Wireless network signal action recognition system and method based on improved support vector machine

Abstract

The present invention discloses a wireless network signal action recognition system and method based on an improved support vector machine in the field of wireless communication technology, including: extracting channel state information caused by action changes through Wi-Fi signals to obtain sample data; obtaining a sample set after preprocessing the sample data; combinatorially optimizing the support vector machine algorithm parameters through an improved differential evolution algorithm; performing model training and testing using the sample set based on the combined optimized support vector machine algorithm to obtain an action recognition training model; and judging the action category of online action data based on the action recognition training model. The present invention can detect actions through Wi-Fi signals to perform highly accurate action recognition.

Description

Technical Field

[0001] The invention relates to a wireless network signal action recognition system and method based on an improved support vector machine, and belongs to the technical field of wireless communications. Background Art

[0002] At present, human motion recognition plays an important role in application fields such as medical rehabilitation, sports fitness, and safety detection. With the continuous development of artificial intelligence technology, people have higher requirements for the accuracy and convenience of motion recognition technology. Existing motion recognition methods are mainly based on sensor technology and computer vision technology based on portable devices. It requires special sensors and camera equipment, such as radar, camera, non-portable optical equipment, etc. The high deployment cost, inconvenience of carrying, and dependence on external lighting environment limit the applicability of these technologies in certain scenarios. In recent years, the method of using Wi-Fi signals to detect motion for motion recognition has become a research hotspot.

[0003] The Wi-Fi signal motion recognition system requires two ports for receiving and transmitting. The motion between the two ports will cause fluctuations in the Wi-Fi signal. CSI comes from the subcarrier decoded in the OFDM system. It is the channel property of the communication link and reflects the attenuation of the signal on each transmission path. Compared with the received signal strength, it has richer information and more obvious amplitude changes, which can better perceive the subtle differences between actions. We extract the CSI containing the motion information from the signal and then perform motion recognition through the algorithm. The implementation of this method only requires low-cost consumer devices, is easy to promote, easy to carry, and has strong applicability, and can achieve high accuracy. However, this method has the following problems:

[0004] Due to the poor sensitivity of such devices, the measured CSI is easily interfered by noise, which affects the action recognition effect. The currently used noise reduction method is difficult to remove interference well, so the system's action recognition ability is very dependent on the classification effect of the classifier algorithm. Popular neural network classifiers such as back propagation neural network, convolutional neural network, and deep neural network have high recognition accuracy, but require a large number of samples and have a long running time. When the environment or action category changes, the system needs to retrain the model, which is inefficient and not suitable for consumer electronics. Support vector machine (SVM) is a two-class classifier with strong generalization ability, only a small number of samples, and fast retraining model speed. It is suitable for this system, but the classification accuracy cannot be guaranteed and still needs to be improved. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a wireless network signal action recognition system and method based on an improved support vector machine, which can detect actions through Wi-Fi signals to perform highly accurate action recognition.

[0006] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0007] In a first aspect, the present invention provides a method for wireless network signal motion recognition based on an improved support vector machine, comprising:

[0008] The channel state information caused by the motion change is extracted through the Wi-Fi signal to obtain sample data;

[0009] After preprocessing the sample data, a sample set is obtained;

[0010] The parameters of the support vector machine algorithm are combined and optimized through the improved differential evolution algorithm;

[0011] Based on the combined optimized support vector machine algorithm, the model is trained and tested using the sample set to obtain the action recognition training model;

[0012] Based on the action recognition training model, the action category of the online action data is judged.

[0013] Furthermore, the sample data is preprocessed to obtain a sample set, including: noise reduction through Butterworth low-pass filtering combined with principal component analysis, and then feature extraction through principal component analysis to form a preprocessed sample set.

