A Design Method for Task-Oriented Secure Semantic Communication Systems

Abstract

The present invention discloses a design method for a task-oriented secure semantic communication system, which relates to the field of communication technologies. The implementation steps are as follows: data set collection, and dividing the data set into a training set and a test set; data preprocessing; constructing a channel environment, and designing a relevant transmission architecture for the secure semantic communication system, including a sending end, a receiving end, and an eavesdropping end; jointly pre-training the constructed secure semantic communication system; loading the pre-trained parameter model, and fine-tuning and training the model with a generative adversarial network loss function and a secure semantic feature loss function; inputting the test set data into the network; outputting the results and verifying the effectiveness of the model. The present invention uses a generative adversarial network and a secure semantic feature loss function to reduce the image reconstruction quality and classification accuracy of the eavesdropping end, ensuring the performance security of downstream AI tasks from a semantic perspective and expanding the application of semantic communication in eavesdropping scenarios.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and particularly to a design method for a semantic communication system oriented to task security. Background Art

[0002] With the continuous evolution of communication technologies towards 6G, traditional communication technologies based on Shannon's information theory are gradually approaching their performance boundaries and are increasingly difficult to meet the stringent requirements of 6G in terms of capacity and energy efficiency. As a novel communication paradigm, semantic communication is considered the key to solving the capacity bottleneck of traditional communication systems, thus attracting extensive attention from researchers. Different from traditional communication systems, semantic communication uses deep learning technologies to encode input signals into representations containing semantic information and then performs semantic interpretation at the receiving end.

[0003] Currently, with the rapid development of artificial intelligence technologies, significant progress has been made in the research of semantic communication based on deep learning in some aspects. However, in semantic communication, the inherent openness and easy accessibility of wireless channels make the transmitted semantic information vulnerable to interception by eavesdroppers. This not only endangers the privacy of senders but also poses significant security risks to legitimate users. Therefore, to ensure the security of semantic communication systems, it is necessary to implement strong security mechanisms to prevent the leakage of semantic information.

[0004] To ensure the security of semantic communication, researchers have proposed various semantic security schemes. For example, a new pixel-level secure mean squared error loss function is used to train semantic encoders and decoders, which can flexibly balance communication efficiency and privacy leakage to achieve secure semantic communication for image transmission. However, the above work only considers the semantic communication security of image reconstruction tasks and does not consider the security of downstream AI tasks. In contrast, another method that uses simultaneously transmitting and reflecting reconfigurable intelligent surfaces (STAR-RIS) in a semantic communication system to achieve privacy protection and reduce the task success rate of eavesdroppers reduces the task success rate of downstream AI tasks of eavesdroppers by converting signals into task-level interference specifically for eavesdroppers, but this method is achieved from a physical layer perspective rather than a semantic perspective. It can be seen that the above methods cannot solve the semantic communication security problem. Therefore, based on the above discussion, it is necessary to design a semantic communication system from a semantic perspective to ensure the performance security of downstream AI tasks. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the related technologies to some extent.

[0006] An object of the present invention is to provide a design method for a task-oriented secure semantic communication system, to establish a task-oriented secure semantic communication system based on a generative adversarial network. By integrating the generative adversarial network into the semantic communication system and designing a security loss function using the semantic-level information required for downstream AI tasks to train the semantic communication system, the performance security of downstream AI tasks is ensured, so as to achieve the goal of task-oriented secure semantic communication.

[0007] To achieve the above object, on the one hand, the present invention provides a design method for a task-oriented secure semantic communication system, including:

[0008] S1. Obtain an image dataset, divide the dataset into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively;

[0009] S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture of a sender, a receiver, and an eavesdropper;

[0010] S3. Use the training set to jointly pre-train the constructed semantic communication model;

[0011] S4. Load the pre-trained semantic communication model, and use the secure semantic feature loss function to fine-tune the parameters of the semantic communication model;

[0012] S5. Verify the effectiveness of the trained semantic communication model through the test set to obtain a task-oriented secure semantic communication system.

[0013] A further preferred technical solution of the present invention is that the data preprocessing of the images in the training set in step S1 includes:

[0014] Pad the images in the training set with zeros of a certain number of pixels around them, and then randomly crop a fixed-size area;

[0015] Horizontally flip the cropped images with a preset probability, and keep the rest of the images unchanged;

[0016] Randomly occlude a fixed-size area on the image;

[0017] Normalize the images in the training set.

