A method for processing industry and finance data across a physically isolated gateway

Abstract

The application provides a kind of industry and finance data processing method across physical isolation gate, comprising: based on the overcomplete dictionary of well-trained, original signal vector of bill image is sparsely decomposed, and original sparse vector is generated;Original sparse vector is compressed and encrypted, and modulated conversion generates optical pulse signal, and optical pulse signal is transmitted from sending end to receiving end across network through physical isolation gate;Optical pulse signal is converted into observation vector, and observation vector is input into deep unfolding network model and outputs reconstruction sparse vector;Reconstruction sparse vector is restored to reconstruction industry and finance data relying on overcomplete dictionary, and double check is completed with original signal vector and electronic voucher data with digital signature, and qualified check pushes industry and finance management system, and unqualified check terminates circulation.The application can at least solve the problems of data flow fragmentation, cross-network transmission security risks, bill and data compliance risks and highly dependent on manual intervention in industry and finance data control.

Description

Technical Field

[0001] This invention relates to the field of business data processing technology, and in particular to a method for processing business data across physical isolation gateways. Background Technology

[0002] With the widespread adoption of fully digital electronic invoices, large group enterprises are rapidly moving towards a stage of deep integration of business, finance, taxation, and invoices in their material settlement processes. Tax authorities have imposed strict compliance requirements on electronic vouchers with digital signatures. Enterprises need to integrate various heterogeneous systems, such as enterprise resource planning systems, material settlement collaboration systems, and financial middleware, to achieve automated and compliant data flow across the entire business and financial chain. Therefore, an integrated intelligent management and control solution is urgently needed.

[0003] Currently, the management and control of corporate financial data still adopts the traditional model of manual multi-node collaboration and offline physical media transfer. Relevant staff complete the invoice processing and data push work by manually entering information, offline verification of documents, and physical media transfer. The entire process is highly dependent on manual operation. This traditional management and control model has the following significant defects: (1) Data flow is fragmented, and manual relay throughout the entire chain is prone to delays and extremely low business processing efficiency; (2) Physical media transfer across internal and external networks is prone to virus intrusion and data leakage, posing serious security risks; (3) The manual comparison process is postponed, which cannot ensure that the entered data is consistent with the underlying logic of the XML voucher, posing compliance risks; (4) The error rate of front-end manual entry is high, which causes the financial middle platform automation function to fail and requires repeated manual intervention.

[0004] There are currently no effective solutions to the problems of fragmented data flow, security risks in cross-network transmission, compliance risks of invoices and data, and high dependence on manual intervention in existing business and financial data management methods. Summary of the Invention

[0005] The present invention provides a method for processing business and financial data across physical isolation gateways, which at least solves the problems of data fragmentation, cross-network transmission security risks, invoice and data compliance risks, and high dependence on manual intervention in the process of business and financial data management methods in related technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for processing business and financial data across physical isolation gateways. The method includes: sparsely decomposing the original signal vector corresponding to the invoice image to be identified based on a trained overcomplete dictionary to generate an original sparse vector; wherein the original signal vector is generated by extracting target field data from the invoice image; transmitting the original sparse vector through compression, encryption, and modulation conversion to obtain an optical pulse signal, which is then transmitted from the transmitting network channel through the physical isolation gateway to the receiving network channel; inputting the observation vector obtained by converting the optical pulse signal into a deep unfolded network model of the receiving network channel, and outputting a reconstructed sparse vector from the deep unfolded network model; wherein the deep unfolded network model is trained based on historical observation vector samples and corresponding original sparse vector samples, combined with a learnable weight matrix and a soft threshold function; restoring the reconstructed sparse vector to reconstruct business and financial data based on the overcomplete dictionary, and performing dual verification with the original signal vector and the electronic voucher data with digital signature; automatically sending the reconstructed business and financial data that passes verification to the business and financial management system, and terminating the flow of the reconstructed business and financial data that fails verification.

[0008] Preferably, before performing sparse decomposition on the original signal vector corresponding to the ticket image based on the trained overcomplete dictionary, the method further includes: constructing a historical signal vector based on the original historical paper tickets after scanning and extracting target field data, and performing preprocessing; wherein, the preprocessing includes mean removal and normalization; performing sparse decomposition on the preprocessed historical signal vector based on the initialized overcomplete dictionary according to a preset sparsity constraint to obtain a sparse vector; updating the dictionary parameters of each column of the overcomplete dictionary column by column based on singular value decomposition, and performing sparse decomposition again to generate the sparse vector after each update of the dictionary parameters; when the reconstruction error corresponding to the sparse vector is less than a preset threshold or the maximum number of iterations is reached, training stops, and a trained overcomplete dictionary is obtained.

[0009] Preferably, based on a trained overcomplete dictionary, the original signal vector corresponding to the invoice image to be identified is sparsely decomposed to generate an original sparse vector, including: obtaining the invoice image to be identified by scanning the original paper invoice; based on the invoice image, locating the region where the target field is located, and extracting the target characters and target numbers in the region to obtain structured financial data; arranging the structured financial data according to a preset rule to generate a one-dimensional original signal vector; and based on the trained overcomplete dictionary, performing sparse decomposition on the original signal vector using an orthogonal matching pursuit algorithm to generate an original sparse vector that meets a preset sparsity constraint.

[0010] Preferably, the optical pulse signal obtained by sequentially compressing, encrypting, and modulating the original sparse vector is transmitted from the transmitting network channel through a physical isolation gateway to the receiving network channel, comprising: performing a linear projection operation on the original sparse vector and a dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector; performing differential phase shift modulation processing on the compressed and encrypted observation vector to generate a modulated electrical signal; performing photoelectric conversion on the modulated electrical signal to generate an optical pulse phase difference signal; and transmitting the optical pulse phase difference signal from the transmitting network channel through a physical isolation gateway to the receiving network channel.

[0011] Preferably, the original sparse vector is subjected to a linear projection operation with the dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector, including: generating a pseudo-random chaotic sequence through chaotic mapping with the timestamp hash value as the initial value; rearranging the pseudo-random chaotic sequence into an initial matrix, and then performing normalization and orthogonalization processes in sequence to obtain a dynamic observation matrix; and performing a linear projection operation on the original sparse vector based on the dynamic observation matrix to obtain a compressed and encrypted observation vector.

