Secret key lossless transmission method based on semantic communication

Abstract

The invention discloses a secret key lossless transmission method based on semantic communication, and the method comprises the steps: 1, segmenting a to-be-transmitted secret key according to batches, adding a batch mark in front of each segment, and generating a corresponding two-dimensional code image through two-dimensional code coding; 2, a channel noise model is adopted to simulate the transmission process of key data, channels comprise additive Gaussian white noise and Rayleigh fading channels, the noise injection intensity is controlled by setting the signal-to-noise ratio, and the influence of channel noise on the key data is simulated; step 3, the receiving end uses the semantic communication model to restore and reconstruct the two-dimensional code image, and the definition and identifiability of the damaged two-dimensional code are improved; step 4, decoding the reconstructed two-dimensional code image, and recovering corresponding secret key data; and step 5, calculating and comparing the hash values of the restored key and the original key, and verifying the content integrity. According to the method, the recovery capability of the two-dimensional code under the condition of low signal-to-noise ratio is enhanced by introducing the semantic communication model, and the method has good robustness.

Description

Technical Field

[0001] This invention relates to the field of network information security technology, specifically to a lossless key transmission method based on semantic communication. Background Technology

[0002] Public key infrastructure (PKI) is the cornerstone of modern cybersecurity, and its core component—the secure and reliable distribution of keys—faces severe challenges in practical deployments. Existing distribution methods based on online protocols or traditional encrypted channels heavily rely on stable network connections and bit-level precision transmission. Their performance deteriorates sharply in complex environments such as offline, weak network, or low signal-to-noise ratio scenarios, making it difficult to meet the needs of edge computing, industrial IoT, and other scenarios.

[0003] Although emerging semantic communication paradigms demonstrate excellent robustness in fields such as images and speech by transmitting and understanding the "semantics" of information rather than precise bits, providing new ideas for solving the aforementioned problems, they face a fundamental contradiction when directly applied to key distribution: the key requires that the transmission result be completely consistent at the bit level, and any tiny error will cause the key to fail; while the goal of traditional semantic communication is "semantic similarity", and its inherent characteristics cannot guarantee this absolute bit precision, thus making it difficult to meet the rigid requirement of "lossless" key pair transmission. Summary of the Invention

[0004] This invention provides a lossless key transmission method based on semantic communication, the core of which lies in utilizing the robustness and adaptability of semantic communication to achieve reliable key transmission in noisy channels.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a lossless key transmission method based on semantic communication, comprising the following steps:

[0006] Step 1, Key Semantic Encapsulation and Carrier Construction: The original key to be transmitted is divided into multiple key fragments according to a preset batch size; a batch marker is added to each key fragment to obtain a marked key fragment; the marked key fragment is encoded into a QR code image to construct a structured semantic information carrier.

[0007] Step 2, Semantic Information Transmission Degradation Simulation: The structured semantic information carrier is input into the channel simulation model, which supports additive white Gaussian noise channel and Rayleigh fading channel. The noise intensity is controlled by the set signal-to-noise ratio, and the degradation process of semantic information in wireless transmission is simulated in the feature domain to generate degraded semantic information.

[0008] Step 3, Degraded semantic information reconstruction: The degraded semantic information is input into the semantic communication model of the receiving end for reconstruction; the semantic communication model includes a semantic encoder, a channel module and a semantic decoder connected in sequence; the semantic encoder is used to extract semantic features and perform channel adaptive modulation; the channel module is used to introduce transmission noise consistent with Step 2 into the feature domain; the semantic decoder is used to decode and reconstruct the damaged semantic features and output the reconstructed QR code image;

[0009] Step 4, Semantic parsing and key recovery: Decode the reconstructed QR code image to extract the marked key fragments, and concatenate them according to the order of the batch markings to obtain the recovery key;

[0010] Step 5, Integrity Verification: Calculate and compare the hash values ​​of the original key and the recovery key respectively. If they match, the transmission is deemed lossless and complete.

