A method and device for measuring the surface temperature of an infrared optical element irradiated by a laser

Abstract

This invention discloses a method for measuring the surface temperature of an infrared optical element irradiated by laser, belonging to the field of infrared thermometry technology. The method includes the following steps: using an infrared optical lens to receive the thermal radiation emitted from the surface of the infrared optical element after laser irradiation and imaging it onto the modulation surface of a spatial light modulator; the spatial light modulator switches a preset binary coding pattern sequence to spatially modulate the thermal radiation image; a single-point infrared detector receives the spatially modulated infrared beam, converts the light intensity signal into a one-dimensional time-series electrical signal, and performs digitization processing; based on the preset coding sequence and the digitized one-dimensional time-series electrical signal, a two-dimensional temperature field distribution on the surface of the infrared optical element under test is obtained through a calculation and reconstruction algorithm. This invention uses the above method to replace traditional area array infrared thermal imagers to achieve high-precision, interference-resistant, and low-cost laser thermal effect measurement, and is particularly suitable for monitoring and analyzing the temperature field of optical elements under strong infrared laser irradiation.

Description

Technical Field

[0001] This invention relates to the field of infrared temperature measurement technology, and in particular to a method and apparatus for measuring the surface temperature of an infrared optical element irradiated by laser. Background Technology

[0002] In recent years, the laser power and energy specifications of infrared laser detection equipment have been continuously increasing, placing higher demands on the laser damage resistance of core infrared optical components. The temperature rise and thermal stress generated by the absorption of energy under high-energy laser irradiation are the fundamental causes of performance degradation and even failure of infrared optical lenses. Therefore, accurately measuring the temperature distribution on the surface of infrared optical components under laser irradiation is of great significance for assessing their laser damage resistance and optimizing optical system design.

[0003] Infrared lenses are typically made of materials such as germanium and zinc selenide, which have high transmittance to infrared radiation. This means the lens itself absorbs almost no infrared radiation, but rather transmits most of it. Therefore, infrared thermal imagers have difficulty directly detecting the surface temperature of the lens. Furthermore, the surface emissivity of infrared lenses, or their coatings, is usually low. Since the temperature measurement principle of infrared thermal imagers relies on the infrared radiation emitted by the object's surface, low emissivity results in a very weak radiation signal received by the imager, making accurate temperature measurement difficult. Additionally, infrared thermal imagers need to be calibrated according to the emissivity of the object being measured; if the emissivity setting is inaccurate, it will lead to significant deviations in the temperature measurement results.

[0004] Furthermore, the high transmittance of infrared lenses or the high reflectivity of high-reflectivity coatings can cause thermal imagers to receive radiation interference from the background or other objects, thus affecting the measurement of the lens surface temperature. The surface of an infrared lens may reflect thermal radiation from the surrounding environment, which the thermal imager may mistake for radiation from the lens itself, further leading to inaccurate temperature measurements. Moreover, the core component of an infrared thermal imager—the infrared detector—is highly sensitive to high-energy radiation and is easily damaged by overheating or ablation, especially when measuring the temperature of optical lenses irradiated by infrared lasers; in such cases, the thermal imager is easily damaged.

[0005] In summary, traditional infrared thermal imagers, when testing the surface temperature of infrared lenses irradiated by high-energy lasers, not only struggle to obtain accurate results due to the special properties of the lenses and the limitations of the imagers themselves, but also easily damage the imager's detection surface. Therefore, in practical applications, there is an urgent need to develop new technical methods to improve the temperature measurement accuracy of infrared lens surfaces irradiated by lasers, while simultaneously protecting the measuring equipment from damage by high-energy lasers. Summary of the Invention

[0006] The purpose of this invention is to provide a new measurement method and a measurement device for implementing the method, so as to solve the technical problems in the prior art that traditional infrared thermal imagers are difficult to accurately measure the surface temperature of infrared optical elements with low emissivity and high transmittance, and that the measurement equipment is easily damaged by high-energy lasers.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: On one hand, the present invention provides a method for measuring the surface temperature of an infrared optical element irradiated by laser, comprising the following steps: S1: Utilize an infrared optical lens to receive the thermal radiation emitted from the surface of an infrared optical element after being irradiated by a laser, and image it onto the modulation surface of the spatial light modulator. S2: By switching a preset binary coding pattern sequence through a spatial light modulator, the thermal radiation image is spatially modulated, and the two-dimensional spatial temperature distribution information is encoded into a one-dimensional time-series signal. S3: Utilize a single-point infrared detector to receive the spatially modulated infrared beam, convert the light intensity signal into a one-dimensional time-series electrical signal, and perform digital processing. S4: Based on the preset encoding sequence and the digitized one-dimensional time-series electrical signal, the two-dimensional temperature field distribution on the surface of the infrared optical element under test is obtained by calculation and reconstruction algorithm.

[0008] Preferably, the binary coded pattern sequence is a Hadamard coded sequence or a Gaussian random coded sequence, and the relationship between the number of coded patterns M and the total number of pixels N in the target temperature field is: M = 0.3N~0.6N.

