LED array spectrum data batch acquisition system and method

Abstract

The invention provides an LED array spectral data batch acquisition system and method. The system comprises a current source, a laser source, a microscope, an AOTF hyperspectral imager, a camera and a processor used for batch acquisition. The microscope comprises a plurality of objective lenses and a mechanical objective table; the processor is used for adjusting the driving frequency of the AOTF hyperspectral imager, performing image acquisition under different wavelengths and outputting a complete spectrum of the LED array; the processor comprises a model training module and a batch acquisition module, the model training module is used for constructing an SVM (Support Vector Machine) model, and the batch acquisition module is used for acquiring LED array spectrums under partial wavelengths and inputting the LED array spectrums into the SVM model to obtain LED array complete spectrums. According to the method, the monochromatic spectral data of a small amount of wavelengths of the LED chips are acquired in batches and input into the SVM model, so that the high-precision LED array full-spectrum data can be mapped, and the method has good practicability.

Description

Technical Field

[0001] The present invention relates to the field of photoelectric detection, and in particular to a system and method for batch collecting and analyzing LED array spectral data with high speed, high efficiency and high precision by using acousto-optic tunable filter (AOTF) hyperspectral and machine learning technology. Background Art

[0002] Faced with the demands of large-scale, efficient and high-precision in the modern photoelectric detection market, traditional light emitting diode (LED) arrays generally use CCD spectrometers to collect the emission spectra of chips one by one, which is inefficient and time-consuming. For monochromatic LED chips, CCD spectrometers collect a large number of invalid band spectra, resulting in a waste of storage space and processing time.

[0003] AOTF uses the refractive index change caused by sound waves in transparent media to adjust and select light of specific wavelengths through the acousto-optic diffraction effect. AOTF is widely used in real-time spectral analysis and hyperspectral imaging due to its advantages such as fast tuning, no mechanical parts and high flexibility.

[0004] At present, some spectral acquisition systems based on AOTF, for example, the invention patent with publication number CN109115339A provides a high-speed hyperspectral full-polarization imaging device and method based on AOTF and intensity modulation, which can achieve high-resolution, high-sensitivity and high-reliability detection. For example, the polarization intensity modulation of the measured spectrum can significantly improve the temporal resolution; for another example, the invention patent with publication number CN117571127A discloses an optical system and method for hyperspectral dual-polarization channel acquisition based on AOTF, which utilizes dual-polarization channel acquisition and a single detector to simultaneously image two polarization state images, thereby improving reliability while ensuring the portability of the system; for another example, the invention patent with publication number CN115077698A provides a common-target multi-channel AOTF hyperspectral real-time imaging system, which can achieve real-time acquisition of continuously changing hyperspectral fusion images of the same target at the same time.

[0005] Nowadays, in order to meet the needs of large-scale production, the actual production line and inspection line of LED require high automation. The introduction of machine learning can significantly improve the automation level, detection accuracy and efficiency of the hyperspectral acquisition system. The combination of machine learning and hyperspectral images can already achieve high-precision predictions. For example, the invention patent with publication number CN115728247A discloses a spectral measurement quality determination method based on machine learning, which ensures the accuracy of spectral data by judging spectral similarity and eliminating noise and other factors. Summary of the invention

[0006] A brief summary of embodiments of the present invention is provided below to provide a basic understanding of certain aspects of the present invention. It should be understood that the following summary is not an exhaustive summary of the present invention. It is not intended to identify key or important parts of the present invention, nor is it intended to limit the scope of the present invention. Its purpose is merely to provide certain concepts in a simplified form as a prelude to a more detailed description discussed later.

[0007] In order to solve the problems of low efficiency and low precision in the existing batch acquisition of LED array spectral data, this application proposes a system for high-speed, efficient and high-precision batch acquisition of LED array spectral data based on AOTF hyperspectral and machine learning.

