Method and apparatus for detecting abnormal biological signals

Abstract

To generate a model for detecting biological signals and to detect abnormal biological signals using the model with a monitor of the biological signal.SOLUTION: A device 1 for detecting abnormal biological signals can communicate with a sensor 2 and one or more external control devices 31, 32, 33, 34. The device 1 receives input data collected by the sensor 2. The sensor 2 emits radio waves toward a human body being the monitoring object, receives radio waves reflected from the surface of the object, and thereby collects characteristics relating to the shape or movement of the object. Thus, the sensor 2 collects changes in the thorax caused by respiration or heartbeat of the human body being the monitoring object. The obtained data is input to a neural network learned on the basis of predetermined learning data in a neuromorphic device that implements the neural network for detecting abnormal biological signals, thereby detecting abnormal biological signals.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] The present invention relates to a method and apparatus for detecting abnormal biological signals, and more particularly to a method and apparatus for generating a model for detecting abnormal biological signals, and a method and apparatus for detecting abnormal biological signals by monitoring the biological signals using the generated model. [Background technology]

[0002] Artificial neural networks mimic biological neural networks, but they are trained by a large amount of input data and are used to estimate or approximate results that are difficult to derive using conventional techniques. Artificial neural networks contain layers of interconnected neurons that exchange signals, and synapses have weights that can be tuned based on learning or experience. To train an artificial neural network, good quality training data is required, and the performance of the artificial neural network can vary depending on the training data.

[0003] Meanwhile, heart disease accounts for a high percentage of the leading causes of death both domestically and internationally. In particular, unintentional sudden death due to suffocation while sleeping occurs most frequently in infants aged 2 to 4 months. In addition, as the rate of lonely deaths gradually increases, there is a growing need to reduce social costs through prevention and management of lonely deaths.

[0004] The above-mentioned background art is technical information that the inventor possessed in order to derive the present invention or that he acquired in the process of deriving the present invention, and is not necessarily publicly known art that was disclosed to the general public prior to the filing of the present invention. Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present disclosure is to provide a method and apparatus for detecting abnormal biological signals. The problems to be solved by the present disclosure are not limited to the technical problems described above, and other technical problems not described will be clearly understood by those skilled in the art from the description of the present disclosure and will be further clearly understood by the embodiments of the present disclosure. It will also be understood that the problems and advantages to be solved by the present disclosure can be achieved by the means and combinations thereof set forth in the claims. [Means for solving the problem]

[0006] As a means for solving the above-mentioned technical problems, a first aspect of the present disclosure can provide a method for detecting an abnormal biological signal, including the steps of: randomly extracting frames having a first data length from raw data containing biological signal data to generate a plurality of sample data; extracting frames having a second data length from the sample data in chronological order to generate a plurality of unit data; generating a training dataset based on the plurality of unit data; and generating an abnormal biological signal detection model by training an artificial intelligence model with the training dataset.

[0007] A second aspect of the present disclosure can provide an apparatus for detecting abnormal biological signals, which includes a memory storing at least one program and a processor that performs calculations by executing the at least one program, wherein the processor randomly extracts frames having a first data length from raw data containing biological signal data to generate a plurality of sample data, extracts frames having a second data length from the sample data in chronological order to generate a plurality of unit data, generates a training dataset based on the plurality of unit data, and generates an abnormal biological signal detection model by training an artificial intelligence model with the training dataset.

[0008] A third aspect of the present disclosure can provide a computer-readable recording medium having recorded thereon a program for causing a computer to execute the method of the first aspect.

[0009] A fourth aspect of the present disclosure can provide a neuromorphic device that realizes a neural network for detecting abnormal biological signals, the neuromorphic device including at least one processor that drives the neural network, and on-chip memory including a crossbar array circuit that receives instructions from the at least one processor and performs in-memory calculations, the at least one processor receiving input data collected by a sensor and inputting the input data to the neural network that has been trained based on predetermined training data to detect the abnormal biological signals, the input data including biological signals and noise signals, the neural network including an input layer to which the input data is input, one or more hidden layers that extract features of the input data, and an output layer that outputs an output signal.

[0010] A fifth aspect of the present disclosure can provide a method for detecting an abnormal biological signal using a neural network, comprising: receiving input data collected by a sensor; and inputting the input data into the neural network trained based on predetermined training data to detect the abnormal biological signal, wherein the input data includes a biological signal and a noise signal; and the neural network includes an input layer to which the input data is input, one or more hidden layers to extract features of the input data, and an output layer to output the abnormal biological signal.

[0011] A sixth aspect of the present disclosure can provide a computer-readable recording medium having recorded thereon a program for causing a computer to execute the method of the fifth aspect.

[0012] In addition to these, other methods and devices for realizing the present invention, and computer-readable recording media having recorded thereon programs for executing the methods can also be provided.

[0013] Other aspects, features, and advantages beyond those described above will be apparent from the accompanying drawings, the claims, and the following detailed description of the invention. [Effects of the Invention]

[0014] According to the above-described means for solving the problems of the present disclosure, it is possible to prevent and manage unintentional sudden deaths or solitary deaths of infants due to suffocation accidents while sleeping, thereby reducing social costs.

[0015] Furthermore, according to the means for solving the problems disclosed herein, erroneous recognition of abnormal biological signals, which significantly increases social costs, can be minimized, and a highly accurate machine learning model can be constructed.