[0009] Preferably, signal preprocessing is also performed in step S3, including: Dark field correction is performed on the digitized one-dimensional time-series electrical signal to eliminate background noise; The signal after dark field correction is normalized to map the signal amplitude to a preset range.

[0010] Preferably, the computational reconstruction algorithm is a compressed sensing reconstruction algorithm, which includes: based on the sparsity of the signal and the low coherence of the measurement matrix, using a pseudo-random binary coding matrix generated by a spatial light modulator as the measurement matrix, compressing and encoding the spatial information of the two-dimensional temperature field into a one-dimensional time-series light intensity signal, and then using a convex optimization algorithm to invert and recover the two-dimensional temperature field from the one-dimensional signal.

[0011] Preferably, the signal sparsity is achieved by sparsely representing the temperature field under a sparse basis, wherein the sparse basis is a wavelet basis, a discrete cosine basis, or a curve wave basis.

[0012] Preferably, the computational reconstruction algorithm is a deep learning reconstruction algorithm, comprising: The digitized one-dimensional time-series electrical signal is expanded and concatenated with the preset encoded sequence to form an input vector; The input vector is normalized and then input into the pre-trained neural network model; The pre-trained neural network model performs forward inference and directly outputs a two-dimensional temperature field distribution matrix, where each pixel value in the matrix is ​​the absolute temperature value at the corresponding location.

[0013] Preferably, the pre-trained neural network model is trained using a training dataset containing samples of different surface emissivity, enabling the neural network model to learn the deviation patterns of one-dimensional time-series signals under different emissivity conditions, and implicitly correcting the unknown surface emissivity of the infrared optical element under test during the reconstruction of the temperature field.

[0014] Preferably, the pre-trained neural network model adopts a hybrid network structure of U-Net and LSTM, wherein the LSTM layer is used to extract the temporal features of the one-dimensional time-series signal, and the encoding and decoding layers of U-Net are used to extract spatial features and recover the two-dimensional temperature field; the training loss function of the hybrid network structure includes mean square error loss, structural similarity loss and physical constraint loss, wherein the physical constraint loss is used to limit the reconstructed temperature within a preset operating temperature range.

[0015] On the other hand, the present invention provides a device for measuring the surface temperature of an infrared optical element irradiated by laser, comprising: The optical imaging and encoding module is used to receive infrared thermal radiation emitted from the surface of the infrared optical element under test, and to spatially modulate the thermal radiation image through the built-in spatial light modulator, encoding the two-dimensional spatial information into a one-dimensional time series signal. The signal acquisition and conversion module includes a single-point infrared detector, which is used to receive the modulated infrared beam and convert it into a one-dimensional time-series electrical signal; The data processing and reconstruction module is used to obtain the two-dimensional temperature field distribution on the surface of the infrared optical element under test by calculating and reconstructing the encoded sequence of the spatial light modulator and the one-dimensional time-series electrical signal.

[0016] Preferably, the spatial light modulator is a digital micromirror device, and the optical imaging and encoding module further includes a converging lens and a narrowband filter. The converging lens is used to collect the infrared beam modulated by the spatial light modulator; the narrowband filter is used to filter out stray light and background interference light of the excitation laser; and the single-point infrared detector is located at the focal point of the converging lens.

[0017] Compared with existing measurement methods, the technical solution provided by this invention has the following advantages: (1) A single-point detector is used to replace the traditional area array infrared detector, and the coding modulation of the spatial light modulator effectively avoids the direct damage of high-energy laser to the detector, while reducing the system cost. (2) By using the principle of computational imaging, the two-dimensional spatial temperature information is encoded into a one-dimensional time-series signal, and then the temperature field is reconstructed by compressed sensing or deep learning algorithms, thus realizing high-precision temperature measurement of low emissivity and high transmittance infrared optical elements. (3) The deep learning reconstruction algorithm implicitly learns the emissivity correction law through the pre-trained model, and can directly output the absolute temperature value without additional measurement or input of emissivity parameters, which simplifies the measurement process.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the measuring device provided by the present invention; Figure 2 This is a schematic diagram illustrating the principle of temperature field calculation, reconstruction, and temperature inversion using the deep learning model of this invention.

