Human body pressure detection method, system, equipment and medium

Abstract

The embodiments of the present application provide a human body pressure detection method, system, device and medium. The EEG feature is divided into a first EEG feature and a second EEG feature through a cross-stage convolution module, and the second EEG feature is used as the input feature of the first dense layer. The dense layer is copied, convolved and spliced ​​according to the second EEG feature to obtain a fifth EEG feature. The first EEG feature and the fifth EEG feature are spliced ​​and then convolved to obtain a sixth EEG feature; the sixth EEG feature is subjected to non-stride convolution to obtain a seventh EEG feature and is encapsulated into a primary capsule unit; pressure classification is performed based on the vector length of the advanced capsule unit obtained by the dynamic routing algorithm to obtain a pressure value; the gradient flow can be truncated through cross-stage convolution to prevent excessive repeated gradient information from being repeatedly learned, thereby greatly reducing the amount of calculation and improving the learning efficiency of the network; and the feature recognition information can be effectively retained through non-stride convolution and dynamic routing.

Description

Technical Field

[0001] The embodiments of the present application relate to, but are not limited to, the field of image processing technology, and in particular, to a human body pressure detection method, system, device, and medium. Background Art

[0002] The psychological stress of the human body can be measured and monitored by the EEG signals of the human body, and the stress can be quantified from the brain wave pattern. The ability of the EEG can be expanded through wireless wearable EEG acquisition equipment to achieve non-invasive assessment of the stress level of the human body. The implementation of EEG signal classification tasks can be facilitated by neural networks and deep learning algorithms. However, there are still the following problems: neural networks have been proven to be particularly powerful when they become deeper and wider; however, expanding the architecture of neural networks usually brings more calculations. When the EEG features are large, strided convolution and pooling can be conveniently skipped, and the model can still learn the features well; but in the case of small EEG features, there is not too much redundant information, and strided convolution and pooling are no longer applicable, and problems of information loss and poor learning of features will be encountered. Most neural networks have difficulty learning the structural information of features. Summary of the invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] The embodiments of the present application aim to at least solve the technical problems existing in the prior art. The embodiments of the present application provide a human body pressure detection method, system, device and medium, which can truncate the gradient flow to prevent too much repeated gradient information from being repeatedly learned, greatly reduce the amount of calculation, and improve the learning efficiency of the network. At the same time, non-stride convolution can be used to downsample the feature map without losing the learnable information; the EEG features are input into the dynamic routing algorithm to identify and judge the work pressure state, and the structural information of the features can be identified through the calculation of vector features.

[0005] An embodiment of the first aspect of the present application, a method for detecting human body pressure, comprises:

[0006] Obtaining human brain electrical signals;

[0007] Extracting features from the EEG signal to obtain EEG features;

[0008] Input the EEG feature to a cross-stage convolution module including multiple dense layers, so that the cross-stage convolution module divides the EEG feature into a first EEG feature and a second EEG feature in the channel dimension, and uses the second EEG feature as the input feature of the first dense layer. The current dense layer convolves the input feature to obtain a third EEG feature, concatenates the input feature and the third EEG feature to obtain a fourth EEG feature, and uses the fourth EEG feature as the input feature of the next dense layer, convolves the fourth EEG features output by multiple dense layers to obtain a fifth EEG feature, and concatenates and re-convolves the first EEG feature and the fifth EEG feature to obtain a sixth EEG feature;

[0009] Performing non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature;

[0010] Performing weighted summation on the primary capsule units formed by encapsulating the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and compressing and activating the predicted feature vector to obtain a high-level capsule unit;

[0011] Pressure classification is performed according to the vector length of the advanced capsule unit to obtain a pressure value.

