A method for detecting based on nonlinear frequency modulation microwave excitation induced thermal wave imaging

Abstract

The application provides a nonlinear frequency modulation microwave excitation induced thermal wave imaging detection method, which comprises the following steps: constructing a composite excitation module of a Gaussian light beam and a dynamic phase compensation, decomposing high-frequency / low-frequency sub-pulse alternate emission, synchronously collecting infrared non-uniform sampling and microwave polarization diversity data, fusing features through a time-domain CNN and a frequency-domain graph neural network, combining a thermal diffusion equation model to iteratively correct propagation parameters, and realizing defect and deep structure detection. Through the composite excitation structure combined with the high-frequency / low-frequency sub-pulse alternate emission, the application significantly improves the micro-defect detection resolution and the deep structure penetration ability, the multi-modal data acquisition and the cross-modal neural network fusion enhance the feature robustness, the depth compensation model and the Kalman filter realize the detection depth self-adaptive optimization, the terahertz spectrum coupling matrix and the Landweber iterative inversion are used to improve the parameter accuracy, and finally, an advanced detection system with efficient excitation, multi-dimensional perception and intelligent analysis is formed.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal wave imaging detection, and in particular to a thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction. Background Art

[0002] With the rapid development of modern industry, the demand for non-destructive testing of materials and structures is increasing. The continuous improvement of thermal theory, the development of thermal imaging technology, and the continuous progress of computer technology and signal processing technology have laid the foundation for the birth and development of thermal wave imaging detection technology, making it possible to use the differences in the thermal properties of objects to detect internal defects and structural features. This technology can quickly and non-contactly detect large areas, and has shown great application potential in many fields such as aerospace, automobile manufacturing, construction engineering, composite materials testing, etc., and has gradually become an important technical means in the field of non-destructive testing.

[0003] Traditional thermal wave imaging detection methods are prone to uneven energy distribution and local thermal damage due to laser and microwave excitation, which limits the detection depth of biological tissues; traditional signal processing technology is constrained by noise interference and difficulties in multi-scale feature fusion, affecting the accuracy of defect identification; although high-frequency excitation has high resolution, the detection depth is insufficient, and low-frequency excitation is the opposite, making it difficult to take into account both deep and micro-defect detection; changes in environmental humidity and temperature significantly affect the thermal wave propagation characteristics of the material, resulting in unstable detection results; the multi-physical field coupling effect increases the complexity of the model, making it difficult to accurately predict complex responses.

[0004] Therefore, it is necessary to design a thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation to solve the problems of existing thermal wave imaging detection methods being susceptible to noise interference, insufficient detection depth, and unstable detection results. Summary of the Invention

[0005] In view of this, the present invention proposes a thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation induction, aiming to solve the problems of existing thermal wave imaging detection methods, such as uneven laser and microwave excitation energy, thermal damage, and limited detection depth of biological tissues; traditional signal processing is interfered by noise and other factors, affecting the accuracy of defect recognition; high and low frequency excitation frequencies have their own shortcomings, making it difficult to take into account both deep and micro defects; ambient temperature and humidity affect the stability of the results; multi-physical field coupling increases model complexity, making it difficult to accurately predict complex responses.

[0006] In one aspect, the present invention proposes a thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation, comprising:

[0007] Based on the nonlinear relationship between the dielectric constant and thermal diffusivity of the material, a composite excitation structure consisting of a Gaussian beam shaping module and a microwave phase compensation module was constructed. The Gaussian beam adopts non-uniform amplitude modulation technology, and the microwave phase compensation is dynamically adjusted based on the real-time dielectric constant inversion results.

[0008] The nonlinear frequency modulation signal is decomposed into a high-frequency sub-pulse sequence and a low-frequency sub-pulse sequence, which are transmitted alternately at a time interval calculated by the thermal diffusion time constant. The high-frequency sub-pulse is used for micro-defect detection, and the low-frequency sub-pulse is used to penetrate deep structures.

[0009] Deploy array infrared detectors and microwave reflectivity sensors to synchronously collect data on thermal wave radiation intensity, phase difference, and dielectric constant changes. The infrared detectors use a non-uniform sampling mode, and the microwave sensors are equipped with polarization diversity receiving units.