[0020] Figure label: 100. Infrared optical lens; 200. Infrared laser source; 300. Collimating lens; 1-1 Infrared optical lens; 1-2 Spatial light modulator; 1-3 Converging lens; 1-4 Narrowband filter; 1-5 Single-point infrared detector; 1-6 Data acquisition and control unit; 1-7 Computing and processing unit. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present application, the appendix in the embodiments of the present application will be described below. Figure 1-2 The technical solutions in the specific embodiments of this application are clearly and completely described. Unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. Terms such as "comprising" or "including" mean that the element or object preceding the word covers the element or object listed following the word and its equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0022] Example 1 First, this invention provides a method for measuring the surface temperature of infrared optical elements by laser irradiation, specifically implemented through the following steps: S1: Utilize an infrared optical lens to receive the thermal radiation emitted from the surface of an infrared optical element after being irradiated by a laser, and image it onto the modulation surface of the spatial light modulator. S2: By switching a preset binary coding pattern sequence through a spatial light modulator, the thermal radiation image is spatially modulated, and the two-dimensional spatial temperature distribution information is encoded into a one-dimensional time-series signal. S3: Utilize a single-point infrared detector to receive the spatially modulated infrared beam, convert the light intensity signal into a one-dimensional time-series electrical signal, and perform digital processing. S4: Based on the preset encoding sequence and the digitized one-dimensional time-series electrical signal, the two-dimensional temperature field distribution on the surface of the infrared optical element under test is obtained by calculation and reconstruction algorithm.

[0023] The measurement method provided by this invention is a computational imaging temperature measurement method based on a combination of single-point detection and spatial light modulation. Unlike traditional infrared thermal imagers that directly use area array detectors for imaging, this method transforms traditional area array imaging into a measurement mode of single-point detection plus algorithm inversion by combining spatial encoding and computational reconstruction, thereby realizing the temperature field measurement of low emissivity, high transmittance infrared optical components. Specifically, the first step uses laser irradiation to induce a local temperature rise in the component under test and emit infrared thermal radiation, providing a signal source for subsequent measurements; the second step uses a spatial light modulator to temporally modulate the thermal radiation image generated in the first step, encoding the two-dimensional spatial temperature distribution information into a one-dimensional time series, achieving information compression at the optical level; the third step uses a single-point infrared detector to collect the encoded and modulated light intensity signal, replacing the traditional area array detector to complete signal acquisition, effectively avoiding damage to the detector from high-energy lasers; finally, based on the known encoding sequence and the collected one-dimensional time series signal, the two-dimensional temperature field distribution is inverted through a computational reconstruction algorithm. The entire method uses the encoded sequence of the spatial light modulator as a bridge to establish a mapping relationship between the one-dimensional measurement signal and the two-dimensional temperature field, and achieves high-precision reconstruction of the temperature field through computational inversion.

[0024] In S2, a digital micromirror device (DMD) can be used as the spatial light modulator. The infrared thermal radiation emitted from the surface of the infrared optical element is captured by the infrared optical lens and clearly imaged on the DMD surface. Under computer control, the DMD rapidly switches a series of pseudo-random binary coded patterns (such as Hadamard or Gaussian random matrices) to spatially modulate the thermal radiation image. In order to meet the sampling theorem requirements, the pattern switching rate of the DMD must be higher than the thermal change rate of the infrared optical element under test.

[0025] Encoding and modulating thermal radiation images using spatial light modulators such as DMDs is a conventional technique, and its principle is the same as that of single-pixel imaging technology. However, the difference is that this invention applies it to temperature field measurement in the mid- and far-infrared bands.

[0026] In the S3, the single-point infrared detector can be either a thermopile detector or a pyroelectric detector. It receives the converged modulated infrared light and converts the total light intensity into a proportional one-dimensional voltage signal. A data acquisition card integrating digital I / O and analog input functions can be used. This acquisition card is connected to the control interface of the DMD and the signal output terminal of the single-point infrared detector via cables.

[0027] Similarly, converting light intensity signals into electrical signals is a conventional photoelectric conversion technology in this field. Thermoelectric detectors and pyroelectric detectors are both mature commercial devices. This invention does not improve their photoelectric conversion principle. The focus of this invention is to combine a single-point detector with a spatial light modulator and realize two-dimensional temperature field measurement through calculation and reconstruction, thereby solving the problem that traditional area array infrared detectors are easily damaged under high-energy laser irradiation conditions.

[0028] The S4 computational reconstruction algorithms include two types: compressed sensing reconstruction algorithm and deep learning reconstruction algorithm. The compressed sensing reconstruction algorithm is based on the sparsity of the temperature field under a sparse basis and uses convex optimization to invert the two-dimensional temperature field from the one-dimensional time series signal. The deep learning reconstruction algorithm uses a pre-trained end-to-end neural network model to directly output the two-dimensional temperature field from the one-dimensional time series signal. The following two examples will be used to illustrate this in detail.

[0029] Example 2 The core principle of compressed sensing reconstruction algorithms is based on signal sparsity and low coherence of the measurement matrix. A pseudo-random binary coding matrix (Hadamard or Gaussian random matrix) generated by a spatial light modulator is used as the measurement matrix to compress and encode the spatial information of the two-dimensional temperature field into a one-dimensional time-series light intensity signal. Then, a convex optimization algorithm is used to invert and recover the two-dimensional temperature field from the one-dimensional signal. This method does not require satisfying the Nyquist sampling theorem, significantly reducing the amount of data acquisition. The specific core algorithm formula is shown below: (a) Establishing a measurement model (b) Sparse representation model (c) Reconstruct the optimization model (d) Temperature field inversion and recovery in, is a one-dimensional time-series observation vector, which is the electrical signal sequence digitized by the data acquisition card output by the detector. The dimension is M×1, where M is the number of spatial light modulator coding patterns, i.e. the number of samplings. Usually, M≪N, where N is the number of columns in the matrix, N=H×W, and H and W represent the number of row pixels and column pixels of the temperature field image, i.e. the target resolution, respectively. x is a one-dimensional expansion vector of the two-dimensional temperature field, i.e. the temperature field of the optical lens surface to be reconstructed.