[0011] Preferably, step 1 specifically includes:

[0012] Step 1.1: Read the original key and divide it into multiple key fragments according to the preset batch size;

[0013] Step 1.2: Add a batch marker to each of the key segments, the batch marker containing the sequence number of the segment and the total number of segments;

[0014] Step 1.3: Encode each key fragment with batch markers into a QR code image to obtain a set of structured semantic information carriers.

[0015] Preferably, step 2 specifically includes:

[0016] Step 2.1: Construct a channel simulation module that supports additive white Gaussian noise channels and Rayleigh fading channels;

[0017] Step 2.2: When the additive white Gaussian noise channel is selected, zero-mean white Gaussian noise is generated according to the set signal-to-noise ratio, and the noise is superimposed on the transmission feature corresponding to the structured semantic information carrier.

[0018] Step 2.3: When a Rayleigh fading channel is selected, a Rayleigh-distributed fading coefficient is generated, and amplitude modulation is performed on the transmission characteristics corresponding to the structured semantic information carrier based on the fading coefficient.

[0019] Step 2.4: The channel simulation module processes the data to generate the degraded semantic information in the semantic feature space.

[0020] Preferably, the semantic communication model in step 3 includes:

[0021] The semantic encoder is configured to extract and adaptively modulate multi-scale semantic features; the channel module is configured to receive the modulated semantic features and simulate the transmission degradation defined in step 2 in the feature domain; and the semantic decoder is configured to reconstruct the degraded semantic features and restore them to a QR code image.

[0022] Preferably, step 4 specifically includes:

[0023] Step 4.1: Detect and decode the reconstructed QR code image;

[0024] Step 4.2: If decoding fails, perform image preprocessing on the reconstructed QR code image and try decoding again. The image preprocessing includes one or more of threshold segmentation, filtering, sharpening, or image scaling.

[0025] Step 4.3: Extract the batch marker from the successfully decoded content, and concatenate the key fragments according to the order indicated by the batch marker to form the recovery key.

[0026] Preferably, step 5 specifically includes:

[0027] Step 5.1: Calculate the hash values ​​of the original key and the recovery key using the same hash algorithm;

[0028] Step 5.2: Compare the hash values ​​of the original key and the recovery key; if they match, it is determined to be lossless transmission.

[0029] Step 5.3: If the hash values ​​are inconsistent, generate an integrity report containing information on failed batches or data differences.

[0030] Preferably, the semantic encoder in step 3 includes:

[0031] The semantic feature extraction module is used to extract multi-scale semantic features based on the window attention mechanism.

[0032] The channel adaptive modulation module is used to dynamically adjust the expression intensity of the semantic features based on the current signal-to-noise ratio using a learnable modulation vector.

[0033] The feature compression module is used to map high-dimensional semantic features to a low-dimensional feature space suitable for channel transmission.

[0034] Preferably, the semantic decoder in step 3 includes:

[0035] The noise-robust feature processing module is used to denoise and enhance the received noisy semantic features; the semantic reconstruction module is used to gradually restore the spatial resolution of semantic information using a reverse hierarchical structure; and the pixel mapping module is used to map the reconstructed semantic features to the reconstructed QR code image.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] (1) This invention introduces a semantic communication model to extract multi-scale features of QR code images and constructs a channel noise simulation module during transmission, so that the model can learn and adapt to the degradation law of images under different signal-to-noise ratio conditions, thereby realizing high-quality reconstruction of degraded QR code images.

[0038] (2) The present invention adopts a batch marking and SHA256 hash comparison mechanism to ensure that the restored key is completely consistent with the original key in terms of logical order, character content and integrity, thus achieving true lossless transmission.

[0039] (3) In response to adverse conditions such as blurriness, low resolution, and noise pollution, this invention adopts a multi-strategy enhanced decoding method, such as adaptive thresholding, filtering, sharpening, and multi-scale scaling, which greatly improves decoding accuracy and stability.