[0012] In certain embodiments of the first aspect of the present application, after acquiring the EEG signal of the human body, the human body stress detection method further includes:

[0013] The EEG signal is preprocessed as follows to obtain a preprocessed EEG signal:

[0014] Filtering the EEG signal through a bandpass filter;

[0015] performing independent component analysis on the EEG signal to eliminate artifacts;

[0016] The electroencephalogram signal is segmented into non-overlapping time windows to obtain a plurality of electroencephalogram signal segments.

[0017] In certain embodiments of the first aspect of the present application, extracting features from the EEG signal to obtain EEG features includes:

[0018] Calculating a phase locking value for the EEG signal;

[0019] A symmetric connection matrix is ​​constructed as the EEG feature according to the phase locking value between each channel of the EEG signal.

[0020] In certain embodiments of the first aspect of the present application, before inputting the EEG feature into a cross-stage convolution module consisting of a transition layer and a plurality of dense layers, the human body stress detection method further includes:

[0021] The EEG features are subjected to preliminary feature extraction through a preliminary convolution layer with a step size equal to a preset step size value.

[0022] In certain embodiments of the first aspect of the present application, the convolution performed by the dense layer on the input features is a zero-padded convolution.

[0023] In certain embodiments of the first aspect of the present application, performing non-stride convolution on the sixth EEG feature to obtain the seventh EEG feature comprises:

[0024] Downsampling the sixth EEG feature according to a preset scale factor to obtain a plurality of sub-EEG features, and splicing the plurality of sub-EEG features to achieve feature transformation to obtain a transformed EEG feature;

[0025] The transformed EEG feature is subjected to non-zero-padded non-stride convolution to obtain a seventh EEG feature.

[0026] In certain embodiments of the first aspect of the present application, the advanced capsule unit is obtained from the primary capsule unit through a capsule network, and the weight parameters of the capsule network are updated by back propagation of a loss function, and the loss function is determined by the category of stress classification judgment, the indicator function of stress classification judgment, the upper and lower boundary values ​​of stress classification judgment, and the seventh EEG feature corresponding to the category of stress classification judgment.

[0027] According to an embodiment of the second aspect of the present application, a human body pressure detection device includes:

[0028] An input unit, used for acquiring brain electrical signals of a human body;

[0029] A feature extraction unit, used to extract features from the EEG signal to obtain EEG features;

[0030] A cross-stage convolution unit, used for dividing the EEG feature into a first EEG feature and a second EEG feature in the channel dimension, taking the second EEG feature as the input feature of the first dense layer, causing the current dense layer to convolve the input feature to obtain a third EEG feature, concatenating the second EEG feature and the third EEG feature to obtain a fourth EEG feature and taking the fourth EEG feature as the input of the next dense layer, convolving the fourth EEG features output by multiple dense layers to obtain a fifth EEG feature, and concatenating and re-convolving the first EEG feature and the fifth EEG feature to obtain a sixth EEG feature;

[0031] a non-stride convolution unit, used for performing non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature;

[0032] A capsule network unit, configured to perform weighted summation on the primary capsule units encapsulated by the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and to compress and activate the predicted feature vector to obtain a high-level capsule unit;

[0033] The pressure classification unit is used to perform pressure classification judgment according to the vector length of the advanced capsule unit to obtain a pressure value.

[0034] An embodiment of the third aspect of the present application is an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the human body pressure detection method as described above when executing the computer program.

[0035] An embodiment of the fourth aspect of the present application is a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the human body pressure detection method as described above.

