Fencing system and information identification method for fencing system

Abstract

The embodiment of the invention provides a fencing system and an information identification method for the fencing system, and the fencing system comprises an information collection module which is used for obtaining the impact information of fencing, generating the impact information when the fencing is hit by the fencing, and transmitting the impact information to a data processing module; the data processing module is used for receiving the hit information, extracting feature information in the hit information and counting the feature information and the number of hit times, and the feature information comprises hit time, hit position and hit strength; and the data analysis module is used for analyzing the hit information so as to identify and classify hit categories of the hit information, and the hit categories comprise puncture, miss and whipping. By providing the real-time hit position, time and strength and classifying the hit categories, the problems of high subjectivity and lack of objective data basis in the judgment of the hit behavior in the fencing match are solved, and the method is helpful for assisting in match penalty.

Description

Technical Field

[0001] The present invention relates to the technical field of sports equipment, and particularly to a fencing system and an information recognition method for a fencing system. Background Art

[0002] Existing fencing referee systems mainly rely on spring sword tips to sense pressure and electrical signal transmission technology. The spring sword tips need to be configured with dedicated lines, have a complex structure and rely on fixed external forces to trigger, and cannot be compatible with the different requirements of épée and foil. Moreover, the existing fencing competition penalties mainly rely on the naked eye observation of referees, which is easily affected by the perspective limitation and human error. Especially in high-speed confrontations, it is difficult to accurately determine the effectiveness of hitting actions (such as stabbing force, hitting position). Wireless technology is easily blocked by athletes' bodies or interfered by the environment, resulting in delays and misjudgments, affecting the accuracy of referees. In addition, during daily training, fencing equipment cannot accurately determine and record the hitting position and method. Athletes lack real-time and multi-dimensional biomechanical feedback (such as hitting force distribution, action coherence) during training, making it difficult to quantitatively analyze technical defects and unable to help athletes train well.

[0003] In addition, in current fencing competitions, high-speed camera video playback technology is usually used to judge the sequence and effectiveness of hits. This method is inefficient and easily affected subjectively. Moreover, during daily training, it cannot reflect the training effect in real time and accurately, and it is difficult to formulate training plans according to the individual differences of athletes. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a fencing system and an information recognition method for a fencing system, which are used to solve the problem of strong subjectivity and lack of objective data basis when using the existing technology to judge hitting behaviors in fencing competitions.

[0005] To achieve the above purpose, the embodiments of the present invention provide a fencing system, which includes: an information collection module, configured to obtain the hitting information of fencing, and generate a hitting information and transmit it to a data processing module when being hit by the fencing; a data processing module, configured to receive the hitting information, extract the feature information in the hitting information, and count the feature information and the number of hits, where the feature information includes hitting time, hitting position, and hitting force; and a data analysis module, configured to analyze the hitting information to identify and classify the hitting category of the hitting information, where the hitting category includes stabbing, missing, and whipping.

[0006] Optionally, the system further includes a data visualization module, which is configured to convert the hit information into an electrical signal map, and generate a corresponding heat map according to the statistically obtained characteristic information and the number of hits at the hit position; when the number of hits at the hit position is greater than or equal to a preset number, the hit position is presented in a first color on the heat map; when the number of hits at the hit position is less than the preset number, the hit position is presented in a second color on the heat map; and the positions not hit by the fencing are presented in a third color on the heat map.

[0007] Optionally, the fencing system further includes a competition and training module, which calculates a score result according to the characteristic information and the hit category according to a preset scoring rule; calculates the hit percentage of each category according to the hit category; and displays the heat map, the score result, the hit percentage, and the characteristic information.

[0008] Optionally, the information collection module includes triboelectric pressure sensors, which are uniformly arranged in an array manner inside the fencing uniform. The triboelectric pressure sensors include: a first support material layer; a positive triboelectric material layer, which is arranged on the first support material layer; a second support material layer; a negative triboelectric material layer, which is arranged on the second support material layer and is arranged opposite to the positive triboelectric material layer, and is configured to contact the positive triboelectric material layer when the tip or the blade of the sword hits the triboelectric pressure sensor, and separate from the positive triboelectric material layer when the tip or the blade of the sword leaves the triboelectric pressure sensor; an induction electrode, which is electrically connected to the negative triboelectric material layer; and an output electrode, which is electrically connected to the induction electrode and is configured to output the electrical signal generated by the contact between the negative triboelectric material layer and the positive triboelectric material layer.

