Wall-penetrating fall detection and early warning system and method based on Lora signal

Abstract

The application discloses a through-wall fall detection and early warning system and method based on a Lora signal, wherein a transmitter and a receiver are placed on the same side under a partition in a room, the receiver is installed with two antennas, can simultaneously receive the signals reflected and diffracted by multiple paths in the environment by the transmitter, and utilizes signal division operation to eliminate the influence of phase offset, so that the transmission signal and the receiving signal clock are synchronized; in order to cope with the problem of weak signal caused by the attenuation of Lora signal transmission through the partition, a time domain aliasing method is used to amplify the signal, the time-frequency spectrum of the signal is extracted, the signal entropy information, the power mutation curve and the signal-to-noise ratio are calculated as the signal characteristics. The data is labeled based on a fuzzy labeling method to train a deep model, and the fall behavior of a person in the room is detected through the model, and if the fall behavior occurs, an alarm signal is sent. The application can distinguish the normal activity of a person from the fall behavior, has a wide sensing range and high sensing accuracy.

Description

Technical Field

[0001] This invention relates to an early warning method, specifically a wall-penetrating fall detection and early warning system and method based on LoRa signals, belonging to the field of wireless sensing vital sign detection technology. Background Technology

[0002] Most older adults are unable to get up on their own after a fall, and research shows that medical outcomes from falls largely depend on reaction time and rescue duration. In some clinical situations, delayed treatment after a fall increases the risk of death; half of patients who experienced prolonged periods lying on the floor (>1 hour) died within 6 months of the incident. For older adults living independently, approximately 50% of falls occur in their own homes. Therefore, timely and automatic fall detection has always been a research goal in assisting older adults with their lives.

[0003] Currently, the existing fall detection technologies mainly include the following:

[0004] 1) Based on wearable sensors, these sensing systems can only work when the user wears the sensor, which results in low accuracy for elderly people with limited mobility.

[0005] 2) Detection devices based on computer vision have privacy intrusion issues. At the same time, the inherently intensive computing cannot be processed in real time, resulting in delayed response of the detection device, which may miss the best response and rescue time.

[0006] 3) Based on WIFI sensing, although this technology is easy to deploy, its range is too small, with a limit of only six meters, which can only monitor the situation of one room.

[0007] Therefore, there is an urgent need for an early warning system and method for elderly falls that is accurate, sensitive, has a wide detection range, and is easy to deploy. Summary of the Invention

[0008] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a wall-penetrating fall detection and early warning system based on LoRa signals, which can detect the activities of elderly people at home, thereby achieving timely early warning of falls. Furthermore, the elderly being detected do not need to wear any devices, and the system boasts a wide sensing range and high sensing accuracy.

[0009] This invention also proposes a method for detecting and warning of falls through walls based on Lora signals.

[0010] According to a first aspect of the present invention, a wall-penetrating fall detection and early warning system based on LoRa signals includes a transmitter, a USRP device, a receiver, and a processor. The transmitter is a LoRa node used to transmit LoRa signals via a transmitting antenna. The receiver is equipped with two receiving antennas for receiving the LoRa signals transmitted by the transmitter. The two receiving antennas are closely connected together, and there is a predetermined distance between the receiving antennas and the transmitting antennas. The receiver is connected to the USRP device. The transmitter and the receiver are placed on the same side of a partition wall environment within the detection space. The USRP device is used to receive the LoRa signals transmitted by the receiver. The processor is used to receive the LoRa signals transmitted by the USRP device and process the received LoRa signals using a fall detection and early warning method pre-stored within it. When a fall signal of an elderly person is detected, an alarm signal is issued.

[0011] A method for detecting and warning of falls through walls based on LoRa signals according to a second aspect of the present invention includes the following steps:

[0012] S1 enhances the wall signal: high-frequency noise in the Lora signal is filtered out to obtain a signal with lower interference. The signal is then subjected to time-domain aliasing to amplify it and increase its amplitude. Finally, the signal entropy method is used to eliminate signal clock asynchrony.

[0013] S2 Lora Signal Feature Extraction: Calculate the power jump curve of the Lora signal after processing in step S1, distinguish between static environmental signals and active signals, select active signal intervals in the active signals, perform video analysis on each active signal interval, calculate the time-frequency graph, and extract Lora signal features from the time-frequency graph: power jump curve, information entropy curve, and signal-to-noise ratio.

