A passive temperature monitoring method based on millimeter wave sensing
Through the passive temperature monitoring method based on millimeter wave perception, the electronic tags and fine-grained authentication model of cholesterol materials are used to solve the problem of battery requirements in traditional wireless temperature sensors, and passive temperature monitoring and thermal image output are realized, with high accuracy and low energy consumption characteristics.
Patent Information
- Application Number
- CN202110409700.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-04-16
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2041-04-16
AI Technical Summary
Traditional wireless temperature sensors require batteries or energy acquisition equipment, which limits the acquisition of temperature information and cannot achieve passive temperature monitoring and thermal imaging image output.
Using a passive temperature monitoring method based on millimeter wave perception, high-precision millimeter wave probe transmits frequency-modulated continuous waves against cholesterol material electronic tags, receives reflected signals, performs fast Fourier transformation and motion compensation, and extracts the displacement of cholesterol material electronic tags. Information, and through clutter suppression and feature extraction, a fine-grained authentication model is constructed to achieve temperature perception.
It realizes the acquisition of passive single-point temperature and thermal imaging images, provides high-precision temperature monitoring, and average perception accuracy reaches more than 99%. It is suitable for portable and low-energy hardware systems.
Smart Images

Figure CN113176001B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless technology, and in particular to a passive temperature monitoring method based on millimeter wave sensing. Background Art
[0002] Since traditional temperature sensing systems are usually based on the thermoelectric effect, wireless temperature sensors usually need to be equipped with batteries or energy harvesting devices to achieve unlimited temperature sensing and monitoring. Due to the development of electronic equipment technology, wireless temperature sensors are ubiquitous in electronic equipment around the world. However, the output of the sensor is usually only a single point temperature, which limits more temperature information resources. Millimeter wave sensing in the field of wireless technology can achieve a passive temperature monitoring system and can simultaneously output single point temperature and thermal imaging images. Summary of the invention
[0003] In order to solve the problem in the background technology, the present invention provides a passive temperature monitoring method based on millimeter wave (mmWave) sensing.
[0004] The technical solution adopted by the present invention is as follows:
[0005] A passive temperature monitoring method based on millimeter wave sensing includes the following steps:
[0006] Step 1: Use a high-precision millimeter wave probe to emit a frequency modulated continuous wave (FMCW) to the surface of the cholesterol material electronic tag. This sensing is non-contact, and then receive the frequency modulated continuous wave with the cholesterol material molecular arrangement information reflected by the cholesterol material electronic tag; the millimeter wave probe converts the periodic linear frequency modulation signal of the received frequency modulated continuous wave into an intermediate frequency (IF) signal, and performs fast Fourier transform and motion compensation on the intermediate frequency signal to obtain the displacement information of the cholesterol material electronic tag;
[0007] The cholesterol material electronic tag in the step 1 is a film-based temperature sensor made of cholesterol material.
[0008] The specific process of obtaining the displacement information of the cholesterol material electronic tag according to the intermediate frequency signal in step 1 is as follows:
[0009] 1.1) Assume a linear frequency modulation period is T r , according to the intermediate frequency signal, the cholesterol material electronic label displacement d(mT) of the mth linear frequency modulation cycle is obtained. r ), the specific calculation formula is:
[0010]
[0011] Wherein, m is the number of the linear frequency modulation cycle, m = 0, 1, 2…, M-1; M is the number of linear frequency modulation cycles; f0 is the carrier frequency; Δψ m It is the result of fast Fourier transform (FFT) of the intermediate frequency signal of the mth linear frequency modulation cycle;
[0012] Thus, the displacement of the cholesterol material sensor of M linear frequency modulation cycles is obtained;
[0013] Δψ m Contains the distance information of the mth linear frequency modulation cycle.
