A blockchain-based smart campus settlement system and method

Abstract

The application discloses a kind of based on blockchain wisdom campus settlement system and method, it is related to machine vision technical field, the system includes: visual perception module, intention mapping module and settlement module;Its technical key points are: obtaining the data set of wisdom campus multi-settlement scene;First machine model is handled with settlement data packet using preset, clustering analysis is executed in combination with on-chain historical settlement event, generates first embedding identifier, behavior trajectory sequence is handled using preset second machine model, identifies the interactive behavior of user and target settlement area, generates second embedding identifier;In combination with the network load data of call, settlement intention feature is handled and generated, and the mapping relationship of settlement intention feature and account state is established;Using mapping relationship, the constraint range of each type of settlement scene is obtained, if meeting constraint range, update account state, and synchronously upload to blockchain;The application improves the intelligent level and response efficiency of multi-scene settlement decision.

Description

Technical Field

[0001] This invention relates to the field of machine vision technology, specifically to a blockchain-based smart campus settlement system and method. Background Technology

[0002] With the advancement of digital transformation in education, smart campus settlement systems have become a core infrastructure connecting student and teacher consumption, campus management, and home-school collaboration. They cover multiple application scenarios such as tuition payment, cafeteria consumption, supermarket shopping, and access control. Existing technologies mostly adopt a centralized information management model, realizing multi-terminal data interaction through the network. Some solutions introduce technologies such as fingerprint verification and blockchain evidence storage to improve security and management efficiency.

[0003] Traditional smart campus payment technologies have the following limitations: Firstly, some solutions incorporate machine vision recognition, but this is limited to single-target detection and recognition. If a user unconsciously looks at or accidentally touches other types of items, the system cannot recognize the user's intent. Especially when network load is extremely unstable, the robustness of identifying target items and areas is poor, making it unsuitable for complex payment scenarios such as peak hours in cafeterias. This leads to timeouts in payment devices and significantly reduces the success rate of payments. Secondly, as the demands of smart campus payment scenarios upgrade, contactless payment technologies have emerged in traditional RFID trays and QR code card swiping. However, current contactless payment technologies use a fixed triggering sequence, such as aligning with a facial recognition frame before reaching for the item. In reality, there may be cross-triggers such as looking and taking, which cannot adapt to multi-person interaction modes. If the user's action does not conform to the predetermined triggering sequence, it is considered an invalid operation, which not only greatly reduces traffic efficiency but also creates a rigid user experience. Furthermore, although some technologies incorporate blockchain technology, it is only used for traceability and recording. In scenarios such as open shelves or multiple payment terminals in cafeterias, there are distributed concurrent security risks. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a blockchain-based smart campus settlement system and method. It acquires a dataset of multiple settlement scenarios within a smart campus, processes settlement data packets using a first machine learning model, and generates a first embedded identifier by clustering historical settlement events on the blockchain. A second machine learning model processes user behavior trajectory sequences, identifies their interaction with the target settlement area, and generates a second embedded identifier. Real-time network load data is combined with the dual embedded identifiers to generate settlement intent features, establishing a mapping relationship between these features and account status. When the settlement intent meets the constraints of the corresponding scenario, the account status is automatically updated, and all process data is synchronized and stored on the blockchain, thus solving the problems mentioned in the background technology.

[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: Firstly, this application provides a blockchain-based smart campus settlement system, the system comprising: The visual perception module acquires data sets from multiple settlement scenarios in the smart campus; these data sets include settlement data packets and behavioral trajectory sequences. The intent mapping module processes settlement data packets using a preset first machine model, performs cluster analysis based on on-chain historical settlement events, generates a first embedded identifier, processes behavioral trajectory sequences using a preset second machine model, identifies user interaction behavior with the target settlement area, and generates a second embedded identifier; and processes the first and second embedded identifiers based on retrieved network load data to generate settlement intent features, and establishes a mapping relationship between settlement intent features and account status. The settlement module uses mapping relationships to obtain the constraint range for each type of settlement scenario. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously.

[0006] Furthermore, acquire data sets from multiple settlement scenarios within the smart campus, including: Multiple edge computing nodes are deployed and connected to several settlement terminals to build a campus sensing topology and establish a campus settlement portal; the settlement terminals include interactive terminals and visual sensing gateways. Several independent settlement data packets are established through the interactive terminal. Each settlement data packet corresponds to a unique identity, account status, and network interface MAC address. The visual perception gateway responds to the capture command and uses an infrared trigger sensor to sense that the user has entered the target settlement area. It activates a multi-angle camera to sample the user's hand operation points and gaze focus in real time. After triggering the settlement event, a sequence of behavioral trajectories is formed. Simultaneously, the identity identifier associated with the current settlement event window is retrieved from the blockchain in real time, encrypted, and fed back to the settlement data packet to perform an account status update.

[0007] Furthermore, the settlement data packet is processed using a preset first machine model, including: The first machine model is configured as a clustering network that includes a multi-scale feature extraction mechanism; Acquire on-chain historical settlement events to build a settlement behavior sample library; import the settlement behavior sample library and settlement data package into the first machine model, use the first feature unit and the second feature unit connected in sequence to extract the amount fluctuation features and frequency distribution features at different scales, and perform non-linear alignment to generate a settlement distribution map; Spatial mapping is performed on the settlement distribution map, and users are assigned to the corresponding credit clusters based on the calculated feature similarity through the pooling layer, and the first embedding identifier is output.

