Encoding method, decoding method, encoding device, decoding device, and electronic device

Abstract

The present disclosure relates to the technical field of communication, and particularly relates to an encoding method, a decoding method, an encoding device, a decoding device and an electronic device. The method comprises the following steps: a sending end extracts semantic encoding information and key feature encoding information from first service information through semantic extraction; based on the type of the semantic encoding information and the current channel transmission environment, a corresponding source channel joint encoding model is selected, the source channel joint encoding model encodes the semantic encoding information, the key feature encoding information, semantic decoding model parameters and channel transmission environment parameters to form second service information, and the second service information is sent to a receiving end; the receiving end receives the second service information, and performs source channel decoding and semantic decoding to obtain semantic information and key feature information; the semantic information is checked by using the key feature information, and after the checking, the semantic information is recovered to obtain third service information. Thus, the communication transmission amount is reduced, and the accuracy of communication transmission is ensured.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of communication technology, in particular to an encoding method, a decoding method, an encoding device, a decoding device and an electronic device. BACKGROUND

[0002] In 1977, Elience et al. first proposed a joint source channel coding scheme. Common traditional source channel coding methods include two-dimensional differential pulse code modulation coding, tree coding and two-dimensional discrete cosine transform coding. Most of the work is carried out in the traditional way for source channel coding. However, with the development of artificial intelligence technology, semantic intelligent extraction coding using artificial intelligence and joint source channel coding based on AI (Artificial Intelligence) model have not been deeply studied.

[0003] In the future Internet of Everything network, network nodes tend to be intelligent. Network node intelligence leads to rapid expansion of information space, dramatic increase in information dimension, and increased difficulty in multi-dimensional information representation, which results in a mismatch between traditional network service capabilities and high-latitude information space. The amount of data for communication transmission is too large, and the information service system cannot continuously meet the needs of people for complex, diverse and intelligent information transmission. In addition, network noise also brings a large transmission error rate. Traditional communication systems focus on the transmission process, but ignore the context-related meaning. The 5G system is close to the Shannon limit, and the increase in data volume will bring a series of problems, such as communication bottlenecks, increased latency and security issues. With the development of society and the progress of the times, communication technology and artificial intelligence are constantly developing. Intelligent network is a communication technology that combines artificial intelligence and communication technology to improve communication performance. By using an artificial intelligence model to encode, transmit and decode business information, the amount of data transmission in communication services can be significantly reduced, greatly improving information transmission efficiency without losing important information. Therefore, a joint source channel encoding and decoding method based on intelligent network is needed to realize adaptive encoding and decoding. In addition, the transmission data can be checked to continuously optimize the encoding and decoding method to ensure the accuracy of communication transmission. SUMMARY

[0004] The present disclosure provides an encoding method, a decoding method, an encoding device, a decoding device, an electronic device and a storage medium for an intelligent network.

[0005] According to an aspect of the present disclosure, an encoding method is provided, comprising:

[0006] The sending end obtains first business information and detects the data type of the first business information.

[0007] The sending end selects an adaptive semantic coding model according to a data type of the first service information;

[0008] The sending end inputs the first service information into the semantic coding model to obtain semantic coding information through semantic extraction, and obtains key feature coding information through feature extraction on the first service information;

[0009] The sending end selects a corresponding source-channel joint coding model based on a type of the semantic coding information and a current channel transmission environment, inputs the semantic coding information, the key feature coding information, parameters of a semantic decoding model and channel transmission environment parameters into the source-channel joint coding model, and codes to form second service information according to an encoding mode output by the source-channel joint coding model and transmits the second service information to the receiving end through a physical channel.

[0010] Optionally, the encoding method further includes that the sending end periodically samples the second service information received by the receiving end, calculates a mean square error of the second service information output by a decoder of the sending end and the second service information output by an encoder of the sending end, and trains the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end by using the calculated mean square error.

[0011] Optionally, the training of the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end by using the calculated mean square error includes:

[0012] calculating a mean square error between semantic information output by a decoder of the sending end and semantic information output by an encoder of the sending end;

[0013]

[0014] wherein, denotes the mean square error; denotes the second service information output by the encoder of the sending end; denotes the second service information received by the decoder of the sending end; denote a length, a width and a channel number of an image respectively;

[0015] Further, a semantic error between an original image and a decoded image is calculated:

[0016]

[0017] wherein, denotes the semantic error; denotes semantic coding information output by a penultimate layer of the semantic coding model;

[0018] The loss equation for training the semantic encoding model in the encoder and the semantic decoding model in the decoder is obtained through the mean square error and the semantic error:

[0019]

[0020] wherein, represents a weight factor, used to represent the semantic error and the mean square error occupy the proportion of each.

[0021] Optionally, the training of the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end further comprises:

[0022] The loss equation between the data signal output by the decoder of the sending end and the data signal output by the encoder of the sending end is calculated through the following formula, which is used to train the source channel joint encoding model in the encoder and / or the source channel joint decoding model in the decoder:

[0023]

[0024] wherein, represents the data signal output by the encoder of the sending end; y represents the data signal output by the decoder of the sending end.

[0025] Optionally, the source channel joint encoding and decoding method further comprises training the semantic encoding model and / or the semantic decoding model through the following steps:

[0026] The similarity between the original image output by the encoder of the sending end and the decoded image output by the decoder of the sending end is calculated through the following formula, so as to evaluate the accuracy of the semantic encoding model and the semantic decoding model:

[0027]

[0028] wherein, represents the mean value, represents the mean value of represents the variance of represents the variance of represents and the covariance of and represent the covariance coefficient;

[0029] Furthermore, when the similarity error between the original image and the decoded image exceeds the tolerance error threshold... At that time, that is The loss equation is calculated. The parameters of the semantic encoding model and / or the semantic decoding model are updated based on the stochastic gradient descent algorithm.

