Storage method and system of intelligent storage cabinet

Abstract

The invention discloses a storage method and system for an intelligent storage cabinet, and the method comprises the steps: receiving authority parameters of an online examination and approval request and user biological characteristics, and generating an encrypted digital token; based on the encrypted digital token, performing identity authentication by adopting a multi-mode biological feature fusion algorithm, associating face features with voiceprint features through a space-time attention mechanism, and outputting a dynamic identity verification code; activating a target storage cabinet group according to the dynamic identity verification code, and generating an article unique identifier set; based on difference analysis of the unique article identifier set and a preset inventory list, outputting an optimal cabinet grid allocation scheme; and driving a modular cabinet group execution mechanism to adjust a physical structure according to the optimal cabinet grid distribution scheme, synchronously encrypting and writing article state data into a distributed account book, and triggering an edge computing node to update a global inventory topological graph. According to the embodiment of the invention, intelligent, self-adaptive and distributed collaborative management of the storage cabinet can be realized on the premise of ensuring the safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent storage, and in particular to a storage method and system for an intelligent safe deposit box. Background Art

[0002] Traditional safe deposit boxes typically use mechanical locks or single biometric technology for identity verification, resulting in crude access control, low item recognition accuracy, and insufficient space utilization. Existing smart safe deposit boxes often rely on fixed RFID or QR code recognition solutions, which struggle to cope with signal interference in complex electromagnetic environments and lack the ability to dynamically optimize the internal space layout. Regarding access control, traditional approval processes typically utilize centralized database storage, which carries the risk of data tampering and cannot meet the requirements of high-security scenarios. Summary of the Invention

[0003] The purpose of the present invention is to provide a storage method and system for an intelligent safe deposit box to address the deficiencies in the prior art and to achieve intelligent, adaptive and distributed collaborative management of the safe deposit box while ensuring security.

[0004] One embodiment of the present application provides a storage method for a smart safe deposit box, the method comprising: Receive the permission parameters and user biometrics of the online approval request, verify the compliance of the approval process through a blockchain consensus algorithm, and generate an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchy topology relationship and timestamp constraints; Based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and facial features and voiceprint features are associated through a spatiotemporal attention mechanism to output a dynamic identity verification code; Activate the target safe deposit box group based on the dynamic authentication code, identify items in the safe deposit box using an anti-interference RFID signal optimization model, and use a spatiotemporal convolutional network to separate the item RFID signal from electromagnetic noise to generate a set of unique item identifiers; Based on the difference analysis between the unique identifier set of the items and the preset inventory list, a reinforcement learning algorithm is used to dynamically optimize the spatial layout of the cabinet group, and a three-dimensional spatial value function is constructed according to the access frequency and volume characteristics of the items to output the optimal cabinet allocation plan; According to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map.

[0005] Optionally, the receiving of the permission parameters and user biometrics of the online approval request, verifying the compliance of the approval process through a blockchain consensus algorithm, and generating an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchical topology relationship and the timestamp constraint, includes: Based on the permission parameters of the online approval request, the node permission thresholds in the approval hierarchy topology are parsed to generate a directed graph of the approval process with a timestamp. The user's biometric hash value is temporally and spatially associated with the approval process directed graph, and the Byzantine fault-tolerant algorithm is used to verify the legitimacy of the signature timing of approval nodes at each level. Based on the verification results, an approval chain hash tree is constructed, and conflicting approval branches are pruned using a timestamp-weighted voting mechanism to generate a compliant approval path hash chain. By integrating the compliance approval path hash chain with the user's biometric entropy value, a dynamic digital token with permission timeliness is generated through the elliptic curve encryption algorithm.

[0006] Optionally, the method of performing identity authentication based on the encrypted digital token using a multimodal biometric fusion algorithm, associating facial features with voiceprint features through a spatiotemporal attention mechanism, and outputting a dynamic identity verification code includes: Extract the pre-stored facial image sequence and voiceprint time-frequency spectrum based on the encrypted digital token, extract the face's spatiotemporal motion features through a three-dimensional convolutional network, and generate a dynamic facial feature tensor; Perform Mel-frequency cepstral coefficient transformation on the voiceprint time spectrum, combine it with the gated recurrent unit to extract the voiceprint temporal features, and output the voiceprint feature vector sequence; Construct a spatiotemporal attention alignment matrix, calculate the cross-modal correlation between the dynamic face feature tensor and the voiceprint feature vector sequence, and generate a feature alignment weight map; Based on the feature alignment weight map, an adaptive gated fusion algorithm is used to dynamically weight the face and voiceprint features, and output a fused biometric feature vector with spatiotemporal consistency. The fused biometric feature vector is input into the two-stream Transformer network, and a one-time dynamic authentication code is generated through a dynamic hash function bound to a timestamp.

[0007] Optionally, activating a target safe deposit box group according to the dynamic identity verification code, identifying items in the safe deposit boxes using an anti-interference RFID signal optimization model, and using a spatiotemporal convolutional network to separate item RFID signals from electromagnetic noise to generate a set of unique item identifiers may include: Parse the spatial permission tag in the dynamic authentication code, activate the radio frequency antenna array of the target safe deposit box group, and transmit multi-band RFID scanning signals; Collect the time-frequency characteristics of the reflected signal in the cabinet, separate the item tag signal from the environmental electromagnetic noise through the spatiotemporal convolutional network, and construct the denoised signal tensor; Adaptive threshold decoding algorithm is used to parse the EPC code in the signal tensor and perform phase difference separation and reconstruction on the overlapping label signals; The EPC code is matched with the pre-stored item fingerprint library using Hamming distance to generate a set of unique item identifiers with confidence scores.

[0008] Optionally, the method of dynamically optimizing the cabinet group spatial layout based on the difference analysis between the set of unique item identifiers and the preset inventory list using a reinforcement learning algorithm, constructing a three-dimensional spatial value function based on the access frequency and volume characteristics of items, and outputting an optimal cabinet allocation plan includes: Based on the differences between the item unique identifier set and the inventory list, the volume, weight, and historical access frequency of the newly added items are extracted to construct an item feature matrix; Define a three-dimensional spatial value function, divide the cabinet into voxel units, and calculate the access efficiency weight, space utilization rate, and item association weight of each voxel; Build a deep reinforcement learning environment, input the item feature matrix and the three-dimensional spatial value function into the policy network, and explore the spatial layout action sequence of the cabinet group through the proximal policy optimization algorithm; Design a course learning mechanism, first train the optimal positioning strategy for a single item, gradually increase the number of items to the actual scene complexity, and generate a set of candidate layout solutions; Based on Pareto front analysis, candidate layout plans are screened, and the comprehensive score is calculated using the entropy weight-TOPSIS algorithm to output the optimal cabinet allocation layout plan that balances space utilization and access efficiency.

[0009] Optionally, the step of driving the modular cabinet group actuator to adjust the physical structure according to the optimal cabinet grid allocation plan, synchronously encrypting and writing the item status data into the distributed ledger, and triggering the edge computing node to update the global inventory topology map includes: The optimal cabinet allocation plan is converted into a sequence of robot arm motion control instructions, and a collision-free path trajectory is planned using an inverse kinematics algorithm. During the driving process, the pressure sensor data is collected in real time and the motor torque is dynamically adjusted using the PID closed-loop control algorithm. Synchronize the item displacement data with the digital twin model and generate a state change proof chain through a zero-knowledge proof algorithm; Trigger the Byzantine broadcast protocol between edge computing nodes, write the encrypted change data into the distributed ledger, and update the spatiotemporal state of the global inventory topology map.