[0036] The embodiments of the present application obtain the EEG signals of the human body; extract the features of the EEG signals to obtain EEG features; input the EEG features to a cross-stage convolution module including multiple dense layers, so that the cross-stage convolution module divides the EEG features into a first EEG feature and a second EEG feature in the channel dimension, uses the second EEG feature as the input feature of the first dense layer, convolves the input feature with the current dense layer to obtain a third EEG feature, concatenates the input feature with the third EEG feature to obtain a fourth EEG feature, and uses the fourth EEG feature as the input feature of the next dense layer, convolves the fourth EEG features output by multiple dense layers to obtain a fifth EEG feature, concatenates the first EEG feature and the fifth EEG feature and then convolves them to obtain a sixth EEG feature; performs non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature; and according to the dynamic The state routing performs weighted summation on the primary capsule units formed by encapsulating the seventh EEG feature to obtain a predicted feature vector, and compresses and activates the predicted feature vector to obtain an advanced capsule unit; the pressure classification is judged according to the vector length of the advanced capsule unit to obtain the pressure value; the gradient flow is split so that the gradient flow is propagated through different network paths, and the propagated gradient information can also have a large correlation difference; the copy operation in the dense layer can retain the feature reuse characteristics and ensure the validity of the output features. At the same time, truncating the gradient flow can prevent excessive repeated gradient information and prevent different dense layers from repeatedly learning the copied gradient information, greatly reducing the amount of calculation, and the learning ability of the network will be greatly improved. At the same time, the use of non-strided convolution ensures that the effective information of the EEG features will not be lost, and the use of dynamic routing enables the network to learn the structural information of the EEG features. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings are used to provide further understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0038] Figure 1 is a step diagram of a human body pressure detection method provided in an embodiment of the present application;

[0039] Figure 2 is a structural diagram of a human body pressure detection device provided in an embodiment of the present application;

[0040] Figure 3 It is a schematic diagram of the principle of the identification network provided in the embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0042] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims or the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0043] The embodiments of the present application are further described below in conjunction with the accompanying drawings.

[0044] An embodiment of the present application provides a method for detecting human body pressure.

[0045] Reference Figure 1 , a human body pressure detection method, including but not limited to the following steps:

[0046] Step S100, obtaining an electroencephalogram signal of a human body;

[0047] Step S200, extracting features from the EEG signal to obtain EEG features;

[0048] Step S300, inputting the EEG feature into a cross-stage convolution module including multiple dense layers, the cross-stage convolution module divides the EEG feature into a first EEG feature and a second EEG feature in the channel dimension, the dense layer convolves the second EEG feature to obtain a third EEG feature, concatenates the second EEG feature and the third EEG feature to obtain a fourth EEG feature and uses the fourth EEG feature as input to the next dense layer, convolves the fourth EEG feature output by the multiple dense layers to obtain a fifth EEG feature, and concatenates the first EEG feature and the fifth EEG feature and then convolves to obtain a sixth EEG feature;

[0049] Step S400, performing non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature;

[0050] Step S500, performing weighted summation on the primary capsule units formed by encapsulating the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and performing compression and activation on the predicted feature vector to obtain a high-level capsule unit;

[0051] Step S600, performing pressure classification judgment according to the vector length of the high-level capsule unit to obtain a pressure value.

[0052] In step S100, an electroencephalogram (EEG) signal is collected from a human body by an EEG collection device.

[0053] The electrical activity of the cerebral cortex can generate EEG signals. The main forms of electrical activity in the cerebral cortex are spontaneous EEG activity and induced electrical activity. Spontaneous potential activity refers to the corresponding brain wave frequency fluctuations that people show autonomously under different emotional states. Evoked potentials need to be generated by external stimulation. EEG signals can be reflected by electroencephalogram (EEG). EEG is a graph obtained by amplifying and recording the spontaneous biopotential of the cerebral cortex from the scalp through precision instruments. It is the spontaneous and rhythmic electrical activity of brain cell groups recorded by electrodes. This electrical activity is a plane diagram with potential as the vertical axis and time as the horizontal axis, thus recording the relationship between potential and time. The frequency (period), amplitude and phase of brain waves constitute the basic characteristics of EEG.

[0054] There are two ways to collect EEG signals, namely embedded collection and non-embedded collection.

[0055] The embedded acquisition method involves opening the brain and implanting sensors inside.

