A ball marker intelligent data management method integrated with bluetooth communication

Abstract

The application discloses a kind of integrated ball marker intelligent data management methods of bluetooth communication, it is related to data management technical field.The application is completed with the same group of equipment networking by BLE5.4 star type Mesh, and the main node is elected by device beacon frame and weighted scoring method, and the reliable transmission of networking is guaranteed by TDMA time slot allocation and AES-128 encryption;Intelligent ball marker is collected data by three-axis sensor, and five types of action recognition are completed by layered TCN-GRU end side lightweight AI algorithm and output structured multi-modal data;In turn, first-level source data filtering, second-level federal cross-checking, third-level global compliance checking are completed in ball marker local, mobile terminal main node, cloud server respectively;After checking, the least square multi-node cross algorithm is solved by UWB ranging and cross algorithm to solve the coordinates of falling ball;Cloud is based on compliance data matching and generates score card, and is issued to mobile terminal and ball marker in two-way, and archives full amount data.The application realizes the intelligent data management of golf ball marker whole process, and has wide application value.

Description

Technical Field

[0001] This invention relates to the field of data management technology, and more specifically to an intelligent data management method for a ball marker with integrated Bluetooth communication. Background Technology

[0002] With the rapid development of smart sports and the intelligent golf industry, the traditional manual scoring and recording management model of golf tournaments is gradually being replaced by intelligent data collection and automatic statistics solutions. Intelligent golf markers, motion sensing and monitoring, wireless data transmission, and automatic scoring technologies have become hot research topics in the industry. Currently, the field of intelligent golf monitoring widely utilizes sensor-based motion data collection, Bluetooth short-range communication networking, and cloud-based data statistics technologies. This enables basic functions such as shot recognition, ball landing position recording, and score statistics, effectively solving the problems of low efficiency, large errors, and poor real-time performance associated with manual recording. Currently, intelligent golf marker data management systems are widely used in daily golf training, amateur tournaments, and professional competitive events, gradually realizing the digital and visual management of golf sports data and providing technical support for the intelligent upgrade of golf. Under this industry trend, how to further improve the accuracy of golf marker motion data collection, optimize the stability of multi-device networking, and solve the problems of data verification distortion and insufficient fairness in tournament results has become a key research focus that those skilled in the art urgently need to overcome.

[0003] Chinese patent (publication number CN105980021A) discloses a golf accessory and a system using the golf accessory. This solution collects player hitting motion data by embedding a sensing module and a Bluetooth communication module in the golf ball and uploads it to a terminal device, realizing basic recognition of hitting motion and simple score statistics, replacing the traditional manual recording method. However, this solution only supports independent data collection from a single device and cannot achieve data interaction and collaborative verification among multiple ball devices in the same group. The single-point sensor is easily affected by the field environment and equipment vibration, resulting in a lot of invalid noise data and low motion recognition accuracy. Furthermore, there is no perfect data error correction mechanism, and the data reliability is poor.

[0004] Chinese patent (publication number CN210114793U) discloses an electronic golf ball marker. This device integrates acceleration sensing and Bluetooth transmission functions, which can collect real-time data on the golf ball's movement status, such as the hitting and placement, and upload the data to a back-end terminal via Bluetooth for data aggregation and storage, thus realizing intelligent monitoring of the ball marker's movement status. However, this device only has basic data collection and uploading functions, cannot identify abnormal data from a single device or logically contradictory data, and lacks a precise algorithm for calculating the ball's landing position, resulting in insufficient positioning accuracy and failing to meet the high-precision data statistics requirements of professional competitions.

[0005] Furthermore, existing technologies generally lack lightweight edge-side intelligent recognition algorithms, making it impossible to accurately distinguish multiple complex motion modes. At the same time, they lack a three-level hierarchical verification system at the edge, within the group, and in the cloud, resulting in low reliability of single-node data. Moreover, the lack of a multi-node cross-positioning algorithm leads to poor accuracy in calculating the ball's landing position, making it difficult to meet the high-precision and high-fairness data management needs of daily training and professional competitions.

[0006] In summary, existing intelligent data management technologies for golf balls are ill-suited to the intelligent, high-precision, and highly fair data management needs of large-scale, professional golf tournaments. Therefore, there is an urgent need to design a new intelligent data management method for golf balls that integrates Bluetooth communication to address the various technical pain points in existing technologies. Summary of the Invention

[0007] Based on the above-mentioned technical problems, this application discloses a smart data management method for a ball marker integrating Bluetooth communication, specifically including:

[0008] A BLE5.4 star-shaped Mesh network is built by connecting the smart ball markers and mobile terminals of players in the same group;

[0009] The smart ball marker collects motion data through a built-in three-axis accelerometer and uses a lightweight AI algorithm on the edge to identify actions such as placing the ball marker, picking up the ball marker, hitting the ball, wind-induced displacement, and accidental touches, thus obtaining multimodal motion data.

[0010] After performing layered verification on the multimodal motion data, including terminal-level data quality verification, group-level federated cross-verification, and cloud-level global consistency verification, UWB mutual ranging is performed between the balls in the same group. Combined with the mobile terminal coordinates, a multi-node cross-verification algorithm is used to calculate the landing position of each ball.

[0011] The mobile terminal uses the ball action recognition results and the ball landing position to count the number of shots and strokes per hole for each player and generate a real-time scorecard.

[0012] The mobile terminal uploads the verified action data and real-time scorecard to the cloud platform to complete the score archiving and data analysis.

[0013] Preferably, the construction of the BLE5.4 star mesh network specifically involves: after all smart beacons and mobile terminals are powered on, broadcasting a beacon frame containing the device ID, signal strength RSSI, and remaining battery power;

[0014] Devices in the same group discover each other via beacon frames and elect a temporary master node using a weighted scoring method.

[0015] The mobile terminal with the highest overall score is designated as the temporary master node, and the remaining mobile terminals are designated as backup nodes. Each smart bead is connected to the network, and a time division multiple access time slot synchronization mechanism is adopted to allocate an independent data transmission time slot to each smart bead connected to the network.

[0016] The master node generates a temporary network key and distributes it to all nodes. Data within the network is transmitted using AES-128 encryption.

[0017] When the master node goes offline, the standby node will elect a new master node to achieve a seamless network switch.