[0010] Another embodiment of the present application provides a storage system for an intelligent safe deposit box, the system comprising: A receiving module, configured to receive permission parameters and user biometrics of an online approval request, verify the compliance of the approval process using a blockchain consensus algorithm that integrates the approval hierarchy topology with timestamp constraints, and generate an encrypted digital token; An authentication module, configured to perform identity authentication based on the encrypted digital token using a multimodal biometric fusion algorithm, associate facial features with voiceprint features through a spatiotemporal attention mechanism, and output a dynamic identity verification code; An identification module is configured to activate a target safe deposit box group based on the dynamic authentication code, identify items within the safe deposit boxes using an anti-interference RFID signal optimization model, and utilize a spatiotemporal convolutional network to separate item RFID signals from electromagnetic noise to generate a set of unique item identifiers. An optimization module is used to dynamically optimize the spatial layout of cabinet groups based on a discrepancy analysis between the unique identifiers of the items and the preset inventory list, using a reinforcement learning algorithm, constructing a three-dimensional spatial value function based on the access frequency and volume characteristics of the items, and outputting an optimal cabinet allocation plan; The execution module is used to drive the modular cabinet group actuator to adjust the physical structure according to the optimal cabinet allocation plan, synchronously encrypt and write the item status data into the distributed ledger, and trigger the edge computing node to update the global inventory topology map.

[0011] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when run.

[0012] Yet another embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the above methods.

[0013] Compared with the existing technology, the present invention provides a storage method for a smart safe deposit box, which receives the permission parameters and user biometrics of an online approval request and generates an encrypted digital token; based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and facial features and voiceprint features are associated through a spatiotemporal attention mechanism to output a dynamic identity verification code; according to the dynamic identity verification code, a target safe deposit box group is activated to generate a set of unique item identifiers; based on the difference analysis between the set of unique item identifiers and a preset inventory list, an optimal cabinet allocation plan is output; according to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map, thereby realizing intelligent, adaptive and distributed collaborative management of safe deposit boxes under the premise of ensuring security. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A hardware structure block diagram of a computer terminal for a storage method of an intelligent safe deposit box provided by an embodiment of the present invention; Figure 2A schematic diagram of a flow chart of a storage method for a smart safe provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a storage system for an intelligent safe deposit box provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0015] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limiting the present invention.

[0016] The embodiment of the present invention first provides a storage method for an intelligent safe deposit box. The method can be applied to electronic devices such as computer terminals, specifically ordinary computers.

[0017] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of the computer terminal of the storage method of the intelligent safe provided by the embodiment of the present invention. Figure 1 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, which, when executed, can cause the processor to execute any one of the storage methods of the smart safe.

[0019] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0020] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any storage method of the smart safe.

[0021] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0022] It should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0023] See also Figure 2 The embodiment of the present invention provides a storage method of a smart safe deposit box, which may include the following steps: S201, receiving permission parameters and user biometrics of an online approval request, verifying compliance with the approval process through a blockchain consensus algorithm, and generating an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchical topology relationship and timestamp constraints; Specifically, the permission parameters of the online approval request can be used to parse the node permission thresholds in the approval hierarchy topology and generate a directed graph of the approval process with a timestamp. The permission parameters for online approval requests typically include the approval type (e.g., financial approval, material procurement), the approval hierarchy (e.g., department manager → director → CEO), the IDs of approvers at each level, and the permission threshold (e.g., maximum amount). During the parsing process, the permission parameters are first encoded into a structured data model, for example, using JSON format to define the attributes of each approval node: { "node_id": "A001", "role": "Department Manager", "threshold": 50000, / / Unit: Yuan "next_nodes": ["B002", "B003"] }.

[0024] The approval hierarchy topology is modeled using a directed acyclic graph (DAG), where nodes represent approvers and edges represent the approval flow. For example, a purchase approval must proceed sequentially from applicant to department manager to finance director to CEO, forming a chain topology. Timestamp constraints are implemented by recording the operation time at each approval node, accurate to the millisecond (Unix timestamp, such as 1630453200000), ensuring a strict sequential approval process.

[0025] The process of generating a directed graph of approval process with timestamps: Node initialization: Create the initial node according to the permission parameters and set the node status to "unprocessed"; Edge connection: Create directed edges between nodes based on the next_nodes field. For example, the edge of node A001 points to B002 and B003. Timestamp embedding: A timestamp field is added to each node to record the approval arrival time (e.g. A001.receive_time=1630453200123); Loop detection: Use depth-first search (DFS) to detect whether there is a circular dependency in the graph. If so, an abnormal interrupt is triggered.

[0026] For example, the permission parameters for a company's purchase approval request define that purchases exceeding 500,000 RMB require CEO approval. The resulting directed graph contains four nodes, with edge weights indicating the approval jump conditions (e.g., if the amount is ≥ 500,000 RMB, the request is directed to the CEO node). The timestamp ensures that if a department manager submits a request at 10:00:00, the CFO must process it after 10:00:00 to prevent time inversion.

[0027] The user's biometric hash value is temporally and spatially associated with the approval process directed graph, and the Byzantine fault-tolerant algorithm is used to verify the legitimacy of the signature timing of approval nodes at each level. User biometrics (such as fingerprints and irises) are converted into fixed-length hash values ​​(such as 0x3a7d...f2c1) using a hash function (such as SHA-3-256). Spatiotemporal correlation coding binds the hash value to the spatiotemporal attributes of the approval process (timestamps and node locations): Time dimension: Concatenate the hash value with the timestamp of the approval node, for example, hash = SHA3(user_bio_hash|| node_A.timestamp); Spatial dimension: Introduce the geographic location of the approval node (such as GPS coordinates). If the node is approved remotely, a location hash value must be attached.

[0028] The Byzantine Fault Tolerance (BFT) algorithm uses the PBFT (Practical Byzantine Fault Tolerance) protocol to ensure that consensus can be reached even when there are at most f malicious nodes (total number of nodes is 3f+1). The verification process is as follows: Proposal phase: The master node (e.g., the first approver) broadcasts the approval request, which includes the user's biometric hash, the approval process directed graph, and a timestamp; Pre-preparation phase: Each node verifies the legitimacy of the proposal, checks the validity of the signature (such as ECDSA signature) and the timing compliance (timestamp increment); Preparation and submission phase: Nodes exchange signed messages and reach consensus after collecting 2f+1 valid responses.

[0029] Signature timing legitimacy verification: Timestamp increment: The processing time of each approval node must be later than that of the predecessor node, for example, node_CEO.timestamp > node_CFO.timestamp; Signature chain integrity: Using chain signatures, each approver uses the private key to re-sign the signature of the predecessor node to form an unalterable signature chain.

[0030] For example: During an approval process, it was detected that the timestamp 1630453200500 of the CFO's node was earlier than the timestamp 1630453200123 of the department manager's node. The BFT consensus determined that the approval sequence was illegal, triggering the process to terminate.

[0031] Based on the verification results, an approval chain hash tree is constructed, and conflicting approval branches are pruned using a timestamp-weighted voting mechanism to generate a compliant approval path hash chain. The approval chain hash tree adopts the Merkle tree structure. The leaf nodes are the hash values ​​of each approval node (including biometric hash, timestamp, and signature), and the non-leaf nodes are the concatenated hashes of the child node hashes.

[0032] When there are conflicting branches (such as two department managers approving at the same time), pruning is required: Conflict detection: traverse the hash tree and identify branches with multiple child nodes at the same level; Timestamp weighted voting: assign a weight to each branch, where weight = the normalized value of the inverse of the timestamp (e.g., branch A has timestamp t1 and weight 1 / t1). The branch with the highest total weight is retained. Pruning operation: remove low-weight branches and update the Merkle tree structure.

