A neural network-based data transmission method, system and device

Abstract

This invention discloses a data transmission method, system, and apparatus based on neural networks, applied in the field of computer data transmission technology. The method includes: determining an environmental parameter set for a transmission scenario between a sending end and a receiving end in response to a transmission request signal; scoring multiple preset candidate encoding schemes based on the environmental parameter set and a data attribute table, and determining a target encoding scheme from among the candidate schemes; encoding the target data according to the target encoding scheme to obtain a first encoded image; displaying the first encoded image through a display component, enabling the receiving end to scan the displayed first encoded image using a scanning component to obtain the target data, and then sending the target data to a second-level network terminal. This achieves dynamic determination of the encoding scheme to complete the data transmission process, achieving adaptation between the encoding scheme and the target data, and improving data transmission capability and anti-interference capability.

Description

Technical Field

[0001] This invention relates to the field of computer data transmission technology, and in particular to a data transmission method, system and apparatus based on neural networks. Background Technology

[0002] As the digital economy permeates all industries, the demand for data transmission has expanded from basic person-to-person communication to all scenarios, including person-to-thing, thing-to-thing, and industry-to-industry communication. Furthermore, the need for data interaction between networks with different security levels is becoming increasingly urgent, requiring not only basic interconnection but also extremely high standards for security, integrity, and accuracy.

[0003] Data transmission relies on both network base stations and transmission technologies. Most existing transmission technologies employ one-way network isolation technology and encoded image transmission technology in fiber optic communication. One-way network isolation technology physically blocks the reverse transmission channel, while encoded image transmission technology enables interaction by encoding the transmitted data into image form. However, existing methods suffer from poor encoding adaptability, low transmission efficiency, limited anti-interference capabilities, and potential security vulnerabilities. Summary of the Invention

[0004] In view of the above problems, a neural network-based data transmission method, system, and apparatus are proposed to overcome or at least partially solve the above problems, comprising: In a first aspect, embodiments of the present invention provide a data transmission method based on a neural network, applied at a sending end, wherein the sending end is communicatively connected to a receiving end, the sending end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal. The method includes: The system receives a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of the target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. In response to the transmission request signal, determine a set of environmental parameters for the transmission scenario between the sending end and the receiving end; The encoding optimization module is invoked to score multiple preset candidate encoding schemes based on the environmental parameter set and the data attribute table, obtain the score result of each candidate encoding scheme, and determine the target encoding scheme from the multiple candidate encoding schemes based on the score result. The encoding optimization module is a convolutional neural network based on ResNet-18. The target data is obtained from the first security level network terminal, and the target data is encoded according to the target encoding scheme to obtain the first encoded image; The first encoded image is displayed by a display component, so that the receiving end scans the displayed first encoded image by a scanning component to obtain the target data and sends the target data to the second security level network terminal.

[0005] Optionally, based on the environmental parameter set and the data attribute table, multiple preset candidate coding schemes are scored to obtain a score result for each candidate coding scheme, including: Based on the environmental parameter set and the data attribute table, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions. The weights of the multiple dimensions are determined, and the sub-scores under the multiple dimensions are weighted and summed according to the weights to obtain the score result of each candidate coding scheme.

[0006] Optionally, based on the environmental parameter set and the data attribute table, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions, including: The environmental parameter set and the data attribute set are combined to obtain the initial parameter set; The initial parameter set is numerically mapped to obtain the input vector; Based on the input vector, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions.

[0007] Optionally, the target data is encoded according to the target encoding scheme to obtain a first encoded image, including: Based on the target encoding scheme, determine the block division rules; According to the segmentation rules, the target data is segmented into multiple data blocks. The multiple data blocks are encoded to obtain a first encoded image.

[0008] Optionally, based on the target encoding scheme, the segmentation rules are determined, including: Obtain the parameters of the target coding scheme, including coding resolution, error correction level, and coding type; Based on the parameters of the target encoding scheme, the block division rules are determined; the block division rules include single block data capacity, total number of blocks, data allocation strategy, and block identifier.

[0009] Optionally, the plurality of data blocks are encoded to obtain a first encoded image, including: The multiple data blocks are encoded block by block to obtain a black and white binary dot matrix; The black and white binary dot matrix is ​​encoded to obtain the first encoded image.

[0010] Optionally, before displaying the first encoded image via the display component, the method further includes: Perform pixel deviation rate detection on the first encoded image; When the deviation rate of the first encoded image is detected to be within acceptable limits, the first encoded image is displayed via the display component.

[0011] Optionally, the method further includes: The receiver receives an abnormal feedback signal, which is a signal fed back to the sender when the receiver calls the abnormal detection module to perform abnormal detection on the first encoded image and the detection result is abnormal. The abnormal detection module includes a lightweight neural network. Obtain historical transmission parameters for the target data, the historical transmission parameters including transmission rate and characteristics of the transmitted data; The breakpoint prediction module is invoked to perform breakpoint detection based on the abnormal feedback signal and the historical transmission parameters, and to determine the data that needs to be retransmitted. The breakpoint prediction module includes an improved long short-term memory network. The data that needs to be retransmitted is re-encoded to obtain a second encoded image; The second encoded image is displayed by the display component, so that the receiving end scans the displayed second encoded image by the scanning component to obtain the retransmitted data, and sends the retransmitted data to the second security level network terminal.

[0012] Secondly, embodiments of the present invention provide a data transmission system based on a neural network, the data transmission system comprising a first-level secure network terminal, a transmitting end, a receiving end, and a second-level secure network terminal; the transmitting end is connected to the first-level secure network terminal, and the receiving end is connected to the second-level secure network terminal, the system comprising: The first security-level network terminal is used to send a transmission request signal to the sending end. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of the target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. The transmitting end is used to determine the target encoding scheme, obtain the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain a first encoded image, and display the first encoded image through a display component; wherein, the target encoding scheme is to determine an environmental parameter set for the transmission scenario between the transmitting end and the receiving end in response to the transmission request signal, call the encoding optimization module, score multiple preset candidate encoding schemes according to the environmental parameter set and the data attribute table, obtain a score result for each candidate encoding scheme, and determine the target encoding scheme from the multiple candidate encoding schemes according to the score results; the encoding optimization module is a convolutional neural network based on ResNet-18. The receiving end is used to scan the first coded image through a scanning component to obtain the target data, and send the target data to the second security level network terminal; The second security level network terminal is used to receive the target data from the receiving end.

[0013] Thirdly, embodiments of the present invention provide a data transmission device based on a neural network, characterized in that it is applied to a transmitting end, wherein the transmitting end is communicatively connected to a receiving end, the transmitting end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal, the device comprising: The transmission request signal acquisition module is used to receive a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. An environment parameter set determination module is used to determine an environment parameter set for the transmission scenario between the sending end and the receiving end in response to the transmission request signal. The target coding scheme determination module is used to call the coding optimization module, score multiple preset candidate coding schemes according to the environmental parameter set and the data attribute table, obtain the score result of each candidate coding scheme, and determine the target coding scheme from the multiple candidate coding schemes according to the score result. The coding optimization module is a convolutional neural network based on ResNet-18. The first encoded image acquisition module is used to acquire the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain the first encoded image; The first encoded image sending module is used to display the first encoded image through a display component, so that the receiving end can scan the displayed first encoded image through a scanning component to obtain the target data and send the target data to the second security level network terminal.

