Bed-tail non-contact periodic sleep periodic limb movement disorder and sleep position combined monitoring device

Abstract

The application discloses a bed tail non-contact periodic sleep limb movement disorder and sleep position combined monitoring device, which realizes the following functions: collecting multi-view continuous echo signals of human leg actions and sleeping postures; respectively performing signal processing on the echo signals and performing fragment segmentation through a dynamic adjustment window algorithm; respectively performing time-distance-frequency multi-spectrum domain combined analysis on the multi-view echo signal fragments, generating time-frequency graphs and time-distance graphs under different views, and then inputting the time-frequency graphs and the time-distance graphs into a sleep body movement recognition model based on a multi-view multi-task architecture, and outputting sleep leg movement type and sleeping posture type judgment results; judging whether the judgment result of the sleep leg movement meets the characteristics of periodic limb movement, and if yes, marking as a periodic leg movement sequence; and judging whether there is a sleep periodic limb movement disorder. The application realizes the combined recognition and monitoring of the periodic sleep limb movement disorder and the sleep position, and provides a flexible and convenient monitoring method for home intelligent sleep monitoring.

Description

Technical Field

[0001] This invention belongs to the field of medical sleep monitoring technology, specifically a non-contact, periodic sleep limb movement disorder and sleep posture combined monitoring device at the foot of the bed. Background Technology

[0002] Periodic Limb Movement Disorder (PLMD) is a sleep disorder that occurs during sleep. The main symptom is recurrent, periodic, involuntary limb movements during sleep, typically manifesting as big toe extension and ankle dorsiflexion, sometimes also affecting the knee and hip joints. PLMD has multiple negative impacts. In terms of sleep, it disrupts sleep continuity and stability, leading to a lack of deep sleep and causing chronic fatigue. In daily life, it causes daytime sleepiness, affecting work and study, causing mood problems, and leading to cognitive decline. In terms of health, it increases the risk of cardiovascular disease, weakens the immune system, and disrupts the endocrine system balance. Sleep posture is an important piece of information in sleep monitoring and is significant for health assessments. In health assessments, it can aid in the diagnosis of diseases such as sleep apnea syndrome; different postures affect the condition, and it can also assess cardiovascular burden; it can also provide early warning of health problems caused by long-term poor posture. Regarding improving sleep quality, choosing suitable bedding based on posture preferences, optimizing the environment, and adjusting habits can help those with sleep disorders sleep better. In terms of safety, measures are taken to prevent accidents from happening to special groups such as the elderly and pregnant women, and to provide nursing support for rehabilitation patients, ensuring correct posture to promote recovery.

[0003] In the field of clinical medicine, periodic sleep limb movement disorders and sleep posture are usually monitored using polysomnography (PSG). This monitoring method has the following disadvantages: (1) High operation difficulty. Polysomnography is a professional medical device containing various types of sensitive sensors and electrodes. The attachment methods for different types of sensors are very different, so professional sleep medicine technicians are required to operate and implement it; (2) High testing cost. PSG devices are expensive and only a few sleep centers have them. They also require professional medical technicians to operate, resulting in high testing costs for patients; (3) Low measurement comfort. When using PSG to measure sleep, a large number of electrodes or sensors need to be worn or attached, which can easily cause discomfort to the subjects, thereby affecting their sleep quality and test results; (4) Large workload for result interpretation. The multiple physiological signals output by PSG currently require professional medical technicians to interpret them according to relevant medical guidelines, which is labor-intensive and inefficient. The above problems directly prevent the widespread application of PSG devices, especially failing to meet the needs of home-based periodic sleep monitoring.

[0004] Radar sensors have been proven to accurately detect the movement of human targets, offering advantages such as non-contact, imperceptible operation, and all-weather, 24 / 7 capability. They are widely used in applications such as human target localization, posture recognition, and gesture recognition. Artificial intelligence technology can automatically classify results and identify targets based on radar echo data, thereby reducing the difficulty and workload of data interpretation and extracting deeper, implicit information from radar echo data.

[0005] In summary, by placing radar sensors at the foot of the bed facing the human body and supplementing them with artificial intelligence, non-contact monitoring and identification of periodic sleep limb movement disorders and sleep positions can be achieved. Summary of the Invention

[0006] The purpose of this invention is to address the problems existing in the prior art by providing a non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed.

[0007] The technical solution to achieve the purpose of this invention is: a non-contact monitoring device for periodic sleep limb movement disorder and sleep posture at the foot of the bed, the device including a signal acquisition module, a signal processing module, a feature extraction module, a classification module, a sleep periodic limb movement sequence judgment module, and a sleep periodic limb movement disorder judgment module;

[0008] The signal acquisition module is used to non-contactly and continuously acquire echo signals of human leg movements and sleeping postures from different perspectives using several distributed FMCW radars.

[0009] The signal processing module is used to process radar echo signals from different perspectives and to identify and segment adaptive leg movement segments through a dynamic window adjustment algorithm.

[0010] The feature extraction module is used to perform time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding perspectives.

[0011] The classification module is used to input the time-frequency map and time-distance map into the deep network model and simultaneously determine the leg movement type and sleeping posture type of the corresponding segment; the deep network model is a leg movement-sleeping posture multi-task classification model based on multi-view and multi-spectral domain information fusion.

[0012] The sleep periodic limb movement sequence judgment module is used to determine whether the leg movement output meets the characteristics of periodic limb movement. If it does, the periodic leg movement sequence is marked.

[0013] The sleep-periodic limb movement disorder (SPLM) detection module is used to determine whether a sleep-periodic limb movement disorder exists.

[0014] Furthermore, the signal acquisition module continuously and non-contactly acquires echo signals of human leg movements and sleeping postures from different perspectives using several distributed FMCW radars, specifically including:

[0015] Step 2-1, Data Acquisition: The subject initially lies flat on the bed, and several FMCW radar sensors are distributed and installed at the foot of the bed, with the antennas all facing the legs of the subject; different FMCW radar sensors collect radar leg echo signals from different angles;

[0016] The echo signal at each viewpoint is represented as follows: ), where r is the original radar signal, and This represents the time sequence number of the slow sampling. This represents the total number of sampling times during slow sampling. The number represents the sequence number of the distance gate, and N represents the total number of distance gates; and These represent the sampling times for the slow and fast times, respectively.