[0010] Constructing a parallel processing structure of a time-domain convolutional neural network and a frequency-domain graph neural network. The time-domain network processes the dynamic response sequence of thermal waves, and the frequency-domain network analyzes the spectral characteristics of microwave reflection coefficients. The time-domain convolutional neural network and the frequency-domain graph neural network perform cross-modal feature fusion through an attention mechanism.

[0011] Based on the Green's function solution of the heat diffusion equation, a nonlinear mapping model of detection depth and frequency response is established. Combined with real-time temperature gradient data, the attenuation coefficient and phase delay parameters of the thermal wave propagation path are iteratively corrected.

[0012] Furthermore, electromagnetic parameter data of the material is obtained by a terahertz spectrometer, and a coupling relationship matrix between dielectric constant and thermal diffusivity is established, and the matrix is ​​used as an input parameter of the Gaussian beam shaping module;

[0013] The non-uniform amplitude modulation technology redistributes the energy density of the Gaussian beam into an annular spot by controlling the phase distribution of the spatial light modulator, wherein the inner diameter of the annular spot is equal to the penetration depth of the surface wave of the material;

[0014] The microwave phase compensation module collects the material surface reflection coefficient data in real time, uses the least squares method to invert the current dielectric constant distribution, generates a compensation signal that matches the electromagnetic properties of the material through a phase modulator, and superimposes it on the output end of the microwave excitation source.

[0015] Furthermore, after the high-frequency sub-pulse is emitted, the infrared detector array completes temperature field acquisition within an interval of 1 / 10 of the thermal diffusion time constant, and then the low-frequency sub-pulse is emitted. At this time, the Gaussian beam shaping module switches to a uniform spot mode.

[0016] Furthermore, the array infrared detector adopts a checkerboard non-uniform sampling mode, and each detector unit works alternately at a preset interval and generates sparse temperature field data;

[0017] Before the sparse temperature field data is input into the frequency domain graph neural network, it is converted into a frequency domain feature graph through a two-dimensional discrete cosine transform;

[0018] The polarization diversity receiving unit of the microwave reflection coefficient sensor synchronously collects the reflection coefficient data of vertical polarization and horizontal polarization, and forms a polarization characteristic spectrum after fast Fourier transformation.

[0019] Furthermore, the time-domain convolutional neural network processes the temperature change sequence of the infrared detector, the frequency-domain graph neural network analyzes the microwave polarization characteristic spectrum, and the time-domain convolutional neural network and the frequency-domain graph neural network achieve feature alignment through a cross-modal attention mechanism;

[0020] The output feature map of the time domain convolutional neural network is multiplied channel by channel with the polarization feature spectrum of the frequency domain graph neural network to generate a fusion feature matrix.

[0021] Furthermore, the theoretical propagation depths of heat waves of different frequencies are calculated based on the Green's function model, and a mapping relationship matrix between spatial position and frequency is generated. This matrix is ​​used as the edge weight parameter of the frequency domain graph neural network to construct the adjacency matrix of the graph convolution layer.

[0022] The defect location probability map output by the time-domain convolutional neural network is weightedly fused with the depth compensation matrix on a pixel-by-pixel basis. Each pixel value of the defect location probability map is multiplied by the depth compensation coefficient of the corresponding position. The depth compensation coefficient is calculated based on the attenuation coefficient and phase delay parameter obtained by inverting the heat diffusion equation. The fused result is input into the classifier for defect type determination.

[0023] Furthermore, after the high-frequency sub-pulse is emitted, the Kalman filter algorithm is used to predict the current depth of the heat wave propagation based on the real-time collected temperature gradient data. This predicted depth is used as the basis for adjusting the low-frequency sub-pulse emission parameters.

[0024] When the predicted depth is less than the preset threshold, the transmission power of the low-frequency sub-pulse is increased and the center frequency is reduced; otherwise, the power is reduced and the frequency is increased;

[0025] The penetration depth of the low-frequency sub-pulse is verified by inversion of the heat diffusion equation. The inversion process adopts the improved Landweber iterative algorithm, and the iteration step size is dynamically adjusted by the temperature gradient data of the infrared detector.

[0026] Furthermore, the microwave reflection coefficient sensor collects dielectric constant distribution data of the material surface in real time, processes the dielectric constant distribution data through a singular value decomposition algorithm, and inputs the processing result into the Gaussian beam shaping module;

[0027] The Gaussian beam shaping module adjusts the phase distribution of the spatial light modulator according to the dielectric constant distribution, and controls the inner diameter of the annular spot within the range of ±10μm;

[0028] The temperature uniformity of the heating area is monitored by an infrared detector array. If the temperature gradient exceeds a set threshold, the microwave phase compensation module is triggered to recalculate the compensation signal.