[0014] Fourthly, embodiments of the present invention provide an electronic device, which includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the method described above.

[0015] Fifthly, embodiments of the present invention provide a readable storage medium on which a program or instructions are stored, and when the program or instructions are executed by a processor, the program or instructions implement the method described above.

[0016] Sixthly, embodiments of the present invention provide a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0017] The embodiments of the present invention have the following advantages: In this embodiment of the invention, a transmission request signal sent by a first-level security network terminal is received. The transmission request signal includes a data attribute table, which is generated by the first-level security network terminal through feature recognition of target data. The target data is the data to be transmitted from the first-level security network terminal to a second-level security network terminal. In response to the transmission request signal, an environmental parameter set for the transmission scenario between the sending and receiving ends is determined. Based on the environmental parameter set and the data attribute table, multiple preset candidate encoding schemes are scored to obtain a score result for each candidate encoding scheme. Based on the score result, a target encoding scheme is determined from the multiple candidate encoding schemes. The target data is obtained from the first-level security network terminal and encoded according to the target encoding scheme to obtain a first encoded image. The first encoded image is displayed through a display component, so that the receiving end scans the displayed first encoded image through a scanning component to obtain the target data and sends the target data to the second-level security network terminal. This achieves dynamic determination of the encoding scheme through a neural network, completes the encoding of data blocks, realizes data transmission, achieves adaptation between the encoding scheme and the target data, and improves data transmission capability and anti-interference capability. Attached Figure Description

[0018] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the steps of a data transmission method based on a neural network, provided in some embodiments of the present invention. Figure 2 This is a structural block diagram of a data transmission system based on a neural network provided in some embodiments of the present invention; Figure 3 This is a structural block diagram of a data transmission device based on a neural network provided in some embodiments of the present invention; Figure 4 This is a parameter flow diagram of three neural network modules provided in some embodiments of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] As the digital economy permeates all industries, the demand for data transmission has expanded from basic person-to-person communication to all scenarios, including person-to-thing, thing-to-thing, and industry-to-industry communication. Furthermore, the need for data interaction between networks with different security levels is becoming increasingly urgent, requiring not only basic interconnection but also extremely high standards for security, integrity, and accuracy.

[0022] Data transmission relies on both network base stations and transmission technologies. Most existing transmission technologies employ one-way network isolation technology and encoded image transmission technology in fiber optic communication. One-way network isolation technology physically blocks the reverse transmission channel, while encoded image transmission technology enables interaction by encoding the transmitted data into image form. However, existing methods suffer from poor encoding adaptability, low transmission efficiency, limited anti-interference capabilities, and potential security vulnerabilities.

[0023] This invention achieves data transmission by dynamically determining the encoding scheme, thereby adapting the encoding scheme to the target data and improving data transmission capability and anti-interference capability.

[0024] Reference Figure 1 The diagram illustrates a flowchart of a data transmission method based on a neural network according to some embodiments of the present invention. The method is applied to a sending end, which is communicatively connected to a receiving end. The sending end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal.

[0025] In some examples, a dense network terminal is an independent device participating in a dense communication network, such as a mobile phone or vehicle-mounted terminal, which is an independent device with complete transmit and receive communication functions. The first-level network terminal is connected to the sending end, and the receiving end is connected to the second-level network terminal. They form a data transmission path, and there is no electrical or physical connection between the sending and receiving ends, ensuring that the sending and receiving ends are completely independent.

[0026] Specifically, it may include the following steps: Step 101: Receive a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of the target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal.

[0027] In some examples, the data attribute table is based on the core attribute information of the target data.

[0028] In practical applications, the first dense network terminal acts as the starting point for data transmission. When data needs to be sent from the first dense network terminal to the second dense network terminal, the first dense network terminal performs feature recognition on the target data, packages it into a transmission request signal, and transmits the transmission request signal to the sending end. For example, in some examples, the target data can be structured text, configuration files, documents, etc. In some examples, a common encoding format such as UTF-8 can be used for standardization to ensure cross-device compatibility. After performing hash value or other integrity checks on the data, it is encapsulated into a "data packet to be transmitted" to avoid damage to the original data. The terminal attribute analysis module automatically identifies data characteristics, including the total data volume accurate to bytes, encoding format, data structure type, and, if it is structured text, the number of characters per line, generating a data attribute table.

[0029] In some examples, the data attribute table is bound to the data packet to be transmitted. The terminal sends a transmission request signal to the sender through a preset link. The signal contains the core information of the data attribute table but does not contain the complete data packet. The terminal waits for the sender to confirm the response.

[0030] Step 102: In response to the transmission request signal, determine the set of environmental parameters for the transmission scenario between the sending end and the receiving end.

[0031] In some examples, the set of environmental parameters for the transmission scenario can be related to factors such as light intensity and transmission distance between the sender and receiver.

[0032] In practical applications, when the core processing unit at the sending end receives a transmission request signal, it can first verify the identity and legitimacy of the first network-dense terminal. After successful verification, it can collect environmental parameters for both the sending and receiving ends, forming an environmental parameter set. For example, in some examples, the core processing unit at the sending end receives the transmission request signal, completes the identity and legitimacy verification through a preset security mechanism, returns a ready signal, and establishes a one-way data transmission link between the terminal and the sending end. In other examples, after establishing the data transmission link, the sending end activates the environmental perception component to collect light intensity, performing multiple samples and averaging to ensure stability; it also collects the straight-line distance between the sending end's display component and the receiving end's scanning component as the transmission distance, generating an environmental parameter table.

[0033] Step 103: Invoke the encoding optimization module, score multiple preset candidate encoding schemes according to the environmental parameter set and the data attribute table, obtain the score result of each candidate encoding scheme, and determine the target encoding scheme from the multiple candidate encoding schemes according to the score result. The encoding optimization module is a convolutional neural network based on ResNet-18.

[0034] In some examples, an encoding scheme is a set of methods for data transformation or rule-based processing of target data during data transmission.

[0035] In practical applications, there are many encoding schemes for target data. By scoring these encoding schemes based on previously collected environmental parameters and the characteristics of the target data, the scheme with the highest comprehensive score is selected as the most suitable encoding scheme for the current environment and target data, thus determining the target encoding scheme.

