A sea clutter prediction method and device based on spatiotemporal correlation

Abstract

The present invention provides a sea clutter prediction method and device based on spatiotemporal correlation. The method collects sea clutter data and corresponding environmental parameter data corresponding to a range unit to be measured and its two adjacent range units, and then applies a time correlation acquisition graph model based on the collected sea clutter data to obtain a temporal dependency feature of the range unit to be measured. The environmental dependency feature of the range unit to be measured is obtained based on the collected environmental parameter data, and then a high-dimensional prediction feature of the range unit to be measured is obtained. The left adjacent dependency feature and the right adjacent dependency feature are then obtained by combining the data corresponding to the two adjacent range units. The temporal correlation is also used within the two adjacent range units to fully exploit the spatial correlation. Thus, a sea clutter prediction result for the range unit to be measured is obtained based on the temporal correlation and spatial correlation, thereby substantially improving the sea clutter prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of sea clutter prediction, and in particular to a sea clutter prediction method and device based on spatiotemporal correlation. Background Art

[0002] Sea clutter is the backscattered echo received by radar when illuminating the sea surface. As the most complex form of radar clutter, its probability distribution typically deviates significantly from a Gaussian distribution, and its statistical characteristics vary over time, exhibiting non-Gaussian and non-stationary characteristics. Due to the influence of gravity waves and other factors, sea clutter within the same range unit exhibits temporal correlation, while sea clutter texture components within different range units within the same time period exhibit spatial correlation. Furthermore, sea clutter is sea-state dependent and significantly influenced by ocean environmental parameters. In applications of detecting sea targets, sudden increases in sea clutter echo amplitude can lead to false alarms and misjudgments. Therefore, it is necessary to conduct sea clutter prediction research and, based on the prediction results, suppress sea clutter interference.

[0003] Currently, sea clutter prediction methods are categorized into traditional model prediction methods, acoustic model prediction methods, and machine learning prediction methods. Traditional model prediction methods include empirical models and statistical models. Empirical models typically make predictions based on sea clutter data from a past period, while statistical models predict sea clutter by establishing mathematical models. Acoustic models are a method that uses acoustic principles to predict sea clutter. These methods typically construct complex acoustic models to simulate sound wave propagation in the ocean environment and use known sea clutter data to train the model, thereby predicting future sea clutter occurrences. Machine learning is an artificial intelligence method that uses a large amount of historical sea clutter data to predict future sea clutter occurrences. However, these methods lack in-depth exploration of the temporal and spatial correlation characteristics of sea clutter during modeling, resulting in limited sea clutter prediction accuracy.

[0004] In view of this, overcoming the defects of the prior art is an urgent problem to be solved in this technical field. Summary of the Invention

[0005] The problem to be solved by the present invention is: how to improve the prediction accuracy of sea clutter.

[0006] In a first aspect, a sea clutter prediction method based on spatiotemporal correlation is provided, comprising:

[0007] Collect sea clutter data and corresponding environmental parameter data corresponding to the distance unit to be measured and its two adjacent distance units, and pre-process the sea clutter data and environmental parameter data of different distance units respectively;

[0008] Obtaining corresponding graph modeling based on time correlation according to the pre-processed sea clutter data of the range unit to be measured, and extracting temporal dependency features according to the graph modeling corresponding to the range unit to be measured;

[0009] Obtaining corresponding environment-dependent features according to the pre-processed environmental parameter data of the distance unit to be measured;

[0010] Obtaining high-dimensional prediction features of the distance unit to be measured based on the time-series dependent features and the environment dependent features;

[0011] Obtaining a left adjacent dependency feature corresponding to the left adjacent distance unit according to the pre-processed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data, and obtaining a right adjacent dependency feature corresponding to the right adjacent distance unit according to the pre-processed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data;

[0012] A sea clutter prediction result corresponding to the range unit to be measured is obtained according to the high-dimensional prediction feature, the left adjacent dependent feature and the right adjacent dependent feature.

[0013] Preferably, the collecting of sea clutter data and corresponding environmental parameter data corresponding to the distance unit to be measured and its two adjacent distance units, and pre-processing the sea clutter data and environmental parameter data of different distance units respectively, specifically includes:

[0014] Querying the collected sea clutter data and environmental parameter data to obtain missing values ​​and / or difference values ​​in the sea clutter data and environmental parameter data;

[0015] Taking each missing value and / or difference value obtained from the sea clutter data and the environmental parameter data as the center value, and interpolating the corresponding missing value and / or difference value with the mean value of the preset span time window around the center value;

[0016] The interpolated sea clutter data and environmental parameter data are subjected to maximum normalization processing to obtain preprocessed sea clutter data and environmental parameter data.