[0009] Optionally, the multiple triboelectric pressure sensors arranged inside the fencing uniform collect the hit information in a real-time synchronous acquisition manner.

[0010] Optionally, the data analysis module analyzes the hit information based on a deep learning model. The construction process of the deep learning model includes: obtaining hit information with known hit categories, and dividing the hit information into a training set, a validation set, and a test set; using a standardization method to perform data normalization on the hit information, and converting the processed hit information into 2D matrix data composed of the number of channels and time steps; after initializing the deep learning model, setting a loss function and an optimizer, and presetting the number of iteration rounds, inputting the training set into the deep learning model for batch training, and performing data augmentation in the way of randomly shuffling the data order; after each round of training, using the validation set to calculate the loss value and accuracy of the model. When the preset number of iteration rounds is reached, if the loss value reaches the preset loss threshold, then save the model weights at this time as the trained model weights, and the training is completed. If the loss value does not reach the preset loss threshold, then continue the training; and inputting the test set into the trained model to output the category prediction score, and calculating the accuracy of the predicted category. When the accuracy reaches the preset accuracy, the model construction is completed. If the accuracy does not reach the preset accuracy, then retrain.

[0011] Optionally, the deep learning model includes: four 1D convolutional layers, a RELU activation function layer, a global average pooling layer, and four fully connected layers. The hit information is input into the deep learning model, and the temporal feature information is extracted through the four 1D convolutional layers, and the number of channels of the feature information is expanded; the RELU activation function layer is used to enhance the feature expression of the feature information, and the enhanced feature information is input into the global average pooling layer to compress its time dimension; and the compressed feature information is input into the fully connected layer, and the dimension is gradually reduced to three dimensions and then output, and the output is the prediction scores of three types of signals: stabbing, missing, and whipping.

[0012] Optionally, the data analysis module is further configured to identify valid hit signals in the hit information according to different competition types.

[0013] On the other hand, the present invention provides an information recognition method for a fencing system. The method includes: obtaining the touch information of fencing, and generating hit information when being hit by the fencing; extracting the feature information in the hit information, and counting the feature information and the number of hits. The feature information includes the hit time, hit position, and hit strength; and analyzing the hit to identify the hit category of the hit information and perform classification, where the hit category includes stabbing, missing, and whipping.

[0014] Optionally, the method further includes: converting the hit information into an electro-signal map, and generating a corresponding heat map according to the statistically obtained feature information and the number of hits, based on the hit position; when the number of hits at the hit position is greater than or equal to a preset number, the hit position is presented in a first color on the heat map, when the number of hits at the hit position is less than the preset number, the hit position is presented in a second color on the heat map, and the position not hit by the fencing is presented in a third color on the heat map.

[0015] Through the above technical solution, the present invention obtains the hit information of fencing through the information collection module, and uses the data processing module to extract the feature information that can determine the hit time, hit position, and hit strength in the hit information, so as to judge the sequence of hits according to the hit time, and determine the validity of the hit based on the hit position and hit strength according to the competition rules. The classification of the hit information is realized through the data analysis module, so that the present invention can judge whether the hit is a valid signal according to different competition types, thereby providing an objective data basis for fencing competitions and assisting in competition penalties.

[0016] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification, and are used to explain the embodiments of the present invention together with the following specific implementation manners, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0018] Figure 1 is a schematic structural diagram of a fencing system provided by an embodiment of the present invention;

[0019] Figure 2 is a schematic structural diagram of a triboelectric pressure sensor provided by an embodiment of the present invention;

[0020] Figure 3 is a schematic structural diagram of a deep learning model provided by an embodiment of the present invention;

[0021] Figure 4 is a schematic diagram of the electro-signal generated by the triboelectric pressure sensor when a fencing hit occurs provided by an embodiment of the present invention;

[0022] Figure 5 is a front view of the triboelectric pressure sensor array arranged on the fencing uniform provided by an embodiment of the present invention;

[0023] Figure 6 is a schematic flowchart of the deep learning model construction process provided by an embodiment of the present invention;

[0024] Figure 7 It is a side view of a hit in the fencing stabbing method provided by an embodiment of the present invention;

[0025] Figure 8 It is a side view of a hit in the fencing whipping method provided by an embodiment of the present invention.

