Bedridden person detection system and method based on long short-term memory neural network

Abstract

The present invention relates to a bedridden person detection system and method based on long short-term memory neural network, the system comprises: a sensor module, a data acquisition module, a data preprocessing module, a feature extraction module, and an output module; the sensor module, the data acquisition module, the data preprocessing module, the feature extraction module, and the output module are connected in sequence. The system of the present invention is placed on a mattress, and is used to continuously, quickly, and accurately detect the bed posture of the elderly, the disabled, and the long-term bedridden patients; and by detecting the posture of the bedridden person, the bedridden patients are efficiently cared for, and the burden of the nursing staff is reduced.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to a bedridden person detection system and method based on a long short-term memory neural network. Background Art

[0002] At present, most developed countries are experiencing a crisis of population aging. China has become the country with the largest elderly population in the world and one of the countries with the fastest aging population. How to effectively fill the gap in the nursing industry has become a top priority. In hospital nursing work, bedridden patients are often accompanied by various complications, such as falling out of bed, pressure bedsores, urinary incontinence and other problems. Therefore, how to efficiently care for bedridden patients and reduce the burden on nursing staff is a new direction to alleviate aging.

[0003] Existing bedridden patient monitoring systems can be divided into posture-recognizing pressure sensors, image recognition, and wearable devices. The pressure sensor method requires the collection of a large amount of information on pressure points, so the equipment maintenance and construction costs are high. Image recognition involves the privacy of patients, and the posture cannot be detected well after the patient is covered with a quilt. Wearable devices are not very comfortable for bedridden patients, and the devices need to be charged regularly, which will cause certain troubles in nursing care.

[0004] Therefore, there is an urgent need for a bedridden person detection system and method based on long short-term memory neural network, which has become a hot topic of concern for researchers today. Summary of the invention

[0005] In order to solve the problems in the above-mentioned background technology, the present invention proposes a bedridden person detection system and method based on a long short-term memory neural network.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] The bedridden person detection system based on long short-term memory neural network includes:

[0008] Sensor module: collects the posture information of the bedridden person based on the information received by the sensor and transmits it to the data acquisition module;

[0009] Data acquisition module: used to collect the posture information of the bedridden person and obtain initial data;

[0010] Data preprocessing module: performs preprocessing operation on the initial data to eliminate abnormal data and obtain preprocessed data, wherein the abnormal data is a sensor value that is higher or lower than a small surrounding range;

[0011] Feature extraction module: extracting features from the preprocessed data to obtain posture information of the bedridden person;

[0012] Output module: connected to the display screen, used to represent the posture of the bedridden person in the form of an image;

[0013] The sensor module, data acquisition module, data preprocessing module, feature extraction module and output module are connected in sequence.

[0014] Preferably, the sensor module includes a plurality of RFID sensors, and the RFID sensors are placed on the surface of the mattress. When the bedridden person lies on the mattress in different postures, the RFID sensors collect posture information of the bedridden person.

[0015] Preferably, the data acquisition module comprises an RFID tag, which automatically identifies the target object and acquires data through the radio frequency signal emitted by the RFID sensor, wherein the RFID tag is a passive RFID tag.

[0016] Preferably, the feature extraction module extracts features from the preprocessed data through a long short-term memory network to obtain posture information of the bedridden person.

[0017] Preferably, the input and output of the memory cells in the long short-term memory neural network are controlled by a forget gate, an update gate and an output gate, so as to learn the representation of time from the extracted features and classify different activity postures.

[0018] Preferably,

[0019] The forget gate is used to remove the previous state of the long short-term memory neural network;

[0020] The update gate is used to control the storage and update of input data;

[0021] The output gate is responsible for generating output results;

[0022] Among them, the forget gate, update gate and output gate are set in sequence.

[0023] Preferably, the process of training the long short-term memory neural network includes:

[0024] Updating the weights in the neural network through a stochastic gradient descent optimizer to minimize the training loss function;

[0025] The data set input into the neural network is divided into a training set data and a test set data set. The neural network is trained and tested according to an iterative window method, and the obtained standardized values ​​are placed in a batch loader to achieve accurate prediction.

