaining a first microwave remote sensing enhanced decision reconstruction vector based on the local optical remote sensing spectral vector associated with the microwave remote sensing band vector in the spectral category vector; obtaining a second microwave remote sensing enhanced decision reconstruction vector based on the local optical remote sensing spectral vector associated with the microwave remote sensing band vector in the spectral band vector; colliding the first microwave remote sensing enhanced decision reconstruction vector and the second microwave remote sensing enhanced decision reconstruction vector to obtain a third microwave remote sensing enhanced decision reconstruction vector; and performing full connection processing on the third microwave remote sensing enhanced decision reconstruction vector to obtain the second enhanced decision reconstruction vector.

[0094] In this embodiment, obtaining the second enhanced decision reconstruction vector is a process based on multiple association and processing steps. This process is mainly performed based on the spectrum category vector in the optical remote sensing spectrum vector and the local optical remote sensing spectrum vector in the spectrum band vector respectively associated with the microwave remote sensing band vector. The second enhanced decision reconstruction vector is finally obtained by respectively obtaining the enhanced decision reconstruction vectors of different parts associated with the microwave remote sensing band vector, and then undergoing collision and full connection processing.

[0095] First, the local optical remote sensing spectral vector associated with the microwave remote sensing band vector in the spectral category vector must be determined. The determination of this association is based on an in-depth understanding of the characteristics of optical remote sensing and microwave remote sensing data in the target saline-alkali land area. For example, in the spectral category vector (V_{category}=[C,H,rho,A]) (where (C) is the vegetation coverage, (H) is the vegetation health index, (rho) is the vegetation category distribution density, and (A) is the quantitative value of the adjacent relationship between the vegetation category and other categories), it may be found that the vegetation coverage (C) is associated with the backscattering coefficient (beta) of a certain microwave band in the microwave remote sensing band vector (set (beta=-5) dB). This association may be due to the fact that vegetation coverage affects the dielectric properties of the soil, thereby affecting the backscattering of microwaves.

[0096] After determining the association, a specific algorithm is used to construct the first microwave remote sensing enhanced decision reconstruction vector. One possible algorithm is based on proportional adjustment. For example, an adjustment coefficient (k_1 = 0.3) is obtained based on experience or data analysis, then the backscatter coefficient of the new microwave band (beta_{new} = beta + k_1*C).

[0097] Assuming (C=0.6), then (beta_{new}=-5+0.3*0.6=-5+0.18=-4.82) dB. In this way, other elements in the microwave remote sensing band vector associated with the spectral category vector are processed to obtain the first microwave remote sensing enhanced decision reconstruction vector (V_{microwave1}).

[0098] The determination of the local optical remote sensing spectral vector associated with the microwave remote sensing band vector in the spectral band vector is also based on the analysis of data characteristics. In the spectral band vector (V_{band}=[r,R_{ratio},sigma_{r}]) (where (r) is the red band reflectivity, (R_{ratio}) is the ratio of the red band reflectivity to the green band reflectivity, and (sigma_{r}) is the standard deviation of the red band reflectivity in the target area), the red band reflectivity (r=0.4) is set to be associated with the backscattering coefficient (beta') of another microwave band in the microwave remote sensing band vector (set (beta'=-3) dB). This association may be because the reflectivity of the soil in the red band is related to certain physical properties of the soil, and these physical properties will affect the scattering characteristics of microwaves.

[0099] Using a method similar to that used to construct the first microwave remote sensing enhanced decision reconstruction vector, for example, using another adjustment coefficient (k_2 = 0.2), the backscatter coefficient of the new microwave band is (beta_{new}' = beta' + k_2*r). Substituting (r = 0.4) and (beta' = -3) dB into (beta_{new}' = -3 + 0.2*0.4 = -3 + 0.08 = -2.92) dB. The other elements in the microwave remote sensing band vector associated with the spectral band vector are processed in a similar manner to obtain the second microwave remote sensing enhanced decision reconstruction vector (V_{microwave2}).

