A sea peak identification method, device and storage medium based on multi-domain feature extraction and machine learning

Abstract

The application discloses a sea spike identification method based on multi-domain feature extraction and machine learning, comprising the following steps: step 1, pre-processing each group of sea surface radar echo data to form a sea spike and sea clutter sample set; step 2, extracting time domain, frequency domain and time-frequency domain features of the samples in the sea spike and sea clutter sample set formed in step 1; step 3, after randomly disturbing the order of the sea spike and sea clutter feature data set formed in step 2, dividing the sea spike and sea clutter feature data set into a training set and a test set according to a ratio of 8:2, training the training set by using an XGBoost model, and outputting a sea spike identification result. The method disclosed by the application overcomes the problem that a traditional sea spike identification method mainly relies on experience criteria such as amplitude threshold, minimum duration and minimum time interval for detection, leading to the problem that the identification result is sensitive to artificial parameter setting and the environmental adaptability is insufficient under complex sea conditions.

Description

Technical Field

[0001] This invention belongs to the field of marine radar signal processing, and specifically relates to a method, device and storage medium for marine spike identification based on multi-domain feature extraction and machine learning. Background Technology

[0002] Sea clutter identification technology is an important research direction in radar sea surface target detection, and it is widely used in fields such as maritime target detection and marine monitoring. Sea clutter is a high-energy transient scattering phenomenon in sea clutter, and its echo characteristics are highly similar to those of small targets on the sea surface, making it one of the important factors leading to false alarms in sea surface radar. Research on sea clutter identification not only helps improve the accuracy of sea clutter modeling and simulation, but also provides reliable feature priors and decision criteria for sea surface target detection and clutter suppression. This has significant theoretical and engineering application value for improving the target detection performance of radar systems in complex marine environments.

[0003] In existing technologies, rule-based criteria are commonly used to identify sea spikes, such as setting amplitude thresholds, minimum durations, and minimum time intervals to determine the echo signal. This method is simple to implement, has clear physical meaning, and is practical in engineering applications. However, its recognition performance is highly sensitive to the selection of threshold parameters, which are usually set based on human experience. This results in poor adaptability to complex sea conditions and varying observation conditions, making it difficult to meet the requirements for recognition accuracy and robustness in practical applications. Summary of the Invention

[0004] To address the shortcomings of traditional sea spike identification techniques in complex sea clutter environments, which rely on manually set threshold parameters and are susceptible to changes in sea state, radar operating parameters, and the observation environment, resulting in insufficient accuracy and environmental adaptability, this invention provides a sea spike identification method, device, and storage medium based on multi-domain feature extraction and machine learning. Feature parameters characterizing sea spikes are extracted from multiple dimensions, including the time, frequency, and time-frequency domains. Machine learning algorithms are used to establish a nonlinear mapping relationship between multi-domain features and sea spike category labels, enabling automatic identification and classification of sea spikes and sea clutter. This provides an effective technical means for sea surface target detection and sea clutter suppression.

[0005] The present invention adopts the following technical solution:

[0006] An improved method for identifying sea peaks based on multi-domain feature extraction and machine learning includes the following steps:

[0007] Step 1: Preprocess each set of sea surface radar echo data to form a sample set of sea spikes and sea clutter.

[0008] Step 2: Extract time-domain, frequency-domain, and time-frequency-domain features from the samples in the sea spike and sea clutter sample set formed in Step 1. Combine the extracted features into a feature matrix and label it to form a sea spike and sea clutter feature dataset.

[0009] Step 3: Randomly shuffle the order of the sea spike and sea clutter feature datasets formed in Step 2, and divide them into training and test sets in an 8:2 ratio. Use the XGBoost model to train the training set, and use the test set to verify the correctness of the trained XGBoost model. Finally, use the trained XGBoost model as the sea spike recognition model and output the sea spike recognition results.

[0010] Furthermore, in step 1:

[0011] Based on the sea surface radar echo data of Group C, let The c-th group of sea surface radar echo data is represented as I× A complex matrix of dimension I, where I represents the number of distance units. Represents the number of pulses, I , The radar echo data over the sea surface is divided into pulse blocks and the pulse block length is set.