[0016] The effects of the embodiments are not limited to those described above, and other effects not described will be clearly understood by those skilled in the art from the description of the present invention. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 1 is a diagram illustrating the architecture of a fully-connected neural network (FCNN) according to an embodiment of the present invention. [Figure 2] 1 is an exemplary diagram for comparing a Von Neumann structure and a Computing-In-Memory (CIM) structure according to an embodiment of the present invention; [Figure 3] 1 is an exemplary diagram for comparing a Von Neumann structure and a Computing-In-Memory (CIM) structure according to an embodiment of the present invention; [Figure 4] FIG. 1 illustrates a neural network according to one embodiment of the present invention. [Figure 5] 1 is an exemplary diagram of an environment for detecting abnormal biological signals according to an embodiment of the present invention; [Figure 6] 1 is an exemplary diagram of an environment for detecting abnormal biological signals according to an embodiment of the present invention; [Figure 7] FIG. 1 is a diagram illustrating raw data according to an embodiment of the present invention. [Figure 8] FIG. 10 is a diagram illustrating a method for generating a plurality of sample data according to an embodiment of the present invention. [Figure 9] 10A and 10B are diagrams illustrating a method for generating a plurality of unit data according to an embodiment of the present invention; [Figure 10] 10A and 10B are diagrams illustrating a method for generating a plurality of unit data according to an embodiment of the present invention; [Figure 11] FIG. 1 is a diagram illustrating a method for generating a training dataset according to an embodiment of the present invention. [Figure 12] FIG. 1 is a diagram illustrating a method for generating a training dataset according to an embodiment of the present invention. [Figure 13] FIG. 1 is a diagram illustrating a method for generating a training dataset according to an embodiment of the present invention. [Figure 14] 1 is a graph showing the results of training an artificial intelligence model with a training dataset according to an embodiment of the present invention. [Figure 15] 1 is a block diagram of an apparatus for detecting abnormal biological signals according to one embodiment of the present invention; DETAILED DESCRIPTION OF THE INVENTION

[0018] In describing the present invention, if it is determined that a specific description of related publicly known technology would obscure the gist of the present invention, the detailed description may be omitted, and unless otherwise defined, all terms used in this specification have the same meaning as commonly understood by a person having ordinary skill in the art to which the present invention belongs.

[0019] The appearances of phrases such as "in one embodiment," "in accordance with one embodiment," and "by implementing one embodiment" in this specification do not necessarily all refer to the same embodiment.

[0020] Since various modifications can be made to the embodiments and various forms can be taken, some embodiments are illustrated in the drawings and described in detail. However, it should be understood that the embodiments are not limited to the specific disclosed forms, but include all modifications, equivalents, and alternatives within the spirit and technical scope of the embodiments. The terms used in the specification are used merely to describe the embodiments and are not intended to limit the embodiments.

[0021] The terms used in the embodiments are generally selected as widely used terms as possible, taking into consideration the functions of the embodiments. However, these may vary depending on the intentions of engineers in the technical field to which the embodiments pertain, legal precedents, the emergence of new technologies, etc. In addition, in certain cases, the applicant may arbitrarily select terms, and in such cases, their meanings will be described in detail in the relevant section. Therefore, the terms used in the embodiments should be defined based on the meanings of the terms and the overall content of the embodiments, rather than simply the names of the terms.

[0022] Some embodiments of the present disclosure may be illustrated by functional block configurations and various processing steps. Some or all of such functional blocks may be implemented by any number of hardware and / or software configurations that perform specific functions. For example, the functional blocks of the present disclosure may be implemented by one or more microprocessors or by circuit configurations for a given function.

[0023] Also, for example, the functional blocks of the present disclosure may be implemented in various programming or scripting languages, or may be implemented as algorithms executed on one or more processors. Furthermore, the present disclosure may employ conventional techniques for electronic configuration, signal processing, and / or data processing, etc.

[0024] Terms such as "database," "element," "means," and "configuration" can be used broadly and are not limited to mechanical and physical configurations. Furthermore, terms such as "unit" and "module" used in the specification refer to a unit that processes at least one function or operation, and can be realized by hardware or software, or a combination of hardware and software.

[0025] Note that the connecting lines or connecting members between components shown in the drawings are merely exemplary of functional and / or physical connections or circuit connections, and in an actual device, connections between components may be represented by various alternative or additional functional, physical, or circuit connections.

[0026] Furthermore, terms including ordinal numbers such as "first" and "second" used in this specification may be used to describe various components, but the components should not be limited by the terms. The terms are used only to distinguish one component from another.

[0027] In addition, the size and proportions of some components in the drawings may be somewhat exaggerated. Furthermore, components shown in one drawing may not be shown in other drawings.

[0028] Throughout the specification, the term "embodiment" is an arbitrary category for easily describing the invention in the present disclosure, and the respective embodiments do not necessarily have to be mutually exclusive. For example, a configuration disclosed in one embodiment can be applied and / or realized in other embodiments, and can be applied and / or realized with modifications without departing from the scope of the present disclosure.

[0029] Furthermore, the terms used in this disclosure are for the purpose of describing the embodiments and are not intended to limit the embodiments. In this disclosure, the singular includes the plural unless otherwise specified.

[0030]

[0030] Hereinafter, with reference to the accompanying drawings, embodiments of the present disclosure will be described in detail so that those skilled in the art can easily implement them. However, the embodiments of the present disclosure may be realized in various different forms and are not limited to the embodiments described in the present disclosure.

[0031] Based on this, the present invention will be described in detail below with reference to the drawings.