[0056] The non-embedded EEG signal acquisition method is to collect EEG signals by placing electrode sensors outside the cerebral cortex. This method is widely used in EEG products. In non-embedded EEG signal acquisition, wet electrodes or dry electrodes can be used. When using wet electrodes to measure EEG signals, the collected signals are highly accurate. The use of wet electrodes requires professional operators to apply conductive paste on the head of the subject to amplify the EEG signals. There are three disadvantages of using wet electrodes. First, the measurement operation is more troublesome. The subject needs a professional to apply the paste, and the subject needs to wash his hair after the test. The steps are cumbersome and time-consuming; second, the stability of the conductive paste as a conductive medium deteriorates over time, so the test time should not be too long; third, the wet electrode needs to be directly connected to the computer using a wire. Compared with using wet electrodes to measure EEG signals, dry electrodes use high input impedance technology and are less dependent on dielectrics. Secondly, dry electrodes use wireless communication technology to collect EEG signals, which reduces the constraints of the environment on them, so they can be applied to a wider range of scenarios. However, it is worth noting that the EEG signals that can be detected by dry electrodes are weaker than those of wet electrodes. In this case, it is necessary to enhance the accuracy of EEG signals by increasing the number of electrodes.

[0057] After acquiring the human body's EEG signal, the EEG signal is preprocessed to obtain a preprocessed EEG signal. Specifically, the EEG signal is filtered through a bandpass filter (0.5Hz-48Hz); then the filtered EEG signal is subjected to independent component analysis to eliminate artifacts; and then the EEG signal is segmented into non-overlapping time windows to obtain multiple EEG signal segments.

[0058] For step S200, feature extraction is performed on the EEG signal to obtain EEG features, including but not limited to the following steps:

[0059] Calculate the phase locking value of the EEG signal;

[0060] A symmetric connection matrix is ​​constructed as the EEG feature according to the phase locking value between each channel of the EEG signal.

[0061] Specifically, the phase lock value can be expressed by the following formula: Among them, n t is the number of time points.

[0062] The PLV value between brain region k and brain region l is in the range [0,1], where 1 reflects perfect phase synchronization between brain regions and 0 reflects no phase synchronization. The PLV is calculated repeatedly between all brain channels. Afterwards, the PLV values ​​between each channel are formed into a symmetric connection matrix.

[0063] The EEG features of the symmetrical connection matrix are subjected to preliminary feature extraction through a preliminary convolution layer with a step size of a preset step size, and the features obtained by the preliminary feature extraction are input into the cross-stage convolution module. The preset step size value can be 1.

[0064] Reference Figure 3 , Figure 3 It is a schematic diagram of the principle of the recognition network. The EEG features are input into the recognition network for recognition and the pressure value is output. For step S300, the EEG features are input into the cross-stage convolution module. The solid line box in the figure represents the cross-stage convolution module. The cross-stage convolution module includes multiple dense layers and transition layers.

[0065] The cross-stage convolution module divides the EEG feature X0 into the first EEG feature X 0′ and the second EEG feature X 0″ The second EEG feature is used as the input feature of the first dense layer.

[0066] The current dense layer convolves the input features to obtain the third EEG feature, concatenates the input features and the third EEG feature in the channel dimension to obtain the fourth EEG feature, and uses the fourth EEG feature as the input feature of the next dense layer. In the dense layer, in order to facilitate concatenation with the original features, you can choose zero-filled convolution.

[0067] After n layers of dense layers, the output [X 0″ , X1, ..., X n ], the output of the nth dense layer can be expressed as follows: X n =w n *[X 0″ , X1, ..., X n-1 ], through feed-forward transfer of weight w n Update the weight w n The update of can be expressed by the following formula: n =f(w n , g 0″ , g1, g2, …, g n-1 ).

[0068] The fourth EEG feature [X 0″ , X1, ..., X n ] is convolved to obtain the fifth EEG feature X T The process can be expressed by the following formula: T =w T *[X 0″ , X1, ..., X n ]. By feed-forward transfer, the weight w T Update the weight w TThe update of can be expressed by the following formula: T =f(w T , g 0″ , g1, g2, ..., g n ).

