CN108875601A - Action identification method and LSTM neural network training method and relevant apparatus - Google Patents

Abstract

The present application discloses an action recognition method and its LSTM neural network training method, system and equipment, and a computer-readable storage medium. The LSTM neural network training method includes: adding to the forward propagation algorithm of the LSTM neural network A second derivative term, and update the backpropagation algorithm of the LSTM neural network according to the increased forward propagation algorithm to construct an improved LSTM neural network; wherein, the second derivative term is the second derivative of cell to time; Acquiring training samples, and training the improved LSTM neural network according to the training samples to obtain a trained improved LSTM neural network. The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow.

Description

Action recognition method, LSTM neural network training method and related device

technical field

The present application relates to the technical field of image processing, and more specifically, relates to an action recognition method and its LSTM neural network training method, system and equipment, and a computer-readable storage medium.

Background technique

In recent years, the research on human action recognition has attracted great attention from the industry, and it has important applications in the fields of video surveillance, games and robots. However, an efficient action recognition algorithm is very challenging: firstly, different moving speeds lead to fluctuations in time for the same action; secondly, many actions are similar, such as throwing high and waving; finally, different people are different in height, Differences in posture and other aspects can also cause difficulties in identification. In the prior art, the LSTM neural network is used to recognize the action sequence, and the recognition result will have the problem of time inaccuracy.

Therefore, how to save the time information of the recognition action sequence and avoid the time inaccuracy of the recognition result is a problem to be solved by those skilled in the art.

Contents of the invention

The purpose of this application is to provide an action recognition method and its LSTM neural network training method, system and equipment, and a computer-readable storage medium, which can save the time information of the recognition action sequence and avoid the time inaccuracy of the recognition result.

In order to achieve the above purpose, the application provides a LSTM neural network training method, including:

In the forward propagation algorithm of the LSTM neural network, increase the secondary derivative term, and update the backpropagation algorithm of the LSTM neural network according to the forward propagation algorithm after the increase, to build an improved LSTM neural network; wherein, the secondary The derivative term is the second derivative of the cell with respect to time;

Acquiring training samples, and training the improved LSTM neural network according to the training samples to obtain a trained improved LSTM neural network.

Among them, also include:

Obtain a test sample, and input the test sample into the improved LSTM neural network that has been trained to obtain an action sequence recognition result;

Calculate the average recognition rate of the test sample according to the recognition rate of each frame image in the test sample.

Wherein, the acquisition of training samples includes:

Acquire raw image data, and perform a preprocessing operation on the original image to obtain the training sample; wherein, the preprocessing operation includes any one or a combination of flipping, downsampling, or cutting operations.

Wherein, training the improved LSTM neural network according to the training samples includes:

Input each frame of image in the training sample into the improved LSTM neural network, and adjust the key parameters of the improved LSTM neural network until the recognition rate of the improved LSTM neural network output reaches a preset value, to obtain training Completed improved LSTM neural network.

Wherein, adjusting the key parameters of the improved LSTM neural network includes:

The key parameters of the improved LSTM neural network are adjusted by cross-validation method and pair-wise algorithm.

Wherein, the key parameters include any one or a combination of epoch, learning rate or learning rate decay.

In order to achieve the above purpose, the application provides an action recognition method, including:

Obtaining original image data, and performing a preprocessing operation on the original image to obtain a sample to be identified;

The sample to be recognized is input into the improved LSTM neural network trained as claimed in claim 1 to obtain the action recognition result.

To achieve the above object, the application provides a LSTM neural network training system, including:

Building block, for increasing the secondary derivative term in the forward propagation algorithm of LSTM neural network, and update the backpropagation algorithm of described LSTM neural network according to the forward propagation algorithm after the increase, to construct improved LSTM neural network; Wherein , the second derivative item is the second derivative of cell with respect to time;

The training module is used to obtain training samples, and train the improved LSTM neural network according to the training samples, so as to obtain the trained improved LSTM neural network.