[0017] Step 2-2, moving target detection: The leg echo signals R collected from each viewpoint are sampled at different speeds and then rearranged to obtain... The echo matrix, where X represents the number of sampling points in fast mode and Y represents the number of sampling points in slow mode;

[0018] Steps 2-3: Use a loop function to iterate through the leg echo signal R. In each loop, subtract the first column of data from the data at fixed time intervals to remove background noise interference.

[0019] Steps 2-4: Filter the leg echo signal R after removing background clutter interference to remove the tail shadow clutter generated by moving target detection, and obtain the preprocessed leg echo signal.

[0020] Furthermore, the signal processing module performs signal processing on radar echo signals from different viewpoints and uses a dynamic window adjustment algorithm to adaptively identify and segment leg movement segments, specifically including:

[0021] Step 3-1: Demodulate the leg echo signal from the fixed-distance door to obtain the phase signal. ;

[0022] Step 3-2, Motion Information Extraction: Perform third-order difference processing on the phase signal to obtain a further enhanced leg motion signal. ;

[0023] Step 3-3, Sliding window detection: Use 2 The sliding window traverses the enhanced leg motion signal Calculate the variance within each window. and the current window The window mean variance of the past N non-leg movement events NLM And determine the current window based on the threshold condition shown in the following formula. Check if the signal is a leg movement event (LM), then proceed with steps 3-4.

[0024]

[0025] If the current window If the signal satisfies the above formula, it indicates a leg movement event (LM); otherwise, the window mean variance is updated. ;

[0026] in, As an empirical threshold, This refers to the segment sampling length;

[0027] Step 3-4, Dynamic Window Adjustment: If the threshold condition in Step 3-3 is met, then the peak and valley points within the window are further detected. If there are two adjacent valley points containing a peak point in between, it indicates a complete leg movement event (LM), and the time segment t seconds before or after the peak point is saved. If not, it indicates that the leg movement event LM is incomplete, and the dynamic window is shifted backward by t seconds to continue detecting peak and valley points, ensuring that a complete leg movement event LM time segment is obtained.

[0028]

[0029] in, , These represent the nth and (n+1)th valley points, respectively. This represents the m-th peak point;

[0030] If the threshold condition is not met in step 3-3, the time segment within the window is saved as a non-leg movement segment NLM, thus obtaining the complete leg movement segment LM and the non-leg movement segment NLM.

[0031] Furthermore, the feature extraction module performs time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding viewpoints, specifically including:

[0032] Step 4-1: Obtain the leg echo signal segment, and perform time-frequency analysis on the fixed-distance gate complex signal through short-time Fourier transform to obtain a complete time-frequency diagram that reflects the positive and negative frequency changes;

[0033] Step 4-2: Obtain leg echo signal segments and use an adaptive neighbor signal enhancement time-distance map optimization algorithm to obtain a high-resolution time-distance map.

[0034] Furthermore, step 4-1 specifically includes:

[0035] Step 4-1-1: Obtain a t-second LM segment containing n sampling points using the energy envelope-based sleep limb activity continuous monitoring algorithm; take the fixed-distance leg movement signal lm[n] during this time period.

[0036] Step 4-1-2: Set the window length, overlap length, and number of points for STFT calculation, and perform a short-time Fourier transform, as shown below:

[0037]

[0038] in, Is Position aligned to The rectangular window function slides based on the window length and overlap length. The number of points calculated for STFT. for The leg movement signals at any moment, For window length, For the time-frequency matrix, ( () represents the coordinate points of the time-frequency matrix;

[0039] Step 4-1-3, obtain the time-frequency matrix Converted into a two-dimensional heatmap This serves as one of the subsequent model inputs.

[0040] Furthermore, step 4-2 specifically includes:

[0041] Step 4-2-1: The acquired leg movement signal segments are formed into a time-distance matrix and converted into a two-dimensional heat map, i.e., a distance map. At the same time, the range of the matrix is ​​restricted to several distance gates to ensure that it covers the distance range of leg movements when a person is lying flat in a sleeping state.

[0042] Step 4-2-2: Perform signal smoothing processing on the signal from step 4-2-1;

[0043] Step 4-2-3, Trajectory Extraction: If two adjacent distance gates detect a signal and the signal strength detected by any other distance gate is lower than the preset threshold (i.e., no signal is detected), then it is determined that the signals detected by the first two distance gates originate from the same leg movement; if non-adjacent distance gates detect a signal, then it is determined that these signals may come from the movement of multiple parts of the body. In this case, the distance gate with the strongest signal is selected as the estimated target position.

[0044] Step 4-2-4, Binarization: Morphological dilation and erosion are introduced to further optimize the time-distance map;

[0045] Step 4-2-5, obtain the time-frequency matrix It is converted into a two-dimensional heatmap, which is then used as one of the inputs for subsequent models.

[0046] Furthermore, the classification module inputs the time-frequency map and time-distance map into the deep network model to simultaneously determine the leg movement type and sleeping posture type of the corresponding segment, specifically including:

[0047] Step 5-1: Input the time-frequency map and time-distance map into the deep network model to obtain the initial leg movement / sleep posture features and their initial predictions.

[0048] Step 5-2, multimodal distillation, includes the calculation and combination of affinity matrices, feature fusion, and feature reconstruction to achieve feature interaction between the two tasks of leg movement classification and sleep position classification. Specifically, it includes:

[0049] Step 5-2-1 involves calculating the pixel-level affinity matrix, specifically including: the first step is dimensionality reduction and activation, and the second step is restoring the number of channels and normalization, thereby obtaining the affinity matrix for the leg movement classification task. Affinity matrix for sleep position classification task ;

[0050] Step 5-2-2: Introduce learnable weight parameters and The affinity matrix for each task is weighted and combined, and expressed as follows:

[0051]

[0052]

[0053] In the formula, , These are the affinity matrices for the weighted leg movement classification task and the sleep posture classification task, respectively.

[0054] Step 5-2-3: Combine the initial features with the combined affinity matrix obtained in step 5-2-2, that is, perform pixel-by-pixel multiplication between the weighted affinity matrix and the original task features to obtain the refined leg movement features. and sleep position characteristics :

[0055]

[0056]

[0057] Step 5-2-4: The refined features are passed to their respective feature reconstruction modules to obtain the reconstructed feature maps; the feature reconstruction modules are specifically:

[0058] The feature reconstruction module is an upsampling module designed to gradually restore the spatial resolution of the feature map through deconvolution operations. During the convolution operation, the input is padded by 1 pixel. After upsampling, ordinary convolution is used to eliminate the interpolation artifacts caused by deconvolution. The ReLU activation function is used to introduce a nonlinear transformation to suppress irrelevant information and alleviate the gradient vanishing problem.

