Data processing method, readable storage medium and electronic equipment

Abstract

The invention relates to the field of communication, in particular to a data processing method, a readable storage medium and electronic equipment. In the method, a first node device in the distributed system receives noise statistical information sent by a decision-making device in the distributed system, receives feature information sent by node devices except the first node device in the distributed system, and carries out training of a neural network model. And obtaining a power adapter sub-model for executing the classification task. It can be understood that the noise statistical information comprises the statistical characteristics of the noise of the channel between each node device and the decision-making device, and the feature information comprises the feature data of each sample in the sample set of the node devices and the classification label of each sample. Therefore, the trained power adaptation sub-model corresponding to each node device can be adapted to the noise in the channel corresponding to each node device, thereby improving the accuracy of the classification result of executing the corresponding classification task in the distributed system.

Description

Technical Field

[0001] This application relates to the field of communications, and in particular, to a data processing method, a readable storage medium, and an electronic device. Background Art

[0002] In some scenarios, a transmitting end (an electronic device that sends signals) can transmit a signal related to a certain classification task (hereinafter referred to as a "task signal") to a receiving end (an electronic device that receives signals) through a channel, so as to achieve information transmission. After receiving the task signal, the receiving end can perform a classification task based on the received signal.

[0003] For example, assume that the transmitting end is a camera and the receiving end can be a base station, and the classification task is to determine the gender (i.e., male or female) of a person in the image captured by the camera. The camera can capture images of people in the surrounding environment to obtain image data, and transmit the signal corresponding to the image data to the base station through a channel. The base station will restore the image captured by the camera according to the received signal, and determine whether the person in the surrounding environment of the camera is male or female based on the restored image.

[0004] However, the signal corresponding to task data (such as image data) is usually interfered by noise signals during the transmission from the transmitting end to the receiving end, and is greatly affected by the noise interference, resulting in a poor classification result determined by the receiving end. Summary of the Invention

[0005] To solve the above problems, embodiments of this application provide a data processing method, an electronic device, and a readable storage medium, which are used to improve the accuracy of the classification result of a classification task.

[0006] In a first aspect, an embodiment of this application provides a data processing method, including: a first node device in a distributed system receives first noise statistical information sent by a decision device in the distributed system, and receives first feature information sent by node devices in the distributed system except the first node device respectively; wherein, the first noise statistical information includes the statistical characteristics of the noise of the channel between the node device and the decision device, and the first feature information includes the feature data of each sample in the first sample set of the node device and the classification label of each sample; and the first noise statistical information and the first feature information are used to determine a first power adaptation sub-model used by each node device to perform a first task.

[0007] It can be understood that the first task can be a certain classification task, not limited to any classification scenario. Each node device in the distributed system can extract features from its own first sample set to obtain the feature data of each sample. Since each sample has a corresponding classification label, the feature data will also have a corresponding classification label at this time. When information can be transmitted between the node devices in the distributed system, any one of the node devices can receive the first feature information sent by other node devices except itself, so that each node device can obtain the feature data of the samples of all node devices. Since the decision-making device can observe the statistical characteristics of the noise of the channels corresponding to each node device, the decision-making device can send the observed statistical characteristics of the noise of the channels corresponding to each node device to each node device. Since the node device has strong computing power, any one of the node devices can determine the first power adaptation sub-model required to execute the first task according to the feature data of the samples of all node devices received and the statistical characteristics of the noise of the channels corresponding to each node device. The power adaptation sub-models of each node device can be used to adjust the data sent by each node device to the decision-making device before sending. For example, the first node device is node 0 in the following text, and the first power adaptation sub-models of each node device can be the power adaptation sub-models corresponding to each node determined by node 0. The feature data of each sample in the first sample set of the node device, as well as the classification label of each sample, can be the label feature data in the following text. It can be understood that since the statistical characteristics of the noise of the channels corresponding to all node devices are considered when determining the first power adaptation sub-model, after using the determined first power adaptation sub-model to adjust the data before sending, the adjusted data can be adapted to the noise of the corresponding channel, and the ability to protect itself from noise interference is stronger. Moreover, the relevance of the data of each node device is also considered, thereby improving the accuracy of the classification result.

[0008] In a possible implementation of the foregoing first aspect, the first power adaptation sub-model used by each node device to execute the first task is determined in the following manner: The first node device trains the first neural network model based on the first noise statistical information and the first feature information corresponding to each node device to obtain the trained first neural network model, where the trained first neural network model includes the first power adaptation sub-model corresponding to each node device.

[0009] It can be understood that the first node device performing neural network model training can be node 0 in the following text performing neural network model training. Since the neural network model has strong fitting ability, and the first power adaptation sub-model is obtained through neural network model training, the scientific nature of the first power adaptation sub-models of each node device obtained is ensured.

[0010] In a possible implementation of the first aspect described above, the statistical characteristics include one or more of the following: mean vector, covariance, arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, midrange, mode, mean absolute deviation, variance.

[0011] In a possible implementation of the first aspect described above, it further includes that the first node device sends first adaptation processing information to the decision-making device, where the first adaptation processing information includes multiple groups of first input-output data pairs, and each group of first input-output data pairs corresponds to a first power adaptation sub-model of a node device respectively.

[0012] It can be understood that the first power adaptation sub-models of each node device can be the power adaptation sub-model corresponding to node 0 and the power adaptation sub-model corresponding to node 1 below. The input-output data pairs of each group can be the key point information below. It can be understood that sending the input-output data pairs corresponding to the first power adaptation sub-models of each node device to the decision-making device can reduce the data transmission volume, thereby saving transmission resources.

[0013] In a possible implementation of the first aspect described above, it further includes that the decision-making device obtains a first power adaptation function corresponding to the first power adaptation sub-model of each node device based on the received first adaptation processing information.

[0014] It can be understood that based on the input-output data pairs corresponding to the first power adaptation sub-models of each node device in the received first adaptation information, the decision-making device can recover the first power adaptation sub-model required for decision-making, that is, the obtained first power adaptation function corresponding to the first power adaptation sub-model is the first power adaptation sub-model recovered according to the input-output data pairs, so as to facilitate the subsequent execution of the first task.

[0015] In a possible implementation of the first aspect described above, the decision-making device obtains a first power adaptation function corresponding to the first power adaptation sub-model of each node device based on the received first adaptation processing information, including: determining the first power adaptation function corresponding to the first power adaptation sub-model of the first node device in the following way: the decision-making device inputs the first input-output data pair corresponding to the first power adaptation sub-model of the first node device into the first parametric function model to obtain the first solution values of the parameters in the first parametric function model; the decision-making device replaces the parameters in the first parametric function model with the corresponding first solution values to obtain the first power adaptation function.

[0016] It can be understood that by selecting an appropriate parametric function model and solving the parameter values of the appropriate parametric model through input-output data pairs, the first power adaptation function corresponding to the first power adaptation sub-model required can be quickly restored.

[0017] In a possible implementation of the above first aspect, after the first node device receives the first feature information respectively sent by the node devices other than the first node device in the distributed system, it further includes: the first node device sends the first label feature statistical information to the decision device, where the first label feature statistical information includes the statistical characteristics corresponding to each of the multiple classification labels, and among the multiple classification labels, the statistical characteristics corresponding to the first classification label are the statistical characteristics of the feature data with the classification label of the first classification label in all the feature data corresponding to the sample set of all the node devices in the distributed system.

[0018] It can be understood that since the first node device can obtain all the feature data corresponding to the first sample set of all the node devices, at this time, statistics can be performed according to the classification labels to obtain the statistical characteristics of the feature data with the same classification label in all the feature data corresponding to the first sample set of all the node devices. Then, the obtained statistical characteristics can be sent to the decision device. At this time, the decision device can simulate the data required in the decision process according to the statistical characteristics in the first label feature statistical information. It can be understood that since the first node device only transmits the first label feature statistical information instead of all the feature data, the amount of data transmitted is reduced.

[0019] In a possible implementation of the above first aspect, it further includes: each node device obtains the first data to be classified corresponding to the first classification object in the first task; after each node device processes the first classification feature data of the first data to be classified into second classification feature data based on its respective first power adaptation sub-model, it sends the second classification feature data to the decision device.

[0020] It can be understood that the first data to be classified can be the task data in the following text; the first classification feature data of the first data to be classified is the feature data obtained after feature extraction of the task data; the second classification feature data is the adjusted feature data obtained after processing the feature data obtained after feature extraction with the power adaptation sub-model in the following text. When each node device adjusts the first classification feature data before sending according to its respective first power adaptation sub-model to obtain the second classification feature data, since the first power adaptation sub-model is obtained by considering the statistical characteristics of the channel noise, the anti-noise ability of the second classification feature data during channel transmission is improved, and the accuracy of the data received by the decision device is increased.

[0021] In a possible implementation of the above first aspect, it further includes that the decision device obtains a classification result for the first classification object based on the first power adaptation function corresponding to the first power adaptation sub-model of each node device, the third classification feature data corresponding to the second classification feature data sent by each node device, and the statistical characteristics corresponding to each classification label among multiple classification labels.

[0022] It can be understood that the third classification feature data is the received-end data received by the decision center below.

[0023] In a possible implementation of the above first aspect, the decision device obtains a classification result for the first classification object based on the first power adaptation function corresponding to the first power adaptation sub-model of each node device, the third classification feature data corresponding to the second classification feature data sent by each node device, and the statistical characteristics corresponding to each classification label among multiple classification labels, including: the decision device determines the likelihood function values corresponding to the first classification object and each classification label respectively based on the first power adaptation function corresponding to each node device, the third classification feature data corresponding to the second classification feature data sent by each node device, and the statistical characteristics corresponding to each classification label among multiple classification labels; and takes the classification category of the classification label corresponding to the maximum likelihood function value among the likelihood function values as the classification result of the first classification object.

[0024] It can be understood that the decision device makes a decision based on the received-end data corresponding to multiple node devices, that is, calculates the likelihood function, so that the classification result takes into account the correlation between multiple node data and improves the accuracy of the decision result.

[0025] In a possible implementation of the above first aspect, it further includes that the first node device sends second adaptation processing information to the decision device, where the second adaptation processing information includes the first power adaptation sub-model for each node device to execute the first task.

[0026] It can be understood that when the first node device sends the first power adaptation sub-model for executing the first task to the decision device, although the transmission amount is large at this time, the accuracy obtained by the decision device according to the first power adaptation sub-model corresponding to each node device is relatively high.

[0027] In a second aspect, an embodiment of the present application provides a data processing method, including: a decision device in a distributed system receives second feature information sent by each node device in the distributed system; where the second feature information includes the feature data of each sample in the second sample set of the node device and the classification label of each sample; and the second feature information is used to determine the second power adaptation sub-model used by each node device to execute the second task.

[0028] It can be understood that the second task can be a certain classification task, not limited to any classification scenario. Each node device in the distributed system can perform feature extraction on its own second sample set to obtain the feature data of each sample. Since each sample has a corresponding classification label, the feature data will also have a corresponding classification label at this time. For example, the label feature data in the following text. When model training needs to be performed on the node devices in the distributed system, at this time, each node device can send the second feature information to the decision-making device. Since the second information includes the feature data of each sample in the second sample set of the corresponding node device and the classification label of each sample, it is convenient for the subsequent decision-making device to perform model training to obtain the second power adaptation sub-model used by each node device to perform subsequent tasks. For example, each node device is node 0 and node 1 in the following text. Node 0 and node 1 respectively send the second feature information to the decision center 00, so that the decision center 00 can determine the power adaptation sub-model corresponding to the node.

[0029] In a possible implementation of the above second aspect, the second power adaptation sub-model used by each node device to perform the second task is determined in the following manner: The decision-making device trains the second neural network model based on the statistical characteristics of the noise of the channels between the decision-making device and each node device, and the second feature information corresponding to each node device, to obtain the trained second neural network model. Among them, the trained second neural network model includes the second power adaptation sub-model corresponding to each node device.

[0030] It can be understood that since the decision-making device trains the neural network model based on the statistical characteristics of the noise of the channels corresponding to each node device and the second feature information corresponding to each node device, when determining the second power adaptation sub-model, the feature data of the second sample sets of all node devices and the statistical characteristics of the noise of the channels corresponding to each node device are fully considered. After the determined second power adaptation sub-model adjusts the data before sending, the adjusted data can be adapted to the noise of the corresponding channel, and the ability to protect itself from noise interference is stronger. At the same time, the relevance of the data of each node device is also considered, thereby improving the accuracy of the classification result.