To achieve the above purpose, the application provides a LSTM neural network training device, including:

memory for storing computer programs;

The processor is configured to implement the steps of the above-mentioned LSTM neural network training method when executing the computer program.

To achieve the above object, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned LSTM neural network training method are implemented.

It can be seen from the above scheme that a kind of LSTM neural network training method provided by the present application includes: adding a secondary derivative term in the forward propagation algorithm of the LSTM neural network, and updating the LSTM neural network according to the increased forward propagation algorithm The backpropagation algorithm of the backpropagation algorithm, to construct the improved LSTM neural network; Wherein, the second-order derivative term is the second-order derivative of cell to time; Obtain training samples, and train the described improved LSTM neural network according to the training samples, to obtain Improved LSTM neural network after training.

The LSTM neural network training method provided by this application improves the original LSTM neural network, adds the second derivative term of the cell to time in the original forward propagation algorithm, and modifies the backward propagation algorithm correspondingly according to the improved forward propagation algorithm . The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow. The application also discloses an LSTM neural network training system and equipment, an action recognition method and a computer-readable storage medium, which can also achieve the above-mentioned technical effects.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present application. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is the flowchart of a kind of LSTM neural network training method disclosed in the embodiment of the application;

Fig. 2 is the structural diagram of the completed LSTM neural network of the training disclosed in the embodiment of the application;

Fig. 3 is the flow chart of another kind of LSTM neural network training method disclosed in the embodiment of the present application;

FIG. 4 is a flowchart of an action recognition method disclosed in an embodiment of the present application;

Fig. 5 is the structural diagram of a kind of LSTM neural network training system disclosed in the embodiment of the present application;

FIG. 6 is a structural diagram of an LSTM neural network training device disclosed in an embodiment of the present application;

FIG. 7 is a structural diagram of another LSTM neural network training device disclosed in the embodiment of the present application.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the application with reference to the drawings in the embodiments of the application. Apparently, the described embodiments are only some of the embodiments of the application, not all of them. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of this application.

The embodiment of the present application discloses a LSTM neural network training method, which improves the supervision efficiency of tourism behavior.

Referring to Fig. 1, a flow chart of a LSTM neural network training method disclosed in the embodiment of the present application, as shown in Fig. 1, includes:

S101: Add a secondary derivative term in the forward propagation algorithm of the LSTM neural network, and update the backpropagation algorithm of the LSTM neural network according to the increased forward propagation algorithm to construct an improved LSTM neural network; wherein, the The second derivative item is the second derivative of the cell with respect to time;

In the specific implementation, on the basis of the forward propagation algorithm of the original LSTM (full name in Chinese: long-term short-term memory network, full name in English: Network-LongShort Term Memory Network) neural network, as shown in Figure 2, the second level of cell versus time is added. Derivative term, from the improved forward propagation algorithm, those skilled in the art can derive the corresponding back propagation algorithm. Since there is a second derivative of the cell to time in the forward propagation algorithm and the back propagation algorithm, it can be well preserved The time information of the action sequence avoids the time inaccuracy of the recognition results.

S102: Obtain a training sample, and train the improved LSTM neural network according to the training sample, so as to obtain a trained improved LSTM neural network.

In the specific implementation, first obtain the original image data of the training sample, and perform a preprocessing operation on the original image to obtain the training sample. This embodiment does not limit the specific preprocessing operation, and those skilled in the art can Choose flexibly. As a preferred implementation manner, the preprocessing operation here includes any one or a combination of flipping, downsampling, or cutting operations. Wherein, the down-sampling operation is to sample a sequence of samples at intervals of several samples.

The training process of the above-mentioned improved LSTM neural network is specifically: input the training samples into the improved LSTM neural network, and adjust the key parameters of the improved LSTM neural network until the output recognition rate reaches a preset value, so as to obtain the improved LSTM after training. Neural Networks.

