Economical crop monitoring data interpolation method and device based on second-order optimization hidden features

Abstract

The invention discloses an economic crop monitoring data interpolation method and device based on second-order optimization hidden features. The method specifically comprises the following steps: S1, obtaining historical economic crop monitoring data from a server and storing the historical economic crop monitoring data; s2, constructing an economic crop monitoring data tensor according to the stored historical economic crop monitoring data; s3, constructing a target loss function according to the historical economic crop monitoring data and the economic crop monitoring data tensor, carrying out iterative training, and extracting a hidden feature matrix; and S4, calculating an economic crop monitoring data interpolation value according to the extracted hidden feature matrix.

Description

Technical Field

[0001] The present invention relates to the technical field of economic crop monitoring data, and in particular to an economic crop monitoring data interpolation method and device based on second-order optimized latent features. Background Art

[0002] With the advancement of agricultural modernization and the in-depth application of smart agriculture technologies, cash crop growth monitoring increasingly relies on multi-source and multi-modal optoelectronic sensing methods, such as RGB cameras, thermal infrared, spectrometers, drones, and satellite remote sensing. These multimodal sensing data provide a rich information foundation for precision agriculture management and are key indicators of cash crop growth status, environmental stress responses, and the effectiveness of agronomic regulation. However, due to factors such as optoelectronic sensing device failures, dispersed deployment, cloud cover, communication interruptions, and errors during data transmission, multimodal optoelectronic sensing data for cash crops is commonly subject to numerous missing values, strong data heterogeneity, and large ranges, severely restricting its efficient utilization.

[0003] To address this challenge, research on efficient and accurate methods for missing data completion is urgently needed. One feasible technical approach is a representation learning method based on latent feature analysis and low-rank modeling. By constructing a high-order incomplete multimodal perception data tensor, it uniformly models multidimensional features such as cash crop type, cash crop monitoring indicators, and monitoring time. Low-rank decomposition of the tensor is then used to extract its intrinsic structural properties, thereby achieving high-quality inference of missing information. This method obtains a low-dimensional latent space embedding representation of multimodal data through training, and its product is as close as possible to the original observation tensor, achieving accurate completion of the missing data.

[0004] However, due to complex data structures, non-convex optimization objectives, and multilinear coupling, traditional first-order gradient-based optimization methods face bottlenecks in both accuracy and efficiency, making it difficult to meet the practical needs of completing high-dimensional, multimodal cash crop data. Therefore, how to efficiently model and accurately complete multimodal incomplete data for cash crop perception scenarios has become a key issue in the current smart agriculture field that urgently needs to be addressed. Summary of the Invention

[0005] Aiming at the problem of incomplete interpolation of high-dimensional multimodal economic crop data in the prior art, the present invention proposes an economic crop monitoring data interpolation method and device based on second-order optimized latent features.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The economic crop monitoring data interpolation method based on second-order optimized latent features specifically includes the following steps:

[0008] S1: Obtain historical economic crop monitoring data from the server and store it;

[0009] S2: constructs the economic crop monitoring data tensor based on the stored historical economic crop monitoring data;

[0010] S3: Construct the target loss function based on the historical cash crop monitoring data and the cash crop monitoring data tensor and perform iterative training to extract the latent feature matrix;

[0011] S4: Calculate the interpolation value of economic crop monitoring data based on the extracted latent feature matrix.

[0012] Preferably, in S1, the received economic crop monitoring data is in the form of a four-tuple data structure Y = (u, s, t, v), where u represents the economic crop type, s represents the specific indicators monitored by various sensors, t represents the time of data collection, and v represents the value of the economic crop u monitored by the economic crop monitoring sensor for the s type indicator at time t.

[0013] Preferably, in said S2, the economic crop monitoring data tensor is V I×J×K , where I is the number of economic crop types, J is the number of various sensor monitoring indicators, and K is the number of time slices for economic crop monitoring data collection.