[0018] Preferably, the data preprocessing of the images in the test set in step S1 includes:

[0019] Use the same mean and standard deviation as the images in the training set to normalize the images in the test set.

[0020] Preferably, the semantic communication model composed of the transmission architecture of the sender, the receiver, and the eavesdropper in step S2 is constructed as:

[0021] After the image at the sending end undergoes semantic encoding and channel encoding, a feature map is output. The feature map is perturbed by the generation network, and the dimension of the feature map remains unchanged after perturbation, which is expressed as:

[0022] ;

[0023] Among them, is the semantic encoder, is the input image, are the neural network parameters of the semantic encoder: is the channel encoder, are the neural network parameters of the channel encoder; is the generation network, are the neural network parameters of the generation network;

[0024] Subsequently, the input signal is transmitted through the constructed AWGN channel, and the signals received by the receiving end and the eavesdropping end are respectively expressed as:

[0025] ;

[0026] ;

[0027] Among them, and respectively represent the signals received by the receiving end and the eavesdropping end, is the input signal of the channel, and respectively represent the independent and identically distributed Gaussian noises of the receiving end and the eavesdropping end;

[0028] The signals received by the receiving end and the eavesdropping end are reconstructed into images through the channel decoder and the semantic decoder. The image reconstruction process is expressed as:

[0029] ;

[0030] ;

[0031] Among them, and are the reconstructed images of the receiving end and the eavesdropping end respectively; and are the channel decoder and the semantic decoder respectively, and are the neural network parameters of the channel decoder of the receiving end and the eavesdropping end respectively, and are the neural network parameters of the semantic decoder of the receiving end and the eavesdropping end respectively.

[0032] Preferably, step S3 uses the semantic communication model constructed by joint pre-training with the training set, specifically: using the mean square error loss function based on pixels to pre-train the semantic communication model composed of the transmission architectures of the sender, receiver, and eavesdropper. The mean square error loss function based on pixels is expressed as:

[0033] ;

[0034] where MSE represents the calculation of the mean square error.

[0035] Preferably, in step S4, after loading the pre-trained semantic communication model, the security semantic feature loss function is used to fine-tune the parameters of the semantic communication model, specifically including:

[0036] Fine-tuning the semantic encoder and channel encoder of the sender, as well as the semantic decoder and channel decoder of the receiver and eavesdropper, using the security semantic feature loss function;

[0037] When using the classification task as the downstream AI task, the original image, the reconstructed images of the receiver and eavesdropper are respectively input into the feature extraction network, which has been pre-trained for the classification task in advance. The security semantic feature loss function is expressed as:

[0038] ;

[0039] where is the th feature map of the th original image, and are respectively the th feature map of the reconstructed image of the receiver and the th feature map of the reconstructed image of the eavesdropper, is the task weight for controlling the balance, is the data volume of each batch.

[0040] Preferably, the receiver of the semantic communication model has a discriminative network corresponding to the generative network of the sender. After the signals received at the receiver and eavesdropper are reconstructed into images through the channel decoder and semantic decoder in step S2, the reconstructed images are input into the discriminative network of the receiver for adversarial training in combination with the generative network.

[0041]

[0041] Preferably, when fine-tuning the parameters of the semantic communication model in step S4, it also includes:

[0042] Jointly training the generative network of the sender and the discriminative network of the receiver using the generative adversarial network loss function;

[0043] The optimization objective of the generation network is to maximize the output of the reconstructed image at the receiving end by the discrimination network, thus deceiving the discrimination network; the optimization objective of the discrimination network is to make the output of the original image as large as possible and the output of the reconstructed image at the receiving end as small as possible to distinguish it from the original image; the loss function of the generative adversarial network is expressed as:

[0044] ;

[0045] ;

[0046] wherein, is the loss function of the generation network, is the loss function of the discrimination network, represents the discrimination network, are the neural network parameters of the discrimination network, is the mean operation, is the hyperparameter of the regularization term, is used to limit the perturbation amplitude norm.

[0047] Preferably, step S5 verifies the effectiveness of the trained semantic communication model through a test set, including:

[0048] From a semantic perspective, FPSNR is used as a performance evaluation index, expressed as:

[0049] ;

[0050] wherein, represents the maximum activation value of the original image feature map; represents the mean square error based on the feature level, expressed as , is the feature map extracted from the original image after passing through the feature extraction network, is the feature map extracted from the reconstructed image after passing through the feature extraction network.