[0033] Compliance approval path hash chain generation: Path backtracking: trace back from the leaf node along the Merkle tree path to the root node, and collect all hash values ​​on the path; Chain splicing: splice the path hashes in timestamp order to form a hash chain (such as H1→H2→RootHash); Digital Signature: The root hash is signed by the system master private key, ensuring the immutability of the chain.

[0034] For example, a network delay creates two conflicting branches during an approval process (department managers A and B approve simultaneously), with timestamps 1630453200123 and 1630453200150, respectively. After weighted voting, the earlier branch (A) receives a higher weight, and its path is retained, while branch B is pruned.

[0035] By integrating the compliance approval path hash chain with the user's biometric entropy value, a dynamic digital token with permission timeliness is generated through the elliptic curve encryption algorithm.

[0036] The entropy of a user's biometrics is calculated using Shannon entropy to quantify the randomness of the biometric information. For example, the entropy calculation steps for a fingerprint image are as follows: Image segmentation: Divide the fingerprint image into 8×8 pixel blocks; Grayscale histogram statistics: calculate the pixel value distribution of each block; Entropy calculation: Use Shannon's formula H = -Σ(p_i * log2(p_i)) for each block and take the average entropy value of all blocks.

[0037] Elliptic Curve Cryptography (ECC) uses the secp256k1 curve (the same as Bitcoin) to generate dynamic digital tokens: Key generation: The system generates a public-private key pair (private key d, public key Q = d * G, where G is the base point of the curve); Token content encoding: concatenate the compliance approval path hash chain (Hex string) with the user's biometric entropy value (floating point number), for example, 0x3a7d...f2c1||0.87; Encryption and Signature: Encrypt the token contents using the recipient's public key (ECIES algorithm); Sign the encrypted content with the system private key (ECDSA algorithm).

[0038] The validity of permissions is achieved through the time window field in the token: Effective time (not_before): token generation time; Expiration time (not_after): Set according to the approval type (for example, the purchase token is valid for 24 hours).

[0039] Dynamic digital token example: { "token_id": "TK202310001", "content": "0x3a7d...f2c1||0.87", "signature": "0x5b9e...d4a3", "not_before": 1630453200000, "not_after": 1630539600000 }.

[0040] Verification process: Decrypt the token content: decrypt with the recipient's private key to obtain the hash chain and entropy value; Signature verification: Use the system public key to verify the legitimacy of the signature; Timeliness check: the current time must be between not_before and not_after; Entropy value comparison: Collect user biometrics in real time to calculate the entropy value, and the error with the entropy value in the token must be less than 5%.

[0041] S202, based on the encrypted digital token, using a multimodal biometric fusion algorithm to perform identity authentication, correlating facial features and voiceprint features through a spatiotemporal attention mechanism, and outputting a dynamic identity verification code; Specifically, the pre-stored facial image sequence and voiceprint time-frequency spectrum can be extracted according to the encrypted digital token, and the spatiotemporal motion features of the face can be extracted through a three-dimensional convolutional network to generate a dynamic facial feature tensor; During the authentication phase, the system first decrypts the user's pre-stored biometric data from an encrypted digital token. The facial image sequence consists of five consecutive frames of infrared-visible dual-mode images (1280×720 resolution, 30fps), capturing subtle head movements (such as blinking and nodding). The voiceprint spectrum is generated from 1.5-second speech segments captured by the microphone array at a 16kHz sampling rate, pre-emphasized (pre-emphasis factor 0.97), and framed (25ms frame length, 10ms frame shift).

[0042] The 3D Convolutional Network (3D-CNN) is an improvement on the ResNet-34 architecture and is designed specifically for spatiotemporal feature extraction: Input layer: 5 frames of images are stacked into a tensor of 1280×720×5; Convolution kernel configuration: The first layer uses a 7×7×3 spatiotemporal convolution kernel (3 represents the time dimension) with a step size of 2×2×1 to extract macroscopic motion patterns (such as head rotation trajectories). Residual module: Contains 4 residual blocks, each containing 3 3×3×3 convolutional layers. The number of channels is gradually increased from 64 to 512, and skip connections are used to alleviate gradient vanishing. Activation function: LeakyReLU (negative slope 0.1) is used to prevent neuron death; Pooling layer: Spatiotemporal maximum pooling (2×2×1) compresses the spatial dimension and preserves temporal continuity.

[0043] During training, transfer learning was performed using the CASIA-3D face dataset (containing 100,000 dynamic sequences), with an initial learning rate of 0.001, a batch size of 32, and an Adam optimizer (β1=0.9, β2=0.999). The network output was a 512-dimensional dynamic facial feature tensor, capturing features such as micro-expression muscle movement frequency (e.g., a 0.2Hz blink cycle) and illumination invariance.

[0044] Perform Mel-frequency cepstral coefficient transformation on the voiceprint time spectrum, combine it with the gated recurrent unit to extract the voiceprint temporal features, and output the voiceprint feature vector sequence; Voiceprint processing uses an architecture that combines MFCC (Mel-Frequency Cepstral Coefficients) and GRU (Gated Recurrent Unit): MFCC extraction: Preprocessing: The speech signal is windowed with a Hanning window, framed, and short-time energy is calculated. Silence frames below -40dB are filtered out. Mel filter bank: 40 triangular filters covering the 0-8kHz range, logarithmic energy compression and discrete cosine transform (DCT) to obtain 13-dimensional MFCC coefficients (including first-order and second-order differences, a total of 39 dimensions); Example parameters: The MFCC spectrum of a user pronouncing "open" shows that coefficients 3 and 7 fluctuate significantly in the 200-400ms range, reflecting the vibration pattern of the vocal cords.

[0045] GRU timing modeling: Network structure: 3-layer bidirectional GRU, 128 hidden units per layer, dropout rate 0.2 to prevent overfitting; Input sequence: Input the 39-dimensional MFCC features in time steps (150 frames), where each time step corresponds to 10ms of speech; Gating mechanism: The update gate controls the proportion of historical information retained, while the reset gate determines the degree to which the current input depends on the historical state. For example, when the plosive sound / p / appears, the reset gate activation value decreases, reducing the interference of the preceding vowel on the current feature.

[0046] The training data used was the VoxCeleb2 dataset (containing 6,000 voices), with a cross-entropy loss and a learning rate of 0.0001. The network output was a 256-dimensional voiceprint feature vector sequence with a temporal resolution of 10ms, capturing individual characteristics such as formant migration trajectories (e.g., the change in F1 and F2 from / i / to / a / ).

[0047] Construct a spatiotemporal attention alignment matrix, calculate the cross-modal correlation between the dynamic face feature tensor and the voiceprint feature vector sequence, and generate a feature alignment weight map; The spatiotemporal attention mechanism achieves cross-modal alignment through multi-head attention: Feature dimension alignment: The facial feature tensor (512 dimensions) is projected to 256 dimensions through a fully connected layer; The voiceprint feature sequence (256 dimensions) maintains its original dimension; Time axis alignment: Downsample facial features to 150 time steps (matching the length of the voiceprint sequence).

[0048] Attention calculation: Query: face feature vector; Key and Value: Voiceprint feature vector; Scaled dot product attention: calculates the similarity score at each time step, with a scaling factor of √256≈16; Multi-head mechanism: 8 attention heads compute in parallel, each with 32 dimensions, capturing correlation patterns at different granularities. For example, head 1 focuses on the synchronization between lip movements and speech fundamental frequency, while head 2 detects the correlation between blink rhythm and sentence pauses.

[0049] Weight graph generation: Softmax normalization: convert similarity scores into probability distributions; Time axis smoothing: Gaussian filtering (σ=3) is used to eliminate instantaneous noise interference; For example, when a user says "apple", the tip of the tongue shows a slight protrusion in the facial feature between 300 and 400 ms, and the corresponding high-frequency energy in the voiceprint feature increases sharply, with an attention weight of 0.92.