[0030] The measurement matrix, also known as the spatial light modulator coding matrix, corresponds to the pseudo-random binary coded sequence generated by the spatial light modulator, and has dimensions M×N; elements (Binary encoding, 0 indicates occlusion, 1 indicates light transmission); if it is Hadamard encoding, This is the binary form of the normalized Hadamard matrix (elements ±1 are mapped to 0 / 1).

[0031] The vector is a one-dimensional expansion of the two-dimensional temperature field. The surface temperature field of the optical lens to be reconstructed has a dimension of N×1. Specifically, the two-dimensional temperature distribution map of H×W is expanded into a one-dimensional vector by rows / columns.

[0032] The noise vector to be measured includes infrared detector noise, thermal noise, shot noise, and data acquisition card noise, with a dimension of M×1, and is usually assumed to be Gaussian white noise.

[0033] For a sparse basis matrix, a sparse basis that adapts to the spatial characteristics of the temperature field is selected, with a dimension of N×N. Wavelet basis or discrete cosine basis is preferred because the temperature field is usually sparse in the frequency domain / wavelet domain, with most of the energy concentrated in a few low-frequency coefficients. If there are local abrupt changes in the temperature field, such as at the edge of a lens or at a defect, a curvelet basis can be selected to adapt to the edge sparse characteristics.

[0034] Let N be the coefficient vector of the temperature field under a sparse basis, with dimension N×1, satisfying sparsity: .

[0035] The regularization parameter balances the weights of the data fitting term and the sparsity constraint, and its value ranges from [value range missing]. ~ It needs to be based on the mid-infrared noise level The adjustment strategy is that the greater the noise, the better. The larger.

[0036] The sparse coefficients are estimated values, and the optimal sparse coefficients are obtained by solving an optimization algorithm.

[0037] The reconstructed two-dimensional temperature field vector, after being restored by the inverse transformation of sparse basis, needs to be reshaped into a two-dimensional matrix of H×W, which is the final temperature distribution map.

[0038] The detailed calculation steps are as follows: S4.1: System parameter initialization, including: Determine the target resolution of the temperature field, specifically by determining the number of rows (H) and columns (W), and then calculate the resolution N = H × W. Determine the encoding parameters of the spatial light modulator, including the number of encoding patterns M, which is typically taken as M = 0.3N~0.6N, to balance reconstruction accuracy with acquisition speed and encoding type (such as Hadamard / Gaussian random matrix), and generate the measurement matrix. ; Choose sparse bases , pre-generate sparse basis matrix Estimating noise levels (Calculated using data collected from the detector's dark field), initialize the regularization parameters. .

[0039] S4.2: Data preprocessing. This specifically includes: Read one-dimensional time-series signal from data acquisition card That is, the digital electrical signal output by the detector; Dark field correction: , This is the dark field signal when the spatial light modulator is fully off, to eliminate background noise; Normalization Normalize the signal amplitude to (Improve and optimize stability) S4.3: Optimization of sparse coefficients (using the FISTA algorithm (Fast Iterative Shrinking Threshold Algorithm), which converges quickly), specifically including: Define a composite function ; Perform initialization, specifically setting the initial sparsity coefficient to 0, the step size parameter to 1, and the number of iterations to be adjusted according to the accuracy requirements, such as one thousand times; Iterative updates are achieved through steps such as gradient calculation, momentum update, soft thresholding, step size update, and convergence judgment.

[0040] S4.4: Temperature field reconstruction and post-processing, specifically including: Inverse sparse coefficient transform We obtain a one-dimensional temperature field vector; Will Remodeled into a two-dimensional H×W matrix , i.e., the temperature field distribution map; Through formula Perform physical constraint corrections, where T min The lowest possible temperature of the lens is determined, eliminating non-physical values ​​such as negative temperatures; finally, Gaussian filtering is used to preserve the details of the temperature field while suppressing reconstruction noise.