[0017] Preferably, the obtaining of corresponding graph modeling based on time correlation according to the pre-processed sea clutter data of the range unit to be measured, and extracting temporal dependency features according to the graph modeling corresponding to the range unit to be measured, specifically includes:

[0018] The pre-processed sea clutter data of the range unit to be measured at multiple times are used as the node feature matrix, and the weighted adjacency matrix of the sea clutter data at different times is obtained;

[0019] Obtaining a graph model of sea clutter data based on time correlation according to the node feature matrix and the weighted adjacency matrix;

[0020] The graph modeling is input into the graph neural network for learning to obtain the temporal dependency features corresponding to the distance unit to be measured.

[0021] Preferably, the step of obtaining the corresponding environment-dependent features based on the pre-processed environmental parameter data of the distance unit to be measured specifically includes:

[0022] Obtaining multi-source input according to the pre-processed environmental parameter data of the distance unit to be measured;

[0023] The multi-source input is input into the GRU model to obtain the environment-dependent features corresponding to the distance unit to be measured.

[0024] Preferably, obtaining the high-dimensional prediction features of the distance unit to be measured based on the time-series dependent features and the environment dependent features specifically includes:

[0025] Obtain the weight of the sea clutter data and the weight of the environmental parameter data corresponding to the range unit to be measured;

[0026] The weight of the sea clutter data is used as the weight corresponding to the time series dependent feature, and the weight of the environmental parameter data is used as the weight corresponding to the environment dependent feature. The time series dependent feature and the environment dependent feature are weighted and summed to obtain the high-dimensional prediction feature of the distance unit to be measured.

[0027] Preferably, obtaining the left adjacent dependency feature corresponding to the left adjacent distance unit according to the pre-processed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data specifically includes:

[0028] Obtaining a corresponding left adjacent graph model based on time correlation according to the preprocessed sea clutter data of the left adjacent range unit, and extracting a left adjacent temporal dependency feature according to the left adjacent graph model;

[0029] Obtain the corresponding left adjacent environment dependency feature according to the pre-processed environmental parameter data of the left adjacent distance unit;

[0030] The left adjacent dependency feature is obtained according to the left adjacent temporal dependency feature and the left adjacent environmental dependency feature.

[0031] Preferably, obtaining the right adjacent dependency feature corresponding to the right adjacent distance unit according to the pre-processed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data specifically includes:

[0032] Acquire a corresponding right adjacent graph model based on time correlation according to the preprocessed sea clutter data of the right adjacent range unit, and extract a right adjacent temporal dependency feature according to the right adjacent graph model;

[0033] Obtaining corresponding right adjacent environment dependency features according to pre-processed environmental parameter data of the right adjacent distance unit;

[0034] The right adjacent dependency feature is obtained according to the right adjacent temporal dependency feature and the right adjacent environmental dependency feature.

[0035] Preferably, obtaining the sea clutter prediction result corresponding to the range unit to be measured according to the high-dimensional prediction feature, the left adjacent dependency feature, and the right adjacent dependency feature specifically includes:

[0036] Get the attention layer parameters corresponding to the distance unit to be measured, the left adjacent distance unit, and the right adjacent distance unit respectively;

[0037] Obtaining the unnormalized attention feature of the distance unit to be measured according to the high-dimensional prediction feature and the attention layer parameter of the distance unit to be measured, obtaining the unnormalized attention feature of the left adjacent distance unit according to the left adjacent dependency feature and the attention layer parameter of the left adjacent distance unit, and obtaining the unnormalized attention feature of the right adjacent distance unit according to the right adjacent dependency feature and the attention layer parameter of the right adjacent distance unit;

[0038] Obtaining the attention weight corresponding to the distance unit to be measured, the attention weight corresponding to the left adjacent distance unit, and the attention weight corresponding to the right adjacent distance unit according to the unnormalized attention feature of the distance unit to be measured, the unnormalized attention feature of the left adjacent distance unit, and the unnormalized attention feature of the right adjacent distance unit;

[0039] The attention vector is obtained by weighted summing the high-dimensional prediction features, left adjacent dependent features, and right adjacent dependent features according to their corresponding attention weights;

[0040] The attention vector is subjected to low-dimensional mapping to obtain a sea clutter prediction result corresponding to the range unit to be measured.

[0041] In a second aspect, a method for predicting sea clutter based on spatiotemporal correlation is provided, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor to perform the method for predicting sea clutter based on spatiotemporal correlation.

[0042] In a third aspect, the present invention further provides a non-volatile computer storage medium, wherein the computer storage medium stores computer-executable instructions, and the computer-executable instructions are executed by one or more processors to complete the method described in the first aspect.