[0014] S3 Elderly Fall Detection: Construct and train the temporal detection model FallNet. Input the LoRa signal features into the trained temporal detection model FallNet. The model processes the LoRa signal features and determines whether a fall has occurred within the detection space covered by the LoRa signal. If a fall has occurred, an early warning signal is issued.

[0015] Compared with the prior art, the present invention has the following advantages:

[0016] 1) This invention uses LoRa signals to detect human activity. LoRa signals have strong anti-interference capabilities, and through wireless signal processing technologies, they can detect falls and provide early warnings of potential falls. This invention has great potential for large-scale deployment in daily life and work scenarios, and is especially suitable for elderly people living alone.

[0017] 2) Since the present invention uses the signal downsampling superposition method to amplify the changes in the sound wave signal, it effectively solves the technical problem of poor sensing accuracy caused by the attenuation of the Lora signal after multiple reflections and diffractions. In addition, the Lora signal used in the present invention has the characteristic of penetrating walls, and it only needs to be deployed in one room.

[0018] 3) This invention uses the time-series detection model FallNet, which is trained and fitted with parameters using data collected from different rooms. Compared with traditional algorithms, it has a stronger scene adaptability. The model is trained and detected based on multiple mixed features. By learning the relationship between different features, it can still maintain high accuracy for extreme samples. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0020] Figure 2 This is the original Lora signal of one embodiment of the present invention;

[0021] Figure 3 This is an amplified LoRa signal according to an embodiment of the present invention.

[0022] Figure 4 This is a fuzzy annotation method according to an embodiment of the present invention;

[0023] Figure 5 This is a fuzzy annotation method according to an embodiment of the present invention;

[0024] Figure 6 This is a FallNet fall detection depth model according to an embodiment of the present invention. Detailed Implementation

[0025] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0026] The following disclosure provides numerous different embodiments or examples for implementing various structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. Additionally, examples of various specific processes and materials are provided in this invention; however, those skilled in the art will recognize the applicability of other processes and / or the use of other materials.

[0027] This embodiment uses the detection of falls in the elderly as an example. Of course, the fall detection and early warning system and method of the present invention are applicable to all people with unsteady gait.

[0028] refer to Figure 1 As shown, a wall-penetrating fall detection and early warning system based on LoRa signals according to an embodiment of the present invention includes a transmitter, a receiver, a USRP device, and a processor. The USRP device is a general-purpose software-defined radio peripheral that allows a regular computer to function like a high-bandwidth software-defined radio device. In this embodiment, the processor is a display with data processing capabilities, and the alarm signal can be displayed on the display. The transmitter is a LoRa node used to transmit LoRa signals through a single transmitting antenna. The receiver is equipped with two receiving antennas for receiving the LoRa signals transmitted by the transmitter, wherein the two receiving antennas are closely connected. The receiving antenna and the transmitting antenna are connected together with a 20cm gap between them. The receiver is connected to the USRP device. The transmitter and the receiver are placed on the same side of the partition wall environment in the detection space. The USRP device is used to receive the Lora signal transmitted by the receiver. The processor is used to receive the Lora signal transmitted by the USRP device and process the received Lora signal through its internally pre-stored fall detection and early warning method. When a fall signal of the elderly is detected, an alarm signal is issued. After receiving the alarm signal, the family members can go to the scene in time to check the elderly's condition, so as to avoid the elderly falling for a long time without being discovered.

[0029] refer to Figure 1 As shown, the fall detection and early warning method based on the above-mentioned early warning system includes the following steps:

[0030] S1 Enhancement of the Partition Wall Signal: A low-pass filter is used to filter out high-frequency noise in the Lora signal to obtain a signal with lower interference. The signal is then subjected to time-domain aliasing to amplify it and increase its amplitude. Finally, the signal entropy method is used to eliminate signal clock asynchrony.

[0031] The temporal aliasing processing method includes the following steps:

[0032] S1.1 Divide the original time series into several sub-sampling sequences at intervals of ΔT = N × Δt seconds, where N = 3 is the number of aliasing windows and Δt = 0.01s is the size of the aliasing window.

[0033] S1.2 Extract the time sequence corresponding to each aliasing window at each corresponding position in each sub-sampling sequence to form a time sequence Q. The number of time sequences Q generated is equal to the number of aliasing windows.

[0034] S1.3 Superimpose all the time sequences Q generated in step S1.2 to form a new time sequence, i.e., the amplified signal.