[0012] Preferably, before inputting the observation vector obtained by converting the optical pulse signal into the deep unfolded network model of the receiving end network channel, the method includes: constructing an initial deep unfolded network model based on a deep unfolded network architecture with a fixed number of layers, combined with a learnable weight matrix and a soft thresholding function; performing end-to-end offline training on the deep unfolded network model with historical observation vector samples as input and reconstructed sparse vectors as output; stopping training until the training loss between the reconstructed sparse vectors and the corresponding original sparse vector samples is less than a preset convergence threshold, or a preset number of training iterations is reached, thereby obtaining a trained deep unfolded network model.

[0013] Preferably, the hierarchical iterative update formula of the deep unfolded network model is: ;in, For the deep unfolded network model, the first The reconstructed sparse vector output by the layer; For the first The first learnable weight matrix corresponding to the layer; This is the transpose of the dynamic observation matrix; To compress and encrypt the observation vector: For the first The second learnable weight matrix corresponding to the layer; For the deep unfolded network model, the first The reconstructed sparse vector output by the layer; For the first The third learnable weight matrix corresponding to the layer; For the first The soft threshold activation function corresponding to the layer.

[0014] Preferably, the reconstructed sparse vector is restored to reconstructed business and financial data based on the overcomplete dictionary, and then double-verified with the original signal vector and the electronic voucher data with digital signature. The reconstructed business and financial data that passes the verification is automatically sent to the business and financial management system; the reconstructed business and financial data that fails the verification is terminated. This includes: restoring the reconstructed sparse vector to reconstructed business and financial data based on the overcomplete dictionary; comparing the reconstructed business and financial data with the original signal vector for numerical consistency verification; comparing the qualified data that passes the numerical consistency verification with the electronic voucher data with digital signature for compliance verification; when both the numerical consistency verification and the compliance verification pass, the data is deemed qualified, and the qualified reconstructed business and financial data is automatically sent to the business and financial management system; when either the numerical consistency verification or the compliance verification fails, the data is deemed unqualified, and the reconstructed business and financial data that fails the verification is terminated.

[0015] Preferably, comparing the reconstructed financial data with the original signal vector to perform numerical consistency verification includes: comparing the reconstructed financial data with the original signal vector and calculating the deviation between the reconstructed financial data and the original signal vector; comparing the deviation with a preset deviation threshold; if the deviation is less than or equal to the preset deviation threshold, the numerical consistency verification passes, otherwise it fails.

[0016] In a second aspect, the present invention provides an electronic device, comprising: a processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the aforementioned business data processing method across a physical isolation gateway.

[0017] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0018] This invention provides a method for processing business and financial data across physical isolation gateways. It extracts target fields from invoice images to generate original signal vectors, then performs complete dictionary sparse decomposition, gateway transmission, and deep network model expansion to restore the data. A unified processing rule connects the entire business process, solving the problems of fragmented and disjointed business and financial data flow, achieving seamless data flow across the entire chain. The original signal vector is converted into a sparse vector, compressed, encrypted, and modulated into an optical pulse signal for cross-gate transmission, abandoning the conventional plaintext transmission mode and addressing security risks in cross-network data transmission, effectively preventing data interception, tampering, and leakage. The receiving network channel obtains a reconstructed sparse vector based on the optical pulse signal, simultaneously verifying it against the original signal vector and digitally signed electronic vouchers, intercepting unqualified data, resolving compliance risks such as inconsistencies, tampering, and voucher mismatches in invoice and financial data, and ensuring compliant and reliable data flow. The entire process relies on automated program and model operation, eliminating the need for manual extraction of invoice information, cross-network data transfer, data verification, and system entry, thus solving the problem of excessive reliance on manual intervention in business processes, saving labor costs and avoiding human error. Attached Figure Description

[0019] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other embodiments based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the core steps of a business data processing method for a cross-physical isolation gateway according to an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the entire process of the business data processing method according to an embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the electronic device created by this invention. Detailed Implementation

[0023] Embodiments of the present invention will now be described in more detail with reference to the accompanying drawings. While some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0024] To address the issues of fragmented data flow, security risks in cross-network transmission, compliance risks of invoices and data, and high dependence on manual intervention in business and financial data management methods in related technologies, the embodiments of this invention provide a business and financial data processing method that crosses physical isolation gateways.

[0025] Among them, such as Figure 1 , Figure 2 As shown, the embodiment of the present invention provides a business data processing method across physical isolation gateways, including the following steps S1 to S4.

[0026] Step S1: Based on the trained overcomplete dictionary, perform sparse decomposition on the original signal vector corresponding to the ticket image to be identified to generate the original sparse vector; wherein, the original signal vector is generated by extracting the target field data from the ticket image.

[0027] Step S2: The optical pulse signal obtained by sequentially compressing, encrypting and modulating the original sparse vector is transmitted from the transmitting network channel through a physical isolation gateway to the receiving network channel.

[0028] Step S3: The observation vector obtained by converting the optical pulse signal is input into the deep unfolding network model of the receiving end network channel, and the deep unfolding network model outputs the reconstructed sparse vector; wherein, the deep unfolding network model is trained based on historical observation vector samples and corresponding original sparse vector samples, combined with a learnable weight matrix and a soft thresholding function.

[0029] Step S4: The reconstructed sparse vector is restored to reconstructed business and financial data based on the overcomplete dictionary, and then double-verified with the original signal vector and the electronic voucher data with digital signature. The reconstructed business and financial data that passes the verification is automatically sent to the business and financial management system, and the reconstructed business and financial data that fails the verification is terminated.

[0030] An overcomplete dictionary is a base set with a number of atoms greater than the vector dimension and possessing redundant representation capabilities. An initialized overcomplete dictionary can be trained using the K-SVD (K-Singular Value Decomposition) algorithm, which adapts to ticket features through sparse coding and iterative atomic updates; alternatively, it can be trained using MOD (Method of Optimal Directions) to adapt for rapid database construction with small batches of samples.

[0031] The image of the document to be identified is the image carrier of a paper document, such as a purchase order or invoice. The image can be obtained by scanning the original paper document with a high-speed scanner, or by taking a photo using a document scanner or mobile device.

[0032] The original signal vector is generated by extracting target field data from the image of the document, splicing the target field data features, and regularizing the dimensions to transform it into a fixed-dimensional original signal vector, thus realizing the conversion of paper information into digital vectors.

[0033] The target field data may include order number, invoice code, invoice date, amount, and buyer / seller name. It can be obtained by locating the corresponding field area in the invoice image and extracting the character and numerical information through OCR (Optical Character Recognition) technology.