[0026] Explanation of reference numerals

[0027] 1. Positive triboelectric material layer; 2. Negative triboelectric material layer; 3. Inductive electrode; 4. Output electrode; 5. First support material layer; 6. Second support material layer; 7. Fencing; 8. Triboelectric pressure sensor; 9. Fencing uniform. Detailed implementation manners

[0028] The following will describe in detail the specific implementation manners of the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific implementation manners described herein are only for explaining and illustrating the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0029] It should be noted that the acquisition, transmission, storage, use, processing, etc. of data in the technical solution of this application all comply with the relevant regulations of laws and regulations. In the embodiments of this application, some industry-existing solutions such as certain software, components, models, etc. may be mentioned. They should be regarded as exemplary, and their purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.

[0030] Figure 2 It is a schematic structural diagram of a triboelectric pressure sensor provided by an embodiment of the present invention. As Figure 2 shown, the triboelectric pressure sensor is a layered structure, including a first support material layer 5; a positive triboelectric material layer 1, which is disposed on the first support material layer 5; a second support material layer 6; a negative triboelectric material layer 2, which is disposed on the second support material layer 6 and is disposed opposite to the positive triboelectric material layer 1, and is used to contact the positive triboelectric material layer 1 when the tip or the body of the sword hits the triboelectric pressure sensor, and separate from the positive triboelectric material layer 1 when the tip or the body of the sword leaves the triboelectric pressure sensor; an inductive electrode 3, electrically connected to the negative triboelectric material layer 2; and an output electrode 4, electrically connected to the inductive electrode 3, and is used to output the electrical signal generated by the contact between the negative triboelectric material layer 2 and the positive triboelectric material layer 1.

[0031] Specifically, the triboelectric pressure sensors are uniformly arranged in the fencing uniform in an array arrangement manner, and its front view is as Figure 5 shown. In some embodiments, the multiple triboelectric pressure sensors disposed in the fencing uniform collect the hitting information in a real-time synchronous acquisition manner.

[0032] It can be understood that when the tip or the body of the fencing touches the fencing uniform, the positive triboelectric material layer and the negative triboelectric material layer of the sensor corresponding to the hit position come into contact and separation. Since the charges carried by the two layers of triboelectric materials are different, a very high potential difference will be generated, driving the electrons in the external circuit to move directionally, thereby generating an electrical signal. In this way, weak mechanical energy can be collected and converted into electrical energy, and this method can achieve self-driving without an external power supply.

[0033] In some embodiments, the positive triboelectric material can be dielectric materials such as PC and Nylon, the negative triboelectric material can be dielectric materials such as FEP, PTFE, and PVDF, the induction electrode 3 can be elastic conductive materials such as conductive sponge and conductive aerogel, the output electrode 4 can be conductive films such as copper and aluminum, and the support material 5 can be elastomers such as polyurethane sponge and EVA sponge.

[0034] Figure 1 It is a schematic structural diagram of a fencing system provided by an embodiment of the present invention. As Figure 1 shown, the fencing system includes: an information collection module, configured to obtain the touch information of the fencing, and when being hit by the fencing, generate a hit information and transmit it to the data processing module; a data processing module, configured to receive the hit information, extract the feature information in the hit information, and perform statistics on the feature information and the number of hits, where the feature information includes the hit time, the hit position, and the hit strength; and a data analysis module, configured to analyze the hit information to identify and classify the hit category of the hit information, where the hit category includes stabbing, missing, and whipping.

