A liquid interference removal method and electronic device based on PINN

Abstract

The present invention discloses a liquid interference removal method and electronic device based on PINN, belonging to the field of signal processing technology, the method includes: collecting radar spectra of cold-blooded living body life signals and liquid interference signals, making a standard data set; constructing a liquid interference signal model and a physical loss function in weak life signal recognition; constructing a PINN network that resists liquid interference, and using the data set for training to generate an anti-interference model, so as to realize the detection and distinction of cold-blooded living body life signals and liquid interference signals in weak life signal recognition. The present invention can effectively solve the problem of misreporting liquid signals as weak life signals in engineering applications, effectively improve the accuracy of cold-blooded living body biological signal recognition and the generalization ability of the overall network, and meet the weak life signal detection needs in complex environments under security inspection scenarios.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and more specifically to a PINN-based liquid interference removal method and electronic device; the method is mainly used in biological signal detection and analysis, especially in the effective processing of liquid interference signals when using millimeter wave radar technology to detect weak life signals in security inspection scenarios. Background Art

[0002] With the rapid development of millimeter-wave radar technology, its application in biological signal detection, environmental monitoring, safety detection and other fields is becoming more and more extensive. However, in practical applications, millimeter-wave radar is often disturbed by liquid flow when detecting micro-motion signals, resulting in reduced detection accuracy of weak life signals. Especially when detecting cold-blooded organisms (such as insects), liquid interference signals may be mistakenly identified as life signals, affecting the reliability of the detection results.

[0003] At present, traditional signal processing methods often rely on empirical rules and threshold judgments when processing liquid interference signals, lacking in-depth exploration and understanding of complex signal features. The existing technology has some challenges and problems in removing liquid signal interference from cold-blooded living organism radar spectrum signals, mainly including the following points: 1. Insufficient signal feature extraction. The existing methods have no targeted solutions for distinguishing between cold-blooded living organism signal radar spectrum and liquid signal radar spectrum. The current liquid feature extraction relies on manually designed features and fails to fully utilize the physical properties of liquid flow and complex signal characteristics, resulting in information loss and low signal processing efficiency. 2. The current solution based on liquid micro-motion signal recognition has a high computational complexity when processing signals and often requires fitting, which makes it difficult to meet the needs of real-time processing, limiting its application effect in dynamic environments. In addition, the objects of recognition of traditional millimeter-wave radar life recognition systems are often warm-blooded animals such as humans and dogs. There are some technical gaps when it comes to distinguishing liquid signals from cold-blooded living signals; 3. There is no public data set for radar data collection of cold-blooded living objects and liquid interference items. Therefore, it is difficult for traditional anti-interference networks to remove interference from the liquid signals described in this case, resulting in a high possibility that the interference signal is mistakenly reported as a life signal, thereby reducing the credibility and practicality of the system.

[0004] Therefore, there is an urgent need for a new signal processing solution that can combine physical knowledge with deep learning technology to accurately identify and remove liquid interference signals, thereby improving the detection capability of weak life signals. Summary of the invention

[0005] In view of this, the present invention provides a PINN-based liquid interference removal method and electronic device that solves at least some of the above-mentioned technical problems. By combining physical knowledge with deep learning technology, liquid interference signals can be accurately identified and removed, thereby improving the detection capability of weak life signals. It effectively solves the deficiencies in the prior art and provides a new idea and solution to meet the needs of high-precision weak life signal detection in practical applications.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] In a first aspect, the present invention provides a method for removing liquid interference based on PINN, the method comprising the following steps:

[0008] S1: Collect radar spectra of cold-blooded living body life signals and liquid interference signals to create a standard data set;

[0009] S2: Construct the liquid interference signal model and physical loss function in weak life signal recognition;

[0010] S3: Construct a PINN network that is resistant to liquid interference, and use the data set for training to generate an anti-interference model, and use the anti-interference model to distinguish between cold-blooded living life signals and liquid interference signals in weak life signal recognition.

[0011] Preferably, in step S1, a millimeter-wave dual radar is used to collect signal data in multiple directions; and signal data collection is performed in a random interference environment.