[0043] In a fourth aspect, a chip is provided, comprising: a processor and an interface, for calling and running a computer program stored in a memory to execute the method of the first aspect.

[0044] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer or a processor, causes the computer or the processor to execute the method of the first aspect.

[0045] In a sixth aspect, a sea clutter prediction system based on spatiotemporal correlation is provided, comprising the sea clutter prediction based on spatiotemporal correlation according to the second aspect and using the sea clutter prediction method based on spatiotemporal correlation according to the first aspect.

[0046] The present invention provides a sea clutter prediction method and device based on spatiotemporal correlation. The method collects sea clutter data and corresponding environmental parameter data corresponding to a range unit to be measured and its two adjacent range units, and then builds a time correlation acquisition graph model based on the collected sea clutter data to obtain a temporal dependency feature of the range unit to be measured. Furthermore, the environmental dependency feature of the range unit to be measured is obtained based on the collected environmental parameter data. Furthermore, a high-dimensional prediction feature of the range unit to be measured is obtained. The left-adjacent dependency feature and the right-adjacent dependency feature are obtained by combining the data corresponding to the two adjacent range units. The method is used to fully exploit the spatial correlation, thereby obtaining a sea clutter prediction result for the range unit to be measured based on the temporal correlation and spatial correlation, thereby substantially improving the sea clutter prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A flow chart of a method for predicting sea clutter based on spatiotemporal correlation provided by an embodiment of the present invention;

[0049] Figure 2 A flow chart of a method for preprocessing collected data in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0050] Figure 3 A flow chart of a method for obtaining time-dependent features in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0051] Figure 4A flow chart of a method for obtaining environment-dependent features in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0052] Figure 5 A flow chart of a method for obtaining high-dimensional prediction features in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0053] Figure 6 A flow chart of a method for obtaining left-neighbor dependency features in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0054] Figure 7 A flow chart of a method for obtaining right adjacent dependent features in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0055] Figure 8 A flow chart of a method for obtaining sea clutter prediction results in a sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention;

[0056] Figure 9 A flow chart of a method for predicting sea clutter based on spatiotemporal correlation provided by an embodiment of the present invention;

[0057] Figure 10 A flowchart of another method for predicting sea clutter based on spatiotemporal correlation provided by an embodiment of the present invention;

[0058] Figure 11 A schematic diagram of a sea clutter prediction device based on spatiotemporal correlation provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0060] In the description of the present invention, it should be understood that the terms "center", "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 disclosure 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 a limitation on the present disclosure.

[0061] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, "multiple" means two or more. In addition, for example, the description may also use the method of adding "A" and "B" at the end to describe the same type of nouns as two independent individuals. In this case, the corresponding features defined as "A" and "B" are only used to distinguish the description purposes of the same type of individuals, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated.

[0062] As used herein, "about," "substantially," or "approximately" includes the stated value and an average value that is within an acceptable range of deviation from the particular value as determined by one of ordinary skill in the art taking into account the measurements in question and the errors associated with the measurement of the particular quantity (i.e., the limitations of the measurement system).

[0063] Unless the context requires otherwise, throughout the specification and claims, the term "including" is to be interpreted as meaning open inclusion, that is, "including, but not limited to". In the description of the specification, the terms "one embodiment", "some embodiments", "exemplary embodiments", "example", "specific example" or "some examples" and the like are intended to indicate that the specific features, structures, materials or characteristics associated with the embodiment or example are included in at least one embodiment or example of the present disclosure. The schematic representation of the above terms does not necessarily refer to the same embodiment or example. In addition, the specific features, structures, materials or characteristics may be included in any one or more embodiments or examples in any appropriate manner, that is, although they may be carried in the embodiments or examples of the above terms due to reasons such as the order and position of appearance, it is not limited to that they can be carried in combination by one embodiment or example.

[0064] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0065] Example 1:

[0066] This embodiment provides a sea clutter prediction method based on spatiotemporal correlation. Figure 1 As shown, including:

[0067] In step 101 , sea clutter data and corresponding environmental parameter data corresponding to the range unit to be measured and its two adjacent range units are collected, and the sea clutter data and environmental parameter data of different range units are preprocessed respectively.

[0068] In this embodiment, the sea clutter data can be collected by radar, and the environmental parameter data can be obtained by a buoy, an anemometer, or an ocean observation station. The environmental parameter data may include one or more of significant wave height, maximum wave height, wave direction, wind speed, wind direction, and sea conditions. It should be noted that the sea clutter data and the environmental parameter data are both historical data of corresponding distance units. Since missing values ​​and / or outliers are inevitable in the process of collecting the above data, and since missing values ​​and outliers may cause errors in subsequent training and prediction-related calculations, in this embodiment, the missing values ​​and / or outliers in the data are processed through the preprocessing, thereby reducing the impact of the missing values ​​and / or outliers in subsequent calculations.