[0035] In some embodiments, the information collection module includes a triboelectric pressure sensor. When the fencing hits the fencing uniform, the triboelectric pressure sensor converts the touch of the fencing into an electrical signal, and this electrical signal is transmitted to the data processing module. Figure 4 It is a schematic diagram of the electrical signal generated by the triboelectric pressure sensor when the fencing is hit provided by an embodiment of the present invention. As Figure 4 can be seen, the characteristics of the electrical signals generated by different hitting methods are different. The characteristic of the stabbing signal is that a single sine wave signal is stronger, the characteristic of the missing signal is that the output signal is very weak, only different from the standby signal, and the characteristic of the whipping signal is a combined signal composed of multiple unit outputs within a short time. From this, it can also be inferred the hitting path of the fencing according to the different signal generation times. Figure 7 It is a side view of the fencing stabbing method hitting provided by an embodiment of the present invention. When the fencing 7 hits the triboelectric pressure sensor 8 under the fencing uniform 9 alone, a peak output electrical signal is generated. The system provided by the present invention collects the electrical signal of the triboelectric pressure sensor 8 and makes a classification by analyzing the characteristics of its electrical signal. Figure 8It is a side view of the hit in the fencing whipping method provided by the embodiment of the present invention. When the fencing sword body sequentially passes over the triboelectric touch sensors 8 located at different positions, peak output electrical signals will be generated sequentially. The system provided by the present invention analyzes the characteristics of the electrical signals for classification and can visually output the hitting path.

[0036] When the electrical signal is transmitted to the data processing module, since the electrical signals generated by different hitting methods have different characteristics, the data processing module extracts the characteristic information in the electrical signal to obtain the hitting time, hitting position, and hitting force of the hit. Among them, the data processing module marks the hitting timestamp according to the time when the electrical signal is received to determine the hitting time. The hitting position is accurately located according to the position of the triboelectric pressure sensor that generates the electrical signal in the array arrangement. Since the voltage values of the electrical signals generated by different hitting forces are different, the hitting force is calculated according to the obtained different voltage values of the electrical signals. At the same time, according to the number of received hitting information times, the number of hits is counted and recorded. The data is transmitted to the data analysis module through the data processing module to analyze the hitting information, so as to analyze and classify the hitting category of the hitting information, where the hitting category includes stabbing, missing, and whipping.

[0037] Since the determination of valid hit signals is different in different fencing competition types, in some embodiments, the data analysis module is further configured to identify valid hit signals in the hitting information according to different competition types. For example, in foil fencing competitions, first, the hitting time is determined. If it is difficult to manually determine when both sides hit at the same time, objective proof can be provided according to the signal generation timing; for the hitting position, only the torso is valid in foil fencing, so only the signals of the torso part are valid signals; for the hitting force, it is stipulated before the game that the hitting force greater than x N (corresponding to a specific voltage value), and the signal exceeding the specific voltage value is considered a valid signal; for the hitting category, only the stabbing signal is a valid signal in foil fencing competitions. The present invention provides data support for objective information in fencing competitions through the automatic analysis and judgment of hitting time, hitting position, hitting force, and hitting category, avoiding the subjectivity of manual judgment and improving the accuracy and efficiency of competition penalties.

[0038] Deep learning is a new branch of machine learning. Based on artificial neural networks, especially deep neural networks such as CNN, RNN, KNN, etc., it can automatically learn multi-level abstract features of data and is applicable to different application scenarios. Specifically, through a multi-layer neural network, features can be automatically extracted from raw data (such as images, audio) without manual intervention, achieving good application effects. The goal of intelligent sports monitoring is to analyze athletes' movements, physical fitness status, and training effects in real time through technical means. Deep learning has become a key driving force due to its powerful data processing ability. Therefore, the data analysis module in the embodiments of the present invention is based on a deep learning model to achieve the analysis and classification of feature information.

[0039] Specifically, the duration of the voltage signal of fencing actions (stabbing, missing, whipping) is 0.2 - 1 second, and the sampling rate of the triboelectric pressure sensor is 2 kHz. Therefore, when acquiring data, the signal data length is intercepted to be 5000 time steps to be consistent. The acquired hit information of known categories is divided into a training set, a validation set, and a test set. Since the voltage values of the electrical signals of different hitting methods are different, for example, the signal peak of stabbing can reach 9V, while the voltage value of the missing signal is close to 0V, it is necessary to preprocess the acquired data.