[0012] Preferably, in step S2, the constructed liquid interference signal model is used to describe the liquid motion state, attenuation characteristics and damping characteristics, and its formal expression includes:

[0013] ;

[0014] Where P is the fluid pressure, is the fluid density, v is the fluid flow rate, and C is a constant;

[0015] ;

[0016] Where A(t) is the signal amplitude at time t, A0 is the initial signal amplitude, is the attenuation coefficient, e is the base of the natural logarithm function;

[0017] ;

[0018] Where m is the equivalent mass of the liquid, c is the viscous damping coefficient, k is the restoring force in the flow, d is the differential operator, and x is the displacement calculated from the original signal;

[0019] ;

[0020] In the formula, is the mass-independent equivalent restoring force, is the frequency.

[0021] Preferably, in step S2, the physical loss function constructed is:

[0022] ;

[0023] In the formula, represents physical loss, N is the total number of samples, is the partial differential symbol.

[0024] Preferably, in step S3, constructing a PINN network resistant to liquid interference includes: a preprocessing layer, a residual block, a Transformer module, a Bernoulli equation module, a damping effect module and a comprehensive loss module.

[0025] Preferably, in step S3, the performance of the PINN network is enhanced by adding physical information, an adaptive normalization factor, and applying a Pre-layer physical inspiration layer to the PINN network loss function.

[0026] Preferably, the comprehensive loss module is used to calculate the total loss value, and the calculation formula is:

[0027] ;

[0028] ;

[0029] Where L is the total loss, , is the adaptive normalized loss coefficient, For data loss, For physical damage; is the phase difference between two time points, v i is the velocity component of the phase change of the ith data point, Δt is the time interval, λ is the wavelength, and x i is the position of the ith data point.

[0030] In a second aspect, the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement the above-mentioned PINN-based liquid interference removal method.

[0031] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0032] In view of the problems in the prior art that liquid interference signals are difficult to effectively remove during detection, the characteristic similarity between liquid-enclosed signals and weak life signals leads to misjudgment, and the model cannot fully utilize the liquid dynamic characteristics, the present invention provides a stable, reliable and efficient anti-liquid interference solution for weak life signal identification, which effectively solves the problem of liquid signals being misreported as weak life signals in engineering applications, effectively improves the recognition accuracy of weak life signals (especially cold-blooded living biological signals) and the generalization ability of the overall network, and meets the needs of weak life signal detection in complex environments under security inspection scenarios.

[0033] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0034] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0037] Figure 1 A schematic flow chart of a PINN-based liquid interference removal method provided by the present invention.

[0038] Figure 2 This is a schematic diagram of dual radar signal acquisition provided by the present invention.

[0039] Figure 3 A schematic diagram of a visual representation of the PINN anti-liquid interference network provided by the present invention.

[0040] Figure 4 This is a schematic diagram of the structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0041] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0042] In the description of the present invention, it should be noted that: in some processes described in the specification and drawings of this application, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be performed in the order in which they appear in this document or may be performed in parallel. In addition, various serial numbers are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0043] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] See also Figure 1 As shown, the embodiment of the present invention provides a liquid interference removal method based on PINN (physical information-based neural network), which collects target life signals (cold-blooded living body signals) and interference signals (liquid signals) through millimeter wave radar; sends the liquid interference signal to the PINN anti-interference network of the host computer for training and generates an anti-interference model; sends the target life signal and the interference signal to the anti-interference model together to complete the distinction between the two, realize the anti-liquid interference signal detection for weak life signal recognition, and improve the accuracy of judging whether the signal contains a living body signal. The method mainly includes the following steps:

[0045] 1: Dataset preparation: Use the millimeter wave radar board to collect 9000 groups of cold-blooded live body signals (without training) and 4200 groups of liquid signals. The collected liquid samples include viscous liquid samples and non-viscous samples. Rigid containers and non-rigid containers are used to carry liquids during collection, and the liquid capacity is changed to simulate the security inspection scene to inspect liquid packages and increase the integrity of the data. In an embodiment of the present invention, when the millimeter wave radar collects data, dual radars are used to collect 270-degree package data to increase the information collection range; the radar signal data collected under different background environmental conditions is stored as a csv file type. The specific content of the csv file is: the row represents the time axis, the column represents the position change, the phase and amplitude can be read from each row and column, and 80% of the liquid data set is randomly taken as the training set, 10% as the verification set, and 10% as the test set to facilitate subsequent data training.