[0018] Preferably, the multimodal motion data specifically refers to: the smart ball marker collects motion data through a built-in three-axis accelerometer, and identifies the actions of placing the ball marker, picking up the ball marker, hitting the ball, wind displacement, and accidental touch based on a lightweight AI algorithm on the edge, thereby obtaining multimodal motion data, which includes structured fusion data of action type labels, action occurrence timestamps, raw three-axis motion data, action confidence parameters, and motion amplitude features, covering five full-scene motion modes: static placement, dynamic picking up, impact hitting, environmental wind disturbance, and human accidental touch.

[0019] Preferably, the lightweight AI algorithm on the edge side specifically refers to: a lightweight TCN-GRU temporal fusion algorithm is constructed to meet the requirements of low computing power, low power consumption and long temporal motion feature capture capability on the edge side, including a data preprocessing layer, a lightweight TCN dilated convolution feature layer, a GRU gated temporal feature layer, a feature fusion layer, and a classification output layer.

[0020] Among them, the data preprocessing layer performs noise reduction and normalization processing on the raw data collected by the triaxial accelerometer;

[0021] The lightweight TCN dilated convolution feature layer expands the receptive field through dilated convolution using a lightweight TCN temporal convolutional network, efficiently capturing long-term continuous motion features of acceleration data, and is suitable for long-term action states such as slow ball and object movement and continuous placement.

[0022] The GRU-gated timing feature layer employs a simplified dual-gated GRU loop unit to capture short-term, high-change instantaneous motion features such as hitting, picking up, and accidental touch. It filters out invalid jitter noise through a gating mechanism and retains effective change features.

[0023] The feature fusion layer combines the long-term steady-state feature vector output by the TCN dilated convolution with the short-term abrupt feature vector output by the GRU to obtain a multi-dimensional global motion feature vector that simultaneously contains steady-state and dynamic features.

[0024] The classification output layer performs global pooling to reduce the dimensionality of the fused global feature vector, compresses redundant feature parameters, and completes multi-class discrimination of five types of actions, namely placement, picking up, hitting, wind-blown displacement, and accidental touch, through the Softmax activation function, and outputs the confidence probability of each type of action.

[0025] A fixed confidence threshold is preset. When the output confidence probability of the target action exceeds the preset threshold, the action is determined to be validly triggered, and standardized, labeled multimodal action data is output to complete the edge-side intelligent action recognition.

[0026] Preferably, the layered verification specifically involves: the first-level data quality verification at the end side is performed on the local end side of the smart ball, which solves the problem of rampant invalid interference data during motion acquisition;

[0027] The secondary cross-verification of the same group of federated networks is performed on mobile terminals in the same group of the temporary master node, which solves the problem of unreliable location of single-node data.

[0028] The cloud-based three-level global consistency check is performed on the cloud server platform, resolving the fairness issue of logical inconsistencies in results during competitions.

[0029] Preferably, the end-side first-level data quality verification specifically involves: verifying whether the three-axis acceleration data of a single frame of motion data is complete, and removing missing or dropped frame data; and setting a reasonable acceleration amplitude range based on the physical characteristics of golf ball movement, and filtering out abrupt interference data that exceeds the range.

[0030] The variance of the sliding window is used to verify the stationarity of the data and filter out the interference data caused by accidental touches due to momentary jitter.

[0031] Output the valid action data after the first-level verification.

[0032] Preferably, the secondary cross-validation of the same group is specifically as follows: each ball in the same group uploads the valid action data and timestamp data after the primary validation to the master node in the group;

[0033] The main node constructs verification rules based on the logic of the golf game scenario to verify that the single shot action of the same player matches the unique ball marker picking and placing sequence logic, and that there are no sequence conflicts in the actions of players in the same group.

[0034] A federated cross-comparison algorithm is used to compare the action timestamps and action types of all balls in the group, and to eliminate erroneous data with temporal inconsistencies and abnormal scene logic.

[0035] Retain valid data for consistency matching within the group, complete secondary cross-validation, and output the standardized action dataset within the group.

[0036] Preferably, the cloud-based three-level global consistency verification specifically involves: addressing the local clock offset and sampling time difference issues that exist for different targets and different mobile terminals, using the cloud standard time as a reference, uniformly aligning and correcting all uploaded action timestamps, location timestamps, and score statistics timestamps;

[0037] The cloud platform builds a global verification rule base based on the official golf competition rules. It collects equipment data, action records, positioning data and score data from all groups in the entire event, establishes a global data association ledger, and performs two-way reconciliation of the player's full-process shot data, ball landing position data and action timing data. It verifies whether the number of actions matches the score statistics, whether the ball landing displacement corresponds to the shot action, whether the data timing chain is complete, and whether the player's actions meet the rule base requirements.

[0038] Based on the logical contradictions found during verification, the cloud platform combines historical equipment status data, site environment data, and multi-user data from the same session to trace the source and determine the cause. It distinguishes between three types of anomalies: equipment malfunction, environmental interference error, and deliberate human cheating, and processes the abnormal data according to the different types.

[0039] Preferably, the step of using a multi-node cross-validation algorithm to calculate the landing position of each ball marker specifically involves: establishing a Cartesian coordinate system for the field with the mobile terminal as the global coordinate origin, and obtaining UWB distance measurement data between each ball marker and the mobile terminal, and between each ball marker;

[0040] Multi-node ranging is performed, a position calculation model is constructed based on the least squares method, and the coordinate parameters are iteratively optimized to obtain the optimal planar coordinate parameters of each smart sphere.

[0041] Preferably, the generation of the real-time scorecard specifically involves: the cloud platform capturing standard multimodal motion data and high-precision ball landing position data after three-level verification, performing global temporal correlation matching on the ball picking action, hitting action, and placing action, and determining a single complete picking-hitting-placing action cycle that conforms to the logic of golf competition as a valid shot;

[0042] Using a single golf hole as an independent statistical unit, the number of valid shots per hole for each player is accumulated and converted into the precise number of strokes per hole;

[0043] It summarizes all hole stroke data and effective shot frequency data, automatically calculates the total score, and generates a standardized real-time scorecard that includes single hole strokes, total score, shot frequency, and shot landing point distribution;

[0044] The cloud will send the generated real-time scorecard data bidirectionally to the bound mobile terminal and the corresponding player's smart ball icon.