[0014] 1.2) The random motion of the cholesterol material electronic tag carrier is the main obstacle to obtaining accurate cholesterol molecular arrangement data. When the electronic tag motion amplitude is greater than half of the range profile resolution, it will be difficult to accurately calculate Δψ m In order to avoid the random motion of the cholesterol material carrier interfering with the result of step 1.1), the displacement of the cholesterol material electronic tag of the M linear frequency modulation cycles obtained in step 1.1) is compensated by translation, that is, the distance information of the M linear frequency modulation cycles is calibrated:
[0015] Definition S m (l) is the distance information of the mth linear frequency modulation period obtained in step 1.1), is the distance information of the mth linear frequency modulation cycle after calibration;
[0016] Wherein, l = 0, 1, 2, .., L-1, L is the number of sampling points in the mth linear frequency modulation period, and l is the number of the sampling point;
[0017] Define Q m (l) is the relative distance information of the mth linear frequency modulation period, specifically:
[0018]
[0019] Among them, Q m-1 (l) is the relative distance information of the m-1th linear frequency modulation cycle;
[0020] The relative distance information Q m (l) is the distance information after calibration The distance information S obtained in step 1.1) m (l) The relative distance between them.
[0021] Define X m For S m (l) translation distance, is the optimal translation distance, according to the relative distance information Q m (l) Calculate the optimal translation distance Optimal translation distance It is obtained by maximizing the following formula:
[0022]
[0023] Among them, S m-1 (lX m ) represents the distance information after the m-1th linear frequency modulation period is shifted;
[0024] The distance information after calibration of the mth linear frequency modulation cycle is:
[0025]
[0026] Where j is a unit complex number, Δ is a vector [0,1,…,L-1];
[0027] The distance information after calibration of M linear frequency modulation cycles is obtained according to the distance information after calibration of the mth linear frequency modulation cycle. The distance information after calibration of M linear frequency modulation cycles is the displacement information of the cholesterol material electronic tag after motion compensation.
[0028] Step 2: clutter suppression processing: perform a second fast Fourier transform on the result of the fast Fourier transform of the intermediate frequency signal of multiple linear frequency modulation cycles in step 1 to obtain a range Doppler matrix, and complete the separation of background clutter and the removal of dynamic clutter by updating the range Doppler matrix;
[0029] The step 2 is specifically as follows:
[0030] 2.1) Separation of background clutter:
[0031] We regard background clutter as being caused by a large number of small objects. The amplitude and phase of these background clutter are random, but their spectral envelope a follows the Weibull distribution:
[0032]
[0033] Therefore they can perform clutter separation as follows.
[0034] Perform a second fast Fourier transform on the range information after the calibration of the M linear frequency modulation cycles in step 1 to obtain a range Doppler matrix (RDM), and then perform logarithmic regularization on the range Doppler matrix;
[0035] Substitute the set clutter separation rate into the definition formula of the clutter separation rate to calculate the threshold μ 0 , the clutter separation rate is set manually, and the clutter separation rate definition formula is:
[0036]
[0037] Where γ is the Euler constant, p cis the clutter separation rate;
[0038] Defining the elasticity matrix Each element in the elastic matrix The definition is as follows:
[0039]
[0040] in, represents the range-Doppler matrix, is the jth element in the i-th row of the elasticity matrix, R ij is the jth element in the i-th row of the range-Doppler matrix, is an unbiased estimate,
[0041] Range Doppler matrix Update to complete the background clutter separation. The specific process of the update is as follows:
[0042]
[0043] J is a 01 matrix, is the Hadamard product, and sgn is the sign function.
[0044] 2.2) Remove dynamic clutter caused by dynamic obstacles detected by FMCW:
[0045] Since dynamic clutter does not follow Weibull distribution, the above steps cannot remove these dynamic clutters. The present invention uses information across multiple RDMs to eliminate dynamic clutter.
[0046] Continuously transmit frequency modulated continuous waves to obtain n continuous range Doppler matrices. Each range Doppler matrix contains M linear frequency modulation cycles. The average moving speed of obstacles in multiple range Doppler matrices is calculated. The values corresponding to obstacles with an average moving speed greater than the range profile resolution ΔRES are removed from the range Doppler matrix, thereby achieving the removal of dynamic clutter.
[0047] The value corresponding to the obstacle is the moving speed of the obstacle.