[0100] The first microwave remote sensing enhanced decision reconstruction vector (V_{microwave1}) and the second microwave remote sensing enhanced decision reconstruction vector (V_{microwave2}) are subjected to a collision operation. The collision operation here can be performed in the form of vector multiplication. For example, if (V_{microwave1} = [beta_{new}, cdots]) (where (beta_{new} = -4.82) dB is the element calculated previously), (V_{microwave2} = [beta_{new}', cdots]) (where (beta_{new}' = -2.92) dB is the element calculated previously), then the element (v_{i}) of the vector (V_{microwave3}) after the collision (set to multiply the elements at the corresponding positions) is (v_{i} = beta_{new}*beta_{new}') (here only the first element is taken as an example), etc., thereby obtaining the third microwave remote sensing enhanced decision reconstruction vector (V_{microwave3}).

[0101] The third microwave remote sensing enhanced decision reconstruction vector (V_{microwave3}) is fully connected. The fully connected process is similar to the fully connected layer operation in a neural network. For example, a fully connected layer is constructed, the number of input nodes is the number of elements (n) of (V_{microwave3}) (set (n=5)), and the number of output nodes is the number of elements (m) of the predetermined second enhanced decision reconstruction vector (set (m=3)). The fully connected layer has a weight matrix (W) (a (m*n) matrix) and a bias vector (b) (an (m)-dimensional vector). The second enhanced decision reconstruction vector (V_{recon2}) is obtained by calculating (y=W*V_{microwave3}+b) (where (y) is the output vector).

[0102] Therefore, by finding the association with the microwave remote sensing band vector from the spectral category vector and the spectral band vector respectively to construct the enhanced decision reconstruction vector, the intrinsic connection between different types of optical remote sensing data and microwave remote sensing data can be fully explored. For example, through the construction process of the association between vegetation coverage and microwave backscattering coefficient and the association between red band reflectivity and microwave backscattering coefficient, the information of multi-source remote sensing data can be more comprehensively integrated. The collision operation can fuse the enhanced information from different sources, and the full connection processing further extracts and transforms the features of the fused information, thereby improving the feature representation ability of the target saline-alkali land area. This helps to more accurately analyze the saline-alkali land conditions in the target area, reduce the recognition errors caused by a single data source or a simple fusion method, and improve the accuracy and reliability of the entire saline-alkali land identification system.

[0103] In other examples, the optical remote sensing distribution vector and the microwave remote sensing distribution vector are both relative distribution vectors between the optical remote sensing detection data and the microwave remote sensing detection data, and the relative distribution vector is used to characterize the priority and correlation between the optical remote sensing detection data and the microwave remote sensing detection data.

[0104] Based on this, the method for obtaining the relative distribution vector includes: determining the distribution information corresponding to each of the multiple optical remote sensings and the multiple microwave remote sensings in the target saline-alkali land area based on the multiple optical remote sensings in the optical remote sensing detection data and the multiple microwave remote sensings in the microwave remote sensing detection data; constructing a global distribution feature relationship network based on the distribution information, and any relationship entity in the global distribution feature relationship network is used to characterize the relative distribution information between the optical remote sensing corresponding to the column to which the entity belongs and the microwave remote sensing corresponding to the row to which the entity belongs; determining the relative distribution vector between any optical remote sensing among the multiple optical remote sensings and any microwave remote sensing among the multiple microwave remote sensings based on the global distribution feature relationship network.

[0105] In this embodiment, the optical remote sensing distribution vector and the microwave remote sensing distribution vector are defined as the relative distribution vector between the optical remote sensing detection data and the microwave remote sensing detection data. The role of this relative distribution vector is to characterize the priority and correlation between the optical remote sensing detection data and the microwave remote sensing detection data. The introduction of this concept aims to more accurately integrate optical remote sensing and microwave remote sensing data to better identify the characteristics of the target saline-alkali land area.

[0106] For multiple optical remote sensing in the optical remote sensing detection data, it is necessary to determine their corresponding distribution information in the target saline-alkali land area. For example, consider an optical remote sensing data with multiple bands (such as blue band, green band, red band, near infrared band, etc.). Taking the blue band as an example, calculate the mean (mu_{b}=0.2) of the blue band reflectance in the target saline-alkali land area (here the value range of the blue band reflectance is set to (0-1)), standard deviation (sigma_{b}=0.05), skewness (s_{b}=0.1), kurtosis (k_{b}=3.2) and other statistics. For other bands, such as the mean (mu_{g}=0.3) and standard deviation (sigma_{g}=0.03) of the green band reflectance, these statistics constitute the distribution information of each optical remote sensing in the target area.