[0030] Optionally, the encoding method further includes training the source-channel joint coding model and / or the source-channel joint decoding model through the following steps:

[0031] The similarity between the data signal output by the transmitter encoder and the data signal output by the transmitter decoder is calculated using the following formula to evaluate the accuracy of the source-channel joint coding model and the source-channel joint decoding model:

[0032]

[0033] in, express mean express The mean, express variance express variance express and The covariance. and Represents the covariance coefficient;

[0034] Furthermore, when the similarity error between the data signal output by the encoder at the transmitting end and the data signal output by the decoder at the transmitting end exceeds the tolerance error threshold... At that time, that is The loss equation is calculated. The parameters of the source-channel joint coding model and the source-channel joint decoding model are updated based on the stochastic gradient descent algorithm.

[0035] Optionally, the step of the sending end processing the first service information to form the second service information specifically includes:

[0036] The first business information extracted using the semantic coding model is normalized, and then the normalized first business information is input into a residual neural network. The first business information is encoded using a multi-layer residual convolutional neural network and a parameterized activation function. Finally, the first business information is regularized to form the second business information.

[0037] The parameterized activation function comprises ReLU or PReLU.

[0038]

[0039] If The activation function PReLU is equivalent to the activation function ReLU.

[0040] Optionally, the semantic encoding model is based on a multi-layer residual network, comprising a bottleneck layer and an expanded bottleneck layer, wherein the number of the bottleneck layer is 2, and the number of the expanded bottleneck layer is 4.

[0041] According to another aspect of the present disclosure, a decoding method is provided, comprising:

[0042] The receiving end receives the second service information sent by the sending end, and decodes the semantic encoding information, the key feature encoding information and the parameters of the semantic decoding model by using the source channel joint decoding model.

[0043] The receiving end constructs the semantic decoding model by using the parameters of the semantic decoding model, and performs semantic decoding on the semantic encoding information and the key feature encoding information by using the semantic decoding model, to obtain semantic information and key feature information.

[0044] The receiving end verifies the semantic information by using the decoded key feature information.

[0045] The receiving end recovers the semantic information to obtain third service information in response to the semantic information passing the verification.

[0046] The receiving end repairs the semantic information by using the key feature information or triggers the sending end to resend the second service information in response to the semantic information failing the verification, until the semantic information passes the verification and is recovered to obtain the third service information.

[0047] Optionally, the step of decoding the second service information by the receiving end comprises:

[0048] The second service information is input into a plurality of deconvolution layers of the semantic decoding model, and is output through an activation function, and finally is subjected to once de-regularization processing; wherein the activation function comprises PReLU, ReLU and Sigmoid activation function.

[0049] According to another aspect of the present disclosure, an encoding device is provided, comprising:

[0050] The first acquisition module is configured to acquire first service information and detect the data type of the first service information.

[0051] a model selection module configured to select an adaptive semantic encoding model according to a data type of the first service information;

[0052] a semantic encoding module configured to input the first service information into the semantic encoding model to obtain semantic encoding information through semantic extraction, and to obtain key feature encoding information through feature extraction on the first service information;

[0053] a source-channel joint encoding module configured to select a corresponding source-channel joint encoding model based on a type of the semantic encoding information and a current channel transmission environment, and to input the semantic encoding information, the key feature encoding information, parameters of a semantic decoding model, and channel transmission environment parameters into the source-channel joint encoding model, and to encode the second service information through an encoding mode output by the source-channel joint encoding model and send the second service information to a receiving end through a physical channel.

[0054] Optionally, the encoding device further comprises a model training module configured to periodically sample the second service information received by the receiving end, calculate a mean square error between the second service information output by a decoder of the sending end and the second service information output by an encoder of the sending end, and train the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end by using the calculated mean square error.

[0055] According to another aspect of the present disclosure, a decoding device is also provided, comprising:

[0056] a source-channel joint decoding module configured to receive the second service information and decode the second service information by using a source-channel joint decoding model to obtain the semantic encoding information, the key feature encoding information, and the parameters of the semantic decoding model;

[0057] a semantic decoding module configured to construct the semantic decoding model by using the parameters of the semantic decoding model, and to decode the semantic encoding information and the key feature encoding information by using the semantic decoding model to obtain semantic information and key feature information;

[0058] an information verification module configured to verify the semantic information by using the key feature information;

[0059] The information processing module is configured to, in response to the semantic information passing the verification, perform recovery processing on the semantic information to obtain third service information; and in response to the semantic information failing to pass the verification, perform repair on the semantic information by using the key feature information, or trigger the sending end to resend the second service information until the semantic information passes the verification and is subjected to the recovery processing to obtain the third service information.

[0060] The present disclosure also provides an electronic device, comprising:

[0061] at least one processor; and

[0062] a memory connected with the at least one processor in communication; wherein

[0063] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the encoding method or the decoding method of any one of the above technical solutions.

[0064] The present disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to perform the encoding method or the decoding method according to any one of the above embodiments.

[0065] The present disclosure also provides a computer program product comprising a computer program, which, when executed by a processor, implements the encoding method or the decoding method according to any one of the above embodiments.

[0066] The present disclosure provides an encoding method, a decoding method, an encoding device, a decoding device, an electronic device, and a storage medium, which combines semantic extraction of artificial intelligence with source-channel joint coding of communication technology, selects a corresponding source-channel joint coding model in combination with the type of source information and the channel transmission environment, reduces the amount of data for communication transmission and ensures the quality of communication transmission, and checks the transmitted service information by using the extracted key feature information, thereby greatly reducing the amount of data for transmission on the premise of ensuring the accuracy of communication transmission.

[0067] It should be understood that the content described in this part is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0068] The accompanying drawings are used to better understand the present scheme and do not limit the present disclosure. Among them:

[0069] Figure 1 is a step diagram of the encoding method in the embodiments of the present disclosure;

[0070] Figure 2 This is a flowchart illustrating the steps of the decoding method in an embodiment of this disclosure;

[0071] Figure 3 This is an overall flowchart of the joint encoding and decoding of the source and channel in the embodiments of this disclosure;

[0072] Figure 4 This is a schematic diagram illustrating the encoder model optimization based on receiver feedback information in this embodiment of the disclosure.