[0059] The feature reconstruction module performs two deconvolution-convolution operations. The first deconvolution layer first doubles the spatial size of the input feature map and halves the number of channels, and then performs a convolution operation while keeping the number of channels unchanged. The second deconvolution layer doubles the feature map again, restores the number of channels to the initial state, and then performs a convolution operation while keeping the number of channels unchanged.

[0060] Step 5-3: Input the reconstructed feature map into the classifier to achieve the final classification and recognition, and output the leg movement type. and types of sleeping positions ;

[0061] Among them, leg movements Includes two types of periodic leg movements, PLM1 and PLM2, double leg twitching, leg crossing, leg extension and flexion, rolling over, and non-leg movement NLM; sleep position types This includes supine (Supine), side-lying (Side), prone (Prone), and non-sleeping position (LM).

[0062] Furthermore, the deep network model described in step 5-1 is specifically as follows:

[0063] The deep learning model includes a shared feature extraction layer, two task-specific classification layers, and a dynamic weight averaging layer; the two task-specific classification layers include a leg movement classification task layer and a sleep position classification task layer.

[0064] The structure of the shared feature extraction layer is as follows:

[0065] (1) ResNet18 is used for pre-training. The initial layer is a 7×7 convolutional layer with a stride of 2, followed by a 3×3 max pooling layer with a stride of 2. Then, it passes through four residual groups in sequence. Each residual group contains two basic residual blocks, and each residual block contains two 3×3 convolutional layers. After the last residual group, a global average pooling layer is used to transform the feature map into a fixed-length feature vector.

[0066] (2) Remove the final fully connected layer in the pre-trained ResNet18 and retain only its convolution and pooling layers to extract features from the input multi-view time-frequency map and time-distance map respectively.

[0067] (3) Add an SE block to each view of the time-frequency map and the time-distance map to perform channel attention calibration on the extracted feature map. Then, add the three input signals of the time-frequency map and the time-distance map with weights to obtain the time-frequency map and time-distance map features after multi-view fusion.

[0068] (4) The time-frequency map and time-distance map after multi-view fusion are calibrated by an SE block, and then the two features are spliced ​​together to obtain the multi-view time-frequency and time-distance fusion features.

[0069] (5) The multi-view time-frequency-time-distance fusion features are passed through a fully connected layer to output the classification result, which is used as the initial training result;

[0070] (6) By using the bottleneck layer, which includes two 1×1 convolutional layers, the number of channels is reduced from 1024 to 512 and 128 respectively, further compressing the feature information;

[0071] (7) Finally, the feature map is flattened by global average pooling;

[0072] In the leg movement classification task layer, the 128-dimensional features are first reduced to 64-dimensional features using a fully connected layer with ReLU activation function, and then another fully connected layer is used to output 6-class prediction results.

[0073] In the sleep posture classification task layer, the same method as the leg movement classification task layer is used to output three types of prediction results.

[0074] In the dynamic weighted averaging layer, the loss for each task is adjusted using a dynamic weighting strategy. The weight for the current task is calculated based on the classification loss from the previous round to balance the training of the two tasks. The formula for calculating the dynamic weight is:

[0075]

[0076] in,

[0077]

[0078] In the formula, leg movement classification and sleeping posture classification correspond to respectively Take 1 and 2, Representative task The weight, The task is in the Losses during the march, It is the first The training speed is calculated step-by-step; a larger value indicates a faster decrease in loss, which allows for a smaller weight, thus reducing the proportion of loss for that task. T is a temperature constant; when T = 1... Without the effect of weight adjustment, the larger T is, the more stable the weight adjustment becomes. After obtaining the dynamic weights, the overall loss L can be calculated.

[0079]

[0080] Finally, based on the weighted network parameters, the leg movement category and sleeping posture category are output in the leg movement classification layer and the sleeping posture classification layer, respectively. When the signal processing module determines that it is in the non-leg movement segment NLM, the leg movement prediction result is directly output as NLM. The sleep posture prediction does not go through the multi-task learning network and is directly output by the sleep posture classifier.

[0081] Furthermore, the sleep periodic limb movement sequence determination module determines whether the leg movement output meets the characteristics of periodic limb movement. If it does, the periodic leg movement sequence is marked, specifically including:

[0082] Step 6-1, determine leg movement output Is it 0 or 1? If so, mark this data segment as a periodic leg movement PLM.

[0083] Step 6-2: When the continuity of the periodic leg movement PLM meets the preset conditions, it is considered a set of periodic leg movement sequences PLMS; at the same time, the sleeping posture output is saved. .

[0084] Furthermore, the sleep-periodic limb movement disorder (SPLM) determination module determines whether sleep-periodic limb movement disorder exists, specifically including:

[0085] Step 7-1: Count the number of times the periodic leg movement sequence (PLMS) occurs during the entire night's sleep, and divide the number by the total sleep time to obtain the periodic limb movement index (PLMI).

[0086] Step 7-2: Determine whether sleep-periodic limb movement disorder exists. If the periodic limb movement index (PLMI) is >5 / h for children and >15 / h for adults, then sleep-periodic limb movement disorder exists; otherwise, sleep-periodic limb movement disorder does not exist.

[0087] Compared with the prior art, the significant advantages of this invention are:

[0088] (1) The non-contact periodic sleep limb movement disorder and sleep posture joint monitoring method proposed in this invention uses non-contact FMCW radar to collect leg movement and sleeping posture data, avoiding interference to the subject from wearing the device, improving the comfort and convenience of monitoring, and enabling long-term monitoring without interfering with the user.

[0089] (2) The radar signal processing flow proposed in this invention includes phase demodulation and short-time Fourier transform, which performs high-resolution time-frequency and time-distance analysis on the radar echo signal, and can extract the feature information of leg movement and sleeping posture. This flow can greatly improve the accuracy of the leg movement and sleeping posture acquisition database, ensure the effectiveness of the model input, and improve the accuracy of leg movement classification and sleeping posture classification.

[0090] (3) The present invention adopts a multi-task learning model, which can share the feature representations of leg movement classification and sleeping posture classification tasks. Through joint training, the performance of classification tasks is improved, and the training efficiency and generalization ability are significantly enhanced.