[0034] Using a well-trained overcomplete dictionary as a basis, the original signal vector is projectively decomposed, redundant noise is filtered out, and the non-zero coefficients of the core fields are retained to generate a simplified original sparse vector.

[0035] The above-mentioned step S1 provided by the embodiments of the present invention overcomes the defects of redundant and noisy feature extraction, compresses the data volume while retaining core information, lays the foundation for subsequent transmission and reconstruction, and adapts to the unified representation of multiple types of invoices.

[0036] Compression encryption is available, with options for sparse domain chaotic encryption, AES (Advanced Encryption Standard) block encryption, and lightweight chaotic sequence encryption. By compressing and encrypting the original sparse vector, high-strength security protection can be achieved while significantly reducing data volume. It combines the dual effects of data compression and encryption, adapts to the narrow channel transmission characteristics of physically isolated gateways, and effectively prevents the risks of data interception, tampering, and leakage during cross-network transmission.

[0037] The encrypted sparse vector code is loaded onto the optical carrier through an electro-optic modulator and converted into an optical pulse signal that can be transmitted across network gates, thereby improving the transmission anti-interference capability.

[0038] The sending network channel is a dedicated isolated transmission link deployed on the business external network. It only carries the encrypted modulation of ticket data and the front-end transmission of the network gateway, and is physically isolated from other links.

[0039] The receiving network uses a dedicated receiving link deployed within the financial intranet to specifically receive and process optical pulse signals transferred by the network gateway, and is logically isolated from the external public network.

[0040] A physical isolation gateway is a security device that completely disconnects internal and external networks. It can only control the data transfer and prevent cross-network attacks. Specifically, it can be a gateway device from NetGuard Nebula, Topsec, or Venustech.

[0041] The above-mentioned step S2 provided in the embodiments of the present invention eliminates the drawbacks of plaintext and mixed-stream transmission, and improves the confidentiality and security of cross-network transmission of business and financial data by combining optical pulse transmission with compression encryption and physical isolation of network gateways.

[0042] The observation vector is a low-dimensional vector obtained by demodulating and restoring the optical pulse signal to a digital sequence through a photodetector at the receiving end, and then performing dimension normalization. It is the representation of the original sparse vector after transmission distortion. At the receiving end, the phase difference signal of the optical pulse is demodulated into an electrical signal through a photodetector, and then normalized to obtain the observation vector input to the adapted depth unfolded network model.

[0043] Historical observation vector samples are distortion-free sparse vectors generated by sparse decomposition of standard tickets. Historical observation vector samples are samples obtained by simulating cross-network transmission and superimposing channel distortion on the original sparse vector. Both types of samples can be generated by batch collection of standard tickets and offline sparse decomposition and simulated transmission, or by collecting and labeling real data from actual cross-network transmission.

[0044] Deep unfolded network models can adopt the LISTA (Learned Iterative Shrinkage-Thresholding Algorithm) architecture, the ISTA (Iterative Shrinkage-Thresholding Algorithm) architecture, or the ADMM (Alternating Direction Method of Multipliers) architecture.

[0045] Using historical observation vector samples as input and corresponding original sparse vector samples as labels, a fixed-layer deep unfolded network model is constructed, embedding a learnable weight matrix and a soft threshold function. An end-to-end offline training method is adopted to iteratively update the weight parameters and soft threshold until the error between the reconstructed sparse vector output by the model and the original sparse vector sample reaches the standard, thus completing the training.

[0046] The above-mentioned step S3 provided in the embodiments of the present invention overcomes the defect of weak anti-interference of reconstruction algorithms in related technologies, accurately restores the original sparse vector based on the customized model, reduces transmission distortion, and ensures the accuracy of data reconstruction.

[0047] Using a well-trained overcomplete dictionary as a base, the reconstructed sparse vectors are subjected to inverse weighted fitting and feature inverse mapping to restore the structured reconstructed financial data, thus realizing the inverse conversion of sparse vectors into financial data.

[0048] Electronic voucher data with digital signatures are standardized electronic business and financial vouchers issued by the business source and affixed with a legal digital signature. They can be obtained synchronously from the electronic tax bureau, banking system, or enterprise business system.

[0049] The dual verification process involves first comparing the reconstructed business and financial data with the original signal vector to perform a numerical consistency verification, and then comparing it with the electronic voucher data bearing a digital signature to perform a compliance verification.

[0050] In the numerical consistency verification, the deviation between the reconstructed business and financial data and the original signal vector is calculated. If the deviation is less than or equal to the preset threshold, it passes. In the compliance verification, the core fields of the reconstructed business and financial data are checked against the core fields of the electronic voucher with digital signature. If the fields are consistent and the digital signature is valid, it passes. If both pass, the verification is qualified. If either fails, the data flow is terminated.

[0051] Business and finance management system, including ERP (Enterprise Resource Planning) system, financial middle platform and other enterprise-level integrated business and finance management and control platform.

[0052] The above-mentioned step S4 provided in the embodiment of the present invention accurately intercepts anomalies such as data tampering and transmission distortion through a dual verification mechanism. Unlike the traditional single verification method, it controls both the authenticity and compliance of the data, ensuring that the reconstructed business and financial data is credible and compliant, while providing reliable support for the subsequent data entry of the business and financial management system.

[0053] The above-mentioned business and financial data processing method across physical isolation gateways provided by the embodiments of the present invention extracts target fields from the invoice image to generate the original signal vector, and then completes the data restoration through complete dictionary sparse decomposition, gateway transmission and deep expansion network model. The unified processing rules connect the entire business process, which solves the problems of fragmented business and financial data flow and disconnected links, and realizes the seamless flow of data across the entire link.

[0054] The method described in the embodiments of this invention converts the original signal vector into a sparse vector, compresses and encrypts it, and modulates it into an optical pulse signal for cross-network transmission. This eliminates the conventional plaintext transmission mode, solves the security risks of cross-network transmission of business and financial data, and effectively prevents data from being intercepted, tampered with, or leaked.

[0055] The method described in the embodiments of this invention obtains a reconstructed sparse vector based on the optical pulse signal through the receiving end network channel, and simultaneously verifies it with the original signal vector and the digitally signed electronic voucher, and intercepts unqualified data. This solves the compliance risks of data discrepancies, tampering, and voucher mismatches in the bill industry, and ensures the compliance and reliability of data flow.