[0014] Preferably, the S3 includes:

[0015] S3-1: Initialize process parameters for interpolation of cash crop monitoring data;

[0016] S3-2: Construct the target loss function based on the historical cash crop monitoring data and the cash crop monitoring data tensor;

[0017] S3-3: Perform training iterative optimization on the target loss function;

[0018] S3-4: Determine whether the target loss function reaches the termination condition during the training iteration process. If not, continue the training iteration. If yes, stop the training iteration and output the latent features.

[0019] Preferably, in S3-1, the process parameters include latent feature matrices U, S and T; the first-order gradient term g ε , the first-order gradient term g ε Including g ε (u i ), g ε (s j ) and g ε (t k ), corresponding to the first-order gradient components of the three latent feature matrices U, S, and T respectively; update the increment vector Δ ε , update the increment vector Δ ε Includes three update increment components vector Δ ε(u i ),Δ ε (s j ) and Δ ε (t k ), corresponding to the updated incremental component vector of each row of the three latent feature matrices U, S and T; the dimension D of the latent feature matrix; the convergence termination threshold δ; and the damping term coefficient γ.

[0020] Preferably, in S3-2, the target loss function ε is expressed by the following formula:

[0021]

[0022] In formula (1), ε represents the target loss function; Γ represents the set of known elements contained in the economic crop monitoring data tensor V; v ijk represents the monitoring value of the jth monitoring indicator of the economic crop u at the kth time in the economic crop monitoring data tensor V; u i,d Represents the value of the d-th dimension of the i-th row vector in the latent feature matrix U of the cash crop type; s j,d Represents the value of the d-th dimension of the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k,d Represents the value of the d-th dimension of the k-th row vector in the temporal latent feature matrix T; f represents the total number of dimensions.

[0023] Preferably, in S3-3, the training iterative optimization method is:

[0024] First, during the training process, the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration round is calculated. ε ;

[0025] Then, during the training process, the update increment of the optimization objective loss function ε latent feature matrix in the current training iteration round is calculated, that is, the first-order gradient information g obtained by the first-order optimization unit is ε Perform nonlinear scaling to obtain the update increment Δ ε ;

[0026] The training iteration formula is as follows:

[0027]

[0028] In formula (2), represents the training iteration of ε in τ-1 for u i The first-order negative gradient information of u i represents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t kRepresents the kth row vector in the temporal latent feature matrix T; represents the training iteration round of ε in τ-1 for s j The first-order negative gradient information; Denotes the effect of ε on t in τ-1 training iterations k The first-order negative gradient information of g ε Indicates that all and The first-order negative gradient information composed of the component vectors; CGD represents the conjugate gradient second-order solver; ρ represents the convergence accuracy of the conjugate gradient second-order solver; M represents the maximum iteration round of the conjugate gradient second-order solver; γ is the damping term coefficient; The i-th row vector in the latent feature matrix U representing the economic crop type for τ training iterations; The i-th row vector in the latent feature matrix U representing the economic crop type for τ-1 training iterations; and Corresponding, used to update the feature vector u i The update increment; and Corresponding, used to update the feature vector s j The update increment; and Corresponding to, used to update the latent feature vector t k The update increment of Δ ε Indicates that all and The component vectors are combined into an update increment vector; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ training iterations; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ-1 training iterations; The kth row vector in the temporal latent feature matrix T representing τ training iterations; represents the kth row vector in the temporal latent feature matrix T of τ-1 training iterations.

[0029] Preferably, in said S4, the calculation formula for the interpolation value of the economic crop monitoring data is:

[0030]

[0031] In formula (3), Indicates the interpolation value of economic crop monitoring data, economic crop type u i Economic crop monitoring sensor data indicators j At time t k Interpolation value of economic crop monitoring data under ui represents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k Represents the kth row vector in the temporal latent feature matrix T.

[0032] The present invention also provides an economic crop monitoring data interpolation device based on second-order optimization latent features, comprising a data acquisition module, a storage module, a tensor construction module, a latent feature extraction module and an output module;

[0033] The data acquisition module is used to obtain historical economic crop monitoring data from the server and store it;

[0034] A storage module, used for storing received historical economic crop monitoring data and economic crop monitoring data interpolation values;

[0035] A tensor construction module is used to construct a cash crop monitoring data tensor based on the stored historical cash crop monitoring data;

[0036] The latent feature extraction module is used to construct a target loss function based on historical cash crop monitoring data and cash crop monitoring data tensors to extract latent features;

[0037] The output module is used to output the interpolation value of economic crop monitoring data based on the extracted latent features.