[0036] In some examples, the encoding optimization module is a convolutional neural network based on an improvement of ResNet-18. ResNet-18 (an 18-layer residual network) was chosen as the basic architecture for several key reasons: residual connections solve the vanishing gradient problem in shallow networks, adapting to the learning of data transmission parameters (low dimensionality, few features); the parameter count is moderate (approximately 11M for the original ResNet-18), making it easy to lightweight and adapt, meeting the low computational power requirements of edge computing at the sending end. Furthermore, improvements were made to the original architecture, such as halving the number of convolutional kernels per layer, reducing the first layer from 64 to 32. This reduced the overall parameter count to ≤8M and inference time by 30%; a channel attention module was added to strengthen the feature weights of environmental parameters (light / distance); the 3D output encoding type, resolution, and error correction level are output as probability distributions using the Softmax activation function. ResNet-18, a lightweight 18-layer deep residual network, uses residual connections to solve the vanishing gradient problem in deep networks. Through skip connections, the input is directly superimposed onto the deep output, learning the residual mapping, and the gradient can be directly backpropagated, easily training deep networks. The improved convolutional neural network is adapted to scenarios that balance computing power and performance, such as cross-security transmission.

[0037] Sub-step 11: Based on the environmental parameter set and the data attribute table, perform multi-dimensional scoring on multiple candidate coding schemes to obtain sub-scores of the candidate coding schemes in multiple dimensions.

[0038] In some examples, multiple dimensions can be transmission efficiency, anti-interference capability, and scene adaptability.

[0039] In practical applications, scoring rules under different dimensions are preset. Under the scoring rules, the matching degree is calculated based on the parameters in different coding schemes, environmental parameters, and data attribute tables to obtain multi-dimensional scores for different coding schemes.

[0040] Sub-step 111: Combine the environmental parameter set and the data attribute set to obtain the initial parameter set.

[0041] In practical applications, the environmental parameter set and the data attribute set are combined to form the initial parameter set for transmission, which is then used as input parameters. For example, the core processing unit sends a loading instruction to the encoding optimization neural network module, the module reads the pre-trained model and completes initialization, returns a "model ready signal," and waits for parameter input.

[0042] Sub-step 112 involves numerically mapping the initial parameter set to obtain the input vector.

[0043] In some examples, the input vector refers to a structured numerical vector that the machine can compute the score for.

[0044] In practical applications, after receiving the initial parameter set, the sending end can standardize the numerical values ​​of each parameter and arrange the transformed values ​​into a multi-dimensional vector in a fixed-dimensional order. For example, the core processing unit inputs the "initial parameter set" into the encoding optimization neural network module. The module performs numerical mapping on the parameters: the total data volume is mapped to values ​​according to intervals, such as 1KB or less = 1, 1KB-10KB = 2, 10KB-100KB = 3, etc.; the light intensity is mapped to values ​​according to ranges, such as low light = 1, medium light = 2, high light = 3; the transmission distance is mapped to values ​​according to intervals, such as short distance = 1, medium distance = 2, long distance = 3; and the encoding format is mapped to corresponding identifier values, forming a multi-dimensional input vector.

[0045] Sub-step 113: Perform multi-dimensional scoring on multiple candidate coding schemes to obtain sub-scores of the candidate coding schemes in multiple dimensions.

[0046] In practical applications, the encoding schemes in the candidate list are scored in multiple dimensions according to the scoring rules, and sub-scores of the candidate encoding schemes in different dimensions are obtained. For example, the scoring dimensions include transmission efficiency (data encoding time, redundancy rate), anti-interference capability (error correction level adaptability, resolution adaptability, scene adaptability, data type and encoding type matching degree, and environmental parameters and encoding parameters matching degree).

[0047] Sub-step 12: Determine the weights of the multiple dimensions, and perform a weighted summation of the sub-scores under the multiple dimensions based on the weights to obtain the score result of each candidate coding scheme.

[0048] In practical applications, weights for different dimensions are obtained. Based on the sub-scores and corresponding weights of different dimensions, the sub-scores of the same coding scheme are weighted to obtain the comprehensive score of each candidate coding scheme. The weights for different dimensions can be dynamically adjusted according to the scenario. For example, the baseline weighting coefficients for the coding scheme scores are fixed, but dynamic adjustment according to the scenario is supported. The specific rules are as follows: Baseline weighting coefficients: transmission efficiency (α=0.4), anti-interference capability (β=0.35), scenario adaptability (γ=0.25), with a weight sum of 1, which is the default configuration. In some examples, if the transmission scenario is a strong interference environment such as "low light / long distance (≥3m)", the values ​​are adjusted to α=0.3, β=0.5, and γ=0.2 (prioritizing anti-interference); if the transmission scenario is "massive text data (≥1GB)", the values ​​are adjusted to α=0.5, β=0.3, and γ=0.2 (prioritizing transmission efficiency). Fine-tuning trigger conditions: the light intensity collected by the environmental perception component at the sending end is ≤50 lux or the transmission distance is ≥3m, or the total amount of data in the data attribute table is ≥1GB. The weight adjustment is automatically triggered without manual intervention.

[0049] Step 104: Obtain the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain the first encoded image.

[0050] In some examples, the encoded image can be a QR code image or a barcode image; the specific type is not limited here.

[0051] In practical applications, after the sending end performs authentication, a one-way data transmission link is established between the terminal and the sending end. The target data is obtained from the first dense network terminal, and the target data is encoded according to the obtained target encoding scheme to obtain the first encoded image after encoding.

[0052] Sub-step 21: Determine the block segmentation rules according to the target encoding scheme.

[0053] In some examples, chunking rules are defined data partitioning strategies for large-volume data transmissions in dense networks.

[0054] In practical applications, the target coding scheme determines specific data block rules to ensure that a single block of coded image can be fully carried without excessive redundancy.

[0055] Sub-step 211: Obtain the parameters of the target coding scheme, including coding resolution, error correction level and coding type.

[0056] Sub-step 212: Determine the block division rules based on the parameters of the target encoding scheme; the block division rules include single block data capacity, total number of blocks, data allocation strategy and block identifier.

[0057] In some examples, the data allocation strategy is how data is allocated to different blocks, based on whether the capacity is full or the remaining amount of data is allocated.

[0058] In practical applications, the encoding resolution, error correction level, and encoding type of the target encoding scheme are obtained. These parameters are then substituted into a standardized calculation process to determine how the data should be divided into blocks. For example, in some examples, the core parameters of the optimal encoding scheme are extracted. From the optimal encoding scheme output by the encoding optimization neural network module, three key parameters are extracted: encoding resolution (e.g., 1024×1024, 2048×2048); error correction level (L-level / 7%, M-level / 15%, H-level / 30%, ultra-high / 40%); and encoding type (QR code / custom dot matrix code / barcode, which only affects the effective pixel ratio and is fixed at 90%).