[0050] The dimension of the final output feature alignment weight map is 150×150 (time step × time step). The weight of the diagonal area is higher than that of the non-diagonal area, reflecting the temporal synchronization of the audiovisual signals.

[0051] Based on the feature alignment weight map, an adaptive gated fusion algorithm is used to dynamically weight the face and voiceprint features, and output a fused biometric feature vector with spatiotemporal consistency. Adaptive gated fusion uses a dynamic weight allocation strategy: Gating signal generation: The attention weight map is compressed into a 150-dimensional vector through global average pooling; Use a two-layer fully connected network (128→64→1) to generate the fusion weights for each time step, with a Sigmoid activation function; Example: When the voiceprint signal-to-noise ratio is less than 15dB, the gating weight favors facial features (weight 0.7); in a quiet environment, the gating weight favors voiceprints (weight 0.6).

[0052] Feature fusion: Weighted by time step: fusion vector = weight × facial feature + (1-weight) × voiceprint feature; Residual connection: add original facial features to prevent information loss; Layer normalization: stabilizes the training process and accelerates convergence.

[0053] Spatiotemporal consistency constraints: Motion continuity loss: Calculate the cosine similarity of feature vectors of adjacent time steps to enforce smooth transitions; Cross-modal consistency loss: compares the KL divergence of face and voiceprint features in the same time window, with the threshold set to 0.1.

[0054] The final output fusion biometric feature vector is 256-dimensional and has a time step of 150. It can represent cross-modal correlation features such as "the nonlinear relationship between the amplitude of the mouth corner raising and the increase in fundamental frequency during pronunciation."

[0055] The fused biometric feature vector is input into the two-stream Transformer network, and a one-time dynamic authentication code is generated through a dynamic hash function bound to a timestamp.

[0056] The two-stream Transformer network consists of a temporal stream and a spatial stream: Time Flow: To handle the temporal dependencies of fused vectors, a 12-layer Transformer encoder is used; Positional encoding uses a learnable sinusoidal embedding with dimension 256; The number of self-attention heads is 8, and the feedforward network dimension is 1024; Capture long-range dependencies such as synchronization between speech word boundaries and turning points of facial expressions.

[0057] Spatial Streams: Fold the time dimension and focus on key biological features through spatial attention; Use a 4-layer Transformer with 4 attention heads; Extract statistical features across time steps (such as voiceprint MFCC mean and face motion variance).

[0058] Dynamic hash function: Input: 512-dimensional joint feature vector of dual-stream output; Timestamp binding: Convert the current UTC time (millisecond level) into a 32-bit binary code and concatenate it with the feature vector; Hash calculation: Use the SHA-3-256 algorithm and intercept the first 128 bits as a one-time verification code; Example: The verification code generated at 2023-10-01 14:05:30.123 is 0x3a7d...f2c1, and is valid for 30 seconds.

[0059] Safety mechanism: Anti-replay attack: The verification code has a strict validity period and will automatically become invalid after expiration; Quantum-resistant: Lattice-based encryption algorithm (Kyber-512) is used to protect the hashing process; Real-time blacklist: Detecting three consecutive authentication failures triggers a security lock.

[0060] S203: Activate the target safe deposit box group according to the dynamic identity verification code, identify items in the safe deposit box using an anti-interference RFID signal optimization model, and use a spatiotemporal convolutional network to separate the item RFID signals from electromagnetic noise to generate a set of unique item identifiers. Specifically, the spatial permission tag in the dynamic authentication code can be parsed to activate the radio frequency antenna array of the target safe deposit box group and transmit a multi-band RFID scanning signal; The dynamic authentication code is a 256-bit encrypted string, with bits 32-64 representing the Space Access Tag (SAT), which identifies the range of safe deposit boxes that the user can access. During parsing, an asymmetric decryption algorithm (such as RSA-2048) is used to extract the SAT field. For example, the SAT value "0x3A7D" corresponds to the 12 safe deposit boxes in safe deposit box group B3 (numbered B3-01 to B3-12).

[0061] When activating the target cabinet group's RF antenna array, the system uses multi-band frequency hopping (FHSS) technology, switching frequencies 50 times per second within the UHF band (860MHz to 960MHz) to mitigate environmental electromagnetic interference. The antenna array consists of eight directional microstrip patch antennas, each with a 60-degree radiation angle and adjustable power from 10dBm to 30dBm. For example, if signal attenuation due to the cabinet's metal structure is detected, the system automatically increases transmit power from 20dBm to 28dBm and switches to a frequency dedicated to anti-metal tags (such as 920MHz).

[0062] The transmission mode of the multi-band RFID scanning signal is dynamically adjusted according to the cabinet group structure: Low frequency band (125kHz): used for close-range (<10cm) object recognition, with high accuracy but weak penetration; High frequency band (13.56MHz): supports ISO / IEC 14443 protocol for reading encrypted item tags; Ultra-high frequency (UHF): wide coverage (1-5 meters), suitable for batch scanning.

[0063] For example, when users need to access high-value items, the system will prioritize high-frequency scanning to ensure data security; while for ordinary items, the UHF band will be used to improve efficiency.

[0064] Collect the time-frequency characteristics of the reflected signal in the cabinet, separate the item tag signal from the environmental electromagnetic noise through the spatiotemporal convolutional network, and construct the denoised signal tensor; The cabinet reflection signal is collected using a software-defined radio (SDR) device (such as the USRP B210) with a sampling rate of 2 MHz and a resolution of 12 bits. The original signal contains: Valid tag signal: coded signal backscattered by the RFID tag (such as EPC code); Environmental noise: Broadband interference from Wi-Fi routers (2.4GHz), Bluetooth devices (2.4GHz), and industrial motors.

[0065] The structure design of the spatiotemporal convolutional network (ST-CNN) is as follows: Input layer: receives I / Q signals (1024 sampling points each) and forms a 2×1024 time-frequency matrix; Spatiotemporal convolution layer: uses a 3×3 convolution kernel to extract features simultaneously in the time dimension (sampling point sequence) and the spatial dimension (frequency channel), and the activation function is LeakyReLU (negative slope 0.01); Attention mechanism: Dynamically enhance the weight of effective signal frequency bands through the SENet (Squeeze-and-Excitation Network) module; Output layer: Generates a denoised signal tensor with a dimension of 1×1024 (time domain waveform) + 64×64 (frequency domain energy spectrum).

[0066] For example, in a certain scan, the signal-to-noise ratio (SNR) of the original signal was -5dB, which was increased to 15dB after ST-CNN processing, effectively separating three overlapping label signals.

[0067] Adaptive threshold decoding algorithm is used to parse the EPC code in the signal tensor and perform phase difference separation and reconstruction on the overlapping label signals; The core of the adaptive threshold decoding algorithm is to dynamically adjust the signal decision threshold: Baseline estimation: Perform a sliding average of the signal tensor's time domain waveform (window width 100 sampling points) and calculate the background noise level (e.g., -65dBm). Peak detection: Set a dynamic threshold (baseline + 6dB) based on the baseline to identify pulse signals exceeding the threshold; Encoding analysis: According to the EPC Gen2 protocol, the pulse width is converted into binary data (for example, a width of 2μs represents "0" and 4μs represents "1").

[0068] For overlapping tag signals (signal aliasing caused by multiple tags responding simultaneously), the following method is used to separate them: IQ quadrature demodulation: Extracts the signal phase information through Hilbert transform and distinguishes the reflection phase difference of different tags; Direction of Arrival (DOA) estimation: Calculates the tag's spatial position using the phase difference of the antenna array and combines it with the MUSIC (Multiple Signal Classification) algorithm to separate spatially overlapping signals. Time domain framing: The aliased signal is framed by windowing (Hamming window, frame length 256 samples), and the time-frequency fingerprint of each label is extracted through short-time Fourier transform (STFT).