[0069] The transition layer converts the first EEG feature X 0′ and the fifth EEG feature X T Splicing and convolution to get the sixth EEG feature X U The process can be expressed by the following formula: U =w U *[X 0′ , X T ]. By feed-forward transfer, the weight w U Update the weight w U The update of can be expressed by the following formula: U =f(w U , g 0′ , g T ). The purpose of the transition layer is to maximize the difference in the combination of gradients.

[0070] In the above formula, * represents the convolution operator, f represents the weight update function, and g i represents the gradient propagated to the i-th dense layer.

[0071] The size of the EEG feature after input preprocessing is L×L, and the feature size obtained after the cross-stage convolution module is L1×L1×C1, where C1 is the number of filters in the convolution.

[0072] In the weight update of the final transition layer, the gradients from the dense layer are integrated separately without passing through the feature map X of the dense layer. 0′ It is also integrated separately, without repeated gradients of other dense layers. By splitting the gradient flow, the gradient flow is propagated through different network paths, and the propagated gradient information can also have a large correlation difference. The copy operation in the dense layer can retain the feature reuse characteristics and ensure the validity of the output features. At the same time, truncating the gradient flow can prevent excessive repeated gradient information and prevent different dense layers from repeatedly learning the copied gradient information, greatly reducing the amount of calculation, and the learning ability of the network will be greatly improved.

[0073] Reference Figure 3 For the portion of the dotted line frame in step S400, performing non-stride convolution on the sixth EEG feature to obtain the seventh EEG feature includes but is not limited to the following steps:

[0074] The sixth EEG feature with a size of L1×L1×C1 is down-sampled according to a preset scale factor (scale) through a space-to-depth (SPD) transformation layer to obtain multiple sub-EEG features, and the multiple sub-EEG features are spliced ​​according to the channel dimension to achieve feature transformation to obtain a transformed EEG feature; the size of each sub-EEG feature is L1 / 2×L1 / 2×C1.

[0075] Specifically, the downsampling process can be expressed by the following formula: 0，0 =X[0:L1:scale,0:L1:scale],f 1，0 =X[1:L1:scale,0:L1:scale],…,f scale-1，0 =X[scale-1:L1:scale,0:L1:scale],f 0，1 =X[0:L1:scale,l:L1:scale],…,f scale-1，1 =X[scale-1:L1:scale,1:L1:scale],…,f 0，sca1e-1 =X[0:L1:scale,scale-1:L1:scale],…,f scale-1，scale-1 =X[scale-1:L1:scale, scale-1:L1:scale].

[0076] The transformed EEG feature is convolved with non-zero padding through a non-strided convolution layer with a C2 filter to obtain the seventh EEG feature of size L2×L2×C2; where C2<scale 2 C1. The stride of the non-strided convolution layer is 1. Using non-strided convolution can retain all discriminative feature information as much as possible. Figure 3 The non-strided convolution in is taken as an example with scale=2.

[0077] In step S500, after the seventh EEG feature is converted into a vector feature in the dimension of the channel through the capsule network, the feature vector is encapsulated into a primary capsule unit u i .

[0078] The primary capsule units are weighted and summed according to the dynamic routing, that is, the primary capsule units and the connection weight matrix W ij Multiply them together to get the predicted feature vector This process can be expressed by the following formula:

[0079] Using the coupling coefficient c ij The predicted feature vector is weighted summed, where the coupling coefficient c ijDetermined by the dynamic routing algorithm. The process can be expressed by the following formula:

[0080] Will s j Perform compression activation to obtain advanced capsule unit v j The process can be expressed by the following formula:

[0081] The advanced capsule unit is obtained from the primary capsule unit through the capsule network, and the weight parameters of the capsule network are updated by back propagation of the loss function.