[0014] Furthermore, the steps of the improved differential evolution algorithm are as follows:

[0015] Randomly and uniformly generate multiple individuals within the solution vector value range to form an initial population;

[0016] After calculating the fitness value of each individual in the initial population, compare and obtain the individual with the best fitness;

[0017] After calculating the scaling operator corresponding to each individual in the initial population, the mutation operation is performed in combination with the individual fitness;

[0018] Calculate the crossover probability corresponding to each individual in the initial population, perform a crossover operation on the individuals after the mutation operation, and obtain a set of crossover individuals;

[0019] Compare the crossover individuals with the corresponding individuals in the parent population, and select individuals with good fitness as individuals at the same position in the offspring according to the greedy rule to form a new population;

[0020] Calculate the fitness value of individuals in the new population and determine whether the fitness value meets the termination condition;

[0021] In response to the fitness value meeting the termination condition, the optimal parameter combination of the action recognition training model is output; otherwise, the scaling operator corresponding to each individual is recalculated, and then the termination condition is re-judged after performing mutation, crossover and selection operations in sequence.

[0022] Furthermore, the composition of the i-th individual in the initial population is as follows:

[0023] X i =[x i,1 ,x i,2 ,…,x i,n ,](i∈1,…,N)

[0024] Among them, n refers to the dimension of a solution vector, and the j-th dimension of the i-th individual in the population is initialized as follows:

[0025] x i,j =rand(0,1)(U j -L j )+L j (i∈1,…,N,j∈1,…,n)

[0026] Among them, L j represents the lower bound of the j-th dimension of the solution vector, U j represents the upper bound of the j-th dimension of the solution vector, rand(0,1) refers to a random number between 0 and 1;

[0027] The fitness value calculation formula is as follows:

[0028]

[0029] Among them, f is the fitness function, L is the number of categories, Y represents the number of test samples in this category, e represents the number of wrong judgments in this category, and i represents the individual serial number.

[0030] Furthermore, the calculation formula of the scaling operator is:

[0031]

[0032]

[0033] Among them, F is the scaling operator, λ m is a parabolic function of the current number of iterations, β i is the adaptive change function of individual fitness, m is the current number of iterations, M is the total number of iterations, and F max and F min is the upper and lower limits of the scaling operator, f i is the fitness value of the i-th individual in this iteration, f min and f max is the minimum and maximum fitness value in this iteration population, f a Yes min and f max The median value of

[0034] The formula for the mutation operation is:

[0035] v i =λ m (x s1 +F(x s2 -x s3 ))+(1-λ m )(x best +F(x s4 -x s5 ))

[0036] Among them, v i is the new individual obtained by mutation, x s1 、x s2 、x s3 、x s4 、x s5 are the individuals in the previous generation that pull different groups, x best It is the individual with the best fitness in the previous generation population.

[0037] Furthermore, the crossover probability calculation formula is as follows:

[0038] CR i =randn(μ CR ,0.05)

[0039] μ CR =(1-c)μ CR +c×mean(S)

[0040]

[0041] Among them, CR i is the crossover probability of the ith individual, c is a parameter between (0, 1), μ CR The initial value is set to 0.5, randn(μ CR ,0.05) represents the mathematical expectation of μ CR , normal distribution function with standard deviation of 0.05, mean(S) is the Lehmer mean of the data in the set, and CR is the crossover probability;

[0042] The cross-individual components are:

[0043]

[0044] Where randi(1,n) represents a randomly selected integer from 1 to n, H i,j (m) is the cross-individual component, V i,j (m) is the new individual obtained by mutation.

[0045] Furthermore, the selection formula for selecting individuals with good fitness as individuals at the same position in the offspring is:

[0046]

[0047] Among them, X i (m) is the individual at the same position in the offspring, X i (m-1) is the corresponding individual in the parent population, H i (m) is the crossover individual, f(H i (m) is the crossover individual fitness value, f(X i (m-1)) is the corresponding individual fitness value in the parent population.