[0091] (4) Data acquisition from multiple perspectives is carried out by multiple FMCW radar sensors installed at the foot of the bed. By fusing signals from multiple perspectives and multiple spectral domains, the detection blind zone is effectively reduced and the signal quality is enhanced. Furthermore, no additional multi-sensor devices or complex hardware systems are required, which reduces hardware costs and system complexity, while improving the convenience and robustness of data acquisition.

[0092] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0093] Figure 1 This is a schematic diagram illustrating the implementation principle of a non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed in one embodiment.

[0094] Figure 2 This is a test scenario diagram from one embodiment.

[0095] Figure 3 Here are time-frequency graphs of different leg movements and sleeping positions in one embodiment, where Figure 3 (a) to (f) are time-frequency diagrams corresponding to PLM1, PLM2, twitching, leg crossing, leg extension and flexion, and rolling over, respectively.

[0096] Figure 4 This is a time-distance graph of different leg movements and sleeping positions in one embodiment, wherein Figure 4 (a) to (f) are time interval diagrams corresponding to PLM1, PLM2, twitching, cross-leg, leg extension and flexion, and rolling over, respectively.

[0097] Figure 5 This is a model block diagram in one embodiment. Detailed Implementation

[0098] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0099] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0100] In one embodiment, a non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed is provided, combined with Figure 1 The device includes: a signal acquisition module, a signal processing module, a feature extraction module, a classification module, a sleep periodic limb movement sequence judgment module, and a sleep periodic limb movement disorder judgment module;

[0101] The signal acquisition module is used to non-contactly and continuously acquire echo signals of human leg movements and sleeping postures from different perspectives using several distributed FMCW radars.

[0102] The signal processing module is used to process radar echo signals from different perspectives and to identify and segment adaptive leg movement segments through a dynamic window adjustment algorithm.

[0103] The feature extraction module is used to perform time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding perspectives.

[0104] The classification module is used to input the time-frequency map and time-distance map into the deep network model and simultaneously determine the leg movement type and sleeping posture type of the corresponding segment; the deep network model is a leg movement-sleeping posture multi-task classification model based on multi-view and multi-spectral domain information fusion.

[0105] The sleep periodic limb movement sequence judgment module is used to determine whether the leg movement output meets the characteristics of periodic limb movement. If it does, the periodic leg movement sequence is marked.

[0106] The sleep-periodic limb movement disorder (SPLM) detection module is used to determine whether a sleep-periodic limb movement disorder exists.

[0107] Furthermore, in one embodiment, the signal acquisition module continuously and non-contactly acquires echo signals of human leg movements and sleeping postures from different perspectives using several distributed FMCW radars, specifically including:

[0108] Step 2-1, Data Acquisition: The subject initially lies flat on the bed, and several FMCW radar sensors are distributed and installed at the foot of the bed, with the antennas all facing the legs of the subject; preferably, the distance between the radar sensor and the part of the subject being measured is about 30 centimeters; different FMCW radar sensors collect radar leg echo signals from different angles;

[0109] Here, for example, the test scenario is as follows: Figure 2 As shown in Table 5.1 and Table 5.2, continuous data were collected on periodic leg movements, common human leg movements, and three basic sleeping positions.

[0110] Table 5.1 Description of Leg Movements

[0111] Leg movements Detailed description Periodic leg movements 1 Brief, stereotyped extension of the big toe and dorsiflexion of the foot. Periodic leg movements 2 Brief, stereotyped extension of the big toe, dorsiflexion of the foot, and simultaneous flexion of the knee and hip, similar to a kick. twitch The slight up-and-down twitching of the calf is a rapid and involuntary slight contraction or shaking of the leg muscles. Crossed legs The legs are crossed, with one leg placed on top of the other. Leg extension and flexion The movement of the leg from bent to straight, or from straight back to bent. Turn over The body rotates laterally, accompanied by the movement and adjustment of the legs.

[0112] Table 5.2 Description of Sleep Positions

[0113] sleeping position Detailed description supine Lie face up on the bed with your back touching the bed surface, your legs and arms naturally placed at your sides, and your feet facing the radar. Side-lying Lie on your side on the bed with your head, shoulders, and hips in a slanted line. Your legs can be bent or straightened, and your feet rotated 90 degrees to align with the radar. prone Lie face down on the bed with your chest and abdomen in contact with the bed surface, head turned to one side, legs and arms can be naturally extended or slightly bent, and feet no longer obstructing the radar.

[0114] The echo signal at each viewpoint is represented as follows: ), where r is the original radar signal, This represents the time sequence number of the slow sampling. This represents the total number of sampling times during slow sampling. The number represents the sequence number of the distance gate, and N represents the total number of distance gates; and These represent the sampling times for the slow and fast times, respectively.

[0115] Step 2-2, moving target detection: The leg echo signals R collected from each viewpoint are sampled at different speeds and then rearranged to obtain... The echo matrix, where X represents the number of sampling points in fast mode and Y represents the number of sampling points in slow mode;

[0116] Steps 2-3: Use a loop function to iterate through the leg echo signal R. In each loop, subtract the first column of data from the data at fixed time intervals to remove background noise interference.

[0117] Steps 2-4 involve filtering the leg echo signal R after removing background clutter interference to remove the tail shadow clutter generated by moving target detection, resulting in a preprocessed leg echo signal. Preferably, median filtering is used.

[0118] Furthermore, in one embodiment, the signal processing module performs signal processing on radar echo signals from different viewpoints and uses a dynamic window adjustment algorithm to adaptively identify and segment leg movement segments, specifically including:

[0119] Step 3-1: Demodulate the leg echo signal from the fixed-distance door to obtain the phase signal. ;

[0120] Step 3-2, Motion Information Extraction: Perform third-order difference processing on the phase signal to obtain a further enhanced leg motion signal. ;

[0121] Step 3-3, Sliding window detection: Use 2 The sliding window traverses the enhanced leg motion signal Calculate the variance within each window. and the current window The window mean variance of the past N non-leg movement events NLM And determine the current window based on the threshold condition shown in the following formula. Check if the signal is a leg movement event (LM), then proceed with steps 3-4.