[0082] The loss function is determined by the category of pressure classification judgment, the indicator function of pressure classification judgment, the upper and lower boundary values ​​of pressure classification judgment, and the high-level capsule unit corresponding to the category of pressure classification judgment.

[0083] Specifically, the loss function can be expressed as: L k =T k max(0,m + -||v k ||) 2 +λ(1-T k )max(0,||v k ||-m - ) 2 ; In the formula, k is the classification category; T k is the indicator function of classification. If it does not belong to category k, T k =1, otherwise T k =0;v k The dynamic routing cluster is a vector unit of category k; m + 、m - are the upper and lower boundaries respectively; λ is the weight to reduce the loss value.

[0084] For step S600, pressure classification is performed according to the vector length of the high-level capsule unit to obtain a pressure value, which can be intuitively displayed through a visualization device.

[0085] An embodiment of the present application also provides a human body pressure detection device.

[0086] Reference Figure 2 The human body pressure detection device includes an input unit 110, a feature extraction unit 120, a strided convolution unit 130, a non-strided convolution unit 140, a capsule network unit 150 and a pressure classification unit 160.

[0087] Among them, the input unit 110 is used to obtain the human body's EEG signal; the feature extraction unit 120 is used to extract features from the EEG signal to obtain EEG features; the cross-stage convolution unit 130 is used to divide the EEG features into a first EEG feature and a second EEG feature in the channel dimension, and use the second EEG feature as the input feature of the first dense layer, so that the current dense layer convolves the input feature to obtain a third EEG feature, concatenates the second EEG feature and the third EEG feature to obtain a fourth EEG feature and uses the fourth EEG feature as the input of the next dense layer, and uses the fourth EEG feature output by multiple dense layers. The first EEG feature is convolved with the fifth EEG feature to obtain the fifth EEG feature, and the first EEG feature and the fifth EEG feature are concatenated and then convolved to obtain the sixth EEG feature; the non-stride convolution unit 140 is used to perform non-stride convolution on the sixth EEG feature to obtain the seventh EEG feature; the capsule network unit 150 is used to perform weighted summation on the primary capsule units formed by encapsulating the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and compress and activate the predicted feature vector to obtain an advanced capsule unit; the pressure classification unit 160 is used to perform pressure classification judgment according to the vector length of the advanced capsule unit to obtain a pressure value.

[0088] It can be understood that the contents in the human body pressure detection method embodiment are applicable to the human body pressure detection device embodiment, the functions specifically implemented by the human body pressure detection device embodiment are the same as those in the human body pressure detection method embodiment, and the beneficial effects achieved are also the same as those achieved by the human body pressure detection method embodiment.

[0089] The embodiment of the present application further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above human body pressure detection method when executing the computer program.

[0090] The memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer executable programs, such as the human body pressure detection in the above-mentioned embodiment of the present invention. The processor implements the human body pressure detection in the above-mentioned embodiment of the present invention by running the non-transitory software programs and programs stored in the memory.

[0091] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data required for executing the human body pressure detection in the above-mentioned embodiment of the present invention, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0092] The embodiment of the present application further provides a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the above human body pressure detection method.

[0093] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. In the above description of this specification, the description of reference terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" and the like means that the specific features, structures, materials or characteristics described in conjunction with the embodiments or examples are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0094] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is limited by the embodiments and their equivalents.

[0095] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the present embodiment.

Claims

1. A method for detecting human body pressure, comprising: Obtaining human brain electrical signals; Extracting features from the EEG signal to obtain EEG features; Input the EEG feature to a cross-stage convolution module including multiple dense layers, so that the cross-stage convolution module divides the EEG feature into a first EEG feature and a second EEG feature in the channel dimension, and uses the second EEG feature as the input feature of the first dense layer. The current dense layer convolves the input feature to obtain a third EEG feature, concatenates the input feature and the third EEG feature to obtain a fourth EEG feature, and uses the fourth EEG feature as the input feature of the next dense layer, convolves the fourth EEG features output by multiple dense layers to obtain a fifth EEG feature, and concatenates and re-convolves the first EEG feature and the fifth EEG feature to obtain a sixth EEG feature; Performing non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature; Performing weighted summation on the primary capsule units formed by encapsulating the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and compressing and activating the predicted feature vector to obtain a high-level capsule unit; Pressure classification is performed according to the vector length of the advanced capsule unit to obtain a pressure value.