[0059] In some examples, the process of determining the target encoding scheme is implemented through an encoding optimization neural network module, which is a convolutional neural network based on an improvement of ResNet-18. ResNet-18 (an 18-layer residual network) was chosen as the basic architecture for the following reasons: residual connections solve the gradient vanishing problem in shallow networks, adapting to the learning of data transmission parameters (low dimensionality, few features); the parameter count is moderate (the original ResNet-18 is approximately 11M), making it easy to lightweightly modify and meeting the low computational power requirements of edge computing at the sending end. Furthermore, improvements were made to the original architecture, such as halving the number of convolutional kernels per layer, reducing the first layer from 64 to 32. The overall parameter count was reduced to ≤8M, and inference time was reduced by 30%. A channel attention module was added to strengthen the feature weights of environmental parameters (light / distance); the 3D output encoding type, resolution, and error correction level are output as probability distributions using a Softmax activation function.

[0060] In some examples, the capacity of a single data block is calculated as follows: Substitute into a unified formula to calculate the maximum effective data capacity that a single block can carry: Single data block capacity (bytes) = (encoding resolution in horizontal pixels × vertical pixels × 90% effective percentage × 1 bit / pixel ÷ 8 bits / byte) × (1 - error correction redundancy rate) × 0.95 security factor. The error correction redundancy rate corresponds one-to-one with the error correction level; 0.95 security factor: 5% space is reserved to avoid data loss due to encoding distortion, and it is a fixed value.

[0061] In some examples, the total number of blocks is calculated as follows: Total number of blocks = Total number of bytes of text data to be transmitted ÷ Capacity of a single block, and the result is rounded up; Data allocation: The first (total number of blocks - 1) blocks are allocated according to the full capacity of a single block, and the last block is allocated according to the remaining data (filled in if insufficient).

[0062] In some examples, the embedded block identifier and format constraints are based on the error correction level of the optimal encoding scheme. A corresponding identifier field (e.g., 0x04 for ultra-high error correction level) is embedded in the header of each data block. Simultaneously, rules such as consecutive block numbers and padding characters appearing only in the last block are enforced. In other examples, rule checks verify whether the segmented data blocks are compatible with the encoding image carrying capacity of the optimal encoding scheme (e.g., single block data size ≤ number of effective bytes carried by the encoded image). If not, the security factor is readjusted (reduced to 0.9) and recalculated until it is compatible.

[0063] Sub-step 42: According to the segmentation rules, the target data is segmented into multiple data blocks. In practical applications, after receiving the target data, the core processing unit first divides the target data into blocks according to the determined block division rules. For example, the core processing unit divides the target data into blocks according to the determined block division rules: calculate the total number of blocks (total data volume ÷ single block capacity, rounded up), the first N-1 blocks are complete capacity data, and the Nth block is the remaining data (if it is less than the single block capacity, it is filled to the standard capacity with a preset padding character); each data block header embeds a block sequence number (incrementing sequence) and a data check bit (calculated based on the data in the block), forming an "identified data block".

[0064] Sub-step 23 involves encoding the multiple data blocks to obtain a first encoded image.

[0065] Sub-step 231 involves encoding the multiple data blocks one by one to obtain a black and white binary dot matrix.

[0066] In practical applications, the encoding optimization neural network module encodes the labeled data blocks one by one, first converting the data into a black and white binary dot matrix, for example, 1 bit corresponds to 1 pixel, and 0 / 1 correspond to black and white pixels respectively.

[0067] Sub-step 232 involves encoding the black-and-white binary dot matrix to obtain the first encoded image.

[0068] In practical applications, further encoding processing of the black and white binary dot matrix can ultimately yield the first encoded image, for example, by combining error correction levels with embedded redundant dot matrices to generate an encoded image.

[0069] Step 105: Display the first encoded image through the display component, so that the receiving end can scan the displayed first encoded image through the scanning component to obtain the target data, and send the target data to the second security level network terminal.

[0070] In practical applications, the core processing unit controls the display component to display the coded image at a fixed refresh rate, ensuring stable and continuous image display and creating favorable scanning conditions for the scanning component at the receiving end. This display component can be on the transmitter's own screen or displayed through other display devices. The main task of the scanning component at the receiving end is to scan the coded image displayed at the transmitter, obtain information from the coded image through scanning, parse this information, restore the coded image to data form, and perform checks. If the check is successful, it is sent to the second dense network terminal.

[0071] Before displaying the first encoded image through the display component, the method further includes: performing pixel deviation rate detection on the first encoded image; and when the deviation rate of the first encoded image is detected to be qualified, performing the step of displaying the first encoded image through the display component.

[0072] In some examples, pixel deviation rate refers to whether the encoded image is accurate.

[0073] In practical applications, after generating the encoded image, it is essential to perform pixel deviation rate detection to ensure the encoding process is correct and free from significant deviations. A deviation rate threshold can be set, and the calculated deviation rate is compared to this threshold. If the deviation rate is less than the threshold, the encoded image is transmitted to the display component for display. The threshold is determined based on different scenarios and data. For example, in some cases, the deviation rate is the core criterion for pass / fail judgment. It is used for format anomaly detection, such as field length deviation rate (passing if less than the threshold, anomaly if greater than the threshold) and latent distortion detection.

[0074] In some examples, the threshold was determined based on experimental data and industry standard fields, including length deviation rate threshold and pixel deviation rate threshold. The process for determining the length deviation rate threshold involved collecting over 100,000 sets of text data blocks covering data volumes from 1KB to 1GB and different encoding schemes; testing the data parsing error rate under different deviation rates: deviation rate ≤ 5%: parsing error rate < 0.1%, which can be corrected using check bits; deviation rate > 5%: parsing error rate rises sharply to over 5%, which cannot be corrected; combined with the specification of "data field deviation tolerance ≤ 5%" in GB / T20279-2020 "Information Security Technology - Security Technical Requirements for Isolation Components of Network and Terminal Equipment", the final threshold was determined to be 5%.

[0075] The process of determining the pixel deviation rate threshold: A simulated test environment was set up (light intensity 50-1000 lux, transmission distance 0.5-5m), and 150,000+ sets of coded image samples were collected; the accuracy of text data restoration under different pixel deviation rates was tested: deviation rate ≤ 8%: data restoration accuracy ≥ 98%, meeting the cross-security level transmission accuracy requirements; deviation rate > 8%: data restoration accuracy < 90%, resulting in garbled text and missing characters; balancing "detection strictness" and "false detection rate": when the threshold is set to 8%, the false detection rate ≤ 0.8% and the missed detection rate ≤ 0.4%, which is the optimal balance point; finally, the threshold was determined to be 8%, and it supports fine-tuning according to the scene (it can be reduced to 6% in low light environment and increased to 10% in strong light environment).

[0076] After displaying the first encoded image via the display component, the method further includes: receiving an anomaly feedback signal from the receiving end, wherein the anomaly feedback signal is a signal fed back to the sending end by the receiving end when the anomaly detection module is invoked to perform anomaly detection on the first encoded image, and the anomaly detection module includes a lightweight neural network; acquiring historical transmission parameters for the target data, wherein the historical transmission parameters include transmission rate and characteristics of the transmitted data; invoking a breakpoint prediction module to perform breakpoint detection based on the anomaly feedback signal and the historical transmission parameters, and determining the data that needs to be retransmitted, wherein the breakpoint prediction module includes an improved long short-term memory network; re-encoding the data that needs to be retransmitted to obtain a second encoded image; and displaying the second encoded image via the display component, so that the receiving end scans the displayed second encoded image via the scanning component to obtain the retransmitted data, and sends the retransmitted data to the second security level network terminal.