[0051] On the other hand, the present invention provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions enable the computer to execute the above-mentioned design method of the task-oriented secure semantic communication system.

[0052] On yet another aspect, the present invention provides an electronic device, including: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus, and the processor calls the logical instructions in the memory to execute the above-mentioned design method of the task-oriented secure semantic communication system.

[0053] In another aspect, the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned method for designing a task-oriented secure semantic communication system.

[0054] Beneficial effects: The present invention discloses a method for designing a task-oriented secure semantic communication system, aiming at the information security problem in the process of image transmission. The method constructs a secure semantic communication architecture including a sender, a receiver, and an eavesdropper, and uses semantic coding, channel coding, and generative adversarial network technologies to extract features and perturb the image. During the transmission process, the present invention uses an AWGN channel to simulate the actual communication environment, and performs signal processing and image reconstruction at the receiver and the eavesdropper respectively. By designing a secure semantic feature loss function and a generative adversarial network loss function, the present invention realizes the joint optimization training of the system, effectively improves the image reconstruction quality of the legitimate receiver, and reduces the reconstruction effect of the eavesdropper at the same time. Finally, the present invention uses the feature-level peak signal-to-noise ratio index to verify the effectiveness of the model, provides a new solution for secure image transmission, and realizes information security while ensuring communication quality. Description of the Drawings

[0055] Figure 1 It is a flowchart of the method for designing a task-oriented secure semantic communication system of the present invention;

[0056] Figure 2 It is a framework diagram of the semantic communication model in the present invention;

[0057] Figure 3 It is a model framework diagram of the sender in Embodiment 1 of the present invention;

[0058] Figure 4 It is a model framework diagram of the receiver in Embodiment 1 of the present invention;

[0059] Figure 5 It is a comparison diagram of FPSNR values of the method of the present invention and other methods under different signal-to-noise ratio conditions in an AWGN channel;

[0060] Figure 6 It is a comparison diagram of classification accuracies of the method of the present invention and other methods under different signal-to-noise ratio conditions in an AWGN channel. Detailed Embodiments

[0061] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention, and they should not be construed as limiting the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts belong to the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be construed as indicating or implying relative importance.

[0062] The following combines Figures 1-6 to describe the design method of a task-oriented secure semantic communication system provided by the present invention.

[0063] Embodiment 1: This embodiment provides a design method for a task-oriented secure semantic communication system. The overall process of this method is as Figure 1 shown and includes:

[0064] S1. Obtain an image dataset, divide the dataset into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively;

[0065] S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture with a sender, a receiver, and an eavesdropper;

[0066] S3. Use the training set to jointly pre-train the constructed semantic communication model;

[0067] S4. Load the pre-trained semantic communication model and design a comprehensive loss function to fine-tune the parameters of the semantic communication model;

[0068] S5. Verify the effectiveness of the trained semantic communication model through the test set to obtain a task-oriented secure semantic communication system.

[0069] The following will explain each step in detail.

[0070] In step S1, first obtain a dataset. This dataset can be directly obtained from the network or constructed according to needs. Then divide the dataset into a training set and a test set.

[0071] Perform preprocessing on the dataset, including preprocessing of the training set and preprocessing of the test set.

[0072] For the images in the training set, first fill 4 pixels of zero values around the image, and then randomly crop a 32×32 area; then flip the image horizontally with a probability of 50%, keep the remaining 50% unchanged, and randomly occlude a 16×16 square area on the image. Finally, perform normalization processing on the image.

[0073] Only normalize the images in the test set, using the same mean and standard deviation as the training set.

[0074] Step S2 is to construct the channel environment and design the transmission architecture related to the semantic communication model, such as Figure 2 shown, including a transmitter, a receiver, and an eavesdropper.

[0075] The transmitter in this embodiment is as Figure 3 shown, and is composed of a semantic encoder, a channel encoder, and a generation network. Among them, the semantic encoder is used to extract the feature map of the input training set image, and the channel encoder reduces the dimension of the feature map output by the semantic encoder for subsequent transmission in the physical channel. Then the output feature map passes through a generation network composed of 4 3×3 convolutional layers, and the generation network adds perturbations to it, and the dimension of the feature map remains unchanged after adding perturbations. The whole process is expressed as:

[0076] ;

[0077] Among them, is the semantic encoder, is the input image, are the neural network parameters of the semantic encoder: is the channel encoder, are the neural network parameters of the channel encoder; is the generation network, are the neural network parameters of the generation network.