[0032] A neuromorphic chip is hardware that mimics the functions of the human brain by creating circuits that mimic the morphology of neurons. In other words, a neuromorphic chip is a computer chip that mimics the structure of the nervous system.

[0033] Neuromorphic chips are composed only of the circuits necessary for neural network operations, enabling gains of hundreds of times in terms of power, area, and speed. Unlike conventional computers, the human brain does not consume much power even when processing large amounts of data. Neuromorphic chips mimic the brain's operating method, configuring a parallel structure connecting neurons and synapses and disconnecting the connections when data is not being processed, thereby saving energy.

[0034] For example, conventional computers with a von Neumann architecture process input data sequentially, making them excellent for executing precisely written programs. However, they have limitations in power consumption and are inefficient in pattern recognition and real-time recognition.

[0035] In contrast, neuromorphic chips do not represent data as digital values ​​such as 0 or 1, but instead use analog operations in which various states change gradually. In other words, artificial neurons configured in parallel operate in an event-driven manner without a clock. This allows them to efficiently process irregular characters, sounds, images, and other data that conventional computers found difficult to intuitively recognize. Specifically, data can be processed in parallel by distributing neurons, which are nerve cells, and synapses, which are connecting lines, using silicon transistor circuits and memory elements.

[0036] In one embodiment, when input data such as an image, sound, or electromagnetic wave is input to a neuromorphic chip, the neuromorphic chip performs an internal calculation on the input data to output predetermined output data. The predetermined output data may include a sound / image / electromagnetic wave recognition result based on a feature classification of the input data. For example, when an electromagnetic wave signal is input as input data, the predetermined output data may include an electromagnetic wave recognition result that classifies the input data as to whether or not it contains a previously learned abnormal signal.

[0037] Meanwhile, the data input to the neuromorphic chip is not limited to the aforementioned images, sounds, or electromagnetic waves, but may include various forms of data such as video and text.

[0038] According to one embodiment, a neuromorphic device may be implemented using an edge AI chip. Edge AI refers to a technology that executes AI algorithms on hardware devices using edge computing based on system-generated data. AI processing is primarily performed in cloud-based data centers, which require enormous computing capacity and are therefore highly dependent on servers. In contrast, edge AI reduces dependency on the cloud (server) by executing AI algorithm operations locally, thereby reducing communication costs and protecting privacy by preventing sensitive personal information from being transmitted to the cloud. Therefore, configuring a neuromorphic device with an edge AI chip not only reduces costs and improves security, but also enables a highly responsive system because operations are immediately processed within the same hardware.

[0039] In the following description, the apparatus for detecting abnormal biological signals according to an embodiment of the present invention may be the aforementioned neuromorphic device, that is, the aforementioned neuromorphic device can function as the apparatus for detecting abnormal biological signals according to an embodiment of the present invention.

[0040] FIG. 1 is a diagram illustrating the architecture of a fully-connected neural network (FCNN) according to one embodiment of the present invention.

[0041] Referring to FIG. 1, FCNN, which is a fully connected layer, is a layer in which all neurons in one layer are connected to all neurons in the next layer, and is included in a convolutional neural network and is used to classify data using a matrix flattened in the form of a one-dimensional array.

[0042] In FCNN, there is a value for each node, and a weight and bias for each neuron. When moving from one layer to another, the value obtained by multiplying each node by its weight and adding the bias becomes the node value of the next layer, but if this calculated value meets certain conditions, the output value that has passed through the activation function is input as the node value of the next layer and activated, and if this calculated value does not meet certain conditions, an activation function intervenes to deactivate the node through the activation function.

[0043] Therefore, since the output value differs depending on the type of activation function, it is important to use an appropriate activation function as needed. Typical examples are the ReLU function and the Softmax function.

[0044] Since only one-dimensional array data can be input to FCNN, if the input data is a three-dimensional image array consisting of vertical, horizontal, and channel (color), it must be flattened into one-dimensional data before being input to FCNN. In other words, the spatial information of the image is ignored, which means that features contained in the shape cannot be extracted. Therefore, FCNN is highly useful when data that can be realized as one-dimensional array data, such as electromagnetic wave data, is used as input. According to one embodiment, the neural network described below may be an FCNN.

[0045] 2 and 3 are exemplary diagrams for comparing a Von Neumann structure and a Computing-In-Memory (CIM) structure according to an embodiment of the present invention.

[0046] Referring to Figure 2, the von Neumann architecture is a computer architecture proposed by John von Neumann, and is a stored-program computer architecture consisting of a typical three-tier structure of a main memory, a central processing unit, and input / output devices.

[0047] The von Neumann architecture has the advantage of greatly improving versatility, since changing a computing device to perform a different task does not require rearranging the hardware (wires, etc.), but rather requires changing only the software (programs). However, because it executes a series of instructions, each of which consists of changing the value of a specific memory location, it poses a serious problem in the design of high-speed computers, known as the von Neumann bottleneck.

[0048] To solve the von Neumann bottleneck phenomenon, alternatives that have been proposed include the Harvard architecture, which divides memory into a section that stores instructions and a section that stores data; the CIM architecture, which not only stores data but also performs data calculations in memory; and neuromorphic computing, which is an integrated circuit in the form of an artificial neural network that mimics the brain structure of higher animals, in which many units that combine calculation and memory functions are connected in parallel in a mesh, and each unit then operates in an event-driven manner.

