Time series classification method, classification terminal, and storage medium

Abstract

The application provides a time sequence classification method, a classification terminal and a storage medium. The method comprises the following steps: preprocessing initial time sequence data, and extracting features of the preprocessed initial time sequence data by using an FCN network to obtain initial features; extracting the initial features by using an IAM network and an AGRes2net network to obtain global features; and classifying based on the global features to obtain a target classification result. The IAM network and the AGRes2net network are combined, multi-scale features can be extracted, and the classification result is more accurate.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a time series classification method, classification terminal, and storage medium. Background Technology

[0002] In various scenarios such as traffic management, air pollution control, healthcare, power load, and securities trading, accurate classification of time series data is often required for further analysis. For example, in the power sector, analyzing daily load curve time series can predict future load curve trends. How to accurately classify time series data across different sectors has become a challenging and crucial issue in current research.

[0003] In existing technologies, deep learning models are commonly used to classify time series data. However, deep learning models extract only a single feature, resulting in inaccurate classification and affecting subsequent data analysis. Summary of the Invention

[0004] This invention provides a time series classification method, classification terminal, and storage medium to solve the problems of single feature extraction and inaccurate classification in existing deep learning models.

[0005] In a first aspect, embodiments of the present invention provide a method for classifying time series data, comprising:

[0006] The initial time series data is preprocessed, and the FCN network is used to extract features from the preprocessed initial time series data to obtain initial features;

[0007] The IAM network and AGRes2net network are used to extract global features from the initial features;

[0008] Classification is performed based on global features to obtain the target classification result.

[0009] In a second aspect, embodiments of the present invention provide a control terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the time series classification method provided in the first aspect or any possible implementation of the first aspect.

[0010] Thirdly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the time series classification method provided in the first aspect or any possible implementation thereof.

[0011] This invention provides a time series classification method, a classification terminal, and a storage medium. The method includes: preprocessing initial time series data and extracting features from the preprocessed initial time series data using an FCN network to obtain initial features; using an IAM network and an AGRES2net network to extract global features from the initial features; and classifying based on the global features to obtain the target classification result. This invention combines the IAM network and the AGRES2net network, integrating their advantages to extract multi-scale features, resulting in more accurate classification results and facilitating subsequent data analysis. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the implementation of a time series classification method provided in an embodiment of the present invention.

[0014] Figure 2 This is a structural diagram of the AGRes2Net network provided in an embodiment of the present invention;

[0015] Figure 3 This is a structural diagram of the FCN network provided in an embodiment of the present invention;

[0016] Figure 4 This is a structural diagram of the IAM network provided in an embodiment of the present invention;

[0017] Figure 5 This is a flowchart illustrating the implementation of another time series classification method provided in this embodiment of the invention.

[0018] Figure 6 This is a schematic diagram of the structure of the time series classification device provided in an embodiment of the present invention;

[0019] Figure 7 This is a schematic diagram of the classification terminal provided in an embodiment of the present invention. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, specific embodiments will be described below in conjunction with the accompanying drawings.

[0022] See Figure 1 The flowchart illustrating the implementation of the mixed time series classification method provided in this embodiment of the invention is described in detail below:

[0023] S101: Preprocess the initial time series data and use the FCN network to extract features from the preprocessed initial time series data to obtain initial features;

[0024] S102: Using IAM network and AGRes2net network, feature extraction is performed on the initial features to obtain global features;

[0025] S103: Classify based on global features to obtain the target classification result.

[0026] In one possible implementation, S102 may include:

[0027] S1021: Input the initial features into the AGRES2net network and extract the output of each residual block in the AGRES2net network;

[0028] S1022: Input the output of each residual block into the IAM network, and use the IAM network to learn the correlation between the outputs of each residual block, and output global features.

[0029] The AGRes2net network (Attentional Gated Res2Net) can extract global and local temporal features, overcoming the shortcomings of the standard CNN architecture and enabling multi-scale feature extraction.

[0030] AGRes2net consists of N AGRes2net residual blocks. The AGRes2Net network is a variant of the residual network, and the structure of an AGRes2Net residual block is as follows: Figure 2 As shown, the residual block of AGRES2net consists of two stages: a convolution stage and an attention stage.