[0029] Compared with the prior art, the beneficial effect of the present invention is that the thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation induced thermal wave imaging detection of the present invention realizes Gaussian beam energy focusing and microwave phase dynamic compensation through a composite excitation structure, and combines the high-frequency / low-frequency sub-pulse alternating emission mechanism to significantly improve the detection resolution and penetration ability of material micro-defects and deep structures. Multimodal data acquisition (non-uniform infrared sampling + polarization diversity microwave) and cross-modal neural network fusion technology effectively enhance the robustness of feature extraction, while the depth compensation model based on the heat diffusion equation and the Kalman filter dynamic adjustment strategy realize the adaptive optimization of the detection depth and the precise correction of the attenuation parameters. In addition, the terahertz spectral coupling matrix, Landweber iterative inversion and other technologies further improve the dielectric constant inversion accuracy and temperature field uniformity, and finally form an advanced detection system integrating efficient excitation, multi-dimensional perception and intelligent analysis, which has significant technical advantages in the field of non-destructive testing of composite materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:

[0031] Figure 1 This is a flow chart of a thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation in accordance with an embodiment of the present invention; DETAILED DESCRIPTION

[0032] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art. It should be noted that, unless there is a conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with the implementation regulations.

[0033] Reference Figure 1As shown, in some embodiments of the present application, a thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation induction includes:

[0034] Based on the nonlinear relationship between the dielectric constant and thermal diffusivity of the material, a composite excitation structure consisting of Gaussian beam shaping and microwave phase compensation was constructed. The Gaussian beam was modulated using non-uniform amplitude modulation, and the microwave phase compensation was dynamically adjusted based on the real-time dielectric constant inversion results.

[0035] The nonlinear frequency modulation signal is decomposed into a high-frequency (1-10 GHz) and a low-frequency (0.1-1 GHz) sub-pulse sequence, which is emitted alternately at time intervals calculated by the thermal diffusion time constant. The high-frequency sub-pulse is used for micro-defect detection, while the low-frequency sub-pulse penetrates deep structures.

[0036] Deploy array infrared detectors and microwave reflectivity sensors to synchronously collect data on thermal wave radiation intensity, phase difference, and dielectric constant changes. The infrared detectors use a non-uniform sampling mode, and the microwave sensors are equipped with polarization diversity receiving units.

[0037] A parallel processing structure of a time-domain convolutional neural network and a frequency-domain graph neural network is constructed. The time-domain network processes the dynamic response sequence of thermal waves, and the frequency-domain network analyzes the spectral characteristics of microwave reflection coefficients. The two are then integrated through an attention mechanism for cross-modal feature fusion.

[0038] Based on the Green's function solution of the heat diffusion equation, a nonlinear mapping model of detection depth and frequency response is established. Combined with real-time temperature gradient data, the attenuation coefficient and phase delay parameters of the thermal wave propagation path are iteratively corrected.

[0039] Specifically, based on the nonlinear relationship between the dielectric constant and thermal diffusivity of the material, a composite excitation structure consisting of Gaussian beam shaping and microwave phase compensation is constructed, in which the Gaussian beam adopts non-uniform amplitude modulation technology, and the microwave phase compensation is dynamically adjusted based on the real-time dielectric constant inversion results. In specific implementation, the electromagnetic parameter data of the material is first obtained through a terahertz spectrometer, and a coupling relationship matrix between the dielectric constant and thermal diffusivity is established. This matrix serves as the input parameter of the Gaussian beam shaping module. The non-uniform amplitude modulation technology redistributes the energy density of the Gaussian beam into an annular spot by controlling the phase distribution of the spatial light modulator. The inner diameter of the ring is equal to the penetration depth of the surface wave of the material. The microwave phase compensation module collects the surface reflection coefficient data of the material in real time, uses the least squares method to invert the current dielectric constant distribution, and generates a compensation signal that matches the electromagnetic properties of the material through the phase modulator, which is superimposed on the output end of the microwave excitation source.