[0049] Referring to FIG. 3, it can be seen that the CIM structure consists of a processor and memory with computing capabilities.

[0050] Unlike the conventional von Neumann architecture, where all data in memory is moved to the processor for calculation, the CIM architecture performs calculations in memory when a processor instruction is received and only the resulting data is sent to the processor, eliminating the need to move large amounts of data, effectively resolving the von Neumann bottleneck phenomenon mentioned above, and also offering the advantage of significantly lower power consumption.

[0051] A neural network according to an embodiment of the present invention can perform operations using only on-chip memory without using external memory. For example, the neural network can perform operations for each layer on a CIM basis using only on-chip memory without using external memory (e.g., off-chip memory), thereby performing operations without memory updates while processing input signals. Specifically, the neural network can perform CIM-based operations in which each memory cell is directly connected to a processor.

[0052] However, compared to general memory such as RAM (Random Access Memory), CIM may have limitations in memory capacity and power consumption of computing devices. Therefore, producing a high-performance CIM-structured chip may require a new hardware structure, which may increase manufacturing costs. Therefore, a CIM structure may be advantageous for performing relatively small or simple calculations. In contrast, a neural network according to an embodiment of the present invention may be configured with a memory capable of implementing multi-bit data to overcome the drawbacks of such a CIM structure. For example, a neural network may be configured with a memory capable of implementing 7 bits and 128 analog memory states. By configuring a neural network with a large capacity, it can process large amounts of data with low power consumption and high performance even over long periods of use, unlike general CIM chips that exhibit problems such as heat generation and performance degradation.

[0053] Meanwhile, on-chip memory can be realized by a crossbar array circuit. That is, the crossbar array can receive instructions from a processor and perform calculations, and CIM calculations can be achieved by integrating memory elements in a crossbar array structure into the on-chip memory. As an example, a processor can detect abnormal biological signals by receiving input data collected by a sensor, driving a neural network circuit trained based on predetermined training data, and receiving abnormal biological signals output by the neural network circuit. Specifically, the processor can detect abnormal biological signals by inputting input data into the neural network. Below, the configuration of the neural network and the training data for generating an abnormal biological signal detection model implemented by the neural network are described in detail.

[0054] FIG. 4 is a diagram illustrating a neural network according to one embodiment of the present invention.

[0055] 4, in one embodiment, the neural network may include an input layer 410 to which input data 400 is input, one or more hidden layers 420, 430, 440 that extract features of the input data 400, and an output layer 450 that outputs an output signal 460. Here, the input layer 410, the one or more hidden layers 420, 430, 440, and the output layer 450 may refer to each layer of the fully connected neural network (FCNN) described above.

[0056] In one embodiment, the neural network can classify the input data 400 into one of a normal biological signal, an abnormal biological signal, and noise. Specifically, the neural network may include an input layer 410 for classifying the input data 400, one or more hidden layers 420, 430, 440, and an output layer 450.

[0057] In one embodiment, input data 400 including a biological signal and a noise signal may be input to the input layer 410 of the neural network. The biological signal is a signal generated by the movement of a human body, and the noise signal may include a simple movement of the monitored subject (e.g., turning over in bed) that is not a monitored biological signal (e.g., breathing or heartbeat), and the movement of a subject that is not a monitored subject, such as a nearby person or a companion animal.

[0058] In one embodiment, the hidden layers 420, 430, 440 of the neural network can extract or classify features of the input data 400 by applying weights to the input data 400. Therefore, the more hidden layers 420, 430, 440 there are, the more sophisticated the processing of the input data 400 can be, improving the performance of the algorithm.

[0059] In one embodiment, the output layer 450 of the neural network may output whether the input data 400 corresponds to a normal biological signal, an abnormal biological signal, or noise based on the features extracted by at least one of the hidden layers 420, 430, and 440. That is, the output layer 450 may output an output signal indicating whether the neural network has classified the input data 400 as a normal biological signal, an abnormal biological signal, or noise.

[0060] 5 and 6 are diagrams illustrating an environment for detecting abnormal biological signals according to an embodiment of the present invention.

[0061] Referring to FIG. 5, an apparatus for detecting abnormal biological signals (hereinafter referred to as "apparatus") 1 can communicate with a sensor 2 and one or more external control devices 31, 32, 33, .

[0062] As an example, device 1 may receive input data collected by sensor 2 .

[0063] Furthermore, when an abnormal biological signal is detected, the device 1 can transmit an emergency signal to an external control device including at least one of a control center 31, a government office 32, an external terminal 33, and an external server .

[0064] In one embodiment, the biological signals and noise signals included in the input data may be collected by a sensor 2 .

[0065] The human chest changes by breathing on the order of 4 to 12 mm, and the chest changes by heartbeat on the order of 0.6 to 2 mm. Therefore, in one embodiment, the sensor 2 may be a millimeter-wave radar (mmWave Radar) that emits radio waves in the millimeter (mm) band or in the frequency band of 24 GHz to 77 GHz. However, the wavelength band and frequency band of the radio waves of the sensor 2 are not limited to those in the above-described embodiment, and the type of radar is not limited to millimeter-wave radar.

[0066] Here, the sensor 2 emits radio waves to the human body being monitored and receives the radio waves reflected by the surface of the body, thereby collecting characteristics related to the shape or movement of the body. Thus, the sensor 2 can collect changes in the thorax caused by breathing or heartbeat of the human body being monitored.