[0031] 1. Convolution stage

[0032] The convolution stage utilizes hierarchical residual convolutional filters to connect smaller convolutional operators, enabling the output to contain different numbers and combinations of feature scales, thus achieving multi-scale feature extraction. Simultaneously, due to the use of a parallel branch network structure, Res2Net has fewer parameters than ResNet, reducing both parameter count and runtime. Furthermore, a gating mechanism is used between different filter groups to capture long-term dependencies between time series data.

[0033] The AGRes2Net convolutional stage inputs the initial time series data into a 1*1 convolutional layer for channel expansion to obtain the input feature map x, and then segments x into (x1, x2, ..., x...). s ), where s is the number of filter groups. The convolution stage obtains the output feature map y. i The calculation formula is as follows:

[0034]

[0035] Among them, C i () represents the i-th 3x3 convolutional module, which consists of a convolutional layer (CONV), batch normalization (BN), and the ReLU activation function. i Let y be the i-th input feature map. i-1 For the (i-1)th output feature map, g i Let i be the i-th gate value. The formula for calculating the gate value is as follows:

[0036] g i =tanh(C(concat(C(y)) i-1 ),C(x i ))))

[0037] Here, concat represents the connection operation, and tanh is the tangent activation function.

[0038] 2. Attention Stage

[0039] The attention phase employs a channel attention mechanism, which can capture temporal features and dependencies at different levels. The feature map contains temporal information within different ranges, and the feature map of each channel is updated according to the relationship between each channel and the feature map. The calculation formula is as follows, representing the updated features based on the attention mechanism:

[0040]

[0041] Among them, F j For the input y i Features, M c (F j ) represents the weight coefficient of the corresponding feature channel. This represents the product operation, where j = 1, 2, ..., s.

[0042] M c (F j The calculation formula for ) is as follows:

[0043]

[0044] Where δ() is the sigmoid function, This represents summing the features of different pieces of information;

[0045] The calculation formula is as follows:

[0046]

[0047] Here, FC represents a fully connected layer. The difference between FC1 and FC2 is that FC1 contains the ReLU activation function. Indicates F j Perform average pooling. Indicates F j Perform maximum pooling.

[0048] F channel For multi-scale features, the calculation formula is as follows:

[0049]

[0050] The AGRes2net network can perform multi-scale feature extraction, but the extraction process does not utilize the interdependencies between different AGRes2net residual blocks, making it prone to losing some global feature information. In contrast, IMA (Inter-Module Adaptive) can learn the correlations between different AGRes2net residual blocks, calculate the correlation index, and adaptively adjust the feature mappings between different AGRes2net residual blocks to extract global features, thereby improving feature representation capabilities and compensating for feature loss.

[0051] This invention combines the advantages of IAM and AGRes2net networks for feature extraction. While performing multi-scale feature extraction, it also improves the expressive power of features, compensates for possible feature loss, and makes the extracted features more comprehensive and accurate. This effectively improves the classification accuracy of time series data and provides a reliable basis for subsequent data analysis.

[0052] Since the initial feature extraction accuracy does not need to be too high, the FCN network (Fully Convolutional Network) is used for initial feature extraction in this embodiment of the invention. The FCN network has a relatively low feature extraction accuracy, but it is fast and suitable for the scenario of this solution.

[0053] Figure 3 A schematic diagram of the FCN network structure is shown. In this application, initial time series data is input into each convolutional block of the FCN network. For example, the number of convolutional blocks can be set to 3, and the structure of each of the 3 convolutional blocks consists of a convolutional layer (CONV), batch normalization (BN), and the ReLU activation function, respectively. Batch normalization (BN) can effectively reduce gradient vanishing, accelerate convergence, and prevent overfitting.

[0054] In one possible implementation, S1022 may include:

[0055] 1. Use the IAM network to perform dimensionality transformation on the output of each residual block to obtain the feature reshaping matrix;

[0056] 2. Based on the feature reshaping matrix, obtain the correlation coefficients between each residual block;

[0057] 3. Based on the feature reshaping matrix and the correlation coefficients between each residual block, a two-dimensional initial matrix is ​​obtained;

[0058] 4. Reshape the initial two-dimensional matrix to construct the global features.

[0059] Figure 4 The diagram shows the structure of the IAM network. The IAM network takes the outputs of N residual blocks as input, performs dimensionality transformation, and reshapes them into a feature reshaping matrix M. Then, it uses the first formula to calculate the correlation coefficient W between the elements in the feature reshaping matrix M. m，n .