[0048] Step 3: Temperature feature extraction:
[0049] Extract the frequency modulation features and frequency modulation detail features describing the molecular arrangement of the cholesterol material and the corresponding temperature information from the range Doppler matrix after removing the clutter in step 2;
[0050] 3.1) Extraction of FM detail features of cholesterol materials: FM signals represent the molecular arrangement of cholesterol material electronic tags and related geometric parameters. The molecular arrangement of cholesterol materials contains real-time temperature information. The residual phase cepstrum coefficients (RPCC) are used to represent the phase information of the potential excitation waveform. The Teager phase cepstrum coefficients (TPCC) are used to measure the spectral amplitude characteristics of the temperature information corresponding to the molecular arrangement of cholesterol materials.
[0051] 3.2) Cholesterol material frequency modulation feature extraction: The cholesterol material electronic tag frequency modulation features used in the present invention include spectral features (such as centroid, energy band, peak, flatness and entropy), Mel frequency cepstral coefficients (MFCC), linear prediction coefficient (LPC), linear prediction cepstral coefficient (LPCC) and line spectrum frequency (LSP).
[0052] Step 4: Construct a fine-grained authentication model, input the frequency modulation features and frequency modulation detail features extracted in step 3 into the fine-grained authentication model to obtain the temperature information perceived by the molecular arrangement of the cholesterol material
[0053] The fine-grained authentication model in step 4 includes a feature selection module based on Fisher Score and a temperature-aware integrated classifier;
[0054] 1) The specific function of the feature selection module based on Fisher score is to use the cutting plane algorithm to select the feature subset with the largest Fisher score in the original features;
[0055] The original features are the FM features and FM detail features extracted in step 3;
[0056] 2) The integrated classifier is obtained by fusion of three models: Gaussian mixture model - general background model, support vector machine and hidden Markov model;
[0057] The specific function of the integrated classifier is to input the feature subset with the largest Fisher score in the original features into the integrated classifier, use the regression model to adjust the weights of the three models of the integrated classifier, and perform weighted summation on the output results of the three models according to the adjusted weights. Finally, the classification result, the single-point temperature and thermal imaging image corresponding to the molecular arrangement of the cholesterol material, is output, thereby completing temperature perception.
[0058] Training of fine-grained authentication model: Select cholesterol material electronic label samples with known single-point temperature and thermal imaging images, extract the temperature characteristics of the cholesterol material electronic label as a sample data set according to steps 1 to 3, and divide the sample data set into a training set and a test set to train the fine-grained authentication model.
[0059] The training optimization of the cutting plane algorithm in the feature selection module uses the ridge regression algorithm and the gradient descent algorithm.
[0060] A passive temperature monitoring system based on millimeter wave sensing includes a hardware module and a software module. The hardware module includes a millimeter wave sensing module of a cholesterol material electronic tag and a clutter suppression processing module, and the software module includes a temperature feature extraction module and a temperature sensing module.
[0061] The beneficial effects of the present invention are:
[0062] The present invention uses millimeter waves to directionally acquire and analyze the molecular arrangement of cholesterol materials, and realizes an end-to-end temperature sensing system from millimeter wave information acquisition, signal processing, feature extraction to single-point temperature and thermal imaging.
[0063] The present invention realizes a passive temperature monitoring system based on a portable, low-energy hardware system, including a passive temperature-sensing electronic tag and a mobile wireless sensing module. It can not only obtain single-point temperature and thermal imaging images in a passive way, but also can be easily migrated to the wireless sensing module of mobile devices such as smart phones. BRIEF DESCRIPTION OF THE DRAWINGS
[0064] Figure 1 It is a system structure diagram of the method of the present invention.
[0065] Figure 2 It is a specific implementation process diagram of the method of the present invention. DETAILED DESCRIPTION
[0066] The present invention will be further described in detail below in conjunction with the accompanying drawings.
[0067] like Figure 1 As shown, the present invention includes a hardware module and a software module. The hardware module includes a millimeter wave sensing module and a clutter suppression processing module of a cholesterol material electronic tag, and the software module includes a temperature feature extraction module and a temperature sensing module.