[0067] Referring to FIG. 6 , in one embodiment, the device 1 may include a communication unit (not shown) that uses at least one communication protocol of a Serial Peripheral Interface (SPI) 12, a Digital Video Port (DVP), a Parallel Serial Peripheral Interface (PSPI), an Inter-Integrated Circuit (I2C), a General Purpose Input / Output (GPIO), a Pulse Width Modulation (PWM), an Inter-IC Sound (I2S), and a Universal Asynchronous Receiver / Transmitter (UART) 13.

[0068] The serial peripheral interface 12 is a synchronous serial communication protocol for communication between multiple peripheral devices and a microcontroller or microprocessor. The serial peripheral interface 12 can be used to exchange data between a single device 1 according to one embodiment and multiple slave devices such as sensors 2 and external memory 21. The universal asynchronous transmitter / receiver 13 is a form of serial communication protocol used to transmit data asynchronously. The universal asynchronous transmitter / receiver 13 can be used for serial communication between a device 1 according to one embodiment and a peripheral device such as an external terminal 33.

[0069] As an example, the device 1 can communicate with the sensor 2 using the serial peripheral interface 12 to receive biological signals and noise signals related to the movement of the thorax of the human body 600 collected by the sensor 2.

[0070] The device 1 can also communicate with the external memory 21 using the serial peripheral interface 12 to send and receive data to and from the external memory 21. The external memory 21 can store at least one of normal biological signals, abnormal biological signals, and noise of the human body 600 detected by the device 1.

[0071] However, when the device 1 communicates with the sensor 2 and / or the external memory 21, it may use at least one of the following communication protocols or communication devices: DVP (Digital Video Port), PSPI (Parallel Serial Peripheral Interface), Universal Asynchronous Transmitter / Receiver (UART), I2C (Inter-Integrated Circuit), GPIO (General Purpose Input / Output), and PWM (Pulse Width Modulation) in addition to the serial peripheral interface 12. That is, the device 1 may include a communication unit (not shown) that uses at least one of the following communication protocols: DVP (Digital Video Port), PSPI (Parallel Serial Peripheral Interface), Universal Asynchronous Transmitter / Receiver (UART), I2C (Inter-Integrated Circuit), GPIO (General Purpose Input / Output), and PWM (Pulse Width Modulation).

[0072] In one embodiment, the device 1 can transmit an emergency signal to an external terminal 33 or the like using a universal asynchronous transceiver 13. Here, the external terminal 33 is merely one embodiment, and it goes without saying that the emergency signal can be transmitted to the control center 31, government agency 32, and external server 34 described above. The external terminal 33 may be a pre-set parent / guardian terminal. The emergency signal can be transmitted to the external terminal 33 in the form of text, image, audio, notification, or the like.

[0073] However, when the device 1 communicates with the external terminal 33, it may use at least one peripheral interface selected from GPIO (General Purpose Input / Output), I2C (Inter-Integrated Circuit), I2S (Inter-IC Sound), and PWM (Pulse Width Modulation) in addition to the universal asynchronous transceiver 13. That is, the device 1 may include a communication unit (not shown) that uses at least one communication protocol selected from GPIO, I2C, I2S, and PWM.

[0074] However, the above-mentioned communication protocol and peripheral interface are merely one embodiment, and the method by which the device 1 communicates with the sensor 2, the external memory 21, the external terminal 33, etc. is not limited thereto.

[0075] In one embodiment, the device 1 is installed in a predetermined space and monitors the biological signals of the human body 600. When an abnormal biological signal is detected from the human body 600, the device 1 can immediately respond to an emergency situation by transmitting an emergency signal to a preset external terminal 33. This can prevent unintentional suffocation accidents among infants and lonely deaths among the elderly.

[0076] FIG. 7 is a diagram for explaining raw data according to an embodiment of the present invention.

[0077] The apparatus according to an embodiment of the present invention may be a neuromorphic device including a neural network, and may be an artificial intelligence model trained based on predetermined training data. That is, the abnormal biological signal detection model may be realized by the neuromorphic device.

[0078] Here, the predetermined training data may be generated based on raw data. The predetermined training data can be generated by processing the raw data according to the following description.

[0079] In one embodiment, the raw data may include normal vital sign data 710 , abnormal vital sign data 720 , and noise data 730 .

[0080] In the following description, the biological signal will be assumed to be a respiratory signal, but this is for convenience of explanation only, and the biological signal may include various human body signals such as a heartbeat signal.

[0081] The normal vital sign data 710 is data that measures only normal respiratory signals. For example, it may be a predetermined amount of data collected by a sensor from a monitored subject who is breathing normally. Similarly, the abnormal vital sign data 720 may be data that measures unstable breathing, such as when a respiratory disorder occurs. For example, it may be a predetermined amount of data collected by a sensor from a monitored subject who is breathing abnormally. Here, the predetermined amount may be varied in various embodiments, such as 3 hours 20 minutes, 10 hours, etc.

[0082] The noise data 730 may be 3 hours and 20 minutes of data measuring signals other than the biological signals of the monitored subject, such as the movements of the monitored subject, the movements of the observer or companion animal, etc. By generating training data for noise as well as normal and abnormal biological signals, higher performance can be achieved when the product is actually applied.

[0083] For example, the amounts of the normal biological signal data 710, the abnormal biological signal data 720, and the noise data 730 may be the same. Therefore, the model can be appropriately trained for normal / abnormal biological signals and noise, and high performance can be achieved. However, the amounts of each data may be different, and the amounts of each data are not limited to those in the above embodiment.