[0077] In practical applications, the receiving end first configures scanning parameters and performs image acquisition and preprocessing. After the receiving end's scanning component acquires the coded image, the raw image is transmitted to the anomaly detection module. The anomaly detection module preprocesses the raw image and outputs a clear coded image. Then, multi-dimensional anomaly detection is performed. If an anomaly is detected, an anomaly control command is generated and fed back to the sending end, triggering retransmission. If the detection is normal, the raw data is extracted to form complete data, which is then sent to the second dense network terminal. For example, the core processing unit at the receiving end configures the scanning module parameters based on the coded image resolution and ambient light parameters fed back from the sending end: scanning resolution (higher than the coded image resolution to ensure complete acquisition), scanning frame rate (e.g., 40fps, higher than the display component refresh rate to avoid missed scans), and exposure parameters (adapted to the current light intensity to ensure moderate image brightness). After the scanning module is started, it continuously acquires the coded images of the display components and transmits the original images to the anomaly detection module. The anomaly detection module preprocesses the original images: it removes noise (such as bright spots caused by ambient light) through filtering algorithms (such as Gaussian filtering) and improves the sharpness of the dot matrix edges through contrast enhancement algorithms (such as histogram equalization), and outputs a clear coded image.

[0078] In some examples, multidimensional detection includes neural network-based format anomaly detection, integrity anomaly detection, and occult distortion anomaly detection.

[0079] In practical applications, the feature extraction submodule of the anomaly detection module (based on a lightweight neural network) extracts spatial features from the "clearly coded image" and maps them to "parsed data blocks" through a fully connected layer. Key fields within the blocks (block number, data checksum, and original data) are extracted. The module checks the field length of the "parsed data block" (compares it with the standard field length; if the deviation exceeds a preset threshold, it is considered a format anomaly) and the consistency of the checksum (recalculates the checksum of the original data; if it does not match the extracted checksum, it is considered a format anomaly). If the format check passes, it proceeds to the next stage; otherwise, it outputs a "format anomaly" result.

[0080] Integrity anomaly detection and block sequence number continuity verification: The receiving end core processing unit records the block sequence numbers that have passed the detection. The current block sequence number should be equal to the previous passed block sequence number + 1. If there is a skipped number (difference ≠ 1), it is determined as "block sequence number missing anomaly". Data missing verification: The length of the original data in the "parsed data block" is detected (compared with the standard single block data length). If the number of missing bytes exceeds a preset threshold (e.g., 10 bytes), it is determined as "data missing anomaly". If the integrity verification passes, it proceeds to the next stage; otherwise, it outputs the result "integrity anomaly".

[0081] Latent distortion anomaly detection and pixel deviation rate calculation: The anomaly detection module iterates through the pixels of the "clearly encoded image," counts the number of abnormal pixels (grayscale pixels that are not pure black / pure white), and calculates the pixel deviation rate (number of abnormal pixels ÷ total number of pixels × 100%). Distortion judgment: If the pixel deviation rate exceeds a preset threshold (e.g., 8%), it is judged as "latent distortion anomaly" (an excessively high deviation rate may lead to data parsing errors); otherwise, it is judged as "distortion acceptable." Final detection result: If all detection steps are qualified, it is judged as "data normal"; if any step is unqualified, it is judged as "data abnormal," and the anomaly type and anomaly block number are output.

[0082] In some examples, the anomaly detection module is implemented using a hybrid network based on an improved MobileNetV3 and fully connected layers. MobileNetV3 is a lightweight convolutional neural network that extracts features from the transmitted data and feeds them to the fully connected layers. The fully connected layers compute through multiple layers of neurons and output the judgment result. The CNN (Convolutional Neural Network) feature extraction layer preprocesses the input 2048×2048×3 RGB image and converts it to grayscale, reducing the dimension to 2048×2048×1. In the example, this architecture is improved, for example, by removing the last two convolutional layers of the original MobileNetV3 and adding a 1×1 convolutional layer to reduce the dimensionality to a 256-dimensional feature map; the output is a 256×256×64 spatial feature map. The fully connected layer has a structure of 3 fully connected layers (256-128-64-5), with the last layer outputting 5 dimensions; in the example, it is improved by adding a Dropout layer (probability 0.2) to prevent overfitting, and the output is 5 dimensions (1 dimension for normal / abnormal + 1 dimension for each of the 4 anomaly types). The model achieves the following results: ≤5M parameters, ≤12ms inference time per image, and supports batch inference (average time per image is ≤5ms when batch size is 8).

[0083] In some examples, the abnormal feedback signal can be a light signal or a pressure signal.

[0084] In practical applications, after detecting an anomaly, the receiving end converts the feedback signal into an optical signal or a pressure signal and transmits it to the transmitting end. In some examples, a "high-low level combination" encoding is used, where a low level represents a "basic identifier" (e.g., 0.5V voltage, lasting 100ms), and a high level represents "data information" (e.g., 3.3V voltage, with the duration corresponding to the anomaly block number range: 200ms for blocks 1-100, 300ms for blocks 101-200, etc.). An "end identifier" (low level lasting 50ms) is added at the end of the signal to ensure signal integrity. The receiving end's feedback signal output module (optical signal transmitter) sends an optical signal to the transmitting end's feedback signal receiving module (optical signal receiver). The transmission path is straight (unobstructed), the transmission distance is the same as the data transmission distance, no electrical connection is required, and signal interaction is achieved through an optical medium. The pressure signal feedback uses a "pressure intensity level + duration" encoding. The pressure intensity level corresponds to the "abnormality type" (e.g., level 1 = format abnormality, level 2 = integrity abnormality, level 3 = latent distortion abnormality), and the duration corresponds to the "abnormal block number range" (e.g., 100ms corresponds to blocks 1-50, 200ms corresponds to blocks 51-100, etc.). A "start indicator" (low-intensity pressure for 50ms) is added at the start of the signal to avoid misidentification. The receiving end feedback signal output module (mechanical pressure application component) and the sending end feedback signal receiving module (pressure sensing component) transmit signals through physical contact. The pressure application component applies pressure of corresponding intensity and duration according to the encoding rules, and the sensing component converts the pressure signal into an electrical signal (without external electrical drive) to achieve signal transmission.

[0085] In practical applications, after a detection result is obtained, the receiving end performs corresponding processing. For example, if it is determined to be "data abnormal," the core processing unit of the receiving end generates an "abnormal control command," controls the feedback signal output module to send a feedback signal, and triggers the retransmission process of the sending end. If it is determined to be "data normal," the core processing unit extracts the original data from the "parsed data block," concatenates it in ascending order of block number to form "complete data," calculates the hash value of the "complete data," and compares it with the original hash value sent by the first-level security terminal. If they match, the data is transmitted to the second-level security network terminal through a preset link; otherwise, it is re-received and detected. The second-level security network terminal receives the "complete data," stores it, or processes it further, and returns a "receive confirmation signal" to the receiving end. Upon receiving the signal, the receiving end ends the current round of data transmission.