[0026] A detection method for a bedridden person detection system based on a long short-term memory neural network, comprising:

[0027] S1. Install RFID tags on the surface of the mattress, arrange RFID reader antennas, build sensor modules, and collect posture information of bedridden persons; wherein each RFID tag is placed symmetrically;

[0028] S2, collecting the posture information of the bedridden person based on the data acquisition module, and obtaining a corresponding data set according to the signal strength indicator RSSI received by the RFID tag;

[0029] S3, removing the obtained data through the data preprocessing module, and clearing the damaged data caused by hardware failure during the data transmission process, to obtain the preprocessed data;

[0030] S4, transmitting the preprocessed data to a feature extraction module, performing feature extraction through a long short-term memory neural network, and obtaining posture information of the bedridden person;

[0031] S5. Sending the posture information of the bedridden person to a display screen through an output module for displaying a posture image of the bedridden person.

[0032] Preferably, in S3, the obtained data is processed based on the three characteristics of EPC value, timestamp TS and signal strength indicator RSSI.

[0033] The beneficial effects of the present invention are:

[0034] The system of the present invention uses radio frequency identification differential array (RFID) tags to receive signal strength index (RSSI) information to collect detailed activity data of bedridden people, and uses long short-term memory network method (LSTM) to extract RSSI features from these activity data to obtain the specific posture of the bedridden person at a certain moment. The system is placed on the mattress and is used to continuously, quickly and accurately detect the posture of the elderly, disabled and long-term bedridden patients in bed; it also detects the posture of bedridden people, efficiently cares for bedridden patients, and reduces the burden on caregivers. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A schematic diagram of module connection relationships in an embodiment of the present invention;

[0037] Figure 2 Schematic diagram of the folded LSTM architecture in an embodiment of the present invention;

[0038] Figure 3 Schematic diagram of the LSTM architecture deployed in an embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the internal architecture of an LSTM unit in an embodiment of the present invention;

[0040] Figure 5 Flow chart of the method in the embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] See attached Figure 1 The present invention provides a bedridden person detection system based on long short-term memory neural network, comprising:

[0044] Sensor module: collects the posture information of the bedridden person based on the information received by the sensor and transmits it to the data acquisition module;

[0045] Data acquisition module: used to collect the posture information of the bedridden person and obtain initial data;

[0046] Data preprocessing module: performs preprocessing operation on the initial data to eliminate abnormal data and obtain preprocessed data;

[0047] Feature extraction module: extracting features from the preprocessed data to obtain posture information of the bedridden person;

[0048] Output module: connected to the display screen, used to represent the posture of the bedridden person in the form of an image;

[0049] The sensor module, data acquisition module, data preprocessing module, feature extraction module and output module are connected in sequence.

[0050] The sensor module and data acquisition module apply radio frequency identification (RFID) technology. The RFID sensors are arranged on the surface of the mattress. When the bedridden person lies on the mattress in different postures, the system collects the sensor signals and obtains initial data.

[0051] Among them, radio frequency identification technology RFID is a commonly used sensing technology in the sensing and interactive layer of the Internet of Things. It is a non-contact automatic recognition technology that automatically identifies the target object through radio frequency signals to obtain data. RFID tags can be divided into three categories according to their working mode: active, semi-active and passive. Among them, passive RFID tags are more widely used in actual production and life because they do not require built-in batteries, have low costs and long service life. Therefore, this embodiment uses passive RFID tags.

[0052] The data preprocessing module and feature extraction module apply the long short-term memory neural network method. First, the collected data is preprocessed to eliminate abnormal data, and then the features are extracted from the processed data to obtain the posture of the bedridden person at that time.

[0053] In the memory cells used by the Long Short-Term Memory artificial neural network LSTM, the input and output are controlled by gates, so LSTM cells can effectively learn temporal representations from statistical features for activity classification. The recurrent connections in LSTM add states or memories to the network, which allows the RNN to learn and exploit the ordered nature of observations in the input sequence to enhance learning performance.

[0054] The network model used in this embodiment is as shown in the attached Figure 2 As shown, the expanded LSTM architecture is shown in the attached Figure 3 As shown in the figure, the operation and calculation activities in each LSTM unit are shown in the attached figure. Figure 4 shown.