[0041] Example 3 The core principle of deep learning reconstruction algorithms is to construct an end-to-end mapping model of "time-series signal → two-dimensional temperature field". The network is trained using a large amount of labeled data; that is, by using the spatial light modulator encoding sequence and the corresponding real temperature field, the nonlinear mapping relationship between the spatial light modulator encoding and the temperature field is learned. This eliminates the need for manually designing sparse bases or optimization functions, resulting in fast reconstruction speed and strong robustness. The specific core formula is based on a U-Net+LSTM hybrid network, and the specific steps are as follows: (a) Network input / output definition , in, , , N=H×W (b) Network forward propagation formula Temporal feature extraction: ; ; Spatial feature encoding (U-Net downsampling): ; Spatial feature decoding (U-Net upsampling): Output layer (temperature field prediction): (c) Loss function (adapted to the physical properties of the temperature field) Wherein, Net(⋅) is the end-to-end reconstruction network, a custom U-Net+LSTM hybrid network, with input being a one-dimensional time-series signal and encoding matrix corresponding to SLM encoding, and output being a two-dimensional temperature field; h t For the LSTM hidden state, the temporal feature vector corresponding to the t-th SLM encoded pattern has a dimension of 128×1 (adjusted according to the training effect). F temp To fuse temporal feature maps, the M LSTM hidden states are concatenated into a two-dimensional feature map with dimensions of 128×32×32 (adapted to U-Net input format). Conv2d / ConvTranspose2d are convolution / transpose convolution operations with kernel size ksize=3, stride s=2, upsampling and downsampling the same, and kernel counts C1=64, C2=128, C3=256, and C4=512. BN / ReLU are batch normalization / activation functions that accelerate network training, alleviate gradient vanishing, and adapt to the distribution characteristics of temperature field data. For mean squared error loss, T gt The actual temperature field (such as infrared thermal imager calibration data); The structural similarity loss measures the structural consistency between the reconstructed temperature field and the real temperature field, with a weight of γ=0.3. This setting emphasizes the accuracy of edge and detail reconstruction. To mitigate physical constraints, the reconstruction temperature is limited to a reasonable range: T min / T max This refers to the operating temperature range on the lens.

[0042] The weights of the loss function are adjusted according to the characteristics of mid-infrared data to balance numerical accuracy, structural consistency, and physical rationality.

[0043] The detailed calculation steps are as follows: S4.1′: Dataset construction, specifically including: real data acquisition, mainly based on the built test system, generating spatial light modulator coding sequences and collected samples; simulation data supplementation, mainly based on the heat conduction equation to generate virtual temperature fields, including scenarios such as uniform temperature, gradient temperature, and local high temperature points; dataset partitioning, including 50,000 training sets, 8,000 validation sets, and 2,000 test sets, divided in a 7:1:1 ratio.

[0044] S4.2′: Network training, specifically including: network initialization; training parameter settings, including setting the optimizer, batch size, and number of iterations; training process, including forward propagation, loss function calculation, back propagation, and model saving; S4.3′: Temperature field reconstruction. The reconstruction process includes data input, preprocessing, grid inference, and post-processing; the reconstruction speed must meet the requirements of real-time monitoring.

[0045] In this embodiment, a pre-trained deep learning reconstruction model (such as the U-Net architecture) is used to accomplish this task, and its contents include: (1) Model input: The one-dimensional signal sequence Y and the encoding matrix A are expanded and combined to form the input vector of the network. The specific steps are as follows: The input data has prior attributes: a one-dimensional signal sequence Y with a dimension of M×1, which physically represents the electrical signal output by the detector and digitized by the data acquisition card; and an encoding matrix A with a dimension of M×N (N=H×W, where H is the number of height pixels of the target temperature field and W is the number of width pixels), which physically represents the Hadamard / Gaussian binary encoding sequence of the spatial light modulator during actual operation (elements are only 0 or 1, where 0 indicates micromirror occlusion and 1 indicates micromirror light transmission).

[0046] The unfolding operation involves directly preserving the M×1 format of the one-dimensional signal Y, or converting it into a 1×M row vector; the encoding matrix A, a two-dimensional matrix, is flattened into a one-dimensional row vector according to the "row priority" rule, and the dimension is converted from M×N to 1×(M×N).

[0047] The combination operation prioritizes horizontally concatenating Y with the flattened A, i.e., concatenating 1×M Y with 1×(M×N) A; the final input vector has a dimension of 1×(M+M×N) and serves as the input to the U-Net model.

[0048] The initial preprocessing is simple and requires no complex operations; it only involves normalization: mapping the concatenated input vector to... The interval is calculated using the following formula: , where Xconcat is the concatenated original input vector, and the normalization parameters (maximum and minimum values) are recorded during the pre-training stage and can be reused directly during inference.

[0049] (2) Model processing: This network learns a complex mapping relationship from encoded measurements to spatial distribution through internal multi-layer nonlinear transformations. The specific steps are as follows: The overall model architecture adopts a core structure of "input layer → encoding layer (downsampling) → decoding layer (upsampling) → intermediate feature fusion," with the number of layers controlled at 5-7 to reduce computational complexity. The specific process is as follows: input layer transformation (solving the adaptation problem of one-dimensional to two-dimensional conversion); encoding layer downsampling (extracting deep mapping features); decoding layer upsampling (restoring two-dimensional spatial resolution); and the core role of multi-layer nonlinear transformation, where the convolution and activation functions of each layer together constitute the nonlinear transformation.