[0012] Based on the sea spike label value, the distance cell where the sea spike is located and the pulse block are... Each sampled complex value is denoted as a sea-peak sample; based on the sea-peak label value, any non-sea-peak pulse block within the same distance unit as the sea-peak is selected. Each sampled complex value is used as a sea clutter sample to form a set of sea spikes and sea clutter samples. The total number of samples in the sample set is denoted as S.

[0013] Furthermore, in step 2:

[0014] The sample from which the features are to be extracted is denoted as , , This represents the complex value of the nth sample;

[0015] right Temporal features were extracted, and the seven extracted temporal features were: relative mean amplitude (RAA), normalized amplitude variance (NAV), normalized amplitude skewness (SK), normalized amplitude kurtosis (KU), amplitude distribution formal parameter (v), amplitude distribution scale parameter (b), and correlation time length (TCL).

[0016] right Frequency domain features were extracted, and the four extracted frequency domain features were: Doppler offset DC, Doppler bandwidth DB, Doppler spectral entropy DSE, and Doppler peak power ratio PPR.

[0017] right Time-frequency domain features were extracted, and the three extracted time-frequency domain features were: time-frequency ridge energy RI, number of connected regions RN, and maximum connected region size LRS;

[0018] The 14 extracted features are combined into a 14-dimensional feature matrix and labeled, with sea spikes labeled as "1" and sea clutter labeled as "0", forming a feature dataset of sea spikes and sea clutter.

[0019] Furthermore, in step 3: a strong classifier is formed by weighted combination of 200 decision trees. During training, each new decision tree uses the negative gradient direction of the current model's loss function as the fitting target, selects the optimal splitting feature and splitting point from the 14-dimensional features, and the maximum depth of a single tree does not exceed 6 layers. After the new tree is trained, it is added to the model with a weight of 0.1. The model uses the log loss function as the optimization target and outputs the trained classifier after 200 iterations.

[0020] An improved method for identifying sea spikes based on multi-domain feature extraction and machine learning includes: a processor; a memory storing executable instructions of the processor; the processor being configured to perform the steps of the method described above by executing the executable instructions.

[0021] A computer-readable storage medium for storing a program, wherein the improvement is that the program, when executed, implements the steps of the above method.

[0022] The beneficial effects of this invention are:

[0023] The method disclosed in this invention overcomes the problems of traditional sea clutter identification methods, which mainly rely on empirical criteria such as amplitude thresholds, minimum durations, and minimum time intervals for detection. This leads to the identification results being sensitive to manual parameter settings and lacking adaptability to complex sea conditions. By constructing a multi-dimensional feature dataset that integrates time-domain, frequency-domain, and time-frequency-domain information, and utilizing the XGBoost model to establish a complex nonlinear mapping relationship between sea clutter and sea spikes, automatic identification and classification of sea spikes are achieved. Compared with traditional methods, this invention can fully utilize the differences between sea spikes and sea clutter in the multi-domain feature space, reduce the influence of human factors on the identification results, and improve the identification accuracy, robustness, stability, and generalization ability of sea spikes in complex sea clutter environments. This provides more accurate and reliable technical support for sea surface radar target detection and sea clutter analysis and suppression. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the method of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the 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 merely illustrative and not intended to limit the invention.

[0026] Example 1: Due to their transient and sudden characteristics, sea spikes differ significantly from sea clutter in multiple feature dimensions, including time-domain waveform characteristics, frequency-domain spectral structure, and time-frequency distribution characteristics. Therefore, sea spike identification can be viewed as a binary classification problem of sea clutter in terms of multi-domain features. This example discloses a sea spike identification method based on multi-domain feature extraction and machine learning. Based on the differences between sea spikes and sea clutter in the time, frequency, and time-frequency domain feature spaces, it preprocesses sea spike and sea clutter samples from sea surface radar echo data and extracts 14-dimensional feature parameters in the time, frequency, and time-frequency domains to construct a multi-dimensional feature dataset for sea spike identification. On this basis, the XGBoost algorithm is used to establish a nonlinear mapping relationship between multi-domain features and sea spike category labels, providing a reliable technical means for accurate identification of sea spikes in complex sea clutter environments. Figure 1 As shown, the specific steps include the following:

[0027] Step 1: Preprocess each set of sea surface radar echo data to form a sample set of sea spikes and sea clutter.