As a preferred implementation manner, key parameters can be adjusted using a cross-validation method and a pair-wise algorithm. The key parameters mentioned here can include Batch_video (the number of videos input each time), Batch_frame (the number of video frames contained in each video), epoch (the number of times to train all data once), learning rate or learning rate decay, etc. The initial values of the above key parameters are not specifically limited here, and those skilled in the art can flexibly set them according to the actual situation, for example, Batch_video=6, Batch_frame=24, epoch=5000-8000, learning rate Learning_rate=0.1, learning rate attenuation lr_decay = 0.1/1000 times.

The LSTM neural network training method provided by the embodiment of the present application improves the original LSTM neural network, adds the second-order derivative term of the cell to time in the original forward propagation algorithm, and correspondingly modifies the backward direction according to the improved forward propagation algorithm. propagation algorithm. The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow.

The embodiment of the present application discloses a LSTM neural network training method. Compared with the previous embodiment, this embodiment further explains and optimizes the technical solution. specific:

Referring to FIG. 3, a flow chart of another LSTM neural network training method provided in the embodiment of the present application, as shown in FIG. 2, includes:

S301: Add a secondary derivative term in the forward propagation algorithm of the LSTM neural network, and update the back propagation algorithm of the LSTM neural network according to the increased forward propagation algorithm, so as to construct an improved LSTM neural network; wherein, the The second derivative item is the second derivative of the cell with respect to time;

S302: Obtain a training sample, and train the improved LSTM neural network according to the training sample to obtain a trained improved LSTM neural network;

S303: Obtain a test sample, and input the test sample into the trained improved LSTM neural network to obtain an action sequence recognition result;

S304: Calculate the average recognition rate of the test sample according to the recognition rate of each frame of image in the test sample.

It can be understood that after the training of the above LSTM neural network is completed, the test sample can also be used to test the trained LSTM neural network. Specifically, all the image frames of the test sample are input into the trained improved LSTM neural network to obtain Action sequence recognition results, and calculate the average recognition rate of all image frames in the test sample to obtain the action recognition accuracy rate of the LSTM neural network. It should be noted that each frame of image in the test sample here has undergone preprocessing operations, namely flipping operations, downsampling operations, and cutting operations.

The following describes an action recognition method provided in this embodiment, which uses the improved LSTM neural network trained in the above embodiment. specific:

Referring to FIG. 4, a flow chart of an action recognition method disclosed in the embodiment of the present application, as shown in FIG. 4, includes:

S401: Obtain original image data, and perform a preprocessing operation on the original image to obtain a sample to be identified;

In the specific implementation, after obtaining the original image data, it is necessary to perform preprocessing operations on the original image data, that is, after the enhancement operation, the samples to be recognized are obtained. Similarly, the preprocessing operations here may include flipping operations, downsampling operations, and cutting operations, etc. .

S402: Input the sample to be recognized into the trained improved LSTM neural network provided by the above embodiment, to obtain an action recognition result.

In a specific implementation, the trained improved LSTM neural network provided in the above embodiment for each frame of image in the sample to be recognized is used to obtain the action sequence recognition result of the sample to be recognized. It can be understood that the action sequence recognition result obtained in this step includes not only the action sequence at the recognition point, but also the recognition rate, that is, the recognition rate of each frame of image is calculated, and the recognition sample is calculated according to the recognition rate of each frame of image average recognition rate.

An LSTM neural network training system provided in the embodiment of the present application is introduced below. The LSTM neural network training system described below and the LSTM neural network training method described above can be referred to each other.

Referring to FIG. 5, a structural diagram of an LSTM neural network training system provided in the embodiment of the present application, as shown in FIG. 5, includes:

Construction module 501, for increasing the secondary derivative term in the forward propagation algorithm of the LSTM neural network, and updating the backpropagation algorithm of the LSTM neural network according to the increased forward propagation algorithm, to construct an improved LSTM neural network; Wherein, the second derivative item is the second derivative of cell to time;

The training module 502 is configured to obtain training samples, and train the improved LSTM neural network according to the training samples, so as to obtain a trained improved LSTM neural network.