[0069] In this embodiment, due to the influence of gravity waves, the sea clutter in different distance units in the same time period has spatial correlation. In this embodiment, by simultaneously acquiring two adjacent distance units (i.e., the left adjacent distance unit and the right adjacent distance unit), and introducing the sea clutter data and environmental parameter data of the two adjacent distance units into the subsequent training and prediction related calculations, the prediction results are guaranteed to have spatial correlation, thereby improving the accuracy of the prediction results.

[0070] In step 102, a corresponding graph model is obtained based on the pre-processed sea clutter data of the range unit to be measured based on time correlation, and a temporal dependency feature is extracted based on the graph model corresponding to the range unit to be measured.

[0071] In this embodiment, due to the influence of gravity waves, sea clutter data at different time points within the same distance unit exhibits temporal correlation. Sea clutter data is collected at different time points for the corresponding distance unit to be measured. Therefore, when acquiring the corresponding graph model, weighting can be performed based on the temporal correlation between different time points to explore the influence between the sea clutter data at different time points. The graph model is then learned using a corresponding graph neural network, outputting corresponding high-order feature vectors as the temporal dependency features. These temporal dependency features are then applied to subsequent prediction-related calculations to ensure that the prediction results exhibit temporal correlation.

[0072] In step 103, the corresponding environment-dependent features are obtained according to the pre-processed environmental parameter data of the distance unit to be measured.

[0073] In this embodiment, since sea clutter data is dependent on sea conditions and is greatly affected by environmental parameter data, the relevant features of the environmental parameter data are indispensable when predicting sea clutter. Therefore, by inputting the environmental parameter data into the corresponding neural network, the outputted environmental dependency features are used to assist in predicting sea clutter at the position of the range unit to be measured.

[0074] In step 104, high-dimensional prediction features of the distance unit to be measured are obtained based on the time-dependent features and the environment-dependent features.

[0075] In this embodiment, the high-dimensional prediction feature is a prediction feature obtained based on the historical environmental parameter data and historical sea clutter data of the predicted distance unit position. Since the historical environmental parameter data and historical sea clutter data are both from multiple time nodes, the high-dimensional prediction feature is a prediction feature that exploits time-related characteristics.

[0076] In step 105, the left adjacent dependency feature corresponding to the left adjacent distance unit is obtained based on the preprocessed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data, and the right adjacent dependency feature corresponding to the right adjacent distance unit is obtained based on the preprocessed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data.

[0077] In this embodiment, the calculation and acquisition process of the left adjacent dependent feature and the right adjacent dependent feature is consistent with the high-order prediction feature, thereby ensuring that the left adjacent dependent feature and the right adjacent dependent feature are obtained by relying on the corresponding environmental parameter data and mining the corresponding time-related features.

[0078] In step 106, a sea clutter prediction result corresponding to the range unit to be measured is obtained according to the high-dimensional prediction feature, the left adjacent dependency feature, and the right adjacent dependency feature.

[0079] In this embodiment, after obtaining the high-dimensional prediction features of the distance unit to be measured, the left-adjacent dependency features of the left-adjacent distance unit, and the right-adjacent dependency features of the right-adjacent distance unit, the attention mechanism is used to assign weights to the above features, and low-dimensional mapping is performed through the fully connected layer. By introducing the left-adjacent dependency features and the right-adjacent dependency features, the spatial correlation is fully exploited in the prediction process, thereby obtaining the prediction result of the sea clutter of the distance unit to be measured. The prediction result obtained at this time fully exploits the temporal correlation and spatial correlation, thereby improving the prediction accuracy of the sea clutter.

[0080] The sea clutter data and the corresponding environmental parameter data corresponding to the distance unit to be measured and its two adjacent distance units are collected, and the sea clutter data and the environmental parameter data of different distance units are pre-processed respectively, such as Figure 2 As shown, including:

[0081] In step 201 , the collected sea clutter data and environmental parameter data are queried to obtain missing values ​​and / or difference values ​​in the sea clutter data and environmental parameter data.

[0082] In this embodiment, the sea clutter data can be acquired by radar, the significant wave height, maximum wave height, and wave direction in the environmental parameter data can be acquired by buoys deployed around the radar, the wind speed and wind direction in the environmental parameter data can be acquired by an anemometer installed around the radar, and other environmental parameter data can be acquired through ocean observation stations. During the above process, errors caused by environmental or equipment factors may result in missing values ​​and / or discrepancies in the acquired data.