[0038] Preferably, the latent feature extraction module includes an initialization unit, a target loss function construction unit, and a training unit;

[0039] Initialization unit, used to initialize the process parameters involved in the interpolation process of economic crop monitoring data;

[0040] A target loss function construction unit is used to construct a target loss function based on historical economic crop monitoring data, economic crop monitoring data tensors, and process parameters;

[0041] The training unit is used to train and optimize the target loss function and extract latent features;

[0042] The training unit includes a first-order optimization unit and a second-order optimization unit:

[0043] The first-order optimization unit is used to calculate the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration during the training process ε ;

[0044] The second-order optimization unit is used to calculate the update increment of the optimization target loss function ε latent feature matrix in the current training iteration round during the training process, that is, the first-order gradient information g obtained by the first-order optimization unit εPerform nonlinear scaling to get the update increment Δ ε , and then complete the training optimization iteration.

[0045] In summary, due to the adoption of the above technical solution, compared with the prior art, the present invention has at least the following beneficial effects:

[0046] This invention provides a method and device for interpolating cash crop monitoring data based on second-order optimization latent features. This method is specifically designed for cash crop monitoring data and can interpolate cash crop monitoring data with high accuracy and in accordance with statistical laws, thereby interpolating missing cash crop monitoring data. The method is widely applicable in fields such as cash crop monitoring and environmental monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of an economic crop monitoring data interpolation method based on second-order optimized latent features according to an exemplary embodiment of the present invention.

[0048] Figure 2 Schematic diagram of an economic crop monitoring data interpolation device based on second-order optimized latent features according to an exemplary embodiment of the present invention.

[0049] Figure 3 FIG. 4 is a schematic diagram of a storage module according to an exemplary embodiment of the present invention.

[0050] Figure 4 Schematic diagram of a latent feature extraction module according to an exemplary embodiment of the present invention.

[0051] Figure 5 FIG. 4 is a schematic diagram of a training unit according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be further described in detail below with reference to the examples and specific implementation methods. However, this should not be understood as limiting the scope of the present invention to the following examples, as all technologies implemented based on the present invention fall within the scope of the present invention.

[0053] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0054] like Figure 1As shown, the present invention provides an economic crop monitoring data interpolation method based on second-order optimized latent features, which specifically includes the following steps:

[0055] S1: Receive the cash crop monitoring data interpolation instruction sent by the server, obtain the historical cash crop monitoring data from the server and store it.

[0056] The server may send the instruction in the following manner: regularly, according to the notification of the device, or according to the notification of a server, the server may send the instruction requiring the economic crop monitoring data to be interpolated to the data receiving module.

[0057] In this embodiment, the received cash crop monitoring data is in the form of a four-tuple data structure Y = (u, s, t, v), where u represents the cash crop type, s represents the specific indicators monitored by various sensors, t represents the time of data collection, and v represents the specific value of the cash crop u monitored by the cash crop monitoring sensor for the indicator of type s at time t;

[0058] S2: Construct an economic crop monitoring data tensor V based on the stored historical economic crop monitoring data.

[0059] In this embodiment, all quadruple groups Y = (u, s, t, v) are constructed into an economic crop monitoring data tensor V of size I × J × K, where I is the number of economic crop types, J is the number of various sensor monitoring indicators, and K is the number of time slices for economic crop monitoring data collection.

[0060] V in tensor ijk represents the specific value of the cash crop monitoring indicator of cash crop type s monitored by the cash crop monitoring sensor at time t, 1≤i≤I, 1≤j≤J, 1≤k≤K, and Γ is the set of known elements contained in the cash crop monitoring data tensor V. Because various sensors may experience data loss during data transmission and cannot synchronously collect all indicators at all time points, the constructed cash crop monitoring data matrix V is a sparse incomplete tensor.

[0061] S3: Construct the target loss function based on the historical cash crop monitoring data and the cash crop monitoring data tensor and perform iterative training to extract the latent feature matrix.