[0040] (4) The semantic communication model has the feature expression capability at the construction level, which enables it to maintain high reconstruction performance when facing different types and degrees of channel degradation, and has both versatility and scalability. Attached Figure Description

[0041] Figure 1 This is a flowchart of a lossless key transmission method based on semantic communication according to the present invention;

[0042] Figure 2 This is an encoder structure diagram of a lossless key transmission method based on semantic communication according to the present invention;

[0043] Figure 3 This is a decoder structure diagram of a lossless key transmission method based on semantic communication according to the present invention;

[0044] Figure 4 This is a channel noise simulation module diagram of a lossless key transmission method based on semantic communication according to the present invention;

[0045] Figure 5 This invention relates to a method for generating integrity verification reports using a lossless key transmission method based on semantic communication. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0047] like Figure 1-5 As shown, the present invention provides a lossless key transmission method based on semantic communication, which is implemented according to the following steps:

[0048] Step 1, Key Semantic Encapsulation and Carrier Construction: The original key to be transmitted is divided into multiple key fragments according to a preset batch; a batch identifier "[Batch i / n]" is added before each key fragment, and a corresponding QR code image is generated through a QR code encoding algorithm to construct a structured semantic information carrier for subsequent semantic communication;

[0049] Step 2, Semantic Information Transmission Degradation Simulation: For the structured semantic information carrier generated in Step 1, a channel simulation model is used to simulate the degradation of the semantic information during wireless transmission. The channel includes an additive white Gaussian noise channel and a Rayleigh fading channel. By setting the signal-to-noise ratio to control the noise injection intensity, degraded semantic information is generated in the feature domain.

[0050] Step 3, Degraded semantic information reconstruction: The degraded semantic information obtained in Step 2 is input into the semantic communication model of the receiving end for reconstruction. A semantic communication model containing an encoder, a channel module and a decoder is constructed. The encoder has a Transformer structure with a sliding window multi-head attention mechanism to extract features and perform signal-to-noise ratio adaptive modulation. The decoder realizes image reconstruction through reverse merging (Patch) and sub-pixel rearrangement mechanism and outputs the reconstructed QR code image.

[0051] Step 4, Semantic parsing and key recovery: The reconstructed QR code image output in Step 3 is quickly detected and recognized. If the QR code detector fails to decode, multiple image enhancement algorithms are automatically called to supplement the decoding. The parsed key data is concatenated according to the marked order to obtain the recovery key.

[0052] Step 5, Integrity Verification and Report Generation: Calculate the hash values ​​of the original key and the recovery key obtained in Step 4 respectively, compare the hash results of the two, and if they match, it is determined that the transmission is lossless and complete.

[0053] Furthermore, step 1 is specifically implemented according to the following steps:

[0054] Step 1.1: Read the original key to be transmitted and divide the data into multiple key segments by line according to the set batch size;

[0055] Step 1.2: Add a batch marker (e.g., Batch i / n) at the beginning of each key segment, where i is the current batch number and n is the total number of batches, forming a marked key segment, which can avoid disordered order and lost key segments;

[0056] Step 1.3: Call the QR code generation library to encode each marked key fragment into a QR code image, generating a file such as: origin_20250101_batch01_of_08.png, which is a set of structured semantic information carriers;

[0057] Furthermore, step 2 is specifically implemented according to the following steps:

[0058] Step 2.1: Construct a channel simulation module that supports two types of channel models: additive white Gaussian noise channel and Rayleigh fading channel. This module is used to simulate degradation in real communication processes and uses the structured semantic information carrier obtained in Step 1 as the input to the transmitter.

[0059] Step 2.2: If an Additive White Gaussian Noise (AWGN) channel is used, zero-mean complex Gaussian noise is generated according to the set noise power, and the noise is superimposed on the transmission feature corresponding to the semantic information carrier output in Step 1 to simulate the random noise interference that the semantic information is subjected to in an ideal wireless channel.

[0060] Step 2.3: If a Rayleigh channel is used, the fading coefficients of the Rayleigh distribution are generated based on two independent zero-mean Gaussian random components. The transmission characteristics corresponding to the semantic information carrier output in Step 1 are amplitude modulated to simulate the transmission degradation process of semantic information under multipath fading environment.

[0061] Step 2.4: Based on the set signal-to-noise ratio (SNR), noise is injected into the encoded features in the feature space to generate corresponding degraded semantic information in the semantic feature space.