[0008] According to the first aspect of the present application, a system for batch acquisition of spectrum data of an LED array is provided, which comprises a current source, a laser source, a microscope, an AOTF hyperspectral imager, a camera and a processor for batch acquisition; the microscope comprises a plurality of objective lenses and a mechanical stage; the AOTF hyperspectral imager is respectively connected to interfaces of the camera and the microscope; a plurality of sample fixtures to be tested are placed on the mechanical stage of the microscope, and the knob of the mechanical stage is turned so that the sample to be tested faces the objective lens; the sample to be tested is energized by the current source to emit light or introduced into the laser source to form a surface light source for batch excitation; the AOTF hyperspectral imager sets a specific driving frequency, and only light energy corresponding to a specific wavelength is emitted from the light outlet of the AOTF hyperspectral imager; the emitted light is imaged on the camera through a lens;

[0009] The processor is used to execute: adjusting the driving frequency of the AOTF hyperspectral imager, collecting images at different wavelengths, and outputting a complete spectrum of the LED array;

[0010] The processor includes a model training module and a batch acquisition module. The model training module is used to build a support vector machine (SVM) model. The batch acquisition module is used to collect the LED array spectrum under partial wavelengths and input the LED array spectrum into the SVM model to obtain the complete LED array spectrum.

[0011] As an implementation scheme, the model training module includes the following units:

[0012] LED array full spectrum data acquisition unit: set the parameters of AOTF hyperspectral imager and camera respectively to acquire the full spectrum data of LED array;

[0013] Preprocessing unit: preprocess the acquired LED array full spectrum data to obtain training set data; preprocessing operations include noise removal, data standardization or normalization to ensure that all features are on the same scale;

[0014] SVM model building unit: set the number of iterations and data range;

[0015] Model training unit: Divide the full spectrum data of the LED array preprocessed in step S1-2 into a feature matrix and a target vector, and use the SVM model constructed in step S1-3 for training. According to the set number of iterations and data range, iteratively update the SVM model parameters and minimize the loss function; after training, the SVM model outputs the complete spectrum of the LED array and the peak wavelength and half-width. The optimized model parameters, including the support vector, kernel function type and its parameters, bias value and related preprocessing steps, are serialized and saved. The model file is stored using Pickle; the output SVM model is a complete body, which can be directly used for machine learning of subsequent batches of data.

[0016] Model evaluation unit: Evaluate the prediction accuracy of the SVM model according to the R2 (coefficient of determination) value; if the R2 value is low, return to step S1-3 to adjust the number of iterations and the data range, and re-execute the model training of step S1-4 until the prediction accuracy of the SVM model meets the requirements.

[0017] The batch acquisition module includes the following units:

[0018] Batch acquisition unit: Modify the parameters of AOTF hyperspectral imager and camera to collect LED array hyperspectral data at a small number of wavelengths;

[0019] Model mapping unit: inputs a small amount of collected wavelengths into the SVM model obtained in the model training process;

[0020] Model output unit: The SVM model outputs the complete spectrum, peak wavelength, and half-width of the LED array.

[0021] According to a second aspect of the present application, a method for batch acquisition of LED array spectrum data is provided, which is applied to the above-mentioned LED array spectrum data batch acquisition system, and includes a model training process and a batch acquisition process, wherein the model training process includes the following steps:

[0022] Step S1-1, respectively setting the parameters of the AOTF hyperspectral imager and the camera to obtain full spectrum data of the LED array;

[0023] Step S1-2, preprocessing the acquired LED array full spectrum data to obtain training set data; the preprocessing operation includes removing noise, data standardization or normalization processing to ensure that all features are on the same scale;

[0024] Step S1-3, construct the SVM model, set the number of iterations and data range;

[0025] Step S1-4, model training: the full spectrum data of the LED array preprocessed in step S1-2 is divided into a feature matrix and a target vector, and the SVM model constructed in step S1-3 is used for training. According to the set number of iterations and data range, the SVM model parameters are iteratively updated to minimize the loss function; after the training is completed, the SVM model can output the complete spectrum of the LED array and the peak wavelength and half-height width;

[0026] Step S1-5, evaluate the model: evaluate the prediction accuracy of the SVM model according to the R2 value; if the R2 value is low, return to step S1-3 to adjust the number of iterations and the data range, and re-execute the model training of step S1-4 until the prediction accuracy of the SVM model meets the requirements.