[0069] For example, the EPC codes of two tags are "0x5A3F" and "0xB7E2" respectively, the phase difference of their reflected signals is 45 degrees, and the spatial position difference is 0.3 meters. After separation, the complete codes of the two tags can be accurately parsed.

[0070] The EPC code is matched with the pre-stored item fingerprint library using Hamming distance to generate a set of unique item identifiers with confidence scores.

[0071] The pre-stored item fingerprint library contains the following information for each item: EPC code: 96-bit unique identifier (e.g., “0x3014BEEF00000000000001”); Physical fingerprint: tag backscatter characteristics (such as resonant frequency offset ±50kHz), installation location (cabinet coordinates); Historical access records: last access time, operator ID, etc.

[0072] The Hamming distance matching process is as follows: Preliminary screening: Calculate the Hamming distance (different number of bits) between the input EPC code and all codes in the fingerprint library, and select candidate codes with a distance ≤ 4; Physical fingerprint verification: Compare the tag's resonant frequency characteristics, with an allowable deviation of ±10kHz; Confidence score: Comprehensive Hamming distance (weight 0.6), frequency matching (weight 0.3), spatial position consistency (weight 0.1), calculation formula:

[0073] Where HD is the Hamming distance, Δf is the frequency deviation, and Δd is the position deviation (meters).

[0074] For example, the matching result for an EPC code "0x3014BEEF..." is: Hamming distance 2 (matching degree 98%); Frequency deviation 8kHz (matching degree 92%); Position deviation 0.2 meters (matching degree 83%); The combined score = 0.6 × 0.98 + 0.3 × 0.92 + 0.1 × 0.83 = 0.953, and the confidence level is marked as "high".

[0075] The format of the final generated item unique identifier set is as follows: [ {EPC: "0x3014BEEF...", Score: 0.953, Position: "B3-05"}, {EPC: "0x5A3F...", Score: 0.872, Position: "B3-08"} ].

[0076] The collection is encapsulated in JSON format and transmitted to the upper-level system for inventory management and access control.

[0077] S204, based on the difference analysis between the set of unique item identifiers and the preset inventory list, a reinforcement learning algorithm is used to dynamically optimize the cabinet group spatial layout, a three-dimensional spatial value function is constructed based on the access frequency and volume characteristics of the items, and an optimal cabinet allocation plan is output; Specifically, based on the difference between the item unique identifier set and the inventory list, the volume, weight, and historical access frequency of the newly added items can be extracted to construct an item feature matrix; After comparing the unique identifiers of items (e.g., EPC codes identified by RFID) with the pre-set inventory list, the system first filters out newly added items (e.g., items not registered in inventory) or items with changed locations. For each newly added item, the following features are extracted: Physical characteristics: Volume: Measure the length, width, and height of an item using a 3D vision sensor (such as the Intel RealSense D415). For example, a package with dimensions of 30cm × 20cm × 15cm has a volume of 9,000 cm³. Weight: Use a pressure sensor (range 0~50kg, accuracy ±10g) to obtain weight data, such as an item weighs 2.3kg; Access behavior characteristics: Access frequency: counts the number of times the item has been accessed in the past 30 days. For example, if a tool is accessed 5 times a week, it is marked as a high-frequency item. Access time: Record the timestamp of each access and analyze whether there is a time preference (for example, certain medical supplies are mostly accessed between 9 and 11 a.m.).

[0078] Item feature matrix construction: The above features are encoded into a matrix form, where each row represents an item and each column corresponds to the feature dimension, as shown in Table 1.

[0079] Table 1

[0080] This matrix serves as the input for subsequent spatial optimization, and its dimension is N×4 (N is the number of items).

[0081] Define a three-dimensional spatial value function, divide the cabinet into voxel units, and calculate the access efficiency weight, space utilization rate, and item association weight of each voxel; The three-dimensional spatial value function guides the optimization of item layout by quantifying the comprehensive value of each cabinet voxel (the smallest spatial unit, such as 5cm×5cm×5cm). The specific definition is as follows: Voxel division: Assume that the internal space of the safe deposit box is 2m×1m×2m (length×width×height), divided into 40×20×40 voxels, and the volume of each voxel is 5cm³; The position coordinates of each voxel are encoded as (x, y, z), for example (10, 5, 30) means the 10th column, 5th row, and 30th layer.

[0082] Value dimension calculation: Access efficiency weight: Based on the distance between the voxel and the cabinet door (the closer the voxel is to the door, the higher the weight), the calculation formula is: Where d is the Euclidean distance from the voxel to the gate, D_max is the farthest voxel distance (e.g., the diagonal distance is 3m), then the voxel weight at 0.5m away from the gate is 0.83; Space utilization rate: calculated based on the volume ratio of items that can be accommodated in a voxel. For example, if a voxel has stored items with a volume of 4000 cm³ (voxel capacity is 5000 cm³), the utilization rate is 80%; Item association weight: Counts the frequency of coexistence of items of the same type in historical data. For example, tools often coexist with repair manuals, and the association weight is set to 0.7.

[0083] Comprehensive value function: Value = 0.5 × W access +0.3×space utilization+0.2×item association weight.

[0084] The value score of each voxel ranges from 0 to 1, and high-scoring voxels are preferentially assigned high-frequency or related items.

[0085] Build a deep reinforcement learning environment, input the item feature matrix and the three-dimensional spatial value function into the policy network, and explore the spatial layout action sequence of the cabinet group through the proximal policy optimization algorithm; The deep reinforcement learning environment is customized based on the OpenAI Gym framework and includes the following core components: State Space: Current cabinet space status (voxel occupancy matrix, 40×20×40); Item feature matrix (N×4); Three-dimensional value matrix (40×20×40×3, including access efficiency, utilization rate, and correlation).

[0086] Action Space: Discrete actions: select the item ID and target voxel coordinates (e.g., "Put EPC001 into (10,5,30)"); Continuous action: Adjust the object's posture (such as rotating it from 0 to 90 degrees) to fit the space.

[0087] Reward function: Positive reward: After successfully placing an item, the reward value is the comprehensive value of the target voxel (e.g. 0.8); Negative reward: When the volume of the item exceeds the remaining space of the voxel, the penalty is -0.5; Time penalty: If each step takes more than 1 second, the penalty is -0.1.

[0088] Policy network and PPO algorithm: Policy network: A 3-layer convolutional network (with 32-64-128 channels) is used to extract spatial features, followed by a fully connected layer to output action probability distribution. Proximal Policy Optimization (PPO): Set key parameters: learning rate: 0.00025; discount factor γ: 0.99; clipping threshold ε: 0.2, limit the policy update amplitude to prevent oscillation.

[0089] Training process: Exploration phase: The agent randomly tries to place actions and accumulates initial experience (about 10,000); Strategy optimization: Each iteration samples 512 data points from the experience pool, calculates the advantage function (GAE, λ = 0.95), and updates the network; Convergence condition: The average reward fluctuation for 10 consecutive rounds is less than 1%.

[0090] Design a course learning mechanism, first train the optimal positioning strategy for a single item, gradually increase the number of items to the actual scene complexity, and generate a set of candidate layout solutions; Curriculum Learning improves model performance by increasing difficulty in stages: Phase 1 (Single Item): Task: Place a single item into the optimal voxel; Example: EPC001 (volume 9000 cm³) should be placed in a voxel near the door with low utilization; Training goal: Make the model learn to match the volume of objects with the remaining space of voxels while maximizing access efficiency.