[0073] Figure 5 This is a flowchart of the joint semantic source-channel coding and decoding process in an embodiment of this disclosure;

[0074] Figure 6 This is a diagram of the multilayer residual network structure used for feature extraction in this embodiment of the disclosure;

[0075] Figure 7 This is a bottleneck layer structure diagram of the residual network in this embodiment of the disclosure;

[0076] Figure 8 This is a diagram of the extended bottleneck layer structure of the residual network in this embodiment of the present disclosure;

[0077] Figure 9 This is a diagram illustrating the key feature extraction and analysis process in an embodiment of this disclosure;

[0078] Figure 10 This is a schematic block diagram of the encoding device in the embodiments of this disclosure;

[0079] Figure 11 This is a schematic block diagram of the decoding device in the embodiments of this disclosure. Detailed Implementation

[0080] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0081] The intelligent and simple network mainly transmits business information through an artificial intelligence model. The artificial intelligence model is used to compress first business information to be transmitted into second business information related to the artificial intelligence model, which greatly reduces the data traffic in the network and the compression efficiency is much higher than that of traditional compression algorithms. The sending end device uses a pre-configured first model to extract the first business information and obtain second business information to be transmitted. The sending end device transmits the second business information to the receiving end device. The receiving end device receives the second business information and uses a pre-configured second model to recover the second business information to obtain third business information. The third business information recovered by the second model has some differences in quality compared with the original first business information, but the content of the two is consistent, and the experience of the user is almost the same. Before the sending end device transmits the second business information to the receiving end device, the method further includes: an updating module judging whether the receiving end device needs to update the second model, and transmitting a pre-configured third model to the receiving end device when it is judged that the second model needs to be updated. The receiving end device updates the second model by using the third model. The artificial intelligence model pre-trained can process business information, which can significantly reduce the data transmission in the communication business and greatly improve the information transmission efficiency. These models are relatively stable and have reusability and transmissibility. The transmission and reuse of the models can help to enhance the network intelligence and reduce the waste of resources. The model can be divided into a plurality of model slices according to different division rules. The model slices can be transmitted between different network nodes, and the model slices can be assembled into a model. The model slices can be stored in multiple network nodes. When a network node finds that it lacks or needs to update a certain model or a certain model slice, it can request the surrounding nodes that may have the slice.

[0082] The transmission of the business information and the transmission of the model occur in a communication network, and the communication transmission is based on a network protocol. The network nodes on the path of the transmission of the business information and the transmission of the model include intelligent and simple routers. The functions of the intelligent and simple routers include but are not limited to business information transmission, model transmission, self-updating of the model, security protection and the like. The transmission function of the intelligent and simple router involves transmitting the business information or the model from a source node to a sink node, and there are multiple paths between the source node and the sink node. The model transmission function of the intelligent and simple router can transmit the model slices, and the model slices can be transmitted through multiple paths to improve the model transmission rate.

[0083] The present disclosure provides an encoding method, as shown in Figure 1 The present disclosure provides an encoding method, as shown in

[0084] In step S101, the sending end acquires first service information and detects the data type of the first service information.

[0085] In step S102, the sending end selects an adaptive semantic coding model according to the data type of the first service information.

[0086] In step S103, the sending end inputs the first service information into the semantic coding model to obtain semantic coding information through semantic extraction, and obtains key feature coding information through feature extraction of the first service information.

[0087] In step S104, the sending end selects a corresponding source-channel joint coding model based on the type of the semantic coding information and the current channel transmission environment, inputs the semantic coding information, the key feature coding information, the parameters of the semantic decoding model and the channel transmission environment parameters into the source-channel joint coding model, and codes to form second service information according to the coding mode output by the source-channel joint coding model and sends the second service information to the receiving end through a physical channel.

[0088] Specifically, the coding method is characterized in that the semantic extraction of the source and the source-channel coding are jointly considered. Figure 3As shown, in terms of semantic extraction and source channel joint coding, a smart transmitter is constructed at the sending end, which mainly includes two modules, one is a semantic coding module, and the other is a source channel joint coding module. In the semantic coding module, an artificial intelligence method is used to extract the semantics of the source data (i.e. the first service information) by using a trained semantic coding model, and to intelligently encode the extracted semantic information and the parameters of the selected semantic coding model. In the semantic extraction process, based on the data type (text or image or video) of the first service information, key feature information is extracted and encoded for transmission, which is used to check the recovered information. The checking result can be used as a basis for judgment to determine whether the information needs to be retransmitted, or as feedback for the training of the semantic coding model and the semantic decoding model at the sending end and the receiving end. In the joint source channel coding module, an artificial intelligence technology is used to model the transmission environment and estimate the modulation and coding channel. Based on the source channel joint coding model, the data to be transmitted is jointly encoded by the source and channel according to the current channel transmission information. Through the above technical solution, the extracted semantic information is combined with the current channel transmission environment. Specific source information is combined with specific channel transmission environment, and a specific source channel joint coding model is selected. For example, assuming that the current transmission source A is required, and the main channel conditions are divided into A, B, C, and D, four source channel joint coding models of A, A, A, and A can be obtained by pre-training. When transmitting source A at a certain time, the current channel condition is C, and the A model can be selected for coding. The model selected for a specific source type and a specific channel transmission environment is beneficial to improve the accuracy of coding and the efficiency of channel transmission.

[0089] As an optional implementation, as shown in Figure 4 As shown, the encoding method further includes: periodically sampling, at the sending end, the second service information received by the receiving end, and calculating the mean square error of the second service information output by the decoder of the sending end and the second service information output by the encoder of the sending end, and training the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end by using the calculated mean square error. The encoder of the sending end and the encoder of the receiving end both include two encoding models of a semantic coding model and a source channel joint coding model, and the decoder of the sending end and the decoder of the receiving end both include two decoding models of a semantic decoding model and a source channel joint decoding model.