The LSTM neural network training system provided by the embodiment of the present application improves the original LSTM neural network, adds the second-order derivative term of the cell to time in the original forward propagation algorithm, and correspondingly modifies the backward direction according to the improved forward propagation algorithm. propagation algorithm. The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow.

On the basis of the foregoing embodiments, as a preferred implementation manner, it also includes:

Obtain a test sample, and input the test sample into the improved LSTM neural network that has been trained to obtain an action sequence recognition result;

Calculate the average recognition rate of the test sample according to the recognition rate of each frame image in the test sample.

On the basis of the above-mentioned embodiments, as a preferred implementation manner, the training module 502 includes:

An acquisition unit, configured to acquire original image data, and perform a preprocessing operation on the original image to obtain the training sample; wherein, the preprocessing operation includes any one or more of a flipping operation, a downsampling operation, or a cutting operation combination of items;

A training unit, configured to train the improved LSTM neural network according to the training samples, so as to obtain a trained improved LSTM neural network.

On the basis of the above-mentioned embodiments, as a preferred implementation manner, the training unit specifically inputs each frame of image in the training sample into the improved LSTM neural network, and adjusts the Key parameters until the recognition rate output by the improved LSTM neural network reaches a preset value, so as to obtain the trained improved LSTM neural network unit.

On the basis of the above embodiments, as a preferred implementation, the training unit specifically inputs each frame of image in the training sample into the improved LSTM neural network, and uses the cross-validation method and pair-wise The algorithm adjusts the unit of key parameters of the improved LSTM neural network.

On the basis of the foregoing embodiments, as a preferred implementation manner, the key parameters include any one or a combination of epoch, learning rate, or learning rate decay.

The present application also provides an LSTM neural network training device, see FIG. 6, a structural diagram of an LSTM neural network training device provided in an embodiment of the present application, as shown in FIG. 6, including:

memory 100, for storing computer programs;

The processor 200 is configured to implement the steps provided in the foregoing embodiments when executing the computer program.

Specifically, the memory 100 includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer-readable instructions, and the internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The processor 200 provides computing and control capabilities for the LSTM neural network training equipment. When executing the computer program stored in the memory 100, the following steps can be implemented: adding a second derivative term in the forward propagation algorithm of the LSTM neural network, and according to The forward propagation algorithm after the increase updates the backpropagation algorithm of the LSTM neural network to construct an improved LSTM neural network; wherein, the second-order derivative item is the second-order derivative of the cell to time; obtain training samples, and according to the The training samples are used to train the improved LSTM neural network to obtain the trained improved LSTM neural network.

The embodiment of the present application improves the original LSTM neural network, adds the second derivative term of cell to time in the original forward propagation algorithm, and modifies the backward propagation algorithm correspondingly according to the improved forward propagation algorithm. The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow.

Preferably, when the processor 200 executes the computer subroutine stored in the memory 100, the following steps can be implemented: obtain a test sample, and input the test sample into the trained improved LSTM neural network to obtain an action sequence recognition Result: calculating the average recognition rate of the test sample according to the recognition rate of each frame image in the test sample.

Preferably, when the processor 200 executes the computer subroutine stored in the memory 100, the following steps can be implemented: acquiring original image data, and performing a preprocessing operation on the original image to obtain the training sample; wherein, the The above preprocessing operations include any one or a combination of flipping operations, downsampling operations, or cutting operations.

Preferably, when the processor 200 executes the computer subroutine stored in the memory 100, the following steps can be implemented: input each frame of image in the training sample into the improved LSTM neural network, and adjust the The key parameters of the LSTM neural network are improved until the recognition rate output by the improved LSTM neural network reaches a preset value, so as to obtain a trained improved LSTM neural network.