[0084] In one embodiment, each raw data may include in-phase raw data ("I-data" in FIG. 7) and phase-shifted raw data ("Q-data" in FIG. 7). As an example, the phase-shifted raw data may be quadrature data. The in-phase raw data may refer to the real part of the raw data expressed as a complex number, and the phase-shifted raw data may refer to the imaginary part of the raw data expressed as a complex number. The in-phase raw data and the phase-shifted raw data may be used together to represent other dimensions of a complex signal.

[0085] FIG. 8 is a diagram illustrating a method for generating a plurality of sample data according to an embodiment of the present invention.

[0086] The device described in Figures 1 to 7 refers to a neuromorphic device that detects abnormal biological signals, while a device that generates an abnormal biological signal detection model realized by a neuromorphic device (hereinafter referred to as a "model generation device") can refer to a device that generates or processes training data for training the abnormal biological signal detection model.

[0087] In one embodiment, the model generating device can extract a plurality of frames from the raw data to generate a plurality of sample data 800. Here, the model generating device can extract a plurality of frames having a first data length 801 from the raw data.

[0088] As an example, the first data length 801 may be 30 seconds. Assuming that the raw data length 802 is 3 hours and 20 minutes and the raw data is sampled at 20 Hz, the number of data points corresponding to the raw data length 802 may be 240,000, and the number of data points corresponding to the first data length 801 of 30 seconds may be 600.

[0089] In one embodiment, the model generating device may randomly extract a plurality of frames from the raw data. That is, instead of sequentially extracting data points, the model generating device may randomly extract a plurality of frames having a first data length 801. The model generating device may also extract a preset number of frames (e.g., 2000 frames) to generate a preset number of sample data 800.

[0090] On the other hand, when the raw data includes in-phase raw data and shifting-phase raw data, the model generating device may randomly extract frames having a first data length 801 from the in-phase raw data to generate a plurality of in-phase sample data 800. Furthermore, the model generating device may extract frames having a first data length 801 from the shifting-phase raw data to correspond to each signal of the plurality of in-phase sample data 800 to generate a plurality of shifting-phase sample data 810. That is, the in-phase sample data 800 may be extracted randomly, and the shifting-phase sample data 810 may extract a signal at a time point corresponding to the in-phase sample data 800. As an example, the model generating device may extract a frame from the shifting-phase raw data, the time point of which is shifted by the phase difference between the shifting-phase raw data and the in-phase low data, to generate the shifting-phase sample data 810 to correspond to the signal of the in-phase sample data 800.

[0091] 9 and 10 are diagrams illustrating a method for generating a plurality of unit data according to an embodiment of the present invention.

[0092] Referring to FIG. 9, a plurality of unit data 900 to 990 generated from sample data having a first data length 801 is shown.

[0093] In one embodiment, the model generating device may generate a plurality of unit data 900-990 by extracting frames having a second data length 901 from the sample data in chronological order. As an example, the second data length 901 may be 7.5 seconds. If the first data length 801 of the sample data is 30 seconds and the first data length 801 has 600 data points due to a data sampling period of 20 Hz, the second data length 901 may have 150 data points.

[0094] In one embodiment, the model generating device can sequentially extract frames having a second data length 901 such that they overlap by a third data length 902 among the plurality of unit data 900-990. Specifically, the model generating device can extract a first frame 900 including a first data point, and then extract a second frame 910 such that it overlaps with the first frame 900 by the third data length 902. Similarly, the model generating device can extract a third frame (not shown) such that it overlaps with the second frame 910 by the third data length 902.

[0095] As an example, the third data length 902 may be 5 seconds. If the second data length 901 is 7.5 seconds, the model generating device may skip 2.5 seconds between each frame extraction and extract the next frame. Here, if the second data length 901 has 150 data points, the third data length 902 may be 100, and a total of 10 frames are extracted, so 10 pieces of unit data 900-990 may be generated. Furthermore, since the multiple pieces of unit data 900-990 are generated for each piece of sample data, the number of pieces of unit data corresponding to one piece of raw data may be generated by multiplying the number of multiple sample data (e.g., 2000) by the number of unit data generated from each sample data (e.g., 10). While FIG. 9 shows that 10 pieces of unit data 900-990 are generated, the number of the multiple piece data 900-990 may vary depending on the third data length 902.

[0096] In one embodiment, the plurality of sample data may include a plurality of in-phase sample data and a plurality of moving phase sample data. Thus, the model generation device can generate a plurality of in-phase unit data corresponding to the plurality of in-phase sample data and a plurality of moving phase unit data corresponding to the plurality of moving phase sample data. That is, the model generation device can generate a plurality of in-phase unit data for each of the in-phase sample data and a plurality of moving phase unit data for each of the moving phase sample data.

[0097] In one embodiment, the model generating device can convert the generated plurality of unit data into frequency components, for example, by performing a Fast Fourier Transform (FFT) on the generated plurality of unit data.

[0098] Referring to FIG. 10, a plurality of unit data 1010 to 1100 converted into frequency components is shown.

[0099] In one embodiment, the model generator can convert a plurality of in-phase unit data into frequency components and a plurality of moving phase unit data into frequency components.

[0100] 11 to 13 are diagrams for explaining a method for generating a training dataset according to one embodiment of the present invention.

[0101] Referring to FIG. 11, a plurality of unit data 1100 to 1190 are shown.