[0068] We use cholesterol material to make electronic tags, which have a side length of 15 cm and a thickness of about 0.01 cm. The electronic tags are attached to the side walls of the cabinet. Then, we aim the millimeter wave probe in the direction of the electronic tag, and the measurement distance starts from 25 cm and gradually increases to 8 meters. The measured temperature range is -10 degrees to 50 degrees.
[0069] The specific implementation process is as follows Figure 2 As shown:
[0070] 1. The millimeter wave probe converts the received periodic linear frequency modulation signal of the frequency modulated continuous wave into an intermediate frequency (IF) signal, and calculates the displacement information of the cholesterol material electronic tag according to the intermediate frequency (IF) signal;
[0071] 1.1) Assume a linear frequency modulation period is T r , the electronic tag displacement d(mT) of the mth linear frequency modulation cycle is obtained according to the intermediate frequency (IF) signal r ) may be calculated as;
[0072]
[0073] Where m is the number of the linear frequency modulation cycle, m = 0, 1, ..., M-1; M is the number of linear frequency modulation cycles, f 0 is the carrier frequency; Δψ m It is the result of fast Fourier transform (FFT) of the intermediate frequency signal of the mth linear frequency modulation cycle;
[0074] 1.2) The random motion of the cholesterol material electronic tag carrier is the main obstacle to obtaining accurate cholesterol material molecular arrangement data. When the carrier motion amplitude is greater than half of the range profile resolution, it will be difficult to accurately calculate Δψ m In order to avoid the influence of random motion of the carrier on the result of step 1.1) and to obtain accurate Fourier transform results of IF signal, we use a fine-grained distance bin alignment method to solve this problem.
[0075] Definition S m (l) is the mth distance information obtained, where m = 0, 1, ..., M-1, l = 0, 1, ..., L-1, M is the number of signal sequences, and L is the number of distance samples. The value represents the distance after calibration. m (l) The translation distance.
[0076] Define the relative distance information Q of the mth distance information m (l) to make past calibration information available as knowledge. m (l) The formal definition is:
[0077]
[0078] The optimal translation distance is obtained by maximizing the following formula
[0079]
[0080] Finally, the calibrated distance information can be obtained:
[0081]
[0082] where Δ is the vector [0,1,…,L-1].
[0083] 2. After obtaining accurate Fourier transform data, we use the elastic algorithm to remove the background clutter that conforms to the Weibull distribution. The Fourier transform result of the IF signal is subjected to a second fast Fourier transform to obtain the range Doppler matrix (RDM). Then the logarithmic regularized Doppler matrix is used to separate the clutter.
[0084] Substitute the artificially set clutter separation rate into the definition formula of the clutter separation rate to calculate the threshold μ 0 , the clutter separation rate is defined as:
[0085]
[0086] Where γ is the Euler constant, p c is the clutter separation rate;
[0087] Define an elasticity matrix represents the Doppler matrix.
[0088] Range Doppler matrix To update, the specific process is as follows:
[0089]
[0090] J is a 01 matrix, is the Hadamard product, and sgn is the sign function.
[0091] Finally, the range Doppler matrix (RDM) after clutter separation can be obtained. However, since the dynamic clutter detected by FMCW does not follow the Weibull distribution, the above steps cannot remove these dynamic clutter. Then, the information across multiple RDMs is used to eliminate dynamic clutter. The range Doppler matrix obtained by the quadratic Fourier transform contains velocity information. We calculate the average moving speed of the targets in multiple Doppler matrices, remove objects whose speed is greater than the range profile resolution, and thus achieve the removal of dynamic clutter.
[0092] 3. Temperature feature extraction: extract frequency modulation features (RPCC, TPCC) and frequency modulation detail features (spectral features, MFCC, LPC, LPCC, LSP).
[0093] 4. Build a fine-grained authentication model
[0094] The fine-grained authentication model includes a feature selection module based on Fisher Score and a temperature-aware ensemble classifier;
[0095] Training of fine-grained authentication model: Select cholesterol material electronic label samples with known single-point temperature and thermal imaging images, extract the temperature characteristics of cholesterol material electronic labels as sample data sets, use the Zero Cross Rate and Root Mean Square in segmentation algorithm to divide the data sets into hundreds of thousands of samples, randomly divide these samples into training sets and test sets, and train the fine-grained authentication model.