[0035] Specifically, for example, each sub-sampling sequence includes aliasing window 1, aliasing window 2, and aliasing window 3. The time sequence corresponding to aliasing window 1 in each sampling sequence is extracted to form a time sequence Q1. Similarly, the time sequences corresponding to aliasing windows 2 and 3 in each sub-sampling sequence are extracted to form time sequences Q2 and Q3, respectively. The time sequences Q1, Q2, and Q3 are superimposed to form a new time sequence, i.e., the amplified signal.

[0036] Although the new time sequence reduces the sampling rate of the original LoRa signal, the high frequency of the LoRa signal means that downsampling will not cause signal distortion. In this case, the signal can still reflect human activity signals. Therefore, this method of LoRa signal amplification will not destroy the human activity information contained in the signal. (Reference) Figure 2 and Figure 3 The image shows a segment of the LoRa signal, illustrating the amplification effect. Figure 2 This is the original LoRa signal. Figure 3 For the amplified LoRa signal, from Figure 2 and Figure 3 It can be seen that the signal amplitude has increased significantly.

[0037] In this embodiment, the transmitter emits a Lora signal with a center frequency of 915MHz and a bandwidth of 125kHz. The USRP device has a sampling rate of 500kHz, and each chirp period is 10ms. A chirp refers to one cycle of frequency change of the Lora signal over time, while the bandwidth is the width of the frequency change interval of the Lora signal within one chirp. To ensure that the two receiving antennas receive the Lora signal synchronously, the signal quotient method is used to achieve clock synchronization of the received signal. The specific method is as follows: the ratio of the Lora signals from the two receiving antennas is combined to form a new clock-synchronized Lora signal, eliminating clock asynchrony. The formula is as follows:

[0038]

[0039] Where R1 and R2 represent the Lora signals received by the two receiving antennas of the USRP device, respectively, and SR represents the reconstructed signal after eliminating clock asynchrony.

[0040] S2 Lora Signal Feature Extraction: Calculate the power jump curve of the Lora signal processed in step S1, distinguish between static environmental signals and active signals, select active signal intervals within the active signals, perform video analysis on each active signal interval, calculate the time-frequency graph, and extract Lora signal features from the time-frequency graph: power jump curve, information entropy curve, and signal-to-noise ratio. The specific steps are as follows:

[0041] S2.1 Calculate the power mutation curve of the Lora signal after processing in step S1, distinguish between static environmental signals and activity signals. In the activity signals, each 10-second interval is considered an activity signal interval. The activity signals refer to the activity signals of the elderly. Specifically, this includes the following steps:

[0042] S2.1.1 During signal propagation, objects in the propagation process will cause changes in signal power. Therefore, a power change curve is used to represent this change process. The power change curve of the Lora signal after processing in step S1 is calculated. When the value of the power change curve is less than 2, it is a static environmental signal, indicating that there is no active target in the detection space (room). When the value of the power change curve is greater than or equal to 2, it is an active signal, indicating that there is an active target in the detection space (room).

[0043] S2.1.2 In the active signal, every 10 seconds of active signal constitutes an active signal interval.

[0044] S2.2 performs video analysis on each active signal interval, calculates the time-frequency plot, and extracts LoRa signal features from the time-frequency plot: power jump curve, information entropy curve, and signal-to-noise ratio. This includes the following steps:

[0045] S2.2.1 In the active signal, each 10-second interval is considered as a window to obtain a 10-second active signal interval. The time-frequency graph is calculated for each 10-second interval, with an interval of 0.5 seconds between calculations. The time-frequency graph is obtained through short-time fast Fourier transform, and the calculation formula is as follows:

[0046]

[0047] Where S(t,f) represents a two-dimensional matrix that represents the relationship between signal time and frequency, h(τ-t) represents a window function, and SR represents the reconstructed signal after eliminating clock asynchrony. In this embodiment, a Hamming window is used.

[0048] S2.2.2 Using the results of short-time fast Fourier transform in the obtained time-frequency graph, the power jump curve of the Lora signal is obtained through formula (3):

[0049]

[0050] Where PBC represents computation over an interval of S(t,f); f u =50Hz, indicating that the maximum value in the frequency range of S(t,f) is selected, where f l Represents the minimum value of the interval;

[0051] S2.2.3 The information entropy curve is obtained through formula (4):

[0052]

[0053] Where p(f,t)=|S(t,f)| 2 When time t is displayed on the time-frequency graph, f represents the instantaneous frequency.