[0048] In a second aspect, the present invention provides a wireless network signal action recognition system based on an improved support vector machine, comprising:

[0049] Sampling module: used to extract channel state information caused by motion changes through Wi-Fi signals to obtain sample data;

[0050] Preprocessing module: used to preprocess sample data to obtain a sample set;

[0051] Optimization module: used to perform combinatorial optimization of support vector machine algorithm parameters through improved differential evolution algorithm;

[0052] Modeling module: used to train and test the model based on the support vector machine algorithm after combinatorial optimization using the sample set to obtain the action recognition training model;

[0053] Recognition and judgment module: used to judge the action category of online action data based on the action recognition training model.

[0054] In a third aspect, the present invention provides a wireless network signal motion recognition device based on an improved support vector machine, including a processor and a storage medium;

[0055] The storage medium is used to store instructions;

[0056] The processor is used to operate according to the instructions to execute the steps of any of the methods described above.

[0057] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of any of the above methods when executed by a processor.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention discloses a method for wireless network signal motion recognition based on an improved support vector machine, which can solve the problem of difficulty in similar motion recognition and the problem of support vector machine algorithm parameter combination optimization, and the motion recognition system has low cost, strong applicability and high accuracy;

[0060] Second, the present invention uses principal component analysis (PCA) combined with Butterworth low-pass filtering to clean CSI data to eliminate noise interference, and uses an improved differential evolution algorithm to optimize SVM to improve the efficiency of SVM, thereby improving the efficiency and accuracy of motion recognition based on cheap consumer Wi-Fi devices;

[0061] 3. The present invention adopts four strategies to improve the differential evolution algorithm. In the first strategy, the proportional factor in the mutation is adaptively adjusted according to the number of iterations to balance the population diversity and the convergence speed. In the second strategy, we divide the individuals in the group into "good" or "bad" according to the fitness, and then adjust the scaling operators of the good individuals and the bad individuals respectively through different methods, retain the good individuals, and change the bad individuals through crossover. In the third strategy, we modified the existing mutation method so that it changes with the number of iterations to further prevent the algorithm from maturing prematurely. The fourth strategy is to adjust the crossover probability by introducing historical information to adaptively change the crossover situation of the parent node to balance the local search ability and the global search ability of the algorithm. The improved DE algorithm not only avoids the algorithm from falling into the local optimum, but also further improves the convergence speed, and can better find the relatively optimal solution of the SVM penalty parameter and the kernel parameter, so that the search efficiency of the SVM is improved;

[0062] Fourth, the present invention applies the improved SVM to the Wi-Fi signal action recognition method. The Wi-Fi signal action recognition method uses cheap consumer equipment, which is low in cost, easy to carry, and has strong applicability. The improved SVM is used as a classifier of the action recognition method, which eliminates the shortcomings of high sensitivity and low classification accuracy of CSI action information extracted by cheap equipment, and can be better applied to various scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 is a model diagram of the action recognition system provided by the first embodiment of the present invention;

[0064] Figure 2 This is the CSI fluctuation caused by the action provided in the first embodiment of the present invention;

[0065] Figure 3 It is a schematic diagram of the process of optimizing support vector machine by improved differential evolution algorithm provided in the first embodiment of the present invention;

[0066] Figure 4 This is a simulation diagram comparing the convergence degree of the improved differential evolution algorithm curve provided in the first embodiment of the present invention with other algorithms;

[0067] Figure 5 This is a schematic diagram comparing the action recognition accuracy of the optimized SVM provided in the first embodiment of the present invention and other methods. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0069] Embodiment 1:

[0070] This embodiment discloses a method for wireless network signal motion recognition based on an improved support vector machine. Figure 1 As shown, it is divided into three parts: signal reception, data preprocessing, and action determination. The method includes the following steps:

[0071] Step 1: For similar specific actions, establish an action recognition scenario. This embodiment designs an action recognition system for similar actions. Taking the biceps curl action (BC) as an example, a standard action and five common incorrect actions are designed. The action details are as follows:

[0072]

[0073] Step 2: Extract the channel state information (CSI) caused by the action change through the Wi-Fi signal. In this embodiment, the sensor consists of two computers equipped with an Intel 5300 wireless network card, which can extract CSI data when the action changes. One of them is set to send mode and the other is set to receive mode. Antennas are used to increase indoor coverage. The tester performs the designed action between the two machines and extracts the action data as a sample set.