[0062] S3-1: Initialize process parameters for interpolation of economic crop monitoring data.

[0063] The process parameters include the latent feature matrices U, S and T; the first-order gradient term g ε , the first-order gradient term g ε Including g ε (u i ), g ε (sj ) and g ε (t k ), corresponding to the first-order gradient components of the three latent feature matrices U, S, and T respectively; update the increment vector Δ ε , update the increment vector Δ ε Includes three update increment components vector Δ ε (u i ),Δ ε (s j ) and Δ ε (t k ), corresponding to the updated incremental component vector of each row of the three latent feature matrices U, S and T; the dimension of the latent feature matrix D; the convergence termination threshold δ; the damping term coefficient γ;

[0064] in,

[0065] The dimension of the latent feature matrix D determines the dimension of the latent feature space of the latent feature matrices U, S, and T, and is initialized to a positive integer;

[0066] The sizes of the latent feature matrices U, S, and T are determined by the dimension values of each order of the corresponding economic crop monitoring data tensor V, that is, U is a latent feature matrix with I rows and D columns, S is a latent feature matrix with J rows and D columns, and T is a latent feature matrix with K rows and D columns. The three latent feature matrices are initialized with small random positive numbers.

[0067] The convergence termination threshold δ is a parameter used to determine whether the iterative process has converged, and is initialized with a very small positive number;

[0068] The damping term coefficient γ is a constant used to control the positive definiteness of the second-order curvature matrix and balance the degree of first-order and second-order approximation, and is initialized to a large positive number.

[0069] S3-2: Construct the target loss function ε based on the historical cash crop monitoring data and the cash crop monitoring data tensor, which is expressed as the following formula:

[0070]

[0071] In formula (1), ε represents the target loss function; Γ represents the set of known elements contained in the economic crop monitoring data tensor V; v ijk represents the monitoring value of the jth monitoring indicator of the economic crop u at the kth time in the economic crop monitoring data tensor V; u i,d Represents the value of the d-th dimension of the i-th row vector in the latent feature matrix U of the cash crop type; s j,d Represents the value of the d-th dimension of the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k,d Represents the value of the d-th dimension of the k-th row vector in the temporal latent feature matrix T; f represents the total number of dimensions.

[0072] In this embodiment, the target loss function uses Euclidean distance as the optimization target.

[0073] S3-3: To minimize the optimization objective loss function ε, the value of the optimization objective loss function ε can be as close to the second-order stable point as possible.

[0074] First, during the training process, the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration round is calculated. ε ;

[0075] Then, during the training process, the optimization objective loss function ε is calculated, and the second-order curvature information of the latent feature matrix in the current training iteration round is calculated. The first-order gradient information g obtained by the first-order optimization unit is ε Perform nonlinear scaling to obtain a more accurate update Δ of the optimization target ε .

[0076] The training iteration formula is as follows:

[0077]

[0078] In formula (2), represents the training iteration of ε in τ-1 for u i The first-order negative gradient information of u i represents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k Represents the kth row vector in the temporal latent feature matrix T; represents the training iteration round of ε in τ-1 for s j The first-order negative gradient information; Denotes the effect of ε on t in τ-1 training iterations k The first-order negative gradient information of g ε Indicates that all and The first-order negative gradient information composed of the component vectors; CGD represents the conjugate gradient second-order solver; ρ represents the convergence accuracy of the conjugate gradient second-order solver; M represents the maximum iteration round of the conjugate gradient second-order solver; γ is the damping term coefficient; The i-th row vector in the latent feature matrix U representing the economic crop type for τ training iterations; The i-th row vector in the latent feature matrix U representing the economic crop type for τ-1 training iterations; and Corresponding, used to update the feature vector u i The update increment; and Corresponding, used to update the feature vector s j The update increment; and Corresponding to, used to update the latent feature vector t k The update increment of Δ ε Indicates that all and The component vectors are combined into an update increment vector; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ training iterations; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ-1 training iterations; The kth row vector in the temporal latent feature matrix T representing τ training iterations; represents the kth row vector in the temporal latent feature matrix T of τ-1 training iterations.