[0107] Similarly, for multiple microwave remote sensing in the microwave remote sensing detection data, determine its distribution information in the target saline-alkali land area. Assume that microwave remote sensing has three bands with different frequencies. For the low-frequency microwave band, calculate the mean (mu_{l}=-10) dB, standard deviation (sigma_{l}=1) dB, coefficient of variation (cv_{l}=0.1) and other statistics of its backscatter coefficient in the target area. Similar statistics are also calculated for the medium-frequency and high-frequency microwave bands, such as the mean (mu_{m}=-5) dB, standard deviation (sigma_{m}=0.5) dB, etc. of the backscatter coefficient of the medium-frequency microwave band. These statistics constitute the distribution information of microwave remote sensing in the target area.

[0108] Based on the above distribution information, a global distribution feature relationship network is constructed. This relationship network is a matrix structure, in which any relationship entity is used to characterize the relative distribution information between the optical remote sensing corresponding to the column to which the entity belongs and the microwave remote sensing corresponding to the row to which the entity belongs. For example, assuming that optical remote sensing has 4 bands (blue, green, red, and near-infrared) and microwave remote sensing has 3 bands (low, medium, and high frequency), the global distribution feature relationship network (R) is a (3*4) matrix. For the elements (r_{ij}) in the matrix ((i=1,2,3) represents the band of microwave remote sensing, and (j=1,2,3,4) represents the band of optical remote sensing), the relative distribution information can be calculated by a specific algorithm.

[0109] One possible algorithm is to calculate the correlation coefficient. For example, for (r_{11}) (the relationship entity between the low-frequency microwave band and the blue band), the Pearson correlation coefficient calculation method can be used. Assume that the backscattering coefficient vector of the low-frequency microwave band is (X_{l}=[x_{l1},x_{l2},cdots,x_{ln}]) ((n) is the number of pixels in the target saline-alkali land area), and the blue band reflectivity vector is (Y_{b}=[y_{b1},y_{b2},cdots,y_{bn}]).

[0110] Pearson correlation coefficient (r_{11}=frac{sum_{k=1}^{n}(x_{lk}-mu_{l})(y_{bk}-mu_{b})}{sqrt{sum_{k=1}^{n}(x_{lk}-mu_{l})^2}sqrt{sum_{k=1}^{n}(y_{bk}-mu_{b})^2}}), set the calculation to get (r_{11}=0.3). All elements in the global distribution feature relationship network are calculated in this way.

[0111] According to the global distribution feature relationship network, the relative distribution vector between any of the multiple optical remote sensing and any of the multiple microwave remote sensing is determined. Taking the blue band as an example, the construction of its relative distribution vector can be based on the relationship with each microwave remote sensing band. Assuming that the blue band has a relationship entity with the low-frequency microwave band (r_{11}=0.3), a relationship entity with the medium-frequency microwave band (r_{12}=-0.2), and a relationship entity with the high-frequency microwave band (r_{13}=0.1) obtained from the global distribution feature relationship network, the relative distribution vector corresponding to the blue band is (D_{b}=[0.3,-0.2,0.1]). For other optical remote sensing bands, such as the green band and the red band, their respective relative distribution vectors are also constructed in the same way.

[0112] In this way, first of all, by determining the distribution information of optical remote sensing and microwave remote sensing in the target saline-alkali area respectively, the distribution characteristics of the two remote sensing data can be comprehensively and meticulously described. Constructing a global distribution characteristic relationship network can clarify the relative distribution information between different optical remote sensing bands and microwave remote sensing bands, which helps to deeply explore the internal relationship between the two remote sensing data. For example, by calculating the correlation coefficient to construct the relationship network, it can be found which optical remote sensing bands and microwave remote sensing bands have strong correlation or dependence. Determining the relative distribution vector further quantifies this relationship, enabling more accurate fusion of optical remote sensing and microwave remote sensing data in the subsequent saline-alkali land identification process. This method improves the rationality and effectiveness of data fusion, helps to enhance the characterization ability of the entire saline-alkali land identification system for the characteristics of the target saline-alkali area, reduces the identification error caused by improper data fusion, and thus improves the accuracy of saline-alkali land identification.