[0096] Temperature perception: First, the cutting plane algorithm is used to select the subset with the largest Fisher score in the original features. The ridge regression algorithm and the gradient descent algorithm are used in the optimization process of the cutting plane algorithm. Finally, the algorithm fusion model of the Gaussian mixture model-universal background model (GMM-UBM), support vector machine and hidden Markov model is used for classification. We use the regression model to adjust the weights of the three models, and use the weighted summation for classification, output the single-point temperature and thermal imaging corresponding to the molecular arrangement of the cholesterol material, and finally realize the perception of temperature.
[0097] According to tests, the average perception accuracy of the present invention for single-point temperature can be maintained above 99% within an error range of 1 degree Celsius, and the average perception accuracy of the thermal imaging image can be maintained above 99% within an error range of 3 degrees Celsius.
Claims
1. A passive temperature monitoring method based on millimeter wave sensing, comprising the following steps: Step 1: Millimeter wave sensing of cholesterol material electronic tags A millimeter wave probe is used to transmit a frequency modulated continuous wave to the surface of the cholesterol material electronic tag, and then the frequency modulated continuous wave with the cholesterol material molecular arrangement information reflected by the cholesterol material electronic tag is received; the millimeter wave probe converts the periodic linear frequency modulation signal of the received frequency modulated continuous wave into an intermediate frequency signal, and performs fast Fourier transform and motion compensation on the intermediate frequency signal to obtain the displacement information of the cholesterol material electronic tag; In the step 1, the specific process of obtaining the displacement information of the cholesterol material electronic tag according to the intermediate frequency signal is as follows: 1.1) Assume a linear frequency modulation period is T r , according to the intermediate frequency signal, the cholesterol material electronic label displacement d(mT) of the mth linear frequency modulation cycle is obtained. r ), the specific calculation formula is: Where m is the number of the linear frequency modulation cycle, m = 0, 1, 2…, M-1; M is the number of linear frequency modulation cycles; f0 is the carrier frequency; Δψ m It is the result of fast Fourier transform of the intermediate frequency signal of the mth linear frequency modulation cycle; Thus, the electronic label displacement of the cholesterol material of M linear frequency modulation cycles is obtained; 1.2) In order to avoid the random motion of the cholesterol material electronic tag carrier interfering with the result of step 1.1), the cholesterol material electronic tag displacement of the M linear frequency modulation cycles obtained in step 1.1) is motion compensated by translation change, that is, the distance information of the M linear frequency modulation cycles is calibrated: Definition S m (l) is the electronic label displacement of cholesterol material in the mth linear frequency modulation cycle obtained in step 1.1), is the distance information of the mth linear frequency modulation cycle after calibration; Wherein, l = 0, 1, 2, .., L-1, L is the number of sampling points in the mth linear frequency modulation period, and l is the number of the sampling point; Define Q m (l) is the relative distance information of the mth linear frequency modulation period, specifically: Among them, Q m-1 (l) is the relative distance information of the m-1th linear frequency modulation cycle; Define X m For S m (l) translation distance, is the optimal translation distance, according to the relative distance information Q m (l) Calculate the optimal translation distance Optimal translation distance It is obtained by maximizing the following formula: Among them, S m-1 (lX m ) represents the distance information after the m-1th linear frequency modulation period is shifted; The distance information after calibration of the mth linear frequency modulation cycle is: Where j is a unit complex number, Δ is a vector [0,1,…,L-1]; According to the distance information after calibration of the mth linear frequency modulation cycle, the distance information after calibration of M linear frequency modulation cycles is obtained, and then according to the distance information after calibration of M linear frequency modulation cycles, the displacement information of the cholesterol material electronic tag after motion compensation is obtained; Step 2: clutter suppression processing: Performing a second fast Fourier transform on the result of the fast Fourier transform of the intermediate frequency signal of multiple linear frequency modulation cycles in step 1 to obtain a range Doppler matrix, and completing the separation of background clutter and the removal of dynamic clutter by updating the range Doppler matrix; Step 3, temperature feature extraction: extract the frequency modulation features and frequency modulation detail features describing the molecular arrangement of the cholesterol material and the corresponding temperature information from the range Doppler matrix after removing the clutter in step 2; 3.1) Extraction of FM detail features of cholesterol materials: FM detail features are residual phase cepstrum coefficients and Teager phase cepstrum coefficients; residual phase cepstrum coefficients are used to represent the phase information of the potential excitation waveform; Teager phase cepstrum coefficients are used to measure the spectral amplitude characteristics of the temperature information corresponding to the molecular arrangement of cholesterol materials; 3.2) FM feature extraction of cholesterol material: FM features are spectral features, Mel frequency cepstral coefficients, linear prediction coefficients, linear prediction cepstral coefficients and line spectrum frequencies. Spectral features are composed of centroid, energy band, peak, flatness and entropy. Step 4: Temperature Sensing: A fine-grained authentication model is constructed, and the frequency modulation features and the frequency modulation detail features extracted in step 3 are input into the fine-grained authentication model to obtain the temperature information sensed by the molecular arrangement of the cholesterol material.