[0045] Compared with the prior art, the technical solution of this application has the following technical effects:

[0046] This invention constructs a star-shaped mesh adaptive networking system based on BLE5.4. It dynamically elects temporary master nodes through device beacon frame broadcasting and a weighted scoring mechanism of power and signal strength. Combined with a time-division multiple access time slot synchronization mechanism and an AES-128 encrypted transmission strategy, it also supports seamless network switching after master node failure. It effectively solves the problems of concurrent data transmission conflicts, packet loss, and network instability of multiple devices in the same group, and greatly improves the real-time transmission and data security of multi-beacon collaborative work. It is suitable for the batch networking needs of multiple players training and playing in the same field.

[0047] This invention is equipped with a lightweight TCN-GRU time-series fusion edge AI algorithm, which performs layered data preprocessing, dual extraction of steady-state and abrupt features, feature fusion and intelligent classification. It can accurately identify five types of motion modalities in all scenarios: placement, picking up, hitting, wind-induced displacement, and accidental touch. Under extremely low computing power and low power consumption edge hardware conditions, it effectively suppresses environmental noise and device jitter interference, and greatly improves the recognition accuracy and stability of multimodal motion actions.

[0048] This invention establishes a three-tiered, progressive data verification system at the terminal, group, and cloud levels. With clear division of labor and step-by-step error correction, the system filters invalid interference data at the source on the smart ball terminal, performs federated cross-verification of single-node data on the mobile main node, and completes global event logic compliance review in the cloud. This eliminates abnormal, contradictory, and false data from the source, local, and global links, completely solving the problems of unreliable single-node data and data distortion.

[0049] This invention integrates UWB mutual ranging and multi-node cross-validation algorithms, and uses the least squares method to iteratively optimize the optimal ball landing coordinates to achieve high-precision position calculation. It effectively corrects the ranging deviation caused by the field environment, significantly improves the accuracy of ball landing positioning, and provides accurate data support for stroke count statistics and landing point data analysis.

[0050] This invention completes the compliance verification of competition data through a cloud-based global rules engine. It can accurately distinguish between three abnormal scenarios: equipment failure, environmental interference, and human cheating, and handle them differently. It eliminates logical contradictions in scores and problems such as malicious stroke manipulation and data tampering, effectively ensuring the authenticity, fairness, and traceability of golf competition results, and is fully compatible with the stringent standards of professional competitive events.

[0051] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0052] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0054] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0055] Figure 1 A flowchart illustrating the overall process of intelligent data management for a ball marker integrating Bluetooth communication;

[0056] Figure 2 This is an architecture diagram of an intelligent data management method for a ball marker integrating Bluetooth communication;

[0057] Figure 3 This is an architecture diagram of the lightweight TCN-GRU time-series fusion algorithm in this application;

[0058] Figure 4 This is a diagram of the layered verification architecture in this application;

[0059] Figure 5 This is a schematic diagram of a BLE5.4 star mesh network in an embodiment of this application;

[0060] Figure 6 This is an action recognition confusion matrix diagram in the embodiments of this application;

[0061] Figure 7 This is a comparison chart of the three-level hierarchical data verification data in the embodiments of this application;

[0062] Figure 8 This is a comparison chart of the positioning error distribution in the embodiments of this application;

[0063] Figure 9 This is a heat map showing the distribution of ball landing points in an embodiment of this application. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0065] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0066] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0067] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0068] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0069] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0070] Example 1 describes a smart data management method for a ball marker integrating Bluetooth communication. Figure 1 , Figure 2 As shown, it specifically includes:

[0071] A BLE5.4 star-shaped Mesh network is built by connecting the smart ball markers and mobile terminals of players in the same group;

[0072] The smart ball marker collects motion data through a built-in three-axis accelerometer and uses a lightweight AI algorithm on the edge to identify actions such as placing the ball marker, picking up the ball marker, hitting the ball, wind-induced displacement, and accidental touches, thus obtaining multimodal motion data.

[0073] After performing layered verification on the multimodal motion data, including terminal-level data quality verification, group-level federated cross-verification, and cloud-level global consistency verification, UWB mutual ranging is performed between the balls in the same group. Combined with the mobile terminal coordinates, a multi-node cross-verification algorithm is used to calculate the landing position of each ball.

[0074] The mobile terminal uses the ball action recognition results and the ball landing position to count the number of shots and strokes per hole for each player and generate a real-time scorecard.

[0075] The mobile terminal uploads the verified action data and real-time scorecard to the cloud platform to complete the score archiving and data analysis.

[0076] Furthermore, the specific steps for building a BLE5.4 star-shaped Mesh network are as follows: after all smart beacons and mobile terminals are powered on, they broadcast a beacon frame containing the device ID, signal strength RSSI, and remaining battery power.

[0077] Devices in the same group discover each other via beacon frames and elect a temporary master node using a weighted scoring method, the formula of which is:

[0078]

[0079] in, The overall network score for the equipment is given. To provide the device with real-time signal strength, To achieve the maximum effective signal strength, This represents the device's current remaining battery power. This represents the device's full charge level. These are signal strength weight and power weight, respectively.

[0080] The mobile terminal with the highest overall score is designated as the temporary master node, and the remaining mobile terminals are designated as backup nodes. Each smart bead is connected to the network, and a time division multiple access time slot synchronization mechanism is adopted to allocate an independent data transmission time slot to each smart bead connected to the network.

[0081] The master node generates a temporary network key and distributes it to all nodes. Data within the network is transmitted using AES-128 encryption.

[0082] When the master node goes offline, the standby node will elect a new master node to achieve a seamless network switch.

[0083] Furthermore, the time-division multiple access time slot synchronization mechanism is as follows: the master node synchronizes the clocks of all network devices with a fixed broadcast period to eliminate clock offset errors of multiple devices; based on the access order of smart spheres within the network, each smart sphere is uniquely numbered, and each smart sphere is assigned a dedicated data transmission time slot that does not overlap with the other spheres. Only the spheres are allowed to upload motion data and status data in their own assigned time slots, while the other time slots remain in a dormant listening state to avoid data transmission conflicts between multiple spheres in the same group.