[0090] In order to continuously optimize and update the models in the encoder and the models in the decoder, as shown in Figure 4As shown, in the embodiment, the decoder model 2A of the receiving end 2 receives the data information (i.e., the second service information) to perform periodic sampling, transmits the data information back to the original sending end 1 through a physical channel, inputs the data into the decoder model 1A of the sending end 1, inputs the decoded symbols together with the original sending symbols output by the encoder model 1B of the sending end into the codec model training unit 1C for comparison, calculates the mean square error value of the two, performs model optimization training of the encoder and the decoder, and thus improves the encoding and decoding capabilities of the encoding model and the decoding model, and guarantees the accuracy of the transmitted data.

[0091] As an optional implementation, the encoder and the decoder of the sending end and the encoder and the decoder of the receiving end are trained by using the calculated mean square error, including:

[0092] The mean square error between the semantic information output by the decoder of the sending end and the semantic information output by the encoder of the sending end is calculated:

[0093]

[0094] wherein, denotes the mean square error; denotes the second service information output by the encoder of the sending end; denotes the second service information received by the decoder of the sending end; denotes the length, the width and the channel number of the image, respectively;

[0095] Further, the semantic error between the original image and the decoded image is calculated:

[0096]

[0097] wherein, denotes the semantic error; denotes the semantic encoding information output by the penultimate layer of the output semantic encoding model;

[0098] The loss equation for training the semantic encoding model in the encoder and the semantic decoding model in the decoder is obtained by the mean square error and the semantic error:

[0099]

[0100] wherein, denotes a weight factor, used to represent the proportion of the semantic error and the mean square error respectively.

[0101] It should be noted that the above loss equation is only used for training the semantic encoding model and the semantic decoding model.

[0102] At the same time, not only the difference between images is considered, but also the difference of data signals, and a loss equation between the data signal output by the decoder of the sending end and the data signal output by the encoder of the sending end is calculated for training the source channel joint encoding model and / or the source channel joint decoding model:

[0103]

[0104] wherein, represents the data signal output by the encoder of the sending end; represents the data signal output by the decoder of the sending end.

[0105] Further, before training the model, the accuracy of the model needs to be evaluated, and the model needs to be trained by using the above loss equation only when the model accuracy is insufficient. The accuracy of the model is evaluated by calculating the similarity between the original image (or data signal) output by the encoder of the sending end and the decoded image (or data signal) output by the decoder of the sending end:

[0106]

[0107] wherein, represents the mean value; represents the mean value of represents the variance of represents the variance of represents the covariance of and represents the covariance coefficient.

[0108] The similarity of the image or the similarity of the data signal between the encoder of the sending end and the decoder of the sending end is obtained by the above evaluation formula, and when the similarity indicates that the difference is greater than a preset threshold, for example, , it indicates that the accuracy of the semantic encoding model or the semantic decoding model is insufficient, and then the model can be trained by using the loss equation ; when , it indicates that the accuracy of the source channel joint encoding model or the source channel joint decoding model is insufficient, resulting in inaccurate encoded or decoded information, and then the model can be trained by using the loss equation .

[0109] Exemplarily, based on the source channel joint encoding overall process shown in Figure 3 , it is assumed that the first service information obtained by the sending end is an image, and a picture source is defined as , wherein​ respectively represent the length, width of the picture and the number of channels. A specific semantic encoder and semantic decoder are first selected based on a picture pre-classifier, denoted as . The semantic encoder includes multiple layers of down-sampling convolutional layers and multiple layers of linear encoding layers to reduce the dimensionality of the multi-dimensional picture source (first service information) to semantic encoding information wherein, represents the dimension of the semantic encoding information. and respectively represent the neural network parameters of the semantic encoder and the semantic decoder. Then, the joint source channel encoded symbol data stream is denoted as , denoted as:

[0110]

[0111] wherein, and respectively represent the encoder and the decoder for joint source channel encoding; parameters represent the parameters of the semantic decoding model.

[0112] Further, the bit data stream (second service information) sent by the sending end undergoes multi-channel channel fading, and the information received by the receiving end is modeled as:

[0113]

[0114] wherein, represents the received symbol information stream; represents an additive white Gaussian noise; is the channel impulse response gain of the multipath channel; represents a convolution operation, assuming that for a symbol signal stream, the channel is fixed, and different between different symbols at the same time. The channel changes over time, and is specifically represented as follows:

[0115]

[0116] wherein, represents the channel gain of the th channel; represents the delay of the th channel; represents the total number of channels.

[0117] The data obtained by the receiving end decoding the second service information is denoted as:

[0118]

[0119] In order to enable the receiving end to decode the information with source information As much as possible, mean square error (MSE) is used as part of the loss equation to train the semantic encoding model and the semantic decoding model, that is, through the above loss equation And / or Train the model.

[0120] The semantic encoder and the semantic decoder are equipped at the transmitting end, and the semantic decoding model parameters need to be transmitted to the receiving end together with the semantic encoding information, wherein the semantic decoding model needs to be configured to have as small data volume as possible. The AI automatic coding based on the convolutional neural network is used to effectively reduce the redundancy of parameters. For a specific channel semantic encoding symbol, the first The semantic expression of the layer feature mapping is:

[0121]

[0122] Among them, Indicates an activation function (the activation function can be a parameterized ReLU or PReLU, Sigmod function); Indicates a 2D convolution operation.

[0123] The full connection layer of the receiving end is symmetrical to the full connection layer of the transmitting end, and the receiving end uses the reconstructed model recovered by the received semantic decoding model parameters to input the semantic bit data stream into the reconstructed semantic decoding model to recover the second service information to obtain third service information (recovered image) highly similar to the first service information. The reconstructed image obtained is as follows:

[0124]

[0125] Among them, Indicates a flip operation.

[0126] Further, in order to evaluate the training accuracy of the model, a structural similarity model is introduced to evaluate the similarity of two images, which is represented as:

[0127]

[0128] Among them, Indicates the mean value; Indicates the mean value of Indicates the variance of Indicates the variance of Indicates the covariance of And And​​​​​ This represents the covariance coefficient.