[0062] Furthermore, step 3 is specifically implemented according to the following steps:

[0063] Step 3.1: The semantic encoder uses the Swing Transformer to extract multi-scale structural features of the QR code, uses a sliding window multi-head attention mechanism to capture local and global information, and introduces an SNR adaptive modulation module to map the signal-to-noise ratio information to the feature vector to achieve dynamic modulation.

[0064] Step 3.2: Map the degraded semantic information obtained in Step 2 to the channel module of the semantic communication model, and introduce transmission noise consistent with that in Step 2 into the feature domain to obtain the degraded features;

[0065] Step 3.3: The semantic decoder uses a reverse patch merging structure to restore the image resolution layer by layer, and uses sub-pixel rearrangement technology to improve the detail restoration of the reconstruction, and finally outputs a high-fidelity reconstructed QR code image.

[0066] Furthermore, step 4 is specifically implemented according to the following steps:

[0067] Step 4.1: Perform preliminary decoding on the reconstructed QR code image obtained in Step 3;

[0068] Step 4.2: If decoding fails, the reconstructed QR code image will enter an enhanced decoding process, including adaptive thresholding, Gaussian or median filtering for noise reduction, and image sharpening to enhance edges.

[0069] Step 4.3: Parse the batch marker "[Batch i / n]" from the successfully decoded content, and concatenate the key fragments according to the order indicated by the batch marker to obtain the recovery key.

[0070] Furthermore, step 5 is specifically implemented according to the following steps:

[0071] Step 5.1: Calculate the hash values ​​of the original key and the recovery key using the SHA256 hash algorithm;

[0072] Step 5.2: If the two hash values ​​match, it means the transmission was lossless; otherwise, mark the batch as failed and output a description of the difference.

[0073] Step 5.3: If the hash values ​​are inconsistent, generate an integrity report containing information on failed batches or data differences; the integrity report includes the generation and decoding status of QR codes for each batch, matching rate statistics, model reconstruction quality indicators, and final integrity conclusions.

[0074] To illustrate the method of the present invention, a specific example is provided:

[0075] First, corresponding to step 1, the input key data is in .txt format. The original key is divided into multiple key fragments according to a preset batch size. A unique batch marker [Batch i / n] is added to each key fragment, where i is the batch number and n is the total batch number. This marker is used to ensure the order of the original key during subsequent reassembly. Each marked key fragment generates a corresponding QR code image through a QR code encoding function, thus obtaining a set of structured semantic information carriers.

[0076] Secondly, corresponding to step 2, noise and fading during wireless transmission are modeled, as shown in the attached figure. Figure 4 As shown, it includes: (1) Additive White Gaussian Noise (AWGN) model: Gaussian noise is superimposed on the signal under a given SNR; (2) Rayleigh fading model: the amplitude of the coded vector is attenuated using a randomly distributed fading coefficient; (3) SNR control module: using noise power control to achieve different noise intensity tests. This module can simulate real wireless transmission environments, such as multipath fading, noise interference, etc.

[0077] Secondly, corresponding to step 3, the encoder uses the Swin Transformer (sliding window self-attention) as the main encoding network, specifically including: patch segmentation, multi-level Swin Transformer encoding, and encoded vector output. This structure fully utilizes the local window self-attention mechanism, exhibiting strong robustness to local features in the QR code image. The decoder structure is shown in the attached figure. Figure 3 As shown, it consists of the following parts: feature reconstruction layer (Linear transformation + Reshape), multi-layer decoder (Swin Transformer), and Patch Reverse. This structure can effectively recover local areas of the image damaged by noise and achieve highly robust reconstruction.

[0078] Then, corresponding to step 4, the reconstructed QR code image is decoded using the general-purpose vision tool OpenCV or the dedicated barcode decoding tool pyzbar. The decoding results are reordered according to batch markers and combined to form a complete key.

[0079] Finally, corresponding to step 5, calculate the SHA256 hash value for both the original key and the restored key. If they match, the transmission is considered successful; otherwise, the system identifies the failed batch and suggests retransmission. An integrity verification report is attached. Figure 5 As shown. Figure 5 The main contents of the key transmission verification report include the total number of QR codes transmitted, matching rate, batch parsing details, content integrity verification, and other information.