[0028] Based on the sea surface radar echo data of Group C, let The c-th group of sea surface radar echo data is represented as I× A complex matrix of dimension I, where I represents the number of distance units. Represents the number of pulses, I , The radar echo data over the sea surface is divided into pulse blocks and the pulse block length is set. ;

[0029] On each distance unit The pulses are divided into the following order: One pulse block:

[0030]

[0031] In the above formula, Indicates rounding down;

[0032] Based on the sea spike label value, the distance cell where the sea spike is located and the pulse block are... Each sampled complex value is denoted as a sea-peak sample; based on the sea-peak label value, any non-sea-peak pulse block within the same distance unit as the sea-peak is selected. Each sampled complex value is used as a sea clutter sample to form a set of sea spikes and sea clutter samples. The total number of samples in the sample set is denoted as S.

[0033] Step 2: Extract time-domain, frequency-domain, and time-frequency-domain features from the samples in the sea spike and sea clutter sample set formed in Step 1. Combine the extracted features into a feature matrix and label it to form a sea spike and sea clutter feature dataset.

[0034] Since the processing steps for sea spikes and sea clutter samples are the same, the sample s for which features are to be extracted is denoted as... , , , This represents the complex value of the nth sample;

[0035] right Temporal features were extracted, and the seven extracted temporal features were: relative mean amplitude (RAA), normalized amplitude variance (NAV), normalized amplitude skewness (SK), normalized amplitude kurtosis (KU), amplitude distribution formal parameter (v), amplitude distribution scale parameter (b), and correlation time length (TCL).

[0036] To calculate the relative average amplitude, reference cell data is required. , yes Two-dimensional complex matrix and ,in, It is the complex echo value of the nth pulse of the m-th distance unit of the reference unit, where M is the number of distance units of the reference unit. The selection criteria are: pulse dimension and To maintain consistency, the distance dimension should be chosen in the same way as... The M nearest neighbors.

[0037] Relative Average Amplitude (RAA) Pass Data with reference unit The ratio of the average amplitude is calculated as follows:

[0038]

[0039] The normalized magnitude variance (NAV) is calculated as follows:

[0040]

[0041]

[0042] In the above formula, var( ) is a built-in function in MATLAB.

[0043] The specific method for calculating normalized amplitude skewness is to normalize the amplitude mean to zero, and then calculate the third moment statistic. The formula for calculating normalized amplitude skewness SK is:

[0044]

[0045] The specific method for calculating normalized amplitude kurtosis is to normalize the amplitude mean to zero, and then use excess kurtosis calculation, which is the fourth moment minus three times the square of the normalized second moment. The formula for calculating normalized amplitude kurtosis (KU) is:

[0046]

[0047] The method used to extract the features of the amplitude distribution shape parameter v and scale parameter b is the K-distribution moment estimation method. First, calculate... First-order origin moment and second-order origin moment and establish and The parsing relationship:

[0048]

[0049]

[0050] In the above formula, It is the Gamma function, and Constructing sample moments ratio Obtain information about shape parameters The nonlinear equation, simplified, yields:

[0051]

[0052] Solve using the lookup table method: (The table is pre-defined in the original text.) Within the range, a lookup table is calculated and constructed with a step size of 0.001. From the table . Solve directly using the second-order moment relation:

[0053]

[0054] Calculating the time-dependent length requires first using the MATLAB built-in function xcorr( Calculate the time autocorrelation vector and :

[0055]

[0056] .right Normalization is performed to obtain the normalized time autocorrelation vector. :

[0057]

[0058] from Start traversal For the first time Corresponding delay Recorded as , This refers to the time-related length, which is then converted to milliseconds (ms):

[0059]

[0060] In the above formula, PRF is the pulse repetition frequency.

[0061] right Frequency domain features were extracted to characterize the spectral structure differences of the sea spikes. The four extracted frequency domain features were: Doppler shift (DC), Doppler bandwidth (DB), Doppler spectral entropy (DSE), and Doppler peak power ratio (PPR).