[0091] Phase 2 (Multiple Items without Conflict): Task: Place 5 items in sequence without overlapping spaces; Difficulty: The relevance between items needs to be considered (e.g. frequently used items should be stored together); Reward adjustment: Increase the relevance weight to 0.3.

[0092] Stage 3 (Full Complexity): Task: Process 50 items simultaneously, allowing dynamic adjustments (such as replacing old items); Introduce random events: simulate sudden access needs (such as an item needs to be taken out urgently) to test the robustness of the strategy.

[0093] Candidate generation: After each stage of training is completed, the 10 best performing strategies (such as single item positioning error <5cm, multi-item conflict rate <2%) are saved to form a candidate layout solution library. For example: Solution A: Place high-frequency small items near the cabinet door and large items at the back; Solution B: Partition by item type (tool area, document area, etc.).

[0094] Based on Pareto front analysis, candidate layout plans are screened, and the comprehensive score is calculated using the entropy weight-TOPSIS algorithm to output the optimal cabinet allocation layout plan that balances space utilization and access efficiency.

[0095] Pareto front analysis: Goal: Maximize space utilization, access efficiency, and relevance; Screening: Eliminate dominated solutions from candidate solutions (i.e., there is another solution that is better in all objectives); Example: Plan A has a utilization rate of 85% and an efficiency of 0.7; Plan B has a utilization rate of 80% and an efficiency of 0.75. If the two do not dominate each other, both are retained.

[0096] Entropy weight method determines weights: Calculate the information entropy of each target. The smaller the entropy value (the greater the data fluctuation), the higher the weight; For example: the space utilization entropy value is the lowest (fluctuates significantly), with a weight of 0.5; the access efficiency is 0.3; and the correlation is 0.2.

[0097] TOPSIS (top-up ranking method): Ideal solution: maximum value of each objective (e.g. utilization 90%, efficiency 0.8, correlation 0.9); Negative ideal solution: minimum value of each objective (e.g. utilization 70%, efficiency 0.5, correlation 0.4); Proximity calculation: ,in, is the distance between solution i and the ideal solution, is the distance to the negative ideal solution.

[0098] Sorting: The higher the closeness (closer to 1), the better the solution.

[0099] Final solution selection: Assuming that the closeness of Plan A is 0.72 and that of Plan B is 0.68, Plan A is selected as the optimal layout. This plan has a space utilization rate of 85% and an average access time that is 30% shorter than the baseline.

[0100] S205: According to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map.

[0101] Specifically, the optimal cabinet allocation plan can be converted into a sequence of robot arm motion control instructions, and the collision-free path trajectory can be planned through the inverse kinematics algorithm; The optimal cabinet allocation solution defines the target location of the items in three-dimensional coordinate form (for example, the coordinates of cabinet B-3 are x=1.2m, y=0.8m, z=0.5m). The generation of the robot arm motion control instructions requires solving the following problems: Inverse kinematics calculation: Convert the target coordinates into the angles of each joint of the robotic arm. For a six-degree-of-freedom robotic arm, numerical iteration methods (such as the Newton-Raphson algorithm) are used to solve for the joint angles θ1 through θ6. For example, the target position corresponds to θ1 = 30°, θ2 = 45°, θ3 = 60°, θ4 = 90°, θ5 = 15°, and θ6 = 0°.

[0102] Collision-free path planning: Environmental modeling: Construct a 3D point cloud map of the cabinet group (accuracy ±5mm) and mark obstacles (such as adjacent cabinets and stored items).

[0103] Path search: RRT* (rapid exploration random tree star algorithm) was used to generate the initial path, the node expansion step was set to 50 mm, and the collision detection frequency was 100 Hz.

[0104] Trajectory optimization: Use B-spline curves to smooth the path and constrain the joint angular acceleration to ≤50rad / s² to avoid mechanical vibration.

[0105] Control instruction sequence encoding: Instruction format: includes target coordinates, joint angles, movement speed (such as 200mm / s), and acceleration curve (S-shaped acceleration and deceleration).

[0106] Example directive: MoveTo(x=1.2, y=0.8, z=0.5), JointAngles(30,45,60,90,15,0), Velocity=200mm / s, AccelerationType=S-Curve.

[0107] After the robot arm controller (such as the EtherCAT bus driver) interprets the instructions, it executes the action through the servo motor (such as the Panasonic MINASA6 series), with a positioning accuracy of ±0.1mm.

[0108] During the driving process, the pressure sensor data is collected in real time and the motor torque is dynamically adjusted using the PID closed-loop control algorithm. A pressure sensor (such as TE Connectivity MS5837-30BA) is deployed at the contact surface between the robot arm gripper and the cabinet to monitor the gripping force in real time (range 0-50N, accuracy ±0.5N). PID control process: Set value: Set the target gripping force according to the weight of the item (e.g. 5N for a 500g item).

[0109] Feedback signal: The sensor uploads the real-time clamping force F_actual at a frequency of 100 Hz.

[0110] PID calculation: Proportional term (P): Kp = 2.0, fast response deviation (for example, when F_actual = 4N, the output adjustment amount Δ = 2 × (5-4) = 2N); Integral term (I): Ki = 0.5, eliminating steady-state error (when the cumulative error reaches 2N·s, Δ+ = 0.5 × 2 = 1N); Differential term (D): Kd=0.1, suppresses overshoot (if the clamping force change rate is -0.5N / s, Δ-=0.1×0.5=0.05N).

[0111] Torque adjustment: Convert the PID output Δ into motor current (e.g. 1N=0.2A) and drive the motor via PWM signal.

[0112] Adaptive Optimization: When object slippage is detected (via gripper displacement sensors such as grating scale feedback position change > 1mm), Kp is automatically increased to 3.0 and the integral term is reset to ensure stable gripping.

[0113] Synchronize the item displacement data with the digital twin model and generate a state change proof chain through a zero-knowledge proof algorithm; Digital Twin Synchronization: Data acquisition: The robot arm encoder (such as Heidenhain EQN1325) provides real-time feedback on joint angles. Combined with a visual positioning system (such as Intel RealSense D455), it captures the object position and generates a displacement data stream (format: timestamp|x|y|z|attitude angle).

[0114] Model update: Data is synchronized to a digital twin platform (such as Siemens MindSphere) via the OPC UA protocol. The 3D model is updated at a frequency of 30Hz, ensuring a virtual-to-real error of less than 2mm.

[0115] Zero-Knowledge Proof (ZKP): Proof Generation: Using the zk-SNARKs (Zero-Knowledge Succinct Non-Interactive Arguments of Knowledge) algorithm, a chain of proof is generated to prove the location change of an item. For example, it can prove the authenticity of "item A was moved from compartment B-3 to C-5" without revealing the details of the movement path.

[0116] Verification process: Trusted setup: pre-generate public parameters (CRS, Common Reference String); Proof generation: Input private data (displacement trajectory) and public constraints (such as coordinate range) to generate proof π; Verification: Other nodes use CRS to verify the validity of π (time < 10ms).

[0117] Example proof chain structure: Block #1234: - Item ID: A-01 - Original location hash: 0x3a7d... - New location hash: 0x5b9e... - ZKP Proof: 0x8c2f... - Previous block hash: 0x4d1a... Trigger the Byzantine broadcast protocol between edge computing nodes, write the encrypted change data into the distributed ledger, and update the spatiotemporal state of the global inventory topology map.

[0118] Byzantine Broadcast Protocol: Node role: Assume 5 edge nodes (N1~N5), and tolerate a maximum of f=2 malicious nodes.

[0119] Broadcasting process: Pre-preparation phase: Master node N1 sends<PRE-PREPARE, blk, v, n, sig> , where v = view number, n = sequence number, sig = digital signature; Preparation stage: each node verifies the signature and broadcasts<PREPARE, blk, v, n, sig> , after receiving 2f+1 valid messages, it enters the submission phase; Submission phase: Broadcast<COMMIT, blk, v, n, sig> , after receiving 2f+1 submissions, the block is confirmed to be valid.