[0084] Furthermore, the multimodal motion data specifically refers to the following: The smart ball marker collects motion data through a built-in three-axis accelerometer and identifies actions such as placing the ball marker, picking up the ball marker, hitting the ball, wind-induced displacement, and accidental touch based on a lightweight AI algorithm on the edge side. This results in multimodal motion data, which includes structured fusion data such as action type labels, action occurrence timestamps, raw three-axis motion data, action confidence parameters, and motion amplitude features. It covers five full-scene motion modes: static placement, dynamic picking up, impact hitting, environmental wind disturbance, and human accidental touch.

[0085] Furthermore, the lightweight AI algorithm for the edge is specifically designed to meet the requirements of low computing power, low power consumption and long-term motion feature capture capabilities on the edge. It constructs a lightweight TCN-GRU temporal fusion algorithm, which includes a data preprocessing layer, a lightweight TCN dilated convolution feature layer, a GRU gated temporal feature layer, a feature fusion layer, and a classification output layer.

[0086] Among them, the data preprocessing layer performs noise reduction and normalization processing on the raw data collected by the triaxial accelerometer;

[0087] The lightweight TCN dilated convolution feature layer expands the receptive field through dilated convolution using a lightweight TCN temporal convolutional network, efficiently capturing long-term continuous motion features of acceleration data, and is suitable for long-term action states such as slow ball and object movement and continuous placement.

[0088] The GRU-gated timing feature layer employs a simplified dual-gated GRU cyclic unit to capture short-duration, high-change instantaneous motion features such as hitting, picking up, and accidental touches. It filters out invalid jitter noise through a gating mechanism, retaining valid change features. The GRU unit state update formula is as follows:

[0089]

[0090]

[0091] in, To update the door, To reset the door, In the candidate hidden state, Hide the current state. It is the sigmoid activation function. These are the weight matrix and the bias term, respectively. This is the element-wise multiplication operation of matrices. These are the learnable weight matrices corresponding to the update gate, reset gate, and candidate hidden states, respectively. These are the biases corresponding to the update gate, reset gate, and candidate hidden state, respectively. for The network input feature vector at time step;

[0092] The feature fusion layer combines the long-term steady-state feature vector output by the TCN dilated convolution with the short-term abrupt feature vector output by the GRU to obtain a multi-dimensional global motion feature vector that simultaneously contains steady-state and dynamic features.

[0093] The classification output layer performs global pooling to reduce the dimensionality of the fused global feature vector, compressing redundant feature parameters. It then uses the Softmax activation function to perform multi-class classification of five actions: placement, picking up, hitting, wind-induced displacement, and accidental touch, outputting the confidence probability of each action. The formula is as follows:

[0094]

[0095] in, For the first Confidence in action recognition The k-th class feature score output by the network. Total number of action categories;

[0096] A fixed confidence threshold is preset. When the output confidence probability of the target action exceeds the preset threshold, the action is determined to be validly triggered, and standardized, labeled multimodal action data is output to complete the edge-side intelligent action recognition.

[0097] Furthermore, such as Figure 3 The diagram shows the architecture of the lightweight TCN-GRU temporal fusion algorithm. The input dimension of the lightweight TCN-GRU temporal fusion algorithm is 3D three-axis acceleration temporal data, and the input size is set to [80,3], that is, a single inference window contains 80 sampling points, and each frame contains x / y / z three-axis data; the model output dimension is 5-dimensional, corresponding to the classification probabilities of five types of actions: placing the ball, picking up the ball, hitting the ball, wind-induced displacement, and accidental touch.

[0098] The data preprocessing layer uses a second-order Butterworth low-pass filter with a cutoff frequency of 20Hz to filter out high-frequency device noise; the unified normalization interval is [0,1], the sampling frequency is fixed at 100Hz, and the single-frame sampling duration is 10ms.

[0099] The lightweight TCN dilated convolution feature layer consists of two stacked one-dimensional dilated convolutions with a uniform kernel size of 3. The first layer has a dilation coefficient d=1, and the second layer has a dilation coefficient d=2, progressively expanding the temporal receptive field. The number of output channels is uniformly set to 16, and the bias term is not enabled. Each convolutional layer is followed by batch normalization (BN) and ReLU activation function with a dropout rate of 0.1 to suppress overfitting of small samples at the edge. This layer does not use pooling downsampling to preserve complete temporal motion details.

[0100] The GRU gated temporal feature layer uses a single-layer, unidirectional, lightweight GRU network with 32 hidden layer neurons to streamline redundant nodes in the traditional GRU. Only the update gate and reset gate are retained, and candidate memory units are discarded to reduce computational cost. The network input channels match the TCN output channels with 16, and the temporal iteration step size matches the input window length. The weight parameters are initialized using a normal distribution, and the bias term is initialized to 0.

[0101] The feature fusion layer adopts a channel-dimensional concatenation fusion method, concatenating the 16-dimensional steady-state feature vector output by TCN with the 32-dimensional mutation time-series feature vector output by GRU into a 48-dimensional fused feature vector; channel dimensionality reduction is performed through 1×1 lightweight convolution, compressing it into a 24-dimensional global feature vector;

[0102] The classification output layer uses global average pooling to compress and fuse feature dimensions, connecting to a single fully connected neural network. The fully connected layer has 24 input nodes and 5 output nodes. It is paired with the Softmax activation function to complete five classifications. The action recognition confidence threshold is preset to 0.85, and only actions with a confidence greater than 0.85 are judged as valid actions.

[0103] Furthermore, such as Figure 4 The architecture diagram of the layered verification shown is as follows: The first-level data quality verification at the end side is performed on the local end side of the smart ball, which solves the problem of the proliferation of invalid interference data during motion acquisition.

[0104] The secondary cross-verification of the same group of federated networks is performed on mobile terminals in the same group of the temporary master node, which solves the problem of unreliable location of single-node data.

[0105] The cloud-based three-level global consistency check is performed on the cloud server platform, resolving the fairness issue of logical inconsistencies in results during competitions.