[0086] In some examples, the breakpoint is the location where an exception occurs.

[0087] In practical applications, after receiving an anomaly feedback signal, the transmitting end can accurately locate the position of the anomaly block by combining historical transmission parameters. Based on the location of the anomaly block, it determines the data block that needs to be retransmitted and encodes the remaining data starting from the position of this anomaly block. The encoded image is then re-displayed on the screen, allowing the receiving end to rescan and retransmit, continuing data transmission. For example, feedback signal reception and parsing: The transmitting end feedback signal receiving module receives the feedback signal and transmits it to the signal recognition neural network module. The module performs noise reduction processing on the signal to filter environmental interference, such as stray light in optical signals and vibration in pressure signals, and parses the anomaly type and anomaly block sequence range, outputting it to the transmitting end core processing unit. Breakpoint location and data filtering: The core processing unit calls the breakpoint prediction neural network module, combining historical transmission parameters such as transmission rate and timing data, and characteristics of transmitted blocks, to accurately locate the breakpoint position down to the starting byte of the anomaly block. Based on the breakpoint position, it filters the abnormal data blocks that need to be retransmitted, only filtering the anomaly and subsequent data blocks, reducing redundancy. Retransmission encoding and detection: The sending end regenerates the encoded image for the data blocks that need to be retransmitted according to the second encoding process and controls the display component to display it; the receiving end scans and detects anomalies in the retransmitted image according to the third detection process until all data blocks are detected normally, and complete data transmission is completed.

[0088] In some examples, breakpoint detection is achieved through a breakpoint prediction model, which is an improved LSTM network. The basic structure consists of two unidirectional LSTM layers and one fully connected layer. LSTM is a Long Short-Term Memory network, an improved version of a recurrent neural network specifically designed for processing temporal data. The two LSTM layers capture the temporal features of the transmitted data. The unidirectional LSTM processes the temporal information sequentially according to the transmission order, identifying the data block distribution at breakpoints. The fully connected layer, after receiving the temporal features extracted by the two LSTM layers, calculates and outputs the specific breakpoint information through neurons. The improvements lie in the initialization of the forget gate weights: setting the initial weight to 0.8 strengthens the memory of historical data (the first 100 transmission parameters); and adding normalization (LayerNorm) to the input layer to adapt to features with large numerical differences, such as transmission rate (MB / s) and timestamp (ms). For example, the input is a 10-dimensional temporal feature sequence (length = the first 100 blocks of transmitted data), including: block number, transmission time, light intensity, transmission distance, whether retransmission is required, encoding scheme, pixel deviation rate, field length deviation rate, feedback signal strength, and transmission rate; the output is a dimensional vector (breakpoint block number, breakpoint start byte), which outputs continuous values ​​through linear activation, and then rounds to obtain the precise breakpoint location. Results: This model can accurately locate breakpoints using temporal features, allowing the sender to retransmit from the breakpoint, improving transmission efficiency, achieving a breakpoint location error ≤ 10 bytes, inference time ≤ 5ms, and location accuracy ≥ 99%.

[0089] In some examples, data transmission efficiency can achieve single-block data encoding time ≤8ms, scanning and detection time ≤12ms, and retransmission data volume reduced by more than 60% compared to traditional solutions. Detection accuracy can achieve an accuracy rate of ≥98% for format anomalies, integrity anomalies, and latent distortion anomalies, with a false detection rate ≤1% and a false negative rate ≤0.5%. Anti-interference performance can achieve a feedback signal recognition accuracy rate of ≥97% (including environmental interference scenarios), and a recognition success rate of ≥95% for encoded images under different lighting conditions.

[0090] In some examples, edge computing units exist at both the sending and receiving ends. Encoding optimization, anomaly detection, and breakpoint prediction modules are all deployed within these edge computing units, which function as core processing units at both the sending and receiving ends. Parameters are interconnected between these three components, such as... Figure 4 As shown, the optimal encoding scheme output by the encoding optimization module serves as the input parameters (such as resolution / error correction level) for the anomaly detection network, guiding the adjustment of the anomaly threshold. The anomaly type / block number of the anomaly detection network serves as the input feature of the LSTM (Long Short-Term Memory) network, improving the accuracy of breakpoint localization. The parameters of the encoding optimization module, anomaly detection module, and breakpoint prediction module are interconnected, improving the overall transmission performance. All networks employ quantization compression + pruning (removing redundant convolutional kernels), with a total model size ≤20MB, meeting the storage / computing power limitations of transmission devices while balancing accuracy and speed, and satisfying the real-time requirements of cross-security level transmission.

[0091] In this embodiment of the invention, a transmission request signal sent by a first-level security network terminal is received. The transmission request signal includes a data attribute table, which is generated by the first-level security network terminal through feature recognition of target data. The target data is the data to be transmitted from the first-level security network terminal to a second-level security network terminal. In response to the transmission request signal, an environmental parameter set for the transmission scenario between the sending and receiving ends is determined. Based on the environmental parameter set and the data attribute table, multiple preset candidate encoding schemes are scored to obtain a score result for each candidate encoding scheme. Based on the score result, a target encoding scheme is determined from the multiple candidate encoding schemes. The target data is obtained from the first-level security network terminal and encoded according to the target encoding scheme to obtain a first encoded image. The first encoded image is displayed through a display component, so that the receiving end scans the displayed first encoded image through a scanning component to obtain the target data and sends the target data to the second-level security network terminal. This achieves the data transmission process by dynamically determining the encoding scheme, thereby achieving the adaptation of the encoding scheme and the target data, and improving the data transmission capability and anti-interference capability.

[0092] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0093] Reference Figure 2 The diagram illustrates a structural block diagram of a neural network-based data transmission system according to some embodiments of the present invention. The system includes: The data transmission system includes a first-level security network terminal 201, a transmitting end 202, a receiving end 203, and a second-level security network terminal 204; the transmitting end 202 is connected to the first-level security network terminal 201, and the receiving end 203 is connected to the second-level security network terminal 204. The first security-level network terminal 201 is used to send a transmission request signal to the sending end. The transmission request signal includes a data attribute table. The data attribute table is generated by the first security-level network terminal through feature recognition of target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. The transmitting end 202 is used to determine a target encoding scheme, obtain the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain a first encoded image, and display the first encoded image through a display component; wherein, the target encoding scheme is, in response to the transmission request signal, determining an environmental parameter set for the transmission scenario between the transmitting end and the receiving end, calling an encoding optimization module, scoring multiple preset candidate encoding schemes according to the environmental parameter set and the data attribute table, obtaining a score result for each candidate encoding scheme, and determining the target encoding scheme from the multiple candidate encoding schemes according to the score results; the encoding optimization module is a convolutional neural network based on a ResNet-18 improvement. The receiving end 203 is used to scan the first encoded image through the scanning component to obtain the target data, and send the target data to the second security level network terminal; The second security level network terminal 204 is used to receive the target data from the receiving end.