[0055] See attached Figure 4 , the LSTM unit consists of a self-connected memory cell c(t) and three different gates:

[0056] (1) Forget gate: f(t), which controls the removal of the previous state;

[0057] (2) Update gate: i(t), which controls the storage and update of input data;

[0058] (3) Output gate: o(t), responsible for generating output results.

[0059] For any given timestamp t, for the generated input data x(t) and output data h(t), the parameters that need to be learned in LSTM are the weights and biases of the sampled data.

[0060] Inside the LSTM cell, there are three main stages:

[0061] The first is the forget stage, which uses f(t) as the forget gate to control which parts of the previous state need to be kept and which need to be forgotten.

[0062] Then comes the selection phase, or the update phase, which uses i(t) as a gating to record the important ones and record less of the unimportant ones.

[0063] The last stage is the output stage, which uses o(t) as the gate. This stage will determine which will be regarded as the output of the current state.

[0064] exist Figure 2 and Figure 4 In the example, the output of LSTM is fed back to two fully connected layers: FCN1 and FCN2. After the data is processed by these two fully connected layers, the output is sent to the SoftMax(Y) layer.

[0065] In the SoftMax(Y) layer, the weights in the neural network are updated by a stochastic gradient descent optimizer to minimize the training loss function;

[0066] According to the iterative window method, the data set input into the neural network is divided into a training data set and a test data set. In this embodiment, the first seven iterative windows of all sampling activities are selected to train the model, and the last three iterative windows are used as test data. At the same time, the standardized value is built into the batch loader to achieve accurate prediction.

[0067] Through the above process, the LSTM model can be trained to improve its efficiency and accuracy in extracting data features.

[0068] The trained data is transmitted to the display screen through the output module and represents the posture of the bedridden person in the form of an image.

[0069] A detection method for a bedridden person detection system based on a long short-term memory neural network, as shown in the attached Figure 5 As shown, including:

[0070] S1. Install RFID tags on the surface of the smart mattress, arrange RFID reader antennas, build sensor modules, and collect posture information of bedridden people; wherein each RFID tag is placed symmetrically;

[0071] S2, collecting the posture information of the bedridden person based on the data acquisition module, and obtaining a corresponding data set according to the RSSI reading of the indicator of the RFID tag;

[0072] S3, removing the obtained data through the data preprocessing module, and clearing the damaged data caused by hardware failure during the data transmission process, to obtain the preprocessed data;

[0073] S4, transmitting the preprocessed data to a feature extraction module, performing feature extraction through a long short-term memory neural network, and obtaining posture information of the bedridden person;

[0074] Because each feature changes with each sampling activity, the three features of electronic product code (EPC value), time stamp TS, and received signal strength indication (RSSI) are selected here to process the data together.

[0075] Perform the following steps:

[0076] First, the Electronic Product Code (EPC value) is mapped to a unique integer ID, with the integer range being 1-228 (corresponding to the total number of RFID tags used in the actual mattress). Each tag ID has a unique integer ID, and other tag IDs are mapped in a similar manner;

[0077] Next, the timestamp TS is uniquely mapped to an integer ID. At the same time, the integer IDs given to TS are in ascending order, that is, for each active data set, the ID of the lowest TS value is 0, and the ID of the next TS is 1, 2, 3, ...;

[0078] Finally, the RSSI values ​​of all IDs under each TS are collected to form RSSI vectors. The length of each vector is 228 (same as above, corresponding to the total number of tags).

[0079] Each active dataset represents a RSSI vector x (t) Sequence, denoted by X K , where K is the total time step, so:

[0080] X K ={x (0) ,x (1) ,x (2) ,x (3) ,…x (K)} (13)

[0081] Denote the RSSI vector x by Q (t) The size of each X K Filled with 0, it is represented as:

[0082]

[0083] Where, X T Represents a single activity training dataset.

[0084] The preprocessed data is input into the above-mentioned trained LSTM model, and the model will automatically extract features and generate the posture of the bedridden person at that time.

[0085] S5. Sending the posture information of the bedridden person to a display screen through an output module for displaying a posture image of the bedridden person.