[0106] Furthermore, the edge-side first-level data quality verification specifically involves: verifying the completeness of the three-axis acceleration data in a single frame of motion data, and removing missing or dropped data; based on the physical characteristics of golf ball motion, setting a reasonable acceleration amplitude range, and filtering out abrupt interference data exceeding the range, using the following formula:

[0107]

[0108] in, These are triaxial acceleration data. These are the preset minimum and maximum effective acceleration amplitudes, respectively.

[0109] The stationarity of the data is verified by using a sliding window variance test, and transient jitter-induced interference data is filtered out. The formula is as follows:

[0110]

[0111] in, The number of sampling points for the sliding window. For the first in the window One sampled data, This represents the average value of the data within the window. The variance is the data variance; if the variance exceeds a preset threshold, the data is considered invalid.

[0112] Output the valid action data after the first-level verification.

[0113] Furthermore, the specific details of the federal secondary cross-verification within the same group are as follows: each ball in the same group uploads the valid action data and timestamp data after the primary verification to the group's master node (mobile terminal).

[0114] The main node constructs verification rules based on the logic of the golf game scenario to verify that the single shot action of the same player matches the unique ball marker picking and placing sequence logic, and that there are no sequence conflicts in the actions of players in the same group.

[0115] A federated cross-comparison algorithm is used to compare the action timestamps and action types of all balls in the group, and to eliminate erroneous data with temporal inconsistencies and abnormal scene logic.

[0116] Retain valid data for consistency matching within the group, complete secondary cross-validation, and output the standardized action dataset within the group.

[0117] Furthermore, the verification rules based on the logic of the golf game scenario are as follows: the timing and behavior verification rules are constructed based on the golf single hole shot competition logic and the ball mark usage specifications. In a single shot process for the same player, only one complete timing chain of picking up the ball mark - hitting the ball - placing the ball mark is allowed. Abnormal behaviors such as hitting the ball before picking up the ball mark, hitting the ball multiple times without interval, ball displacement due to not picking up the ball mark, and ending the shot process without placing the ball mark are prohibited.

[0118] Furthermore, the cloud-based three-level global consistency verification specifically addresses the local clock offset and sampling time difference issues that exist for different targets and mobile terminals. It uses the cloud-based standard time as a benchmark to uniformly align and correct all uploaded action timestamps, location timestamps, and score statistics timestamps.

[0119] The cloud platform builds a global verification rule base based on the official golf competition rules. It collects equipment data, action records, positioning data and score data from all groups in the entire event, establishes a global data association ledger, and performs two-way reconciliation of the player's full-process shot data, ball landing position data and action timing data. It verifies whether the number of actions matches the score statistics, whether the ball landing displacement corresponds to the shot action, whether the data timing chain is complete, and whether the player's actions meet the rule base requirements.

[0120] Based on the logical contradictions found during verification, the cloud platform combines historical equipment status data, site environment data, and multi-user data from the same session to trace the source and determine the cause. It distinguishes between three types of anomalies: equipment malfunction, environmental interference error, and deliberate human cheating, and processes the abnormal data according to the different types.

[0121] Furthermore, the system corrects and compensates for equipment error data, reduces noise and re-evaluates environmental interference data, and directly resets invalid scores for cheating behaviors such as data tampering, false shots, and malicious strokes, while marking abnormal records to retain evidence for tracing the source.

[0122] After completing all global verification and anomaly correction, the cloud encrypts and solidifies the compliant data, generating tamper-proof standardized competition results data. This completely resolves issues such as multi-level data deviations, logical conflicts, and results dishonesty, providing an authoritative and reliable data foundation for subsequent results archiving, data statistics, and competition announcements.

[0123] Furthermore, the multi-node cross-validation algorithm is used to calculate the landing position of each ball marker. Specifically, a Cartesian coordinate system is established on the field plane with the mobile terminal as the global coordinate origin, and UWB distance measurement data between each ball marker and the mobile terminal, and between each ball marker, are obtained.

[0124] Multi-node ranging is performed, and a position calculation model is constructed based on the least squares method. The coordinate parameters are iteratively optimized to obtain the optimal planar coordinate parameters for each intelligent sphere. The formula is as follows:

[0125]

[0126]

[0127] in, The distance between the sphere and the reference node is measured by UWB. For the first Sphere coordinates, For reference node coordinates, It is a least squares iterative optimization function. This represents the number of nodes participating in cross-validation. Calculate the distance measurement value for the model.

[0128] Furthermore, the generation of real-time scorecards is specifically as follows: the cloud platform captures standard multimodal motion data and high-precision ball landing position data after three levels of verification, performs global temporal correlation matching on the ball picking action, hitting action, and placing action, and judges a single complete picking-hitting-placing action cycle that conforms to the logic of golf competition as a valid shot;

[0129] Using a single golf hole as an independent statistical unit, the number of valid shots per hole for each player is accumulated and converted into the precise number of strokes per hole;

[0130] It summarizes all hole stroke data and effective shot frequency data, automatically calculates the total score, and generates a standardized real-time scorecard that includes single hole strokes, total score, shot frequency, and shot landing point distribution;

[0131] The cloud will send the generated real-time scorecard data bidirectionally to the bound mobile terminal and the corresponding player's smart ball icon.

[0132] Furthermore, the cloud platform's data analysis includes player hitting habit analysis, court difficulty statistics, and sports data trend analysis, specifically:

[0133] The cloud platform performs batch analysis on archived performance data, ball placement data, and shot data, statistically analyzes the score deviation and ball deviation patterns of players on different holes, and generates individual shot habit reports; it aggregates multi-user course sports data and analyzes the shot difficulty coefficient of different holes; based on time-series data, it iteratively analyzes the changing trends of players' stage-by-stage sports performance, and outputs standardized data analysis reports to provide data support for player training optimization and course operation management.

[0134] Furthermore, when the distance between the ball and the bound mobile terminal exceeds a preset threshold, an anti-loss alarm is triggered. The lost ball is then retrieved remotely by uploading its location information through other users' mobile terminals. Specifically:

[0135] Real-time monitoring of the distance between the ball and the bound mobile terminal based on the BLE5.4 Bluetooth ranging algorithm;

[0136] When the monitoring interval D is continuously greater than the preset distance threshold and the duration exceeds the preset time threshold, the terminal triggers a dual anti-loss alarm with vibration and sound and light.