[0056] The method described in the embodiments of this invention is fully automated through programs and models, eliminating the need for manual extraction of invoice information, cross-network data transfer, data verification, and system entry. This solves the problem of excessive reliance on manual intervention in business processes, saves labor costs, and avoids human error.

[0057] Furthermore, the method provided in the embodiments of the present invention preferably further includes, before performing step S1: constructing a historical signal vector based on the original historical paper tickets after scanning and extracting target field data, and performing preprocessing; wherein, the preprocessing includes mean removal and normalization; performing sparse decomposition on the preprocessed historical signal vector based on the initialized overcomplete dictionary according to a preset sparsity constraint to obtain a sparse vector; updating the dictionary parameters of each column of the overcomplete dictionary column by column based on singular value decomposition, and performing sparse decomposition again to generate a sparse vector after each update of the dictionary parameters; when the reconstruction error corresponding to the sparse vector is less than a preset threshold or the maximum number of iterations is reached, training stops, and a trained overcomplete dictionary is obtained.

[0058] Specifically, the system first scans and collects data based on the original historical paper documents. During the system initialization phase, a sufficient number of document samples are collected, and target field data such as order number, invoice code, and amount are extracted. The data are then arranged according to rules to construct historical signal vectors and training sample sets. All historical signal vectors are subjected to mean removal and normalization preprocessing to eliminate the influence of DC components and dimensions.

[0059] Using the initialized overcomplete dictionary, and according to the preset sparsity constraints, the preprocessed historical signal vector is sparsely decomposed using the Orthogonal Matching Pursuit (OMP) algorithm to obtain the corresponding original sparse vector.

[0060] The pre-set sparsity constraint is set according to the number of target fields and feature redundancy of the document. In the preferred embodiment of the present invention, S=5, where S represents the pre-set sparsity constraint. This can filter out nonlinear distortion noise, retain core field features, reduce the sparse vector dimension, and improve compression transmission efficiency.

[0061] The dictionary parameters of each column of the complete dictionary are updated column by column based on K-Singular Value Decomposition (K-SVD). After each dictionary parameter update, the historical signal vector is sparsely decomposed again to generate a new sparse vector.

[0062] The sparse decomposition and dictionary update steps are executed repeatedly. When the reconstruction error corresponding to the sparse vector is less than the preset threshold, or the number of iterations reaches the maximum number of iterations, the training terminates. Finally, a well-trained overcomplete dictionary that adapts to the nonlinear distortion features of the invoice and can improve the sparse representation capability of the signal is obtained.

[0063] The preset threshold is selected based on the accuracy of financial data and the requirements for signal fidelity. In a preferred embodiment of this invention... It can strictly control reconstruction deviations and meet the requirements of financial compliance and high-precision data verification.

[0064] The maximum number of iterations is selected by considering the convergence characteristics of the K-SVD algorithm, training accuracy, and system efficiency. In this invention, the preferred embodiment is 100 iterations, which can ensure that the dictionary converges to the optimal state and control the training time, thus meeting the needs of rapid deployment in engineering.

[0065] The method described in the embodiments of this invention constructs and preprocesses signal vectors based on historical paper tickets, and uses the K-singular value decomposition algorithm combined with orthogonal matching pursuit sparse decomposition and singular value decomposition to update the dictionary column by column and iteratively converge to train. This method can accurately adapt to the fixed format of the tickets and nonlinear distortion features such as tilt and wrinkles, and greatly improves the sparse representation capability of the ticket signals and the accuracy of data reconstruction.

[0066] Further, step S1 of the present invention preferably includes: obtaining an image of the document to be identified by scanning the original paper document; locating the area where the target field is located based on the document image, and extracting the target characters and target numbers in the area content to obtain structured financial data; arranging the structured financial data according to preset rules to generate a one-dimensional original signal vector; and performing sparse decomposition on the original signal vector using an orthogonal matching pursuit algorithm based on a trained overcomplete dictionary to generate an original sparse vector that meets preset sparsity constraints.

[0067] Specifically, in the preferred embodiment of this invention, the original paper business documents such as purchase orders and invoices are scanned and collected using high-speed scanners, document scanners, and other equipment to obtain images of the documents to be identified, which contain complete document information.

[0068] Next, relying on OCR optical character recognition technology, the target field regions of the invoice image are located, and the target fields such as invoice number, invoice date, amount, buyer and seller names, and order number are accurately identified. The character and numerical information in the corresponding regions are extracted, and the unstructured information on the invoice is organized into structured business and financial data with standardized format and complete fields.

[0069] According to the preset field arrangement and dimension regularization rules, the structured business and financial data is feature spliced ​​and formatted, and converted into a one-dimensional original signal vector with fixed dimensions. This completes the conversion of paper ticket information into a standardized digital vector, which meets the input conditions for subsequent sparse decomposition.

[0070] By utilizing a pre-trained, overcomplete dictionary adapted to the nonlinear distortion features of invoices, the orthogonal matching pursuit algorithm is used to perform sparse decomposition on the original signal vector. Under the preset sparsity constraint, redundant noise caused by image distortion is filtered out, core business data features are retained, and a concise original sparse vector that meets the constraint requirements is generated.

[0071] The method provided in the embodiments of this invention, through document scanning and acquisition, OCR field extraction, signal vector normalization, and overcomplete dictionary combined with orthogonal matching pursuit sparse decomposition, can achieve nonlinear distortion adaptation of documents, efficient data structuring, and lightweight sparse vector representation.

[0072] Furthermore, step S2 of the method provided in the present invention preferably includes: performing a linear projection operation on the original sparse vector and the dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector; performing differential phase shift modulation processing on the compressed and encrypted observation vector to generate a modulated electrical signal; performing photoelectric conversion on the modulated electrical signal to generate an optical pulse phase difference signal; and transmitting the optical pulse phase difference signal from the transmitting end network channel through a physical isolation gateway to the receiving end network channel.

[0073] Specifically, the method provided in the embodiments of the present invention first performs a linear projection operation on the dynamic observation matrix generated by the original sparse vector and the chaotic mapping, which can simultaneously complete the dimensionality compression and security encryption of the data, and obtain a compressed and encrypted observation vector that meets the requirements of cross-network transmission, adapting to the narrow bandwidth transmission characteristics of the physical isolation gateway.