[0031] In a possible implementation of the above second aspect, the statistical characteristics include one or more of the following: mean vector, covariance, arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, midrange, mode, mean absolute deviation, variance.

[0032] In a possible implementation of the second aspect described above, it further includes that the decision-making device separately sends third adaptation processing information to each node device, where the third adaptation processing information includes a set of second input-output data pairs, and each set of second input-output data pairs corresponds to the second power adaptation sub-model of the sending node device.

[0033] In a possible implementation of the second aspect described above, it further includes: each node device obtains a second power adaptation function corresponding to the second power adaptation sub-model of the node device based on the received third adaptation processing information.

[0034] In a possible implementation of the second aspect described above, each node device obtains a second power adaptation function corresponding to the second power adaptation sub-model of the node device based on the received third adaptation processing information, including: the first node device obtains a second power adaptation function corresponding to the second power adaptation sub-model of the first node device based on the received third adaptation processing information: the first node device inputs the second input-output data pair into the second parametric function model to obtain the second solution values of the parameters in the second parametric function model; the first node device replaces the parameters in the second parametric function model with the corresponding second solution values to obtain the second power adaptation function.

[0035] In a possible implementation of the second aspect described above, it further includes: each node device obtains second data to be classified corresponding to the second classification object of the second task; each node device processes the fourth classification feature data of the second data to be classified into fifth classification feature data respectively based on its own second power adaptation function and then sends it to the decision-making device.

[0036] In a possible implementation of the second aspect described above, it further includes that the decision-making device obtains a classification result for the second classification object based on the second power adaptation sub-model corresponding to each node device, the sixth classification feature data corresponding to the fifth classification feature data sent by each node device, and the feature data of each sample in the second sample set of all node devices.

[0037] In a possible implementation of the second aspect described above, the decision-making device obtains a classification result for the second classification object based on the second power adaptation sub-models corresponding to each node device, the sixth classification feature data corresponding to the fifth classification feature data sent by each node device, and the feature data of each sample in the second sample set of all node devices, including: the decision-making device determines the likelihood function values corresponding to the second classification object and each classification label based on the second power adaptation sub-models corresponding to each node device, the sixth classification feature data corresponding to the fifth classification feature data sent by each node device, and the feature data of each sample in the second sample set of all node devices; and takes the classification category of the classification label corresponding to the maximum likelihood function value among the likelihood function values as the classification result of the second classification object.

[0038] In a possible implementation of the second aspect described above, the decision-making device further sends fourth adaptation processing information to each node device, where the fourth adaptation processing information includes the second power adaptation sub-model corresponding to the node device.

[0039] In a third aspect, an embodiment of the present application provides an electronic device, including: one or more processors; one or more memories; and one or more memories store one or more instructions, when the one or more instructions are executed by the one or more processors, enabling the electronic device to execute the data transmission method provided in the first aspect and any possible implementation of the first aspect, or the data transmission method provided in the second aspect and any possible implementation of the second aspect.

[0040] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which instructions are stored, and when the instructions are executed on a computer, enabling the computer to execute the data transmission method provided in the first aspect and any possible implementation of the first aspect, or the data transmission method provided in the second aspect and any possible implementation of the second aspect.

[0041] In a fifth aspect, an embodiment of the present application provides a computer program product containing instructions, and when the computer program product runs on a computer, enabling the computer to execute the data transmission method provided in the first aspect and any possible implementation of the first aspect, or the data transmission method provided in the second aspect and any possible implementation of the second aspect. Description of the Drawings

[0042] Figure 1A According to some embodiments of the present application, a schematic diagram of a scenario for transmission based on the same task in a distributed system is shown;

[0043] Figure 1B According to some embodiments of the present application, a schematic diagram of an image T1 is shown;

[0044] Figure 1C According to some embodiments of the present application, a schematic diagram of an image T2 is shown;

[0045] Figure 1D According to some embodiments of the present application, a schematic diagram of an image T3 is shown;

[0046] Figure 1E According to some embodiments of the present application, a schematic diagram of an image T4 is shown;

[0047] Figure 2A According to some embodiments of the present application, a schematic diagram of a process by linear filtering processing is shown;

[0048] Figure 2B According to some embodiments of the present application, a schematic diagram of a process by BPSK processing is shown;

[0049] Figure 2C According to some embodiments of the present application, a schematic diagram of an interval for binary classification is shown;

[0050] Figure 2D According to some embodiments of the present application, a schematic diagram of an interval for ternary classification is shown;

[0051] Figure 2E According to some embodiments of the present application, a schematic diagram of interval scaling is shown;

[0052] Figure 3A According to some embodiments of the present application, a schematic diagram of a decentralized distributed system framework system 300 is shown;

[0053] Figure 3B According to some embodiments of the present application, a schematic diagram of a process in which node 0 and node 1 adjust the feature data corresponding to the task data and send it to the decision center 00, and the decision center 00 makes a decision is shown;

[0054] Figure 4A According to some embodiments of the present application, a schematic diagram of a framework system 300 is shown;

[0055] Figure 4B According to some embodiments of the present application, a schematic diagram of a non-linear structure corresponding to a power adaptation function is shown;

[0056] Figure 5A According to some embodiments of the present application, a schematic diagram of a process in which node 0 receives the label feature data from node 1 and the noise variance of the decision center and performs neural network model training is shown;

[0057] Figure 5BAccording to some embodiments of the present application, a schematic diagram of feature extraction for sample data of node j is shown;

[0058] Figure 5C According to some embodiments of the present application, a schematic diagram of information interaction between node j and other nodes j' is shown;

[0059] Figure 5D According to some embodiments of the present application, a schematic diagram of the decision center 00 sending the variance σ of noise 2 to node j is shown;

[0060] Figure 5E According to some embodiments of the present application, a schematic diagram of the process of model training is shown;

[0061] Figure 6A According to some embodiments of the present application, a schematic diagram of the process of adapting the function of transmitting power through key point information is shown;

[0062] Figure 6B According to an embodiment of the present application, a schematic diagram of a curve corresponding to the power adaptation function f 0 is shown;

[0063] Figure 6C According to some embodiments of the present application, a schematic diagram of node j sending key point information of the power adaptation function to the decision center 00 is shown;

[0064] Figure 7 According to an embodiment of the present application, a schematic diagram of a centralized distributed system framework system 700 is shown;

[0065] Figure 8A According to an embodiment of the present application, a schematic diagram of LeNet-5 is shown;

[0066] Figure 8B According to an embodiment of the present application, a schematic diagram of a fully connected neural network is shown;

[0067] Figure 9 According to an embodiment of the present application, a schematic diagram of the hardware structure of an electronic device 10 is shown. Detailed implementation manners

[0068] Illustrative embodiments of the present application include but are not limited to a data processing method, a readable storage medium, and an electronic device, etc.

[0069] It can be understood that the data processing method provided in the embodiments of the present application can be applied to any wireless communication scenario, including but not limited to being applied in a wireless communication system with sensing functions, such as wireless communication systems like 5G - the fifth generation of mobile communication, 6G - the sixth generation of mobile communication, etc., as well as short - range wireless communication systems, such as Wi - Fi and ultra - wideband wireless communication technology (UWB), etc., which are not limited here. Moreover, the data processing method provided in the embodiments of the present application can be widely used in any terminal device or network - side device in a wireless communication scenario, which is not limited here.

[0070] The embodiments of the present application will be specifically described below with reference to the accompanying drawings.

[0071] Figure 1A According to an embodiment of the present application, a schematic diagram of a scenario for transmission based on the same task in a distributed system is shown.

[0072] It can be understood that in a distributed system, there will be a group of nodes (i.e., node devices) that cooperate for the same classification task, and a decision - making center (i.e., decision - making device) that can perform integrated analysis and processing on signals from different nodes. Each node can collect different task data of the classification object (for example, taking images of different angles of a certain environment, collecting sounds at different positions in a certain environment, etc.), and send the task data collected by each node to the decision - making center, and the decision - making center classifies the classification object according to the received task data.

[0073] It can be understood that the classification object in the embodiments of the present application can include but not be limited to a certain environment in an image, a certain thing in an image, the speaker in an audio, the location where the audio is generated, the description object in a text, the text emotion in a text, the semantics in a text, etc. The task data can include but not be limited to text data, audio data, image data, etc.

[0074] It can be understood that the nodes proposed in the embodiments of the present application can be any electronic device with processing capabilities, including but not limited to mobile phones, tablet computers, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants (PDAs), servers, cameras, etc. The decision-making centers proposed in the embodiments of the present application can be any electronic device with decision-making capabilities, including but not limited to base stations, routers, computer room hosts, mobile phones, tablet computers, in-vehicle devices, augmented reality / virtual reality devices, ultra-mobile personal computers (UMPCs), netbooks, personal digital assistants, servers, etc., which are not limited herein.

[0075] The following takes the classification object being a person in an image, the node being a camera, and the decision-making center being a base station as an example for elaboration.

[0076] As Figure 1A shown, a distributed system with multiple cameras such as camera A1, camera A2,..., camera An as nodes and a base station B as the decision-making center, and the multiple cameras A1 - An cooperate with the base station B to complete the classification task. The following takes the example where the multiple cameras A1 - An and the base station B need to perform a binary classification task for elaboration.

[0077] Specifically, cameras A1 - An capture the classification object G in environment 1. Since cameras A1 - An are located at different positions on the road, cameras A1 - An will capture the classification object G at different angles, thereby obtaining different image data. For example, Figure 1B and Figure 1C respectively show an image T1 and an image T2 of the classification object G captured by camera A1. Figure 1D and Figure 1E respectively show an image T3 and an image T4 of the classification object G captured by camera A2. It can be understood that due to different shooting angles, the image qualities of images T1, T2, T3, and T4 are different, and the image of the classification object G captured in image T1 is significantly clearer than that captured in images T2, T3, and T4. Then, each camera processes the captured image data into a task signal that can be transmitted and transmits it to the base station B through a channel. As the decision-making center, the base station B makes a final classification decision based on the task signals received from each camera to determine whether the gender of the classification object G in the current environment 1 is male or female.

[0078] In some embodiments, each camera transmits the image content of the corresponding classification task to base station B, enabling the base station to make decisions based on the image content. For example, camera A1 transmits the content of images T1 and T2 to base station B, and other cameras use the same method, which will not be elaborated here. Base station B makes decisions based on the task signals received from each camera. Specifically, as Figure 2A shown, for camera A1, camera A1 can process the collected images T1 and T2 through encoder 01 and linear filter 02 to obtain the processed data. And the processed data is transformed into a transmissible task signal and transmitted to base station B through the channel. Base station B restores the data obtained from camera A1 through decoder 03 to obtain the data for decision-making. For other cameras, the interaction process with base station B is essentially the same as that of camera A1 and will not be elaborated here. Base station B makes decisions based on the task data received from each corresponding node.

[0079] However, during the process of camera A1 transmitting the task signals corresponding to images T1 and T2 to base station B, the task signals are affected by channel noise (such as the distribution of the channel noise being Z 0 ~N(0, σ 2 ))), resulting in the signals received by base station B including both task signals and noise signals. Thus, in some scenarios, due to the large influence of noise on the signals received by base station B, some classification data restored by the receiving end from the received signals is inaccurate, leading to a suboptimal accuracy of the task results determined by the receiving end.

[0080] For example, the signals received by base station B include not only the task signals corresponding to images T1 and T2 but also noise signals. Specifically, for image T2, before transmission, according to the task data of image T2, it can be classified as male. However, since image T2 is a back view of a person and shows less obvious male characteristics. During the transmission process, it is affected by noise interference. At this time, base station B receives signals that include not only the task signal of image T2 but also noise signals. At this time, base station B will, based on the signals received corresponding to image T2, misclassify the original male characteristics as female characteristics, thus making it very likely that the classification result is misclassified from male to female.

[0081] In some other embodiments, each camera may first classify the currently acquired image data to obtain a classification result. For example, whether the gender is male or female, and then send the classification result to base station B. Base station B makes a decision based on the classification results sent by each camera to obtain a final decision result. For example, compare the number of classification results corresponding to "male" and the number of classification results corresponding to "female" among the classification results sent by each camera, and use the classification result with the larger number as the final decision result. Specifically, as Figure 2B shown Figure 1A each camera in determines whether the object in the current environment 1 is male or female (i.e., obtains a classification result) based on the image-related data captured by itself. Then, the obtained classification result is encoded by encoder 11 to obtain task data. For example, the classification result "male" can be encoded as 0 and the classification result "female" can be encoded as 1, processed by binary phase shift keying (BPSK) 12, and then the classification result is sent to base station B. After base station B restores the classification result through decoder 13, it makes a decision to obtain a decision result.