[0080] Furthermore, semantic communication, as an emerging communication paradigm, prioritizes ensuring the receiver can recover or understand the semantic information sent by the sender rather than pursuing bit-level precision transmission. It has demonstrated excellent performance in areas such as image transmission and voice communication. Applying semantic communication to key distribution can effectively address the aforementioned problems faced by traditional methods: robust semantic representation significantly reduces dependence on channel quality, maintaining high information recovery capability even under low signal-to-noise ratio conditions; continuous feature modulation enables stepless channel adaptation, overcoming the limitations of coarse-grained adjustment in traditional modulation and coding schemes. However, direct application of semantic communication to key distribution still faces key challenges: First, keys require complete bit-level consistency; any minor error will lead to key failure, while traditional semantic communication pursues "semantic similarity" rather than "bit precision," making it difficult to meet the lossless requirement of key transmission; second, key distribution scenarios often require offline, cross-platform transmission capabilities, and simple wireless channel transmission cannot meet practical deployment needs, lacking an effective out-of-band verification mechanism to establish initial trust; furthermore, existing semantic communication models lack an effective information carrier that can both carry structured key data and adapt to deep learning frameworks, as well as an end-to-end integrity verification mechanism. These challenges limit the practical application of semantic communication in distribution scenarios.

[0081] To address the aforementioned challenges, this invention innovatively proposes using QR codes as information carriers to construct a key transmission method based on semantic communication. QR codes, as a mature visual encoding technology, naturally possess offline transmission capabilities, enabling cross-platform, network-independent key transmission via screen display, paper printing, and other methods. Simultaneously, their visual characteristics provide a natural out-of-band verification mechanism, allowing users to establish initial trust through manual verification. More importantly, QR codes encode keys into structured visual information, providing an ideal information carrier for semantic communication models, making deep learning-driven end-to-end optimization possible. Based on this innovative combination, this invention constructs a complete semantic encoding-transmission-decoding model: fine-grained channel adaptation is achieved through continuous feature modulation based on signal-to-noise ratio (SNR), maintaining high-reliability transmission even under low SNR conditions (SNR < 10dB); deep learning-driven intelligent image reconstruction technology overcomes the traditional 30% error correction rate limitation of QR codes, successfully recovering data even under conditions of re-degradation, blurring, or noise contamination; combined with a hash verification mechanism, bit-level accurate key recovery is ensured, perfectly unifying the robustness and accuracy requirements of semantic communication.

[0082] This invention deeply integrates QR code carriers with semantic communication technology, solving problems such as network dependence, high implementation complexity, and poor signal-to-noise ratio performance of traditional distribution methods. It also overcomes key challenges in semantic communication applications, including lossless transmission, offline deployment, and information carrier selection. This provides a novel technical path for secure and reliable key transmission in complex environments, possessing significant theoretical value and broad application prospects. Furthermore, considering that QR codes, as a structured visual encoding method, have a capacity limit (typically no more than 3KB of text content), and that the decoding accuracy of semantic communication models under extreme degradation conditions is still affected by the complexity and structural interference of the input image, directly encoding the entire key into a single QR code image not only easily leads to encoding failure but also affects the integrity and reliability of overall key recovery due to single-point failures. Therefore, adopting a batch-segmented key approach with batch identifiers not only significantly improves the encoding and decoding success rate but also provides a more easily learned structured input for the semantic communication model, which is beneficial for the accuracy of model feature extraction and reconstruction. The method of this invention is used to achieve high-reliability encoding, transmission, and reconstruction of public keys under channel noise interference conditions. This method is applicable to various resource-constrained or network-constrained scenarios such as IoT devices, smart terminals, and embedded systems, and aims to improve the security, reliability, and deployment flexibility of key distribution.

[0083] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in the details for the sake of brevity.