[0074] Chaotic mappings can include logistic chaotic mappings, tent chaotic mappings, Henon chaotic mappings, Lorentz chaotic mappings, Chua's chaotic mappings, etc. Specifically, if the embodiments of this invention employ logistic chaotic mapping, a uniformly distributed chaotic sequence is generated iteratively by setting initial iteration values ​​and control parameters, and then reconstructed and normalized according to the target row and column dimensions to generate a dynamic observation matrix adapted for projection operations. If the embodiments of this invention employ tent chaotic mapping, an aperiodic pseudo-random chaotic sequence is generated based on piecewise linear iteration rules, and after numerical interval mapping and dimensional pruning and regularization, a dynamic observation matrix for sparse vector compression and encryption is constructed.

[0075] Next, differential phase-shift modulation (DPSK) is applied to the compressed and encrypted observation vector, using a Mach-Zehnder modulator (MZM) to convert the numerical amplitudes of the vector elements into phase changes of the optical pulses. For the first... Each symbol has a phase of 1. What is actually transmitted is the phase difference between adjacent symbols, that is... The numerical characteristics of the vector are converted into a standard electrical signal form, generating a modulated electrical signal with stronger anti-interference capabilities, thus ensuring the stability of the signal during transmission.

[0076] The modulated electrical signal is converted into an optical pulse phase difference signal through a photoelectric conversion device. Relying on the advantages of optical signal transmission, it is adapted to the optical domain transmission mode of the physical isolation gate to complete the safe conversion of electrical signals to optical signals.

[0077] The optical pulse phase difference signal is transmitted securely to the receiving network channel of the financial intranet via a dedicated channel of the external network transmitter and a physical isolation gateway, thus completing the cross-network transmission of business and financial data in a physically isolated environment between the internal and external networks.

[0078] The method provided by the embodiments of the present invention generates a dynamic observation matrix through chaotic mapping, performs linear projection compression and encryption, differential phase shift modulation, photoelectric conversion to generate optical pulse phase difference signals, and transmits them across networks via a physical isolation gateway. This enables integrated compression and encryption processing, reduces the time delay of step-by-step processing, and transmits in the form of optical pulse phase difference, which greatly improves the anti-electromagnetic interference capability and data transmission concealment of cross-network transmission.

[0079] Furthermore, the method described in the embodiments of the present invention, which involves performing a linear projection operation on the original sparse vector and the dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector, preferably includes: generating a pseudo-random chaotic sequence using a timestamp hash value as the initial value through chaotic mapping; rearranging the pseudo-random chaotic sequence into an initial matrix, and then performing normalization and orthogonalization processes sequentially to obtain a dynamic observation matrix; and performing a linear projection operation on the original sparse vector based on the dynamic observation matrix to obtain a compressed and encrypted observation vector.

[0080] The preferred embodiment of the present invention is logical chaotic mapping. Since the present invention requires the rapid generation of pseudo-random sequences and the real-time iterative construction of dynamic observation matrices, and requires low computational complexity and uniform and controllable sequence distribution, logical chaotic mapping can rapidly generate irregular chaotic sequences, support dynamic updates of the observation matrix, improve the security level of sparse vector projection encryption, reduce computation time, and ensure efficient cross-network transmission of business and financial data.

[0081] Specifically, using the timestamp hash value as the initial value, a pseudo-random chaotic sequence is generated through chaotic mapping: Logistic chaotic mapping is adopted and full chaotic state parameters are set. The current frame timestamp is hashed using SHA-256 and the feature bits are truncated and mapped to a preset interval as the initial value for iteration. A pseudo-random chaotic sequence of fixed length is generated iteratively, and the transient iteration values ​​in the early stage are discarded to eliminate the initial disturbance interference.

[0082] The pseudo-random chaotic sequence is rearranged into an initial matrix, and then normalized and orthogonalized sequentially to obtain the dynamic observation matrix: the effective pseudo-random chaotic sequence is reconstructed according to the row priority rule to generate the initial matrix, the initial matrix is ​​normalized by columns and then Gram-Schmidt orthogonalized, and finally the size-adapted dynamic observation matrix is ​​generated.

[0083] A compressed and encrypted observation vector is obtained by performing a linear projection operation on the original sparse vector based on the dynamic observation matrix: the dimensionality reduction transformation from high-dimensional vector to low-dimensional vector is completed by performing a linear projection calculation on the dynamic observation matrix and the original sparse vector, and the data encryption is realized simultaneously, and the compressed and encrypted observation vector is output.

[0084] The method provided by the embodiments of the present invention initializes the chaotic sequence through timestamp hashing, constructs a dynamic observation matrix through matrix reconstruction, normalization and orthogonalization, and performs linear projection on the original sparse vector. This enables the observation matrix to be dynamically and randomly variable, and simultaneously completes data dimensionality reduction and compression and chaotic encryption, effectively preventing matrix reverse engineering and improving the confidentiality and non-replicability of business and financial data transmitted across networks.

[0085] Furthermore, the method provided in the embodiments of the present invention preferably includes, before step S3: constructing an initial deep unfolded network model based on a deep unfolded network architecture with a fixed number of layers, combined with a learnable weight matrix and a soft thresholding function; performing end-to-end offline training on the deep unfolded network model with historical observation vector samples as input and reconstructed sparse vectors as output; stopping training until the training loss between the reconstructed sparse vectors and the corresponding original sparse vector samples is less than a preset convergence threshold, or the preset number of training iterations is reached, thereby obtaining a trained deep unfolded network model.

[0086] Specifically, the fixed-layer depth unfolded network architecture decomposes the traditional sparse reconstruction iterative algorithm layer by layer and maps it into a neural network topology with a fixed number of layers. Each layer corresponds to one round of sparse reconstruction iterative operation, and each layer is independently configured with learnable parameters and threshold functions, thus realizing the organic integration of sparse reconstruction algorithm and neural network structure.

[0087] The present invention preferably adopts the LISTA architecture. Since the sparse reconstruction iterative algorithm is deeply expanded into a neural network, the weights and thresholds can be adaptively learned, converge faster, and have stronger resistance to transmission distortion. It can adapt to the sparse features of invoices and has high reconstruction accuracy.

[0088] The fixed number of layers can be selected based on the complexity of the sparse features of the invoice, the reconstruction accuracy index, and the model inference latency. In this embodiment of the invention, it is preferably set to 8 to 12 layers. By setting this range of layers, it can fully fit the inherent sparse features of the invoice and the distortion law of cross-network channel transmission, ensuring high-precision reconstruction of sparse vectors, while avoiding the problems of parameter redundancy and increased inference latency caused by too many layers, thus balancing the model fitting ability and the real-time performance of engineering implementation.