[0074] Step 3: After preprocessing the sample data, a sample set is obtained. In this embodiment, the CSI data is preprocessed by first using Butterworth low-pass filtering combined with principal component analysis to reduce noise, and then extracting features through principal component analysis (PCA) to form a preprocessed sample set. The PCA whitening feature extraction process first calculates a low-dimensional feature matrix through PCA, and then normalizes the standard deviation of the features of each dimension so that all features have the same variance, reducing the redundancy of the input data and facilitating classification. The waveform of the CSI data before and after noise reduction is as follows: Figure 2 shown.

[0075] Step 4: Improve the differential evolution algorithm, and realize the combined optimization of the support vector machine algorithm parameters through the improved differential evolution algorithm. Based on the obtained improved support vector machine algorithm, use the sample set to train and test the model, find the optimal action recognition training model (SVM model), and realize the classification and recognition of similar actions. In this embodiment, the processed sample set is extracted features by PCA, divided into training set and test set, and then put into the improved SVM training model. The optimal model can be obtained by the improved differential evolution algorithm (DE algorithm). The online extracted data can directly test the results. The comparison chart of the recognition accuracy of each category of action in the experiment is as follows: Figure 5 As shown, the improved SVM recognition accuracy proposed in the present invention is higher than that of other improved algorithms.

[0076] The standard differential evolution algorithm is an intelligent global search algorithm. It evaluates and optimizes each individual in the group, and then searches from the best individual. The combination and competition between individuals intelligently guide the direction of the optimization search. The differential evolution algorithm process is to first randomly generate an initial population, and then generate a new population through mutation, crossover, and selection to continue to find the optimal solution until the termination condition is reached, that is, the set number of iterations is reached or a solution vector that meets the conditions is found. The differential evolution algorithm solves an optimization problem. First, the fitness function f(x) and algorithm parameters must be determined. f(x) reflects the quality of the individual. The smaller the fitness, the better the individual. When the algorithm is continuously iterated, the population diversity decreases and premature convergence is prone to occur. We try to enrich the population diversity in the early stages to prevent premature convergence. In the later stages, convergence needs to be accelerated. We propose an improved differential evolution algorithm, which comprehensively considers the iterations in the population and the differences in individual fitness to solve this problem. The function convergence effect of the improved differential evolution algorithm is compared with other improved algorithms. Figure 4 As shown, the improvement method is as follows:

[0077] (1) The scaling operator F controls the magnification and reduction of the deviation variable, which affects the diversity of the population. The larger the F, the greater the individual variation in the population and the richer the population. The smaller the F, the faster the convergence speed. We propose to adjust the scaling operator F according to the iteration situation and the quality of the individual. The method is as follows:

[0078]

[0079]

[0080] Among them, F is the scaling operator, λ m is a parabolic function of the current number of iterations, β i is the adaptive change function of individual fitness, m is the current number of iterations, M is the total number of iterations, and F max and Fmin is the upper and lower limits of the scaling operator, f i is the fitness value of the i-th individual in this iteration, f min and f max is the minimum and maximum fitness value in this iteration population, f a Yes min and f max The middle value of .

[0081] We use the parabolic function λ m Adjust F at different stages of iteration to improve the algorithm to prevent premature maturity. The scaling operator changes with the function, gradually decreases as the current number of iterations increases, and the rate of decrease gradually decreases, and finally stabilizes at the set minimum value. In the early stage of iteration, F is large, the population diversity is rich, and the global search ability is strong; in the middle of the iteration, the rate of decrease of F slows down and the convergence speed is accelerated; in the late stage of iteration, F gradually stabilizes at a set smaller value, the convergence speed reaches the maximum while ensuring the diversity of the population.