[0079] S3-4: Determine whether the training iteration process of the target loss function ε on the known data set Λ reaches the termination condition. If not, continue the training iteration. If so, stop the training iteration and output the latent feature tensors U, S, and T.

[0080] In this step, the convergence judgment condition is that the absolute value of the difference between the calculated value of the target loss function ε and the value of the target loss function ε in the current round of 10 rounds of training and the value of the target loss function ε in the previous round is less than the threshold δ.

[0081] S4: Calculate the interpolation value of the economic crop monitoring data according to the extracted latent feature matrix, and store the interpolated economic crop monitoring data.

[0082] When the target loss function ε converges on the known data set Γ, the latent feature matrices U, S, and T obtained by training when the target loss function ε reaches the minimum value are used to calculate the economic crop type u with the highest accuracy. i Economic crop monitoring sensor data indicators j At time t k Interpolation values of economic crop monitoring data under

[0083]

[0084] In formula (3), Indicates the interpolation value of economic crop monitoring data, economic crop type u i Economic crop monitoring sensor data indicators j At time t k Interpolation value of economic crop monitoring data under u irepresents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k Represents the kth row vector in the temporal latent feature matrix T.

[0085] This invention provides a method and device for interpolating cash crop monitoring data based on second-order optimization latent features. This method is specifically designed for cash crop monitoring data and can interpolate cash crop monitoring data with high accuracy and in accordance with statistical laws, thereby interpolating missing cash crop monitoring data. The method is widely applicable in fields such as cash crop monitoring and environmental monitoring.

[0086] Based on the above-mentioned economic crop monitoring data interpolation method based on second-order optimization latent features, e.g. Figure 2 As shown, the present invention also provides an economic crop monitoring data interpolation device based on second-order optimization latent features, including a data acquisition module 1, a storage module 2, a tensor construction module 3, a latent feature extraction module 4 and an output module 5.

[0087] The output end of the data acquisition module 1 is connected to the first input end of the storage module 2, the output end of the storage module 2 is respectively connected to the input end of the tensor construction module 3, the output end of the tensor construction module 3 is connected to the input end of the latent feature extraction module 4, the output end of the latent feature extraction module 4 is connected to the input end of the output module 5, and the output end of the output module 5 is connected to the second input end of the storage module 2.

[0088] The data acquisition module 1 is used to acquire and store historical economic crop monitoring data from the server.

[0089] The storage module 2 is used to store the received historical economic crop monitoring data and the economic crop monitoring data interpolation values.

[0090] In this embodiment, Figure 3 As shown, the storage module 2 includes a first storage unit 21 and a second storage unit 22;

[0091] The first storage unit 21 is used to store the received historical cash crop monitoring data, which is stored and protected in a four-tuple data structure. The four-tuple data structure is in the form of Y = (u, s, t, v), where u represents the cash crop type, s represents the specific indicators monitored by each sensor, t represents the time of data collection, and v represents the specific value of cash crop u monitored by the cash crop monitoring sensor for indicator type s at time t.

[0092] The second storage unit 22 is used to store the interpolated value of the economic crop monitoring data output by the output module 5 .

[0093] In this embodiment, the tensor construction module 3 is used to construct the economic crop monitoring data tensor according to the stored historical economic crop monitoring data: V I×J×K , where I is the number of economic crop types, J is the number of various sensor monitoring indicators, and K is the number of time slices for economic crop monitoring data collection.

[0094] The latent feature extraction module 4 is used to construct a target loss function based on the historical economic crop monitoring data and the economic crop monitoring data tensor to extract latent features.

[0095] like Figure 4 As shown, the latent feature extraction module 4 includes an initialization unit 41, a target loss function construction unit 42, and a training unit 43; the output end of the initialization unit 41 is connected to the input end of the target loss function construction unit 42, the output end of the target loss function construction unit 42 is connected to the input end of the training unit 43, and the output end of the training unit 43 is connected to the output module 5.

[0096] Initialization unit 41, used for initializing process parameters involved in the interpolation process of economic crop monitoring data;

[0097] a target loss function construction unit 42 for constructing a target loss function based on historical cash crop monitoring data, cash crop monitoring data tensors, and process parameters;

[0098] The training unit 43 is used to train and optimize the target loss function and extract latent features.