[0089] Based on the fixed-level LISTA architecture, and following the hierarchical iterative update logic, multiple sets of learnable weight matrices are embedded in each layer of the network, and an independent soft threshold function is matched for each layer. The transpose of the dynamic observation matrix, the compressed and encrypted observation vector, and the reconstructed output vector of the previous layer are used as the input for each layer. The forward inference logic is built based on the inter-layer iterative correlation mapping relationship to complete the network topology construction and initial parameter configuration, thereby constructing the initial deep unfolded network model.

[0090] Using historical observation vector samples with channel transmission distortion as input to the deep unfolded network model, and matching the corresponding original sparse vector samples as standard output labels, an end-to-end offline training method is adopted to drive the deep unfolded network to autonomously learn the nonlinear mapping relationship from observation vectors to sparse vectors, and adaptively fit the channel distortion characteristics of cross-network transmission and the inherent sparse distribution law of tickets.

[0091] The loss error between the reconstructed sparse vector output by the deep unfolded network model and the standard original sparse vector sample is used as the optimization criterion. When the training loss converges to the accuracy threshold range, or when the training iteration reaches the set upper limit, the parameter iteration update is terminated. The learnable weights and soft threshold parameters of each layer of the network are solidified, and finally a deep unfolded network model with distortion compensation and accurate reconstruction capabilities is obtained.

[0092] The preset convergence threshold is selected based on the accuracy of business and financial invoice data verification and the requirements for signal fidelity and compliance. The preferred value in this invention is [value missing]. It can strictly constrain the error deviation between the reconstructed sparse vector and the original sparse vector, and meet the requirements of accurate matching of financial fields. The preset number of training iterations is set in combination with the convergence characteristics of the LISTA architecture and the efficiency of offline training. The present invention is preferably set to 200 to 300 times, which can ensure that the model parameters fully converge to the optimal state, and avoid ineffective iterations that waste computing power and lengthen the model training cycle.

[0093] By performing sparse decomposition on standard ticket samples and superimposing cross-network channel transmission distortion, 100,000 sets of data pairs matching historical observation vectors with the original sparse vectors are generated in batches as the offline training sample set for the model. The Adam optimizer is used for end-to-end training of the deep unfolded network, adaptively iteratively updating the weight matrices and soft threshold parameters of each layer to accelerate model convergence. After model training, it is put into online business inference, where sparse reconstruction of the observation vectors transmitted through the network gateway is performed, with the time consumption of a single inference cycle controlled within [timeframe value missing]. Within this range, it can support high-concurrency, low-latency cross-network real-time processing of various types of business and financial documents.

[0094] The method provided by the embodiments of the present invention, by constructing a fixed-depth unfolded network that integrates a learnable weight matrix and a soft threshold function and conducting end-to-end offline convergence training, enables the network to autonomously learn cross-network channel distortion and ticket sparsity features, achieve adaptive optimization of model parameters, overcome the shortcomings of poor adaptability of traditional reconstruction algorithms, and effectively improve the accuracy of sparse vector anti-distortion reconstruction while taking into account training efficiency.

[0095] Furthermore, the formula for the hierarchical iterative update of the deep unfolded network model is:

[0096] .

[0097] in, For the first deep unfolding network model The reconstructed sparse vector output by the layer; For the first The first learnable weight matrix corresponding to the layer; This is the transpose of the dynamic observation matrix; To compress and encrypt the observation vector: For the first The second learnable weight matrix corresponding to the layer; For the first deep unfolding network model The reconstructed sparse vector output by the layer; For the first The third learnable weight matrix corresponding to the layer; For the first The soft threshold activation function corresponding to the layer.

[0098] The iterative update formula provided in the embodiments of this invention uses three sets of learnable weight matrices for each layer as the basis for feature regulation, integrates the transpose of the dynamic observation matrix, the input compressed and encrypted observation vector, and the reconstructed sparse vector output from the previous layer, and then completes sparse feature screening and noise suppression through the corresponding layer soft threshold function. Through multi-layer iterative correction, it continuously approximates the original sparse vector without distortion.

[0099] This invention incorporates the aforementioned iterative update formula into a deep unfolded network model and applies it to the scenario of reconstructing business and financial invoice data across physical isolation gateways. Relying on the independent learnable weight matrices configured at each level of the deep unfolded network, it adaptively fits the transmission distortion patterns of internal and external network channels. Combined with the soft threshold function set by the network layers, it accurately preserves the core sparse features of the invoices and filters out transmission noise. By leveraging the multi-layer iterative mechanism of the network, it continuously optimizes the reconstruction results, thus overcoming the shortcomings of related technologies in reconstruction algorithms, such as weak anti-distortion ability and poor invoice adaptability.

[0100] Further, step S4 of the method provided in the embodiments of the present invention preferably includes: restoring the reconstructed sparse vector to reconstructed business and financial data based on an overcomplete dictionary; comparing the reconstructed business and financial data with the original signal vector to perform numerical consistency verification; comparing the qualified data that passes the numerical consistency verification with the electronic voucher data with digital signature to perform compliance verification; when both the numerical consistency verification and the compliance verification pass, it is determined that the verification is qualified, and the qualified reconstructed business and financial data is automatically sent to the business and financial management system; when the numerical consistency verification fails or the compliance verification fails, it is determined that the verification is unqualified, and the unqualified reconstructed business and financial data is terminated from circulation.

[0101] Specifically, relying on the trained overcomplete dictionary, the reverse sparse representation operation is performed on the reconstructed sparse vector. The sparse vector is reversed to the original signal dimension through dictionary atomic weighted fitting, and the structured reconstructed business and financial data containing key fields such as order number, amount, and invoice code is restored, thus completing the restoration and transformation from sparse representation to actual business and financial information.

[0102] The reconstructed financial data is compared with the original signal vector generated from the extracted document image by comparing the values ​​of key fields. The deviation between the two is calculated to verify the numerical fidelity of the reconstructed data and the original document extracted data, and to investigate numerical distortion problems caused during cross-network transmission and signal reconstruction.