[0129] Furthermore, the semantic encoding model and semantic decoding model are trained based on the similarity evaluation results of the original image and the decoded image. The training steps include:

[0130] Input: A set of images S obtained from an image database, and a model selection classifier. Tolerance error threshold ;

[0131] While do;

[0132] calculate ;

[0133] calculate ;

[0134] Calculate the loss equation ;

[0135] Updating neural network parameters based on stochastic gradient descent algorithm ;

[0136] End while

[0137] Output: Semantic encoder and decoder model , .

[0138] Simultaneously, based on the similarity evaluation results of the data signals, the source-channel joint coding model and the source-channel joint decoding model are trained. The training steps include:

[0139] Input: A set of images S obtained from an image database, and a model selection classifier. Tolerance error threshold Semantic coding model and semantic decoding model Fading channel ,noise ;

[0140] While do;

[0141] Computation symbolic data stream ;

[0142] Calculate the received symbolic data stream ;

[0143] Based on the decoded bit data The decoder model parameters and their semantic encoding data are obtained, and are represented as follows: ;

[0144] Computing the loss function ;

[0145] Updating the neural network parameters based on the stochastic gradient descent algorithm ;

[0146] End While

[0147] Output: get the trained network model: , .

[0148] As an optional implementation, as shown in Figure 5 , the sending end forms the second service information by processing the first service information, and the step specifically includes:

[0149] The first service information is extracted by using a semantic encoding model for normalization processing, and then the normalized first service information is input into a residual neural network. The first service information is encoded by using a multi-layer residual convolutional neural network and a parameterized activation function. Finally, the first service information is regularized to form the second service information.

[0150] The parameterized activation function includes ReLU or PReLU:

[0151]

[0152] If , the activation function PReLU is equivalent to the activation function ReLU.

[0153] The step of decoding the second service information by the receiving end includes:

[0154] The second service information is input into multiple deconvolution layers of a semantic decoding model, and is output through an activation function. Finally, a de-regularization processing is performed. The activation function includes PReLU, ReLU, and Sigmoid activation function.

[0155] Specifically, in this embodiment, a neural network is introduced to map the extracted semantic information and the corresponding semantic extraction model parameters to bit data. On the basis of the convolutional neural network, a residual network is introduced to parameterize the source encoder and the decoder. The specific joint source-channel encoding process is shown in Figure 5 , and the parameterization process is specifically described as follows:

[0156] The sending end normalizes the extracted source data (first service information), and then inputs the normalized data into a residual neural network. The residual neural network adopts a parameterized activation function PReLU. The source data is encoded by an encoding equation based on a multi-layer residual convolutional neural network. After passing through the multi-layer convolutional neural network, the data is normalized and input into a physical channel to be sent to the receiving end. The PReLU activation function is specifically represented as:

[0157]

[0158] If The activation function PReLU is equivalent to the activation function ReLU.

[0159] In addition, the key features of the first service information also need to be extracted, and the transmission redundancy is increased to increase the decoding accuracy of the receiving end. In the aspect of feature extraction, a multi-layer residual network generated by ResNet50 is as shown in Figure 6 , wherein Res1-Res4 are residual convolutional layers generated by ResNet50, Res5 is an additional convolutional layer, and F is the encoding output information. Among them, Res1-Res4 have different convolutional structures, and usually need to be adjusted based on the output format of F. Res1 initialization generally adopts the (7X7 conv, 64, stride 2) parameter mode (convolution is 7x7, 64 pixels, and stride 2). The advantages of using the above residual network structure are: (1) low complexity, few parameters required; (2) deeper network depth, no gradient disappearance; (3) increased classification accuracy; (4) solves the network degradation problem in the training process.

[0160] As shown in Figure 5 , the receiving end inputs the received data into the deconvolutional layer corresponding to the transmission convolutional layer, and outputs through the activation function, and then performs a reverse regularization processing again to decode the semantic bit data, the bit data of the semantic decoding model parameters, and the bit data of the key feature extraction encoding information. Based on the semantic decoding model parameters, the semantic information is decoded. The activation function described above includes but is not limited to PReLU, ReLU, Sigmoid activation function, etc. The deconvolutional neural network of the decoder corresponds to the neural network of the encoder.

[0161] In the aspect of key feature extraction, two types of residual networks are introduced, one is a bottleneck layer, and the other is an expanded bottleneck layer. The bottleneck layer and the expanded bottleneck layer structure are as shown in Figure 7 and Figure 8 . The key feature analysis process based on the residual neural network is as shown in Figure 9 . The receiving end inputs the received semantic information into two bottleneck layers and four expanded bottleneck layers for semantic segmentation and semantic parsing, respectively.

[0162] The present disclosure also provides a decoding method, as shown, comprising: Figure 2

[0163] Step S201, the receiving end receives the second service information sent by the sending end, and decodes the semantic encoding information, the key feature encoding information and the parameters of the semantic decoding model by using the source channel joint decoding model;

[0164] Step S202, the receiving end constructs the semantic decoding model by using the parameters of the semantic decoding model, and performs semantic decoding on the semantic encoding information and the key feature encoding information by using the semantic decoding model, to obtain the semantic information and the key feature information;

[0165] Step S203, the receiving end verifies the semantic information by using the decoded key feature information;

[0166] Step S204, the receiving end performs recovery processing on the semantic information in response to the semantic information passing the verification, to obtain the third service information;

[0167] Step S205, the receiving end performs repair on the semantic information by using the key feature information in response to the semantic information failing the verification, or triggers the sending end to retransmit the second service information, until the semantic information passes the verification, and then the receiving end performs recovery processing on the semantic information to obtain the third service information.