[0094] In some embodiments of the present invention, multiple preset candidate coding schemes are scored based on the environmental parameter set and the data attribute table to obtain a score result for each candidate coding scheme, including: Based on the environmental parameter set and the data attribute table, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions. The weights of the multiple dimensions are determined, and the sub-scores under the multiple dimensions are weighted and summed according to the weights to obtain the score result of each candidate coding scheme.

[0095] In some embodiments of the present invention, multiple candidate coding schemes are scored in multiple dimensions based on the environmental parameter set and the data attribute table to obtain sub-scores of the candidate coding schemes in multiple dimensions, including: The environmental parameter set and the data attribute set are combined to obtain the initial parameter set; The initial parameter set is numerically mapped to obtain the input vector; Based on the input vector, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions.

[0096] In some embodiments of the present invention, the target data is encoded according to the target encoding scheme to obtain a first encoded image, including: Based on the target encoding scheme, determine the block division rules; According to the segmentation rules, the target data is segmented into multiple data blocks. The multiple data blocks are encoded to obtain a first encoded image.

[0097] In some embodiments of the present invention, determining the block segmentation rules according to the target encoding scheme includes: Obtain the parameters of the target coding scheme, including coding resolution, error correction level, and coding type; Based on the parameters of the target encoding scheme, the block division rules are determined; the block division rules include single block data capacity, total number of blocks, data allocation strategy, and block identifier.

[0098] In some embodiments of the present invention, encoding the plurality of data blocks to obtain a first encoded image includes: The multiple data blocks are encoded block by block to obtain a black and white binary dot matrix; The black and white binary dot matrix is ​​encoded to obtain the first encoded image.

[0099] In some embodiments of the present invention, prior to displaying the first encoded image via the display component, the method further includes: Perform pixel deviation rate detection on the first encoded image; When the deviation rate of the first encoded image is detected to be within acceptable limits, the first encoded image is displayed via the display component.

[0100] In some embodiments of the present invention, it further includes: The receiver receives an abnormal feedback signal, which is a signal fed back to the sender when the receiver calls the abnormal detection module to perform abnormal detection on the first encoded image and the detection result is abnormal. The abnormal detection module includes a lightweight neural network. Obtain historical transmission parameters for the target data, the historical transmission parameters including transmission rate and characteristics of the transmitted data; The breakpoint prediction module is invoked to perform breakpoint detection based on the abnormal feedback signal and the historical transmission parameters, and to determine the data that needs to be retransmitted. The breakpoint prediction module includes an improved long short-term memory network. The data that needs to be retransmitted is re-encoded to obtain a second encoded image; The second encoded image is displayed by the display component, so that the receiving end scans the displayed second encoded image by the scanning component to obtain the retransmitted data, and sends the retransmitted data to the second security level network terminal.

[0101] Reference Figure 3 The diagram illustrates a schematic of a neural network-based data transmission device according to some embodiments of the present invention. This device is applied at a transmitting end, which is communicatively connected to a receiving end. The transmitting end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal. Specifically, it may include the following modules: The transmission request signal acquisition module 301 is used to receive a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. The environmental parameter set determination module 302 is used to determine the environmental parameter set for the transmission scenario between the sending end and the receiving end in response to the transmission request signal. The target coding scheme determination module 303 is used to call the coding optimization module to score multiple preset candidate coding schemes according to the environmental parameter set and the data attribute table, obtain the score result of each candidate coding scheme, and determine the target coding scheme from the multiple candidate coding schemes according to the score result. The coding optimization module is a convolutional neural network based on ResNet-18. The first encoded image acquisition module 304 is used to acquire the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain the first encoded image; The first encoded image sending module 305 is used to display the first encoded image through a display component, so that the receiving end can scan the displayed first encoded image through a scanning component to obtain the target data and send the target data to the second security level network terminal.

[0102] In some embodiments of the present invention, the target coding scheme determination module includes: The sub-score determination submodule is used to perform multi-dimensional scoring on multiple candidate coding schemes based on the environmental parameter set and the data attribute table, and obtain the sub-scores of the candidate coding schemes in multiple dimensions. The scoring result determination submodule is used to confirm the weights of the multiple dimensions, and to perform a weighted summation of the sub-scorings under the multiple dimensions based on the weights to obtain the scoring result of each candidate coding scheme.

[0103] In some embodiments of the present invention, the sub-scoring determination submodule includes: An initial parameter set determination unit is used to combine the environmental parameter set and the data attribute set to obtain an initial parameter set; An input vector determination unit is used to perform numerical mapping on the initial parameter set to obtain an input vector; The sub-score determination unit performs multi-dimensional scoring on multiple candidate coding schemes based on the input vector to obtain the sub-scores of the candidate coding schemes in multiple dimensions.

[0104] In some embodiments of the present invention, the first encoded image acquisition module 304 includes: The block rule determination submodule is used to determine the block rules based on the target encoding scheme; The data block determination submodule is used to divide the target data into multiple data blocks according to the block division rules. The first encoded image determination submodule is used to encode the plurality of data blocks to obtain the first encoded image.

[0105] In some embodiments of the present invention, the block rule determination submodule includes: The parameter determination unit is used to obtain the parameters of the target coding scheme, including coding resolution, error correction level and coding type; The block rule determination unit is used to determine the block rules based on the parameters of the target encoding scheme; the block rules include single block data capacity, total number of blocks, data allocation strategy and block identifier.

[0106] In some embodiments of the present invention, the first coded image determination submodule includes: A black-and-white binary dot matrix unit is used to encode the multiple data blocks block by block to obtain a black-and-white binary dot matrix. The first coded image unit is used to encode the black and white binary dot matrix to obtain the first coded image.

[0107] In some embodiments of the present invention, before the first encoded image sending module 305, the apparatus further includes: The deviation rate detection module is used to detect the pixel deviation rate of the first encoded image; The first encoded image display module is used to display the first encoded image through the display component when the deviation rate of the first encoded image is detected to be qualified.

[0108] In some embodiments of the present invention, the apparatus further includes: An abnormal feedback signal receiving module is used to receive an abnormal feedback signal from the receiving end. The abnormal feedback signal is a signal fed back to the sending end when the receiving end calls the abnormal detection module to perform abnormal detection on the first encoded image and the detection result is abnormal. The abnormal detection module includes a lightweight neural network. The historical transmission parameter acquisition module is used to acquire historical transmission parameters for the target data, the historical transmission parameters including transmission rate and characteristics of the transmitted data; The retransmission data determination module is used to call the breakpoint prediction module to perform breakpoint detection based on the abnormal feedback signal and the historical transmission parameters, and determine the data that needs to be retransmitted. The breakpoint prediction module includes an improved long short-term memory network. The second encoded image determination module is used to re-encode the data that needs to be retransmitted to obtain a second encoded image; The second encoded image display module is used to display the second encoded image through a display component, so that the receiving end can scan the displayed second encoded image through a scanning component to obtain the retransmitted data and send the retransmitted data to the second security level network terminal.