[0046] 2: Design a liquid interference signal model for weak life signal recognition: For liquids in packages in security inspection scenarios, the project goal is to detect liquid signals in millimeter-wave radar signals rather than accurately describe the complex flow behavior inside the liquid. The Bernoulli equation is used as a condition to capture the liquid flow characteristics; the phase and amplitude changes of radar echoes in liquids are used as one of the conditions for distinguishing signal characteristics: the phase shift of the millimeter-wave radar signal caused by the liquid medium is obtained through the Doppler effect; the amplitude transformation of the radar signal is obtained using the damping effect of the liquid, so that the relevant parameters of the liquid dynamics equation are used as physical constraints.

[0047] 3: Design of liquid-interference-resistant PINN network: Use three different parts to enhance PINN performance; specifically: Add physical information to the loss function to ensure that the network fits radar data and extracts features under the premise of following the physical laws of liquid dynamics; Adaptive normalization factor, adaptively adjust the relative weights of each loss term in the loss function to ensure that each loss term has a reasonable contribution during the training process; Pre-layer physical inspiration layer design, introduce a preprocessing layer before entering the main network, add a damped sine wave-based feature extraction module to the input layer, and simulate the analytical solution of liquid flow; The entire network uses the transformer architecture combined with resnet to achieve efficient recognition of millimeter-wave radar anti-interference liquid data.

[0048] The specific implementation manner and working principle of the method of the present invention are described in detail below:

[0049] In step 1, radar spectra of cold-blooded living body vital signals and liquid interference signals are collected to create a standard data set.

[0050] In a specific implementation, two radars are arranged as follows during acquisition: Figure 2 In the direction shown, 270-degree effective radar spectrum was collected, 9,000 groups of cold-blooded live body signals and 4,200 groups of liquid signals were collected. The cold-blooded live body samples collected are the customs biological entry categories. The collected liquid samples include viscous liquid samples and non-viscous samples. Rigid containers and non-rigid containers are used to carry liquids during collection, and the liquid capacity is changed to simulate the security inspection scene to inspect liquid packages, thereby increasing the integrity of the data. The interference data collected under different background environmental conditions is stored as a csv file type. The specific content of the csv file is: rows represent the time axis, columns represent position changes, and the phase and amplitude can be read from each row and column. 80% of the data set is randomly selected as the training set, 10% as the validation set, and 10% as the test set to facilitate subsequent data training.

[0051] In step 2, a liquid interference signal model for weak life signal detection is designed, and the physical constraint equation is designed through the radar spectrum echo signal.

[0052] In a specific embodiment, the fluid dynamics equation Bernoulli equation (1) is used to describe the liquid motion state. For the liquid signal in the security inspection scenario, it is approximately assumed that the height difference h between the initial position of the liquid (on the conveyor belt) and the detection position is small, especially in the case of short-distance transmission. Therefore, h is ignored and the Bernoulli equation is simplified to equation (2):

[0053] (1);

[0054] (2);

[0055] Where P is the fluid pressure, is the fluid density, v is the fluid flow rate, and C is a constant. During the flow of liquid, the relationship between flow rate and pressure reflects the state of motion of the fluid. For example, when the flow rate of the liquid increases, the pressure will decrease accordingly, and vice versa. Millimeter wave radar can capture the tiny phase and amplitude changes of the liquid during the flow process. By constraining the physical loss in the network, the model can make the recognition of liquid signals more accurate.

[0056] In one embodiment, the empirical formula (3) is used to describe the specific attenuation characteristics that liquid signals usually exhibit, so as to help the model distinguish the gradually weakening signal characteristics during liquid flow, while life signals usually do not have such significant attenuation.

[0057] (3);

[0058] Where A(t) is the signal amplitude at a certain moment, A0 is the initial signal amplitude, is the attenuation coefficient.

[0059] In a specific embodiment, when the liquid flows in a restricted manner in a bottle, due to factors such as the friction between the bottle wall and the liquid and the interaction of the liquid in the bottle, the signal will exhibit damping characteristics of attenuation or slow change. A viscous damping model, i.e., formula (4), is introduced to detect the phase and amplitude changes in the signal.