[0168] The embodiment constructs an intelligent receiver at the receiving end, which can include an AI model-based source channel decoder and an AI model-based semantic decoder. The intelligent receiver decodes the received second service information (the second service information is bit data) according to the AI model parameters, to obtain the semantic decoding model parameters, the semantic encoding information and the key feature encoding information used for semantic decoding at the receiving end, constructs the semantic decoding model by using the semantic decoding model parameters, and performs semantic decoding to obtain the semantic information and the key feature information corresponding to the first service information. In order to verify the accuracy of the received information, the decoded semantic information needs to be verified. The present disclosure provides two verification methods: (1) important information in the semantic information is analyzed and compared with the received key feature information; (2) the semantic information is input into a verification network, and it is judged whether the output elements of the verification network are in the key feature information. If the verification passes, it means that the semantic information decoded by the receiving end is correct, otherwise the semantic information is incorrect. The key feature information is used to repair the semantic information that fails the verification, or the sending end can be triggered to retransmit or partially retransmit the service information in a necessary case, to ensure the accuracy of the transmitted service information.

[0169] ​As an optional implementation, the step of decoding the second service information at the receiving end comprises: inputting the second service information into multiple deconvolution layers of a semantic decoding model, and outputting through an activation function, and finally performing a de-regularization process; wherein the activation function comprises a PReLU, ReLU, Sigmoid activation function. Finally, the receiving end inputs the decoded semantic information into a feature verification network to verify the recovered semantic information. The processing process of the feature verification network comprises: performing pooling processing on the decoded feature information, and adding the signal-to-noise ratio of the channel to the pooled information for splicing, inputting into a fully connected layer, and outputting through an activation function PReLU, and then inputting into a next fully connected layer, and outputting through a Sigmoid activation function. The matching degree of the data output by the feature verification network and the key feature information is matched, so as to determine whether the decoded source image information is correct.

[0170] The present disclosure also provides an encoding device, as shown in Figure 10 comprising:

[0171] The first acquisition module 101 is configured to acquire first service information and detect the data type of the first service information.

[0172] The model selection module 102 is configured to select an adaptive semantic encoding model according to the data type of the first service information.

[0173] The semantic encoding module 103 is configured to input the first service information into the semantic encoding model to extract semantics and obtain semantic encoding information, and to extract features from the first service information to obtain key feature encoding information.

[0174] The source channel joint encoding module 104 is configured to select a corresponding source channel joint encoding model based on the type of the semantic encoding information and the current channel transmission environment, and to input the semantic encoding information, the key feature encoding information, the parameters of the semantic decoding model, and the channel transmission environment parameters into the source channel joint encoding model, and to encode the second service information according to the encoding mode output by the source channel joint encoding model and send it to the receiving end through the physical channel.

[0175] Specifically, the source channel joint encoding method is characterized by jointly considering semantic extraction of the source and source channel encoding. The first acquisition module 101, the model selection module 102, the semantic encoding module 103, and the source channel joint encoding module 104 can be provided in the sending end 1. In terms of semantic extraction and source channel joint encoding, a smart sender is constructed in the sending end, which mainly includes two modules, one is a semantic encoding module, and the other is a source channel joint encoding module. Figure 3The diagram shows the overall flowchart of semantic encoding transmission at the sending end and semantic reconstruction at the receiving end. The semantic encoding module 103 employs artificial intelligence methods, utilizing a trained semantic encoding model to extract semantics from the source data (i.e., the first service information), and intelligently encodes the extracted semantic information and the parameters of the selected semantic encoding model. During semantic extraction, based on the data type (text, image, or video) of the first service information, key feature information is extracted and encoded for transmission. This key feature is used to verify the reconstructed information; the verification result can serve as a basis for determining whether the information needs to be retransmitted, and can also serve as feedback for the training of the semantic encoding and decoding models at the sending and receiving ends. In the source-channel joint encoding module 104, artificial intelligence technology is used to perform perceptual modeling of the transmission environment and channel estimation of the modulation-coding channel. Based on the AI ​​model and the current channel transmission information, source-channel joint encoding is performed on the data to be transmitted. Through the above encoding device, the extracted semantic information is combined with the current channel transmission environment, and the AI ​​model outputs a better encoding mode, which helps improve the encoding accuracy and channel transmission efficiency.

[0176] As an optional implementation, the encoding apparatus further includes a model training module 105, configured to periodically sample the second service information received by the receiving end, calculate the mean square error between the second service information output by the decoder at the transmitting end and the second service information output by the encoder at the transmitting end, and train the encoder and decoder at the transmitting end and the encoder and decoder at the receiving end using the calculated mean square error. The model training method in this embodiment is consistent with the model training method in the above-described encoding method embodiments, and the model is trained by calculating the loss equation; therefore, it will not be described in detail below.

[0177] This disclosure also provides a decoding device, such as Figure 11 As shown, it includes:

[0178] The source-channel joint decoding module 201 is configured to receive the second service information and use the source-channel joint decoding model to decode it to obtain semantic coding information, key feature coding information and parameters of the semantic decoding model;

[0179] The semantic decoding module 202 is configured to construct a semantic decoding model using the parameters of the semantic decoding model, and to perform semantic decoding on the semantic encoded information and key feature encoded information using the semantic decoding model to obtain semantic information and key feature information.

[0180] The information verification module 203 is configured to verify semantic information using key feature information;

[0181] The information processing module 204 is configured to, in response to the semantic information passing the verification, perform recovery processing on the semantic information to obtain third service information; and in response to the semantic information not passing the verification, perform repair on the semantic information by using the key feature information, or trigger the sending end to resend the second service information, until the semantic information passes the verification and is subjected to the recovery processing to obtain the third service information.