[0120] Data encryption and writing: Encryption algorithm: AES-256-GCM mode is used to encrypt displacement data, and the key is dynamically distributed by KMS (Key Management System); Ledger structure: Each block contains: Block header: version number, timestamp, previous block hash, Merkle root; Transaction list: encrypted item displacement record; Consensus signature: joint signature of PBFT nodes.

[0121] Global inventory topology map updated: Spatiotemporal indexing: Encodes cabinet locations into 12-bit strings (such as wx4g0b) based on the GeoHash algorithm, supporting fast range queries. Incremental Update: Only the changed area (such as the status of cabinet C-5) is synchronized to reduce network load. For example, the status of cabinet C-5 in the original topology is "Empty", but after the update, it is "Occupied | Item A-01".

[0122] Example execution effect: Latency: From the completion of the robot arm movement to the update of the global topology map, the end-to-end latency is less than 500ms; Reliability: Byzantine fault tolerance ensures that consensus can be reached even when two nodes fail. Security: Zero-knowledge proof and AES-256 encryption ensure data privacy and integrity.

[0123] It can be seen that the permission parameters and user biometrics of the online approval request are received to generate an encrypted digital token; based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and the facial features and voiceprint features are associated through the spatiotemporal attention mechanism to output a dynamic identity verification code; according to the dynamic identity verification code, the target safe deposit box group is activated to generate a set of unique item identifiers; based on the difference analysis between the set of unique item identifiers and the preset inventory list, the optimal cabinet allocation plan is output; according to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map, so that the intelligent, adaptive and distributed collaborative management of the safe deposit box can be realized under the premise of ensuring security.

[0124] Another embodiment of the present invention provides a smart safe deposit box storage system, see Figure 3 , the system may include: Receiving module 301, configured to receive permission parameters and user biometrics of an online approval request, verify the compliance of the approval process using a blockchain consensus algorithm that integrates the approval hierarchy topology with timestamp constraints, and generate an encrypted digital token; Authentication module 302, configured to perform identity authentication based on the encrypted digital token using a multimodal biometric fusion algorithm, associate facial features with voiceprint features through a spatiotemporal attention mechanism, and output a dynamic identity verification code; Identification module 303, configured to activate a target safe deposit box group based on the dynamic authentication code, identify items within the safe deposit boxes using an anti-interference RFID signal optimization model, and utilize a spatiotemporal convolutional network to separate item RFID signals from electromagnetic noise to generate a set of unique item identifiers; Optimization module 304 is configured to dynamically optimize the cabinet group spatial layout based on the difference analysis between the set of unique item identifiers and the preset inventory list using a reinforcement learning algorithm, construct a three-dimensional spatial value function based on the access frequency and volume characteristics of the items, and output an optimal cabinet allocation plan; The execution module 305 is used to drive the modular cabinet group actuator to adjust the physical structure according to the optimal cabinet allocation plan, synchronously encrypt the item status data and write it into the distributed ledger, and trigger the edge computing node to update the global inventory topology map.

[0125] It can be seen that the permission parameters and user biometrics of the online approval request are received to generate an encrypted digital token; based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and the facial features and voiceprint features are associated through the spatiotemporal attention mechanism to output a dynamic identity verification code; according to the dynamic identity verification code, the target safe deposit box group is activated to generate a set of unique item identifiers; based on the difference analysis between the set of unique item identifiers and the preset inventory list, the optimal cabinet allocation plan is output; according to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map, so that the intelligent, adaptive and distributed collaborative management of the safe deposit box can be realized under the premise of ensuring security.

[0126] An embodiment of the present invention further provides a storage medium storing a computer program, wherein the computer program is configured to execute the steps of any one of the above method embodiments when running.

[0127] Specifically, in this embodiment, the above-mentioned storage medium may be configured to store a computer program for performing the following steps: S201, receiving permission parameters and user biometrics of an online approval request, verifying compliance with the approval process through a blockchain consensus algorithm, and generating an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchical topology relationship and timestamp constraints; S202, based on the encrypted digital token, using a multimodal biometric fusion algorithm to perform identity authentication, correlating facial features and voiceprint features through a spatiotemporal attention mechanism, and outputting a dynamic identity verification code; S203: Activate the target safe deposit box group according to the dynamic identity verification code, identify items in the safe deposit box using an anti-interference RFID signal optimization model, and use a spatiotemporal convolutional network to separate the item RFID signals from electromagnetic noise to generate a set of unique item identifiers. S204, based on the difference analysis between the set of unique item identifiers and the preset inventory list, a reinforcement learning algorithm is used to dynamically optimize the cabinet group spatial layout, a three-dimensional spatial value function is constructed based on the access frequency and volume characteristics of the items, and an optimal cabinet allocation plan is output; S205: According to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map.

[0128] It can be seen that the permission parameters and user biometrics of the online approval request are received to generate an encrypted digital token; based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and the facial features and voiceprint features are associated through the spatiotemporal attention mechanism to output a dynamic identity verification code; according to the dynamic identity verification code, the target safe deposit box group is activated to generate a set of unique item identifiers; based on the difference analysis between the set of unique item identifiers and the preset inventory list, the optimal cabinet allocation plan is output; according to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map, so that the intelligent, adaptive and distributed collaborative management of the safe deposit box can be realized under the premise of ensuring security.

[0129] An embodiment of the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the steps in any one of the above method embodiments.

[0130] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0131] Specifically, in this embodiment, the processor may be configured to execute the following steps through a computer program: S201, receiving permission parameters and user biometrics of an online approval request, verifying compliance with the approval process through a blockchain consensus algorithm, and generating an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchical topology relationship and timestamp constraints; S202, based on the encrypted digital token, using a multimodal biometric fusion algorithm to perform identity authentication, correlating facial features and voiceprint features through a spatiotemporal attention mechanism, and outputting a dynamic identity verification code; S203: Activate the target safe deposit box group according to the dynamic identity verification code, identify items in the safe deposit box using an anti-interference RFID signal optimization model, and use a spatiotemporal convolutional network to separate the item RFID signals from electromagnetic noise to generate a set of unique item identifiers. S204, based on the difference analysis between the set of unique item identifiers and the preset inventory list, a reinforcement learning algorithm is used to dynamically optimize the cabinet group spatial layout, a three-dimensional spatial value function is constructed based on the access frequency and volume characteristics of the items, and an optimal cabinet allocation plan is output; S205: According to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map.

[0132] It can be seen that the permission parameters and user biometrics of the online approval request are received to generate an encrypted digital token; based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and the facial features and voiceprint features are associated through the spatiotemporal attention mechanism to output a dynamic identity verification code; according to the dynamic identity verification code, the target safe deposit box group is activated to generate a set of unique item identifiers; based on the difference analysis between the set of unique item identifiers and the preset inventory list, the optimal cabinet allocation plan is output; according to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map, so that the intelligent, adaptive and distributed collaborative management of the safe deposit box can be realized under the premise of ensuring security.

[0133] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.