[0122]

[0123] If the current window If the signal satisfies the above formula, it indicates a leg movement event (LM); otherwise, the window mean variance is updated. ;

[0124] in, As an empirical threshold, This refers to the segment sampling length;

[0125] Here, according to the AASM periodic leg movement interpretation standard, the period length between leg movement events (LM) is 5~90s, so the segment length is set. A time of 5 seconds is reasonable, as is the number of sampling points in the segment. ;

[0126] Step 3-4, Dynamic Window Adjustment: If the threshold condition in Step 3-3 is met, then the peak and valley points within the window are further detected. If there are two adjacent valley points containing a peak point in between, it indicates a complete leg movement event (LM), and the time segment t seconds before or after the peak point is saved. If not, it indicates that the leg movement event LM is incomplete, and the dynamic window is shifted backward by t seconds to continue detecting peak and valley points, ensuring that a complete leg movement event LM time segment is obtained.

[0127]

[0128] in, , These represent the nth and (n+1)th valley points, respectively. This represents the m-th peak point;

[0129] If the threshold condition is not met in step 3-3, the time segment within the window is saved as a non-leg movement segment NLM, thus obtaining the complete leg movement segment LM and the non-leg movement segment NLM. The obtained leg movement segment is then input into the multi-task learning model for simultaneous classification of leg movement and sleep position, while the non-leg movement segment is directly passed through the classifier and only outputs the sleep position classification.

[0130] Here, preferably, t seconds is 5 seconds.

[0131] Furthermore, in one embodiment, the feature extraction module performs time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding viewpoints, specifically including:

[0132] Step 4-1: Obtain the leg echo signal segment, and perform time-frequency analysis on the fixed-distance gate complex signal through short-time Fourier transform to obtain a complete time-frequency diagram that reflects the positive and negative frequency changes;

[0133] Step 4-2: Obtain leg echo signal segments and use an adaptive neighbor signal enhancement time-distance map optimization algorithm to obtain a high-resolution time-distance map.

[0134] Furthermore, in one embodiment, step 4-1 specifically includes:

[0135] Step 4-1-1: Obtain a LM segment of t seconds using the energy envelope-based continuous monitoring algorithm for sleep limb activity, containing n sampling points (preferably, t is 2 and n is 250); take the fixed-distance leg movement signal lm[n] during this time period;

[0136] Step 4-1-2: Set the window length, overlap length, and number of points for STFT calculation, and perform a short-time Fourier transform, as shown below:

[0137]

[0138] in, Is Position aligned to The rectangular window function slides based on the window length and overlap length. The number of points calculated for STFT. for The leg movement signals at any moment, For window length, It is a time-frequency matrix;

[0139] Step 4-1-3, obtain the time-frequency matrix Converted into a two-dimensional heatmap This serves as one of the subsequent model inputs. Because leg movement and sleeping posture information are coupled in the time-frequency plots, it is difficult to directly observe the characteristics of different sleeping postures; therefore, time-frequency plots for different sleeping postures are not displayed. Figure 3 The time-frequency diagrams of different leg movements are shown.

[0140] Furthermore, in one embodiment, step 4-2 specifically includes:

[0141] Step 4-2-1: The acquired leg movement signal segments are formed into a time-distance matrix and converted into a two-dimensional heat map, i.e., a distance map. At the same time, the range of the matrix is ​​restricted to several distance gates to ensure that it covers the distance range of leg movements when a person is lying flat in a sleeping state.

[0142] Preferably, the matrix range is limited to 1 to 3 distance gates, i.e., 0 to 1.32 meters;

[0143] Step 4-2-2: Perform signal smoothing processing on the signal from step 4-2-1;

[0144] Preferably, two-dimensional median filtering is first performed using a 3×3 sliding window to suppress isolated noise points, and then Gaussian filtering is used to remove noise.

[0145] Step 4-2-3, Trajectory Extraction: If two adjacent distance gates detect a signal and the signal strength detected by any other distance gate is lower than the preset threshold (i.e., no signal is detected), then it is determined that the signals detected by the first two distance gates originate from the same leg movement; if non-adjacent distance gates detect a signal, then it is determined that these signals may come from the movement of multiple parts of the body. In this case, the distance gate with the strongest signal is selected as the estimated target position.

[0146] Step 4-2-4, Binarization: Morphological dilation and erosion are introduced to further optimize the time-distance map;

[0147] Step 4-2-5, obtain the time-frequency matrix The data was converted into a two-dimensional heatmap and used as one of the inputs for subsequent models. Because leg movement and sleeping posture information are coupled in the time-distance graph, it is difficult to directly observe the characteristics of different sleeping postures. Therefore, the time-distance graphs for different sleeping postures were not displayed. Figure 4 The time-distance graphs for different leg movements are shown.

[0148] Furthermore, in one embodiment, the classification module inputs the time-frequency map and time-distance map into the deep network model to simultaneously determine the leg movement type and sleeping posture type of the corresponding segment, specifically including:

[0149] Step 5-1: Input the time-frequency map and time-distance map into the deep network model to obtain the initial leg movement / sleep posture features and their initial predictions.

[0150] Step 5-2, multimodal distillation, includes the calculation and combination of affinity matrices, feature fusion, and feature reconstruction to achieve feature interaction between the two tasks of leg movement classification and sleep position classification. Specifically, it includes:

[0151] Step 5-2-1 involves calculating the pixel-level affinity matrix, specifically including: the first step is dimensionality reduction and activation, and the second step is restoring the number of channels and normalization, thereby obtaining the affinity matrix for the leg movement classification task. Affinity matrix for sleep position classification task ;

[0152] Here, the feature extractor in ResNet18 downsamples the input multiple times, resulting in a final output feature map of size 2×2 with 128 channels, i.e., a size of [128, 2, 2]. Preferably, the calculation of the affinity matrix includes two steps: the first step is dimensionality reduction and activation, using a 1×1 convolutional kernel to reduce the input channels to 8 and applying the ReLU activation function; the second step is channel restoration and normalization, using another 1×1 convolutional kernel to restore the number of channels to 128 and applying the Sigmoid activation function to constrain the output to the range [0, 1].

[0153] Step 5-2-2: Introduce learnable weight parameters and The affinity matrix for each task is weighted and combined, and expressed as follows:

[0154]

[0155]

[0156] In the formula, , These are the affinity matrices for the weighted leg movement classification task and the sleep posture classification task, respectively.

[0157] Step 5-2-3: Combine the initial features with the combined affinity matrix obtained in step 5-2-2, that is, perform pixel-by-pixel multiplication between the weighted affinity matrix and the original task features to obtain the refined leg movement features. and sleep position characteristics :

[0158]

[0159]

[0160] Step 5-2-4: The refined features are passed to their respective feature reconstruction modules to obtain the reconstructed feature maps; the feature reconstruction modules are specifically:

[0161] The feature reconstruction module is an upsampling module designed to gradually restore the spatial resolution of the feature map through deconvolution operations (preferably, deconvolution uses a 4×4 kernel with a stride of 2), and pads the input with 1 pixel during the convolution operation. After upsampling, ordinary convolution is used to eliminate the interpolation artifacts caused by deconvolution. The ReLU activation function is used to introduce a nonlinear transformation to suppress irrelevant information and alleviate the gradient vanishing problem.