[0062] Frequency domain feature extraction requires solving... Spectrum :right conduct Point Fast Fourier Transform, where That is, taking a power of 2, which satisfies the requirements of the radix-2 FFT algorithm. =1024 launched =1024. Calculate. power spectrum :

[0063]

[0064] In the above formula, This is a built-in function in MATLAB. Regarding the power spectrum Energy normalization was performed to obtain the normalized Doppler power spectrum. ,in:

[0065]

[0066] Doppler shift characterizes the degree of shift of the echo spectrum energy center relative to the zero-frequency position. This invention employs a minimum weighted second-order moment estimation algorithm for calculation. By cyclically shifting the spectrum, it finds the position that minimizes the spectral weighted second-order moment, and uses the frequency corresponding to this position as the estimated Doppler frequency shift value. First, a spectral weighted second-order moment objective function is constructed. :

[0067]

[0068] In the above formula, It is the weighted sum of squares of the spectral energy distance from the center. This represents the shortest cyclic distance from the frequency index to the center of the spectrum. This is a built-in MATLAB function that performs the remainder operation. It searches for the position where the objective function reaches its minimum value. Finally, the Doppler shift value is obtained by converting it into angular frequency. The unit is :

[0069]

[0070] The Doppler bandwidth, building upon the previous method, is characterized by the square root of the least weighted second moment, representing the extent of the spectrum's spread around the center frequency. The formula for calculating the Doppler bandwidth (dB) is:

[0071]

[0072] This invention uses MATLAB built-in functions Normalized Doppler power spectrum Doppler spectral entropy The calculation.

[0073]

[0074] Doppler peak power ratio The calculation formula is:

[0075]

[0076] In the above formula, This is a built-in function in MATLAB.

[0077] right Time-frequency domain features were extracted, and the three extracted time-frequency domain features were: time-frequency ridge energy RI, number of connected regions RN, and maximum connected region size LRS;

[0078] The calculation is performed using the Smoothed Pseudo Wigner-Ville Distribution (SPWVD). The time-frequency distribution of the vector. This invention uses the wvd function in the MATLAB Signal Processing Toolbox to calculate the smoothed pseudo-Wigner-Ville distribution, with the input being... The pulse repetition frequency PRF = 1000, and 'smoothedPseudo' is specified to calculate the smoothed pseudo-Wigner-Ville distribution. The window lengths of the time window and frequency window are automatically selected by the function based on the pulse block length. Let... The output time-frequency distribution matrix, ,right The elements are moduloed and squared, and then the zero-frequency component is shifted to the center to obtain the time-frequency energy distribution matrix. :

[0079]

[0080]

[0081] In the above formula, Perform spectrum shifting using MATLAB built-in functions. Perform modulo operations using MATLAB's built-in functions.

[0082] Calculate reference cell data The time-frequency distribution matrix is ​​used to obtain the reference time-frequency distribution set, and the mean matrix is ​​calculated pixel by pixel from this set. Sum of standard deviation matrix ,right Perform Z-score normalization. This yields the normalized time-frequency energy distribution. ,and :

[0083]

[0084] Time-frequency spine energy The calculation method is as follows:

[0085]

[0086] That is, extract the most prominent frequency component at that moment, then sum it over time to quantize the overall energy intensity of the time-frequency ridge throughout the entire pulse block.

[0087] The number of connected components represents the number of significant energy clusters in the time-frequency plot. The calculation method is as follows:

[0088] Construct a binary matrix :

[0089]

[0090] In the above formula, Let L be the Lth largest energy value at time t, where L is an empirically set parameter, here taken as 5. Then, use the MATLAB built-in function bwconncomp to... Performing 8-neighbor connectivity analysis, the final output is the number of connected regions. :

[0091]

[0092] The maximum connected component size is calculated by applying the `numel` function to each cell of the cell array using the built-in MATLAB function `cellfun(@numel,())`. The number of elements with a value of 1 in each region is counted, and the maximum value is taken.

[0093]

[0094] In the above formula, Indicates the first The number of elements (area) contained within a connected region.

[0095] The 14 extracted features are combined into a 14-dimensional feature matrix and labeled, with sea spikes labeled as "1" and sea clutter labeled as "0", forming a feature dataset of sea spikes and sea clutter.

[0096] Step 3: Randomly shuffle the order of the sea spike and sea clutter feature datasets generated in Step 2 to eliminate the influence of data arrangement order on model training. The shuffled dataset contains sea spike samples and sea clutter samples, each containing 14-dimensional features and a 1-dimensional class label. A hierarchical random partitioning method is used to divide the feature dataset into training and test sets in an 8:2 ratio. This hierarchical partitioning ensures that the ratio of sea spike samples to sea clutter samples in the training and test sets is consistent with the original dataset, avoiding an imbalance in the proportion of one class of samples in the training or test sets due to random partitioning.