[0103] The reconstructed business and financial data that has passed numerical verification is matched and verified against the core fields of the electronic voucher data obtained synchronously from the electronic tax bureau and accompanied by a valid digital signature. At the same time, the legality of the digital signature of the electronic voucher is verified to ensure that the reconstructed data is consistent with the legal electronic voucher information and to avoid the risk of data tampering and voucher mismatch.

[0104] Both verifications passed, indicating that the reconstructed data values ​​are true and the vouchers are compliant. The data is automatically pushed to business and financial management systems such as ERP and financial middle platform, realizing full-link automated integration of data without the need for manual secondary entry and verification.

[0105] If any verification step fails, indicating that the data is distorted, tampered with, or non-compliant, the data flow will be immediately intercepted and terminated to filter abnormal data at the source, ensuring the data quality and compliance of the business and finance management system.

[0106] The method provided by the embodiments of the present invention obtains business and financial data by reconstructing sparse vectors through inverse reconstruction of an overcomplete dictionary, and sequentially performs numerical consistency verification and electronic voucher compliance verification. Based on the verification results, the data flow path is diverted, which can achieve dual protection of the numerical authenticity of the reconstructed data and the compliance of vouchers, accurately intercept abnormal situations such as data distortion and tampering, avoid the risk of non-compliance of business and financial data, and at the same time ensure the automated flow of qualified data, thereby improving data quality and integration efficiency.

[0107] The embodiments of this invention also provide a business and financial data processing system across physical isolation gateways, applied to the business and financial data processing method across physical isolation gateways provided in the above embodiments. Optionally, the business and financial data processing system across physical isolation gateways provided in the embodiments of this invention preferably includes: a document settlement module, an OCR module, a transmission modulation module, a physical isolation gateway, an inspection module, a collaboration module, and a business and financial management system.

[0108] Specifically, the document delivery module is responsible for document image acquisition, supporting image input in step S1; the OCR module has a built-in complete dictionary to realize sparse decomposition in step S1 and generation of the original sparse vector in step S2; the transmission modulation module works with the physical isolation gateway to realize chaotic encryption, optical domain modulation, and cross-network transmission in step S2; the verification module is equipped with a deep unfolded network to complete sparse reconstruction, data restoration, and dual verification in step S3; the business and finance management system receives qualified data and blocks unqualified data; and the collaboration module coordinates the data interaction of various modules to ensure full-process linkage.

[0109] The document delivery module integrates an OCR module to complete the scanning, recognition, and data preprocessing of paper documents. The OCR module uses the PyTorch framework to train the model and convert it to ONNX format for deployment. It uses the YOLOv5 model to achieve document region localization, with an input image resolution greater than or equal to 300 DPI and an output bounding box coordinate accuracy greater than or equal to 99%.

[0110] The YOLOv5 model was iterated for 200 rounds with a learning rate of 0.01 and a batch size of 32, and Mosaic data augmentation was used. A directional classification model was built with VGG16 as the backbone network, the first 10 layers were frozen, and iterated for 150 rounds with a learning rate of 0.001 and a batch size of 64.

[0111] The text recognition function of the OCR module adopts the SVTR (Single Visual model for Scene Text Recognition) model based on Transformer and CTC (Connectionist Temporal Classification) loss function. It achieves a recognition accuracy of ≥98.5% and a single-image processing time of less than 2 seconds. The SVTR model iterates for 300 rounds with a learning rate of 0.0005 and a batch size of 128. It achieves a 3:1 compression ratio through INT8 (8-bit Integer) quantization and is deployed on NVIDIA Jetson Orin NX.

[0112] The verification module uses a 10Gbps unidirectional optical shutter with a transmission delay of less than or equal to 1ms. On the external network side, it calls the electronic tax bureau interface every 5 minutes via HTTPS protocol to pull the invoice XML file. On the internal network side, a DOM4J parser and verification rule engine are deployed to verify the digital signature and status of electronic vouchers.

[0113] For the specific implementation process, please refer to steps S1 to S4 of the business data processing method across the physical isolation gateway in the above embodiment, which will not be repeated here.

[0114] The system provided by the embodiments of this invention, by adopting a modular architecture of bill settlement, OCR, transmission modulation, physical isolation gateway, inspection, collaboration and business and financial management system, integrates high-precision OCR recognition, chaotic dynamic encryption, deep expansion network rapid reconstruction and dual verification mechanism, can realize secure cross-network transmission of business and financial bill data, high-precision rapid reconstruction and accurate compliance verification, eliminate data distortion and compliance risks, ensure automated flow of data throughout the entire link, and adapt to the deep integration scenario of business, finance, taxation and bills.

[0115] Embodiments of the present invention also provide a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of the present invention.

[0116] Embodiments of the present invention also provide a computer program product, including a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform the method of an embodiment of the present invention.

[0117] An embodiment of the present invention also provides an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which, when executed by the at least one processor, causes the electronic device to perform the method of the embodiment of the present invention.

[0118] refer to Figure 3The present invention will now describe a structural block diagram of an electronic device that can serve as an embodiment of the present invention, serving as an example of a hardware device applicable to various aspects of the present invention. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0119] like Figure 3 As shown, the electronic device includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 502 or loaded from storage unit 508 into RAM (Random Access Memory) 503. RAM 503 can also store various programs and data required for the operation of the electronic device. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. An I / O interface (Input / Output Interface) 505 is also connected to bus 504.

[0120] Multiple components in the electronic device are connected to I / O interface 505, including: input unit 506, output unit 507, storage unit 508, and communication unit 509. Input unit 506 can be any type of device capable of inputting information into the electronic device. Input unit 506 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 507 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 508 may include, but is not limited to, disks and optical discs. Communication unit 509 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and / or wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0121] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a CPU (Central Processing Unit), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of the present invention can be implemented as a computer program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 502 and / or communication unit 509. In some embodiments, the computing unit 501 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0122] Computer programs for implementing the methods of embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0123] In the context of embodiments of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, or infrared systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0124] It should be noted that the term "comprising" and its variations used in the embodiments of this invention are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "a plurality" mentioned in the embodiments of this invention are illustrative and not restrictive, and those skilled in the art should understand that unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0125] The steps described in the method embodiments provided by the present invention can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of the present invention is not limited in this respect.