[0060] The first formula can be shown below:

[0061]

[0062] Among them, W m,n Let M be the correlation coefficient between the m-th residual block and the n-th residual block, and M be the feature reshaping matrix. T This is the transpose of the eigenvalue reshaping matrix; The function is a sigmoid function, where m, n = 1, 2, ..., N, and N is the number of residual blocks in the AGRES2net network.

[0063] The two-dimensional initial matrix FM is obtained by using the feature reshaping matrix and the correlation coefficients between the elements in the matrix through the second formula.

[0064] The second formula can be shown below:

[0065]

[0066] Among them, W m,nLet M be the correlation coefficient between the m-th residual block and the n-th residual block. n,m is the nth row and mth column of the feature reshaping matrix; M is the feature reshaping matrix.

[0067] Finally, the initial two-dimensional matrix FM is reshaped to construct the global feature F. global The calculation formula is as follows:

[0068] F global =H CONCV (Re(FM))

[0069] Among them, H CONCV () represents a one-dimensional convolution operation, and Re represents a reshaping operation.

[0070] In one possible implementation, S102 may further include:

[0071] S1023: Input the initial features into the AGRes2net network and output multi-scale features;

[0072] S103 may include:

[0073] S1031: Concatenate multi-scale features with global features to obtain concatenated features;

[0074] S1032: Input the spliced ​​features into the global average pooling layer to adjust the dimensionality information and obtain the adjusted spliced ​​features;

[0075] S1033: Classify based on the adjusted splicing features to obtain the target classification result.

[0076] In this embodiment of the invention, the classification results are more accurate by combining the multi-scale features extracted by the AGRes2net network and the global features adjusted based on the dependency relationship between each residual block.

[0077] In one possible implementation, S1033 may include:

[0078] 1. Based on the adjusted concatenated features, the softmax function is used for classification to obtain the target classification result.

[0079] Specifically, the formula for calculating the probability of each classification result when using the softmax function is as follows:

[0080] Y = softmax(con(F) channel ,F gl o bal ))

[0081] In one possible implementation, S101 may include:

[0082] S1011: Perform convolutional pooling on the initial time series data to obtain the first time series data;

[0083] S1012: Sample the initial time series data and perform convolutional pooling on the sampled data to obtain the second time series data;

[0084] S1013: Filter the initial time series data and perform convolutional pooling on the filtered data to obtain the third time series data;

[0085] S1014: The first time series data, the second time series data, and the third time series data are concatenated to obtain the preprocessed initial time series data.

[0086] To preserve the authenticity of the data, the preprocessing stage directly processes the initial time series to obtain a time series dataset. The formula for the first time series data T1 is shown below:

[0087] T1 = {t1, t2, ..., t} i ,…t n}

[0088] Among them, t i Let n be the data at time i, and n be the dimension of the first time series data.

[0089] A single time series dataset contains information of varying dimensions and time scales. Long-term characteristics of a time series reflect the overall picture, while short-term characteristics reflect local changes. In time series data, the data features within a specific time period are most representative. For example, in electricity load data, electricity load measurements are taken over a 24-hour day, yielding data from 24 locations. In daily life, the main features of electricity data are often contained between 8 and 22 points. Therefore, the series within this time period can be used as the main data, removing features with weaker representation. Based on this, this application obtains a second time series dataset by sampling the data, selecting multiple time scales, and performing convolutional pooling. The calculation formula for the second time series dataset is shown below:

[0090]

[0091] Where k = 1, 2, ..., n.

[0092] To remove noise from time series data, mean smoothing filtering is used. This effectively suppresses and prevents interference signals by filtering out specific frequency bands in the signal. Given input time series data, new time series with different smoothness are generated using moving averages with different window sizes. For a filter with a window size of l, the time series data from the i-th data point to the (i+l-1)-th data point is filtered using the following formula:

[0093]

[0094] Where l represents the filter window size, t i This represents the data at time i. After filtering, it undergoes convolutional pooling to obtain...

[0095] Finally, the three time series data were concatenated, and the preprocessing was completed.

[0096] Figure 5 A flowchart illustrating a time series classification method is shown, explaining the specific implementation process of the above method, which will not be elaborated further here.