[0008] Furthermore, the behavioral trajectory sequence is processed using a pre-defined second machine model, including: The contour features and pupil features are input into the second machine model, and asynchronous feature extraction is performed using the first and second channels that are connected in sequence: First channel: The geometric shape of the contour features is analyzed using a continuous convolution operator, and the first execution feature of the first closed chain relative to the second closed chain is extracted along the time series, including: deformation rate; Second channel: The gaze projection coordinates of the pupil features are identified using a single convolution operator, and the second execution feature of the gaze projection coordinates relative to the second closed chain is extracted along the time series, including: dwell time and distribution density; By concatenating the first and second execution features, and using sequential multi-layer convolutional layers for correlation analysis along the time series, a second embedding identifier is output.

[0009] Furthermore, the settlement intent features are generated, including: Generate at least one group of interactive intent vectors based on the first embedded identifier and the second embedded identifier; The Euclidean distance between the gaze focus and the hand operation point is calculated in real time. Cases where the Euclidean distance is less than a standard distance threshold and this condition lasts for at least 2 frames are filtered out, triggering an interaction intent recognition. By combining the campus topology network, the topological relationships in the visual window are dynamically obtained based on the current settlement event. The target settlement items are identified through semantic segmentation, and candidate semantic features are extracted. Under the condition of network stability, the maximum points of the candidate semantic features are identified. If a certain type of target settlement item exists in consecutive frames, it is used as the final semantic feature and recorded in the interaction intent vector group. The network-stable scalar operator is extracted, and the modulus of the interaction intent vector group is dynamically scaled to generate the settlement intent feature of the current settlement event.

[0010] Furthermore, the process of recognizing an interaction intent also includes: Before an interaction intent is recognized, the identity verification process begins. When the settlement event progresses to the point of being triggered, the current user is determined to be in the target settlement area and selected as the visual window. Identity verification is used as the calibration condition. After confirming that the calibration condition is met, a feedback instruction is sent to the blockchain to update the current account status to frozen and lock transaction permissions. At the same time, the visual perception gateway is immediately activated to collect hand gestures and gaze focus and perform an interaction intent recognition.

[0011] Furthermore, the criteria for judging network stability include: Retrieve network load data, including bandwidth, latency, and packet loss rate; convert the network load data into a scalar operator using a weight adjustment coefficient; the weight adjustment coefficient is a dynamic value determined by the particle swarm optimization algorithm; if the scalar operator is less than the standard stability threshold, mark the current scalar operator as 'a' and assign it to the first level, and combine 'a' with the first level to generate network instability information; if the scalar operator is greater than or equal to the standard stability threshold, mark the current scalar operator as 'b' and assign it to the second level, and combine 'b' with the second level to generate network stability information.

[0012] Furthermore, a mapping relationship is established between settlement intention characteristics and account status, including: Retrieve the current account status, and parse and output the first embedded identifier and the second embedded identifier according to the settlement intent characteristics; create a first time tag for the first embedded identifier according to the time series to form a first mapping relationship between the settlement intent characteristics and the account status; create a second time tag for the second embedded identifier according to the time series to form a second mapping relationship between the settlement intent characteristics and the account status.

[0013] The constraints must be satisfied, including: Based on the mapping relationship, the timing triggering features of the first time tag and the second time tag are extracted, including synchronous triggering, sequential triggering and cross triggering; Identify the current settlement scenario, retrieve the smart contract for the current settlement scenario from the blockchain, and the smart contract contains an association constraint graph; map the time-series triggering features onto the association constraint graph, and through structural isomorphic search, if at most one connected path is found, it is determined that the constraint range is met, and drive the smart contract to change the account status to unfrozen.

[0014] Secondly, this application provides a blockchain-based smart campus settlement method, the method comprising: Acquire a dataset of settlement scenarios in a smart campus; the dataset includes settlement data packets and behavior trajectory sequences. The system processes settlement data packets using a pre-defined first machine model, performs cluster analysis based on on-chain historical settlement events, generates a first embedded identifier, processes behavioral trajectory sequences using a pre-defined second machine model, identifies user interaction with the target settlement area, and generates a second embedded identifier. Based on the retrieved network load data, the first and second embedded identifiers are processed to generate settlement intent features, and a mapping relationship between settlement intent features and account status is established. By utilizing the mapping relationship, the constraint range of each type of settlement scenario is obtained. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously.