2. A human body pressure detection method according to claim 1, characterized in that: After obtaining the brain wave signal of the human body, the human body stress detection method further includes: The EEG signal is preprocessed as follows to obtain a preprocessed EEG signal: Filtering the EEG signal through a bandpass filter; performing independent component analysis on the EEG signal to eliminate artifacts; The electroencephalogram signal is segmented into non-overlapping time windows to obtain a plurality of electroencephalogram signal segments.

3. A human body pressure detection method according to claim 1, characterized in that: The step of extracting features from the EEG signal to obtain EEG features includes: Calculating a phase locking value for the EEG signal; A symmetric connection matrix is ​​constructed as the EEG feature according to the phase locking value between each channel of the EEG signal.

4. A human body pressure detection method according to claim 1, characterized in that: Before inputting the EEG feature into a cross-stage convolution module consisting of a transition layer and a plurality of dense layers, the human body stress detection method further includes: The EEG features are subjected to preliminary feature extraction through a preliminary convolution layer with a step size equal to a preset step size value.

5. A human body pressure detection method according to claim 1, characterized in that: The dense layer performs convolution on the input features in a zero-filled convolution manner.

6. A human body pressure detection method according to claim 1, characterized in that: The performing non-stride convolution on the sixth EEG feature to obtain the seventh EEG feature includes: Downsampling the sixth EEG feature according to a preset scale factor to obtain a plurality of sub-EEG features, and splicing the plurality of sub-EEG features to achieve feature transformation to obtain a transformed EEG feature; The transformed EEG feature is subjected to non-zero-padded non-stride convolution to obtain a seventh EEG feature.

7. A human body pressure detection method according to claim 1, characterized in that: The advanced capsule unit is obtained from the primary capsule unit through a capsule network, and the weight parameters of the capsule network are updated by back propagation of a loss function, wherein the loss function is determined by the category of pressure classification judgment, the indicator function of pressure classification judgment, the upper and lower boundary values ​​of pressure classification judgment, and the advanced capsule unit corresponding to the category of pressure classification judgment.

8. A human body pressure detection device, characterized in that: include: An input unit, used for acquiring brain electrical signals of a human body; A feature extraction unit, used to extract features from the EEG signal to obtain EEG features; A cross-stage convolution unit, used for dividing the EEG feature into a first EEG feature and a second EEG feature in the channel dimension, taking the second EEG feature as the input feature of the first dense layer, causing the current dense layer to convolve the input feature to obtain a third EEG feature, concatenating the second EEG feature and the third EEG feature to obtain a fourth EEG feature and taking the fourth EEG feature as the input of the next dense layer, convolving the fourth EEG features output by multiple dense layers to obtain a fifth EEG feature, and concatenating and re-convolving the first EEG feature and the fifth EEG feature to obtain a sixth EEG feature; a non-stride convolution unit, used for performing non-stride convolution on the sixth EEG feature to obtain a seventh EEG feature; A capsule network unit, configured to perform weighted summation on the primary capsule units encapsulated by the seventh EEG feature according to dynamic routing to obtain a predicted feature vector, and to compress and activate the predicted feature vector to obtain a high-level capsule unit; The pressure classification unit is used to perform pressure classification judgment according to the vector length of the advanced capsule unit to obtain a pressure value.

9. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the human body pressure detection method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to execute the human body pressure detection method as described in any one of claims 1 to 7.

Images