[0078] Subsequently, the input signal is transmitted through the constructed AWGN channel, and the signals received by the receiver and the eavesdropper are respectively expressed as:

[0079] ;

[0080] ;

[0081] Among them, and respectively represent the signals received by the receiver and the eavesdropper, is the input signal of the channel, and respectively represent the independent and identically distributed Gaussian noises of the receiver and the eavesdropper.

[0082] The receiver in this embodiment is as Figure 4As shown in the figure, it consists of a channel decoder, a semantic decoder, and a discriminant network. The channel decoder upsamples the received signal and then inputs it into the semantic decoder for image reconstruction. The reconstructed image also needs to be input into the discriminant network to prepare for adversarial training of the subsequent joint generation network. The architecture of the eavesdropping end is basically the same as that of the receiving end, but it only has a channel decoder and a semantic decoder. The image reconstruction process can be expressed as:

[0083] ;

[0084] ;

[0085] Among them, and are the reconstructed images of the receiving end and the eavesdropping end respectively; and are the channel decoder and the semantic decoder respectively, and are the neural network parameters of the channel decoders of the receiving end and the eavesdropping end respectively, and are the neural network parameters of the semantic decoders of the receiving end and the eavesdropping end respectively.

[0086] Step S3 is to jointly pre-train the constructed semantic communication model using the training set.

[0087] Based on the semantic communication model constructed in step S2, use the pixel-level mean squared error (MSE) loss function to pre-train the semantic communication model framework including the sending end, the receiving end, and the eavesdropping end to ensure the image reconstruction quality of the receiving end and the eavesdropping ability of the eavesdropping end. The pixel-level mean squared error loss function is expressed as:

[0088] ;

[0089] Among them, MSE represents the calculation of the mean squared error.

[0090] Step S4 is to load the pre-trained model and fine-tune the training model using the generative adversarial network loss function and the secure semantic feature loss function.

[0091] Based on the obtained pre-trained model, use specialized loss functions to train and fine-tune different architecture parts.

[0092] For the semantic encoder and the channel encoder of the sending end, and the semantic decoder and the channel decoder of the receiving end and the eavesdropping end, use the secure semantic feature loss function composed of feature maps for fine-tuning. In this embodiment, taking the classification task as an example of the downstream AI task, input the original image, the reconstructed images of the receiving end and the eavesdropping end into the feature extraction network respectively. The feature extraction network has been pre-trained for the classification task in advance. The secure semantic feature loss function is expressed as:

[0093] ;

[0094] in, For the The original image feature map, and Respectively The receiving end reconstructs the image and The eavesdropping end reconstructs the image feature map, is the task weight that controls the balance, is the amount of data in each batch.

[0095] The generative network at the sender and the discriminative network at the receiver are jointly trained using the generative adversarial network loss function. The principle is that for the perturbed feature map generated by the generative network at the sender, the goal is to jointly perform adversarial training with the discriminative network at the receiver to ensure that the generated perturbation has minimal impact on the image reconstruction task at the receiver, while maintaining the image reconstruction quality at the receiver as much as possible. However, since the eavesdropper lacks a discriminative network, the perturbation will reduce the image reconstruction quality at the eavesdropping end. The optimization goal of the generative network is to maximize the output of the discriminative network at the receiver to reconstruct the image, thereby deceiving the discriminative network; the optimization goal of the discriminative network is to make the output of the original image as large as possible and the output of the reconstructed image at the receiver as small as possible to distinguish it from the original image. Therefore, the generative adversarial network loss function is expressed as:

[0096] ;

[0097] ;

[0098] in, is the loss function of the generated network, is the loss function of the discriminant network, represents the discriminant network, is the neural network parameter of the discriminant network; To take the mean value, is the hyperparameter of the regularization term, To limit the disturbance amplitude Norm.

[0099] The ultimate training goal is to ensure that the perturbations introduced by the generator network reduce the image reconstruction quality of the eavesdropping end while maintaining the image reconstruction quality of the receiving end as much as possible. Finally, the overall comprehensive loss function of the model is expressed as:

[0100] .