[0109] Some embodiments of the present invention also provide an electronic device, including a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the method described above.

[0110] Some embodiments of the present invention also provide a computer-readable storage medium on which a computer program is stored, and which, when executed by a processor, implements the method described above.

[0111] Some embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the method described above.

[0112] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0114] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0115] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0116] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0117] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0119] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0120] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes the aforementioned element.

[0121] The above provides a detailed description of a data transmission method, system, and apparatus based on a neural network. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for data transmission based on a neural network, characterized in that, Applied to a sending end, wherein the sending end and the receiving end are communicatively connected, the sending end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal, the method includes: The system receives a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of the target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. In response to the transmission request signal, determine a set of environmental parameters for the transmission scenario between the sending end and the receiving end; The encoding optimization module is invoked to score multiple preset candidate encoding schemes based on the environmental parameter set and the data attribute table, obtain the score result of each candidate encoding scheme, and determine the target encoding scheme from the multiple candidate encoding schemes based on the score result. The encoding optimization module is a convolutional neural network based on ResNet-18. The target data is obtained from the first security level network terminal, and the target data is encoded according to the target encoding scheme to obtain the first encoded image; The first encoded image is displayed by a display component, so that the receiving end scans the displayed first encoded image by a scanning component to obtain the target data and sends the target data to the second security level network terminal.

2. The method according to claim 1, characterized in that, Based on the environmental parameter set and the data attribute table, multiple preset candidate coding schemes are scored to obtain a score result for each candidate coding scheme, including: Based on the environmental parameter set and the data attribute table, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions. The weights of the multiple dimensions are determined, and the sub-scores under the multiple dimensions are weighted and summed according to the weights to obtain the score result of each candidate coding scheme.

3. The method according to claim 2, characterized in that, Based on the environmental parameter set and the data attribute table, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions, including: The environmental parameter set and the data attribute set are combined to obtain the initial parameter set; The initial parameter set is numerically mapped to obtain the input vector; Based on the input vector, multiple candidate coding schemes are scored in multiple dimensions to obtain sub-scores of the candidate coding schemes in multiple dimensions.

4. The method according to any one of claims 1-3, characterized in that, The target data is encoded according to the target encoding scheme to obtain a first encoded image, including: Based on the target encoding scheme, determine the block division rules; According to the segmentation rules, the target data is segmented into multiple data blocks. The multiple data blocks are encoded to obtain a first encoded image.

5. The method according to claim 4, characterized in that, Based on the target encoding scheme, the block segmentation rules are determined, including: Obtain the parameters of the target coding scheme, including coding resolution, error correction level, and coding type; Based on the parameters of the target encoding scheme, the block division rules are determined; the block division rules include single block data capacity, total number of blocks, data allocation strategy, and block identifier.

6. The method according to claim 4, characterized in that, Encoding the plurality of data blocks to obtain a first encoded image includes: The multiple data blocks are encoded block by block to obtain a black and white binary dot matrix; The black and white binary dot matrix is ​​encoded to obtain the first encoded image.

7. The method according to any one of claims 1-3, characterized in that, Before displaying the first encoded image via the display component, the method further includes: Perform pixel deviation rate detection on the first encoded image; When the deviation rate of the first encoded image is detected to be within acceptable limits, the first encoded image is displayed via the display component.

8. The method according to any one of claims 1-3, characterized in that, The method further includes: The receiver receives an abnormal feedback signal, which is a signal fed back to the sender when the receiver calls the abnormal detection module to perform abnormal detection on the first encoded image and the detection result is abnormal. The abnormal detection module includes a lightweight neural network. Obtain historical transmission parameters for the target data, the historical transmission parameters including transmission rate and characteristics of the transmitted data; The breakpoint prediction module is invoked to perform breakpoint detection based on the abnormal feedback signal and the historical transmission parameters, and to determine the data that needs to be retransmitted. The breakpoint prediction module includes an improved long short-term memory network. The data that needs to be retransmitted is re-encoded to obtain a second encoded image; The second encoded image is displayed by the display component, so that the receiving end scans the displayed second encoded image by the scanning component to obtain the retransmitted data, and sends the retransmitted data to the second security level network terminal.

9. A data transmission system based on a neural network, characterized in that, The data transmission system includes a first-level security network terminal, a sending end, a receiving end, and a second-level security network terminal; the sending end is connected to the first-level security network terminal, and the receiving end is connected to the second-level security network terminal. The first security-level network terminal is used to send a transmission request signal to the sending end. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of the target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. The transmitting end is used to determine the target encoding scheme, obtain the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain a first encoded image, and display the first encoded image through a display component; wherein, the target encoding scheme is to determine an environmental parameter set for the transmission scenario between the transmitting end and the receiving end in response to the transmission request signal, call the encoding optimization module, score multiple preset candidate encoding schemes according to the environmental parameter set and the data attribute table, obtain a score result for each candidate encoding scheme, and determine the target encoding scheme from the multiple candidate encoding schemes according to the score results; the encoding optimization module is a convolutional neural network based on ResNet-18. The receiving end is used to scan the first coded image through a scanning component to obtain the target data, and send the target data to the second security level network terminal; The second security level network terminal is used to receive the target data from the receiving end.

10. A data transmission device based on a neural network, characterized in that, The device is applied to a transmitting end, which is communicatively connected to a receiving end. The transmitting end is connected to a first-level security network terminal, and the receiving end is connected to a second-level security network terminal. The device includes: The transmission request signal acquisition module is used to receive a transmission request signal sent by the first security-level network terminal. The transmission request signal includes a data attribute table, which is generated by the first security-level network terminal through feature recognition of target data. The target data is the data that the first security-level network terminal wants to transmit to the second security-level network terminal. An environment parameter set determination module is used to determine an environment parameter set for the transmission scenario between the sending end and the receiving end in response to the transmission request signal. The target coding scheme determination module is used to call the coding optimization module, score multiple preset candidate coding schemes according to the environmental parameter set and the data attribute table, obtain the score result of each candidate coding scheme, and determine the target coding scheme from the multiple candidate coding schemes according to the score result. The coding optimization module is a convolutional neural network based on ResNet-18. The first encoded image acquisition module is used to acquire the target data from the first security level network terminal, and encode the target data according to the target encoding scheme to obtain the first encoded image; The first encoded image sending module is used to display the first encoded image through a display component, so that the receiving end can scan the displayed first encoded image through a scanning component to obtain the target data and send the target data to the second security level network terminal.

11. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the method as described in any one of claims 1 to 8.

13. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 8.

Images