[0027] Furthermore, the step S1-1 specifically includes:

[0028] Set the camera acquisition mode to continuous acquisition in the trigger mode. The camera starts acquisition at a certain delay relative to the wavelength setting of the AOTF hyperspectral imager, and the sum of the exposure time and the delay time is less than the interval between adjacent wavelength settings of the AOTF hyperspectral imager. Collect the AOTF hyperspectral data of the sample to be tested (LED array to be tested), and obtain the full spectrum data of the sample to be tested in each array (or LED array full spectrum data, LED array complete spectrum).

[0029] Furthermore, the step S1-2 specifically includes:

[0030] The spectrum contains noise and other uncertainties, so it is necessary to preprocess the acquired full spectrum data to remove random noise in the image, smooth the image, and normalize the existing data to highlight the characteristic peaks in the spectrum to improve the training effect of the model and make subsequent machine learning training more effective. The preprocessed full spectrum data is used as the training set;

[0031] Furthermore, the step S1-3 specifically includes:

[0032] Use the Multilayer Perception (MLP) model to initialize model parameters, including the number of iterations, the range of data to be acquired, and the selection of an appropriate weight initialization method. After evaluating the data, determine the number of hidden layers and neurons.

[0033] Furthermore, the step S1-4 specifically includes:

[0034] The full spectrum data is used as input data for training, and is input into the one-dimensional vector SVM model classifier for training. The Gaussian kernel function is used as the kernel function of the SVM model. The Gaussian kernel function can handle nonlinear mapping relationships and performs well in reconstructing spectra in high-dimensional space. The activation value of the hidden layer is extracted as a feature vector; the SVM model optimizes the performance and generalization ability of this model by finding the optimal hyperplane and adjusting the bandwidth of the kernel function, and then maps the monochromatic spectrum data of a small amount of wavelength covering the LED emission peak to the data points corresponding to the full spectrum. According to the principle of spectral superposition, the monochromatic LED spectral distribution can be linearly superimposed to generate a new spectral distribution and generate continuous full spectrum data. The mapping relationship of the SVM model can be expressed by the following formula:

[0035]

[0036] Among them, K(x·x i ) is the kernel function, x i are training data points, and b * is a model parameter. The SVM model can reconstruct the complete spectrum curve by learning the mapping relationship between the input data and the target data. LED spectrum data is usually smooth, and the intensity distribution of monochromatic light can often reflect the intensity trend of adjacent wavelengths. SVM captures these nonlinear relationships through this method and realizes the inference of the complete spectrum from partial wavelengths. The SVM model is trained with multiple complete spectrum data to learn the mapping relationship between the input spectrum points and the complete spectrum. In this way, in the inference stage, even if only a few wavelengths of light intensity are input, the model can reconstruct the complete spectrum using the patterns learned in training. The complete spectrum curve is reconstructed by the SVM model, and after eliminating the noise caused by the equipment or environment, the peak wavelength is obtained according to the maximum position of the detected spectrum curve, and the half-peak width of the spectrum curve is obtained by the half-height width. After the SVM model training is completed, the optimized model parameters, including support vectors, kernel function types and their parameters, bias values, and related preprocessing steps, are serialized and saved. The model file is stored using Pickle to ensure its cross-platform compatibility in subsequent applications. The saved model file has the ability to nonlinearly map from the input spectrum data points to the complete spectrum curve, and can be loaded into the online analysis platform to achieve efficient spectrum reconstruction and key parameter extraction.

[0037] Furthermore, the step S1-5 specifically includes:

[0038] Calculate the R2 value to evaluate the performance of the SVM model, obtain the successfully trained SVM model, and enter the batch acquisition module stage.

[0039] The batch collection process includes the following steps:

[0040] Step S2-1, batch acquisition: modify the parameters of the AOTF hyperspectral imager and the camera to collect LED array hyperspectral data at a small number of wavelengths;

[0041] Step S2-2, model mapping: inputting a small amount of collected wavelengths into the SVM model obtained in the model training process;

[0042] Step S2-3: The SVM model outputs the complete spectrum, peak wavelength, and half-width of the LED array.