[0082] It can be understood that, similarly, although the signals received by base station B not only include task signals corresponding to each classification result, but also include noise signals, such as Figure 2B the noise shown in (such as the distribution of channel noise is Z 0 ~N(0, σ 2 ^2)). Since only the classification result needs to be transmitted, under binary phase shift keying, only two phases are needed for transmission, that is, the classification result can be distinguished by the phase difference between the two. Therefore, the small changes generated by the task signal affected by noise will not affect the transmission result.

[0083] However, since base station B can only make a final decision based on the classification results processed by each camera, it cannot consider the correlation of the image data collected between each camera, nor can it determine the reliability of the classification results obtained by each node. Therefore, the obtained decision result is too one-sided and the accuracy of the decision result is not high enough. For example, if three cameras classify the classified object G as male based on the side and back of the object in the images they captured, while one camera captures the front face of the object and classifies it as female, the base station will determine it as male based on the results received from each camera, resulting in a misjudgment. Moreover, each camera needs to process the image data to obtain a classification result, which consumes more resources when the number of cameras is large.

[0084] In some embodiments, for a classification task of classifying an object into K categories, each transmitter can extract features from the task data of the classified object it collects, obtain the feature data corresponding to the task data, and send the feature signal corresponding to the feature data to the receiver. Among them, the feature data is used to indicate the classification corresponding to the task data collected by the transmitter, and the amplitude of the feature signal is positively correlated with the magnitude of the feature data value. After receiving the feature signals sent by multiple transmitters, the receiver can restore the feature data corresponding to each feature signal based on the amplitudes of the received feature signals. Then, the receiver can determine the category of the classified object based on the restored feature data.

[0085] It should be understood that the specific manner in which the receiver can determine the category of the classified object based on the restored feature data will be introduced below and is not limited here.

[0086] It should be understood that in some embodiments, since there is a power limit on the feature signals that the transmitter can send, and the amplitude of the feature signal is positively correlated with the magnitude of the feature data, if the value of the feature data is too large, the transmitter will not have enough power to send the feature signal corresponding to the feature data. Therefore, in some embodiments, the feature data of the task will be adjusted before transmission. In some embodiments, the feature data will be scaled proportionally. After the transmitter sends the feature signal corresponding to the scaled feature data to the receiver. However, since the feature signal is greatly affected by noise, the data obtained by the receiver after restoring the received signal is inaccurate.

[0087] Based on this, the present application proposes a data processing method. In this method, after the transmitter collects the feature data corresponding to the task data, the feature data is adjusted non-linearly before transmission to obtain the adjusted feature data, and the adjusted feature data is sent to the receiver.

[0088] In some embodiments, a power adaptation sub-model capable of performing pre-transmission adjustment can be determined through deep learning. This power adaptation sub-model can non-linearly adjust the feature data. Specifically, the power adaptation sub-model used for pre-transmission adjustment of each node device at the transmitting end is determined in advance through a neural network model. Among them, when training the neural network model, the statistical characteristics of the noise of the channels of each node and the feature data of the samples of all nodes need to be considered. At this time, since the feature data of the samples of all nodes and the noise of the channels of each node are considered, not only the correlation between the data used to perform the same task among the nodes is considered during the training process, but also the adaptation to the noise of the channels of each node is considered. Thus, the anti-noise interference ability of the feature data obtained after pre-transmission adjustment using the corresponding power adaptation sub-model by each node is improved, and further, when the subsequent decision-making center performs a classification task based on the received data, the accuracy of the obtained classification result is improved.

[0089] It can be understood that in a distributed system, since both the transmitting end and the receiving end have strong computing and processing capabilities, when training the neural network model, the neural network model can be trained at the transmitting end or at the receiving end.

[0090] In some embodiments, since information can be transmitted between each node in the distributed system, for convenience, the neural network model can be trained at the transmitting end to determine the power adaptation sub-model required for each node. And when the transmitting end determines the power adaptation sub-model required for each node, the power adaptation sub-model required for each node needs to be sent to the receiving end so that the receiving end can make classification decisions based on the power adaptation sub-model corresponding to each node. In other embodiments, since information cannot be transmitted between each node in the distributed system, for convenience, the neural network model can be trained at the receiving end to determine the power adaptation sub-model required for each node.

[0091] Similarly, when the receiving end determines the power adaptation sub-model required for each node, the power adaptation sub-model required for each node needs to be sent to each node at the transmitting end respectively, so that each node at the transmitting end can perform pre-transmission adjustment according to the corresponding power adaptation sub-model.

[0092] It can be understood that in some embodiments, in order to reduce the data transmission volume and thus save transmission resources, when the transmitting end obtains the power adaptation sub-model corresponding to each node after training through the neural network model, the input-output data pair corresponding to the power adaptation sub-model corresponding to the node can be sent to the receiving end; the receiving end can restore the required power adaptation sub-model according to the received input-output data pair corresponding to the power adaptation sub-model corresponding to the node, so as to be used for subsequent decision-making.

[0093] In some implementations, the decision-making device can input the input-output data pairs corresponding to the power adaptation sub-models of each node into a parametric function model to obtain the solution values of the parameters in the parametric function model corresponding to each node; replace the parameters in the parametric function model of each node with the corresponding solution values to obtain a power adaptation function, and then use the obtained power adaptation functions of each node as the restored power adaptation sub-models. The parametric function model can be, but is not limited to, an exponential model with parameters, a Tanh function model, an Arctan function model, a power series model, etc.

[0094] Similarly, when the receiving end obtains the power adaptation sub-models corresponding to each node after training through a neural network model, it can send the input-output data pairs corresponding to the power adaptation sub-models corresponding to each node to the corresponding nodes in the sending end respectively; the corresponding nodes in the sending end can also restore the required power adaptation sub-models according to the received input-output data pairs corresponding to their own power adaptation sub-models, so as to make adjustments before sending. And each node obtains the corresponding power adaptation function in the same way and uses the obtained power adaptation function as the restored power adaptation sub-model.

[0095] It can be understood that in some embodiments, the statistical characteristics of the channel noise can be a mean vector, covariance, arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, midrange, mode, mean absolute deviation, variance.

[0096] In addition, when the neural network model is trained at the sending end, the sending end also needs to send the statistical characteristics corresponding to each classification label among the classification labels corresponding to multiple classification categories to the decision-making device, where the statistical characteristics corresponding to each classification label among the multiple classification labels are the statistical characteristics of the feature data that are the feature data corresponding to the corresponding classification label in all the feature data of the sample set of all the nodes in the distributed system. For example, it can be the mean vector and covariance of the feature data corresponding to the classification labels corresponding to each classification category, etc., so as to facilitate subsequent decision-making by the decision center. It can be understood that the statistical characteristics corresponding to each classification label among the classification labels corresponding to multiple classification categories can include, in addition to the mean vector and covariance, but are not limited to arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, midrange, mode, mean absolute deviation, variance, etc.

[0097] It can be understood that in some embodiments, the neural network model to be trained can include, but is not limited to, a fully connected neural network, a convolutional neural network, etc.

[0098] In addition, it can be understood that in a classification task, the value range of the numerical values of the feature data can be divided into K feature value intervals, each feature value interval corresponding to a classification category, and the critical value between two adjacent feature value intervals can be called a classification boundary. For example, for a binary classification task, as Figure 2C shown, assuming that the value range of the numerical values of the feature data is [A1, B1], then the value range of the numerical values of the feature data can be divided into 2 feature value intervals, namely the feature value interval [A1, C1] corresponding to category 1 and the feature value interval (C1, B1] corresponding to category 2. The boundary between the feature value interval [A1, C1] and the feature value interval (C1, B1] is C1, that is, C1 is the classification boundary. Another example, for a ternary classification task, as Figure 2D shown, assuming that the value range of the numerical values of the feature data is [A2, B2], then the value range of the numerical values of the feature data can be divided into 3 feature value intervals, namely the feature value interval [A2, C2] corresponding to category 1, the feature value interval (C2, D2] corresponding to category 2, and the feature value interval (D2, B2] corresponding to category 3, where A2 < C2 < D2 < B1. At this time, the classification boundary between the feature value interval [A2, C2] and the feature value interval (C2, D2] is C2, that is, C2 is classification boundary 1; the classification boundary between the feature value interval (C2, D2] and the feature value interval (D2, B2] is D2, that is, D2 is classification boundary 2.

[0099] It should be understood that in some embodiments, due to the limitation of the power of the feature signal that the transmitting end can send, and the amplitude of the feature signal is positively correlated with the size of the feature data. If the numerical value of the feature data is too large, it will cause the transmitting end not to have enough power to send the feature signal corresponding to the feature data. Therefore, in some embodiments, if the value range of the feature data is too large, the transmitting end needs to be able to scale the feature data so that the transmitting end can send all the feature data within the value range of the feature data. For example, assuming that the value range of the feature data is [A1, B1], and the value range of the feature data that the transmitting end can send is [A3, B3], then when the transmitting end sends within the range of [A1, B1], it maps it to the range of [A3, B3] (such as scaling in proportion) and then sends the scaled feature signal.

[0100] It should be understood that if the feature data is scaled, the corresponding classification boundary will also be scaled. Referring to Figure 2E , corresponding to the value range of the feature data being [A1, B1], and the value range of the feature data that the transmitting end can send being [A3, B3], the classification boundary C1 within the value range of [A1, B1] will be scaled to the classification boundary C3 as shown in the figure.

[0101] It can be understood that after the originating end collects the feature data corresponding to the task data, it uses the power adaptation sub-model obtained through neural network training to non-linearly adjust the feature data of the task data, and then obtains the adjusted feature data. In some cases, the scaling ratio between the adjusted feature data and the feature data before adjustment changes with the change of the feature data before adjustment. Assume that the value range of the feature data before adjustment is the first value range, and the value range of the output data of the power adaptation sub-model is the second value range, and the first classification boundary in the first value range (for example, Figure 2E C1 in Figure 2E ) after being input into the power adaptation sub-model, the output data is the second classification boundary (for example, C3 in

[0102] ). At this time, the ratio of the second distance between the adjusted feature data and the second classification boundary to the first distance between the feature data before adjustment and the first classification boundary increases as the first distance decreases. That is, the feature data close to the classification boundary after adjustment moves away from the classification boundary. In this way, the distance between the feature data close to the classification boundary after the original equal-proportion scaling and the classification boundary can be increased, avoiding the numerical value of the feature data close to the classification boundary changing from the feature value interval corresponding to one classification category to the feature value interval corresponding to another classification category due to noise signals during the transmission process.

[0103] For example, in the scenario shown in Figure 1A , the camera A1 extracts features from the images T1 and T2 through the pre-trained LeNet-5 model, and obtains the one-dimensional feature data J1 and J2 corresponding to the images T1 and T2 respectively.

[0104] Among them, since the image T2 is a back view of the classification object G, the male characteristics of the person reflected by the image T2 are not obvious enough, and the obtained one-dimensional feature data J2 will be close to the classification boundary. Referring to the above Figure 2C, assume that category 1 is male and category 2 is female. At this time, the one-dimensional feature data J2 will be close to the classification boundary C1. The one-dimensional feature data J2 is numerically adjusted to obtain the one-dimensional feature data J2'. At this time, the one-dimensional feature data J2' will be farther from the classification boundary C1 than the one-dimensional feature data J2.

[0105] Since the image T1 is a front view of the classification object G, the male characteristics of the person reflected by the image T1 are relatively obvious, and the obtained one-dimensional feature data J1 will be far from the classification boundary C1. In addition, in some cases, after adjustment by the power adaptation sub-model, the feature data far from the classification boundary can also be mapped to the data close to the classification boundary, but the category of the feature data is not changed. For example, the one-dimensional feature data J1 is numerically adjusted to obtain the one-dimensional feature data J1'. At this time, the one-dimensional feature data J1' will be slightly closer to the classification boundary C1 than the one-dimensional feature data J1. However, the value of the one-dimensional feature data J1' will still be greater than that of the one-dimensional feature data J2'. The camera A1 sends the feature signals corresponding to the adjusted one-dimensional feature data J1' and J2' to the base station B. Other cameras can achieve the same effect after processing the collected images and send them to the base station B.