[0040] Figure 6 It is a schematic flowchart of the process for constructing the deep learning model provided by the embodiments of the present invention. As Figure 6 shown, the process for constructing the deep learning model includes: acquiring hit information of known hit categories and dividing the hit information into a training set, a validation set, and a test set; using a standardization method to perform data normalization on the hit information and converting the processed hit information into 2D matrix data composed of the number of channels and time steps; after initializing the deep learning model, setting a loss function and an optimizer and presetting the number of iteration rounds, inputting the training set into the deep learning model for batch training and performing data augmentation in the way of randomly shuffling the data order; after each round of training, using the validation set to calculate the loss value and accuracy of the model. When the preset number of iteration rounds is reached, if the loss value reaches the preset loss threshold, then save the model weights at this time as the trained model weights and the training is completed. If the loss value does not reach the preset loss threshold, then continue the training; and inputting the test set into the trained model to output the category prediction score and calculate the accuracy of the predicted category. When the accuracy reaches the preset accuracy, the model construction is completed. If the accuracy does not reach the preset accuracy, then retrain.

[0041] Specifically, in this embodiment, a standardization method is used to perform data normalization on the hit information, making its mean 0 and standard deviation 1, thereby optimizing the numerical stability of model training. It should be noted that in this example, a 2×4 sensor array is used, and the single hit information is converted into a 2D matrix of 8 channels × 5000 time steps to conform to the input format of the convolutional neural network (CNN), that is, the original shape is adjusted to a structure of (number of channels, time step) to adapt to the feature extraction method of the deep learning model.

[0042] After initializing the used deep learning model, setting the cross-entropy loss function, Adam optimizer, and a preset number of 200 training epochs, the training set is input into the deep learning model for batch training, and data augmentation is performed in the way of randomly shuffling the data order to enhance the generalization ability of the model. After each training epoch ends, the performance of the model is evaluated using the validation set data, and the loss value and prediction accuracy of the model at this time are calculated. When the preset number of epochs is reached, if the loss value reaches the preset loss threshold, then the model weights at this time are saved as the trained model weights, and the training is completed. If the loss value does not reach the preset loss threshold, then continue training. Finally, load the trained model, input the test set data into the model, calculate the accuracy of the predicted category according to the output category prediction scores. When the accuracy reaches the preset accuracy, the model construction is completed. If the accuracy does not reach the preset accuracy, then retrain until the accuracy reaches the preset value.

[0043] Figure 3 It is a schematic structural diagram of a deep learning model provided by an embodiment of the present invention. As Figure 3 shown, the structure of the deep learning model includes: four 1D convolutional layers (Conv), RELU activation function, global average pooling layer (Global average pooling), and four fully connected layers (FC). Specifically, the deep learning model receives the electrical signals of 5000 time steps for each channel. These electrical signals extract temporal feature information (mean, standard deviation, peak, etc.) through four 1D convolutional layers, gradually expanding the number of channels to 512, and enhancing the feature expression of the feature information in combination with the RELU activation function. Subsequently, the global average pooling is used to compress the time dimension of the feature information and reduce the computational complexity. Then, the compressed feature information is gradually reduced to a 3D output through four fully connected networks, and this output corresponds to the prediction scores of three types of signals: stabbing, missing, and whipping.

[0044] In some embodiments, the system further includes a data visualization module, which is configured to convert the hit information into an electro-signal graph and generate a corresponding heat map according to the statistically obtained feature information and the number of hits, based on the hit positions; when the number of hits at a certain hit position is greater than or equal to a preset number, the hit position is presented in a first color on the heat map, when the number of hits at the hit position is less than the preset number, the hit position is presented in a second color on the heat map, and the positions not hit by the fencing are presented in a third color on the heat map.