[0015] (III) Beneficial Effects This invention provides a blockchain-based smart campus settlement system and method, which has the following beneficial effects: This invention constructs a campus sensing topology, collects data from multiple settlement scenarios through settlement terminals and visual sensing gateways, and allows for the free addition of these terminals and gateways via the campus portal, facilitating remote configuration and data aggregation and processing. This significantly reduces the maintenance costs of the smart management platform. By setting up a blockchain, authentication and encryption of the current visual window are immediately initiated after a settlement event is triggered, ensuring the security of the settlement process. Simultaneously, centralized data management on a single smart management platform improves the efficiency of subsequent analysis and the timeliness of problem-solving. This invention utilizes a first machine model to generate a first embedded identifier using credit clusters, and a second machine model to generate a second embedded identifier by filtering the Euclidean distance between the gaze focus and hand operation. By determining network stability information, and combining interactive intent vector groups with the campus topology network, semantic feature embedding is performed to generate settlement intent features, effectively filtering out non-settlement interference actions. Simultaneously, under the condition of determining network stability information, the particle swarm optimization algorithm is used to dynamically optimize the scalar operator, realizing dynamic scaling of the magnitude of the intent feature vector, executing smart contract consensus, significantly improving the payment pass rate during peak campus periods, and automatically switching to lightweight computational analysis when the network is unstable, ensuring the security of the settlement process. This invention automatically analyzes the timing characteristics of synchronous, sequential, and cross-triggered events using first and second time tags, enhancing the system's adaptability to diverse interaction scenarios. By invoking smart contracts, the extracted timing characteristics are mapped onto an association constraint graph. A complete connected path is found through structural isomorphism search, driving the smart contract to update the account status, significantly reducing the time consumption of settlement events. To a certain extent, this ensures the data consistency of the smart campus settlement system and improves the intelligence level and response efficiency of settlement decisions in multiple scenarios. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a smart campus settlement system according to an exemplary embodiment; Figure 2 This is a flowchart illustrating a smart campus settlement method according to an exemplary embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] The core of this invention lies in: acquiring a data set of multiple settlement scenarios in a smart campus; processing settlement data packets through a first machine model and generating a first embedded identifier by clustering historical settlement events on the blockchain; processing user behavior trajectory sequences through a second machine model to identify their interaction with the target settlement area and generating a second embedded identifier; combining real-time network load data and fusing the dual embedded identifiers to generate settlement intent features and establishing a mapping relationship between them and account status; to a certain extent, this improves the accuracy of settlement intent recognition; when the settlement intent meets the constraints of the corresponding scenario, the account status is automatically updated, and the entire process data is synchronized to the blockchain for evidence storage, facilitating the traceability of settlement events.

[0019] Example 1: This embodiment of the invention provides a blockchain-based smart campus settlement system; Figure 1 This is a schematic diagram of a smart campus settlement system according to an exemplary embodiment; please refer to... Figure 1 The system includes a visual perception module, an intent mapping module, and a settlement module, and the visual perception module, intent mapping module, and settlement module are interconnected. The following is an explanation of each module involved: Visual perception module: Acquires data sets from multiple settlement scenarios in the smart campus; the data sets include settlement data packets and behavior trajectory sequences; Acquire data sets for multiple settlement scenarios in a smart campus, including: deploying multiple edge computing nodes and connecting several settlement terminals within the smart campus, building a campus perception topology and establishing a campus settlement portal; wherein, the settlement terminals include interactive terminals and visual perception gateways; In this embodiment, the campus perception topology is a logical topology network built upon the smart campus. The smart campus is divided into a set of multiple settlement scenarios. These settlement scenarios are abstracted as scenario nodes, each corresponding to a specific edge computing node and visual perception gateway within the smart campus. For example, settlement scenarios include, but are not limited to, cafeteria windows, unmanned supermarkets, and self-service borrowing machines. When a new settlement scenario or terminal is accessed, it is marked as an increase in resource volume, requiring the addition of a corresponding scenario node to the set of multiple settlement scenarios and updating the corresponding logical topology network. Conversely, when a new student or faculty member performs their first settlement or identity activation at any scenario node, it is marked as an access count. As the system grows, it adds corresponding user nodes to the user node set. It captures user-triggered settlement events through a visual perception gateway, and uses data captured by visual sensors to build topological edges between user nodes and scene nodes, thus constructing a campus perception topology. By refining scene nodes, such as from the cafeteria to the cafeteria window, it can achieve sharded processing of large-scale settlement data and parallel consensus with the blockchain. It should be noted that the campus settlement portal is configured for data aggregation and processing, including but not limited to remote configuration of the campus perception topology, user management, and blockchain management. It should also be noted that the sensors or devices mentioned above are not shown in the diagram and are installed adaptively according to actual conditions; therefore, they will not be elaborated upon here. Several independent settlement data packets are established through the interactive terminal. Each settlement data packet corresponds to a unique identity, account status, and network interface MAC address. The intelligent management platform sends a behavior capture command to the visual perception gateway. The visual perception gateway responds to the behavior capture command by using an infrared trigger sensor to detect when a user enters the target settlement area. It then activates a multi-angle camera to sample the user's hand operation points and gaze focus in real time. After a settlement event is triggered, the hand operation points are excited to generate a serialized displacement. In subsequent analysis, the fingertip of the index finger is analyzed, and a behavior trajectory sequence is formed based on the dynamic changes of the hand operation points and gaze focus. Simultaneously, the identity identifier associated with the current visual window is retrieved from the blockchain in real time. After encryption processing, such as using a hash algorithm to encrypt the identity identifier, it is fed back to the settlement data packet to perform subsequent account status updates.

[0020] By building a campus-wide sensing topology, data from multiple settlement scenarios is collected through settlement terminals and visual sensing gateways. These terminals and gateways can be freely added to the campus portal, facilitating remote configuration and data aggregation and processing. This significantly reduces the maintenance costs of the smart management platform. By setting up a blockchain, authentication and encryption of the current visual window are immediately initiated after a settlement event is triggered, ensuring the security of the settlement process. At the same time, data is centrally managed on a single smart management platform, improving the efficiency of subsequent analysis and the timeliness of problem-solving.