[0086] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A bedridden person detection system based on long short-term memory neural network, characterized in that: include: Sensor module: collects the posture information of the bedridden person based on the information received by the sensor and transmits it to the data acquisition module; Data acquisition module: used to collect the posture information of the bedridden person and obtain initial data; Data preprocessing module: performs preprocessing operation on the initial data to eliminate abnormal data and obtain preprocessed data, wherein the abnormal data is a sensor value that is higher or lower than a small surrounding range; The data preprocessing module processes the obtained data based on the three features of the electronic product code EPC value, the timestamp TS and the signal strength indicator RSSI, including: mapping the electronic product code EPC value to a unique integer ID, uniquely mapping the timestamp TS to the integer ID, and the integer IDs given to the timestamp TS are in ascending order, collecting the signal strength indicator RSSI values ​​of all IDs under each timestamp TS to form a signal strength indicator RSSI vector; Feature extraction module: extracting features from the preprocessed data to obtain posture information of the bedridden person; The feature extraction module extracts features from the preprocessed data through a long short-term memory network to obtain posture information of the bedridden person: controls the input and output of memory cells in the long short-term memory neural network through a forget gate, an update gate, and an output gate, so as to learn the representation of time from the extracted features and classify different activity postures; Output module: connected to the display screen, used to represent the posture of the bedridden person in the form of an image; The sensor module, data acquisition module, data preprocessing module, feature extraction module, and output module are connected in sequence; The sensor module includes a plurality of RFID sensors, which are placed on the surface of the mattress. When the bedridden person lies on the mattress in different postures, the RFID sensors collect posture information of the bedridden person; The data acquisition module includes an RFID tag, which automatically identifies the target object and acquires data through the radio frequency signal emitted by the RFID sensor, wherein the RFID tag is a passive RFID tag, and each RFID tag is symmetrically placed; The process of training the long short-term memory neural network includes: Updating the weights in the neural network through a stochastic gradient descent optimizer to minimize the training loss function; The data set input into the neural network is divided into a training set data and a test set data set. The neural network is trained and tested according to an iterative window method, and the obtained standardized values ​​are placed in a batch loader to achieve accurate prediction.

2. The bedridden person detection system based on long short-term memory neural network according to claim 1 is characterized in that: The forget gate is used to remove the previous state of the long short-term memory neural network; The update gate is used to control the storage and update of input data; The output gate is responsible for generating output results; Among them, the forget gate, update gate and output gate are set in sequence.

3. A detection method for a bedridden person detection system based on a long short-term memory neural network according to any one of claims 1 to 2, characterized in that: include: S1. Install RFID tags on the surface of the mattress, arrange RFID reader antennas, build sensor modules, and collect posture information of the bedridden person; wherein each RFID tag is placed symmetrically; S2, collecting the posture information of the bedridden person based on a data acquisition module, and obtaining a corresponding data set according to the signal strength indicator RSSI received by the RFID tag, wherein the RFID tag is a passive RFID tag; S3, removing the obtained data through the data preprocessing module, and clearing the damaged data caused by hardware failure during the data transmission process, to obtain the preprocessed data; In the S3, the obtained data is processed based on the three features of the electronic product code EPC value, the timestamp TS and the signal strength indicator RSSI, including: mapping the electronic product code EPC value to a unique integer ID, uniquely mapping the timestamp TS to the integer ID, and the integer IDs given to the timestamp TS are in ascending order, collecting the RSSI values ​​of all IDs under each timestamp TS to form a signal strength indicator RSSI vector; S4, transmitting the preprocessed data to a feature extraction module, performing feature extraction through a long short-term memory neural network, and obtaining posture information of the bedridden person; Controlling the input and output of the memory cells in the long short-term memory neural network through a forget gate, an update gate, and an output gate, so as to learn the representation of time from the extracted features and classify different activity postures; The process of training the long short-term memory neural network includes: Updating the weights in the neural network through a stochastic gradient descent optimizer to minimize the training loss function; The data set input into the neural network is divided into a training set data and a test set data set, the neural network is trained and tested according to an iterative window method, and the obtained standardized values ​​are placed in a batch loader to achieve accurate prediction; S5. Sending the posture information of the bedridden person to a display screen through an output module for displaying a posture image of the bedridden person.

Images