[0102] In one embodiment, the model generation device can generate a training data set based on a plurality of unit data 1100 to 1190. Specifically, the model generation device can synthesize the plurality of unit data 1100 to 1190 converted into frequency components in the order of extraction.

[0103] This allows one piece of learning data to be generated corresponding to one piece of sample data. The reason why the multiple unit data 1100 to 1190 that have already been generated by overlapping each other are synthesized is that if a signal such as a delta function is included in the signal, it can be confirmed whether it is noise or not depending on whether there is continuity with adjacent unit data and reflected in the learning data.

[0104] Referring to FIG. 12, a plurality of in-phase unit data 1100 to 1190 and a plurality of shifted phase unit data 1101 to 1191 are shown.

[0105] In one embodiment, the model generating device can cross-combine a plurality of in-phase unit data 1100-1190 and a plurality of moving phase unit data 1101-1191. Cross-combine can mean combining in order, such that the first moving phase unit data 1101 is positioned after the first in-phase unit data 1100, the second in-phase unit data 1110 is positioned after the first moving phase unit data 1101, and the second in-phase unit data 1111 is positioned after the second in-phase unit data 1110. Similarly, the third in-phase unit data 1120 and the third moving phase unit data 1121 can be positioned after the second moving phase unit data 1111, and depending on the number of unit data, the last in-phase unit data 1190 and the last moving phase unit data 1191 can be positioned at the end.

[0106] Referring to FIG. 13, the raw data includes normal biological signal data, abnormal biological signal data, and noise data, so the model generating device can generate normal biological signal training data 1310 from the normal biological signal data 710, generate abnormal biological signal training data 1320 from the abnormal biological signal data 720, and generate noise signal training data 1330 from the noise data 730.

[0107] A plurality of sample data and a plurality of unit data may be generated for each of the normal biological signal data, the abnormal biological signal data, and the noise data. For example, the model generating device can generate N (e.g., 2000) in-phase sample data 800 and N (e.g., 2000) shifted-phase sample data 810 from the normal biological signal data 710.

[0108] In addition, multiple unit data (not shown) generated from one sample data 800 are further synthesized into one learning data, so that N (e.g., 2000) pieces of normal biological signal learning data 1310, abnormal biological signal learning data 1320, and noise signal learning data 1330 can also be generated.

[0109] In one embodiment, the training data set generated by the model generator may include normal vital sign training data 1310, abnormal vital sign training data 1320, and noise signal training data 1330.

[0110] Meanwhile, in one embodiment, the model generating device can generate an abnormal biological signal detection model by training an artificial intelligence model with a training dataset.

[0111] FIG. 14 is a graph showing the results of training an artificial intelligence model with a training data set according to one embodiment of the present invention.

[0112] In one embodiment, the model generator can adjust parameters of an artificial intelligence model with a training dataset, where the artificial intelligence model may be a machine learning model.

[0113] Referring to Figure 14, it can be seen that the training loss function and validation loss function of a machine learning model change as the number of times (number of epochs) trained on a training dataset increases, as shown in Figure 14. The training loss function (train_loss) is a value that indicates the loss of the training dataset, and the more accurate the machine learning model makes predictions on the training dataset, the more the training loss function decreases. The validation loss function (val_loss) is a value that indicates the loss of new data (i.e., validation data) that the machine learning model has not seen before, and the more the machine learning model generalizes, the more the validation loss function decreases.

[0114] In one embodiment, the model generating device may select the machine learning model with the lowest loss function and the highest model accuracy among them, and determine it as the abnormal biological signal detection model.

[0115] The abnormal biological signal detection model thus generated is realized in a neural network, and the processor of the neuromorphic device inputs the input data collected by the sensor into the abnormal biological signal detection model and is able to detect abnormal biological signals based on the output of the abnormal biological signal detection model.

[0116] FIG. 15 is a block diagram of an apparatus for detecting abnormal biological signals according to one embodiment of the present invention.

[0117] 15, a device 1500 may include a communication unit 1510, a processor 1520, and a DB 1530. Only components related to the embodiment are shown in the device 1500 in Fig. 15. Therefore, a person skilled in the art would understand that the device 1500 may further include other general-purpose components in addition to the components shown in Fig. 15.

[0118] The communication unit 1510 may include one or more components that enable wired / wireless communication with an external server or device. For example, the communication unit 1510 may include at least one of a short-range communication unit (not shown), a mobile communication unit (not shown), and a broadcast receiving unit (not shown). In one embodiment, the communication unit 1510 can use at least one communication protocol of a serial peripheral interface (SPI) and a universal asynchronous receiver / transmitter (UART). In another embodiment, the communication unit 1510 can communicate with sensors, external memories, and external control devices.

[0119] The DB 1530 is hardware that stores various data processed within the device 1500, and can store programs for processing and controlling the processor 1520.

[0120] DB1530 may include random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), CD-ROM, Blu-ray or other optical disc storage, hard disk drive (HDD), solid state drive (SSD), or flash memory.

[0121] The processor 1520 controls the overall operation of the device 1500. For example, the processor 1520 may generally control an input unit (not shown), a display (not shown), a communication unit 1510, a DB 1530, etc. by executing a program stored in the DB 1530. The processor 1520 may generally control the operation of the device 1500 by executing a program stored in the DB 1530.

[0122] The processor 1520 can control at least part of the operation of the device for detecting abnormal biological signals or the model generating device described above in FIGS.

[0123] The processor 1520 may be implemented using at least one of application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, micro-controllers, microprocessors, or other electrical units for performing functions.