[0021] Intent Mapping Module: Processes settlement data packets using a preset first machine model, performs cluster analysis based on on-chain historical settlement events to generate a first embedded identifier, processes behavioral trajectory sequences using a preset second machine model to identify user interaction with the target settlement area, and generates a second embedded identifier; combines retrieved network load data to process the first and second embedded identifiers to generate settlement intent features, and establishes a mapping relationship between settlement intent features and account status. The process of processing settlement data packets using a pre-defined first machine model includes: configuring the pre-defined first machine model as a clustering network containing a multi-scale feature extraction mechanism; acquiring on-chain historical settlement events to construct a settlement behavior sample library; wherein, the steps for acquiring historical settlement events include: initiating a smart contract query in the blockchain based on the identity identifier in the current settlement data packet, selecting all on-chain settlement event records of the user within a pre-defined historical time window, such as the past 30 days; arranging the on-chain settlement event records in chronological order, extracting the settlement amount and settlement time interval from each settlement event record, constructing a two-dimensional settlement behavior sample library matrix, and inputting it and the current settlement data packet as a whole into the pre-defined first machine model; The algorithm extracts monetary fluctuation features and frequency distribution features at different scales using a first and second feature unit connected internally: the first feature unit is configured with a small convolution operator, such as a 3×1 one-dimensional convolutional layer, which slides across the time series with a small stride, such as 10 minutes, to capture continuous high-frequency payment behaviors or sudden jumps in monetary amounts within a similar time period, outputting monetary fluctuation features; the second feature unit is configured with a larger convolution operator, which slides across the time series with a larger stride to perform downsampling analysis to filter out short-term transaction noise, such as 24 hours, extracting the user's basic consumption baseline and periodic payment frequency within a month, outputting frequency distribution features; by performing nonlinear alignment on the monetary fluctuation features and frequency distribution features, the algorithm uses MLP to map the monetary fluctuation features and frequency distribution features to the same latent high-order features. The system constructs a feature space and performs channel alignment using bilinear interpolation. The aligned features are then concatenated and stitched together to generate a settlement distribution map. It should be noted that the clustering network is based on K-means clustering. The settlement distribution map is then spatially mapped to a predefined credit feature space. In this credit feature space, the feature similarity between the K credit cluster centers and the settlement distribution map is calculated. A pooling layer selects the credit cluster center with the smallest geometric distance as the matching target, determining it to have the highest feature similarity. This center is then assigned to the corresponding credit cluster, and the first embedding identifier is output. Here, K is a positive integer, typically 3, representing high-quality credit clusters, volatile credit clusters, and potential risk clusters. However, this analysis is merely an example; the specific value needs to be set according to the actual situation, which will not be elaborated upon here. The behavior trajectory sequence is processed using a pre-defined second machine model, including: the pre-defined second machine model is configured as a visual tracking network containing a time-series correlation mechanism; computer vision algorithms are used to analyze the user's interaction behavior with the target checkout area, the behavior trajectory sequence is pre-processed, and contour features and pupil features that trigger the checkout event are extracted; wherein, the pre-processing includes, but is not limited to, contrast enhancement and structure enhancement: contrast enhancement: for each frame of the real-time acquired behavior trajectory sequence, a pre-defined sliding window is used to extract the local pixel neighborhood, the ratio of the pixel extreme value difference in the neighborhood to the local average brightness is calculated, and it is marked as the local contrast factor; structure enhancement The system employs a preset brightness response threshold and performs convolution mapping between the local contrast factor and the brightness response threshold. By increasing the gradient change rate of the edge regions in the behavior trajectory sequence, it achieves structured enhancement at the junction of the head and background, and the junction of the target settlement area and the background. Feature extraction involves performing multi-scale gradient operator extraction on the enhanced image to obtain the edge response of the image at different resolutions, thereby suppressing visual artifacts caused by ambient light fluctuations. Based on the edge response, closed geometric edges are extracted. Morphological dilation and erosion operations are used to identify the first closed chain representing the edge of the user's hand and the second closed chain representing the edge of the target settlement area, which are then determined as contour features. The contour features and pupil features are input into the second machine model, and asynchronous feature extraction is performed using the first and second channels connected internally: First channel: Two consecutive 3×1 convolutional layers are set up, and the geometric morphology of the contour features is analyzed using consecutive convolution operators. The first execution feature of the first closed chain relative to the second closed chain is extracted along the time series, including: deformation rate. It should be noted that the first execution feature characterizes the dynamic evolution of hand operation points, such as calculating the deformation rate of the geometric morphological changes of the closed chain between adjacent frames. Specifically, the user's interaction behavior with the target settlement area is analyzed by examining the user's hand extension and contraction. Dynamic changes occur in the checkout scenario. When a user reaches for an item, the shape of their hand undergoes a rapid change from open to closed. High deformation rate represents explicit interactive actions, such as placing or holding, while low deformation rate represents relatively static or slow movement, such as waving the hand while passing by a dish. To a certain extent, this indirectly determines whether a checkout event has been triggered. The second channel uses a 1×1 convolution operator to analyze the gaze projection coordinates of pupil features. It extracts the dwell time and distribution density of the gaze projection coordinates relative to the second closed chain along the time series, and trains the signal through weighted aggregation. The weight of the change in gaze projection coordinates is calculated. It should be noted that the second execution feature represents the dynamic evolution of the user's gaze focus. For example, the change in gaze projection coordinates of the closed chain between adjacent frames is calculated, and the user's interaction with the target checkout area is specifically the dynamic change of the user's gaze focus. The dwell time represents the continuous duration for which the gaze projection coordinates fall within the second closed chain. The user's gaze focus may jump rapidly. If the dwell time exceeds a certain threshold, it is marked as if the user may intend to buy. In this embodiment, it is set to more than 2 consecutive frames, and the specific value is set according to the actual situation. The distribution density represents the degree of spatial aggregation of gaze projection coordinate points within the second closed chain in the visual window. High distribution density means that the user's pupils repeatedly and accurately lock onto the same target checkout item, such as staring at a chicken leg, which can express the directionality of the checkout intention. Low distribution density means that the gaze is scattered and there are multiple target checkout items. To a certain extent, it can indirectly determine whether a checkout event has been triggered. Through the splicing of the first execution feature and the second execution feature, a multi-layer 3×1 convolution operator is set. By using the correlation analysis along the time series of continuous multi-layer convolution layers, the first execution feature and the second execution feature are flattened and correlated and decoded to output the second embedding identifier.