[0106] The base station B receives the signals sent by each transmitting end to obtain the receiving end signal, restores the receiving end signal to obtain the corresponding receiving end data, that is, the image feature data, and performs gender prediction based on the receiving end data from each node to obtain the gender result.

[0107] It can be understood that after adjusting the feature data at the classification boundary, the anti-interference ability of the corresponding feature signal during transmission can be improved, which can avoid the problem that in some scenarios, the receiving end determines the task result based on the data that is more critical for classification, and due to the greater influence of noise, the accuracy is not high, resulting in the accuracy of the determined task result being affected. And in a distributed scenario, the transmitting end (such as each node) sends the feature signal corresponding to the task content to the receiving end (such as the decision center). At this time, when the receiving end makes a decision, it fully considers the relevance of the data of each node, avoids the obtained decision result being too one-sided, and also avoids the problem that each node needs to process the collected task data to obtain its own task result, which consumes more resources when the number of nodes is large.

[0108] It can be understood that during the task execution process, before the feature data corresponding to the task data collected by the sender is sent, it is adjusted through the corresponding power adaptation sub-model to obtain the adjusted feature data. At this time, the signal corresponding to the adjusted feature data will be sent to the receiver. After the receiver receives the feature signals sent by each sender, the receiver can receive the signals from each sender (hereinafter referred to as "receiver signals" for convenience of description), and restore the receiver signals to obtain the corresponding data (for convenience of description, the data corresponding to the "receiver signals" is simply referred to as "receiver data"), and make a decision based on the receiver data from each node, such as performing class prediction to obtain a decision result. It should be understood that in some cases, due to the non-linear adjustment through the power adaptation sub-function, and the adjusted feature data still belongs to the same classification category, the receiver uses the receiver data affected by noise interference to make a decision, and the accuracy of the obtained decision result is relatively high.

[0109] For the convenience of understanding, the following will first take a distributed system including 2 nodes and 1 decision center as an example, and after determining the power adaptation sub-models of each node, each node uses the determined power adaptation sub-model to adjust the feature data, and the specific process of the decision center making a decision based on the data corresponding to the received signal will be introduced.

[0110] Figure 3A According to some embodiments of the present application, a schematic diagram of a framework system 300 is shown in which two nodes are transmitted based on their respective corresponding power adaptation sub-models to the decision center 00, enabling the decision center 00 to make a decision.

[0111] As Figure 3A shown, node 0 includes a feature extraction module 301 and a power adaptation module 302. Similarly, node 1 includes a feature extraction module 311 and a power adaptation module 312. Among them, the feature extraction module 301 is used to extract features from the task data obtained by node 0 based on a feature extractor (such as a pre-trained model) to obtain feature data (such as x 0 ) shown in the figure). The power adaptation module 312 can perform numerical adjustment on different input feature data based on the determined power adaptation sub-model corresponding to node 0 (such as the power adaptation sub-model f 0 ) corresponding to node 0 to obtain the adjusted feature data (such as the adjusted feature data x 0 ') shown in the figure). For example, the feature data close to the classification boundary can be adjusted to be farther from the classification boundary than before, and the classification category remains unchanged.

[0112] The feature extraction module 311 and the power adaptation module 312 are substantially the same as the feature extraction module 301 and the power adaptation module 302 respectively, and will not be described in detail here.

[0113] It can be understood that the power adaptation modules of each node (such as the power adaptation module 302 and the power adaptation module 312 above) are based on the corresponding power adaptation module. After numerically adjusting each piece of feature data, the average power of the transmitted adjusted data will not exceed the power allowed by the node (such as the rated power). For different nodes, due to the different distributions of the feature data of each node and the allowed power (hereinafter referred to as "allowed power"), each adaptation module will process based on different power adaptation sub-models, and after adjusting the corresponding feature data, there will be a certain range.

[0114] For node 0, the feature signal corresponding to the adjusted feature data will be transmitted to the decision center 00. At this time, the decision center 00 will receive the received signal W0 from node 0. The received signal W0 not only includes the adjusted feature signal but also includes, for example, Figure 3A the channel noise Z shown in 0 (Z 0 ~N(0, σ 2 ²)). Similarly, for node 1, the decision center 00 can receive the received signal W1 from node 1. After the decision center 00 receives the signals of each node, it makes a decision using the data corresponding to the received signals to obtain a decision result. For example, taking a two-classification task as an example, the decision center 00 calculates the likelihood function value using the data corresponding to the received signals and determines the specific category according to the likelihood function value. Next, in combination with Figure 3A the framework system 300 shown, the technical solution of the present application will be introduced.

[0115] Figure 3B According to some embodiments of the present application, a schematic diagram of the process in which node 0 and node 1 adjust the feature data corresponding to the task data and then send it to the decision center 00, and the decision center 00 makes a decision is shown.

[0116] S101, Node 0 collects task data R0.

[0117] In some embodiments, node 0 collects task data R0 related to the target task (i.e., the data to be classified). Among them, the task data R0 can include multiple task data or can be a single task data.

[0118] For example, in the scenario shown above Figure 1A , node 0 collects image data a0 for a two-classification task. Among them, the image data a0 is the data corresponding to one image or multiple images, that is, the task data R0 is the image data a0.

[0119] S102, Node 0 extracts features from the task data R0 to obtain feature data x 0 .

[0120] In some embodiments, node 0 performs feature extraction on task data R0 through feature extraction module 301 to obtain feature data x 0 (i.e., as classification feature data).

[0121] In some embodiments, the feature data x 0 is one-dimensional data, i.e., a single value. For example, when the task data R0 is multiple data, after performing feature extraction on the task data R0, multiple one-dimensional data will be obtained, i.e., the feature data x 0 is multiple data.

[0122] For example, node 0 performs image feature extraction on image data a0 corresponding to multiple images through the LeNet-5 model to obtain multiple one-dimensional image feature data m0, i.e., each one-dimensional image feature data m0 corresponds to one image, and the one-dimensional image feature data m0 corresponding to the image data a0 is the feature data x 0 .

[0123] S103, node 0 adjusts the feature data x 0 based on the power adaptation sub-model f 0 to obtain the adjusted feature data x 0 '.

[0124] It can be understood that the power adaptation sub-model f 0 can adjust the corresponding input data according to its situation with respect to the classification boundary.

[0125] In some embodiments, the feature data x 0 includes multiple feature data. When multiple feature data x 0 with different values are input into the power adaptation module 302, adjusted feature data x 0 ' corresponding to each input feature data will be output. Since the power adaptation sub-model f 0 used in the power adaptation module 302 is a non-linear model. For example, referring to the above Figure 2C , assuming the classification boundary is C1, at this time, the feature data x 0 with a value of b is on the right side of C1 and relatively close to the classification boundary C1. At this time, through the power adaptation sub-model f 0 , it will be mapped to a value farther from C1. However, at this time, the feature data x 0 still belongs to category 2.

[0126] For example, when multiple one-dimensional image feature data m0 with different values are input into the power adaptation module 302, multiple adjusted image feature data m0' corresponding to each input can be output.

[0127] S104, Node 0 transmits the adjusted feature data x to Decision Center 00 0 ' and the corresponding feature signal V0.

[0128] In some embodiments, Node 0 transmits the adjusted feature data x to Decision Center 00 through a radio frequency device 0 ' and the corresponding feature signal V0, and the feature signal V0 is subject to channel noise Z during transmission 0 interference.

[0129] For example, Node 0 transmits the image feature signal corresponding to the adjusted image feature data m0' to Decision Center 00 through a radio frequency device, and the image feature signal is subject to channel noise Z during transmission 0 interference.

[0130] S105, Decision Center 00 obtains the received signal W0 corresponding to the adjusted feature data x 0 '.

[0131] In some embodiments, Decision Center 00 obtains the received signal W0 corresponding to the adjusted feature signal x 0 ', and the received signal W0 includes the feature signal V0 corresponding to the adjusted feature data x 0 ' and channel noise Z 0 .

[0132] For example, Decision Center 00 obtains the received signal corresponding to the adjusted image feature data m0', and the received signal includes the adjusted image feature signal and channel noise Z 0 .

[0133] S106, Node 1 collects task data R1.

[0134] S107, Node 1 extracts features from the task data R1 to obtain feature data x 1 .

[0135] S108, Node 1 adjusts the feature data x based on the power adaptation sub-model f 1 , to obtain the adjusted feature data x 1 '. 1 '.

[0136] S109, Node 1 transmits the feature signal V1 corresponding to the adjusted feature data x 1 ' to Decision Center 00.

[0137] S110, Decision Center 00 receives the received signal W1 corresponding to the feature data x 1 '.

[0138] It can be understood that the above S106 - S110 is substantially the same as the above S101 - S105, and will not be elaborated here.

[0139] S111. The decision - making center 00 makes a decision based on the received - end data s0 and s1 corresponding to the received - end signal W0 and the received - end signal W1 respectively, and obtains a decision result.

[0140] In some embodiments, the decision - making center 00 performs class prediction based on the received - end data s0 and s1 corresponding to the received - end signal W0 and the received - end signal W1 respectively, and obtains a decision result according to the predicted class. It can be understood that since the received - end signal W0 and the received - end signal W1 are an electromagnetic wave, it is necessary to demodulate the received - end signal W0 and the received - end signal W1 to obtain the corresponding received - end data s0 and s1.

[0141] In some implementation manners, calculate the likelihood function value of the received - end data (s 0 , s 1 , …, s d-1 ), and perform class prediction according to the likelihood function value. Calculate the likelihood function values corresponding to each label (i.e., classification category), and take the category with the larger likelihood function value as the decision result.

[0142] It can be understood that when the decision - making center 00 calculates the likelihood function value of the received received - end data, it is necessary to consider the statistical characteristics of the noise and the statistical characteristics of the characteristic data transmitted by each node. The following formula (1) shows an expression for calculating the likelihood function value corresponding to each classification category according to the received - end data (s 0 , s 1 , …, s d-1 ) obtained in a classification scenario.

[0143]

[0144] Where is the characteristic data generated according to the distribution N(μ k , ∑ k ) given by the label k, that is, the simulated characteristic data obtained according to the distribution situation corresponding to the statistical characteristics of the training samples. Among them, the label k is the number of the specific classification category. For example, k = 1 corresponds to category 1, which is male, and k = 2 corresponds to category 2, which is female. The μ k and ∑ k in this distribution N are the mean vector and covariance of the characteristic data corresponding to each label obtained based on the training samples of all nodes during the training process respectively, σ 2 is the variance of the channel noise observed by the decision - making center 00, and f j (*) is the power adaptation sub - model corresponding to each node. The specific explanation will be introduced in the following training process and will not be described in detail here.

[0145] For example, in the above binary classification scenario, when the likelihood function value corresponding to the male category obtained by the decision center 00 based on the received data s0 and the received data s1 is 0.4, and the likelihood function value corresponding to the female category is 0.7, then it is determined that the corresponding category is female, that is, the decision result is female.

[0146] It can be understood that the execution order of the above steps S101 to S1011 is only for illustration. In some other embodiments, other execution orders can also be adopted, and some steps can also be split or combined, which are not limited herein. For example, the execution order of the process where the above node 1 executes S106 to S109 and the process where node 0 executes S101 to S104 can be changed, that is, node 1 executes S106 to S109 before node 0, or can be executed in parallel with node 0, and there is no strict order in the execution process. The order in which the decision center 00 receives signals from node 0 and node 1 is not fixed either, and it can also receive them simultaneously.

[0147] It can be understood that after processing the feature data through the power adaptation sub-model, the anti-interference ability of the feature data during transmission can be improved, that is, its transmission quality can be improved, and it can avoid the problem that in some scenarios, the received important feature data has poor anti-noise interference ability and low accuracy, resulting in the accuracy of the determined task result being affected. This can save resources while ensuring a high accuracy rate as much as possible. Moreover, the decision center utilizes the correlation between the data of each node during decision-making to help the decision center make better decisions.

[0148] It can be understood that during the above task execution process, if the neural network model training is carried out at the transmitting end in advance, and the transmitting end sends the first adaptation processing information to the receiving end, where the first adaptation processing information includes the input-output data pairs corresponding to each node, then the receiving end will use the power adaptation function obtained according to the input-output data pairs as the restored power adaptation sub-model to make decisions, that is, the power adaptation sub-model used for likelihood function calculation in the above step S111 is the power adaptation function.