[0101] Step S5: Input the test set data into the network, output the results, and verify the effectiveness of the model.

[0102] For the evaluation of the reconstructed image quality, since the model framework is trained using a feature-level secure semantic feature loss function, the traditional pixel-level PSNR (Peak Signal-to-Noise Ratio) may not be suitable for evaluating the image reconstruction quality under this model. In view of this, a new performance metric FPSNR is proposed from a semantic perspective, which is expressed as:

[0103] ;

[0104] where represents the maximum activation value of the original image feature map; represents the feature-level mean square error, which is expressed as is the feature map extracted from the original image after passing through the feature extraction network, is the feature map extracted from the reconstructed image after passing through the feature extraction network. For the classification task evaluation, the classification accuracy metric is used, which is defined as the ratio of the number of correctly predicted samples to the total number of samples.

[0105] The following further illustrates the effect of the task-oriented secure semantic communication system constructed by the present invention in combination with simulation experiments.

[0106] 1. Simulation conditions and parameter settings:

[0107] The simulation experiment of the present invention is carried out on a simulation platform based on Python 3.9 and Pytorch 1.13.0+cu117. The dataset used is the CIFAR-10 dataset, which contains 60,000 32×32 color images, divided into 10 categories, with 6,000 images in each category. It is divided into 50,000 training sets (including 5 sub-batches) and 10,000 test sets. The pre-trained ResNet-18 is used as the feature extraction network for the classification task and the classifier for the downstream AI task. In the simulation of this embodiment, as Figure 3 and Figure 4 shown, the semantic encoder consists of a 5×5 convolutional layer and three residual blocks, the semantic decoder consists of a 5×5 convolutional layer, three residual blocks and a 3×3 convolutional layer, and the discriminant network consists of 2 3×3 convolutional layers and a fully connected layer. The channel encoder and the channel decoder are a 3×3 convolutional neural network with 64 neurons in each layer, both using the ReLU activation function, and the semantic decoder and the channel decoder architectures at the receiving end and the eavesdropping end are the same. For the channel environment, in this embodiment, it is set that the channel environment conditions at the receiving end and the eavesdropping end are the same.

[0108] ​The system is pre-trained in the pre-training stage using a pixel-level MSE loss function, with the Adamw optimizer and a learning rate of . After the training loss value converges, the obtained parameters are loaded, and then the network is fine-tuned. For the security semantic feature loss function, is set to 0.5, and the learning rate is ; for the generative adversarial network, the learning rate of the generator network is , and the learning rate of the discriminator network is . The data volume of each batch is 256.

[0109] 2. Simulation content:

[0110] In the simulation of the present invention, the comparison benchmarks all have a receiver and an eavesdropper. Figure 5 shows the relationship between the FPSNR value and the signal-to-noise ratio of the receiver and eavesdropper of different benchmarks in the AWGN channel environment. Figure 5 The abscissa in Figure 5 represents different signal-to-noise ratios (dB), ranging from -6 to 9 dB in steps of 3 dB, and the ordinate represents the FPSNR value. The broken line marked with a rectangle represents the FPSNR value curve of the receiver using the method of the present invention, the broken line marked with a circle represents the FPSNR value curve of the eavesdropper using the method of the present invention, and the broken lines marked with triangles and inverted triangles respectively represent the FPSNR value curves of the receiver and eavesdropper of the pixel-level MSE security loss function. The system architecture of the pixel-level benchmark is the same as that of the system in this embodiment, except that the pixel-level security loss function is used, and the channel environment and parameter settings are the same as those of the method proposed in the present invention. It can be seen from

[0111] Figure 6 that as the signal-to-noise ratio increases, the FPSNR value of the eavesdropper using the method of the present invention is always lower than that of the pixel-level eavesdropper, because the pixel-level benchmark does not consider the semantic-level information required for the classification task. Further, at each signal-to-noise ratio, the FPSNR value of the receiver using the method of the present invention is always about 4 dB higher than that of the eavesdropper using the method of the present invention. It can also be seen that compared with the pixel-level benchmark, the FPSNR value of the receiver using the method of the present invention is always higher than that of the pixel-level receiver. The above results show that the method of the present invention effectively reduces the semantic information of the reconstructed image at the eavesdropper, reduces its image reconstruction quality, and at the same time ensures the image reconstruction quality of the receiver. It also shows that the pixel-level security loss function benchmark is not as good as the method of the present invention.