[0022] The process of generating settlement intent features includes: generating at least one set of interaction intent vectors based on a first embedded identifier and a second embedded identifier; calculating the Euclidean distance between the gaze focus and the hand operation point in real time, filtering cases where the Euclidean distance is less than a standard distance threshold, and such cases must last for at least 2 frames; specifically, calculating the Euclidean distance between the gaze focus and the fingertip of the index finger: extracting the coordinates of the fingertip of the index finger and the gaze focus using computer vision algorithms, performing temporal alignment and coordinate system unification, and using linear interpolation to ensure that each set of fingertip coordinates and gaze focus coordinates corresponds to the same moment; comparing the Euclidean distance with the standard distance threshold: if the Euclidean distance is greater than or equal to the standard distance threshold, interaction intent recognition is not triggered; if the Euclidean distance is less than the standard distance threshold, and such cases must last for at least 2 frames, then interaction intent recognition is activated once; it should be noted that the interaction intent is represented as follows: if the user has a clear interaction intent, they will focus their gaze on the target settlement item, and then their hand will naturally move towards the target settlement item, and at the moment the action is executed, the gaze focus and the hand operation point will be highly coincident in space; The standard distance threshold is derived from analyzing Euclidean distance data from past operations in a specific settlement scenario. Statistical analysis of the collected data determines the mean and standard deviation of the Euclidean distance data under activated interaction intent recognition. The standard distance threshold is set to the mean plus a multiple of the standard deviation. It should be noted that the value of the multiple of 2 is just an example and should be set according to the actual situation, which will not be elaborated here. Based on the dynamic acquisition of the topological relationships in the visual window of the current settlement event, target settlement items are identified through semantic segmentation, so that multiple target settlement items present peaks in the feature space, and candidate semantic features are extracted. The entire feature space is traversed, and the maximum point of the candidate semantic features is searched under the condition of network stability information. If a certain type of target settlement item appears continuously in consecutive frames and the recognition probability is stable, it is taken as the final semantic feature and recorded in the interaction intent vector group. The scalar operator of network stability is extracted, and the modulus of the interaction intent vector group is dynamically scaled to generate the settlement intent feature of the current settlement event. The judgment criteria for network stability information include: retrieving network load data, including bandwidth, latency rate and packet loss rate; standardizing the network load data, normalizing it to the interval of 0 to 1, eliminating the influence of units, and making the subsequent calculations have physical meaning. Using weighted adjustment coefficients, network load data is weighted and aggregated to generate a comprehensive load coefficient: For normalized network load data, the network load data is weighted and summed using weighted adjustment coefficients. The normalized bandwidth is multiplied by the first weighted adjustment coefficient, the reciprocal of the normalized latency rate is taken, and the result is multiplied by the second weighted adjustment coefficient. The reciprocal of the normalized packet loss rate is taken, and the result is multiplied by the third weighted adjustment coefficient. The results of the three multiplications are added together to generate a scalar operator. The formula is: Comprehensive load coefficient = Normalized bandwidth × First weighted adjustment coefficient + (1 / Normalized latency rate) × Second weighted adjustment coefficient + (1 / Normalized packet loss rate) × Third weighted adjustment coefficient; where the value range of the scalar operator is 0 to 1, the weighted adjustment coefficients are dynamic values, and the values ​​of the first, second, and third weighted adjustment coefficients are all in the range of 0 to 1. The sum of the first, second, and third weighted adjustment coefficients is 1. The weight adjustment coefficients are dynamic values, determined by the particle swarm optimization algorithm. The algorithm involves: constructing a high-dimensional space and generating a reference vector set to provide direction for population evolution; randomly generating an initial population, setting the iteration count to t_max (greater than 0); assigning each particle a unique position and velocity vector; where the position vector typically represents a potential solution, and the velocity vector indicates the search direction; combining the particle's position in space as weights, including a first, second, and third weight adjustment coefficient; introducing binary competitive selection, crossover, and mutation operations to select the best-fit particles and add them to the mating pool, repeating this process until the mating pool reaches its maximum size; randomly selecting two particles from the mating pool and adding them to a subpopulation; filling the subpopulation with the maximum number of individuals; merging the subpopulation with the parent population to form a mixed population; finding the minimum value of each objective in the mixed population as the ideal point and driving particles to move towards the ideal point; thus calculating the optimal first, second, and third weight