[0050] Key processing features: End-to-end processing: After the input vector is fed into the model, no manual intervention is required in the intermediate process. The model automatically completes feature extraction, mapping learning, and resolution restoration, directly outputting a two-dimensional temperature field. Robustness to noise: U-Net's deep feature extraction capabilities can filter thermal noise and shot noise during mid-infrared detection, eliminating the need for additional denoising modules and simplifying the processing flow.

[0051] (3) Model output: The network directly outputs a two-dimensional matrix, where each pixel value represents the absolute temperature value at the corresponding location of the measured target. In other words, during the reconstruction of the temperature field image, the model simultaneously corrects for the emissivity of the unknown surface and directly outputs the temperature. The specific steps are as follows: Output the core attributes of the data. The output format is a two-dimensional matrix. The matrix has dimensions H×W and is a single-channel grayscale matrix (each pixel corresponds to a temperature value). The physical meaning of a pixel is: the value of each pixel in the matrix. The absolute temperature value corresponding to position (i,j) on the surface of the tested optical lens can be expressed in ℃ or K. After pre-training, the average temperature error can be controlled within ±1℃ (meeting the actual needs of optical lens temperature field monitoring), and the error in edge areas and local defect areas does not exceed ±2℃.

[0052] The implementation logic of automatic emissivity correction: The model does not require additional input emissivity parameters or a separately designed correction algorithm. Instead, emissivity correction is completed synchronously during the pre-training phase through the coverage of the labeled dataset and the implicit learning capability of U-Net. This includes: dataset support, i.e., the pre-training dataset contains optical lens samples with different surface emissivity (e.g., coated lenses, uncoated lenses, and lenses with scratches / stains, emissivity range 0.1~0.9), and each sample is labeled with the true absolute temperature field collected by a high-precision infrared thermal imager (with the influence of emissivity eliminated); and implicit learning, i.e., during the multi-layer nonlinear transformation process, the model automatically learns the deviation pattern of different emissivity → time-series signal Y and incorporates this pattern into the network parameters; during the inference phase, the model automatically offsets the temperature deviation caused by the unknown emissivity based on the characteristics of the input Y, and finally outputs the corrected absolute temperature.

[0053] Post-processing of the output includes denormalization, physical constraint correction, and format conversion. Denormalization, if normalization was performed during the input stage, uses the normalization parameters (maximum and minimum values) recorded during the pre-training stage to remap the output pixel values ​​in the [0,1] range to the actual temperature values. Physical constraint correction simply filters out non-physical temperature values, such as cropping pixel values ​​below the lens's minimum operating temperature (e.g., -20℃) or above its maximum operating temperature (e.g., 300℃) to their corresponding extreme values ​​to avoid invalid output. Format conversion converts the two-dimensional matrix into common image formats (e.g., TIFF, PNG) for easy visualization or subsequent data analysis; this step is necessary.

[0054] In summary, based on the content of Embodiments 2 and 3, both algorithms, when combined with the mid- and far-infrared spatial light modulator-encoded temperature field reconstruction scenario, can solve specific technical problems: First, the measurement matrix binarization of compressed sensing and the encoding matrix auxiliary input of deep learning are both designed for the physical working mode of SLM. Secondly, the noise adaptive regularization of compressed sensing and the construction of mid-infrared noise datasets by deep learning have both solved the problem of high noise in the mid-infrared band. Finally, the selection of the curvilinear basis in compressed sensing and the physical constraint loss in deep learning are both designed to address the spatial smoothness and temperature range limitations of the lens temperature field.

[0055] Example 4 This embodiment also provides a measuring device for implementing the aforementioned measurement method, specifically including: an optical imaging and encoding module, a signal acquisition and conversion module, and a data processing and reconstruction module. It also includes a triggering structure, whereby the infrared laser source 200, after being collimated by the collimating lens 300, compresses the infrared laser divergence angle, ensuring that the collimated beam output irradiates the surface of the infrared optical lens 100 to be measured. After absorbing laser energy, the infrared optical lens 100 experiences a local temperature increase, and this area thus generates and emits infrared radiation corresponding to its temperature distribution.

[0056] The optical imaging and coding module is responsible for capturing target radiation and completing spatial coding. Its core equipment includes an infrared optical lens 1-1, a spatial light modulator 1-2, a converging lens 1-3, and a narrowband filter 1-4.

[0057] Infrared optical lens 1-1, serving as the objective lens of the system, is aimed at the target under test to receive and converge the infrared radiation emitted from the target surface, forming an intermediate infrared image on its image plane. The main function of infrared optical lens 1-1 is to clearly image the target temperature distribution onto the micromirror surface of spatial light modulator 1-2, and the size of the image must match the effective optical area of ​​spatial light modulator 1-2.