[0060] (4);

[0061] Where m is the equivalent mass of the liquid, c is the viscous damping coefficient, k is the restoring force in the flow, and x is the displacement calculated from the original signal. Considering that the damping characteristics of different liquids may be different, the damping coefficient c can be set as a learnable parameter to allow the network to automatically adjust and match different types of liquid signals. The mass-independent equivalent restoring force can be directly expressed by the square of the frequency, as shown in Equation (5).

[0062] (5);

[0063] In the formula, is the mass-independent equivalent restoring force, is the frequency.

[0064] In a specific embodiment, the mass m and density m proposed in the above formula are As for the liquid in the black box of the security inspection scene, it is impossible to directly give the restoring force k. Set the feature mapping (amplitude attenuation feature A(t), frequency change feature ), so that the network uses the characteristics to establish the Bernoulli equation and damping equation constraints, establish the physical loss function, and use the restoring force term, damping term and flow characteristic term to constrain the network, as shown in formula (6).

[0065] (6);

[0066] Where, L Physics represents physical loss, N is the total number of samples, is the partial differential symbol.

[0067] In step 3, a PINN network LiquidPhysicsInformedNN that is resistant to liquid interference is designed, and the interference data and weak life signal data are put into the network for training, and the best_model is trained to achieve the separation of cold-blooded living data and liquid interference signals.

[0068] In a specific implementation, the designed LiquidPhysicsInformedNN mainly includes six parts: PreprocessingLayer, ResNetblock, TransformerModule, BernoulliPINN, DampedSineBubblePINN and TotleLoss. Its network structure is as follows: Figure 3 As shown in Figure 2, the LiquidPhysicsInformedNN network combines physics-inspired information and deep learning techniques to ensure effective fitting of radar data while following the physical laws of liquid dynamics.

[0069] In one specific implementation, in the PreprocessingLayer, the maximum amplitude of the signal is first extracted by torch.max (AmplitudeExtraction), and the maximum value is found from the input signal sequence to ensure that only the intensity information of the signal is retained. Subsequently, torch.fft.fft (FrequencyExtraction) is used to perform a fast Fourier transform (FFT) to convert the time domain signal to the frequency domain to obtain the main frequency component. The attenuation coefficient of the signal is calculated using scipy.optimize.curve_fit (DecayCoefficientExtraction) attenuation coefficient extraction function, and the attenuation characteristics of the signal over time are analyzed by exponential fitting. Use torch.cat to merge the extracted signal features and output them as a tensor of (batch_size, num_features) (batch size, feature number size).

[0070] In a specific implementation, in the ResNetBlock (residual block) structure, the first convolution layer Conv1 is included: the convolution kernel size is 3 (kernel_size=3), the stride is 1 (stride=1), the padding is 1 (padding=1), and the bias is set to True. This layer is used to extract local features, and the output feature map dimension is (batch_size, num_filters, sequence_length); the batch normalization layer BatchNorm1D1 uses nn.BatchNorm1d to normalize the output of the convolution layer to accelerate training and improve the stability of the model; the ReLU activation function is applied, and nonlinear features are introduced through nn.ReLU() to enhance the expressiveness of the model; the second convolution layer Conv2: the convolution kernel size is 3 (kernel_size=3), the stride is 1 (stride=1), the padding is 1 (padding=1), and the bias is True to further extract features. After the output feature map is processed by BatchNorm1D2 (batch normalization again), the input signal is added to the output of ResNetBlock (residual block) through the residual connection to prevent information loss and enhance the learning ability of the network to form the final output feature, thereby deepening the network depth and alleviating the gradient vanishing problem.

[0071] In a specific embodiment, the Transformer module structure includes MultiHeadAttention (multi-head self-attention mechanism, using nn.MultiheadAttention (class of multi-head attention mechanism)) and FeedForwardNetwork (feedforward neural network) to capture long-term dependencies and sequence features in the signal. The TransformerModule parameters are set to 8 heads (num_heads=8), the dimension of each head is 64 (head_dim=64), and the output dimension is 512 (embed_dim=512); the output of the self-attention mechanism is passed to FeedForwardNetwork: it is a series of fully connected layers (constructed by nn.Sequential), including two linear transformations, the activation function is ReLU, the output dimension of the first layer is 2048 (hidden_dim=2048), and the output dimension of the second layer is the input feature dimension. To prevent overfitting, LayerNormalization is used in this module to normalize the data to ensure the stability of the training process. In a specific embodiment, in the subspace module (Subspace) structure: it includes a UNetConvBlock (UNet convolution block), the input data first passes through a 3*3 convolution layer, and the activation function usually uses LeakyReLU to introduce nonlinearity. The core of the subspace module is to remap the input feature map to a low-dimensional representation space, thereby reducing noise sensitivity and making the feature map have a higher information density and a more effective representation space.