[0113] In a preferred embodiment, determining the saline-alkali state label to which the target saline-alkali area belongs according to the saline-alkali state attention discrimination vector includes: performing feature mapping on the saline-alkali state attention discrimination vector to obtain a saline-alkali state mapping vector; performing quantization coding on the saline-alkali state mapping vector to obtain multiple discrimination scores for the target saline-alkali area belonging to multiple saline-alkali state labels; and screening the saline-alkali state label corresponding to the discrimination score that meets the detection requirements as the saline-alkali state label to which the target saline-alkali area belongs.

[0114] In the process of determining the saline-alkali state label to which the target saline-alkali area belongs according to the saline-alkali state attention discrimination vector, a feature mapping operation is first performed. The purpose of feature mapping is to convert the information in the saline-alkali state attention discrimination vector into a form more suitable for quantization and classification. The saline-alkali state attention discrimination vector contains the comprehensive feature information about the target saline-alkali area obtained by fusing optical remote sensing and microwave remote sensing data through various methods before, but this information may have a complex structure and interrelationship, which is not convenient for directly judging the saline-alkali state label.

[0115] For example, assume that the saline-alkali state attention discrimination vector is \((Z = [z_1, z_2, \cdots, z_n])\) (where \((n)\) is the dimension of the vector). Adopting a linear feature mapping method, let the mapping matrix \((M)\) be an \((m*n)\) matrix (\((m < n)\), for example, \((m = 5)\), \((n = 10)\)), where the element \((m_{ij})\) of the matrix \((M)\) is the optimal parameter obtained by analyzing a large number of known saline-alkali land sample data. The saline-alkali state mapping vector \((Z_{map} = [z_{map1}, z_{map2}, \cdots, z_{mapm}])\) is obtained through matrix multiplication \((Z_{map} = M*Z)\). Specifically, numerically, if:

[0116] (M=begin{bmatrix}0.1&0.2&cdots&0.10.3&0.1&cdots&0.2cdots&cdots&cdots&cdots0.2&0.3&cdots&0.1end{bmatrix}), (Z=[0.5,0.3,cdots,0.2]), then (z_{map1}=(0.1*0.5+0.2*0.3+cdots+0.1*0.2)), calculate all elements in (Z_{map}) according to this method.

[0117] After obtaining the salinity state mapping vector, a quantization encoding operation is performed. The purpose of quantization encoding is to convert the feature information in the salinity state mapping vector into a discrimination score for the target salinity land area belonging to multiple salinity state labels. This process is to establish a connection between the features after feature mapping and different salinity state labels, and determine the possibility score of each label in a quantitative way.

[0118] One possible quantization coding method is based on distance metric. Assume that there are three known salinity state labels: mild salinity, moderate salinity and severe salinity. For each salinity state label, there is a predetermined standard vector. For example, the standard vector of mild salinity (L = [l_1, l_2, cdots, l_m]) (where the element value is the statistical characteristic value calculated based on a large amount of mild salinity sample data, such as (l_1 = 0.2), (l_2 = 0.3), etc.), the standard vector of moderate salinity (M = [m_1, m_2, cdots, m_m]), and the standard vector of severe salinity (H = [h_1, h_2, cdots, h_m]).

[0119] Calculate the distance between the salinity state mapping vector (Z_{map}) and each standard vector. For example, use the Euclidean distance formula (d_{L}=sqrt{sum_{i=1}^{m}(z_{mapi}-l_i)^2}), (d_{M}=sqrt{sum_{i=1}^{m}(z_{mapi}-m_i)^2}), (d_{H}=sqrt{sum_{i=1}^{m}(z_{mapi}-h_i)^2}). Then calculate the discrimination score based on the distance. For example, an exemplary scoring function (s_{L}=1-frac{d_{L}}{max(d_{L}, d_{M}, d_{H})}), (s_{M}=1-frac{d_{M}}{max(d_{L}, d_{M}, d_{H})}), (s_{H}=1-frac{d_{H}}{max(d_{L}, d_{M}, d_{H})}) can be used to obtain the discrimination scores (s_{L}), (s_{M}) and (s_{H}) of the target saline-alkali land area belonging to slightly saline-alkali land, moderately saline-alkali land and severely saline-alkali land.