[0045] Specifically, the data visualization module receives the hit information processed by the data processing module. In the present invention, this hit information is an electro-signal, and it converts the electro-signal into an electro-signal graph, thereby converting the information on the hit process of the fencing into a schematic diagram. In some embodiments, the deep learning model used by the data analysis module can process the electro-signal or the converted electro-signal graph. Since the data visualization module can generate a corresponding heat map according to the feature information of the hits and the statistically obtained corresponding number of hits, based on the hit positions, it should be noted that when the information collection module is not hit by the fencing, no hit information is generated, and at this time, the positions not hit, that is, the positions where no hit information is generated, are presented in a third color on the heat map, and the third color is defaulted to white. Specifically, during daily actual combat training, the data processing module counts the number of hits in each area of the fencing uniform. When the number of hits at a certain position is greater than or equal to the preset number, it means that this position has been hit more times. This hit position is presented in a first color on the heat map. The first color can be dark red, indicating that this position is a weak defensive area of the athlete and requires strengthening of defensive training. When the number of hits at a certain position is less than the preset number, it means that this position has been hit relatively fewer times. This hit position is then presented in a second color on the heat map. The second color can be light red, indicating that this position is an area where the athlete's attack is not precise and requires strengthening of attack training.

[0046] In addition, in some embodiments, the total number of touches and the number of hits can be separately counted. When the percentage of the number of hits in the total number of touches is greater than P1, the corresponding hit position is displayed in red on the heat map, representing the weak defensive area of the athlete. When the percentage of the number of hits in the total number of touches is less than P2, the corresponding hit position is displayed in blue on the heat map, representing the area where the athlete's attack is not precise. Thus, based on the generated heat map, the training habits and defensive weaknesses of the athlete can be evaluated, helping the trainer to formulate a training plan, assisting in daily training, thereby optimizing the attack strategy and providing targeted technical improvement suggestions.

[0047] In some embodiments, the fencing system further includes a competition and training module, which calculates a scoring result according to the feature information and the hit category according to a preset scoring rule; calculates the hit percentage of each category according to the hit category; and displays the heat map, the scoring result, the hit percentage, and the feature information.

[0048] Specifically, the competition and training module receives the data processed by the data processing module, and can preset scoring rules according to different competition types. For example, in a foil competition, hitting the whole body is regarded as a valid hit, and in a sabre competition, only hitting the torso is a valid hit. It automatically calculates the valid score according to the received hit time, hit position, and hit strength, and prompts the referee to confirm, reducing human misjudgment, which is beneficial to improving the accuracy of the referee's judgment on controversial penalties and ensuring the fairness of the competition.

[0049] In daily sabre practice training, both whipping and stabbing are valid hits. Since it is different from foil and epee, there is a lack of targeted testing and training for the mastery of the whipping technical movement. After N1 times of hitting practice, the data analysis module classifies each hit, and the competition and training module outputs the percentages of whipping, stabbing, and missed hits in this training. When the whipping hit rate is greater than S1, the stabbing hit rate is less than S2, and the miss rate is less than S3, it proves that the training meets the standard. In special training, specific practice can be carried out on a single area, and the competition and training module outputs the hit percentage. When the percentage of the specific training times in the total times is greater than Q, it proves that the hit training meets the standard.

[0050] In addition, the competition and training module can also be connected to the visualization module to display the heat map generated by the visualization module, and display the calculated scoring result, hit percentage, and feature information of the hit information. The competition and training module also supports the display of the first hit time and the action classification result, and supports slow motion playback and retrospective review of controversial penalties during the competition.

[0051] In a second aspect, the present invention provides an information recognition method for a fencing system, the method includes: obtaining the touch information of fencing, and generating hit information when being hit by the fencing; extracting the feature information from the hit information, and statistically analyzing the feature information and the number of hits, where the feature information includes hit time, hit position, and hit strength; and analyzing the hit to identify and classify the hit category of the hit information, where the hit category includes stabbing, missing, and whipping.

[0052] In some embodiments, the method further includes: converting the hit information into an electrical signal map, and generating a corresponding heat map according to the statistically obtained feature information and the number of hits, based on the hit positions; when the number of hits at the hit position is greater than or equal to a preset number, the hit position is presented as a first color on the heat map, when the number of hits at the hit position is less than the preset number, the hit position is presented as a second color on the heat map, and the positions not hit by the fencing are presented as a third color on the heat map.