[0082] We propose to adjust the individual variation parameters according to different strategies based on the individual fitness. a The better individual, β i The value of varies between 0.5 and 1, f i The smaller the β i The smaller the value, the smaller the scaling operator, that is, the better the individual is, the smaller the corresponding scaling operator is, so that the best individuals can be retained as much as possible; greater than f a The inferior individuals are β i The maximum value is 1, which makes the F of these individuals in this iteration relatively large and ensures the diversity of the population.

[0083] (2) In this embodiment, we improve the mutation strategy. The commonly used mutation strategies DE / rand / 1 and DE / best / 1 are shown in the formula:

[0084] v i =x s1 +F(x s2 -x s3 ) (2)

[0085] v i =x best +F(x s4 -x s5 ) (3)

[0086] Where x s1 、x s2 、x s3 、x s4 、x s5 are the individuals in the previous generation that pull different groups, xbest is the individual with the best fitness in the previous generation population. The DE / rand / 1 strategy is conducive to the enrichment of the population, but the convergence speed is slow. The DE / best / 1 strategy is the opposite. It has the best information of the population and can converge quickly, but it is easy to fall into the local optimum. We have made improvements based on the iteration situation and combined the advantages of the two strategies, as shown in the formula:

[0087] v i =λ m (x s1 +F(x s2 -x s3 ))+(1-λ m )(x best +F(x s4 -x s5 )) (4)

[0088] Among them, v i is the new individual obtained by mutation, using the λ m In the early stage of iteration, the DE / rand / 1 strategy is dominant, with strong global search capability and not prone to premature. As the iteration progresses, the DE / best / 1 strategy plays a greater role, which helps accelerate convergence when the range of the global optimal solution is slowly determined in the later stage.

[0089] (3) The crossover probability CR affects the convergence speed and population richness of the algorithm. In order to make the crossover operation balance the global search capability and local search efficiency, we adopt an adaptive update strategy based on historical information.

[0090] A history set S is set, which stores the crossover probability of individuals successfully selected after crossover in the last iteration, and the Lehmer mean mean(S) of the data in the set is calculated, as shown in the formula:

[0091]

[0092] The crossover probability update formula is as follows:

[0093] CR i =randn(μ CR ,0.05) (6)

[0094] μ CR =(1-c)μ CR +c×mean(S) (7)

[0095] Among them, CR i is the crossover probability of the ith individual, c is a parameter between (0, 1), μ CR The initial value is set to 0.5, randn(μ CR ,0.05) represents the mathematical expectation of μCR , a normal distribution function with a standard deviation of 0.05. If the value exceeds the range of CR [0, 1], it will be replaced with the closer upper and lower limits.

[0096] The crossover probability changes with each generation of the historical set, ensuring that the value is within a reasonable range. The normal distribution increases the randomness and diversity of the values. The Lehmer average effectively avoids the μ CR Too small to prevent the algorithm from falling into local search easily.

[0097] The judgment accuracy of the system model depends largely on the parameters C and G of the nonlinear SVM. The error penalty factor C affects the generalization ability of the system. When C is higher, the penalty for error is heavier, and overfitting is prone to occur. When C is smaller, the tolerance for error is greater, and underfitting is prone to occur. G is the kernel parameter of the Gaussian kernel function, which affects the range of action of the Gaussian corresponding to each support vector. When G is larger, the Gaussian distribution becomes higher and narrower, the stronger the division ability of the Gaussian kernel function, the poorer the classification effect for unknown samples, and the weaker the generalization ability. When G is smaller, the smoothing effect will be greater, resulting in inaccurate sample classification.