[0072] In one specific implementation, the BernoulliPINN module focuses on the processing of fluid dynamics: InputLayer receives the relative velocity obtained by phase change, passes through FullyConnectedLayer, the input dimension of the first layer is 2, the output dimension is set to 128, and the activation function is ReLU. The output dimension of the second layer is set to 2, representing the predicted flow rate and pressure respectively, and no activation function is used to maintain the accuracy of the predicted value.

[0073] In one embodiment, the DampedSinePINN module processes the prediction of the liquid amplitude, receives input through the InputLayer and enters the FullyConnectedLayer, which consists of two fully connected layers. The input dimension of the first fully connected layer is 2, the output dimension is set to 128, and the activation function is ReLU (nn.ReLU()) to introduce nonlinear features. The output dimension of the second fully connected layer is set to 1, indicating the predicted amplitude (Amplitude). The output layer does not use an activation function to keep the output range wide.

[0074] In one specific implementation, the TotalLoss module combines the various losses and calculates the total loss value: AdaptiveNormalizationFactor is used to dynamically adjust the weights of each loss item (the loss items include main loss, physical loss, and regularization loss). Specifically, AdaptiveNormalizationFactor adjusts the weights of each loss item according to the ratio of the current loss value to the historical average value to improve the effect and convergence speed of network training; CombinedLoss combines damped_sine_loss with bernoulli_loss and data_loss to form a total loss. This total loss is averaged by torch.mean() for optimization. The optimizer uses Adam (torch.optim.Adam) and implements parameter updates by setting an appropriate learning rate (lr). The details are as follows:

[0075] (7);

[0076] (8);

[0077] Where L is the total loss, , is the adaptive normalized loss coefficient, For data loss, For physical damage; is the phase difference between two time points, v i is the velocity of the ith data point (the velocity component of the phase change), Δt is the time interval used to calculate the increment of the phase change over time, λ is the wavelength used to relate the phase change to the velocity, and x i is the position of the ith data point.

[0078] In a specific implementation, in TrainingProcess, the argparse parsing block is used to parse the training parameters, and the data loading, transformation, and training processes are set. Specifically, the preprocessed dataset is loaded using torch.utils.data.DataLoader (class), and transform is applied for data enhancement. In each training cycle, the loss function and optimizer are called, and the model parameters are updated using back propagation (loss.backward) to achieve efficient learning of features and continuous improvement of model performance.

[0079] From the description of the above embodiments, those skilled in the art can know that the main technical means and technical advantages adopted by the present invention are as follows:

[0080] 1. Self-made cold-blooded living weak life data set and interference liquid data set, using millimeter-wave dual radar to collect data in multiple directions, accurately collect effective signals, and reduce the interference caused by signal collection errors; set up multi-variable and multi-type samples to enrich interference signals: collect different types of living signals under random interference environments; collect viscous liquid and non-viscous liquid signals in rigid containers and non-rigid containers, and control the liquid capacity to collect rich interference signals; there are 9000 groups of cold-blooded living signals and 4200 groups of liquid signals in the experimental data.

[0081] 2. A liquid interference signal model for weak life signal recognition is designed. The liquid radar spectrum signal in the package under security inspection is used to constrain the network training process using liquid dynamics equations. The Bernoulli equation is used to capture the liquid flow characteristics, and the damping effect is used to capture the amplitude and phase transformation of the liquid signal radar echo, which specifically solves the problem of false positives in weak life signal recognition caused by the single liquid signal fitting in traditional detection methods.

[0082] 3. A PINN network resistant to liquid interference was designed: by adding physical information to the PINN network loss function, the network is ensured to fit radar data and extract features under the premise of following the physical laws of liquid dynamics; through the adaptive normalization factor, the relative weights of each loss term in the loss function are adaptively adjusted to ensure that each loss term has a reasonable contribution during the training process; through the application of the Pre-layer physical inspiration layer, a preprocessing layer is introduced before entering the main network, and a feature extraction module based on damped sine waves is added to the input layer to simulate the analytical solution of liquid flow to enhance the network learning ability; the entire network uses the transformer architecture combined with resnet to enhance the model's recognition accuracy and efficiency for liquid signals and weak life signals in complex scenarios.