[0053] In some embodiments, the method further includes: calculating a score result according to the feature information and the hit category, based on a preset scoring rule; calculating the hit percentage of each category according to the hit category; and displaying the heat map, the score result, the hit percentage, and the feature information.

[0054] In some embodiments, the method obtains the touch information of the fencing through triboelectric pressure sensors, which are uniformly arranged in the fencing suit in an array layout, and a plurality of the triboelectric pressure sensors arranged in the fencing suit collect the hit information in a real-time synchronous acquisition manner.

[0055] In some embodiments, based on a deep learning model, the feature information is analyzed, and the method further includes the construction process of the deep learning model: obtaining the hit information with known hit categories, and dividing the hit information into a training set, a validation set, and a test set; using a standardization method to perform data normalization processing on the hit information, and converting the processed hit information into 2D matrix data composed of the number of channels and the time step; after initializing the deep learning model, setting the loss function and the optimizer and presetting the number of iteration rounds, inputting the training set into the deep learning model for batch training, and performing data augmentation in a manner of randomly shuffling the data order; after each round of training, using the validation set to calculate the loss value and the accuracy of the model, when the preset number of iteration rounds is reached and the loss value reaches a preset loss threshold, saving the model weights at this time as the trained model weights, and the training is completed, if the loss value does not reach the preset loss threshold, continue the training; and inputting the test set into the trained model to output the category prediction score, and calculating the accuracy of the predicted category, when the accuracy reaches a preset accuracy, the model construction is completed, if the accuracy does not reach the preset accuracy, retrain.

[0056] Specifically, the deep learning model includes: four 1D convolutional layers, a RELU activation function layer, a global average pooling layer, and four fully connected layers. The hit information is input into the deep learning model, and the temporal feature information is extracted through the four 1D convolutional layers, and the number of channels of this feature information is expanded; the RELU activation function layer is used to enhance the feature expression of the feature information, and the enhanced feature information is input into the global average pooling layer to compress its time dimension; and the compressed feature information is input into the fully connected layer, and the dimension is gradually reduced to three dimensions and then output, and this output is the prediction scores of three types of signals: stabbing, missing, and whipping.

[0057] Through the above technical solutions, the problems of subjectivity in touch judgment in fencing competitions, lack of objective data basis, as well as the problems of single data evaluation and lack of personalization in training programs are solved. Visual penalty support for real-time hit position, time, and action mode is provided, which helps to assist in competition penalties. And it breaks through the limitation of the lack of biomechanical feedback in traditional training. Through high-precision action classification and multi-dimensional data analysis, it helps athletes optimize their technical movements and improve their tactical decision-making abilities, and thus assists athletes in training.

[0058] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0059] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate for implementing in the process Figure 1 one process or multiple processes and / or blocks Figure 1 a device for the functions specified in one block or multiple blocks.

[0060] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and this instruction device implements in the processFigure 1 one or more processes and / or blocks Figure 1 the functions specified in one or more blocks.

[0061] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more processes and / or blocks Figure 1 one or more blocks.

[0062] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and memory.

[0063] The memory may include non-permanent memory in the computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0064] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can store information by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0065] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

[0066] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A fencing system, characterized in that, The fencing system includes: An information collection module for obtaining the hitting information of fencing. When being hit by the fencing, it generates a hit information and transmits it to the data processing module; A data processing module for receiving the hit information, extracting the feature information in the hit information, and statistically analyzing the feature information and the number of hits. The feature information includes the hit time, hit position, and hit strength; and A data analysis module for analyzing the hit information to identify and classify the hit categories in the hit information, where the hit categories include stabbing, missing, and whipping.

2. The fencing system according to claim 1, wherein, The system further includes a data visualization module, for converting the hit information into an electro-signal graph, and generating a corresponding heat map according to the statistically analyzed feature information and the number of hits, according to the hit position; When the number of hits at the hit position is greater than or equal to a preset number, the hit position is presented in a first color on the heat map. When the number of hits at the hit position is less than the preset number, the hit position is presented in a second color on the heat map. The position not hit by the fencing is presented in a third color on the heat map.