[0098] The optimization of SVM based on Gaussian kernel function is to optimize the penalty parameter C and kernel parameter G in combination. The fitness function is set to the total classification error rate. The formula is as follows:

[0099]

[0100] Among them, f is the fitness function, L is the number of categories, Y represents the number of test samples in this category, and e represents the number of wrong judgments in this category. The flowchart of improving DE algorithm to optimize SVM for action recognition is as follows Figure 3 shown.

[0101] Specifically, the specific steps of improving the DE algorithm to optimize SVM for action recognition in step 4 include:

[0102] Step S1: Split the processed offline sample set into a training set and a test set.

[0103] Step S2: Population initialization: Using the penalty parameter C and kernel parameter G of SVM as parameters, generate the initial population according to the requirements of the differential evolution algorithm and set relevant parameters.

[0104] N individuals are randomly and uniformly generated within the solution vector value range. These individuals form the initial population. The composition of the i-th individual is as follows:

[0105] X i =[x i,1 ,x i,2 ,…,x i,n ,](i∈1,…,N) (9-1)

[0106] Where n refers to the dimension of a solution vector.

[0107] The value of each individual is used as the penalty parameter C and the kernel parameter G to train the SVM model. The fitness of each individual is calculated according to the fitness function set in formula (8). The j-th dimension value of the i-th individual in the population is initialized as follows:

[0108] x i,j =rand(0,1)(U j -L j )+L j i∈1,…,N j∈1,…,n (9-2)

[0109] Among them, L j represents the lower bound of the j-th dimension of the solution vector, U j Represents the upper bound of the j-th dimension of the solution vector, and rand(0,1) refers to a random number between 0 and 1. Calculate the fitness value of each individual and compare to get the optimal value.

[0110] Step S3: Mutation operation: By adding the difference between two individuals in the population to other individuals, a new mutant individual is generated. We combine two commonly used mutation strategies and improve the strategy with iterative algebra. At the same time, the scaling operator F corresponding to each individual in this iteration is adjusted according to the iteration situation and the quality of the individual. That is, the scaling operator F corresponding to each individual is calculated by formula (1), and the individual mutation is performed through the mutation strategy in formula (4).

[0111] Step S4: Crossover operation: exchange the components of each individual in the parent population with the components in the corresponding mutant individual, that is, calculate the crossover probability CR corresponding to each individual by formula (6), and perform the crossover operation of the individual. In the mth iteration, the crossover individual components are as follows:

[0112]

[0113] Where randi(1,n) represents a randomly selected integer from 1 to n, ensuring that at least one component of the crossover individual is provided by the corresponding individual in the parent generation. In order to balance the global search capability and local search efficiency in the crossover operation, we use an adaptive update strategy based on historical information to adjust the crossover probability CR corresponding to each individual. i .

[0114] Step S5: Selection operation: A group of crossover individuals are obtained through the crossover operation. The selection is to compare them with the corresponding individuals in the parent population, and follow the greedy rule to keep the one with the best fitness as the individual at the same position in the offspring. The selection formula is as follows:

[0115]

[0116] Among them, X i (m) is the individual at the same position in the offspring, X i (m-1) is the corresponding individual in the parent population, H i (m) is the crossover individual, f(H i (m)) is the crossover individual fitness value, f(X i (m-1)) is the corresponding individual fitness value in the parent population.

[0117] The selection operation ensures that the new generation of population is better than the previous generation, eliminates poor individuals, retains the original excellent individuals, and guides the algorithm to approach the optimal solution.

[0118] Step S6: Store the CR corresponding to the successfully selected individual into the history set, and calculate the fitness value of the individual in the new population.

[0119] Step S7: If the termination condition is met, the iteration ends, the optimal SVM parameter combination is output and the optimal model is generated, otherwise return to step S3.

[0120] Step S8: Put the online test data into the optimal model for classification and recognition.