adjustment coefficients that maximize the search under stable network conditions. It should be noted that normalized bandwidth represents throughput capacity, and its value is positively correlated with network stability; the larger the bandwidth, the more data can be processed. Normalized latency represents system response lag, and its value is negatively correlated with network stability; the higher the latency, the slower the processing speed and the lower the timeliness of subsequent intent recognition. Normalized packet loss rate represents data noise, and its value is negatively correlated with network stability; the higher the packet loss rate, the more likely smart contracts on the blockchain will fail. It should also be noted that the values ​​of the above-mentioned data are all between 0 and 1. Calculated using the above formula, the larger the value of the comprehensive load coefficient, the more stable the network; the smaller the value of the comprehensive load coefficient, the less stable the network. The comprehensive load coefficient is converted into a scalar operator using the Sigmoid activation function: Sigmoid(m·(comprehensive load coefficient - standard load threshold)); where Sigmoid() represents the activation function, m represents the gain coefficient, and its value ranges from 0 to 1. It is used to adjust the system's response gradient to sudden changes in network load. For example, in an open campus environment with frequent network signal jitter, the value of m is reduced to increase the smoothness of the mapping; conversely, in a cafeteria peak period scenario with extremely low tolerance for payment delay, its specific value is obtained by removing the top peak through particle swarm optimization and using gradient descent. The system has a parameter correction interface, allowing real-time collected network load data to fine-tune the value of m, which will not be elaborated here. It should be noted that the scalar operator acts on the subsequent settlement intention feature, and its value ranges from 0 to 1. If the scalar operator approaches 1, it indicates network stability; if the scalar operator approaches 0, it indicates network instability. If the scalar operator is less than the standard stability threshold, the current scalar operator is marked as 'a' and assigned the first level. 'a' is then combined with the first level to generate network instability information. If the scalar operator is greater than or equal to the standard stability threshold, the current scalar operator is marked as 'b' and assigned the second level. 'b' is then combined with the second level to generate network stability information. It should be noted that the standard stability threshold and the standard load threshold are obtained through cross-validation. The specific steps include: aligning historical settlement event logs and network quality logs by timestamp to obtain a sample set; extracting successfully settled events as positive samples and unsuccessfully settled events as negative samples. Multi-fold cross-validation is used to optimize the threshold range. The sample set is divided into 5 subsets, with 1 subset used as the validation set and the other 5 subsets used as the training set. For the combination of standard stable threshold and standard load threshold, settlement events are simulated on the training set, and the threshold range corresponding to the threshold combination is recorded. Evaluation indicators are calculated on the validation set, and this process is repeated 5 times. In each iteration, the goal is to maximize the settlement success rate. By calculating the mean and standard deviation of all validation frequencies, the parameter combination with the most stable performance distribution is selected as the final standard stable threshold and standard load threshold, thereby eliminating the risk of overfitting in a single settlement scenario. The dynamic scaling of the modulus of the interaction intent vector group includes: calculating the modulus of the interaction intent vector group using the L2 norm to characterize the intensity of the interaction intent; introducing a scalar factor and multiplying it by the modulus to achieve linear scaling, obtaining the updated modulus, and generating the final settlement intent feature; it should be noted that during the calculation process, the data involved in the calculation needs to be normalized to eliminate the influence of dimensions and make the result physically meaningful; in particular, the process of one interaction intent recognition also includes: before the first interaction intent recognition, entering the identity identification preparation stage; when the settlement event progresses to the waiting stage, determining that the current user is in the target settlement area and selecting it as the visual window; using the identity identification verification as the calibration condition, after confirming that the calibration condition is met, sending a feedback instruction to the blockchain to update the current account status to frozen and lock the transaction permissions; at the same time, immediately activating the visual perception gateway to perform data collection of the user's hand operation points and gaze focus and one interaction intent recognition; Establishing a mapping relationship between settlement intent features and account status includes: retrieving the current account status, parsing and outputting a first embedded identifier and a second embedded identifier according to the settlement intent features; creating a first time tag for the first embedded identifier according to the time series to form a first mapping relationship between settlement intent features and account status; and creating a second time tag for the second embedded identifier according to the time series to form a second mapping relationship between settlement intent features and account status.