[0162] The feature reconstruction module performs two deconvolution-convolution operations. The first deconvolution layer first doubles the spatial size of the input feature map and reduces the number of channels by half (from 128 to 64), and then performs a convolution operation while keeping the number of channels (64) unchanged. The second deconvolution layer doubles the feature map again, restores the number of channels to the initial state, and then performs a convolution operation while keeping the number of channels unchanged.

[0163] Step 5-3: Input the reconstructed feature map into the classifier to achieve the final classification and recognition, and output the leg movement type. and types of sleeping positions ;

[0164] Among them, leg movements This includes two types of periodic leg movements (PLM1, PLM2), double leg twitching, leg crossing, leg extension and flexion, rolling over (NPLM1, NPLM2, NPLM3, NPLM4), and non-leg movement NLM, labeled 0, 1, 2, 3, 4, 5, and 6 respectively; sleep position types. The positions included supine (Supine), lateral (Side), and prone (Prone), labeled 0, 1, and 2 respectively.

[0165] Furthermore, in one embodiment, combined with Figure 5 The deep network model described in step 5-1 is specifically as follows:

[0166] The deep learning model includes a shared feature extraction layer, two task-specific classification layers, and a dynamic weight averaging layer; the two task-specific classification layers include a leg movement classification task layer and a sleep position classification task layer.

[0167] The structure of the shared feature extraction layer is as follows:

[0168] (1) ResNet18 is used for pre-training. The initial layer is a 7×7 convolutional layer with a stride of 2, followed by a 3×3 max pooling layer with a stride of 2. Then, it passes through four residual groups in sequence. Each residual group contains two basic residual blocks, and each residual block contains two 3×3 convolutional layers. After the last residual group, a global average pooling layer is used to transform the feature map into a fixed-length feature vector.

[0169] (2) Remove the final fully connected layer in the pre-trained ResNet18 and retain only its convolution and pooling layers to extract features from the input multi-view time-frequency map and time-distance map respectively.

[0170] (3) Add an SE block to each view of the time-frequency map and the time-distance map to perform channel attention calibration on the extracted feature map. Then, add the three input signals of the time-frequency map and the time-distance map with weights to obtain the time-frequency map and time-distance map features after multi-view fusion.

[0171] (4) The time-frequency map and time-distance map after multi-view fusion are calibrated by an SE block, and then the two features are spliced ​​together to obtain the multi-view time-frequency and time-distance fusion features.

[0172] (5) The multi-view time-frequency-time-distance fusion features are passed through a fully connected layer to output the classification result, which is used as the initial training result;

[0173] (6) By using the bottleneck layer, which includes two 1×1 convolutional layers, the number of channels is reduced from 1024 to 512 and 128 respectively, further compressing the feature information;

[0174] (7) Finally, the feature map is flattened by global average pooling;

[0175] In the leg movement classification task layer, the 128-dimensional features are first reduced to 64-dimensional features using a fully connected layer with ReLU activation function, and then another fully connected layer is used to output 6-class prediction results.

[0176] In the sleep posture classification task layer, the same method as the leg movement classification task layer is used to output three types of prediction results.

[0177] In the dynamic weighted averaging layer, the loss for each task is adjusted using a dynamic weighting strategy. The weight for the current task is calculated based on the classification loss from the previous round to balance the training of the two tasks. The formula for calculating the dynamic weight is:

[0178]

[0179] in,

[0180]

[0181] In the formula, leg movement classification and sleeping posture classification correspond to respectively Take 1 and 2, Representative task The weight, The task is in the Losses during the march, It is the first The training speed is calculated step-by-step; a larger value indicates a faster decrease in loss, which allows for a smaller weight, thus reducing the proportion of loss for that task. T is a temperature constant; when T = 1... Without the effect of weight adjustment, the larger T is, the more stable the weight adjustment becomes. After obtaining the dynamic weights, the overall loss L can be calculated.

[0182]

[0183] Finally, based on the weighted network parameters, the leg movement category and sleeping posture category are output in the leg movement classification layer and the sleeping posture classification layer, respectively. When the signal processing module determines that it is in the non-leg movement segment NLM, the leg movement prediction result is directly output as NLM. The sleep posture prediction does not go through the multi-task learning network and is directly output by the sleep posture classifier.

[0184] Furthermore, in one embodiment, the sleep periodic limb movement sequence determination module determines whether the leg movement output meets the characteristics of periodic limb movement; if it does, it marks the periodic leg movement sequence, specifically including:

[0185] Step 6-1, determine leg movement output Is it 0 or 1? If so, mark this data segment as a periodic leg movement PLM.

[0186] Step 6-2: When the continuity of the periodic leg movement PLM meets the preset conditions, it is considered a set of periodic leg movement sequences PLMS; at the same time, the sleeping posture output is saved. .

[0187] Preferably, the preset condition is that the periodic leg movement PLM occurs continuously more than 4 times, that is, the interval between adjacent periodic leg movement PLM is between 5 and 90 seconds.

[0188] Furthermore, in one embodiment, the sleep-periodic limb movement disorder (SPLM) determination module determines whether SPLM exists, specifically including:

[0189] Step 7-1: Count the number of times the periodic leg movement sequence (PLMS) occurs during the entire night's sleep, and divide the number by the total sleep time to obtain the periodic limb movement index (PLMI).

[0190] Step 7-2: Determine whether sleep-periodic limb movement disorder exists. If the periodic limb movement index (PLMI) is >5 / h for children and >15 / h for adults, then sleep-periodic limb movement disorder exists; otherwise, sleep-periodic limb movement disorder does not exist.

[0191] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention without departing from its spirit and scope should be included within the protection scope of the present invention.