[0083] Advantages: In view of the problems in the existing technology that liquid interference signals are difficult to effectively remove during detection, the characteristic similarity between liquid-enclosed signals and weak life signals leads to misjudgment, and the model cannot fully utilize the liquid dynamic characteristics, the present invention provides a stable, reliable and efficient anti-liquid interference solution for the recognition of weak life signals by millimeter-wave radar, which effectively solves the problem of misreporting liquid signals as weak life signals in engineering applications, effectively improves the recognition accuracy of weak life signals (especially cold-blooded living biological signals) and the generalization ability of the overall network, and meets the needs of weak life signal detection in complex environments under security inspection scenarios.

[0084] Further, refer to Figure 4 As shown, an embodiment of the present invention further provides an electronic device, which may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, and the processor executes the computer program to implement a PINN-based liquid interference removal method in the above method embodiment.

[0085] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions of the electronic device and processes data by running or executing programs or modules stored in the memory 11, and calling data stored in the memory 11.

[0086] The electronic device provided in the embodiment of the present invention produces the same technical effect as the aforementioned method embodiment. For the sake of brief description, for parts not mentioned in this embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment, which will not be repeated here.

[0087] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, devices or computer program products, etc. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] It should be noted that the word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several distinct components, and by means of a suitably programmed computer.

[0089] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0090] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A liquid interference removal method based on PINN, characterized in that: The method comprises the following steps: S1: Collect radar spectra of cold-blooded living body life signals and liquid interference signals to create a standard data set; S2: Construct a liquid interference signal model and a physical loss function in weak life signal recognition; the constructed liquid interference signal model is used to describe the liquid motion state, attenuation characteristics and damping characteristics, and its formal expression includes: In the formula, P is the fluid pressure, ρ is the fluid density, v is the fluid flow rate, and C is a constant; A(t)=A0e -αt ; Where A(t) is the signal amplitude at time t, A0 is the initial signal amplitude, α is the attenuation coefficient, and e is the base of the natural logarithm function: Where m is the equivalent mass of the liquid, c is the viscous damping coefficient, k is the restoring force in the flow, d is the differential operator, and x is the displacement calculated from the original signal; k eff =f 2 ; In the formula, k eff is the equivalent restoring force without mass dependence, f is the frequency; The constructed physical loss function is: Where, L Physics represents physical loss, N is the total number of samples, is the symbol for partial differential; S3: Construct a PINN network that is resistant to liquid interference, and use the data set for training to generate an anti-interference model, and use the anti-interference model to distinguish between cold-blooded living life signals and liquid interference signals in weak life signal recognition.

2. A PINN-based liquid interference removal method according to claim 1, characterized in that: In step S1, the millimeter-wave dual radar is used to collect signal data in multiple directions; and the signal data is collected in a random interference environment.

3. A method for removing liquid interference based on PINN according to claim 1, characterized in that: In step S3, a PINN network resistant to liquid interference is constructed, including: a preprocessing layer, a residual block, a Transformer module, a Bernoulli equation module, a damping effect module, and a comprehensive loss module.

4. The method for removing liquid interference based on PINN according to claim 3, characterized in that: In step S3, the performance of the PINN network is enhanced by adding physical information, an adaptive normalization factor, and applying a Pre-layer physical inspiration layer to the PINN network loss function.

5. The method for removing liquid interference based on PINN according to claim 3, characterized in that: The comprehensive loss module is used to calculate the total loss value, and the calculation formula is: L=λ1L data +λ2L physics ; Where L is the total loss, λ1 and λ2 are adaptive normalized loss coefficients, and L data is the data loss, L physics is the physical loss; Δφ i is the phase difference between two time points, v i is the velocity component of the phase change of the ith data point, Δt is the time interval, λ is the wavelength, and x i is the position of the ith data point.

6. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores machine executable instructions that can be executed by the processor, and the processor executes the machine executable instructions to implement a PINN-based liquid interference removal method as described in any one of claims 1-5.

Images