[0058] Spatial light modulators 1-2, in this embodiment, are digital micromirror devices (DMDs). Under computer control, they switch a series of pseudo-random binary coded patterns at high speeds, typically on the order of kilohertz. This is because the pattern switching speed of the DMD must be much higher than the thermal change rate of the target being measured to satisfy the sampling theorem. The pseudo-random binary coded patterns include Hadamard patterns or Gaussian random matrix coded patterns. After spatially modulating the incident thermal radiation image, the DMD encodes the two-dimensional spatial information into a one-dimensional time-series signal. Specifically, the DMD is precisely placed at the image plane of the infrared optical lens to receive the intermediate infrared image formed by it. Since each micromirror unit of the DMD can be deflected independently, high-speed, programmable spatial modulation of the image is possible. The open-state beam reflected by the specific pattern of the DMD is collected by converging lens 1-3.

[0059] Converging lenses 1-3 are high numerical aperture converging lenses or lens groups, which are set in the reflected light path of the DMD to collect and converge the infrared beam that has been modulated and reflected by the DMD to a single focal point.

[0060] Narrowband filters 1-4 are placed close to the sensor surface in the signal acquisition and conversion module to receive optical signals. Their center wavelength is matched with the thermal radiation band, aiming to strictly filter out stray laser light and other background interference light from the excitation source, ensuring that only the target thermal radiation signal is detected.

[0061] The signal acquisition and conversion module is responsible for converting optical signals into processable electrical signals. Its core components include: single-point infrared detectors 1-5 and data acquisition and control units 1-6.

[0062] The single-point infrared detector 1-5 has good performance, simple structure and low cost, and is preferably a thermopile detector or a pyroelectric detector. The single-point infrared detector 1-5 is precisely placed at the focal point of the converging lens 1-3 to receive the converged modulated infrared light and convert its total light intensity into a one-dimensional voltage signal proportional to it. That is, the narrowband filter 1-4 is placed close to the front of the single-point infrared detector 1-5.

[0063] In this embodiment, the data acquisition and control unit 1-6 can be a data acquisition card that integrates digital I / O and analog input functions. This acquisition card is connected to the control interface of the DMD and the signal output terminal of the single-point infrared detector 1-5 via cables, respectively, for synchronously controlling the encoding switching of the DMD and acquiring the electrical signals output by the single-point infrared detector 1-5.

[0064] The core of the data processing and reconstruction module is the computing processing unit 1-7, which in this embodiment is an industrial computer. This computer is electrically connected to the data acquisition and control unit via a bus, and internally stores the software program and control logic for implementing the method of this invention. The single-point infrared detector 1-5 converts the received time-series light intensity signal into an electrical signal, which is then digitized by the data acquisition and control unit 1-6 and uploaded to the computing processing unit 1-7. The computing processing unit 1-7 uses computational reconstruction algorithms such as compressed sensing or deep learning, combined with a known DMD encoding sequence, to invert and reconstruct a two-dimensional high-resolution temperature field distribution map of the optical lens surface from this one-dimensional signal.

[0065] The measurement process using the aforementioned measuring device can be divided into four core stages: Phase 1: A high-power infrared laser beam precisely irradiates a specific area on the surface of the optical component under test. The laser energy is absorbed by the material and converted into heat energy, resulting in a localized temperature rise. The heated area emits corresponding mid-wave or long-wave infrared thermal radiation according to its temperature distribution.

[0066] The second stage involves the infrared thermal radiation emitted from the surface being measured being captured by an infrared optical lens and clearly imaged onto the DMD surface. Under computer control, the DMD rapidly switches pseudo-random binary coded patterns at a high rate of several kilohertz to spatially modulate the thermal radiation image. This process encodes two-dimensional spatial temperature distribution information into a one-dimensional time series, achieving compressed sensing at the optical level.

[0067] The third stage: The open-state beam modulated by the DMD is collected by a high numerical aperture converging lens, and after being rigorously filtered by a narrow-band filter to remove stray light from the excitation laser, it is focused onto an infrared single-point detector. The detector converts the light intensity signal into a time-series electrical signal, which is amplified, filtered, and digitized by a data acquisition card. After digitization, dark-field correction is performed to eliminate background noise, and normalization is performed to map the signal amplitude to a preset range.

[0068] The fourth stage: The computer uses compressed sensing or deep learning algorithms, combined with known DMD encoding sequences, to reconstruct a two-dimensional temperature distribution image from the acquired one-dimensional time-series signal. The deep learning algorithm implicitly performs emissivity correction through a pre-trained model, eliminating the influence of uncertainties in the emissivity of the material surface, and finally outputs high spatial resolution absolute temperature field measurement results, achieving sub-pixel-level visualization of temperature distribution.

[0069] The entire system achieves high-precision, anti-interference, and low-cost laser thermal effect measurement, which is difficult to accomplish with traditional area array infrared thermal imagers, through an innovative combination of single-point detection and spatial light modulation. It is particularly suitable for monitoring and analyzing the temperature field of optical components under strong infrared laser irradiation.