[0111] Figure 6 shows the relationship between the classification accuracy and the signal-to-noise ratio of the method of the present invention and the pixel-level benchmark method in the AWGN channel environment. Figure 6 The ordinate in Figure 5Keep consistent. It can be seen from Figure 6 that when the signal-to-noise ratio is 9 dB, the classification accuracy of the receiving end of the method of the present invention reaches about 80%, while the eavesdropping end of the method of the present invention performs poorly, and the classification accuracy is less than 20%. And the classification accuracy of the eavesdropping end based on the pixel-level benchmark always remains at a relatively high level at each signal-to-noise ratio. This indicates that the pixel-level-based benchmark cannot guarantee the performance security of the classification task, and as the signal-to-noise ratio decreases, the classification accuracy of the receiving end of the method of the present invention still maintains a relatively high precision performance, while the classification accuracy of the eavesdropping end of the method of the present invention always remains at a low level, less than 20%. This proves that the method of the present invention effectively reduces the classification accuracy of the eavesdropping end and realizes the security of the semantic communication system in terms of the classification task.

[0112] Based on the above simulation results and analysis, the design method of the task-oriented secure semantic communication system based on the generative adversarial network proposed by the present invention can effectively ensure the performance security of downstream AI tasks, which enables the present invention to be better applied in actual communication scenarios.

[0113] Embodiment 2: This embodiment provides a non-transitory computer-readable storage medium, on which computer instructions are stored, and the computer instructions cause the computer to execute the design method of the task-oriented secure semantic communication system, and the method includes the following steps:

[0114] S1. Obtain an image data set, divide the data set into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively;

[0115] S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture of a sending end, a receiving end, and an eavesdropping end;

[0116] S3. Use the training set to jointly pre-train the constructed semantic communication model;

[0117] S4. Load the pre-trained semantic communication model, and design a comprehensive loss function to fine-tune the parameters of the semantic communication model;

[0118] S5. Verify the effectiveness of the trained semantic communication model through the test set to obtain a task-oriented secure semantic communication system.

[0119] Embodiment 3: This embodiment provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete mutual communication through the communication bus. The processor can call the logical instructions in the memory to execute the design method of the task-oriented secure semantic communication system, and the method includes the following steps:

[0120] S1. Obtain an image dataset, divide the dataset into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively;

[0121] S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture of a transmitter, a receiver, and an eavesdropper;

[0122] S3. Use the training set to jointly pre-train the constructed semantic communication model;

[0123] S4. Load the pre-trained semantic communication model and design a comprehensive loss function to fine-tune the parameters of the semantic communication model;

[0124] S5. Verify the effectiveness of the trained semantic communication model through the test set to obtain a task-oriented secure semantic communication system.

[0125] In addition, when the logical instructions in the above-mentioned memory are implemented in the form of software functional units and sold or used as an independent product, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0126] Embodiment 4: This embodiment provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for designing a task-oriented secure semantic communication system. The method includes the following steps:

[0127] S1. Obtain an image dataset, divide the dataset into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively;

[0128] S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture of a transmitter, a receiver, and an eavesdropper;

[0129] S3. Use the training set to jointly pre-train the constructed semantic communication model;

[0130] S4. Load the pre-trained semantic communication model, and design a comprehensive loss function to fine-tune the parameters of the semantic communication model;

[0131] S5. Verify the effectiveness of the trained semantic communication model through a test set to obtain a semantic communication system for task security.