3. The fencing system according to claim 2, characterized in that, The fencing system further includes a competition and training module, calculating a score result according to the feature information and the hit category according to a preset scoring rule; calculating the hit percentage of each category according to the hit category; and displaying the heat map, the score result, the hit percentage, and the feature information.

4. The fencing system according to claim 1, wherein The information collection module includes triboelectric pressure sensors, which are uniformly arranged in an array manner inside the fencing suit. The triboelectric pressure sensors include: A first support material layer; A positive triboelectric material layer, which is arranged on the first support material layer; A second support material layer; A negative triboelectric material layer, which is arranged on the second support material layer and is arranged opposite to the positive triboelectric material layer, and is used to contact the positive triboelectric material layer when the tip or the blade of the sword hits the triboelectric pressure sensor, and separate from the positive triboelectric material layer when the tip or the blade of the sword leaves the triboelectric pressure sensor; An induction electrode, electrically connected to the negative triboelectric material layer; and An output electrode, electrically connected to the induction electrode, for outputting the electrical signal generated by the contact between the negative triboelectric material layer and the positive triboelectric material layer.

5. The fencing system according to claim 4, wherein, Multiple triboelectric pressure sensors arranged inside the fencing suit collect the hit information in a real-time synchronous acquisition manner.

6. The fencing system according to claim 1, characterized in that, The data analysis module analyzes the hit information based on a deep learning model. The construction process of the deep learning model includes: Obtaining the hit information with known hit categories, and dividing the hit information into a training set, a validation set, and a test set; Using a standardization method to perform data normalization processing on the hit information, and converting the processed hit information into 2D matrix data composed of the number of channels and the time step; After initializing the deep learning model, setting the loss function and optimizer, and presetting the iteration rounds, the training set is input into the deep learning model for batch training, and data enhancement is performed by randomly disrupting the data order; After each round of training, the loss value and accuracy of the model are calculated using the validation set. When the preset iteration round is reached, if the loss value reaches the preset loss threshold, the model weight at this time is saved as the trained model weight, and the training is completed. If the loss value does not reach the preset loss threshold, the training continues; and The test set is input into the trained model to output the category prediction score and calculate the accuracy of the predicted category. When the accuracy reaches the preset accuracy, the model construction is completed. If the accuracy does not reach the preset accuracy, retraining is performed.

7. The fencing system according to claim 6, wherein The deep learning model includes: four 1D convolutional layers, RELU activation function, global average pooling layer and four fully connected layers. The hit information is input into the deep learning model, and the temporal feature information is extracted through the four 1D convolutional layers, and the number of channels of the feature information is expanded; Using the RELU activation function to enhance the feature expression of the feature information, and inputting the enhanced feature information into the global average pooling layer to compress its time dimension; and The compressed feature information is input into the fully connected layer, and the dimension is gradually reduced to three dimensions before being output, which is the prediction score of the three types of signals: stab, miss and whip.

8. The fencing system according to claim 1, wherein The data analysis module is also used to identify effective hitting signals in the hitting information according to different game types.

9. An information recognition method for a fencing system, characterized in that, The method comprises: Acquiring the striking information of the fencing, and generating the striking information when being struck by the fencing; Extracting characteristic information from the hitting information, and counting the characteristic information and the number of hits, wherein the characteristic information includes the hitting time, the hitting position, and the hitting force; and The hits are analyzed to identify and classify hit categories of the hit information, wherein the hit categories include piercing hits, misses, and whiplash.

10. The recognition method according to claim 9, wherein The method further comprises: Converting the hit information into an electrical signal graph, and generating a corresponding heat map according to the hit position based on the statistical feature information and the hit times; When the number of hits counted at the hitting position is greater than or equal to a preset number, the hitting position is presented as a first color on the heat map; when the number of hits counted at the hitting position is less than a preset number, the hitting position is presented as a second color on the heat map; the position not hit by the fencing is presented as a third color on the heat map.

Images