[0126] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily imply the same embodiment, nor does it imply independence or alternativeity from other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0127] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for processing business and financial data across physical isolation gateways, characterized in that, The method includes: Based on a trained overcomplete dictionary, the original signal vector corresponding to the ticket image to be identified is sparsely decomposed to generate an original sparse vector; wherein, the original signal vector is generated by extracting target field data from the ticket image; The optical pulse signal obtained by sequentially compressing, encrypting and modulating the original sparse vector is transmitted from the transmitting network channel through a physical isolation gateway to the receiving network channel. The observation vector obtained by converting the optical pulse signal is input into the deep unfolded network model of the receiving end network channel, and the deep unfolded network model outputs a reconstructed sparse vector; wherein, the deep unfolded network model is trained based on historical observation vector samples and corresponding original sparse vector samples, combined with a learnable weight matrix and a soft thresholding function. The reconstructed sparse vector is restored to reconstructed business and financial data based on the overcomplete dictionary, and then double-verified with the original signal vector and the electronic voucher data with digital signature. The reconstructed business and financial data that passes the verification is automatically sent to the business and financial management system, and the reconstructed business and financial data that fails the verification is terminated.

2. The business data processing method for cross-physical isolation gateways according to claim 1, characterized in that, Before performing sparse decomposition on the original signal vector corresponding to the ticket image based on the trained overcomplete dictionary, the method further includes: Based on historical paper ticket originals, after scanning and extracting target field data, a historical signal vector is constructed and preprocessed; wherein, the preprocessing includes mean removal and normalization. Based on the initialized overcomplete dictionary, the preprocessed historical signal vector is sparsely decomposed according to the preset sparsity constraints to obtain a sparse vector. Based on singular value decomposition, the dictionary parameters of each column of the overcomplete dictionary are updated column by column, and after each update of the dictionary parameters, sparse decomposition is performed again to generate the sparse vector. Training stops when the reconstruction error corresponding to the sparse vector is less than a preset threshold or when the maximum number of iterations is reached, and the trained overcomplete dictionary is obtained.

3. The business data processing method for cross-physical isolation gateways according to claim 1, characterized in that, Based on a pre-trained overcomplete dictionary, the original signal vector corresponding to the ticket image to be recognized is sparsely decomposed to generate the original sparse vector, including: The image of the document to be identified is obtained by scanning the original paper document; Based on the invoice image, the region where the target field is located is located, and the target characters and target numbers in the region are extracted to obtain structured business and financial data; The structured business data is arranged according to a preset rule to generate a one-dimensional original signal vector; Based on the trained overcomplete dictionary, the orthogonal matching pursuit algorithm is used to perform sparse decomposition on the original signal vector to generate an original sparse vector that meets the preset sparsity constraints.

4. The business data processing method for cross-physical isolation gateways according to claim 1, characterized in that, The optical pulse signal obtained by sequentially compressing, encrypting, and modulating the original sparse vector is transmitted from the transmitting network channel through a physical isolation gateway to the receiving network channel, including: The original sparse vector is linearly projected onto the dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector. The compressed and encrypted observation vector is subjected to differential phase shift modulation processing to generate a modulated electrical signal; The modulated electrical signal is photoelectrically converted to generate an optical pulse phase difference signal; The optical pulse phase difference signal is transmitted from the transmitting network channel through a physical isolation gateway to the receiving network channel.

5. The business data processing method for cross-physical isolation gateways according to claim 4, characterized in that, The original sparse vector is linearly projected onto the dynamic observation matrix generated based on chaotic mapping to obtain a compressed and encrypted observation vector, including: Using timestamp hash values ​​as initial values, pseudo-random chaotic sequences are generated through chaotic mapping; The pseudo-random chaotic sequence is rearranged into an initial matrix, and then normalized and orthogonalized sequentially to obtain the dynamic observation matrix; Based on the dynamic observation matrix, a linear projection operation is performed on the original sparse vector to obtain a compressed and encrypted observation vector.

6. The business data processing method for cross-physical isolation gateways according to claim 1, characterized in that, Before inputting the observation vector obtained by converting the optical pulse signal into the deep unfolded network model of the receiving network channel, the method includes: Based on a fixed-layer deep unfolded network architecture, an initial deep unfolded network model is constructed by combining a learnable weight matrix and a soft thresholding function. Using historical observation vector samples as input and reconstructed sparse vectors as output, the deep unfolded network model is trained end-to-end offline. Training continues until the training loss between the reconstructed sparse vector and the corresponding original sparse vector sample is less than a preset convergence threshold, or until a preset number of training iterations is reached, at which point training stops and a well-trained deep unfolded network model is obtained.

7. The business data processing method for cross-physical isolation gateways according to claim 6, characterized in that, The hierarchical iterative update formula for the deep unfolded network model is as follows: ； in, For the deep unfolded network model, the first The reconstructed sparse vector output by the layer; For the first The first learnable weight matrix corresponding to the layer; This is the transpose of the dynamic observation matrix; To compress and encrypt the observation vector: For the first The second learnable weight matrix corresponding to the layer; For the deep unfolded network model, the first The reconstructed sparse vector output by the layer; For the first The third learnable weight matrix corresponding to the layer; For the first The soft threshold activation function corresponding to the layer.

8. The business data processing method for cross-physical isolation gateways according to claim 1, characterized in that, The reconstructed sparse vector is restored to reconstructed business and financial data based on the overcomplete dictionary, and then double-verified with the original signal vector and the digitally signed electronic voucher data. The reconstructed business and financial data that passes verification is automatically sent to the business and financial management system; the reconstructed business and financial data that fails verification is terminated. This includes: Based on the overcomplete dictionary, the reconstructed sparse vector is restored to the reconstructed business data; The reconstructed financial data is compared with the original signal vector to perform numerical consistency verification. The qualified data that passes the numerical consistency check will be compared with the electronic certificate data with digital signature to perform a compliance check. When both the numerical consistency check and the compliance check pass, the data is deemed to be qualified, and the reconstructed business and financial data that has passed the check is automatically sent to the business and financial management system. If the numerical consistency check fails or the compliance check fails, the data is deemed unqualified, and the reconstructed financial data that fails the check will be terminated from circulation.

9. The business data processing method for cross-physical isolation gateways according to claim 8, characterized in that, The reconstructed financial data is compared with the original signal vector to perform numerical consistency verification, including: The reconstructed financial data is compared with the original signal vector, and the deviation between the reconstructed financial data and the original signal vector is calculated. The deviation is compared with a preset deviation threshold. If the deviation is less than or equal to the preset deviation threshold, the numerical consistency check passes; otherwise, it fails.

10. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the business data processing method across a physical isolation gateway according to any one of claims 1 to 9.

Images