[0099] like Figure 5 As shown, the training unit 43 includes a first-order optimization unit 431 and a second-order optimization unit 432; specifically:

[0100] The first-order optimization unit 431 is used to calculate the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration round during the training process. ε .

[0101] The second-order optimization unit 432 is used to calculate the update increment of the optimization target loss function ε latent feature matrix in the current training iteration round during the training process, that is, the first-order gradient information g obtained by the first-order optimization unit ε Perform nonlinear scaling to obtain a more accurate update increment Δ of the optimization target ε , and then complete the training optimization iteration.

[0102] The output module 5 is used to output the interpolation value of the economic crop monitoring data according to the extracted latent features.

[0103] The present invention also provides an electronic device, which includes a processor, and the processor is used to run a computer program stored in a memory, so that the electronic device implements the steps of the economic crop monitoring data interpolation method based on second-order optimized latent features in the above embodiment.

[0104] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program runs on a processor, it implements the steps of the economic crop monitoring data interpolation method based on second-order optimized latent features in the above embodiment.

[0105] A computer program includes computer program code, which may be in source code form, object code form, executable files, or some intermediate form. Computer-readable media may include at least any entity or device capable of carrying computer program code to an electronic device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunications signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, due to legislation and patent practice, computer-readable media cannot be electrical carrier signals or telecommunications signals.

[0106] Those skilled in the art will appreciate that the above-mentioned embodiments are specific examples for implementing the present invention, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present invention.

Claims

1. The economic crop monitoring data interpolation method based on second-order optimization latent features is characterized by: The specific steps include: S1: Obtain historical economic crop monitoring data from the server and store it; S2: constructs the economic crop monitoring data tensor based on the stored historical economic crop monitoring data; S3: Construct the target loss function based on the historical cash crop monitoring data and the cash crop monitoring data tensor and perform iterative training to extract the latent feature matrix; S4: Calculate the interpolation value of economic crop monitoring data based on the extracted latent feature matrix.

2. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 1 is characterized in that: In S1, the received economic crop monitoring data is in the form of a four-tuple data structure Y = (u, s, t, v), where u represents the economic crop type, s represents the specific indicators monitored by various sensors, t represents the time of data collection, and v represents the value of the economic crop u monitored by the economic crop monitoring sensor for the s type indicator at time t.

3. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 1 is characterized in that: In S2, the economic crop monitoring data tensor is V I×J×K , where I is the number of economic crop types, J is the number of various sensor monitoring indicators, and K is the number of time slices for economic crop monitoring data collection.

4. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 1 is characterized in that: The S3 includes: S3-1: Initialize process parameters for interpolation of cash crop monitoring data; S3-2: Construct the target loss function based on the historical cash crop monitoring data and the cash crop monitoring data tensor; S3-3: Perform training iterative optimization on the target loss function; S3-4: Determine whether the target loss function reaches the termination condition during the training iteration process. If not, continue the training iteration. If yes, stop the training iteration and output the latent features.

5. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 4 is characterized in that: In S3-1, the process parameters include the latent feature matrices U, S, and T; the first-order gradient term g ε , the first-order gradient term g ε Including g ε (u i ), g ε (s j ) and g ε (t k ), corresponding to the first-order gradient components of the three latent feature matrices U, S, and T respectively; update the increment vector Δ ε , update the increment vector Δ ε Includes three update increment components vector Δ ε (u i ),Δ ε (s j ) and Δ ε (t k ), corresponding to the updated incremental component vector of each row of the three latent feature matrices U, S and T; the dimension D of the latent feature matrix; the convergence termination threshold δ; and the damping term coefficient γ.

6. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 4 is characterized in that: In S3-2, the target loss function ε is expressed by the following formula: In formula (1), ε represents the target loss function; Γ represents the set of known elements contained in the economic crop monitoring data tensor V; v ijk represents the monitoring value of the jth monitoring indicator of the economic crop u at the kth time in the economic crop monitoring data tensor V; u i,d Represents the value of the d-th dimension of the i-th row vector in the latent feature matrix U of the cash crop type; s j,d Represents the value of the d-th dimension of the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k,d Represents the value of the d-th dimension of the k-th row vector in the temporal latent feature matrix T; f represents the total number of dimensions.

7. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 4 is characterized in that: In S3-3, the training iterative optimization method is: First, during the training process, the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration round is calculated. ε ; Then, during the training process, the update increment of the optimization objective loss function ε latent feature matrix in the current training iteration round is calculated, that is, the first-order gradient information g obtained by the first-order optimization unit is ε Perform nonlinear scaling to obtain the update increment Δ ε ; The training iteration formula is as follows: In formula (2), represents the training iteration of ε in τ-1 for u i The first-order negative gradient information; u i represents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k Represents the kth row vector in the temporal latent feature matrix T; represents the training iteration round of ε in τ-1 for s j The first-order negative gradient information; Denotes the effect of ε on t in τ-1 training iterations k The first-order negative gradient information of g ε Indicates that all and The first-order negative gradient information composed of the component vectors; CGD represents the conjugate gradient second-order solver; ρ represents the convergence accuracy of the conjugate gradient second-order solver; M represents the maximum iteration round of the conjugate gradient second-order solver; γ is the damping term coefficient; The i-th row vector in the latent feature matrix U representing the economic crop type for τ training iterations; Represents the i-th row vector in the latent feature matrix U of the economic crop type for τ-1 training iterations; and Corresponding, used to update the feature vector u i The update increment; and Corresponding, used to update the feature vector s j The update increment; and Corresponding to, used to update the latent feature vector t k The update increment of Δ ε Indicates that all and An update increment vector composed of component vectors; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ training iterations; Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators for τ-1 training iterations; The kth row vector in the temporal latent feature matrix T representing τ training iterations; represents the kth row vector in the temporal latent feature matrix T of τ-1 training iterations.

8. The economic crop monitoring data interpolation method based on second-order optimized latent features according to claim 1 is characterized in that: In S4, the calculation formula for the interpolation value of economic crop monitoring data is: In formula (3), Indicates the interpolation value of economic crop monitoring data, economic crop type u i Economic crop monitoring sensor data indicators j At time t k Interpolation value of economic crop monitoring data under u i represents the i-th row vector in the latent feature matrix U of the cash crop type; s j Represents the j-th row vector in the latent feature matrix S of various sensor monitoring indicators; t k Represents the kth row vector in the temporal latent feature matrix T.

9. An economic crop monitoring data interpolation device based on second-order optimized latent features according to the method of any one of claims 1 to 8, characterized in that: It includes data acquisition module, storage module, tensor construction module, latent feature extraction module and output module; The data acquisition module is used to obtain historical economic crop monitoring data from the server and store it; A storage module, used for storing received historical economic crop monitoring data and economic crop monitoring data interpolation values; A tensor construction module is used to construct a cash crop monitoring data tensor based on the stored historical cash crop monitoring data; The latent feature extraction module is used to construct a target loss function based on historical cash crop monitoring data and cash crop monitoring data tensors to extract latent features; The output module is used to output the interpolation value of economic crop monitoring data based on the extracted latent features.

10. The economic crop monitoring data interpolation device based on second-order optimization latent features according to claim 9, characterized in that: The latent feature extraction module includes an initialization unit, a target loss function construction unit, and a training unit; Initialization unit, used to initialize the process parameters involved in the interpolation process of economic crop monitoring data; A target loss function construction unit is used to construct a target loss function based on historical economic crop monitoring data, economic crop monitoring data tensors, and process parameters; The training unit is used to train and optimize the target loss function and extract latent features; The training unit includes a first-order optimization unit and a second-order optimization unit: The first-order optimization unit is used to calculate the first-order gradient information g of the optimization target loss function ε latent feature matrix in the current training iteration during the training process ε ; The second-order optimization unit is used to calculate the update increment of the optimization target loss function ε latent feature matrix in the current training iteration round during the training process, that is, the first-order gradient information g obtained by the first-order optimization unit ε Perform nonlinear scaling to get the update increment Δ ε , and then complete the training optimization iteration.

Images