[0070] It should be noted that, for this invention, the core inventive concept is to replace the traditional area array thermal imager with computational imaging principles, namely, single-point detection combined with DMD spatial coding combined with deep learning / compressed sensing reconstruction. This is emphasized in the specification. Some well-known techniques in the calculation process, such as using pseudo-random binary coding patterns for spatial adjustment and photoelectric conversion, are not described in detail. Similarly, conventional mechanical and electrical structures and connection methods in the device section of this application are also briefly or omitted. It should be understood that the omission of these contents in detail will not affect the completeness of the disclosure in this specification. Those skilled in the art can clearly and without doubt understand all the technical solutions of this invention.

[0071] Finally, it should be noted that the described embodiments are merely some, not all, of the embodiments of the present invention. Those skilled in the art will understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents; that is, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A method for measuring the surface temperature of an infrared optical element irradiated by laser, characterized in that, Includes the following steps: S1: Utilize an infrared optical lens to receive the thermal radiation emitted from the surface of an infrared optical element after being irradiated by a laser, and image it onto the modulation surface of the spatial light modulator. S2: By switching a preset binary coding pattern sequence through a spatial light modulator, the thermal radiation image is spatially modulated, and the two-dimensional spatial temperature distribution information is encoded into a one-dimensional time-series signal. S3: Utilize a single-point infrared detector to receive the spatially modulated infrared beam, convert the light intensity signal into a one-dimensional time-series electrical signal, and perform digital processing. S4: Based on the preset encoding sequence and the digitized one-dimensional time-series electrical signal, the two-dimensional temperature field distribution on the surface of the infrared optical element under test is obtained by calculation and reconstruction algorithm.

2. The measurement method according to claim 1, characterized in that, The binary coded pattern sequence is a Hadamard coded sequence or a Gaussian random coded sequence, and the relationship between the number of coded patterns M and the total number of pixels N in the target temperature field is: M = 0.3N~0.6N.

3. The measurement method according to claim 1, characterized in that, S3 also includes signal preprocessing, including: Dark field correction is performed on the digitized one-dimensional time-series electrical signal to eliminate background noise; The signal after dark field correction is normalized to map the signal amplitude to a preset range.

4. The measurement method according to claim 1, characterized in that, The computational reconstruction algorithm is a compressed sensing reconstruction algorithm, including: Based on the sparsity of the signal and the low coherence of the measurement matrix, a pseudo-random binary coding matrix generated by a spatial light modulator is used as the measurement matrix to compress and encode the spatial information of the two-dimensional temperature field into a one-dimensional time-series light intensity signal. Then, the two-dimensional temperature field is recovered from the one-dimensional signal by a convex optimization algorithm.

5. The method according to claim 4, characterized in that, The signal sparsity is achieved by sparsely representing the temperature field under a sparse basis, wherein the sparse basis is a wavelet basis, a discrete cosine basis, or a curve wave basis.

6. The measurement method according to claim 1, characterized in that, The computational reconstruction algorithm is a deep learning reconstruction algorithm, including: The digitized one-dimensional time-series electrical signal is expanded and concatenated with the preset encoded sequence to form an input vector; The input vector is normalized and then input into the pre-trained neural network model; The pre-trained neural network model performs forward inference and directly outputs a two-dimensional temperature field distribution matrix, where each pixel value in the matrix is ​​the absolute temperature value at the corresponding location.

7. The measurement method according to claim 6, characterized in that, The pre-trained neural network model is trained using a training dataset containing samples of different surface emissivity, enabling the neural network model to learn the deviation patterns of one-dimensional time-series signals under different emissivity conditions, and implicitly correcting the unknown surface emissivity of the infrared optical element under test during the reconstruction of the temperature field.

8. The measurement method according to claim 6 or 7, characterized in that, The pre-trained neural network model adopts a hybrid network structure of U-Net and LSTM, where the LSTM layer is used to extract the temporal features of the one-dimensional time-series signal, and the encoding and decoding layers of U-Net are used to extract spatial features and recover the two-dimensional temperature field. The training loss function of the hybrid network structure includes mean square error loss, structural similarity loss and physical constraint loss, and the physical constraint loss is used to limit the reconstructed temperature within a preset operating temperature range.

9. A device for measuring the surface temperature of an infrared optical element irradiated by laser, characterized in that, include: The optical imaging and encoding module is used to receive infrared thermal radiation emitted from the surface of the infrared optical element under test, and to spatially modulate the thermal radiation image through the built-in spatial light modulator, encoding the two-dimensional spatial information into a one-dimensional time series signal. The signal acquisition and conversion module includes a single-point infrared detector, which is used to receive the modulated infrared beam and convert it into a one-dimensional time-series electrical signal; The data processing and reconstruction module is used to obtain the two-dimensional temperature field distribution on the surface of the infrared optical element under test by calculating and reconstructing the encoded sequence of the spatial light modulator and the one-dimensional time-series electrical signal.

10. The measuring device according to claim 9, characterized in that, The spatial light modulator is a digital micromirror device. The optical imaging and encoding module further includes a converging lens and a narrowband filter. The converging lens is used to collect the infrared beam modulated by the spatial light modulator. The narrowband filter is used to filter out stray light and background interference light from the excitation laser. The single-point infrared detector is located at the focal point of the converging lens.

Images