[0149] It can be understood that during the above task execution process, if the neural network model training is carried out at the transmitting end in advance, and the transmitting end sends the second adaptation processing information to the receiving end, where the second adaptation processing information includes the power adaptation sub-models corresponding to each node, then the receiving end will directly use the power adaptation sub-models to make decisions, that is, the power adaptation sub-model used for likelihood function calculation in the above step S111 is the received power adaptation sub-model.

[0150] It can be understood that during the execution of the above task, if the neural network model is trained at the receiving end in advance, and the receiving end sends third adaptation processing information to each sending end (i.e., each node device) respectively, where the third adaptation processing information includes the input-output data pairs of the power adaptation sub-model corresponding to the node. At this time, each sending end will obtain the power adaptation function according to the input-output data pairs, and use the power adaptation function as the power adaptation sub-model to adjust the feature data before sending. That is, the power adaptation sub-model used in the above steps S103 and S108 is the power adaptation function.

[0151] It can be understood that during the execution of the above task, if the neural network model is trained at the receiving end in advance, and the receiving end sends fourth adaptation processing information to each sending end (i.e., each node device) respectively, where the fourth adaptation processing information includes the power adaptation sub-model corresponding to the node. At this time, each sending end will adjust the feature data before sending according to the obtained power adaptation sub-model. That is, the power adaptation sub-model used in the above steps S103 and S108 is the received power adaptation sub-model.

[0152] The following specifically introduces the process of determining the power adaptation sub-model of each node according to the deep learning method.

[0153] For the convenience of elaboration, the following still takes a distributed system including 2 nodes and 1 decision center as an example for specific introduction.

[0154] The following combines the attached Figures 4A to 6C Elaborate on the determination method of the power adaptation sub-model.

[0155] In some embodiments, when adopting deep learning technology, a neural network model is used for training, where the neural network model includes the power adaptation sub-models corresponding to each node. It can be understood that after training the neural network model, the trained neural network model can be obtained, thereby determining the power adaptation sub-models corresponding to each node.

[0156] It can be understood that as mentioned above, for the training stage, there will be two situations:

[0157] (1) Information can be exchanged between nodes. For example, for Figure 1A the shown cameras, due to the short distance, signal interaction can be achieved, such as through information sharing. For the convenience of elaboration, the following will call a distributed system including the required decision center and each node that can exchange information with each other a "decentralized distributed system".

[0158] (2) Information cannot be exchanged between nodes. For example, for Figure 1AInformation interaction is not allowed between the cameras shown. For the convenience of description, the distributed system including the required decision-making center and each node that cannot perform information interaction will be referred to as a "centralized distributed system" in the following text.

[0159] (1) Decentralized Distributed System

[0160] For the above situation (1), since information interaction is allowed between nodes, at this time, the node (i.e., the transmitting end) can obtain the characteristic data of all nodes, and thus perform neural network model training at the transmitting end to determine the required power adaptation sub-model.

[0161] Combined with Figures 4A - 6C First, the neural network model training is carried out under the "decentralized distributed system" to determine their respective power adaptation sub-models. And after the power adaptation sub-model is determined at the node (i.e., the transmitting end), the power adaptation sub-model needs to be synchronized at the receiving end so that the receiving end can make decisions during use. Referring to Figure 3B In step S111 of, the receiving end (i.e., decision-making center 00) needs to use the trained power adaptation sub-models of each node for class prediction.

[0162] Figure 4A According to some embodiments of the present application, a framework schematic diagram 400 of a decentralized distributed system is shown. The Figure 4A For the convenience of comparison with Figure 3A the framework schematic diagram of the actual use process after training shown, still taking Figure 3A two nodes and one decision-making center shown as an example. Among them, information interaction is allowed between the two nodes.

[0163] As Figure 4A shown, node 0 includes a feature extraction module 301a and a power adaptation module 302a. Node 1 includes a feature extraction module 311a and a power adaptation module 312a.

[0164] Among them, in some embodiments, the feature extraction module 301a is used to perform feature extraction on each sample data in the sample set of node 0 (i.e., sample 0 in the figure) based on a feature extractor (such as a pre-trained model) to obtain feature data. And since each sample data in the sample set has a classification label, the corresponding feature data has a corresponding classification label. At this time, label feature data x corresponding to each classification label can be obtained k；0 The subscript 0 in represents node 0, and the subscript k represents the number of the classification label corresponding to the classification category. For example, taking a binary classification task as an example, k = 1 and 2, k = 1 represents male, and k = 2 represents female. It can be understood that the feature extractor used here is the same as the Figure 3A feature extractor used above.

[0165] The power adaptation module 302a includes a neural network model that needs to be trained. Before training, the parameters in the neural network model can be any values. After training, the parameters in the neural network model are values that make the feature data of each node 0 match the noise of the channel. The power adaptation module 302a can perform a non-linear adjustment on the labeled feature data x k；0 to obtain the adjusted labeled feature data x' k；0 . For example, as Figure 4B shown, the labeled feature data x of each node k；j is non-linearly adjusted through the power adaptation sub-model to obtain f j (x k；j ). Where j is the number of each node. For node 0, it is 0. f j (x k；j ) is the adjusted labeled feature data x' k；0 .

[0166] It can be understood that the functional essence of the feature extraction module 311a and the power adaptation module 312a is the same as that of the feature extraction module 301a and the power adaptation module 302a, and will not be elaborated here.

[0167] It can be understood that for the case where information can be exchanged between each node, since data can be exchanged between each node, each node can receive the feature data from other nodes, so each node can obtain the feature data of all nodes. At this time, in order to save resources, as long as one node is selected to obtain the feature data of all nodes and the neural network model is trained according to the feature data, the power adaptation sub-model corresponding to each node can be determined. And, in some implementation manners, each node will transmit the obtained power adaptation sub-model of its own to the corresponding decision center 00, so that the decision center 00 can be used in subsequent decisions. In other implementation manners, it can also be the node (assumed to be Figure 4A node 0 in) that performs the training process to transmit the power adaptation sub-model corresponding to each node to the corresponding decision center 00.

[0168] Specifically, taking the specific training at node 0 as an example to elaborate the training process, the following Figure 5A According to some embodiments of the present application, a schematic diagram of the process of node 0 receiving the labeled feature data from node 1 and the statistical characteristics (such as noise variance) of the noise sent by the decision center and performing neural network model training is shown. Among them, steps S201 - S205 are the specific training processes.

[0169] S201, node 0 extracts features from sample 0 to obtain the labeled feature data x corresponding to each label k；0 .

[0170] It is understandable that each node will obtain a large amount of sample data for the same classification task, and each sample number will correspond to a classification label. Moreover, each node can obtain one-dimensional feature data of the sample data through a pre-trained feature extractor (for example, for image tasks, an existing LeNet-5 model can be used). This one-dimensional feature data has determined the corresponding classification category, thereby removing redundant information in the sample data. It is understandable that the feature extractor used here is the same as the feature extractor in the actual use process after training is completed.

[0171] For example, as Figure 5B shown, each node j will perform feature extraction on its respective sample data, and each sample data will have a corresponding label, obtaining label feature data x corresponding to each label k；j , where the subscript k represents the category of the task label, and j represents the node number.

[0172] In some embodiments, node 0 can obtain sample data related to the task (i.e., sample 0). Each sample data has a corresponding label. The obtained sample data is subjected to feature extraction using the feature extraction module 301a to obtain feature data. Moreover, according to the task label information, label feature data x corresponding to each task label is obtained k；0 , where the subscript k represents the category number of the task label, and the subscript 0 represents node 0. Taking a binary classification task as an example, k can be 1 and 2.

[0173] For example, in Figure 1A the scenario shown, node 0 obtains a large number of image sample data 0. Each image sample data corresponds to an image, and each image has a corresponding classification category, for example, male or female. Node 0 performs feature extraction on multiple image samples 0 to obtain multiple one-dimensional feature data, such as one one-dimensional feature data corresponding to one image data. At this time, according to the labels corresponding to each image sample 0, the multiple one-dimensional feature data are divided into 2 categories, obtaining label feature data x corresponding to the label male (for example, k = 1) 0；0 ; and obtaining label feature data x corresponding to the label female (for example, k = 2) 1；0 .

[0174] S202. Node 1 performs feature extraction on sample 1 to obtain label feature data x corresponding to each label k；1 .

[0175] It is understandable that for node 1, this process is substantially the same as the process through step S201, and will not be elaborated here.

[0176] S203, Node 1 sends the first feature information to Node 0, where the first feature information includes label feature data x k；1 .

[0177] It can be understood that data interaction can occur between Node 0 and Node 1. As Figure 5C shown, each node j can interact with other nodes j' for label feature data. Node j sends its own label feature data x k；j to other nodes j', and other nodes j' can send the label feature data x k；j to Node j.

[0178] For the interaction between Node 1 and Node 0, as Figure 4A shown, Node 0 and Node 1 can send the first feature information to each other respectively. The first feature information corresponding to Node 0 includes the obtained label feature data x k；0 , and the first feature information corresponding to Node 1 includes the obtained label feature data x k；1 . Since taking the neural network model training on Node 0 as an example, Node 1 will send the label feature data x k；1 to Node 0 so that Node 0 can perform training.

[0179] In some implementation manners, it is default that the interaction between nodes is not affected by channel noise. For example, other data sharing methods can be used for transmission.

[0180] S204, Decision Center 00 sends the first noise statistic information to Node 0, where the first noise statistic information includes the statistical characteristics of the noise corresponding to each node.

[0181] It can be understood that since the purpose of training the neural network model is to adapt to channel noise, channel noise needs to be considered during the training of the neural network model. Since only Decision Center 00 can observe the channel noise, Decision Center 00 will send the statistical characteristics corresponding to the channel noise, for example, variance σ 2 to each node. Referring to Figure 5D , Decision Center 00 sends the variance σ 2 of the noise to Node j, where Node j can be any node.

[0182] For example, taking the channel noise as Gaussian noise, Figure 4A Decision Center 00 sends the variance σ 2 corresponding to the channel noise to each node.

[0183] It can be understood that in this embodiment, the neural network model is trained on Node 0. At this time, Decision Center 00 can send only the noise variance σ 2. It is understandable that the decision center 00 can also send the noise variance σ to all nodes. 2 , and no specific requirements are made here. Then, node 0 can receive the channel noise variance required for training. It is understandable that the variances of different channel noises can be different values. For the convenience of description, in some embodiments of the present application, hereinafter, it is assumed that the variance value of each channel noise is the same σ 2 value is taken as an example for description.

[0184] S205. Node 0 trains the neural network model according to the preset loss function and the label feature data of each node to obtain the trained neural network model (i.e., the power adaptation sub-model corresponding to each node), where the loss function involves the allowable power of each node and the statistical characteristics of the noise.

[0185] It is understandable that the received data corresponding to each node received by the simulation decision center 00 is determined according to the label feature data and the statistical characteristics of the channel noise (e.g., variance). Since the received signal of the decision center 00 includes the label feature signal after each node adjusts the label feature data using the power adaptation sub-model and the channel noise signal, the simulated received data includes the label feature data processed by the power adaptation sub-model and the data corresponding to the noise signal. For the specific method of obtaining the simulated received data of the decision center 00, refer to the description of the following formula (3), and no further details are provided here.

[0186] At this time, the mutual information between the simulated received data and the classification label can be maximized as the objective function to train the neural network model to obtain the power adaptation sub-model corresponding to each node. For the specific description of the mutual information, refer to the description of the following formula (2), and no further details are provided here.

[0187] Moreover, since the power of each node is determined, it is necessary to consider that when each node transmits the signal corresponding to the adjusted feature data, the power corresponding to the transmitted feature signal cannot exceed the allowable power. Therefore, a penalty is added to the objective function of the corresponding mutual information to constrain the power. For the specific description of the mutual information under power constraint, refer to the descriptions of the following formulas (4) to (6), and no further details are provided here.

[0188] Specifically,

[0189] ① Denote the power adaptation sub-model of each node as f j , j is the number of each node. For example, j = 0,..., d - 1, and j is a natural number. Denote the mutual information between the corresponding task label and the received data of the decision center as I(H; s 0 ,..., s d-1 ).