Claims

1. A storage method for an intelligent safe deposit box, characterized in that: The method comprises: Receive the permission parameters and user biometrics of the online approval request, verify the compliance of the approval process through a blockchain consensus algorithm, and generate an encrypted digital token, wherein the blockchain consensus algorithm integrates the approval hierarchy topology relationship and timestamp constraints; Based on the encrypted digital token, a multimodal biometric fusion algorithm is used for identity authentication, and facial features and voiceprint features are associated through a spatiotemporal attention mechanism to output a dynamic identity verification code; Activate the target safe deposit box group based on the dynamic authentication code, identify items in the safe deposit box using an anti-interference RFID signal optimization model, and use a spatiotemporal convolutional network to separate the item RFID signal from electromagnetic noise to generate a set of unique item identifiers; Based on the difference analysis between the unique identifier set of the items and the preset inventory list, a reinforcement learning algorithm is used to dynamically optimize the spatial layout of the cabinet group, and a three-dimensional spatial value function is constructed according to the access frequency and volume characteristics of the items to output the optimal cabinet allocation plan; According to the optimal cabinet allocation plan, the modular cabinet group actuator is driven to adjust the physical structure, and the item status data is simultaneously encrypted and written into the distributed ledger, triggering the edge computing node to update the global inventory topology map.

2. The method according to claim 1, characterized in that The authorization parameters and user biometrics of the online approval request are received, and the compliance of the approval process is verified through a blockchain consensus algorithm to generate an encrypted digital token. The blockchain consensus algorithm integrates the approval hierarchical topology relationship and the timestamp constraint, including: Based on the permission parameters of the online approval request, the node permission thresholds in the approval hierarchy topology are parsed to generate a directed graph of the approval process with a timestamp. The user's biometric hash value is temporally and spatially associated with the approval process directed graph, and the Byzantine fault-tolerant algorithm is used to verify the legitimacy of the signature timing of approval nodes at each level. Based on the verification results, an approval chain hash tree is constructed, and conflicting approval branches are pruned using a timestamp-weighted voting mechanism to generate a compliant approval path hash chain. By integrating the compliance approval path hash chain with the user's biometric entropy value, a dynamic digital token with permission timeliness is generated through the elliptic curve encryption algorithm.

3. The method according to claim 2, characterized in that The method uses a multimodal biometric fusion algorithm to perform identity authentication based on the encrypted digital token, associates facial features with voiceprint features through a spatiotemporal attention mechanism, and outputs a dynamic identity verification code, including: Extract the pre-stored facial image sequence and voiceprint time-frequency spectrum based on the encrypted digital token, extract the face's spatiotemporal motion features through a three-dimensional convolutional network, and generate a dynamic facial feature tensor; Perform Mel-frequency cepstral coefficient transformation on the voiceprint time spectrum, combine it with the gated recurrent unit to extract the voiceprint temporal features, and output the voiceprint feature vector sequence; Construct a spatiotemporal attention alignment matrix, calculate the cross-modal correlation between the dynamic face feature tensor and the voiceprint feature vector sequence, and generate a feature alignment weight map; Based on the feature alignment weight map, an adaptive gated fusion algorithm is used to dynamically weight the face and voiceprint features, and output a fused biometric feature vector with spatiotemporal consistency. The fused biometric feature vector is input into the two-stream Transformer network, and a one-time dynamic authentication code is generated through a dynamic hash function bound to a timestamp.

4. The method according to claim 3, characterized in that The method of activating a target safe deposit box group according to the dynamic identity verification code, identifying items in the safe deposit box using an anti-interference RFID signal optimization model, and using a spatiotemporal convolutional network to separate item RFID signals from electromagnetic noise to generate a set of unique item identifiers includes: Parse the spatial permission tag in the dynamic authentication code, activate the radio frequency antenna array of the target safe deposit box group, and transmit multi-band RFID scanning signals; Collect the time-frequency characteristics of the reflected signal in the cabinet, separate the item tag signal from the environmental electromagnetic noise through the spatiotemporal convolutional network, and construct the denoised signal tensor; Adaptive threshold decoding algorithm is used to parse the EPC code in the signal tensor and perform phase difference separation and reconstruction on the overlapping label signals; The EPC code is matched with the pre-stored item fingerprint library using Hamming distance to generate a set of unique item identifiers with confidence scores.

5. The method according to claim 4, characterized in that The method uses a reinforcement learning algorithm to dynamically optimize the cabinet group space layout based on the difference analysis between the unique identification set of items and the preset inventory list, constructs a three-dimensional space value function based on the access frequency and volume characteristics of items, and outputs the optimal cabinet allocation plan, including: Based on the differences between the item unique identifier set and the inventory list, the volume, weight, and historical access frequency of the newly added items are extracted to construct an item feature matrix; Define a three-dimensional spatial value function, divide the cabinet into voxel units, and calculate the access efficiency weight, space utilization rate, and item association weight of each voxel; Build a deep reinforcement learning environment, input the item feature matrix and the three-dimensional spatial value function into the policy network, and explore the spatial layout action sequence of the cabinet group through the proximal policy optimization algorithm; Design a course learning mechanism, first train the optimal positioning strategy for a single item, gradually increase the number of items to the actual scene complexity, and generate a set of candidate layout solutions; Based on Pareto front analysis, candidate layout plans are screened, and the comprehensive score is calculated using the entropy weight-TOPSIS algorithm to output the optimal cabinet allocation layout plan that balances space utilization and access efficiency.

6. The method according to claim 5, characterized in that The method drives the modular cabinet group actuator to adjust the physical structure according to the optimal cabinet allocation plan, simultaneously encrypts and writes the item status data into the distributed ledger, and triggers the edge computing node to update the global inventory topology map, including: The optimal cabinet allocation plan is converted into a sequence of robot arm motion control instructions, and a collision-free path trajectory is planned using an inverse kinematics algorithm. During the driving process, the pressure sensor data is collected in real time and the motor torque is dynamically adjusted using the PID closed-loop control algorithm. Synchronize the item displacement data with the digital twin model and generate a state change proof chain through a zero-knowledge proof algorithm; Trigger the Byzantine broadcast protocol between edge computing nodes, write the encrypted change data into the distributed ledger, and update the spatiotemporal state of the global inventory topology map.

7. A storage system for an intelligent safe deposit box, characterized in that: The system comprises: A receiving module, configured to receive permission parameters and user biometrics of an online approval request, verify the compliance of the approval process using a blockchain consensus algorithm that integrates the approval hierarchy topology with timestamp constraints, and generate an encrypted digital token; An authentication module, configured to perform identity authentication based on the encrypted digital token using a multimodal biometric fusion algorithm, associate facial features with voiceprint features through a spatiotemporal attention mechanism, and output a dynamic identity verification code; An identification module is configured to activate a target safe deposit box group based on the dynamic authentication code, identify items within the safe deposit boxes using an anti-interference RFID signal optimization model, and utilize a spatiotemporal convolutional network to separate item RFID signals from electromagnetic noise to generate a set of unique item identifiers. An optimization module is used to dynamically optimize the spatial layout of cabinet groups based on a discrepancy analysis between the unique identifiers of the items and the preset inventory list, using a reinforcement learning algorithm, constructing a three-dimensional spatial value function based on the access frequency and volume characteristics of the items, and outputting an optimal cabinet allocation plan; The execution module is used to drive the modular cabinet group actuator to adjust the physical structure according to the optimal cabinet allocation plan, synchronously encrypt and write the item status data into the distributed ledger, and trigger the edge computing node to update the global inventory topology map.

8. The system according to claim 7, characterized in that The receiving module is specifically configured to: Based on the permission parameters of the online approval request, the node permission thresholds in the approval hierarchy topology are parsed to generate a directed graph of the approval process with a timestamp. The user's biometric hash value is temporally and spatially associated with the approval process directed graph, and the Byzantine fault-tolerant algorithm is used to verify the legitimacy of the signature timing of approval nodes at each level. Based on the verification results, an approval chain hash tree is constructed, and conflicting approval branches are pruned using a timestamp-weighted voting mechanism to generate a compliant approval path hash chain. By integrating the compliance approval path hash chain with the user's biometric entropy value, a dynamic digital token with permission timeliness is generated through the elliptic curve encryption algorithm.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 6 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 6.

Images