[0084] This invention is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A lossless key transmission method based on semantic communication, characterized in that, Includes the following steps: Step 1, Key Semantic Encapsulation and Carrier Construction: The original key to be transmitted is divided into multiple key fragments according to a preset batch size; a batch marker is added to each key fragment to obtain a marked key fragment; the marked key fragment is encoded into a QR code image to construct a structured semantic information carrier. Step 2, Semantic Information Transmission Degradation Simulation: The structured semantic information carrier is input into the channel simulation model, which supports additive white Gaussian noise channel and Rayleigh fading channel. The noise intensity is controlled by the set signal-to-noise ratio, and the degradation process of semantic information in wireless transmission is simulated in the feature domain to generate degraded semantic information. Step 3, Degraded semantic information reconstruction: The degraded semantic information is input into the semantic communication model of the receiving end for reconstruction; the semantic communication model includes a semantic encoder, a channel module and a semantic decoder connected in sequence; the semantic encoder is used to extract semantic features and perform channel adaptive modulation; the channel module is used to introduce transmission noise consistent with Step 2 into the feature domain; the semantic decoder is used to decode and reconstruct the damaged semantic features and output the reconstructed QR code image; Step 4, Semantic parsing and key recovery: Decode the reconstructed QR code image to extract the marked key fragments, and concatenate them according to the order of the batch markings to obtain the recovery key; Step 5, Integrity Verification: Calculate and compare the hash values ​​of the original key and the recovery key respectively. If they match, the transmission is deemed lossless and complete.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Read the original key and divide it into multiple key fragments according to the preset batch size; Step 1.2: Add a batch marker to each of the key segments, the batch marker containing the sequence number of the segment and the total number of segments; Step 1.3: Encode each key fragment with batch markers into a QR code image to obtain a set of structured semantic information carriers.

3. The method according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Construct a channel simulation module that supports additive white Gaussian noise channels and Rayleigh fading channels; Step 2.2: When the additive white Gaussian noise channel is selected, zero-mean white Gaussian noise is generated according to the set signal-to-noise ratio, and the noise is superimposed on the transmission feature corresponding to the structured semantic information carrier. Step 2.3: When a Rayleigh fading channel is selected, a Rayleigh-distributed fading coefficient is generated, and amplitude modulation is performed on the transmission characteristics corresponding to the structured semantic information carrier based on the fading coefficient. Step 2.4: The channel simulation module processes the data to generate the degraded semantic information in the semantic feature space.

4. The method according to claim 1, characterized in that, The semantic communication model in step 3 includes: The semantic encoder is configured to extract and adaptively modulate multi-scale semantic features; the channel module is configured to receive the modulated semantic features and simulate the transmission degradation defined in step 2 in the feature domain; and the semantic decoder is configured to reconstruct the degraded semantic features and restore them to a QR code image.

5. The method according to claim 1, characterized in that, Step 4 specifically includes: Step 4.1: Detect and decode the reconstructed QR code image; Step 4.2: If decoding fails, perform image preprocessing on the reconstructed QR code image and try decoding again. The image preprocessing includes one or more of threshold segmentation, filtering, sharpening, or image scaling. Step 4.3: Extract the batch marker from the successfully decoded content, and concatenate the key fragments according to the order indicated by the batch marker to form the recovery key.

6. The method according to claim 1, characterized in that, Step 5 specifically includes: Step 5.1: Calculate the hash values ​​of the original key and the recovery key using the same hash algorithm; Step 5.2: Compare the hash values ​​of the original key and the recovery key; if they match, it is determined to be lossless transmission. Step 5.3: If the hash values ​​are inconsistent, generate an integrity report containing information on failed batches or data differences.

7. The method according to claim 1, characterized in that, The semantic encoder in step 3 includes: The semantic feature extraction module is used to extract multi-scale semantic features based on the window attention mechanism. The channel adaptive modulation module is used to dynamically adjust the expression intensity of the semantic features based on the current signal-to-noise ratio using a learnable modulation vector. The feature compression module is used to map high-dimensional semantic features to a low-dimensional feature space suitable for channel transmission.

8. The method according to claim 1, characterized in that, The semantic decoder in step 3 includes: The noise-robust feature processing module is used to denoise and enhance the received noisy semantic features; the semantic reconstruction module is used to gradually restore the spatial resolution of semantic information using a reverse hierarchical structure; and the pixel mapping module is used to map the reconstructed semantic features to the reconstructed QR code image.

Images