[0054] S2.2.4 The signal-to-noise ratio is obtained through formula (5):

[0055]

[0056] Among them, A signal Original signal amplitude, A noise Noise signal amplitude.

[0057] S3 Elderly Fall Detection: Construct and train the temporal detection model FallNet, and input the Lora signal features extracted in step S2 into the trained temporal detection model FallNet. The model processes the Lora signal features and determines whether a fall has occurred within the detection space covered by the Lora signal. If a fall has occurred, an early warning signal is issued. The specific steps include:

[0058] S3.1 Constructing the Temporal Detection Model FallNet: The entire temporal detection model FallNet is divided into two parts: upsampling and downsampling. The input signal features go through four downsampling stages and four upsampling stages. Downsampling is performed through one-dimensional convolution and stride, with the length of the convolution kernel set to 2 data points and the stride set to 2 data points. Upsampling is performed through inverse integration to restore the input length of the previous stage. In each deconvolution stage, the feature layers output from each downsampling stage are directly concatenated to the corresponding upsampling stage in the upsampling process, and the softmax function is used to calculate the probability of the fall activity.

[0059] For details, please refer to Figure 6The diagram shows the structure of the FallNet temporal detection model. The input consists of three time sequences of length 1024: a power jump curve, an information entropy curve, and a signal-to-noise ratio. The output is a time sequence of the same length, where each value represents a fall probability. The gray rectangles represent layers within the neural network. Figure 6 The numbers in the text represent the dimensions of each layer, following the format "number of channels per channel × length". The arrows indicate operations applied between layers, and their meanings are as follows: Figure 6 The bottom right corner indicates that the input signal features undergo four downsampling stages and four upsampling stages. Downsampling is performed through one-dimensional convolution and stride. The convolution kernel length is set to 2 data points, and the stride step is set to 2 data points. Upsampling is performed using inverse involution to restore the input length of the previous stage. Each deconvolution stage directly connects the downsampled output features to the right layer without going through deeper layers, improving convergence during training. The gray rectangles with dashed boundaries represent features directly copied from skipped connections, and the probability of fall activity is calculated using the softmax function.

[0060] S3.2 Training the Temporal Detection Model FallNet:

[0061] S3.2.1 Sample Acquisition: Lora signals are acquired within the detection space. The Lora signals include four types of signals: falling, standing still, walking, and standing. Each signal sample contains a 10-second Lora signal sample. The signal sample includes three Lora signal features: information entropy curve, power change curve, and signal-to-noise ratio, each with a time sequence length of 1024. The Lora signal samples are labeled using a fuzzy labeling method.

[0062] When labeling samples, the reaction time of individuals during sample collection can cause labeling errors. Experiments have shown that the error in determining the fall time does not exceed 50 milliseconds. Therefore, this invention uses a fuzzy labeling method to label the samples. The specific method is as follows: A 100-millisecond window is used. When labeling fall probability data, the moment when the elderly person falls is taken as the center. Each point within the window represents the probability of whether a fall has occurred. A Gaussian distribution curve is fitted around the center of the window. The mean of this curve is 0, and the standard deviation is 10. Finally, the values ​​of all points are multiplied by 25, so that the fall probability values ​​within the window decrease from 1 to 0 from the center outwards. (Reference) Figure 4 As shown; for normal activities that do not involve falling, the probability value on the entire label is marked as 0, for reference. Figure 5 As shown.

[0063] S3.2.2 Training Model: The cross-entropy function is used as the loss function, and the stochastic gradient descent method is used to optimize the training of the time-series detection model FallNet. 300 LoRa signal samples are used to train the time-series detection model FallNet. The number of training batches is set to 200, and 4 LoRa signal samples are input for each batch. The learning rate is set to 0.01.

[0064] S3.2.3 Input the Lora signal samples from step S3.2.1, namely the information entropy curve, power change curve, and signal-to-noise ratio of the three time-series sequences with a length of 1024, into the FallNet time-series detection model trained in step 3.2.2), and output a time-series sequence of the same length. Each value in the output time-series sequence is a fall probability value, and the softmax function is used to calculate the probability of a fall.

[0065] S3.3 Set the fall probability threshold to 0.5 in the trained temporal detection model FallNet. Input the signal features extracted in step S2 into the trained temporal detection model FallNet. The model processes the Lora signal features and determines whether a fall has occurred within the detection space covered by the Lora signal. If the predicted fall probability sequence output by the model exceeds the threshold, it is considered that an elderly person has fallen and a warning signal is issued.