[0040] It can be understood that the composite excitation structure constructed above can utilize the nonlinear relationship between the dielectric constant and thermal diffusivity of the material, redistribute the energy density into a specific annular spot through non-uniform amplitude modulation of the Gaussian beam, match the surface wave penetration depth of the material, and dynamically adjust the phase based on real-time dielectric constant inversion in combination with microwave phase compensation, thereby improving the detection accuracy and stability of the thermal wave imaging detection method induced by nonlinear frequency modulation microwave excitation, effectively overcoming the shortcomings of traditional methods, and more accurately detecting internal defects of materials.

[0041] Specifically, the Gaussian beam shaping module dynamically adjusts the inner diameter of the annular spot according to the dielectric constant distribution output by the microwave phase compensation module, so that the laser energy concentration area is consistent with the propagation direction of the microwave heat wave. The time-sharing multi-frequency excitation sequence separates the thermal response of the surface layer of the material and the deep defects on the time axis by controlling the emission timing of high-frequency sub-pulses (1-10GHz) and low-frequency sub-pulses (0.1-1GHz). In specific implementation, after the high-frequency sub-pulse is emitted, the infrared detector array is triggered to complete the temperature field acquisition within an interval of 1 / 10 of the thermal diffusion time constant, and then the low-frequency sub-pulse is emitted. At this time, the Gaussian beam shaping module switches to a uniform spot mode to avoid interference of the high-frequency heating area with the low-frequency signal.

[0042] It can be understood that the Gaussian beam shaping module dynamically adjusts the inner diameter of the annular spot based on the dielectric constant distribution of the microwave phase compensation module, so that the laser energy concentration area is consistent with the propagation direction of the microwave thermal wave. The time-sharing multi-frequency excitation sequence separates the thermal responses of defects at different depths of the material by controlling the emission timing of high and low frequency sub-pulses, and implements low-frequency emission and spot switching after medium and high frequency acquisition to avoid interference, significantly improving the detection accuracy and resolution of the thermal wave imaging detection method induced by nonlinear frequency modulation microwave excitation, and can accurately locate defects at different depths.

[0043] Specifically, the array infrared detector uses a checkerboard non-uniform sampling pattern, with each detector unit working alternately at preset intervals to generate sparse temperature field data. Before inputting this data into the frequency domain graph neural network, it is first converted into a frequency domain feature map via a two-dimensional discrete cosine transform to reduce the data dimension. The polarization diversity receiving unit of the microwave sensor synchronously collects reflection coefficient data for vertical and horizontal polarization, which is then transformed into a polarization characteristic spectrum after fast Fourier transformation. The time domain convolutional neural network processes the temperature change sequence of the infrared detector, while the frequency domain graph neural network analyzes the microwave polarization characteristic spectrum. The two achieve feature alignment through a cross-modal attention mechanism. In specific implementation, the output feature map of the time domain network is point-multiplied with the polarization characteristic spectrum of the frequency domain network on a channel-by-channel basis to generate a fused feature matrix.

[0044] It can be understood that the checkerboard non-uniform sampling pattern of the array infrared detector is combined with the frequency domain graph neural network. The temperature field data dimension is compressed through the two-dimensional discrete cosine transform. The microwave polarization diversity receiving unit synchronously collects the dual-polarization reflection coefficient and converts it into a characteristic spectrum. After being processed by the time domain convolution network and the frequency domain graph network respectively, the feature alignment and fusion are realized through the cross-modal attention mechanism, which effectively improves the utilization efficiency of multimodal data based on the nonlinear frequency modulation microwave excitation induced thermal wave imaging detection method, enhances the ability to identify complex defects, reduces redundant information interference, and significantly improves the detection accuracy and anti-environmental noise performance.

[0045] Specifically, the depth compensation algorithm calculates the theoretical propagation depths of thermal waves of different frequencies based on the Green's function model, generating a mapping matrix between spatial position and frequency. This matrix serves as the edge weight parameter for the frequency-domain graph neural network and is used to construct the adjacency matrix of the graph convolution layer. The defect location probability map output by the time-domain convolutional neural network is then weightedly fused with the depth compensation matrix on a pixel-by-pixel basis. In implementation, each pixel value in the probability map is multiplied by the depth compensation coefficient for the corresponding position. This coefficient is calculated from the attenuation coefficient and phase delay parameter derived from the inversion of the heat diffusion equation. The fused result is then input into a classifier to determine the defect type.