[0023] The system utilizes a first machine model to generate a first embedded identifier using credit clusters, and a second machine model to generate a second embedded identifier by filtering the Euclidean distance between the gaze focus and hand operation. By determining network stability information, and combining the interaction intent vector group with the campus topology network, semantic feature embedding is performed to generate settlement intent features, effectively filtering out non-settlement interference actions. Simultaneously, under the condition of determining network stability information, the particle swarm optimization algorithm is used to dynamically optimize the scalar operator, realizing dynamic scaling of the magnitude of the intent feature vector, executing smart contract consensus, significantly improving the payment pass rate during peak campus periods, and automatically switching to lightweight computational analysis when the network is unstable, ensuring the security of the settlement process.

[0024] Settlement module: Utilizes mapping relationships to obtain the constraint range for each type of settlement scenario. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously. The constraints include: extracting the timing triggering features of the first and second time tags based on the mapping relationship, including synchronous triggering, sequential triggering, and cross-triggered triggering. The following is an explanation of some of the terms involved: Synchronous triggering: Calculate the time deviation between the first and second time tags. If the time deviation is less than 200ms, it indicates that the first and second time tags are triggered synchronously. Sequential triggering: Calculate the time deviation between the first and second time tags. If the time deviation is greater than or equal to 200ms, and the timestamp corresponding to the first time tag is less than the timestamp corresponding to the second time tag, it indicates that the first and second time tags are triggered sequentially. Cross-triggered triggering: Calculate the time deviation between the first and second time tags. If the time deviation is greater than or equal to 200ms and the first and second time tags intersect in the time series, it indicates that the first and second time tags are cross-triggered. It should be noted that the specific data mentioned above is an example; the actual settings should be based on the specific circumstances and will not be elaborated upon here. The current settlement scenario is identified, and the smart contract for that scenario is retrieved from the blockchain. This smart contract contains an association constraint graph, where nodes represent the progress state of a settlement event, such as entering the target settlement area or identifying an intent. Edges in the association constraint graph represent temporal triggering features. These extracted temporal triggering features are mapped onto the association constraint graph. Through structural isomorphism search, these features are encapsulated into subgraphs. Subgraphs with identical structures to the association constraint graph are searched, and their corresponding nodes are obtained. A complete connected path is found, indicating that the constraints are met, and the smart contract is then driven to change the account status to unfrozen.

[0025] By automatically analyzing the timing characteristics of synchronous, sequential, and cross-triggered events using first and second time tags, the system demonstrates its adaptability to diverse interaction scenarios. By invoking smart contracts, the extracted timing characteristics are mapped onto an association constraint graph. A complete connected path is found through structural isomorphism search, driving the smart contract to update the account status, significantly reducing the time consumption of settlement events. To a certain extent, this ensures the data consistency of the smart campus settlement system and improves the intelligence level and response efficiency of settlement decisions in multiple scenarios.

[0026] Example 2: This embodiment of the invention provides a blockchain-based smart campus settlement method; Figure 2 This is a flowchart illustrating a smart campus settlement method according to an exemplary embodiment; please refer to [link / reference]. Figure 2 The method includes the following steps: acquiring a data set of multiple settlement scenarios in a smart campus; wherein, the data set includes settlement data packets and behavior trajectory sequences; The system processes settlement data packets using a pre-defined first machine model, performs cluster analysis based on on-chain historical settlement events, generates a first embedded identifier, processes behavioral trajectory sequences using a pre-defined second machine model, identifies user interaction with the target settlement area, and generates a second embedded identifier. Based on the retrieved network load data, the first and second embedded identifiers are processed to generate settlement intent features, and a mapping relationship between settlement intent features and account status is established. By utilizing the mapping relationship, the constraint range of each type of settlement scenario is obtained. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously.

[0027] In the application, the various formulas mentioned are all calculated by removing dimensions and taking their numerical values. The formulas are derived from the most recent real-world situation by collecting a large amount of data and conducting software simulations. The formulas are set by those skilled in the art according to the actual situation.

[0028] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0029] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0030] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A blockchain-based smart campus settlement system, characterized in that, The system includes: The visual perception module acquires data sets from multiple settlement scenarios in the smart campus; these data sets include settlement data packets and behavioral trajectory sequences. The intent mapping module processes settlement data packets using a preset first machine model, performs cluster analysis based on on-chain historical settlement events, generates a first embedded identifier, processes behavioral trajectory sequences using a preset second machine model, identifies user interaction behavior with the target settlement area, and generates a second embedded identifier; and processes the first and second embedded identifiers based on retrieved network load data to generate settlement intent features, and establishes a mapping relationship between settlement intent features and account status. The settlement module uses mapping relationships to obtain the constraint range for each type of settlement scenario. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously.

2. The smart campus settlement system based on blockchain according to claim 1, characterized in that, Obtain data sets from multiple settlement scenarios in the smart campus, including: Multiple edge computing nodes are deployed and connected to several settlement terminals to build a campus sensing topology and establish a campus settlement portal; the settlement terminals include interactive terminals and visual sensing gateways. Several independent settlement data packets are established through the interactive terminal. Each settlement data packet corresponds to a unique identity, account status, and network interface MAC address. The visual perception gateway responds to the capture command and uses an infrared trigger sensor to sense that the user has entered the target settlement area. It activates a multi-angle camera to sample the user's hand operation points and gaze focus in real time. After triggering the settlement event, a sequence of behavioral trajectories is formed. Simultaneously, the identity identifier associated with the current settlement event window is retrieved from the blockchain in real time, encrypted, and fed back to the settlement data packet to perform an account status update.