[0124] In one embodiment, device 1500 may be a server. The server may be implemented as a computer device or multiple computer devices that communicate over a network to provide instructions, code, files, content, services, etc. The server may receive data necessary to reduce the neural network model and reduce the neural network model based on the received data.

[0125] Meanwhile, embodiments of the present invention may be realized in the form of a computer program executable by various components on a computer, and such a computer program may be recorded on a computer-readable medium, including magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program instructions, such as ROMs, RAMs, and flash memories.

[0126] On the other hand, the computer program may be one specially designed and constructed for the present invention, or it may be one that is well known and available to those skilled in the art of computer software. Examples of computer programs include not only machine language code such as that produced by a compiler, but also high-level language code that is executed by a computer using an interpreter, etc.

[0127] According to one embodiment, methods according to various embodiments of the present disclosure can be provided in a computer program product. The computer program product can be traded as a commodity between a seller and a buyer. The computer program product can be distributed in the form of a machine-readable storage medium (e.g., a compact disc read only memory (CD-ROM)) or can be distributed through an application store (e.g., the Play Store). TM ) or directly between two user devices. In the case of online distribution, at least a portion of the computer program product may be at least temporarily stored on or temporarily generated by a machine-readable storage medium, such as the memory of a manufacturer's server, an application store server, or an intermediary server.

[0128] Unless explicitly stated or stated to the contrary, steps constituting a method according to the present invention may be performed in any suitable order. The present invention is not necessarily limited to the order of steps described above. The use of all examples or exemplary terms in the present invention is merely for the purpose of explaining the present invention in detail, and the scope of the present invention is not limited by the examples or exemplary terms unless limited by the claims. Furthermore, those skilled in the art will understand that various modifications, combinations, and variations can be made within the scope of the claims or their equivalents, depending on design conditions and factors.

[0129] Therefore, the concept of the present invention should not be limited to the above-described embodiments, and not only the scope of the appended claims, but also all scopes equivalent to or modified equivalently from the scope of the claims, can be said to fall within the scope of the concept of the present invention.

Claims

1. generating a plurality of sample data by randomly extracting frames having a first data length from raw data including biological signal data; extracting frames having a second data length from the sample data in time order to generate a plurality of unit data; generating a learning data set based on the plurality of unit data; and generating an abnormal biological signal detection model by training an artificial intelligence model with the training data set; The step of generating a plurality of unit data includes: A method for detecting an abnormal biological signal, comprising the step of sequentially extracting frames having the second data length so that the plurality of unit data overlap by a third data length.

2. the plurality of sample data and the plurality of unit data, generated for each of the normal biological signal data, the abnormal biological signal data, and the noise data included in the raw data, The training data set is The method for detecting an abnormal biological signal according to claim 1 , comprising normal biological signal training data generated from the normal biological signal data, abnormal biological signal training data generated from the abnormal biological signal data, and noise signal training data generated from the noise data.

3. The step of generating a plurality of unit data includes: The method for detecting an abnormal biological signal according to claim 1 , further comprising the step of converting the generated plurality of unit data into frequency components.

4. The step of generating the training data set includes: The method for detecting an abnormal biological signal according to claim 3, further comprising the step of combining the converted plurality of unit data in the order of extraction.

5. The raw data is Including in-phase raw data and phase-shifted raw data, The step of generating a plurality of sample data includes: randomly extracting frames having the first data length from the in-phase raw data to generate a plurality of in-phase sample data; 5. The method for detecting an abnormal biological signal according to claim 4, further comprising a step of extracting a frame having the first data length from the moving phase raw data so as to correspond to each signal of the plurality of in-phase sample data, thereby generating a plurality of moving phase sample data.

6. The step of generating the training data set includes: The method for detecting an abnormal biological signal according to claim 5, further comprising a step of crossing and synthesizing a plurality of in-phase unit data corresponding to the plurality of in-phase sample data and a plurality of moving phase unit data corresponding to the plurality of moving phase sample data.

7. The step of generating the abnormal biological signal detection model includes: adjusting parameters of the artificial intelligence model with the training data set; The method for detecting an abnormal biological signal according to claim 1 , further comprising a step of generating the abnormal biological signal detection model based on a loss function according to the number of adjustments (epochs) and the accuracy of the artificial intelligence model.

8. The abnormal biological signal detection model is This is realized by neuromorphic devices, The neuromorphic device comprises: at least one processor that drives the abnormal vital sign detection model; a crossbar array circuit that receives instructions from the at least one processor and performs in-memory operations; The at least one processor Inputting input data collected by the sensor into the abnormal biological signal detection model; The method for detecting an abnormal physiological signal according to claim 1 , wherein the abnormal physiological signal is detected based on an output of the abnormal physiological signal detection model.

9. a memory having at least one program stored therein; a processor that performs calculations by executing the at least one program; The processor: randomly extracting frames having a first data length from raw data including biological signal data to generate a plurality of sample data; extracting frames having a second data length from the sample data in chronological order to generate a plurality of unit data; generating a learning dataset based on the plurality of unit data; generating an abnormal biological signal detection model by training an artificial intelligence model with the training data set; The processor: an apparatus for detecting an abnormal biological signal, which generates a plurality of unit data by sequentially extracting frames having the second data length so that the plurality of unit data overlap by a third data length;

10. A computer-readable recording medium having recorded thereon a program for executing the method of claim 1 on a computer.

Images