Preferably, when the processor 200 executes the computer subroutine stored in the memory 100, the following steps can be implemented: using the cross-validation method and pair-wise algorithm to adjust the key parameters of the improved LSTM neural network.

On the basis of the foregoing embodiments, as a preferred embodiment, referring to Fig. 7, the LSTM neural network training device also includes:

The input interface 300 is connected with the processor 200 , and is used for acquiring externally imported computer programs, parameters and instructions, and storing them in the memory 100 under the control of the processor 200 . The input interface 300 can be connected with an input device to receive parameters or instructions manually input by a user. The input device may be a touch layer covered on the display screen, or may be a button, a trackball or a touch pad provided on the terminal shell, or may be a keyboard, a touch pad, or a mouse.

The display unit 400 is connected to the processor 200 and used for displaying data sent by the processor 200 . The display unit 400 may be a display screen on a PC, a liquid crystal display screen, or an electronic ink display screen. Specifically, in this embodiment, the action sequence recognition result of the sample to be recognized may be displayed through the display unit 400 .

The network port 500 is connected with the processor 200 and used for communicating with various external terminal devices. The communication technology used in the communication connection can be wired communication technology or wireless communication technology, such as mobile high-definition link technology (MHL), universal serial bus (USB), high-definition multimedia interface (HDMI), wireless fidelity technology (WiFi), Bluetooth communication technology, low-power Bluetooth communication technology, communication technology based on IEEE802.11s, etc. Specifically, in this embodiment, the original LSTM neural network model and the like may be imported to the processor 200 through the network port 500 .

The video collector 600 is connected to the processor 200 for acquiring video data, and then sends the video data to the processor 200 for data analysis and processing, and the subsequent processor 200 can send the processing result to the display unit 400 for display, or transmit it to The processor 100 saves it, or can send it to a preset data receiving terminal through the network port 500 . Specifically, in this embodiment, the video collector 600 may acquire samples to be identified, training samples, test samples, and the like.

The present application also provides a computer-readable storage medium, which may include: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic Various media that can store program codes such as discs or optical discs. A computer program is stored on the storage medium, and the following steps are implemented when the computer program is executed by a processor: adding a second-order derivative term in the forward propagation algorithm of the LSTM neural network, and updating the described forward propagation algorithm according to the increase The backpropagation algorithm of LSTM neural network, to build improved LSTM neural network; Wherein, described secondary derivative item is the secondary derivative of cell to time; Obtain training sample, and train described improved LSTM neural network according to described training sample , to obtain the improved LSTM neural network that has been trained.

The embodiment of the present application improves the original LSTM neural network, adds the second derivative term of cell to time in the original forward propagation algorithm, and modifies the backward propagation algorithm correspondingly according to the improved forward propagation algorithm. The improved LSTM neural network is used to identify the action sequence. Since the second derivative of the cell to time exists in the forward propagation algorithm and the back propagation algorithm, the time information of the action sequence can be well preserved and the time loss of the recognition result can be avoided. allow.

Preferably, when the computer subroutine stored in the computer-readable storage medium is executed by the processor, the following steps can be implemented specifically: obtaining a test sample, and inputting the test sample into the trained improved LSTM neural network to obtain an action Sequence recognition result: calculating the average recognition rate of the test sample according to the recognition rate of each frame image in the test sample.

Preferably, when the computer subroutine stored in the computer-readable storage medium is executed by the processor, the following steps can be implemented specifically: acquiring original image data, and performing a preprocessing operation on the original image to obtain the training sample; wherein , the preprocessing operation includes any one or a combination of a flipping operation, a downsampling operation, or a cutting operation.

Preferably, when the computer subroutine stored in the computer-readable storage medium is executed by the processor, the following steps can be specifically implemented: input each frame of image in the training sample into the improved LSTM neural network, and adjust Key parameters of the improved LSTM neural network until the recognition rate output by the improved LSTM neural network reaches a preset value, so as to obtain a trained improved LSTM neural network.