[0137] The sphere continuously broadcasts Bluetooth identification information. After other users' mobile terminals scan and identify it, the sphere's real-time location and scanning time are automatically uploaded to the cloud platform. The cloud platform aggregates the location data from multiple terminals, calculates the sphere's precise location, and pushes it to the bound user's terminal, enabling long-distance retrieval.

[0138] This embodiment details a smart data management method for ball markers integrating Bluetooth communication. It establishes a network of devices within the same group using a BLE5.4 star mesh, elects a master node using device beacon frames and a weighted scoring method, and ensures reliable network transmission through TDMA time slot allocation and AES-128 encryption. The smart ball marker collects data via a triaxial sensor and uses a layered TCN-GRU lightweight AI algorithm to recognize five types of actions and output structured multimodal data. First-level source data filtering, second-level federated cross-validation, and third-level global compliance verification are sequentially performed on the ball marker, the mobile master node, and the cloud server. After verification, the ball's coordinates are calculated using UWB ranging and a least-squares multi-node cross-validation algorithm. The cloud server matches the compliance data with shot logic, calculates the number of strokes, generates a scorecard, and sends it bidirectionally to the mobile device and the ball marker, while also archiving all data.

[0139] Example 2, based on Example 1, describes in detail the process of using this method for intelligent data management of ball markers on a standard 18-hole amateur golf course with four players, as follows:

[0140] The venue is equipped with 4 smart ball markers, 4 smartphones (mobile terminals) and 1 cloud server. The smart ball markers have built-in Nordic-nRF52840-BLE5.4 modules, DW1000-UWB ranging modules and ADXL345 triaxial accelerometers with a sampling frequency of 100Hz.

[0141] After all smart spheres and mobile terminals have completed power-on initialization, they broadcast beacon frames containing device ID, RSSI, and remaining battery power at 100ms intervals. Devices in the same group scan and parse the beacon frames, then use a weighted scoring method. The overall score of each device is calculated, and Player A's mobile phone score is the highest (S=0.92), so it is selected as the temporary master node, and the other 3 mobile phones are backup nodes. The master node is assigned the numbers 1-4 to the 4 smart balls, corresponding to time slots 0-3, with each time slot lasting 20ms. Data within the network is transmitted using AES-128 encryption. If the master node goes offline, the backup node can be re-elected within 500ms, achieving seamless network switching.

[0142] like Figure 5The BLE5.4 star mesh network diagram shown depicts a star topology layout. A temporary master node (smartphone) icon is drawn at the center, surrounded by three backup mobile terminal icons and four smart beacon icons. Solid lines represent BLE5.4 wireless communication links, with the master node establishing point-to-point connections with all backup terminals and all smart beacons. There are no direct communication lines between backup terminals and smart beacons. A simplified diagram of a cloud server is included in the corners of the diagram. The master node connects to the cloud via a wireless link. The legend labels four categories: master node, backup terminals, smart beacons, and wireless communication links, visually illustrating the star network topology of this solution. This demonstrates a clear network architecture, proving that multi-device collaborative networking can be quickly achieved by coordinating communication across all devices through the central master node.

[0143] The smart ball collected motion data at a frequency of 100Hz using a triaxial accelerometer. After second-order Butterworth low-pass filtering (cutoff frequency 20Hz) and normalization, the data was input into a lightweight TCN-GRU temporal fusion algorithm. The algorithm extracted long-term steady-state features through two layers of dilated convolution (d=1, d=2) and extracted short-term abrupt features through a single-layer 32-neuron GRU. After feature fusion, the algorithm output the action confidence score through Softmax classification. A total of 1236 valid action data points were collected in this competition. The accuracy rates of various action recognition types are shown in Table 1 below.

[0144] Table 1 Accuracy of various action recognition rates

[0145] Action type Number of samples collected Correctly identify numbers Recognition accuracy Place ball marker 312 309 99.0% Pick up the ball 312 308 98.7% Hitting the ball 312 310 99.4% Wind-blown displacement 156 152 97.4% Accidental touch 144 141 97.9%

[0146] According to Table 1 and Figure 6 As shown in the motion recognition confusion matrix diagram, this is a 5×5 square matrix. The horizontal axis represents the predicted motion type, and the vertical axis represents the actual motion type. The rows and columns are arranged in order for five categories: placing a ball, picking up a ball, hitting the ball, being moved by the wind, and accidental touch. The number of recognized samples for each category is entered in the matrix cells. The color intensity of the cell corresponds to the sample size. The diagonal line represents correctly recognized samples, and the off-diagonal line represents misclassified samples. There are 312 actual samples for placing a ball, with 309 correctly recognized and only 3 misclassified to other categories. There are 312 actual samples for picking up a ball, with 308 correctly recognized and 4 misclassified cases. There are 312 samples for hitting the ball, with only 2 misclassified and 310 correctly recognized. There are 156 samples for being moved by the wind, with 152 correctly recognized and the remaining 4 confused. There are 144 samples for accidental touch, with 141 correctly recognized, resulting in the fewest misclassifications. The values ​​of the diagonal cells are generally much higher than those of the other cells in the same row and column, which intuitively shows the distribution of recognition errors for the five types of motions, proving that the algorithm of this invention can accurately distinguish multiple complex motion modes.

[0147] First-level data quality verification on the device side was conducted locally on the smart ball, verifying the integrity of single-frame data, acceleration amplitude (Amin=0.1g, Amax=20g), and sliding window variance (window size 10, threshold 0.5). A total of 153 invalid and interfering data entries were removed, accounting for 12.3% of the total collected data. Second-level federated cross-verification within the same group was performed on the mobile master node. Based on the logic of the golf game scene, the master node verified the consistency of the timing of the actions of players in the same group, and used a federated cross-comparison algorithm (Scoreth=0.7) to compare the timestamps of each ball action. Based on the data type, 39 data entries with temporal inconsistencies were removed, accounting for 3.1% of the data after the first-level verification. The cloud-based three-level global consistency verification was deployed on the cloud server. The cloud aligned all data based on the standard time, performed a global logical review based on the competition rule base, and established a global data association ledger for two-way reconciliation. Two abnormal data entries were found. One was a temporal disorder caused by equipment failure, which was corrected and retained. The other was a logical inconsistency caused by a player maliciously swiping their score, which was directly cleared and the evidence was preserved. The three-level hierarchical data verification data diagram is shown in Table 2 below:

[0148] Table 2. Data Validation Chart for Three-Level Hierarchical Data

[0149] Verification phase Input data volume Remove outlier data Data accuracy Raw collected data 1401 items - 84.2% After end-side first-level verification 1248 153 articles 92.5% After secondary verification in the same group 1209 items 39 articles 98.1% After cloud-based three-level verification 1207 items 2 99.8%

[0150] According to Table 2 and Figure 7 As shown in the comparison chart of the three-level hierarchical data verification, the data accuracy gradually increases from 84.2% of the original data to 99.8% after the three-level verification. This intuitively presents the trend of data accuracy gradually increasing with the verification level, which is consistent with the data in the table above, proving that the verification system of the present invention can significantly improve data reliability.