[0083] In step 202, each missing value and / or difference value obtained from the sea clutter data and the environmental parameter data is used as a center value, and the corresponding missing value and / or difference value is interpolated using the mean value of a preset span time window around the center value.

[0084] The preset span time window is set by those skilled in the art according to actual conditions. In this embodiment, the preset span time window can be set to 10.

[0085] In step 203 , maximum normalization processing is performed on the interpolated sea clutter data and environmental parameter data to obtain pre-processed sea clutter data and environmental parameter data.

[0086] In this embodiment, the maximum value normalization is calculated as follows:

[0087]

[0088] Among them, x is the original data of the collected sea clutter data and environmental parameter data, x min is the minimum value corresponding to the original data, x max is the maximum value corresponding to the original data.

[0089] The sea clutter data after preprocessing of the distance unit to be measured is used to obtain the corresponding graph model based on time correlation, and the temporal dependency feature is extracted based on the graph model corresponding to the distance unit to be measured, such as Figure 3 As shown, the method flow includes:

[0090] In step 301, the pre-processed sea clutter data of the range unit to be measured at multiple moments are used as a node feature matrix, and a weighted adjacency matrix of the sea clutter data at different moments is obtained.

[0091] In this embodiment, the node feature matrix is:

[0092]

[0093] in, is the sea clutter data of the range unit to be measured at the Nth moment, is the node feature matrix.

[0094] The weighted adjacency matrix is ​​A mid , A mid The relative weights of the elements at different time nodes are A ij , where A ij The expression is:

[0095]

[0096] Among them, t i is the i-th time node, t j is the j-th time node, exp(*) is the natural exponential function, and σ is the scale factor.

[0097] In step 302, a graph model of sea clutter data based on time correlation is obtained according to the node feature matrix and the weighted adjacency matrix.

[0098] In this embodiment, the graph is modeled as

[0099] In step 303, the graph modeling is input into a graph neural network for learning to obtain the temporal dependency features corresponding to the distance unit to be measured.

[0100] In this embodiment, the graph modeling can be learned using an M-layer ChebNet network, and the calculation method is as follows:

[0101]

[0102] Among them, ChebNet M (*) is the M-layer ChebNet network, is the timing dependency characteristic corresponding to the distance unit to be measured.

[0103] The corresponding environment-dependent features are obtained based on the pre-processed environmental parameter data of the distance unit to be measured, such as Figure 4 As shown, the method flow includes:

[0104] In step 401, multi-source input is obtained according to the pre-processed environmental parameter data of the distance unit to be measured.

[0105] In this embodiment, the multi-source input is:

[0106] in, and are the effective wave height, maximum wave height, wave direction, wind speed, wind direction and sea conditions corresponding to the distance unit to be measured at time k.

[0107] In step 402, the multi-source input is input into the GRU model to obtain the environment-dependent features corresponding to the distance unit to be measured.

[0108] In this embodiment, the calculation method corresponding to the GRU model is as follows:

[0109]

[0110] Among them, GRU M (*) is the M-layer GRU model, is the environment-dependent feature.

[0111] The high-dimensional prediction features of the distance unit to be measured are obtained according to the time-dependent features and the environment-dependent features, such as Figure 5 As shown, the method flow includes:

[0112] In step 501, the weight of the sea clutter data and the weight of the environmental parameter data corresponding to the range unit to be measured are obtained.

[0113] In step 502, the weight of the sea clutter data is used as the weight corresponding to the time series dependent feature, and the weight of the environmental parameter data is used as the weight corresponding to the environment dependent feature. The time series dependent feature and the environment dependent feature are weighted and summed to obtain the high-dimensional prediction feature of the distance unit to be measured.

[0114] The weighted sum is calculated as follows:

[0115]

[0116] in, is the weight of the sea clutter data, is the weight of environmental parameter data, is the high-dimensional prediction feature.

[0117] The left adjacent dependency feature corresponding to the left adjacent distance unit is obtained based on the pre-processed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data, such as Figure 6 As shown, the method flow includes:

[0118] In step 601, a corresponding left adjacent graph model is obtained based on the pre-processed sea clutter data of the left adjacent range unit based on time correlation, and a left adjacent temporal dependency feature is extracted based on the left adjacent graph model.

[0119] In this embodiment, the left adjacent graph is modeled as The corresponding left adjacent temporal dependency feature is calculated as follows:

[0120]

[0121] The left adjacent graph is modeled The definition of and the calculation method of the left adjacent time series dependency feature can refer to the graph modeling and time series dependency features of the distance unit to be predicted mentioned above, and will not be elaborated here.

[0122] In step 602, the corresponding left adjacent environment dependent feature is obtained according to the pre-processed environmental parameter data of the left adjacent distance unit.