[0097] The XGBoost model is used to train the training set, and the correctness of the trained XGBoost model is verified using the test set. Finally, the trained XGBoost model is used as the sea crest recognition model, and the sea crest recognition results, accuracy, precision, recall and F1 score are output.

[0098] The model is built using the Python language environment. The XGBoost gradient boosting framework is used to build the sea spike classifier. The scikit-learn library is used to complete the data partitioning (stratified random partitioning, with the test set accounting for 20%), standardization preprocessing, and model performance evaluation. The classification results are visualized using matplotlib and seaborn libraries.

[0099] This method uses the XGBoost gradient boosting tree model as the classifier. XGBoost is a decision tree model based on the idea of ​​boosting ensemble learning. It progressively adds weak learners (decision trees) and fits the residuals from the previous step, finally combining the predictions of all weak learners in a weighted manner to obtain a strong classifier. The hyperparameters of the XGBoost classifier are set as follows:

[0100] The number of base learners (decision trees), i.e., the number of Boosting iterations n_estimators=200, indicates training 200 decision trees; the maximum depth of each decision tree max_depth=6, limiting the tree depth prevents overfitting of a single tree, and the value of 6 strikes a balance between model expressiveness and generalization performance; the learning rate (step shrinkage factor) learning_rate=0.1 controls the contribution weight of each tree to the final prediction result. A smaller learning rate combined with a larger number of trees is beneficial to improving the model's generalization ability; the proportion of randomly sampled samples when training each tree subsample=0.8 means that each tree is trained using 80% of the training samples, introducing randomness to prevent overfitting; random seed random_state=42 ensures that the results of each training session are reproducible; the evaluation metric is the log loss function. The calculation formula is:

[0101]

[0102] In the above formula, It is a real label (0 or 1). It is the probability that the model predicts that a sample belongs to class 1. , The log loss measures the deviation between the predicted probability and the true label; a smaller log loss indicates better classification performance.

[0103] A classifier was built and trained based on the XGBoost gradient boosting framework. The model employs an additive ensemble strategy, gradually adding decision trees to fit the prediction residuals of the previous stage. A strong classifier is ultimately formed by a weighted combination of 200 decision trees. During training, each new decision tree uses the negative gradient direction of the current model's loss function as its fitting target, selecting the optimal splitting feature and splitting point from 14-dimensional features. The maximum depth of a single tree does not exceed 6 layers. After training, new trees are added to the model with a learning rate of 0.1. The model uses a logarithmic loss function as its optimization objective and outputs the trained classifier after 200 iterations.

[0104] After training, the feature matrix of the test set is input into the trained XGBoost model, and the model outputs the predicted probability value p of each test sample belonging to the sea clutter category. Set a decision threshold. :when When, determine that the current pulse block is a sea spike; when At that time, the current pulse block is determined to be sea clutter.

[0105] Using accuracy ( ), accuracy ( ), recall rate ) and F1 score ( The metrics are used to evaluate the classification performance of the model on the test set. First, TP is defined as the number of samples with a label of 1 and a prediction of 1; TN is the number of samples with a label of 0 and a prediction of 0; FP is the number of samples with a label of 0 and a prediction of 1; and FN is the number of samples with a label of 1 and a prediction of 0.

[0106] The formula for calculating accuracy is:

[0107]

[0108] The formula for calculating accuracy is:

[0109]

[0110] The formula for calculating recall rate is:

[0111]

[0112] The formula for calculating the F1 score is:

[0113]

[0114] The effectiveness of the sea spike identification method disclosed in this invention is further illustrated below with actual measurement data:

[0115] Based on 2668 echo data files from a certain X-band radar, covering sea condition levels 1-5, 27554 sea spike samples and 27554 sea clutter samples were extracted. Through multi-dimensional feature extraction, 55108 14-dimensional feature vectors were formed. These feature vectors were labeled, with sea spikes marked as "1" and sea clutter as "0", and denoted as label vectors. These label vectors were then merged with the 14-dimensional feature vectors to form the feature dataset to be trained.