[0120] The detection requirements are standards set based on the accuracy and reliability of saline-alkali land identification. This standard can be adjusted according to actual conditions. For example, the discrimination score must reach a certain threshold to be considered a reliable judgment result. The setting of detection requirements is usually based on empirical values ​​obtained by testing and verifying a large amount of sample data.

[0121] The threshold is set to (0.6). If (s_{L}=0.7), (s_{M}=0.5), (s_{H}=0.4), because (s_{L}) is greater than the threshold (0.6), the slightly saline-alkali land is selected as the saline-alkali state label of the target saline-alkali land area. If no discriminant score meets the detection requirements, it may be necessary to readjust the parameters of the feature mapping or quantization encoding, or collect more sample data to re-determine the standard vector and other operations to improve the accuracy of saline-alkali land identification.

[0122] In this way, firstly, the salinity state attention discriminant vector is converted into the salinity state mapping vector through feature mapping, which can simplify and optimize the data structure, highlight the feature information that is important for the judgment of salinity state, and reduce the noise and redundancy in the data. The quantitative coding process uses specific algorithms such as distance measurement to calculate the discriminant score, so that there is a quantitative evaluation basis for the possibility that the target salinity land area belongs to different salinity state labels, which improves the scientificity and accuracy of the judgment. Finally, the salinity state labels corresponding to the discriminant scores that meet the detection requirements are screened, which can effectively avoid misjudgment and ensure the reliability of the salinity land identification results. This step-by-step processing method comprehensively considers the conversion of data features, quantitative evaluation, and screening judgment, which helps to improve the performance of the entire salinity land identification system and more accurately determine the salinity state of the target salinity land area.

[0123] In summary, the embodiment of the present invention obtains the optical remote sensing spectrum vector and the microwave remote sensing band vector, and constructs an enhanced decision reconstruction vector to determine the saline-alkali state attention discriminant vector to obtain the saline-alkali state label. Its beneficial effect is remarkable, integrating optical remote sensing and microwave remote sensing data to improve the accuracy of saline-alkali land identification in many aspects. The two vectors respectively represent different remote sensing detection data, and the correlation between the two is mined to construct an enhanced vector, making full use of the complementarity of the two data. By determining the discriminant vector through the dual enhanced decision reconstruction vector, the characteristics of saline-alkali land can be deeply mined by integrating multi-source information. Finally, the saline-alkali state label is accurately determined to reduce misjudgment, providing a reliable basis for saline-alkali land management and monitoring.

[0124] An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the saline-alkali land identification method based on remote sensing decision is implemented.

[0125] An embodiment of the present invention provides a processor, which is used to run a program, wherein the saline-alkali land identification method based on remote sensing decision is executed when the program is running.

[0126] In the embodiment of the present invention, Figure 2 As shown, the saline-alkali land identification system 100 includes at least one processor 101, and at least one memory 102 and a bus 103 connected to the processor 101; wherein the processor 101 and the memory 102 communicate with each other through the bus 103; the processor 101 is used to call the program instructions in the memory 102 to execute the above-mentioned saline-alkali land identification method based on remote sensing decision.

[0127] The embodiments of the present invention are described with reference to the flowcharts and / or block diagrams of the methods, saline-alkali land identification systems (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0128] In a typical configuration, the saline-alkali land identification system includes one or more processors (CPU), a memory and a bus. The saline-alkali land identification system may also include an input / output interface, a network interface and the like.

[0129] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0130] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage computer readable storage media or any other non-transmission media that can be used to store information that can be accessed by the salt-alkali land identification system. As defined in this article, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0131] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or computer-readable storage medium that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such process, method, commodity or computer-readable storage medium. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the presence of other identical elements in the process, method, commodity or computer-readable storage medium that includes the elements.