[0190] The following formula (2) shows an expression for maximizing mutual information:

[0191]

[0192] where I(*) represents mutual information, and s 0 , …, s d-1 represents the received data of each node received by the decision center, and f j is the neural network model that maximizes I, that is, the power adaptation sub-model corresponding to each node.

[0193] ② Based on the label feature data and the statistical characteristics of the channel noise (e.g., the variance of the noise), the theoretically received data (i.e., the data corresponding to the signals of each node received by the decision center) can be obtained as s 0 , …, s d-1 . Among them, each s j (j = 0,......, d - 1) includes the received data s k；j corresponding to each label.

[0194] For example, the following formula (3) shows the expression of the received data s k；j corresponding to each label:

[0195] s k；j = f j (x k；j ) + z j , z j ~N(0, σ 2 ) Formula (3)

[0196] where z j represents the noise data, k is the label category, that is, s k；j represents the data obtained after the label feature data of each node is affected by the power adaptation sub-model and interfered by the channel noise.

[0197] ③ Based on the introductions in ① and ② above, a loss function under power constraint for each node can be obtained as shown in the following formula (4), where the power is constrained by adding a penalty term to the loss function, and the loss function is

[0198]

[0199] where is the specific expression formula of the above I(H; s 0 , …, s d-1 ). represents the power used when the transmitting end (i.e., each node) transmits the signal corresponding to the label feature data adjusted by the power adaptation sub-model, and p jIndicates the allowable power of each node, that is, the power allowed to pass through the channel. The weight λ j is a hyperparameter that needs to be debugged, and under the weight λ j on the training set is not greater than p j .

[0200] The explanations of each term in the loss function shown in the above formula (4) are as follows:

[0201] (1) Introduction to the mutual information term . The mutual information term reflects the mutual information between the label and the received data at the receiver. It is calculated through the label feature data x of each label k k；j and the received data s of the corresponding label feature data x k；j . k；j Calculated.

[0202] Assume that the corresponding label feature data x of each node k；j and the received data s at the receiver k；j respectively include N f,k is the different data in the label feature data of each node under label k. N s,k is the different data in the received data of each node at the receiver under label k.

[0203] The following formula (5) is the specific expansion formula of in the above formula (4):

[0204]

[0205] Among them, among them The sequence of and

[0206] Among them, k is the category number of the label, K is the total number of labels, and is the number of the last label.

[0207] (2) Introduction to the power constraint term . Indicates that when the power used is less than the allowable power p j , at this time this term is 0, then there is no power constraint, and at this time the objective function is mutual information.

[0208] When the power used is greater than the allowable power pj, the power constraint term is not 0, then a penalty is required, and at this time the loss function includes this power constraint term.

[0209] Specifically, subtract from the upper limit p j and if the loss is positive, otherwise it is zero.

[0210] The following formula (6) shows the expression of the usage power of

[0211]

[0212] Figure 5E According to some embodiments of the present application, a schematic diagram of the training process of a neural network model including each power adaptation sub-model is shown. This process is described by taking node 0 as the execution subject.

[0213] S205A: Use the labeled feature data as the input data of the neural network model to obtain the model output result, and calculate the loss function.

[0214] In some embodiments, initialize the neural network model including each power adaptation sub-model, and input the labeled feature data into the neural network model to obtain the model output result. Calculate the loss function according to the model output result and the labeled feature data.

[0215] For example, calculate the loss function value corresponding to the above formula (5) according to the model output result and the labeled feature data.

[0216] S205B: Iterate the neural network model for a preset number of rounds, and stop training after reaching the preset number of training rounds.

[0217] In some embodiments, the parameters in the model can be iterated using stochastic gradient, and stop training after reaching the preset number of training rounds (e.g., 50 times).

[0218] S205C: Use the neural network model with the minimum loss function value in each round as the trained neural network model, where the trained neural network model includes each required power adaptation sub-model.

[0219] In some embodiments, use the neural network model corresponding to the round with the minimum loss function value as the trained neural network model, where the trained neural network model includes each required power adaptation sub-model.

[0220] ​It can be understood that the loss function is calculated based on the label feature data, noise variance, and allowable power of each node corresponding to the label. The neural network model corresponding to the round with the minimum loss function value is used as the trained neural network model, so as to obtain the rate adaptation sub-model corresponding to each node. It can be understood that when the loss function is minimized, the mutual information is as large as possible.

[0221] It can be understood that the execution order of the above steps S201 to S205 is only for illustration. In some other embodiments, other execution orders can also be adopted, and some steps can be split or combined, which will not be limited here.

[0222] It can be understood that during training, the feature data of the samples of each node are all involved in the training process, and the power adaptation sub-models corresponding to each node are jointly trained. Therefore, the correlation between the signals of each node is considered during the training process, making the obtained rate adaptation sub-models of each node more scientific and effective.

[0223] In addition, when node 0 obtains the label feature data of all nodes, the joint distribution of the label feature data corresponding to all nodes can be obtained. Specifically, the statistical characteristics of the label feature data corresponding to all nodes can be obtained. For example, the mean vector μ k , covariance ∑ k . Where k is the number corresponding to the label category. For example, in the case of binary classification, when k = 1, the mean vector μ 0 , covariance ∑ 0 of the label feature data corresponding to all nodes and corresponding to k = 1 are obtained. The situation is essentially the same when k = 2 and will not be elaborated here.

[0224] When node 0 obtains the statistical characteristics of the label feature data corresponding to all nodes, such as the mean vector μ k , covariance ∑ k , it can send the first label feature statistical information to the decision center. The first label feature statistical information includes the statistical characteristics of the label feature data of all nodes. Referring to the above Figure 4A , for node 0, the statistical characteristics of the label feature data, such as the mean vector μ k , covariance ∑ k are sent to the decision center, so that the decision center 00 generates simulated data for decision-making during use based on the mean vector μ k , covariance ∑ k , for example, referring to the introduction in step S111 above. It can be understood that since the data between each node can be transmitted, other nodes, such as node 1 shown in the figure, can also send the mean vector μ k, covariance ∑ k Send it to the decision center 00, which is not required here.

[0225] The power adaptation sub-model will be introduced for synchronization at the receiving and transmitting ends below.

[0226] After the power adaptation sub-model is determined at the node (i.e., the transmitting end), the power adaptation sub-model needs to be synchronized at the receiving end so that the receiving end can make decisions during use.

[0227] In some embodiments, the node may send second adaptation processing information to the decision center 00. The second adaptation processing information includes each power adaptation sub-model, so that the decision center 00 can obtain the required power adaptation sub-model. It can be understood that after the node 0 determines the power adaptation sub-models corresponding to each node, it will send the corresponding power adaptation sub-models to other nodes (such as node 1). And, in some implementation manners, the node 0 may send the power adaptation sub-models corresponding to all nodes to the decision center 00 for use in the decision-making process. In other implementation manners, each node may send the received power adaptation sub-models corresponding to itself to the decision center 00 for use in the decision-making process. The specific manner is not limited here.

[0228] In other embodiments, the data volume corresponding to each power adaptation sub-model is relatively large. To save transmission resources, multiple groups of input-output data pairs (i.e., key points) related to the neural network model (i.e., the determined power adaptation sub-model) can be selected and transmitted to the decision center 00. The decision center 00 reconstructs the required power adaptation function according to the key points. Specifically, the node may send first adaptation processing information to the decision center 00. The first adaptation processing information includes multiple groups of input-output data pairs, and each group of input-output data pairs corresponds to a power adaptation sub-model of a node, so that the decision center 00 can restore the required power adaptation sub-model.

[0229] Figure 6A According to some embodiments of the present application, a schematic diagram of the process of transmitting the power adaptation sub-model through key point information is shown, where, in Figure 4A the non-centralized distribution system shown, taking the node 0 transmitting the key point data of the power adaptation sub-models corresponding to each node to the decision center 00, and then the decision center 00 restoring each power adaptation sub-model according to the received key point information to obtain the power adaptation sub-models corresponding to each node as an example for introduction.

[0230] S301, the node 0 determines the key point information corresponding to the input-output of the power adaptation sub-models of each node.

[0231] In some embodiments, it is necessary to determine the number of key points to be transmitted. According to the preset number of key points and the trained power adaptation sub-model, key point information of the preset number is obtained.

[0232] Specifically, preset the number of key points to be transmitted, obtain the curves corresponding to the power adaptation sub-models of each node, and according to the curves corresponding to the trained power adaptation sub-models, obtain the corresponding input-output key point coordinates (x i , y i ) of each node as the key point information for transmission.

[0233] It can be understood that the following formula (7) shows a general model of a power adaptation sub-model as:

[0234] y = f j (x; θ) Formula (7)

[0235] Where y is the output data after the action of the power adaptation sub-model f j , x is the input data to be adapted, j represents the node number corresponding to the current power adaptation sub-model, and the parameter θ is the parameter of the neural network model. At this time, select the preset number of key points to obtain the corresponding input-output key point coordinates (x i , y i ), i represents the subscript of the key point coordinates, and send them to the decision center 00.

[0236] For example, for the power adaptation sub-model of node 0 is f 0 , the power adaptation sub-model f 0 is a non-linear function processing, and the power adaptation sub-model will increase the data close to the classification boundary and decrease the data far from the classification boundary, and the size order of the adjusted data remains the same as that of the data before adjustment.

[0237] Figure 6B According to the embodiments of the present application, a schematic diagram of the curve corresponding to the power adaptation sub-model f 0 is shown. The horizontal asymptote ordinates of the input-output curve of the current power adaptation sub-model f 0 are m and M (m < M) respectively, that is, m is the minimum value of the output value and M is the maximum value of the output value.

[0238] Taking the preset number of key points as 15 as an example, 15 key point coordinates of the input-output data corresponding to the power adaptation sub-model f 0 can be obtained as (x 1 , y 1 ), (x 2 , y 2 ), (x 3 , y 3),..., (x 15 , y 15 ). It can be understood that 15 key points can be selected by any existing method, such as equidistant sampling of the abscissa, equidistant sampling of the ordinate, etc., and no requirements are made here.

[0239] In some implementation manners, the selection can be performed by the following selection method. For example, within the range of [m, M] on the vertical axis, 15 key points are selected at equal intervals. Figure 6B The 15 key points (x 1 , y 1 ) obtained by the foregoing method are shown in 2 , y 2 ), (x 3 , y 3 ),..., (x 15 , y 15 ). Specifically, since the differences between the output values are the same, the following formula (8) shows the expression of the ordinate yi corresponding to the 15 key points.

[0240]

[0241] Where i = 1, 2,..., 15, corresponding to the numbers of the respective key points.

[0242] The following formula (9) shows the expression of the abscissa x 1 , …, x 15 of the 15 key points obtained according to the above formula (8):

[0243]

[0244] S302. Node 0 sends first adaptation processing information to decision center 00, where the first adaptation processing information includes the key point information of the power adaptation sub-model corresponding to each node.

[0245] In some embodiments, node 0 sends multiple groups of key point coordinates to decision center 00, each group corresponding to a power adaptation sub-model and each group including the coordinates of 15 key points. For example, Figure 6C As shown, node j will send the key point information of each power adaptation sub-model to decision center 00. When the node is 0, it means that node 0 sends the key point information of the power adaptation sub-model of each node to decision center 00.

[0246] S303. Decision center 00 reconstructs the power adaptation sub-model according to the key point information to obtain the restored power adaptation sub-models.

[0247] In some embodiments, for each power adaptation sub-model to be restored, a parameterized model is selected. For example, a parameterized model can be any one of an exponential model, a Tanh function model, an Arctan function model, and a power series model. The power adaptation function is obtained by solving each parameterized model according to the key point information, and the solved power adaptation function is used as the restored power adaptation sub-model.

[0248] (1) Obtain the parameterized model.

[0249] Some specific parameterized models will be introduced below.

[0250] Specifically, the following formulas (10) to (13) respectively show the expressions of the exponential model, the Tanh function model, the Arctan function model, and the power series model:

[0251] ①. Exponential model

[0252]

[0253] ②. Tanh function model

[0254]

[0255] ③. Arctan function model

[0256]

[0257] ④. Power series model

[0258]

[0259] Among them, a l , b l , c l in the above formulas (10) to (13) are all parameters to be solved. And, l is the parameter subscript, the value of m can represent the number of parameters, and m is a positive integer.

[0260] For example, an exponential model can be selected as the parameterized model for subsequent calculations.

[0261] (2) According to the received preset number of key point information and the preset parameterized model, obtain the power adaptation function, and use the obtained power adaptation function as the restored power adaptation sub-model.