[0182] Specifically, the source channel joint decoding module 201, the semantic decoding module 202, the information verification module 203, and the information processing module 204 can be arranged in the receiving end 2. In the receiving end, a smart receiver is constructed, which can include an AI model-based source channel decoder and an AI model-based semantic decoder. The source channel joint decoding module 201 decodes the received second service information (the second service information is bit data) according to AI model parameters to obtain semantic decoding model parameters, semantic encoding information, and key feature encoding information used for semantic decoding of the receiving end, and the semantic decoding module 202 constructs a semantic decoding model by using the semantic decoding model parameters to perform semantic decoding to obtain semantic information and key feature information corresponding to the first service information. In order to verify the accuracy of the received information, the information verification module 203 is needed to verify the decoded semantic information, and the present disclosure provides two verification methods: (1) important information in the semantic information is analyzed and compared with the received key feature information; (2) the semantic information is input into a verification network, and it is judged whether the output elements of the verification network are in the key feature information. If the verification passes, it indicates that the semantic information decoded by the receiving end is correct, and the information processing module 204 performs semantic recovery on the correct semantic information to obtain third service information. For example, the first service information is an image, the second service information contains semantic information obtained by performing semantic extraction on the first service information, and the semantic information is recovered to obtain an image highly similar to the first service information, i.e., the third service information, which greatly reduces the amount of data transmitted and improves the communication efficiency. On the contrary, the semantic information is incorrect, the key feature information is used to repair the semantic information that does not pass the verification, or the sending end is triggered to resend or partially resend the service information if necessary, until the semantic information decoded by the receiving end passes the verification of the information verification module 203, so as to ensure the accuracy of the transmitted service information.

[0183] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program product.

[0184] In particular, an electronic device is intended to represent a wide variety of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. An electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular telephones, smartphones, wearable devices, and other similar computing devices. The components shown here, their connections and relationships, and their functions, are meant to be examples only, and are not intended to limit the implementations of the present disclosure described and / or claimed in this document.

[0185] The device includes a computing unit that can perform various appropriate actions and processes in accordance with a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for the operation of the device can also be stored. The computing unit, the ROM, and the RAM are connected to each other through a bus. An input / output (I / O) interface is also connected to the bus.

[0186] A plurality of components in the device are connected to the I / O interface, including an input unit such as a keyboard, a mouse, and the like; an output unit such as various types of displays, a speaker, and the like; a storage unit such as a magnetic disk, an optical disk, and the like; and a communication unit such as a network card, a modem, a wireless communication transceiver, and the like. The communication unit allows the device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0187] The computing unit can be various general and / or special purpose processing components with processing and computing capabilities. Some examples of the computing unit include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, and the like. The computing unit performs various methods and processes described above, such as the encoding method or the decoding method in the above-described embodiments. For example, in some embodiments, the encoding method or the decoding method can be implemented as a computer software program that is tangibly embodied in a machine-readable medium, such as the storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via the ROM and / or the communication unit. When the computer program is loaded into the RAM and executed by the computing unit, one or more steps of the encoding method or the decoding method described above can be performed. Alternatively, in other embodiments, the computing unit can be configured to perform the encoding method or the decoding method by any other appropriate means, such as by means of firmware.

[0188] The various embodiments of the systems and techniques described above can be implemented in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a load programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0189] Program code for carrying out methods of the present disclosure, or for carrying out operations of the methods, can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions / operations specified in the flow charts and / or block diagrams to be implemented. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as part of a standalone software package, or entirely on a remote machine or server.

[0190] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0191] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0192] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0193] The computer system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server is generally established by computer programs running on the respective computers and having a client-server relationship to each other. The servers can be cloud servers, servers of a distributed system, or servers combined with a blockchain.

[0194] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be performed in parallel, in series, or in a different order, without departing from the desired results of the technology disclosed in the present disclosure, which are not limited herein.

[0195] The specific embodiments described above are not intended to be limiting, and persons skilled in the art will appreciate that various modifications, combinations, sub-combinations and alternatives can be made to the specific embodiments without departing from the spirit and scope of the disclosure. Any modifications, equivalent substitutions, improvements, and the like, made within the spirit and principles of the disclosure, are intended to be included in the scope of the disclosure.

Claims

1. An encoding method, characterized in that, include: The sending end obtains the first service information and detects the data type of the first service information; The sending end selects an appropriate semantic coding model based on the data type of the first service information; The sending end inputs the first service information into the semantic coding model to extract semantic information and obtains semantic coding information, and performs feature extraction on the first service information to obtain key feature coding information; Based on the type of the semantic coding information and the current channel transmission environment, the transmitting end selects the corresponding source-channel joint coding model, and inputs the semantic coding information, the key feature coding information, the parameters of the semantic decoding model, and the channel transmission environment parameters into the source-channel joint coding model. The transmitting end encodes the second service information according to the coding mode output by the source-channel joint coding model and sends it to the receiving end through the physical channel. The specific steps of the sending end processing the first service information to form the second service information include: The first business information extracted using the semantic coding model is normalized, and then the normalized first business information is input into a residual neural network. The first business information is encoded using a multi-layer residual convolutional neural network and a parameterized activation function. Finally, the first business information is regularized to form the second business information. The parameterized activation function includes ReLU or PReLU: like The activation function PReLU is equivalent to the activation function ReLU; x represents the output of the multilayer residual convolutional neural network.

2. The encoding method according to claim 1, characterized in that, Also includes: The transmitting end periodically samples the second service information received by the receiving end, and calculates the mean square error between the second service information output by the decoder of the transmitting end and the second service information output by the encoder of the transmitting end. The calculated mean square error is used to train the encoder and decoder of the transmitting end and the encoder and decoder of the receiving end.

3. The encoding method according to claim 2, characterized in that, The training of the encoder and decoder at the transmitting end and the encoder and decoder at the receiving end includes: Calculate the mean square error between the semantic information output by the decoder at the transmitting end and the semantic information output by the encoder at the transmitting end: in, This represents the mean square error; This refers to the second service information output by the encoder at the transmitting end; This indicates that the decoder at the sending end has received the second service information; These represent the image's length, width, and number of channels, respectively. Furthermore, the semantic error between the original image and the decoded image is calculated: in, This indicates the semantic error; This represents the semantic coding information output by the penultimate layer of the semantic coding model; The loss equations for training the semantic encoding model in the encoder and the semantic decoding model in the decoder are obtained using the mean squared error and the semantic error: in, The weighting factor is used to represent the semantic error. and the mean square error The respective proportions.