[0132] It should be understood by those skilled in the art that the embodiments of the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the embodiments of the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0133] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A saline-alkali land identification method based on remote sensing decision-making, characterized in that: Applied to a saline-alkali land identification system, the method comprises: Acquire an optical remote sensing spectrum vector and a microwave remote sensing band vector of a target saline-alkali land area, wherein the optical remote sensing spectrum vector is used to characterize the optical remote sensing detection data corresponding to the target saline-alkali land area, and the microwave remote sensing band vector is used to characterize the microwave remote sensing detection data corresponding to the target saline-alkali land area; For the optical remote sensing spectrum vector, obtaining a first enhanced decision reconstruction vector according to a local microwave remote sensing band vector associated with the optical remote sensing spectrum vector in the microwave remote sensing band vector, wherein the first enhanced decision reconstruction vector is used to represent the optical remote sensing spectrum vector integrated with the microwave remote sensing band vector; For the microwave remote sensing band vector, obtaining a second enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector, wherein the second enhanced decision reconstruction vector is used to represent the microwave remote sensing band vector integrated with the optical remote sensing spectrum vector; Determining a salinity state attention discrimination vector through the first reinforcement decision reconstruction vector and the second reinforcement decision reconstruction vector; The salinity state label to which the target saline-alkali land area belongs is determined according to the salinity state attention discrimination vector.

2. The method according to claim 1, characterized in that The step of obtaining the optical remote sensing spectrum vector and the microwave remote sensing band vector of the target saline-alkali land area includes: Extracting optical remote sensing spectrum state elements of the optical remote sensing detection data, combining the optical remote sensing spectrum state elements with the optical remote sensing distribution vector to obtain the optical remote sensing spectrum vector; The microwave remote sensing environment state element of the microwave remote sensing detection data is extracted, and the microwave remote sensing environment state element is combined with a microwave remote sensing distribution vector to obtain the microwave remote sensing band vector.

3. The method according to claim 2, characterized in that The optical remote sensing detection data includes spectrum category data and spectrum band data; the optical remote sensing spectrum vector includes spectrum category vector and spectrum band vector; The extracting of the optical remote sensing spectrum state element of the optical remote sensing detection data comprises: extracting the spectrum category spectrum state element of the spectrum category data and the spectrum band spectrum state element of the spectrum band data; The combining of the optical remote sensing spectrum state element and the optical remote sensing distribution vector to obtain the optical remote sensing spectrum vector includes: combining the spectrum category spectrum state element and the spectrum category distribution vector to obtain the spectrum category vector; combining the spectrum band spectrum state element and the spectrum band distribution vector to obtain the spectrum band vector.

4. The method according to claim 3, characterized in that The first enhanced decision reconstruction vector includes a spectrum category enhanced decision reconstruction vector and a spectrum band enhanced decision reconstruction vector, and the obtaining of the first enhanced decision reconstruction vector for the optical remote sensing spectrum vector according to a local microwave remote sensing band vector associated with the optical remote sensing spectrum vector in the microwave remote sensing band vector includes: For the spectral category vector, according to at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector, obtaining the spectral category enhanced decision reconstruction vector, wherein the spectral category enhanced decision reconstruction vector is used to represent the spectral category vector after integrating the spectral band vector and the microwave remote sensing band vector; For the spectral band vector, the spectral band enhancement decision reconstruction vector is obtained based on at least part of the vectors in the spectral category vector and the microwave remote sensing band vector that are respectively associated with the spectral band vector. The spectral band enhancement decision reconstruction vector is used to represent the spectral band vector after integrating the spectral category vector and the microwave remote sensing band vector.