[0121] Embodiment 2:

[0122] The wireless network signal action recognition system based on the improved support vector machine can implement the wireless network signal action recognition method based on the improved support vector machine in the first embodiment, including:

[0123] Sampling module: used to extract channel state information caused by motion changes through Wi-Fi signals to obtain sample data;

[0124] Preprocessing module: used to preprocess sample data to obtain a sample set;

[0125] Optimization module: used to perform combinatorial optimization of support vector machine algorithm parameters through improved differential evolution algorithm;

[0126] Modeling module: used to train and test the model based on the support vector machine algorithm after combinatorial optimization using the sample set to obtain the action recognition training model;

[0127] Recognition and judgment module: used to judge the action category of online action data based on the action recognition training model.

[0128] Embodiment three:

[0129] An embodiment of the present invention also provides a wireless network signal action recognition device based on an improved support vector machine, which can implement the wireless network signal action recognition method based on the improved support vector machine in the first embodiment, including a processor and a storage medium;

[0130] The storage medium is used to store instructions;

[0131] The processor is used to operate according to the instructions to execute the steps of the following method:

[0132] Extract the channel state information caused by action changes through the Wi-Fi signal to obtain sample data;

[0133] After preprocessing the sample data, obtain a sample set;

[0134] Use the improved differential evolution algorithm to perform combined optimization on the support vector machine algorithm parameters;

[0135] Based on the support vector machine algorithm after combined optimization, use the sample set for model training and testing to obtain an action recognition training model;

[0136] Based on the action recognition training model, judge the action category of the online action data.

[0137] Embodiment 4:

[0138] An embodiment of the present invention also provides a computer-readable storage medium, which can implement the wireless network signal action recognition method based on the improved support vector machine in the first embodiment. A computer program is stored thereon, and when the program is executed by a processor, it implements the steps of the following method:

[0139] Extract the channel state information caused by action changes through the Wi-Fi signal to obtain sample data;

[0140] After preprocessing the sample data, obtain a sample set;

[0141] Use the improved differential evolution algorithm to perform combined optimization on the support vector machine algorithm parameters;

[0142] Based on the support vector machine algorithm after combined optimization, use the sample set for model training and testing to obtain an action recognition training model;

[0143] Based on the action recognition training model, judge the action category of the online action data.

[0144] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0145] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0146] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0147] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0148] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. Wireless network signal action recognition method based on improved support vector machine, Its characteristics are: include: The channel state information caused by the motion change is extracted through the Wi-Fi signal to obtain sample data; After preprocessing the sample data, a sample set is obtained; The parameters of the support vector machine algorithm are combined and optimized through the improved differential evolution algorithm; Based on the combined optimized support vector machine algorithm, the model is trained and tested using the sample set to obtain the action recognition training model; Based on the action recognition training model, the action category of the online action data is judged; The improved differential evolution algorithm steps are as follows: Randomly and uniformly generate multiple bodies within the solution vector value range to form an initial population; After calculating the fitness value of each individual in the initial population, compare and obtain the individual with the best fitness; After calculating the scaling operator corresponding to each individual in the initial population, the mutation operation is performed in combination with the individual fitness; Calculate the crossover probability corresponding to each individual in the initial population, perform a crossover operation on the individuals after the mutation operation, and obtain a set of crossover individuals; Compare the crossover individuals with the corresponding individuals in the parent population, and select individuals with good fitness as individuals at the same position in the offspring according to the greedy rule to form a new population; Calculate the fitness value of individuals in the new population and determine whether the fitness value meets the termination condition; In response to the fitness value meeting the termination condition, the optimal parameter combination of the action recognition training model is output; otherwise, the scaling operator corresponding to each individual is recalculated, and then the termination condition is re-judged after performing mutation, crossover and selection operations in sequence.

2. According to claim 1, the wireless network signal action recognition method based on the improved support vector machine, Its characteristics are: The sample data is preprocessed to obtain a sample set, including: noise reduction through Butterworth low-pass filtering combined with principal component analysis, and feature extraction through principal component analysis to form a preprocessed sample set.