Preferably, when the computer subroutine stored in the computer-readable storage medium is executed by the processor, the following steps can be specifically implemented: using the cross-validation method and pair-wise algorithm to adjust the key parameters of the improved LSTM neural network.

Each embodiment in this specification is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other.

The above description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Each embodiment in the description is described in a progressive manner, each embodiment focuses on the difference from other embodiments, and the same and similar parts of each embodiment can be referred to each other. As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and for the related information, please refer to the description of the method part. It should be pointed out that those skilled in the art can make some improvements and modifications to the application without departing from the principles of the application, and these improvements and modifications also fall within the protection scope of the claims of the application.

It should also be noted that in this specification, relative terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that these entities or operations There is no such actual relationship or order between the operations. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims (10)

1. a kind of LSTM neural network training method, which is characterized in that including：

Increase second level derivative term in the propagated forward algorithm of LSTM neural network, and more according to the propagated forward algorithm after increase The back-propagation algorithm of the new LSTM neural network improves LSTM neural network with building；Wherein, the second level derivative term is Second level derivative of the cell to the time；

Training sample is obtained, and according to the training sample training improvement LSTM neural network, to obtain training completion Improve LSTM neural network.

2. LSTM neural network training method according to claim 1, which is characterized in that further include：

Test sample is obtained, and the test sample is inputted in the improvement LSTM neural network that training is completed, obtains movement sequence Column recognition result；

The average recognition rate of the test sample is calculated according to the discrimination of frame image each in the test sample.

3. LSTM neural network training method according to claim 1, which is characterized in that the acquisition training sample, including：

Raw image data is obtained, and pretreatment operation is carried out to the original image and obtains the training sample；Wherein, described Pretreatment operation includes the combination of any one of turning operation, down-sampling operation or cutting operation or several.

4. any one of -3 LSTM neural network training method according to claim 1, which is characterized in that according to the trained sample This training improvement LSTM neural network, including：

Each frame image in the training sample is inputted in the improvement LSTM neural network, and adjusts the improvement LSTM The key parameter of neural network is until the discrimination of the improvement LSTM neural network output reaches preset value, to obtain having trained At improvement LSTM neural network.

5. LSTM neural network training method according to claim 4, which is characterized in that adjust the improvement LSTM nerve net The key parameter of network, including：

The key parameter for improving LSTM neural network is adjusted using cross validation method and pair-wise algorithm.

6. LSTM neural network training method according to claim 5, which is characterized in that the key parameter include epoch, Learning rate or any one or several combinations of learning rate decaying.

7. a kind of action identification method, which is characterized in that including：

Raw image data is obtained, and pretreatment operation is carried out to the original image and obtains sample to be identified；

By the sample input improvement LSTM neural network that training is completed as described in claim 1 to be identified, obtains movement and know Other result.

8. a kind of LSTM neural metwork training system, which is characterized in that including：

Module is constructed, for increasing second level derivative term in the propagated forward algorithm of LSTM nerve net, and according to before after increase The back-propagation algorithm of the LSTM neural network is updated to propagation algorithm, and LSTM neural network is improved with building；Wherein, described Second level derivative term is second level derivative of the cell to the time；

Training module, for obtaining training sample, and according to the training sample training improvement LSTM neural network, to obtain The improvement LSTM neural network completed to training.

9. a kind of LSTM neural metwork training equipment, which is characterized in that including：

Memory, for storing computer program；

Processor realizes that LSTM neural network is instructed as described in any one of claim 1 to 6 when for executing the computer program The step of practicing method.

10. a kind of computer readable storage medium, which is characterized in that be stored with computer on the computer readable storage medium Program realizes the LSTM neural metwork training as described in any one of claim 1 to 6 when the computer program is executed by processor The step of method.