[0132] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

[0133] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on such an understanding, the above technical solutions, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or equivalently replace some of the technical features. These modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A design method for a task-oriented secure semantic communication system, characterized in that Including: S1. Obtain an image dataset, divide the dataset into a training set and a test set, and perform data preprocessing on the images in the training set and the test set respectively; S2. Construct a channel environment and design a semantic communication model composed of a transmission architecture of a transmitter, a receiver, and an eavesdropper; the semantic communication model composed of the transmission architecture of the transmitter, the receiver, and the eavesdropper is constructed as follows: After the image at the transmitter undergoes semantic encoding and channel encoding, a feature map is output. The feature map is perturbed by a generation network, and the dimension of the feature map remains unchanged after perturbation, which is expressed as: ； Among them, is a semantic encoder, is the input image, is the neural network parameter of the semantic encoder: is a channel encoder, is the neural network parameter of the channel encoder; is a generation network, is the neural network parameter of the generation network; Subsequently, the input signal is transmitted through the constructed AWGN channel, and the signals received by the receiver and the eavesdropper are respectively expressed as: ； ； Among them, and respectively represent the signals received by the receiving end and the eavesdropping end, is the input signal of the channel, and respectively represent the independent and identically distributed Gaussian noises of the receiving end and the eavesdropping end; The signals received by the receiver and the eavesdropper are reconstructed into images through a channel decoder and a semantic decoder, and the image reconstruction process is expressed as: ； ； wherein, and are the reconstructed images at the receiving end and the eavesdropping end respectively; and are the channel decoder and the semantic decoder respectively, and are the neural network parameters of the channel decoders at the receiving end and the eavesdropping end respectively, and are the neural network parameters of the semantic decoders at the receiving end and the eavesdropping end respectively; S3. Jointly pre-train the constructed semantic communication model using the training set; S4. Load the pre-trained semantic communication model and fine-tune the parameters of the semantic communication model using a secure semantic feature loss function; specifically including: Fine-tune the semantic encoder and channel encoder at the transmitter, and the semantic decoder and channel decoder at the receiver and the eavesdropper using the secure semantic feature loss function; When using a classification task as a downstream AI task, input the original image, the reconstructed images at the receiver and the eavesdropper into a feature extraction network respectively. The feature extraction network has been pre-trained for the classification task in advance, and the secure semantic feature loss function is expressed as: ； in, For the The original image feature map, and Respectively The receiving end reconstructs the image and The eavesdropping end reconstructs the image feature map, is the task weight that controls the balance, is the amount of data in each batch; The receiver of the semantic communication model has a discriminant network corresponding to the generation network at the transmitter. After the signals received by the receiver and the eavesdropper are reconstructed into images through a channel decoder and a semantic decoder in step S2, the reconstructed images are input into the discriminant network at the receiver, and adversarial training is performed jointly with the generation network; Jointly train the generation network at the transmitter and the discriminant network at the receiver using a generative adversarial network loss function; The generative adversarial network loss function is expressed as: ； ； Among them, is the loss function of the generation network, is the loss function of the discriminative network, represents the discriminative network, is the neural network parameter of the discriminative network, is the operation of taking the mean, is the hyperparameter of the regularization term, is used to limit the perturbation amplitude of norm; S5. Verify the effectiveness of the trained semantic communication model through the test set to obtain a task-oriented secure semantic communication system.

2. The design method of the task-oriented secure semantic communication system according to claim 1, characterized in that The data preprocessing of the images in the training set in step S1 includes: Pad the images in the training set with zeros for a certain number of pixels around them, and then randomly crop a fixed-size area; Horizontally flip the cropped images with a preset probability, and keep the rest of the images unchanged; Randomly occlude a fixed-size area on the images; Normalize the images in the training set.

3. The design method of the task-oriented secure semantic communication system according to claim 2, wherein The data preprocessing of the images in the test set in step S1 includes: Normalize the images in the test set using the same mean and standard deviation as the images in the training set.

4. The design method of the task-oriented secure semantic communication system according to claim 1, wherein, Step S3 uses the training set to jointly pre-train the constructed semantic communication model. Specifically: use a pixel-level mean squared error loss function to pre-train the semantic communication model composed of the transmission architecture of the transmitter, the receiver, and the eavesdropper. The pixel-level mean squared error loss function is expressed as: ； Where MSE represents the calculation of the mean squared error.

5. The method for designing a task-oriented secure semantic communication system according to claim 1, wherein Step S5 verifies the effectiveness of the trained semantic communication model through the test set, including: From a semantic perspective, use FPSNR as a performance evaluation index, which is expressed as: ； Among them, represents the maximum activation value of the original image feature map; represents the mean square error based on the feature level, expressed as , is the feature map extracted after the original image passes through the feature extraction network, is the feature map extracted after the reconstructed image passes through the feature extraction network.

6. An electronic device, characterized in that, Including: A processor, a communication interface, a memory, and a communication bus. Among them, the processor, the communication interface, and the memory complete communication with each other through the communication bus. The processor calls the logical instructions in the memory to execute the method for designing a task-oriented secure semantic communication system described in any one of claims 1-5.

Images