5. The method according to claim 4, characterized in that The obtaining of the spectral category enhanced decision reconstruction vector according to at least part of the vectors in the spectral band vector and the microwave remote sensing band vector respectively associated with the spectral category vector comprises: Obtaining a first spectral category enhanced decision reconstruction vector according to at least a portion of the vectors in the spectral band vector that are associated with the spectral category vector; Obtaining a second spectral category enhanced decision reconstruction vector according to at least a portion of vectors in the microwave remote sensing band vector that are associated with the spectral category vector; The first spectrum category enhanced decision reconstruction vector and the second spectrum category enhanced decision reconstruction vector are collided to obtain a third spectrum category enhanced decision reconstruction vector; Fully connect the third spectrum category enhancement decision reconstruction vector to obtain the spectrum category enhancement decision reconstruction vector.

6. The method according to claim 4, characterized in that The obtaining of the spectral band enhancement decision reconstruction vector according to at least part of the vectors in the spectral category vector and the microwave remote sensing band vector respectively associated with the spectral band vector comprises: Obtaining a first spectral band enhancement decision reconstruction vector according to at least a portion of vectors in the spectral category vector associated with the spectral band vector; Obtaining a second spectral band enhancement decision reconstruction vector according to at least a portion of vectors in the microwave remote sensing band vector that are associated with the spectral band vector; The first spectral band enhancement decision reconstruction vector and the second spectral band enhancement decision reconstruction vector are collided to obtain a third spectral band enhancement decision reconstruction vector; Fully connect the third spectral band enhancement decision reconstruction vector to obtain the spectral band enhancement decision reconstruction vector.

7. The method according to claim 3, characterized in that The obtaining of a second enhanced decision reconstruction vector for the microwave remote sensing band vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the optical remote sensing spectrum vector comprises: For the microwave remote sensing band vector, obtaining the second enhanced decision reconstruction vector according to the spectrum category vector and the local optical remote sensing spectrum vectors respectively associated with the microwave remote sensing band vector in the spectrum band vector; The acquiring the second enhanced decision reconstruction vector according to the local optical remote sensing spectrum vectors respectively associated with the microwave remote sensing band vector in the spectrum category vector and the spectrum band vector comprises: Obtaining a first microwave remote sensing enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the spectrum category vector; Obtaining a second microwave remote sensing enhanced decision reconstruction vector according to a local optical remote sensing spectrum vector associated with the microwave remote sensing band vector in the spectrum band vector; The first microwave remote sensing enhanced decision reconstruction vector and the second microwave remote sensing enhanced decision reconstruction vector are collided to obtain a third microwave remote sensing enhanced decision reconstruction vector; The third microwave remote sensing enhanced decision reconstruction vector is fully connected to obtain the second enhanced decision reconstruction vector.

8. The method according to claim 2, characterized in that: The optical remote sensing distribution vector and the microwave remote sensing distribution vector are both relative distribution vectors between the optical remote sensing detection data and the microwave remote sensing detection data, and the relative distribution vector is used to characterize the priority and correlation between the optical remote sensing detection data and the microwave remote sensing detection data; The relative distribution vector is obtained by: Determine, based on a plurality of optical remote sensings in the optical remote sensing detection data and a plurality of microwave remote sensings in the microwave remote sensing detection data, distribution information corresponding to each of the plurality of optical remote sensings and the plurality of microwave remote sensings in the target saline-alkali land area; According to the distribution information, a global distribution feature relationship network is constructed, wherein any relationship entity in the global distribution feature relationship network is used to characterize the relative distribution information between the optical remote sensing corresponding to the column to which the entity belongs and the microwave remote sensing corresponding to the row to which the entity belongs; A relative distribution vector between any one of the plurality of optical remote sensings and any one of the plurality of microwave remote sensings is determined based on the global distribution feature relationship network.

9. The method according to claim 1, characterized in that: Determining the salinity state label of the target saline-alkali land area according to the salinity state attention discrimination vector includes: Performing feature mapping on the salinity state attention discrimination vector to obtain a salinity state mapping vector; Quantizing and encoding the salinity state mapping vector to obtain multiple discrimination scores of the target salinity land area belonging to multiple salinity state labels; The salinity state label corresponding to the discrimination score that meets the detection requirements is selected as the salinity state label to which the target saline-alkali land area belongs.

10. A saline-alkali land identification system, characterized in that: It comprises a processor, a memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the saline-alkali land identification method based on remote sensing decision-making as described in any one of claims 1-9.

Images