3. The smart campus settlement system based on blockchain according to claim 2, characterized in that, The settlement data packet is processed using a preset first machine model, including: The first machine model is configured as a clustering network that includes a multi-scale feature extraction mechanism; Acquire on-chain historical settlement events to build a settlement behavior sample library; import the settlement behavior sample library and settlement data package into the first machine model, use the first feature unit and the second feature unit connected in sequence to extract the amount fluctuation features and frequency distribution features at different scales, and perform non-linear alignment to generate a settlement distribution map; Spatial mapping is performed on the settlement distribution map, and users are assigned to the corresponding credit clusters based on the calculated feature similarity through the pooling layer, and the first embedding identifier is output.

4. A blockchain-based smart campus settlement system according to claim 2, characterized in that, The behavior trajectory sequence is processed using a pre-defined second machine model, including: The contour features and pupil features are input into the second machine model, and asynchronous feature extraction is performed using the first and second channels that are connected in sequence: First channel: The geometric shape of the contour features is analyzed using a continuous convolution operator, and the first execution feature of the first closed chain relative to the second closed chain is extracted along the time series, including: deformation rate; Second channel: The gaze projection coordinates of the pupil features are identified using a single convolution operator, and the second execution feature of the gaze projection coordinates relative to the second closed chain is extracted along the time series, including: dwell time and distribution density; By concatenating the first and second execution features, and using sequential multi-layer convolutional layers for correlation analysis along the time series, a second embedding identifier is output.

5. A blockchain-based smart campus settlement system according to claim 4, characterized in that, Generate settlement intent features, including: Generate at least one group of interactive intent vectors based on the first embedded identifier and the second embedded identifier; The Euclidean distance between the gaze focus and the hand operation point is calculated in real time. Cases where the Euclidean distance is less than a standard distance threshold and this condition lasts for at least 2 frames are filtered out, triggering an interaction intent recognition. By combining the campus topology network, the topological relationships in the visual window are dynamically obtained based on the current settlement event. The target settlement items are identified through semantic segmentation, and candidate semantic features are extracted. Under the condition of network stability, the maximum points of the candidate semantic features are identified. If a certain type of target settlement item exists in consecutive frames, it is used as the final semantic feature and recorded in the interaction intent vector group. The network-stable scalar operator is extracted, and the modulus of the interaction intent vector group is dynamically scaled to generate the settlement intent feature of the current settlement event.

6. A blockchain-based smart campus settlement system according to claim 5, characterized in that, The process of recognizing an interaction intent also includes: Before an interaction intent is recognized, the identity verification process begins. When the settlement event progresses to the point of being triggered, the current user is determined to be in the target settlement area and selected as the visual window. Identity verification is used as the calibration condition. After confirming that the calibration condition is met, a feedback instruction is sent to the blockchain to update the current account status to frozen and lock transaction permissions. At the same time, the visual perception gateway is immediately activated to collect hand gestures and gaze focus and perform an interaction intent recognition.

7. A blockchain-based smart campus settlement system according to claim 5, characterized in that, The criteria for judging network stability include: Retrieve network load data, including bandwidth, latency, and packet loss rate; convert the network load data into a scalar operator using a weight adjustment coefficient; wherein the weight adjustment coefficient is a dynamic value determined by the particle swarm optimization algorithm; If the scalar operator is less than the standard stability threshold, the current scalar operator is marked as 'a' and assigned the first level. The 'a' is then combined with the first level to generate network instability information. If the scalar operator is greater than or equal to the standard stability threshold, the current scalar operator is marked as 'b' and assigned the second level. The 'b' is then combined with the second level to generate network stability information.

8. A smart campus settlement system based on blockchain according to claim 1, characterized in that, Establish a mapping relationship between settlement intention characteristics and account status, including: Retrieve the current account status, and parse and output the first embedded identifier and the second embedded identifier according to the settlement intent characteristics; create a first time tag for the first embedded identifier according to the time series to form a first mapping relationship between the settlement intent characteristics and the account status; create a second time tag for the second embedded identifier according to the time series to form a second mapping relationship between the settlement intent characteristics and the account status.

9. A blockchain-based smart campus settlement system according to claim 8, characterized in that, The constraints must be satisfied, including: Based on the mapping relationship, the timing triggering features of the first time tag and the second time tag are extracted, including synchronous triggering, sequential triggering and cross triggering; Identify the current settlement scenario, retrieve the smart contract for the current settlement scenario from the blockchain, and the smart contract contains an association constraint graph; map the time-series triggering features onto the association constraint graph, and through structural isomorphic search, if at most one connected path is found, it is determined that the constraint range is met, and drive the smart contract to change the account status to unfrozen.

10. A method applied to a blockchain-based smart campus settlement system as described in any one of claims 1-9, characterized in that, The method includes: acquiring a dataset of multiple settlement scenarios in a smart campus; wherein the dataset includes settlement data packets and behavior trajectory sequences; The system processes settlement data packets using a pre-defined first machine model, performs cluster analysis based on on-chain historical settlement events, generates a first embedded identifier, processes behavioral trajectory sequences using a pre-defined second machine model, identifies user interaction with the target settlement area, and generates a second embedded identifier. Based on the retrieved network load data, the first and second embedded identifiers are processed to generate settlement intent features, and a mapping relationship between settlement intent features and account status is established. By utilizing the mapping relationship, the constraint range of each type of settlement scenario is obtained. If the constraint range is met, the account status is updated and uploaded to the blockchain simultaneously.

Images