[0066] This embodiment detects whether an elderly person has fallen using three LoRa signal features: information entropy curve, power change curve, and signal-to-noise ratio. These features exhibit morphological differences in their changes during normal activity and when a person falls. The impact of human activity on the LoRa signal is related to the shape of the information entropy curve. Therefore, this embodiment uses deep learning to analyze the temporal characteristics of the signal features to predict the timing of the fall, thereby detecting the elderly person's fall behavior.

[0067] Other components of the wall-penetrating fall detection and early warning system based on Lora signals according to embodiments of the present invention, such as transmitters, USRP devices, transmitting antennas, and receiving antennas, as well as their operation, are known to those skilled in the art and will not be described in detail here.

[0068] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0069] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0070] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection, an electrical connection, or a communication connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0071] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first feature is in direct contact with the second feature, or that the first feature is in indirect contact with the second feature through an intermediate medium. Furthermore, "above," "over," and "on top" of the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0072] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0073] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for detecting and warning of falls through walls based on LoRa signals, wherein the detection and warning system comprises a transmitter, a receiver, a USRP device, and a processor, wherein the transmitter is a LoRa node used to transmit LoRa signals through a transmitting antenna; the receiver is equipped with two receiving antennas for receiving the LoRa signals transmitted by the transmitter, wherein... Two receiving antennas are tightly connected together, with a predetermined distance between the receiving antenna and the transmitting antenna. The receiver is connected to the USRP device, and the transmitter and receiver are placed on the same side of the partitioned environment within the detection space. The USRP device is used to receive the Lora signal transmitted by the receiver. The processor is used to receive the Lora signal transmitted by the USRP device and process the received Lora signal using a pre-stored fall detection and early warning method. When a fall signal is detected in the elderly, an alarm signal is issued. The system is characterized by the following steps: S1 enhances the wall signal: high-frequency noise in the Lora signal is filtered out to obtain a signal with lower interference. The signal is then subjected to time-domain aliasing to amplify it and increase its amplitude. Finally, the signal entropy method is used to eliminate signal clock asynchrony. S2 Lora Signal Feature Extraction: Calculate the power jump curve of the Lora signal after processing in step S1, distinguish between static environmental signals and active signals, select active signal intervals in the active signals, perform video analysis on each active signal interval, calculate the time-frequency graph, and extract Lora signal features from the time-frequency graph: power jump curve, information entropy curve, and signal-to-noise ratio. S3 Elderly Fall Detection: Construct and train the temporal detection model FallNet. Input the LoRa signal features into the trained temporal detection model FallNet. The model processes the LoRa signal features and determines whether a fall has occurred within the detection space covered by the LoRa signal. If a fall has occurred, an early warning signal is issued.

2. The method for detecting and warning of falls through walls based on Lora signals according to claim 1, characterized in that, The temporal aliasing processing method in step S1 includes the following steps: S1.1 Divide the original time series into several sub-sampling sequences at intervals of ΔT = N × Δt seconds, where N is the number of aliasing windows and Δt is the size of the aliasing window; S1.2 Extract the time sequence corresponding to each aliasing window at each corresponding position in each sub-sampling sequence to form a time sequence Q. The number of time sequences Q generated is equal to the number of aliasing windows. S1.3 Superimpose all the time sequences Q generated in step S1.2 to form a new time sequence, i.e., the amplified signal.

3. The method for detecting and warning of falls through walls based on Lora signals according to claim 2, characterized in that, In step S1, the received signal clock synchronization is achieved using the signal quotient method, as follows: The ratio of the LoRa signals from the two receiving antennas is combined to form a new clock-synchronized LoRa signal, eliminating clock asynchrony. The formula is as follows: ； Where R1 and R2 represent the Lora signals received by the two receiving antennas of the USRP device, respectively, and SR represents the reconstructed signal after eliminating clock asynchrony.

4. The method for detecting and warning of falls through walls based on Lora signals according to claim 2, characterized in that, The specific steps of step S2 are as follows: S2.1 Calculate the power mutation curve of the Lora signal after processing in step S1, distinguish between static environmental signals and active signals. In the active signals, the active signal every T seconds is an active signal interval, and a person falls in each active signal interval. The active signal is the person's active signal. S2.2 performs video analysis on each active signal interval, calculates the time-frequency graph, and extracts LoRa signal features from the time-frequency graph: power jump curve, information entropy curve, and signal-to-noise ratio.