Claims

1. A non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed, characterized in that, The device includes a signal acquisition module, a signal processing module, a feature extraction module, a classification module, a sleep periodic limb movement sequence judgment module, and a sleep periodic limb movement disorder judgment module. The signal acquisition module is used to non-contactly and continuously acquire echo signals of human leg movements and sleeping postures from different perspectives using several distributed FMCW radars. The signal processing module is used to process radar echo signals from different perspectives and to identify and segment adaptive leg movement segments through a dynamic window adjustment algorithm. The feature extraction module is used to perform time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding perspectives. The classification module is used to input the time-frequency map and time-distance map into the deep network model and simultaneously determine the leg movement type and sleeping posture type of the corresponding segment; the deep network model is a leg movement-sleeping posture multi-task classification model based on multi-view and multi-spectral domain information fusion. The sleep periodic limb movement sequence judgment module is used to determine whether the leg movement output meets the characteristics of periodic limb movement. If it does, the periodic leg movement sequence is marked. The sleep-periodic limb movement disorder (SPLM) detection module is used to determine whether a sleep-periodic limb movement disorder exists.

2. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 1, characterized in that, The signal acquisition module uses several distributed FMCW radars to non-contactly and continuously acquire echo signals of human leg movements and sleeping postures from different perspectives, specifically including: Step 2-1, Data Acquisition: The subject initially lies flat on the bed, and several FMCW radar sensors are distributed and installed at the foot of the bed, with the antennas all facing the legs of the subject; different FMCW radar sensors collect radar leg echo signals from different angles; The echo signal at each viewpoint is represented as follows: ), where r is the original radar signal, and This represents the time sequence number of the slow sampling. This represents the total number of sampling times during slow sampling. The number represents the sequence number of the distance gate, and N represents the total number of distance gates; and These represent the sampling times for the slow and fast times, respectively. Step 2-2, moving target detection: The leg echo signals R collected from each viewpoint are sampled at different speeds and then rearranged to obtain... The echo matrix, where X represents the number of sampling points in fast mode and Y represents the number of sampling points in slow mode; Steps 2-3: Use a loop function to iterate through the leg echo signal R. In each loop, subtract the first column of data from the data at fixed time intervals to remove background noise interference. Steps 2-4: Filter the leg echo signal R after removing background clutter interference to remove the tail shadow clutter generated by moving target detection, and obtain the preprocessed leg echo signal.

3. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 1, characterized in that, The signal processing module processes radar echo signals from different viewpoints and uses a dynamic window adjustment algorithm to adaptively identify and segment leg movement segments, specifically including: Step 3-1: Demodulate the leg echo signal from the fixed-distance door to obtain the phase signal. ; Step 3-2, Motion Information Extraction: Perform third-order difference processing on the phase signal to obtain a further enhanced leg motion signal. ; Step 3-3, Sliding window detection: Use 2 The sliding window traverses the enhanced leg motion signal Calculate the variance within each window. and the current window The window mean variance of the past N non-leg movement events NLM And determine the current window based on the threshold condition shown in the following formula. Check if the signal is a leg movement event (LM), then proceed with steps 3-4. ； If the current window If the signal satisfies the above formula, it indicates a leg movement event (LM); otherwise, the window mean variance is updated. ; in, As an empirical threshold, This refers to the segment sampling length; Step 3-4, Dynamic Window Adjustment: If the threshold condition in Step 3-3 is met, then the peak and valley points within the window are further detected. If there are two adjacent valley points containing a peak point in between, it indicates a complete leg movement event (LM), and the time segment t seconds before or after the peak point is saved. If not, it indicates that the leg movement event LM is incomplete, and the dynamic window is shifted backward by t seconds to continue detecting peak and valley points, ensuring that a complete leg movement event LM time segment is obtained. ； in, , These represent the nth and (n+1)th valley points, respectively. This represents the m-th peak point; If the threshold condition is not met in step 3-3, the time segment within the window is saved as a non-leg movement segment NLM, thus obtaining the complete leg movement segment LM and the non-leg movement segment NLM.

4. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 1, characterized in that, The feature extraction module performs time-frequency analysis and time-distance analysis on the echo signal segments to generate time-frequency maps and time-distance maps from the corresponding viewpoints, specifically including: Step 4-1: Obtain the leg echo signal segment, and perform time-frequency analysis on the fixed-distance gate complex signal through short-time Fourier transform to obtain a complete time-frequency diagram that reflects the positive and negative frequency changes; Step 4-2: Obtain leg echo signal segments and use an adaptive neighbor signal enhancement time-distance map optimization algorithm to obtain a high-resolution time-distance map.

5. The non-contact periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 4, characterized in that, Step 4-1 specifically includes: Step 4-1-1: Obtain a t-second LM segment containing n sampling points using the energy envelope-based sleep limb activity continuous monitoring algorithm; take the fixed-distance leg movement signal lm[n] during this time period. Step 4-1-2: Set the window length, overlap length, and number of points for STFT calculation, and perform a short-time Fourier transform, as shown below: ； in, Is Position aligned to The rectangular window function slides based on the window length and overlap length. The number of points calculated for STFT. for The leg movement signals at any moment, For window length, For the time-frequency matrix, ( () represents the coordinate points of the time-frequency matrix; Step 4-1-3, obtain the time-frequency matrix Converted into a two-dimensional heatmap This serves as one of the subsequent model inputs.

6. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 4, characterized in that, Step 4-2 specifically includes: Step 4-2-1: The acquired leg movement signal segments are formed into a time-distance matrix and converted into a two-dimensional heat map, i.e., a distance map. At the same time, the range of the matrix is ​​restricted to several distance gates to ensure that it covers the distance range of leg movements when a person is lying flat in a sleeping state. Step 4-2-2: Perform signal smoothing processing on the signal from step 4-2-1; Step 4-2-3, Trajectory Extraction: If two adjacent distance gates detect a signal and the signal strength detected by any other distance gate is lower than the preset threshold (i.e., no signal is detected), then it is determined that the signals detected by the first two distance gates originate from the same leg movement; if non-adjacent distance gates detect a signal, then it is determined that these signals may come from the movement of multiple parts of the body. In this case, the distance gate with the strongest signal is selected as the estimated target position. Step 4-2-4, Binarization: Morphological dilation and erosion are introduced to further optimize the time-distance map; Step 4-2-5, obtain the time-frequency matrix It is converted into a two-dimensional heatmap, which is then used as one of the inputs for subsequent models.

7. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 1, characterized in that, The classification module inputs the time-frequency map and time-distance map into the deep network model to simultaneously determine the leg movement type and sleeping posture type of the corresponding segment, specifically including: Step 5-1: Input the time-frequency map and time-distance map into the deep network model to obtain the initial leg movement / sleep posture features and their initial predictions. Step 5-2, multimodal distillation, includes the calculation and combination of affinity matrices, feature fusion, and feature reconstruction to achieve feature interaction between the two tasks of leg movement classification and sleep position classification. Specifically, it includes: Step 5-2-1 involves calculating the pixel-level affinity matrix, specifically including: the first step is dimensionality reduction and activation, and the second step is restoring the number of channels and normalization, thereby obtaining the affinity matrix for the leg movement classification task. Affinity matrix for sleep position classification task ; Step 5-2-2: Introduce learnable weight parameters and The affinity matrix for each task is weighted and combined, and expressed as follows: ； ； In the formula, , These are the affinity matrices for the weighted leg movement classification task and the sleep posture classification task, respectively. Step 5-2-3: Combine the initial features with the combined affinity matrix obtained in step 5-2-2, that is, perform pixel-by-pixel multiplication between the weighted affinity matrix and the original task features to obtain the refined leg movement features. and sleep position characteristics : ； ； Step 5-2-4: The refined features are passed to their respective feature reconstruction modules to obtain the reconstructed feature maps; the feature reconstruction modules are specifically: The feature reconstruction module is an upsampling module designed to gradually restore the spatial resolution of the feature map through deconvolution operations. During the convolution operation, the input is padded by 1 pixel. After upsampling, ordinary convolution is used to eliminate the interpolation artifacts caused by deconvolution. The ReLU activation function is used to introduce a nonlinear transformation to suppress irrelevant information and alleviate the gradient vanishing problem. The feature reconstruction module performs two deconvolution-convolution operations. The first deconvolution layer first doubles the spatial size of the input feature map and halves the number of channels, and then performs a convolution operation while keeping the number of channels unchanged. The second deconvolution layer doubles the feature map again, restores the number of channels to the initial state, and then performs a convolution operation while keeping the number of channels unchanged. Step 5-3: Input the reconstructed feature map into the classifier to achieve the final classification and recognition, and output the leg movement type. Types of sleeping positions ; Among them, leg movements Includes two types of periodic leg movements, PLM1 and PLM2, double leg twitching, leg crossing, leg extension and flexion, rolling over, and non-leg movement NLM; sleep position types This includes supine (Supine), side-lying (Side), prone (Prone), and non-sleeping position (LM).

8. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 3 or 7, characterized in that, The deep network model described in step 5-1 is specifically as follows: The deep learning model includes a shared feature extraction layer, two task-specific classification layers, and a dynamic weight averaging layer; the two task-specific classification layers include a leg movement classification task layer and a sleep position classification task layer. The structure of the shared feature extraction layer is as follows: (1) ResNet18 is used for pre-training. The initial layer is a 7×7 convolutional layer with a stride of 2, followed by a 3×3 max pooling layer with a stride of 2. Then, it passes through four residual groups in sequence. Each residual group contains two basic residual blocks, and each residual block contains two 3×3 convolutional layers. After the last residual group, a global average pooling layer is used to transform the feature map into a fixed-length feature vector. (2) Remove the final fully connected layer in the pre-trained ResNet18 and retain only its convolution and pooling layers to extract features from the input multi-view time-frequency map and time-distance map respectively. (3) Add an SE block to each view of the time-frequency map and the time-distance map to perform channel attention calibration on the extracted feature map. Then, add the three input signals of the time-frequency map and the time-distance map with weights to obtain the time-frequency map and time-distance map features after multi-view fusion. (4) The time-frequency map and time-distance map after multi-view fusion are calibrated by an SE block, and then the two features are spliced ​​together to obtain the multi-view time-frequency and time-distance fusion features. (5) The multi-view time-frequency-time-distance fusion features are passed through a fully connected layer to output the classification result, which is used as the initial training result; (6) By using the bottleneck layer, which includes two 1×1 convolutional layers, the number of channels is reduced from 1024 to 512 and 128 respectively, further compressing the feature information; (7) Finally, the feature map is flattened by global average pooling; In the leg movement classification task layer, the 128-dimensional features are first reduced to 64-dimensional features using a fully connected layer with ReLU activation function, and then another fully connected layer is used to output 6-class prediction results. In the sleep posture classification task layer, the same method as the leg movement classification task layer is used to output 3 types of prediction results; In the dynamic weighted averaging layer, the loss for each task is adjusted using a dynamic weighting strategy. The weight for the current task is calculated based on the classification loss from the previous round to balance the training of the two tasks. The formula for calculating the dynamic weight is: ； in, ； In the formula, leg movement classification and sleeping posture classification correspond to respectively Take 1 and 2, Representative task The weight, The task is in the Losses during the march, It is the first The training speed is calculated step-by-step; a larger value indicates a faster decrease in loss, which allows for a smaller weight, thus reducing the proportion of loss for that task. T is a temperature constant; when T = 1... Without the effect of weight adjustment, the larger T is, the more stable the weight adjustment becomes. After obtaining the dynamic weights, the overall loss L can be calculated. ； Finally, based on the weighted network parameters, the leg movement category and sleeping posture category are output in the leg movement classification layer and the sleeping posture classification layer, respectively. When the signal processing module determines that it is in the non-leg movement segment NLM, the leg movement prediction result is directly output as NLM. The sleep posture prediction does not go through the multi-task learning network and is directly output by the sleep posture classifier.

9. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 1, characterized in that, The sleep periodic limb movement sequence determination module determines whether the leg movement output meets the characteristics of periodic limb movement. If it does, it marks the periodic leg movement sequence, specifically including: Step 6-1, determine leg movement output Is it 0 or 1? If so, mark this data segment as a periodic leg movement PLM. Step 6-2: When the continuity of the periodic leg movement PLM meets the preset conditions, it is considered a set of periodic leg movement sequences PLMS; at the same time, the sleeping posture output is saved. .

10. The non-contact, periodic sleep limb movement disorder and sleep posture joint monitoring device at the foot of the bed according to claim 9, characterized in that, The sleep-periodic limb movement disorder (SPLM) determination module determines whether sleep-periodic limb movement disorder exists, specifically including: Step 7-1: Count the number of times the periodic leg movement sequence (PLMS) occurs during the entire night's sleep, and divide the number by the total sleep time to obtain the periodic limb movement index (PLMI). Step 7-2: Determine whether sleep-periodic limb movement disorder exists. If the periodic limb movement index (PLMI) is >5 / h for children and >15 / h for adults, then sleep-periodic limb movement disorder exists; otherwise, sleep-periodic limb movement disorder does not exist.

Images