[0046] It can be understood that the depth compensation algorithm constructs a mapping relationship matrix between spatial position and frequency through the Green's function model, which is used as the adjacency matrix parameter of the frequency domain graph neural network. It effectively integrates the defect probability map output by the time domain convolutional network and the depth compensation coefficient inverted by the heat diffusion equation, adjusts the defect position confidence pixel by pixel, enhances the feature expression ability of defects at different depths, and significantly improves the defect depth positioning accuracy and type recognition accuracy based on the nonlinear frequency modulation microwave excitation induced thermal wave imaging detection method, solving the misjudgment problem caused by depth information loss and attenuation effect in traditional methods.

[0047] Specifically, after a high-frequency sub-pulse is emitted, a dynamic depth compensation algorithm uses a Kalman filter to predict the maximum depth of the current thermal wave propagation based on real-time temperature gradient data. This predicted depth serves as the basis for adjusting the emission parameters of the low-frequency sub-pulse. In practice, if the predicted depth is less than a preset threshold, the transmission power of the low-frequency sub-pulse is increased and the center frequency is reduced; otherwise, the power is reduced and the frequency is increased. The penetration depth of the low-frequency sub-pulse is verified by inverting the heat diffusion equation. This inversion process utilizes a modified Landweber iterative algorithm, with the iteration step size dynamically adjusted by the infrared detector's temperature gradient data. The inversion results are fed back to the excitation sequence module to optimize the sub-pulse parameters for the next time period.

[0048] It can be understood that the dynamic depth compensation algorithm predicts the thermal wave propagation depth in real time through Kalman filtering, dynamically adjusts the power and frequency parameters of the low-frequency sub-pulse, and optimizes the balance between detection depth and resolution; combined with the improved Landweber algorithm to invert and verify the penetration depth and feedback the optimized excitation sequence, it significantly improves the adaptability of the thermal wave imaging detection method induced by nonlinear frequency modulation microwave excitation, effectively solves the problem that traditional fixed-frequency excitation is difficult to take into account both deep and micro defect detection, and enhances the recognition accuracy and detection stability of defects of different depths in complex structures.

[0049] Specifically, a microwave reflectivity sensor collects real-time data on the dielectric constant distribution of the material surface. This data is processed using a singular value decomposition algorithm and then fed into the Gaussian beam shaping module of the electromagnetic-thermal coupling excitation model. The shaping module adjusts the phase distribution of the spatial light modulator based on the dielectric constant distribution, precisely controlling the inner diameter of the annular spot within a ±10μm range. The temperature uniformity of the heated area is simultaneously monitored using an infrared detector array. If the temperature gradient exceeds a set threshold, the microwave phase compensation module is triggered to recalculate the compensation signal, forming a closed-loop optimization process. The response time of this closed-loop process is determined by the time constant of the heat diffusion equation, ensuring that the heating process is synchronized with the thermal response characteristics of the material.

[0050] It should be noted that:

[0051] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known structures and technologies are not shown in detail so as not to obscure the understanding of this description.

[0052] Furthermore, those skilled in the art will appreciate that although some embodiments described herein include certain features and not other features included in other embodiments, the combination of features from different embodiments is meant to be within the scope of this application and to form different embodiments.

[0053] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A thermal wave imaging detection method based on nonlinear frequency-modulated microwave excitation, characterized in that: include: Based on the nonlinear relationship between the dielectric constant and thermal diffusivity of the material, a composite excitation structure consisting of a Gaussian beam shaping module and a microwave phase compensation module was constructed. The Gaussian beam adopts non-uniform amplitude modulation technology, and the microwave phase compensation is dynamically adjusted based on the real-time dielectric constant inversion results. The nonlinear frequency modulation signal is decomposed into a high-frequency sub-pulse sequence and a low-frequency sub-pulse sequence, which are transmitted alternately at a time interval calculated by the thermal diffusion time constant. The high-frequency sub-pulse is used for micro-defect detection, and the low-frequency sub-pulse is used to penetrate deep structures. Deploy array infrared detectors and microwave reflectivity sensors to synchronously collect data on thermal wave radiation intensity, phase difference, and dielectric constant changes. The infrared detectors use a non-uniform sampling mode, and the microwave sensors are equipped with polarization diversity receiving units. Constructing a parallel processing structure of a time-domain convolutional neural network and a frequency-domain graph neural network. The time-domain convolutional neural network processes the dynamic response sequence of thermal waves, and the frequency-domain graph neural network analyzes the spectral characteristics of microwave reflection coefficients. The time-domain convolutional neural network and the frequency-domain graph neural network perform cross-modal feature fusion through an attention mechanism. Based on the Green's function solution of the heat diffusion equation, a nonlinear mapping model of detection depth and frequency response is established. Combined with real-time temperature gradient data, the attenuation coefficient and phase delay parameters of the thermal wave propagation path are iteratively corrected.

2. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: Acquiring electromagnetic parameter data of the material through a terahertz spectrometer, establishing a coupling relationship matrix between dielectric constant and thermal diffusivity, and using the matrix as an input parameter of the Gaussian beam shaping module; The non-uniform amplitude modulation technology redistributes the energy density of the Gaussian beam into an annular spot by controlling the phase distribution of the spatial light modulator, wherein the inner diameter of the annular spot is equal to the penetration depth of the surface wave of the material; The microwave phase compensation module collects the material surface reflection coefficient data in real time, uses the least squares method to invert the current dielectric constant distribution, generates a compensation signal that matches the electromagnetic properties of the material through a phase modulator, and superimposes it on the output end of the microwave excitation source.

3. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: After the high-frequency sub-pulse is emitted, the infrared detector array completes temperature field acquisition within an interval of 1 / 10 of the thermal diffusion time constant, and then the low-frequency sub-pulse is emitted. At this time, the Gaussian beam shaping module switches to a uniform spot mode.

4. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: The array infrared detector adopts a checkerboard non-uniform sampling mode, and each detector unit works alternately at a preset interval and generates sparse temperature field data; Before the sparse temperature field data is input into the frequency domain graph neural network, it is converted into a frequency domain feature graph through a two-dimensional discrete cosine transform; The polarization diversity receiving unit of the microwave reflection coefficient sensor synchronously collects the reflection coefficient data of vertical polarization and horizontal polarization, and forms a polarization characteristic spectrum after fast Fourier transformation.

5. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: The time-domain convolutional neural network processes the temperature change sequence of the infrared detector, the frequency-domain graph neural network analyzes the microwave polarization characteristic spectrum, and the time-domain convolutional neural network and the frequency-domain graph neural network achieve feature alignment through a cross-modal attention mechanism; The output feature map of the time domain convolutional neural network is multiplied channel by channel with the polarization feature spectrum of the frequency domain graph neural network to generate a fusion feature matrix.

6. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: The theoretical propagation depths of heat waves of different frequencies are calculated based on the Green's function model, and a mapping matrix between spatial position and frequency is generated. This matrix is ​​used as the edge weight parameter of the frequency domain graph neural network to construct the adjacency matrix of the graph convolution layer. The defect location probability map output by the time-domain convolutional neural network is weightedly fused with the depth compensation matrix on a pixel-by-pixel basis. Each pixel value of the defect location probability map is multiplied by the depth compensation coefficient of the corresponding position. The depth compensation coefficient is calculated based on the attenuation coefficient and phase delay parameter obtained by inverting the heat diffusion equation. The fused result is input into the classifier for defect type determination.

7. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: After the high-frequency sub-pulse is emitted, the Kalman filter algorithm is used to predict the current depth of the heat wave propagation based on the real-time collected temperature gradient data. This predicted depth is used as the basis for adjusting the low-frequency sub-pulse emission parameters. When the predicted depth is less than the preset threshold, the transmission power of the low-frequency sub-pulse is increased and the center frequency is reduced; otherwise, the power is reduced and the frequency is increased; The penetration depth of the low-frequency sub-pulse is verified by inversion of the heat diffusion equation. The inversion process adopts the improved Landweber iterative algorithm, and the iterative step size is dynamically adjusted by the temperature gradient data of the infrared detector.

8. The thermal wave imaging detection method based on nonlinear frequency modulation microwave excitation induction according to claim 1 is characterized in that: The microwave reflection coefficient sensor collects dielectric constant distribution data of the material surface in real time, processes the dielectric constant distribution data through a singular value decomposition algorithm, and inputs the processing result into the Gaussian beam shaping module; The Gaussian beam shaping module adjusts the phase distribution of the spatial light modulator according to the dielectric constant distribution, and controls the inner diameter of the annular spot within the range of 10 μm; The temperature uniformity of the heating area is monitored by an infrared detector array. If the temperature gradient exceeds a set threshold, the microwave phase compensation module is triggered to recalculate the compensation signal.

Images