[0116] The dataset was shuffled and stratified into training and testing sets in an 8:2 ratio. The training set was then fed into the XGBoost model for training, resulting in a trained classification model. The testing set was then fed into the model, which outputs a predicted label for each sample. The predicted labels were compared with the true labels on the test set to calculate the accuracy, precision, recall, and F1 score for Haijianfeng identification.

[0117] XGBoost 0.9086 0.9027 0.9159 0.9093

[0118] As can be seen from the table above, the recognition accuracy reached 90.86%, verifying the effectiveness of the method of the present invention.

[0119] This embodiment also discloses a peak recognition device based on multi-domain feature extraction and machine learning, comprising: a processor; a memory storing executable instructions of the processor; the processor being configured to perform the steps of the above-described method by executing the executable instructions; and a computer-readable storage medium for storing a program that, when executed, implements the steps of the above-described method.

Claims

1. A method for identifying sea peaks based on multi-domain feature extraction and machine learning, characterized in that, Includes the following steps: Step 1: Preprocess each set of sea surface radar echo data to form a sample set of sea spikes and sea clutter. Step 2: Extract time-domain, frequency-domain, and time-frequency-domain features from the samples in the sea spike and sea clutter sample set formed in Step 1. Combine the extracted features into a feature matrix and label it to form a sea spike and sea clutter feature dataset. Step 3: Randomly shuffle the order of the sea spike and sea clutter feature datasets formed in Step 2, and divide them into training and test sets in an 8:2 ratio. Use the XGBoost model to train the training set, and use the test set to verify the correctness of the trained XGBoost model. Finally, use the trained XGBoost model as the sea spike recognition model and output the sea spike recognition results.

2. The method for identifying sea peaks based on multi-domain feature extraction and machine learning according to claim 1, characterized in that, In step 1: Based on the sea surface radar echo data of Group C, let The c-th group of sea surface radar echo data is represented as I× A complex matrix of dimension I, where I represents the number of distance units. Represents the number of pulses, I , The radar echo data over the sea surface is divided into pulse blocks and the pulse block length is set. Based on the sea spike label value, the distance cell where the sea spike is located and the pulse block are... Each sampled complex value is denoted as a sea-peak sample; based on the sea-peak label value, any non-sea-peak pulse block within the same distance unit as the sea-peak is selected. Each sampled complex value is used as a sea clutter sample to form a set of sea spikes and sea clutter samples. The total number of samples in the sample set is denoted as S.

3. The method for identifying sea peaks based on multi-domain feature extraction and machine learning according to claim 2, characterized in that, In step 2: The sample from which the features are to be extracted is denoted as , , This represents the complex value of the nth sample; right Temporal features were extracted, and the seven extracted temporal features were: relative mean amplitude (RAA), normalized amplitude variance (NAV), normalized amplitude skewness (SK), normalized amplitude kurtosis (KU), amplitude distribution formal parameter (v), amplitude distribution scale parameter (b), and correlation time length (TCL). right Frequency domain features were extracted, and the four extracted frequency domain features were: Doppler offset DC, Doppler bandwidth DB, Doppler spectral entropy DSE, and Doppler peak power ratio PPR. right Time-frequency domain features were extracted, and the three extracted time-frequency domain features were: time-frequency ridge energy RI, number of connected regions RN, and maximum connected region size LRS; The 14 extracted features are combined into a 14-dimensional feature matrix and labeled, with sea spikes labeled as "1" and sea clutter labeled as "0", forming a feature dataset of sea spikes and sea clutter.

4. The method for identifying sea peaks based on multi-domain feature extraction and machine learning according to claim 3, characterized in that, In step 3: A strong classifier is formed by weighted combination of 200 decision trees. During training, each new decision tree uses the negative gradient direction of the current model's loss function as the fitting target, selects the optimal splitting feature and splitting point from the 14-dimensional features, and the maximum depth of a single tree does not exceed 6 layers. After the new tree is trained, it is added to the model with a weight of 0.

1. The model uses the log loss function as the optimization target and outputs the trained classifier after 200 iterations.

5. A peak recognition device based on multi-domain feature extraction and machine learning, characterized in that, include: A processor; a memory storing executable instructions of the processor; the processor being configured to perform the steps of the method of claim 1 by executing the executable instructions.

6. A computer-readable storage medium for storing a program, characterized in that: When the program is executed, it performs the steps of the method of claim 1.

Images