[0151] A Cartesian coordinate system was established on the field with the temporary master node as the origin. UWB distance measurements were performed between each smart ball and the master node, and between the balls themselves, acquiring multiple sets of distance measurement data. A position calculation model was constructed based on the least squares method, and the coordinate parameters were iteratively optimized to obtain the optimal ball landing coordinates for each ball. A total of 312 ball landing positioning tests were conducted in this competition. The positioning accuracy of this invention compared with existing mainstream dynamic base station positioning methods is shown in Table 3 below.

[0152] Table 3. Comparison of positioning accuracy between the present invention and existing mainstream dynamic base station positioning methods.

[0153] Technical solution Average positioning error Maximum positioning error 95% confidence interval error This invention 15.2mm 28.7mm ≤25.7mm Existing technology (dynamic base station positioning method) 52.6mm 89.3mm ≤70.4mm

[0154] According to Table 3 and Figure 8As shown in the positioning error distribution comparison chart, the errors of the present invention are mainly concentrated in the 10-25mm range, with concentrated probability peaks and a narrow distribution range, indicating that most positioning results can be stably maintained within a small error range. In contrast, the errors of the prior art are mainly concentrated in the 40-70mm range, with a more dispersed probability distribution, and the high error range still has a high probability of occurrence, indicating that its positioning results fluctuate greatly and have poor stability. In addition, the present invention has covered about 95% of the positioning samples within 25.7mm, while the prior art only covers a small number of samples within the same error range. This further proves that the present invention, through UWB collaborative ranging and least squares iterative optimization positioning mechanism, can significantly reduce positioning errors and improve positioning consistency, making the positioning results more concentrated and reliable. The overall positioning accuracy is significantly better than the existing dynamic base station positioning method, which can meet the application requirements for high-precision positioning of the ball landing point in golf tournament scenarios.

[0155] The cloud-based system captures compliant data that has undergone three levels of verification. A complete cycle of picking up the ball, hitting it, and placing it is considered a valid shot. The system accumulates the strokes of each player on each hole and generates a real-time scorecard that includes the strokes on each hole, total strokes, shot frequency, and landing point distribution. This scorecard is then distributed bidirectionally to mobile devices and smart ball markers. Simultaneously, the cloud analyzes all the data to generate individual player shot habits reports and course difficulty statistics reports. Player B's average stroke on hole 3 is 1.2 strokes higher than other players, and the highest average stroke on hole 12 (4.8 strokes) is achieved. This provides data support for player training and course operation.

[0156] like Figure 9 The heat map showing the ball landing point distribution, with a standard golf course layout as the background, clearly marks the tee box, fairway, green, bunker, and water hazard areas. The heat map uses a color gradient of red, orange, yellow, green, and blue to represent the landing point density, with red areas indicating the most concentrated landing points and blue areas indicating the sparsest landing points. The fairway area 100 meters in front of the green shows obvious red hotspots, while the bunker area on the right side of the green shows orange secondary hotspots. This map proves that the present invention can achieve accurate landing point data analysis.

[0157] This embodiment details the process of using this method for intelligent data management of golf markers on a standard 18-hole amateur golf tournament course for four players. During the process, the network setup time is less than 2 seconds, the data transmission packet loss rate is less than 0.5%, the motion recognition accuracy reaches 98.7%, and the data accuracy is improved to 99.8% after three levels of verification. The average positioning error is 15.2cm, and the score statistics error is 0, effectively preventing human cheating and ensuring the fairness of the competition. The implementation results verify the feasibility and superiority of the technical solution of this invention, realizing intelligent data management of golf markers throughout the entire process, and fully adapting to the needs of daily training and professional competitions.

[0158] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A method for intelligent data management of a ball marker integrating Bluetooth communication, characterized in that, include: A BLE5.4 star-shaped Mesh network is built by connecting the smart ball markers and mobile terminals of players in the same group; The smart ball marker collects motion data through a built-in three-axis accelerometer and uses a lightweight AI algorithm on the edge to identify actions such as placing the ball marker, picking up the ball marker, hitting the ball, wind-induced displacement, and accidental touches, thus obtaining multimodal motion data. After performing layered verification on the multimodal motion data, including terminal-level data quality verification, group-level federated cross-verification, and cloud-level global consistency verification, UWB mutual ranging is performed between the balls in the same group. Combined with the mobile terminal coordinates, a multi-node cross-verification algorithm is used to calculate the landing position of each ball. The mobile terminal uses the ball action recognition results and the ball landing position to count the number of shots and strokes per hole for each player and generate a real-time scorecard. The mobile terminal uploads the verified action data and real-time scorecard to the cloud platform to complete the score archiving and data analysis.

2. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 1, characterized in that, The construction of the BLE5.4 star mesh network specifically involves: after all smart spheres and mobile terminals are powered on, they broadcast beacon frames containing device ID, signal strength RSSI, and remaining battery power; Devices in the same group discover each other via beacon frames and elect a temporary master node using a weighted scoring method. The mobile terminal with the highest overall score is designated as the temporary master node, and the remaining mobile terminals are designated as backup nodes. Each smart bead is connected to the network, and a time division multiple access time slot synchronization mechanism is adopted to allocate an independent data transmission time slot to each smart bead connected to the network. The master node generates a temporary network key and distributes it to all nodes. Data within the network is transmitted using AES-128 encryption. When the master node goes offline, the standby node will elect a new master node to achieve a seamless network switch.

3. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 1, characterized in that, The multimodal motion data specifically refers to the following: The smart ball marker collects motion data through a built-in three-axis accelerometer and identifies actions such as placing the ball marker, picking up the ball marker, hitting the ball, wind-induced displacement, and accidental touch based on a lightweight AI algorithm on the edge side. This results in multimodal motion data, which includes structured fusion data such as action type labels, action occurrence timestamps, raw three-axis motion data, action confidence parameters, and motion amplitude features. It covers five full-scene motion modes: static placement, dynamic picking up, impact hitting, environmental wind disturbance, and human accidental touch.

4. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 3, characterized in that, The lightweight AI algorithm for the edge is specifically designed to meet the requirements of low computing power, low power consumption and long-term motion feature capture capabilities on the edge. It constructs a lightweight TCN-GRU temporal fusion algorithm, which includes a data preprocessing layer, a lightweight TCN dilated convolution feature layer, a GRU gated temporal feature layer, a feature fusion layer, and a classification output layer. Among them, the data preprocessing layer performs noise reduction and normalization processing on the raw data collected by the triaxial accelerometer; The lightweight TCN dilated convolution feature layer expands the receptive field through dilated convolution using a lightweight TCN temporal convolutional network, efficiently capturing long-term continuous motion features of acceleration data, and is suitable for long-term action states such as slow ball and object movement and continuous placement. The GRU-gated timing feature layer employs a simplified dual-gated GRU loop unit to capture short-term, high-change instantaneous motion features such as hitting, picking up, and accidental touch. It filters out invalid jitter noise through a gating mechanism and retains effective change features. The feature fusion layer combines the long-term steady-state feature vector output by the TCN dilated convolution with the short-term abrupt feature vector output by the GRU to obtain a multi-dimensional global motion feature vector that simultaneously contains steady-state and dynamic features. The classification output layer performs global pooling to reduce the dimensionality of the fused global feature vector, compresses redundant feature parameters, and completes multi-class discrimination of five types of actions, namely placement, picking up, hitting, wind-blown displacement, and accidental touch, through the Softmax activation function, and outputs the confidence probability of each type of action. A fixed confidence threshold is preset. When the output confidence probability of the target action exceeds the preset threshold, the action is determined to be validly triggered, and standardized, labeled multimodal action data is output to complete the edge-side intelligent action recognition.

5. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 1, characterized in that, The layered verification specifically refers to: the first-level data quality verification at the end side is performed at the local end side of the smart ball, which solves the problem of the proliferation of invalid interference data during motion acquisition; The secondary cross-verification of the same group of federated networks is performed on mobile terminals in the same group of the temporary master node, which solves the problem of unreliable location of single-node data. The cloud-based three-level global consistency check is performed on the cloud server platform, resolving the fairness issue of logical inconsistencies in results during competitions.

6. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 5, characterized in that, The edge-side first-level data quality verification specifically involves: verifying whether the three-axis acceleration data of a single frame of motion data is complete, and removing missing or dropped frame data; and setting a reasonable acceleration amplitude range based on the physical characteristics of golf ball movement, and filtering out abrupt interference data that exceeds the range. The variance of the sliding window is used to verify the stationarity of the data and filter out the interference data caused by accidental touches due to momentary jitter. Output the valid action data after the first-level verification.

7. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 6, characterized in that, The aforementioned group-based federal secondary cross-validation specifically involves each ball in the same group uploading its valid action data and timestamp data after primary validation to the group's master node. The main node constructs verification rules based on the logic of the golf game scenario to verify that the single shot action of the same player matches the unique ball marker picking and placing sequence logic, and that there are no sequence conflicts in the actions of players in the same group. A federated cross-comparison algorithm is used to compare the action timestamps and action types of all balls in the group, and to eliminate erroneous data with temporal inconsistencies and abnormal scene logic. Retain valid data for consistency matching within the group, complete secondary cross-validation, and output the standardized action dataset within the group.

8. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 7, characterized in that, The cloud-based three-level global consistency verification specifically addresses the local clock offset and sampling time difference issues that exist for different targets and mobile terminals. It uses the cloud standard time as a reference to uniformly align and correct all uploaded action timestamps, location timestamps, and score statistics timestamps. The cloud platform builds a global verification rule base based on the official golf competition rules. It collects equipment data, action records, positioning data and score data from all groups in the entire event, establishes a global data association ledger, and performs two-way reconciliation of the player's full-process shot data, ball landing position data and action timing data. It verifies whether the number of actions matches the score statistics, whether the ball landing displacement corresponds to the shot action, whether the data timing chain is complete, and whether the player's actions meet the rule base. Based on the logical contradictions found during verification, the cloud platform combines historical equipment status data, site environment data, and multi-user data from the same session to trace the source and determine the cause. It distinguishes between three types of anomalies: equipment malfunction, environmental interference error, and deliberate human cheating, and processes the abnormal data according to the different types.

9. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 8, characterized in that, The multi-node cross-validation algorithm is used to calculate the landing position of each ball marker. Specifically, a Cartesian coordinate system is established on the field plane with the mobile terminal as the global coordinate origin, and UWB distance measurement data between each ball marker and the mobile terminal, and between each ball marker, are obtained. Multi-node ranging is performed, a position calculation model is constructed based on the least squares method, and the coordinate parameters are iteratively optimized to obtain the optimal planar coordinate parameters of each smart sphere.

10. The intelligent data management method for a ball marker integrating Bluetooth communication according to claim 1, characterized in that, The generation of the real-time scorecard is specifically as follows: the cloud platform captures standard multimodal action data and high-precision ball landing position data after three levels of verification, performs global temporal correlation matching on the ball picking action, hitting action, and placing action, and determines a single complete picking-hitting-placing action cycle that conforms to the logic of golf competition as a valid shot; Using a single golf hole as an independent statistical unit, the number of valid shots per hole for each player is accumulated and converted into the precise number of strokes per hole; It summarizes all hole stroke data and effective shot frequency data, automatically calculates the total score, and generates a standardized real-time scorecard that includes single hole strokes, total score, shot frequency, and shot landing point distribution; The cloud will send the generated real-time scorecard data bidirectionally to the bound mobile terminal and the corresponding player's smart ball icon.

Images