[0262] The received key point values (x i , y i ), i = 1,..., e (the number e depends on the number of parameters of the parameterized model) can be substituted into the parameterized model, and the equations are solved to obtain each parameter to be solved in the parameterized model.

[0263] For example, taking the exponential model as an example, the following formula (14) shows an exponential parametric model including 15 parameters to be solved.

[0264]

[0265] At this time, according to the coordinates of 15 key points (x 1 , y 1 ), (x 2 , y 2 ), (x 3 , y 3 ),......, (x 15 , y 15 ) and substituting them into the above formula (14), that is, referring to the following formula (15), a set of feasible solutions for a l , b l , c l can be obtained, and thus the restored power adaptation function can be obtained.

[0266]

[0267] In addition, combined with the above formula (8), formula (15) can be transformed into the following formula (16). At this time, only the coordinates x i of 15 key points need to be substituted into the following formula (16)

[0268]

[0269] to obtain a set of feasible solutions for the parameters a l , b l , c l . Substituting this set of feasible solutions into the parametric model f(x; a l , b l , c l ) gives the power adaptation function at the receiving end.

[0270] It can be understood that the execution order of the above steps S301 to S303 is only for illustration. In some other embodiments, other execution orders can also be adopted, and some steps can also be split or combined, which is not limited here.

[0271] It can be understood that in the embodiments of the present application, when it is necessary to synchronize the power adaptation sub-model between the node and the decision center, only a small amount of data needs to be transmitted to achieve the synchronization of the power adaptation sub-model. Especially in the scenario where the number of nodes is large and the data volume of the power adaptation sub-model is large, the synchronization of the power adaptation function sub-model can be effectively achieved by transmitting relatively few curve key points, thereby overcoming the problems of large amount of data to be transmitted and complexity, and also improving the transmission efficiency and saving transmission resources.

[0272] (2) Centralized distributed system

[0273] In the case where data interaction between nodes is not possible, at this time, only the decision center can obtain the feature data of all nodes, and neural network model training needs to be performed on the decision center to determine the required power adaptation sub-model.

[0274] Figure 7 According to some embodiments of the present application, a framework schematic diagram 700 of a centralized distributed system is shown. Figure 7 Still taking Figure 3A the 2 nodes and 1 decision center shown as an example, a framework schematic diagram related to the training process is shown, so as to facilitate comparison with Figure 3A the framework schematic diagram of the actual usage process after training shown.

[0275] As Figure 7 shown, node 0 includes a feature extraction module 301b and a power adaptation module 302b. Node 1 includes a feature extraction module 311b and a power adaptation module 312b. It can be understood that the feature extraction module 301b and the power adaptation module 302b, as well as the feature extraction module 311b and the power adaptation module 312b, Figure 4A have the same functional essence as the feature extraction module 301a and the power adaptation module 302a, as well as the feature extraction module 311a and the power adaptation module 312a shown in

[0276] Specifically, during the training process, for node 0, after node 0 and node 1 obtain the labeled feature data x k；0 and the labeled feature data x k；1 through the feature extraction module 301b and the feature extraction module 311b, they will respectively send second feature information to the decision center 00. The second feature information includes the feature data of each sample in the sample set of each node and the classification label of each sample. It can be understood that this method is default to be sent without using channel transmission.

[0277] Then, the decision center 00 trains the pre-set neural network model based on the second feature information received from all nodes. For example, the label feature data in the second feature information, and the statistical characteristics of the channel noise between each node and the decision center 00 observed, such as the noise variance. This training method is the same as the specific training process of step S205 shown in the above-mentioned 5A, and will not be elaborated here. Therefore, the trained neural network models corresponding to each node are obtained, and thus the power adaptation sub-models corresponding to each node are obtained. Then, the obtained power adaptation sub-models corresponding to each node are synchronized to each node. This sending process is essentially the same as the process of the node 0 synchronizing the power adaptation sub-model to the decision center 00 under the above non-centralized system. The specific process can be referred to Figure 6A for description and will not be elaborated here.

[0278] It can be understood that since the decision center 00 can obtain the label feature data of all nodes, the decision center can obtain the statistical characteristics of the label feature data corresponding to all nodes. For example, the mean vector μ k and the covariance ∑ k to make decisions during actual use. It can be understood that in the above step S111, for the centralized system, since the decision center 00 can obtain the label feature data of all nodes, the label feature data is directly used when calculating the likelihood function value, and there is no need to use the data simulated according to the mean vector μ k and the covariance ∑ k , thus making the decision effect better.

[0279] The experimental verification data of applying the data transmission method shown in the embodiments of the present application to the classification problem of pictures passing through a noisy channel is introduced below, thus comprehensively proving that the technical solutions of the embodiments of the present application can achieve good effects. The effects obtained when using the MNIST dataset and the CIFAR-10 dataset are introduced respectively.

[0280] (1) Experimental introduction using the MNIST dataset:

[0281] 1. The MNIST dataset mainly includes a handwritten digit recognition dataset, with a total of 10 classes of digits from 0 to 9, that is, 10 labels. Taking two classes of data, namely digits 0 and 1, that is, 2 labels as an example. The dataset contains 50,000 28x28 black and white pictures, which are used for offline training of the feature extractor and the power adaptation function, as well as the receiver performance evaluation in actual compression and transmission tests.

[0282] 2. LeNet-5 is used as the feature extractor to extract the one-dimensional feature data of the pictures. The neural network structure of LeNet-5 is as Figure 8AAs shown. Among them, the basic structure of LeNet-5 includes a 7-layer network structure (excluding the input layer), including 2 convolutional layers, 2 downsampling layers (pooling layers), 2 fully connected layers and an output layer.

[0283] 3. Use a single linear layer for linear enhancement (as a comparison benchmark); use a fully connected neural network with an activation layer for non-linear power adaptation. Among them, the neural network structure is as Figure 8B shown, where the input is one-dimensional data, the output is one-dimensional data, and it includes at least one hidden layer.

[0284] 4. Generate data that follows a normal distribution based on the mean and variance of the extracted features for training the power adaptation sub-model.

[0285] In the scenarios of pairwise combinations of the rated powers of three different channels, the performance of the binary classification decision at the receiving end was compared when using three types of methods:

[0286] 1) Use the non-linear power adaptation sub-model to combat noise (i.e., the method proposed in the embodiments of the present application);

[0287] 2) Use the method of linear enhancement to combat noise;

[0288] 3) Classify at the transmitting end and use BPSK to transmit the results.

[0289] The experimental results are shown in Table 1 below (the vertical columns of the table correspond to the classification accuracy results obtained by using linear enhancement, BPSK, and non-linear power adaptation methods from top to bottom in sequence; the horizontal columns are the upper limits of the transmission power sizes from two nodes to the decision center; the numbers are the classification accuracies).

[0290] Referring to the classification results corresponding to Table 1, the first and second rows are control groups, which respectively use the traditional linear filtering method, and the method of judging the label at the transmitting end and performing BPSK transmission. The third and fourth rows are experimental groups for the method proposed in the embodiment of the present application, corresponding to the decentralized system and the centralized system respectively. The last row is a high-traffic control group. Assuming that there are sufficient communication conditions, the power adapter model is directly transmitted to the receiving end / transmitting end, which can prove the reliability of saving the communication cost of the adaptation function. Through the classification accuracy results shown in Table 1, it can be seen that under different rated powers, the classification accuracy results obtained by the embodiment scheme of the present application (such as the decentralized system, centralized system and power adapter model lossless transmission in Table 1) are better than the accuracy of the scheme using BPSK and linear filtering. When the power adapter model is synchronized at both ends, if the power adapter model is directly synchronized, the classification result obtained at this time (the power adapter model lossless transmission of the 5th row in Table 1) will be better than the classification result obtained by transmitting key point information (the decentralized system of the 3rd row in Table 1, the centralized system of the 4th row).

[0291] Table 1

[0292]

[0293] (II) Using the CIFAR-10 dataset

[0294] 1. The CIFAR-10 dataset is mainly an object recognition dataset. Take the two categories of data, "airplanes and birds", i.e. two labels, as an example. The two categories in the dataset each contain 1,000 32x32 color (three-channel) images, which are used for offline training of feature extractors and power adapter models, as well as for receiving end performance evaluation in actual compression and transmission tests.

[0295] 2. Use LeNet-5 as a feature extractor to extract one-dimensional feature data of the image (the network structure is the same as before).

[0296] 3. Use a single linear layer for linear enhancement (as a comparison benchmark); use a fully connected neural network with an activation layer for nonlinear power adaptation.

[0297] 4. Based on the mean and variance of the extracted feature data, generate data that obeys the normal distribution and train the power adaptation sub-model.

[0298] In the scenario of three different channel rated power combinations, the performance of the receiving end binary classification decision when using the three methods is compared:

[0299] 1) Using a nonlinear power adaptation sub-model to combat noise (i.e., the method proposed in the embodiment of the present application);

[0300] 2) Method of using linear enhancement to counter noise;

[0301] 3) Method of classifying at the transmitting end and using BPSK to transmit the result.

[0302] The experimental results are shown in the following table (the vertical columns of the table correspond to using linear enhancement, BPSK, and non - linear power adaptation from top to bottom in sequence; the horizontal columns are respectively the upper limits of the transmission power magnitudes from two nodes to the decision center; the numbers are the classification accuracies). The experimental results are shown in Table 2. Specifically, the first and second rows are the control groups, which use the traditional linear filtering method and the method of judging the label at the transmitting end and performing BPSK transmission respectively. The third and fourth rows are the experimental groups, corresponding to the decentralized system and the centralized system respectively. The last row is the high - traffic control group. Assuming sufficient communication conditions, directly transmitting the power adaptation sub - model to the receiving end / transmitting end can prove the reliability of saving communication costs. From the classification accuracy results shown in Table 2, it can be seen that at different rated powers, the classification accuracy results corresponding to the embodiment solutions of the present application (such as the decentralized system, the centralized system, and the lossless transmission of the adaptation function in Table 2) are better than the accuracies of the solutions using BPSK and linear filtering. When synchronizing the power adaptation sub - model at both ends, if directly synchronizing the power adaptation sub - model, the classification results obtained at this time (the lossless transmission of the power adaptation sub - model in the 5th row of Table 2) will be better than the classification results obtained by the method of transmitting key point information (the decentralized system in the 3rd row and the centralized system in the 4th row of Table 2).

[0303] Table 2

[0304]

[0305] Figure 9 is a hardware structure block diagram of an electronic device provided according to an embodiment of the present application. As Figure 9 shown, the electronic device 10 includes a processor 110, a communication module 120, a screen 130, an interface module 140, a memory 150, a power module 160, as well as an audio module 170, a camera 180, and a sensor module 190, where the audio module 170 includes a speaker 171 and a microphone 172.

[0306] Among them: The processor 110 may include one or more processing units. For example, it may include a processing module or processing circuit such as a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), a micro-programmed control unit (MCU), an artificial intelligence (AI) processor, or a field programmable gate array (FPGA). Among them, different processing units may be independent devices or integrated in one or more processors. In some embodiments, the processor 110 may perform feature extraction on data, adjust the feature data before sending, or perform decision-making processing by executing a program related to the data processing method of this application.

[0307] The communication module 120 may include various wired or wireless communication modules, such as a Bluetooth module (BT), a wireless local area networks (WLAN) module, etc., for providing solutions for wired or wireless communications such as wireless fidelity (Wi-Fi), Bluetooth (BT), global navigation satellite system (GNSS), frequency modulation (FM), near field communication (NFC), infrared technology (IR), wide area network (WAN), etc. In some embodiments, the processor 110 may execute a program related to the data processing method of this application.

[0308] The screen 130 includes a display panel. The display panel can be a liquid crystal display (LCD), an organic light-emitting diode (OLED), an active-matrix organic light-emitting diode (AMOLED), a flexible light-emitting diode (fled), a Mini LED, a Micro LED, a Micro OLED, a quantum dot light-emitting diode (QLED), etc.

[0309] The interface module 140 can include various forms of input or output interfaces. The electronic device 10 can transmit video and / or audio data to other electronic devices through the output interface and receive video and / or audio data from other electronic devices through the input interface. In some embodiments, the input / output interface can include: S / PDIF interface, HDMI interface, LAN (local area networks) interface, universal serial bus (USB) interface, AV interface, etc.