[0097] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0098] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0099] Figure 6 A schematic diagram of the time series classification device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below:

[0100] like Figure 6 As shown, the time series classification device includes:

[0101] The first feature extraction module 21 is used to preprocess the initial time series data and use the FCN network to extract features from the preprocessed initial time series data to obtain initial features;

[0102] The second feature extraction module 22 is used to extract global features from the initial features using the IAM network and the AGRes2net network.

[0103] Classification module 23 is used to perform classification based on global features to obtain the target classification result.

[0104] In one possible implementation, the second feature extraction module 22 may include:

[0105] The residual output unit is used to input the initial features into the AGRES2net network and extract the output of each residual block in the AGRES2net network.

[0106] The correlation learning unit is used to input the output of each residual block into the IAM network and use the IAM network to learn the correlation between the outputs of each residual block, and output global features.

[0107] In one possible implementation, the relevance learning unit may include:

[0108] The dimension transformation subunit is used to perform dimension transformation on the output of each residual block using the IAM network to obtain the feature reshaping matrix.

[0109] The correlation coefficient extraction subunit is used to reshape the matrix based on features to obtain the correlation coefficients between each residual block;

[0110] The matrix construction sub-unit is used to obtain a two-dimensional initial matrix based on the feature reshaping matrix and the correlation coefficients between each residual block;

[0111] The reshaping sub-unit is used to reshape the two-dimensional initial matrix to construct the global features.

[0112] In one possible implementation, the correlation coefficient extraction subunit can be specifically used for:

[0113] Based on the feature reshaping matrix, the correlation coefficients between each residual block are calculated using the first formula.

[0114] The first formula can be:

[0115]

[0116] Among them, W m,n Let M be the correlation coefficient between the m-th residual block and the n-th residual block, and M be the feature reshaping matrix. T This is the transpose of the eigenvalue reshaping matrix; The function is a sigmoid function, where m, n = 1, 2, ..., N, and N is the number of residual blocks in the AGRES2net network.

[0117] In one possible implementation, the matrix construction subunit can be specifically used for:

[0118] Based on the feature reshaping matrix and the correlation coefficients between each residual block, the two-dimensional initial matrix is ​​calculated using the second formula.

[0119] The second formula can be:

[0120]

[0121] Among them, W m,n Let M be the correlation coefficient between the m-th residual block and the n-th residual block. n,m is the nth row and mth column of the feature reshaping matrix; M is the feature reshaping matrix.

[0122] In one possible implementation, the second feature extraction module 22 may further include:

[0123] The multi-scale feature extraction unit is used to input the initial features into the AGRes2net network and output multi-scale features.

[0124] Classification module 23 may include:

[0125] The first concatenation unit is used to concatenate multi-scale features with global features to obtain concatenated features;

[0126] The dimension adjustment unit is used to input the concatenated features into the global average pooling layer to adjust the dimension information and obtain the adjusted concatenated features.

[0127] The classification output unit is used to classify based on the adjusted concatenated features to obtain the target classification result.

[0128] In one possible implementation, the classification output unit can be specifically used to: classify based on the adjusted concatenated features using the softmax function to obtain the target classification result.

[0129] In one possible implementation, the first feature extraction module 21 may include:

[0130] The first preprocessing unit is used to perform convolutional pooling on the initial time series data to obtain the first time series data.

[0131] The second preprocessing unit is used to sample the initial time series data and perform convolutional pooling on the sampled data to obtain the second time series data.

[0132] The third preprocessing unit is used to filter the initial time series data and perform convolutional pooling on the filtered data to obtain the third time series data.

[0133] The second splicing unit is used to splice the first time series data, the second time series data, and the third time series data to obtain the preprocessed initial time series data.

[0134] Figure 7 This is a schematic diagram of the classification terminal provided in an embodiment of the present invention. Figure 7As shown, the classification terminal 3 in this embodiment includes a processor 30 and a memory 31. The memory 31 stores a computer program 32, and the processor 30 calls and runs the computer program 32 stored in the memory 31 to execute the steps in the various time series classification method embodiments described above, for example... Figure 1 The steps S101 to S103 are shown. Alternatively, the processor 30 is used to call and run the computer program 32 stored in the memory 31 to implement the functions of each module / unit in the above-described device embodiments, for example... Figure 6 The functions of modules 21 to 23 are shown.

[0135] For example, computer program 32 can be divided into one or more modules / units, one or more of which are stored in memory 31 and executed by processor 30 to complete the present invention. One or more modules / units can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 32 in the classified terminal 3. For example, computer program 32 can be divided into... Figure 6 Modules / units 21 to 23 are shown.