[0123] The multi-source input of the pre-processed environmental parameter data of the left adjacent distance unit is defined as The definition method refers to the multi-source input of the pre-processed environmental parameter data of the distance unit to be measured. The calculation method of the left adjacent environment dependent feature is:

[0124]

[0125] in, It is the left adjacent environment dependent feature.

[0126] In step 603, the left adjacent dependency feature is obtained according to the left adjacent temporal dependency feature and the left adjacent environmental dependency feature.

[0127] Get the weight of the sea clutter data of the left adjacent range unit and weights of environmental parameter data The left adjacent dependency feature is obtained by weighted summation, and the calculation method is as follows:

[0128]

[0129] in, is the left adjacent dependent feature.

[0130] The right adjacent dependency feature corresponding to the right adjacent distance unit is obtained according to the pre-processed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data, such as Figure 7 As shown, the method flow includes:

[0131] In step 701, a corresponding right adjacent graph model is obtained based on time correlation according to the pre-processed sea clutter data of the right adjacent range unit, and a right adjacent temporal dependency feature is extracted according to the right adjacent graph model.

[0132] In this embodiment, the right adjacent graph is modeled as The corresponding right adjacent temporal dependency feature is calculated as follows:

[0133]

[0134] The right adjacent graph model The definition of and the calculation method of the right adjacent time series dependency feature can refer to the graph modeling and time series dependency features of the distance unit to be predicted mentioned above, and will not be elaborated here.

[0135] In step 702, the corresponding right adjacent environment dependent feature is obtained according to the pre-processed environmental parameter data of the right adjacent distance unit.

[0136] The multi-source input of the pre-processed environmental parameter data of the right adjacent distance unit is defined as The definition method refers to the multi-source input of the pre-processed environmental parameter data of the distance unit to be measured. The calculation method of the right adjacent environment dependent feature is:

[0137]

[0138] in, It is the right adjacent environment dependent feature.

[0139] In step 703, the right adjacent dependency feature is obtained according to the right adjacent temporal dependency feature and the right adjacent environmental dependency feature.

[0140] Get the weight of the sea clutter data of the right adjacent range unit and weights of environmental parameter data The right adjacent dependency feature is obtained by weighted summation, and the calculation method is as follows:

[0141]

[0142] in, is the right adjacent dependent feature.

[0143] The sea clutter prediction result corresponding to the distance unit to be measured is obtained according to the high-dimensional prediction feature, the left adjacent dependent feature and the right adjacent dependent feature, such as Figure 8 As shown, the method flow includes:

[0144] In step 801, the attention layer parameters corresponding to the distance unit to be measured, the left adjacent distance unit and the right adjacent distance unit are obtained respectively.

[0145] In step 802, the unnormalized attention features of the distance unit to be measured are obtained based on the high-dimensional prediction features and the attention layer parameters of the distance unit to be measured, the unnormalized attention features of the left adjacent distance unit are obtained based on the left adjacent dependent features and the attention layer parameters of the left adjacent distance unit, and the unnormalized attention features of the right adjacent distance unit are obtained based on the right adjacent dependent features and the attention layer parameters of the right adjacent distance unit.

[0146] Get the unnormalized attention feature corresponding to the distance unit to be measured. The calculation method is:

[0147]

[0148] Among them, P mid is the unnormalized attention feature corresponding to the distance unit to be measured, w mid and b mid These are the attention layer parameters corresponding to the distance unit to be measured.

[0149] Get the unnormalized attention feature of the left adjacent distance unit, calculated as:

[0150]

[0151] Among them, P left is the unnormalized attention feature corresponding to the left adjacent distance unit, w left and b left They are all attention layer parameters corresponding to the left adjacent distance unit.

[0152] Get the unnormalized attention feature of the right adjacent distance unit, calculated as:

[0153]

[0154] Among them, P right is the unnormalized attention feature corresponding to the right adjacent distance unit, w right and b right They are all attention layer parameters corresponding to the right adjacent distance unit.

[0155] In step 803, the attention weight corresponding to the distance unit to be measured, the attention weight corresponding to the left adjacent distance unit, and the attention weight corresponding to the right adjacent distance unit are obtained based on the unnormalized attention features of the distance unit to be measured, the unnormalized attention features of the left adjacent distance unit, and the unnormalized attention features of the right adjacent distance unit.

[0156] Get the attention weight a corresponding to the distance unit to be measured mid , calculated as:

[0157]

[0158] Get the attention weight a corresponding to the left adjacent distance unit left , calculated as:

[0159]

[0160] Get the attention weight a corresponding to the left adjacent distance unit right , calculated as:

[0161]

[0162] In step 804, the high-dimensional prediction features, the left-adjacent dependent features, and the right-adjacent dependent features are weighted and summed according to their respective corresponding attention weights to obtain an attention vector.