3. According to the wireless network signal action recognition method based on improved support vector machine according to claim 1, Its characteristics are: The composition of the i-th individual in the initial population is as follows: ， Among them, n refers to the dimension of a solution vector, and the j-th dimension of the i-th individual in the population is initialized as follows: ， in, represents the lower bound of the j-th dimension of the solution vector, represents the upper bound of the j-th dimension of the solution vector, rand(0,1) refers to a random number between 0 and 1; The fitness value calculation formula is as follows: ， in, is the fitness function, L is the number of categories, Y represents the number of test samples in this category, e represents the number of wrong judgments in this category, and i represents the individual number.

4. According to claim 1, the wireless network signal action recognition method based on improved support vector machine, Its characteristics are: The calculation formula of the scaling operator is: ， = ， = ， Where F is the scaling operator, is a parabolic function of the current number of iterations, is the adaptive change function of individual fitness, m is the current number of iterations, M is the total number of iterations, and are the upper and lower limits of the scaling operator. is the fitness value of the i-th individual in this iteration, and is the minimum and maximum fitness value in this iteration population, yes and The median value of The formula for the mutation operation is: ， in, is a new individual obtained by mutation. , , , , It is the individuals in the previous generation population that pull different forces. It is the individual with the best fitness in the previous generation population.

5. According to claim 1, the wireless network signal action recognition method based on improved support vector machine, Its characteristics are: The crossover probability calculation formula is as follows: ， ， ， in, is the crossover probability of the ith individual, c is a parameter between (0, 1), The initial value is set to 0.

5. Represents the mathematical expectation as , a normal distribution function with a standard deviation of 0.05, is the Lehmer mean of the data in the set, CR is the crossover probability; The cross-individual components are: ， in represents a randomly selected integer from 1 to n, is the cross-individual component, is the new individual obtained by mutation.

6. The wireless network signal motion recognition method based on improved support vector machine according to claim 1, Its characteristics are: The selection formula for selecting individuals with good fitness as individuals at the same position in the offspring is: ， in, is the individual at the same position in the offspring, is the corresponding individual in the parent population, For crossover individuals, is the crossover individual fitness value, is the fitness value of the corresponding individual in the parent population.

7. Wireless network signal action recognition system based on improved support vector machine, Its characteristics are: include: Sampling module: used to extract channel state information caused by motion changes through Wi-Fi signals to obtain sample data; Preprocessing module: used to preprocess sample data to obtain a sample set; Optimization module: used to perform combinatorial optimization of support vector machine algorithm parameters through improved differential evolution algorithm; Modeling module: used to train and test the model based on the support vector machine algorithm after combinatorial optimization using the sample set to obtain the action recognition training model; Recognition and judgment module: used to judge the action category of online action data based on the action recognition training model; The improved differential evolution algorithm steps are as follows: Randomly and uniformly generate multiple bodies within the solution vector value range to form an initial population; After calculating the fitness value of each individual in the initial population, compare and obtain the individual with the best fitness; After calculating the scaling operator corresponding to each individual in the initial population, the mutation operation is performed in combination with the individual fitness; Calculate the crossover probability corresponding to each individual in the initial population, perform a crossover operation on the individuals after the mutation operation, and obtain a set of crossover individuals; Compare the crossover individuals with the corresponding individuals in the parent population, and select individuals with good fitness as individuals at the same position in the offspring according to the greedy rule to form a new population; Calculate the fitness value of individuals in the new population and determine whether the fitness value meets the termination condition; In response to the fitness value meeting the termination condition, the optimal parameter combination of the action recognition training model is output; otherwise, the scaling operator corresponding to each individual is recalculated, and then the termination condition is re-judged after performing mutation, crossover and selection operations in sequence.

8. Wireless network signal motion recognition device based on improved support vector machine, Its characteristics are: including processor and storage medium; The storage medium is used to store instructions; The processor is used to operate according to the instructions to execute the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, Its characteristics are: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are implemented.

Images