2. The passive temperature monitoring method based on millimeter wave sensing according to claim 1 is characterized in that: The cholesterol material electronic tag in the step 1 is a film-based temperature sensor made of cholesterol material.
3. The passive temperature monitoring method based on millimeter wave sensing according to claim 1 is characterized in that: The step 2 is specifically as follows: 2.1) Separation of background clutter: Perform a second fast Fourier transform on the range information after the M linear frequency modulation cycles are calibrated in step 1 to obtain a range Doppler matrix, and then perform logarithmic regularization on the range Doppler matrix; Substitute the set clutter separation rate into the definition formula of the clutter separation rate to calculate the threshold μ0. The definition formula of the clutter separation rate is: Where γ is the Euler constant, p c is the clutter separation rate; Defining the elasticity matrix Each element in the elastic matrix The definition is as follows: in, represents the range-Doppler matrix, is the jth element in the i-th row of the elasticity matrix, R ij is the jth element in the i-th row of the range-Doppler matrix, is an unbiased estimate, Range Doppler matrix Update to complete the background clutter separation; 2.2) Remove dynamic clutter caused by dynamic obstacles detected by FMCW: Continuously transmit frequency modulated continuous waves to obtain n continuous range Doppler matrices. Each range Doppler matrix contains M linear frequency modulation cycles. The average moving speed of obstacles in multiple range Doppler matrices is calculated. The values corresponding to obstacles with an average moving speed greater than the range profile resolution ΔRES are removed from the range Doppler matrix, thereby achieving the removal of dynamic clutter.
4. The passive temperature monitoring method based on millimeter wave sensing according to claim 3 is characterized in that: The range-Doppler matrix The specific process of updating is as follows: J is a 01 matrix, is the Hadamard product, and sgn is the sign function.
5. The passive temperature monitoring method based on millimeter wave sensing according to claim 1 is characterized in that: The fine-grained authentication model in step 4 includes a feature selection module based on Fisher's score and a temperature-aware integrated classifier; 1) The specific function of the feature selection module based on Fisher score is to use the cutting plane algorithm to select the feature subset with the largest Fisher score in the original features; The original features are the FM features and FM detail features extracted in step 3; 2) The integrated classifier is obtained by fusion of three models: Gaussian mixture model - general background model, support vector machine and hidden Markov model; The specific function of the integrated classifier is: input the feature subset with the largest Fisher score in the original features into the integrated classifier, use the regression model to adjust the weights of the three models of the integrated classifier, perform weighted summation on the output results of the three models according to the adjusted weights, and finally output the classification result to complete temperature perception. The output classification result is the single-point temperature and thermal imaging image corresponding to the molecular arrangement of the cholesterol material.
6. The passive temperature monitoring method based on millimeter wave sensing according to claim 5 is characterized in that: The training optimization of the cutting plane algorithm in the feature selection module uses the ridge regression algorithm and the gradient descent algorithm.