4. The encoding method according to claim 2, characterized in that, The training of the encoder and decoder at the transmitting end and the encoder and decoder at the receiving end further includes: The loss equation between the data signal output by the decoder at the transmitting end and the data signal output by the encoder at the transmitting end is calculated using the following formula, and is used to train the source-channel joint coding model in the encoder and / or the source-channel joint decoding model in the decoder: in, This represents the data signal output by the encoder at the transmitting end; M represents the data signal output by the decoder at the transmitting end; M represents the number of data signals output by the encoder at the transmitting end.

5. The encoding method according to claim 3, characterized in that, It also includes training the semantic encoding model and / or the semantic decoding model through the following steps: The similarity between the original image output by the encoder at the transmitting end and the decoded image output by the decoder at the transmitting end is calculated using the following formula to evaluate the accuracy of the semantic encoding model and the semantic decoding model: in, express mean express The mean, express variance express variance express and covariance; and The variable represents the covariance coefficient; x represents the original image; y represents the decoded image; Furthermore, when the similarity error between the original image and the decoded image exceeds the tolerance error threshold... At that time, that is The loss equation is calculated. The parameters of the semantic encoding model and / or the semantic decoding model are updated based on the stochastic gradient descent algorithm.

6. The encoding method according to claim 4, characterized in that, It also includes training the source-channel joint coding model and / or the source-channel joint decoding model through the following steps: The similarity between the data signal output by the transmitter encoder and the data signal output by the transmitter decoder is calculated using the following formula to evaluate the accuracy of the source-channel joint coding model and the source-channel joint decoding model: in, express mean express The mean, express variance express variance express and covariance; and y represents the covariance coefficient; x represents the data signal output by the encoder at the transmitting end; y represents the data signal output by the decoder at the transmitting end. Furthermore, when the similarity error between the data signal output by the encoder at the transmitting end and the data signal output by the decoder at the transmitting end exceeds the tolerance error threshold... At that time, that is The loss equation is calculated. The parameters of the source-channel joint coding model and the source-channel joint decoding model are updated based on the stochastic gradient descent algorithm.

7. The encoding method according to claim 1, characterized in that, The semantic coding model is based on a multi-layer residual network, including two types of residual networks: a bottleneck layer and an extended bottleneck layer. The number of bottleneck layers is 2, and the number of data in the extended bottleneck layer is 4.

8. A decoding method, characterized in that, include: The receiving end receives the second service information sent by the sending end and uses the source-channel joint decoding model to decode it to obtain semantic coding information, key feature coding information, and parameters of the semantic decoding model; The receiving end constructs the semantic decoding model using the parameters of the semantic decoding model, and performs semantic decoding on the semantic encoded information and the key feature encoded information using the semantic decoding model to obtain semantic information and key feature information; The receiving end uses the key feature information obtained from decoding to verify the semantic information; In response to the semantic information passing the verification, the receiving end performs recovery processing on the semantic information to obtain the third service information; In response to the semantic information failing verification, the receiving end repairs the semantic information using the key feature information, or triggers the sending end to resend the second service information until the semantic information passes verification and is then restored to obtain the third service information. The steps for the receiving end to decode the second service information include: The second business information is input into multiple deconvolutional layers of the semantic decoding model and output through activation functions, and finally subjected to an inverse regularization process; wherein, the activation functions include PReLU, ReLU, and Sigmoid activation functions.

9. An encoding device, characterized in that, include: The first acquisition module is configured to acquire first business information and detect the data type of the first business information; The model selection module is configured to select an appropriate semantic coding model based on the data type of the first business information. The semantic encoding module is configured to input the first business information into the semantic encoding model to extract semantic information and obtain semantic encoded information, and to extract key feature information from the first business information. The source-channel joint coding module is configured to select the corresponding source-channel joint coding model based on the type of the semantic coding information and the current channel transmission environment, and input the semantic coding information, the key feature coding information, the parameters of the semantic decoding model and the channel transmission environment parameters into the source-channel joint coding model, and encode the second service information according to the coding mode output by the source-channel joint coding model and send it to the receiving end through the physical channel. The first business information extracted using the semantic coding model is normalized, and then the normalized first business information is input into a residual neural network. The first business information is encoded using a multi-layer residual convolutional neural network and a parameterized activation function. Finally, the first business information is regularized to form the second business information. The parameterized activation function includes ReLU or PReLU: like The activation function PReLU is equivalent to the activation function ReLU; x represents the output of the multilayer residual convolutional neural network.

10. The encoding device according to claim 9, characterized in that, Also includes: The model training module is configured to periodically sample the second service information received by the receiving end, calculate the mean square error between the second service information output by the decoder of the transmitting end and the second service information output by the encoder of the transmitting end, and use the calculated mean square error to train the encoder and decoder of the transmitting end and the encoder and decoder of the receiving end.

11. A decoding device, characterized in that, include: The source-channel joint decoding module is configured to receive the second service information and use the source-channel joint decoding model to decode it to obtain semantic coding information, key feature coding information, and parameters of the semantic decoding model. The semantic decoding module is configured to construct the semantic decoding model using the parameters of the semantic decoding model, and to perform semantic decoding on the semantic encoded information and the key feature encoded information using the semantic decoding model to obtain semantic information and key feature information; The information verification module is configured to verify the semantic information using the key feature information; The information processing module is configured to, in response to the semantic information passing the verification, perform recovery processing on the semantic information to obtain third business information; In response to the semantic information failing verification, the key feature information is used to repair the semantic information, or the sending end is triggered to resend the second service information until the semantic information passes verification and is then restored to obtain the third service information. The source-channel joint decoding module is also used to input the second service information into multiple deconvolution layers of the semantic decoding model, output it through an activation function, and finally perform an inverse regularization process; wherein, the activation function includes PReLU, ReLU, and Sigmoid activation functions.

12. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the encoding method of any one of claims 1-7 or the decoding method of claim 8.

13. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to execute the encoding method according to any one of claims 1-7 or the decoding method according to claim 8.

14. A computer program product comprising a computer program that, when executed by a processor, implements the encoding method according to any one of claims 1-7 or the decoding method according to claim 8.

Images