[0310] The memory 150 can be used to store data, software programs, and modules. It can be a volatile memory, such as a random-access memory (RAM); or a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid-state drive (SSD); or a combination of the above types of memories, or it can also be a removable storage medium, such as a secure digital (SD) memory card. Specifically, in some embodiments of the present application, the memory can be used to store relevant instructions for executing the data processing method of the present application.

[0311] The power module 160 can include a power button, an IR receiver, etc., and is used to turn on or off the power of the electronic device 10 according to the user's operation.

[0312] The audio module 170 can convert a digital audio signal into an analog audio signal for output, or convert an analog audio input into a digital audio signal. It can also transmit the digital audio signal and / or the analog audio signal to other electronic devices through the interface module 140. In some embodiments, the audio module 170 may include a speaker 171 and a microphone 172.

[0313] The camera 180 is used to acquire a static image or video. The optical image generated by the scene through the lens is projected onto the surface of the image sensor, and then converted into an electrical signal. After being converted by analogue-to-digital conversion (A / D), it becomes a digital image signal, which is then sent to the digital signal processing chip for processing.

[0314] The sensor module 190 may include a magnetic sensor, an acceleration sensor, a temperature sensor, a voice sensor, etc.

[0315] It can be understood that the structure of the electronic device 10 shown in FIG. 3 is only an example. In some other embodiments, the electronic device 10 may also include more or fewer modules, and some modules may be combined or split. The embodiments of the present application do not make any limitations.

[0316] It can be understood that the electronic device 10 may be a specific hardware structure diagram of a node or a specific hardware structure diagram of a decision center, which is not required here. When the electronic device is a node, it can perform the operations executed by the node in the data processing method proposed in the embodiments of the present application. When the electronic device is a decision center, it can perform the operations executed by the decision center in the data processing method proposed in the embodiments of the present application. It can be understood that the structure illustrated in the embodiments of the present invention does not constitute a specific limitation on the electronic device 10.

[0317] The embodiments of the mechanism disclosed in the present application can be implemented in hardware, software, firmware, or a combination of these implementation methods. The embodiments of the present application can be implemented as a computer program or program code executed on a programmable system, which includes at least one processor, a storage system (including volatile and non-volatile memories and / or storage elements), at least one input device, and at least one output device.

[0318] The program code can be applied to the input instructions to execute the various functions described in the present application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of the present application, the processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0319] The program code can be implemented in a high-level procedural language or an object-oriented programming language to communicate with the processing system. When needed, the program code can also be implemented in assembly language or machine language. In fact, the mechanisms described in this application are not limited to the scope of any specific programming language. In any case, the language can be a compiled language or an interpreted language.

[0320] In some cases, the disclosed embodiments can be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments can also be implemented as instructions carried or stored on one or more transient or non-transitory machine-readable (e.g., computer-readable) storage media, which can be read and executed by one or more processors. For example, the instructions can be distributed via a network or via other computer-readable media. Thus, machine-readable media can include any mechanism for storing or transmitting information in a form readable by a machine (e.g., a computer), including but not limited to, floppy disks, optical disks, optical discs, compact disc read-only memories (CD-ROMs), magneto-optical discs, read-only memories (ROMs), random access memories (RAMs), erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), magnetic or optical cards, flash memory, or tangible machine-readable memories for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) in electrical, optical, acoustic, or other forms using the Internet. Thus, machine-readable media includes any type of machine-readable media suitable for storing or transmitting electronic instructions or information in a form readable by a machine (e.g., a computer).

[0321] In the drawings, some structural or method features may be shown in a particular arrangement and / or order. However, it should be understood that such a particular arrangement and / or ordering may not be required. Rather, in some embodiments, these features may be arranged in a different manner and / or order than shown in the illustrative drawings. Additionally, the inclusion of a structural or method feature in a particular figure does not imply that such a feature is required in all embodiments, and in some embodiments, these features may not be included or may be combined with other features.

[0322] It should be noted that each unit / module mentioned in the device embodiments of the present application is a logical unit / module. Physically, a logical unit / module can be a physical unit / module, a part of a physical unit / module, or can be implemented as a combination of multiple physical units / module. The physical implementation manner of these logical units / modules themselves is not the most important. The combination of the functions implemented by these logical units / modules is the key to solving the technical problems proposed by the present application. In addition, in order to highlight the innovative part of the present application, the above device embodiments of the present application do not introduce units / modules that are not closely related to solving the technical problems proposed by the present application. This does not mean that there are no other units / modules in the above device embodiments.

[0323] It should be noted that in the examples and the description of the present patent, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one" does not exclude the existence of another identical element in the process, method, article or device including the element.

[0324] Although the present application has been illustrated and described by referring to certain preferred embodiments of the present application, those of ordinary skill in the art should understand that various changes can be made in form and detail without departing from the spirit and scope of the present application.

Claims

1. A data processing method, characterized in that, it includes: The first node device in the distributed system receives the first noise statistical information sent by the decision-making device in the distributed system, and receives the first feature information sent by the node devices in the distributed system except the first node device respectively; wherein, the first noise statistical information includes the statistical characteristics of the noise of the channel between the node device and the decision-making device, and, the first feature information includes the feature data of each sample in the first sample set of the node device, and the classification label of each sample, and the first noise statistical information and the first feature information are used to determine the first power adaptation sub-model used by each node device to execute the first task.

2. The method according to claim 1, characterized in that, the first power adaptation sub-model used by each node device to execute the first task is determined by the following method: The first node device trains the first neural network model based on the first noise statistical information and the first feature information corresponding to each node device, and obtains the trained first neural network model, wherein the trained first neural network model includes the first power adaptation sub-model corresponding to each node device.

3. The method according to claim 1, characterized in that, the statistical characteristics include one or more of the following: mean vector, covariance, arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, midrange, mode, mean absolute deviation, variance.

4. The method according to any one of claims 1-3, characterized in that, it further includes that the first node device sends first adaptation processing information to the decision-making device, wherein the first adaptation processing information includes multiple groups of first input-output data pairs, and each group of first input-output data pairs corresponds to the first power adaptation sub-model of a node device respectively.

5. The method according to claim 4, characterized in that, it further includes: The decision-making device obtains the first power adaptation function corresponding to the first power adaptation sub-model of each node device based on the received first adaptation processing information.

6. The method according to claim 5, characterized in that, the decision-making device obtains the first power adaptation function corresponding to the first power adaptation sub-model of each node device based on the received first adaptation processing information, including: determining the first power adaptation function corresponding to the first power adaptation sub-model of the first node device by the following method: The decision-making device inputs the first input-output data pair corresponding to the first power adaptation sub-model of the first node device into the first parametric function model, and obtains the first solution value of each parameter in the first parametric function model; The decision-making device replaces the parameters in the first parametric function model with the corresponding first solution values to obtain the first power adaptation function.

7. The method according to claim 6, characterized in that, After the first node device receives the first feature information respectively sent by the node devices in the distributed system other than the first node device, it further includes: The first node device sends first label feature statistical information to the decision device, where the first label feature statistical information includes statistical characteristics corresponding to each classification label in a plurality of classification labels, and the statistical characteristic corresponding to the first classification label in the plurality of classification labels is the statistical characteristic of the feature data with the classification label being the first classification label in all the feature data corresponding to the sample set of all the node devices in the distributed system.

8. The method according to claim 7, wherein, it further includes: Each of the node devices acquires first data to be classified corresponding to a first classification object in the first task; After each of the node devices processes the first classification feature data of the first data to be classified into second classification feature data based on its respective first power adaptation sub-model, it sends the second classification feature data to the decision device.

9. The method according to claim 8, wherein, it further includes that the decision device obtains a classification result for the first classification object based on a first power adaptation function corresponding to the first power adaptation sub-model of each node device, third classification feature data corresponding to the second classification feature data sent by each node device, and statistical characteristics corresponding to each classification label in a plurality of classification labels.

10. The method according to claim 9, wherein, The decision device obtains a classification result for the first classification object based on a first power adaptation function corresponding to the first power adaptation sub-model of each node device, third classification feature data corresponding to the second classification feature data sent by each node device, and statistical characteristics corresponding to each classification label in a plurality of classification labels, including: The decision device determines likelihood function values respectively corresponding to the first classification object and each classification label based on the first power adaptation function corresponding to each node device, third classification feature data corresponding to the second classification feature data sent by each node device, and statistical characteristics corresponding to each classification label in a plurality of classification labels; Taking the classification category of the classification label corresponding to the maximum likelihood function value among the likelihood function values as the classification result of the first classification object.

11. The method according to any one of claims 1-3, wherein, it further includes that the first node device sends second adaptation processing information to the decision device, where the second adaptation processing information includes the first power adaptation sub-models for each node device to execute the first task.

12. A data processing method, wherein, it includes: The decision device in the distributed system receives second feature information sent by each node device in the distributed system; wherein, the second feature information includes the feature data of each sample in the second sample set of the node device and the classification label of each sample; and The second feature information is used to determine a second power adaptation sub-model used by each node device to execute the second task.

13. The method according to claim 12, wherein, the second power adaptation sub - models used by the respective node devices to execute the second task are determined by the following method: the decision - making device trains a second neural network model based on the statistical characteristics of the noise of the channels between the decision - making device and the respective node devices, and the second feature information corresponding to the respective node devices, to obtain a trained second neural network model, wherein the trained second neural network model includes the second power adaptation sub - models corresponding to each node device.

14. The method according to any one of claims 13, wherein, the statistical characteristics include one or more of the following: mean vector, covariance, arithmetic mean, geometric mean, harmonic mean, weighted mean, quadratic mean, median, mid - range, mode, mean absolute deviation, variance.

15. The method according to any one of claims 12 - 14, wherein, further comprising that the decision - making device separately sends third adaptation processing information to the respective node devices, wherein the third adaptation processing information includes a set of second input - output data pairs, and each set of second input - output data pairs corresponds to the second power adaptation sub - model of the sending node device.

16. The method according to claim 15, wherein, further comprising: each of the node devices obtains a second power adaptation function corresponding to the second power adaptation sub - model of the node device based on the received third adaptation processing information.

17. The method according to claim 16, wherein, each of the node devices obtains a second power adaptation function corresponding to the second power adaptation sub - model of the node device based on the received third adaptation processing information, including: the first node device obtains a second power adaptation function corresponding to the second power adaptation sub - model of the first node device based on the received third adaptation processing information: the first node device inputs the second input - output data pair into a second parameter - containing function model to obtain second solution values of the parameters in the second parameter - containing function model; the first node device replaces the parameters in the second parameter - containing function model with the corresponding second solution values to obtain the second power adaptation function.

18. The method according to claim 17, wherein, further comprising: each of the node devices acquires second data to be classified corresponding to the second classification object of the second task; each of the node devices processes the fourth classification feature data of the second data to be classified into fifth classification feature data based on its respective second power adaptation function and then sends it to the decision - making device.

19. The method according to claim 18, wherein, further comprising that the decision - making device obtains a classification result for the second classification object based on the second power adaptation sub - models corresponding to the respective node devices, the sixth classification feature data corresponding to the fifth classification feature data sent by the respective node devices, and the feature data of each sample in the second sample set of all the node devices.

20. The method according to claim 19, wherein, the decision-making device obtains a classification result for the second classification object based on the second power adaptation sub-models corresponding to the respective node devices, the sixth classification feature data corresponding to the fifth classification feature data sent by the respective node devices, and the feature data of each sample in the second sample set of all the node devices, including: the decision-making device determines likelihood function values corresponding to the second classification object and each classification label respectively based on the second power adaptation sub-models corresponding to the respective node devices, the sixth classification feature data corresponding to the fifth classification feature data sent by the respective node devices, and the feature data of each sample in the second sample set of all the node devices; The classification category of the classification label corresponding to the maximum likelihood function value among the respective likelihood function values is used as the classification result of the second classification object.

21. The method according to any one of claims 12-14, wherein, further comprising that the decision-making device respectively sends fourth adaptation processing information to the respective node devices, wherein the fourth adaptation processing information includes the second power adaptation sub-model corresponding to the node device.

22. An electronic device, wherein, comprising: one or more processors; one or more memories; the one or more memories store one or more instructions, and when the one or more instructions are executed by the one or more processors, the electronic device executes the data processing method according to any one of claims 1 to 21.

23. A computer-readable storage medium, wherein, instructions are stored on the storage medium, and when the instructions are executed on a computer, the computer executes the data processing method according to any one of claims 1 to 21.

Images