[0163] The attention vector V all The calculation method is:

[0164]

[0165] In step 805, low-dimensional mapping is performed on the attention vector to obtain a sea clutter prediction result corresponding to the range unit to be measured.

[0166] In this embodiment, the attention vector V is processed by L fully connected layers. all For low-dimensional mapping, the corresponding calculation method is:

[0167]

[0168] Among them, FC L (*) is the L-layer fully connected layer, is the sea clutter prediction result corresponding to the range unit to be measured.

[0169] Figure 9 and Figure 10 These are all flowcharts of the sea clutter prediction method based on spatiotemporal correlation provided by this embodiment.

[0170] This embodiment can achieve the following beneficial effects through the above method:

[0171] (1) Considering the temporal correlation between sea clutter data in the same distance unit, the sea clutter data is modeled and the sea clutter graph data is learned to deeply explore the temporal correlation characteristics between sea clutter data. In addition, the corresponding environmental parameter data is extracted and weightedly fused with the sea clutter data to make the feature extraction richer and more conducive to obtaining accurate sea clutter prediction results.

[0172] (2) Considering that the sea clutter texture components of different distance units in the same time period have spatial correlation, and the closer the distance, the stronger the correlation, the sea clutter data of the left and right adjacent distance units are introduced for auxiliary prediction. At the same time, the attention mechanism is used to assign weights to the influence degree to further improve the prediction accuracy of sea clutter.

[0173] Example 2:

[0174] like Figure 11FIG2 is a schematic diagram of a sea clutter prediction device based on spatiotemporal correlation according to an embodiment of the present invention. The sea clutter prediction device based on spatiotemporal correlation according to this embodiment includes one or more processors 41 and a memory 42 .

[0175] The processor 41 and the memory 42 may be connected via a bus or other means. Figure 11 The bus connection is taken as an example.

[0176] Memory 42, as a nonvolatile computer-readable storage medium, can be used to store nonvolatile software programs and nonvolatile computer-executable programs, such as the sea clutter prediction method based on spatiotemporal correlation described in the above embodiment. Processor 41 executes the sea clutter prediction method based on spatiotemporal correlation by running the nonvolatile software programs and instructions stored in memory 42.

[0177] The memory 42 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some embodiments, the memory 42 may optionally include a memory remotely located relative to the processor 41, and such remote memory may be connected to the processor 41 via a network. Examples of such networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0178] The program instructions / modules are stored in the memory 42 and, when executed by the one or more processors 41, perform the sea clutter prediction method based on spatiotemporal correlation in the above embodiment, for example, perform the above described Figure 1-Figure 5 The steps shown.

[0179] An embodiment of the present invention further provides a computer storage medium having computer program instructions stored thereon; when the computer program instructions are executed by a processor, the sea clutter prediction method based on spatiotemporal correlation provided by an embodiment of the present invention is implemented.

[0180] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A sea clutter prediction method based on spatiotemporal correlation, characterized in that: include: Collect sea clutter data and corresponding environmental parameter data corresponding to the distance unit to be measured and its two adjacent distance units, and pre-process the sea clutter data and environmental parameter data of different distance units respectively; Obtaining corresponding graph modeling based on time correlation according to the pre-processed sea clutter data of the range unit to be measured, and extracting temporal dependency features according to the graph modeling corresponding to the range unit to be measured; Obtaining corresponding environment-dependent features according to the pre-processed environmental parameter data of the distance unit to be measured; Obtaining high-dimensional prediction features of the distance unit to be measured based on the time-series dependent features and the environment dependent features; Obtaining a left adjacent dependency feature corresponding to the left adjacent distance unit according to the pre-processed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data, and obtaining a right adjacent dependency feature corresponding to the right adjacent distance unit according to the pre-processed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data; A sea clutter prediction result corresponding to the range unit to be measured is obtained according to the high-dimensional prediction feature, the left adjacent dependent feature and the right adjacent dependent feature.

2. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The collecting of sea clutter data and corresponding environmental parameter data corresponding to the distance unit to be measured and its two adjacent distance units, and pre-processing of the sea clutter data and environmental parameter data of different distance units specifically includes: Querying the collected sea clutter data and environmental parameter data to obtain missing values ​​and / or difference values ​​in the sea clutter data and environmental parameter data; Taking each missing value and / or difference value obtained from the sea clutter data and the environmental parameter data as the center value, and interpolating the corresponding missing value and / or difference value with the mean value of the preset span time window around the center value; The interpolated sea clutter data and environmental parameter data are subjected to maximum normalization processing to obtain preprocessed sea clutter data and environmental parameter data.

3. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The method of obtaining a corresponding graph model based on the time correlation of the pre-processed sea clutter data of the distance unit to be measured, and extracting a temporal dependency feature based on the graph model corresponding to the distance unit to be measured, specifically includes: The pre-processed sea clutter data of the range unit to be measured at multiple times are used as the node feature matrix, and the weighted adjacency matrix of the sea clutter data at different times is obtained; Obtaining a graph model of sea clutter data based on time correlation according to the node feature matrix and the weighted adjacency matrix; The graph modeling is input into the graph neural network for learning to obtain the temporal dependency features corresponding to the distance unit to be measured.

4. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The step of obtaining the corresponding environment-dependent features based on the pre-processed environmental parameter data of the distance unit to be measured specifically includes: Obtaining multi-source input according to the pre-processed environmental parameter data of the distance unit to be measured; The multi-source input is input into the GRU model to obtain the environment-dependent features corresponding to the distance unit to be measured.

5. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The step of obtaining the high-dimensional prediction features of the distance unit to be measured based on the time-series dependent features and the environment dependent features specifically includes: Obtain the weight of the sea clutter data and the weight of the environmental parameter data corresponding to the range unit to be measured; The weight of the sea clutter data is used as the weight corresponding to the time series dependent feature, and the weight of the environmental parameter data is used as the weight corresponding to the environment dependent feature. The time series dependent feature and the environment dependent feature are weighted and summed to obtain the high-dimensional prediction feature of the distance unit to be measured.

6. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The step of obtaining the left adjacent dependency feature corresponding to the left adjacent distance unit according to the pre-processed sea clutter data of the left adjacent distance unit and the corresponding environmental parameter data specifically includes: Obtaining a corresponding left adjacent graph model based on time correlation according to the preprocessed sea clutter data of the left adjacent range unit, and extracting a left adjacent temporal dependency feature according to the left adjacent graph model; Obtain the corresponding left adjacent environment dependency feature according to the pre-processed environmental parameter data of the left adjacent distance unit; The left adjacent dependency feature is obtained according to the left adjacent temporal dependency feature and the left adjacent environmental dependency feature.

7. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The obtaining of the right adjacent dependency feature corresponding to the right adjacent distance unit according to the pre-processed sea clutter data of the right adjacent distance unit and the corresponding environmental parameter data specifically includes: Acquire a corresponding right adjacent graph model based on time correlation according to the preprocessed sea clutter data of the right adjacent range unit, and extract a right adjacent temporal dependency feature according to the right adjacent graph model; Obtaining corresponding right adjacent environment dependency features according to pre-processed environmental parameter data of the right adjacent distance unit; The right adjacent dependency feature is obtained according to the right adjacent temporal dependency feature and the right adjacent environmental dependency feature.

8. The sea clutter prediction method based on spatiotemporal correlation according to claim 1, characterized in that: The obtaining of the sea clutter prediction result corresponding to the range unit to be measured according to the high-dimensional prediction feature, the left adjacent dependent feature, and the right adjacent dependent feature specifically includes: Get the attention layer parameters corresponding to the distance unit to be measured, the left adjacent distance unit, and the right adjacent distance unit respectively; Obtaining the unnormalized attention feature of the distance unit to be measured according to the high-dimensional prediction feature and the attention layer parameter of the distance unit to be measured, obtaining the unnormalized attention feature of the left adjacent distance unit according to the left adjacent dependency feature and the attention layer parameter of the left adjacent distance unit, and obtaining the unnormalized attention feature of the right adjacent distance unit according to the right adjacent dependency feature and the attention layer parameter of the right adjacent distance unit; Obtaining the attention weight corresponding to the distance unit to be measured, the attention weight corresponding to the left adjacent distance unit, and the attention weight corresponding to the right adjacent distance unit according to the unnormalized attention feature of the distance unit to be measured, the unnormalized attention feature of the left adjacent distance unit, and the unnormalized attention feature of the right adjacent distance unit; The attention vector is obtained by weighted summing the high-dimensional prediction features, left adjacent dependent features, and right adjacent dependent features according to their corresponding attention weights; The attention vector is subjected to low-dimensional mapping to obtain a sea clutter prediction result corresponding to the range unit to be measured.

9. A sea clutter prediction device based on spatiotemporal correlation, characterized in that: The system comprises at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor to execute the sea clutter prediction method based on spatiotemporal correlation according to any one of claims 1 to 8.

10. A non-volatile computer storage medium, characterized in that The computer storage medium stores computer program instructions, which, when executed by one or more processors, implement the sea clutter prediction method based on spatiotemporal correlation according to any one of claims 1 to 8.

Images