5. The method for detecting and warning of falls through walls based on Lora signals according to claim 4, characterized in that, The following methods can be used to distinguish between static environmental signals and active signals: S2.1.1 Calculate the power jump curve of the Lora signal after processing in step S1. When the value of the power jump curve is less than 2, it is a static environment signal. When the value of the power jump curve is greater than or equal to 2, it is an active signal. S2.1.2 In the active signal, the active signal every T seconds is an active signal interval.

6. The method for detecting and warning of falls through walls based on Lora signals according to claim 4, characterized in that, The specific method for step S2.2 is as follows: S2.2.1 In the active signal, each T second is considered as a window to obtain the T-second active signal interval. The time-frequency diagram within each T-second interval is calculated, with a calculation interval of T1 seconds. The time-frequency diagram is obtained through short-time fast Fourier transform, and the calculation formula is as follows: ； Where S(t,f) represents a two-dimensional matrix that shows the relationship between the signal's time and frequency. The window function is represented by SR, which represents the reconstructed signal after eliminating clock asynchrony. S2.2.2 Using the results of short-time fast Fourier transform in the obtained time-frequency graph, the power jump curve of the Lora signal is obtained through formula (3): ； Where PBC represents the computation performed over an interval of S(t,f); , indicating that the maximum value in the frequency range of S(t,f) is selected. Represents the minimum value of the interval; S2.2.3 The information entropy curve is obtained through formula (4): ； in, When time t is displayed on the time-frequency graph, f represents the instantaneous frequency. S2.2.4 The signal-to-noise ratio is obtained through formula (5): ； in, Original signal amplitude, Anoise Noise signal amplitude.

7. A method for detecting and warning of falls through walls based on Lora signals according to claim 2, characterized in that, Step S3 specifically includes the following steps: S3.1 Constructing the Temporal Detection Model FallNet: The entire temporal detection model FallNet is divided into two parts: upsampling and downsampling. The input signal features go through four downsampling stages and four upsampling stages. Downsampling is performed through one-dimensional convolution and stride. The length of the convolution kernel is set to 2 data points, and the stride step is set to 2 data points. Upsampling is performed through inverse integration to restore the input length of the previous stage. Each deconvolution stage directly concatenates the feature layers output from each downsampling stage in the downsampling process to the corresponding upsampling stage in the upsampling process, and uses the softmax function to calculate the probability of the fall activity. S3.2 Training the Temporal Detection Model FallNet: S3.2.1 Sample Acquisition: Lora signals are acquired within the detection space. The Lora signals include four types of signals: falling, standing still, walking, and standing. Each signal sample contains a Lora signal sample with a duration of T seconds. The signal sample includes three Lora signal features: information entropy curve, power change curve, and signal-to-noise ratio, each with a time sequence length of 1024. The Lora signal samples are labeled using a fuzzy labeling method. S3.2.2 Training Model: The cross-entropy function is used as the loss function, and the stochastic gradient descent method is used to optimize the training of the time-series detection model FallNet. Signal samples are used to train the time-series detection model FallNet, and the number of training batches, the number of LoRa signal samples in each batch of training and the learning rate are set. S3.2.3 Input the Lora signal samples from step S3.2.1, namely the information entropy curve, power change curve, and signal-to-noise ratio of the three time series with a length of 1024, into the FallNet time series detection model trained in step 3.2.2), and output a time series of the same length. Each value in the output time series is a fall probability value. The softmax function is used to calculate the probability of the fall activity. S3.3 Set a fall probability threshold in the trained temporal detection model FallNet. Input the signal features extracted in step S2 into the trained temporal detection model FallNet. The model processes the Lora signal features and determines whether a fall has occurred in the detection space covered by the Lora signal. If the predicted fall probability sequence output by the model exceeds the threshold, it is considered that someone has fallen and a warning signal is issued.

8. The method for detecting and warning of falls through walls based on Lora signals according to claim 7, characterized in that, The specific labeling method for the fuzzy annotation is as follows: A window of length N milliseconds is used, where M > 50. When labeling the fall probability data, the moment when a person is observed to fall is taken as the center. Each point in the window represents the probability of whether a fall has occurred. A Gaussian distribution curve is fitted around the center of the window. The mean of the curve is 0 and the standard deviation is M / 10. Finally, the values ​​of all points are multiplied by M / 4, so that the fall probability values ​​in the window decrease from 1 to 0 from the middle to both sides. For normal activities that do not involve falling, the probability value on the entire label is marked as 0.

Images