[0136] The classification terminal 3 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The classification terminal 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art will understand that... Figure 7 This is merely an example of the classification terminal 3 and does not constitute a limitation on the classification terminal 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, the terminal may also include input / output devices, network access devices, buses, etc.

[0137] The processor 30 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0138] The memory 31 can be an internal storage unit of the sorting terminal 3, such as a hard disk or RAM in the sorting terminal 3. The memory 31 can also be an external storage device of the sorting terminal 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the sorting terminal 3. Furthermore, the memory 31 can include both internal storage units and external storage devices of the sorting terminal 3. The memory 31 is used to store computer programs and other programs and data required by the terminal. The memory 31 can also be used to temporarily store data that has been output or will be output.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0140] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0141] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0142] In the embodiments provided by this invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0143] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0145] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0146] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for classifying time series data, characterized in that, include: The initial time series data is preprocessed, and the FCN network is used to extract features from the preprocessed initial time series data to obtain initial features; The IAM network and AGRes2net network are used to extract features from the initial features to obtain global features; Classification is performed based on the global features to obtain the target classification result; The preprocessing of the initial time series data includes: The initial time series data is subjected to convolutional pooling to obtain the first time series data; The initial time series data is sampled, and the sampled data is subjected to convolutional pooling to obtain the second time series data; The initial time series data is filtered, and the filtered data is then subjected to convolutional pooling to obtain the third time series data. The first time series data, the second time series data, and the third time series data are concatenated to obtain the preprocessed initial time series data. The process employs IAM and AGRes2net networks to extract global features from the initial features, including: The initial features are input into the AGRES2net network, and the outputs of each residual block in the AGRES2net network are extracted. The outputs of each residual block are input into the IAM network, and the IAM network is used to learn the correlation between the outputs of each residual block to output the global features.

2. The time series classification method according to claim 1, characterized in that, The process of inputting the outputs of each residual block into the IAM network, and using the IAM network to learn the correlation between the outputs of each residual block, and outputting the global features, includes: The IAM network is used to perform dimensionality transformation on the output of each residual block to obtain the feature reshaping matrix; Based on the feature reshaping matrix, the correlation coefficients between each residual block are obtained; Based on the feature reshaping matrix and the correlation coefficients between the residual blocks, a two-dimensional initial matrix is ​​obtained; The global feature is obtained by reshaping the two-dimensional initial matrix.

3. The time series classification method according to claim 2, characterized in that, The step of obtaining the correlation coefficients between each residual block based on the feature reshaping matrix includes: Based on the feature reshaping matrix, the correlation coefficients between the residual blocks are calculated using the first formula. The first formula is: in, For the first The residual block and the first The correlation coefficient between the residual blocks For the feature reshaping matrix, This is the transpose of the feature reshaping matrix; For the sigmoid function, , The number of residual blocks in the AGRes2net network.

4. The time series classification method according to claim 2, characterized in that, The two-dimensional initial matrix is ​​obtained based on the feature reshaping matrix and the correlation coefficients between the residual blocks, including: Based on the feature reshaping matrix and the correlation coefficients between the residual blocks, the two-dimensional initial matrix is ​​calculated using the second formula. The second formula is: in, For the first The residual block and the first The correlation coefficient between the residual blocks The first feature reshaping matrix Line 1 List; The feature reshaping matrix is ​​given.

5. The time series classification method according to claim 1, characterized in that, The step of using IAM network and AGRes2net network to extract features from the initial features to obtain global features also includes: The initial features are input into the AGRes2net network, which outputs multi-scale features. The classification based on the global features to obtain the target classification result includes: The multi-scale features are concatenated with the global features to obtain the concatenated features; The spliced ​​features are input into a global average pooling layer to adjust the dimensionality information, resulting in the adjusted spliced ​​features. The target classification result is obtained by classifying based on the adjusted splicing features.

6. The time series classification method according to claim 5, characterized in that, The classification based on the adjusted splicing features to obtain the target classification result includes: Based on the adjusted splicing features, the softmax function is used for classification to obtain the target classification result.

7. A classification terminal, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to call and run the computer program stored in the memory to perform the steps of the time series classification method as described in any one of claims 1 to 6.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the time series classification method as described in any one of claims 1 to 6.

Images