[0043] Furthermore, the step S2-1 specifically includes:

[0044] Implement batch acquisition of LED arrays, adjust the driving frequency of the AOTF hyperspectral imager to collect a small number of wavelengths, and adjust the exposure time of the camera accordingly to collect monochromatic spectral data of a small number of wavelengths covering the LED emission peak; move the mechanical stage to perform batch spectral data and image acquisition of the LED array.

[0045] Furthermore, the step S2-2 specifically includes:

[0046] The monochromatic spectral data of a small amount of wavelengths in the LED array obtained by batch collection are input into the trained SVM model; the SVM model reconstructs a complete spectral curve from the monochromatic spectral data of a small amount of wavelengths according to the learned mapping relationship.

[0047] Furthermore, the step S2-3 specifically includes:

[0048] The SVM model outputs the complete spectrum of the LED array and simultaneously extracts the peak wavelength and half-width.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] 1. The AOTF spectrum scanning speed is fast, generally in the microsecond range, so the spectrum scanning of the entire spectrum is very rapid. The AOTF hyperspectral imager and camera timing design realize continuous batch high-speed acquisition of LED array spectra.

[0051] 2. During batch acquisition, the AOTF hyperspectral imager is set to collect only a small amount of monochromatic spectral data at a small wavelength, eliminating redundant data of the full spectrum.

[0052] 3. Compared with traditional training models, the SVM model has better generalization ability, uses kernel functions to deal with nonlinear mapping problems, maps data to a higher-dimensional feature space, and effectively establishes a mapping relationship from a small number of wavelengths to the full spectrum.

[0053] 4. The trained SVM model contains the learned mapping relationship from monochromatic spectrum data of a small number of wavelengths to full spectrum data. By batch collecting monochromatic spectrum data of a small number of wavelengths of a single LED chip and inputting them into the SVM model, high-precision full spectrum data of the LED array can be mapped. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention may be better understood by referring to the following description given in conjunction with the accompanying drawings, wherein the same or similar reference numerals are used throughout the drawings to represent the same or similar components. The accompanying drawings, together with the following detailed description, are included in and form a part of this specification and are used to further illustrate the preferred embodiments of the present invention and to explain the principles and advantages of the present invention. In the drawings:

[0055] Figure 1 This is a diagram of the system device of the present invention;

[0056] Figure 2 Schematic diagram of the AOTF hyperspectral imager frequency and camera timing set for the PC;

[0057] Figure 3 This is a schematic diagram of the batch collection process of LED array spectrum data of the present invention;

[0058] Figure 4 This is a timing diagram of machine learning model training during the specific implementation of batch acquisition of LED array spectrum data in the present invention;

[0059] The reference numerals are explained as follows: 1 is a constant current power supply, 2 is a microscope, 3 is an AOTF hyperspectral imager, 4 is a camera, 5 is a PC, 6 is an AOTF hyperspectral imager driver, 7 is a mechanical stage, 8 is an LED array, and 9 is a laser light source. DETAILED DESCRIPTION

[0060] Embodiments of the present invention will be described below with reference to the accompanying drawings. Elements and features described in one drawing or one embodiment of the present invention may be combined with elements and features shown in one or more other drawings or embodiments. It should be noted that for the purpose of clarity, the representation and description of components and processes that are not related to the present invention and are known to those of ordinary skill in the art are omitted in the drawings and descriptions.

[0061] As a specific example, see Figure 1The LED array spectral data batch acquisition system of the present application includes a constant current power supply 1 (current source), a microscope 2, an AOTF hyperspectral imager 3, a camera 4, a PC (computer) 5 (including a processor for batch acquisition), an AOTF hyperspectral imager driver 6 for driving the AOTF hyperspectral imager 3, and a laser light source 9 (laser source); the microscope includes a plurality of objective lenses and a mechanical stage 7; the AOTF hyperspectral imager is respectively connected to the interfaces of the camera and the microscope; a plurality of sample fixtures to be tested are placed on the mechanical stage of the microscope, and the knob of the mechanical stage is turned so that the sample to be tested (LED array 8) is facing the objective lens; the AOTF hyperspectral imager is pre-set with a specific driving frequency, and only light of a specific wavelength can be emitted from the light outlet of the AOTF hyperspectral imager; the emitted light is imaged on the camera through the lens.

[0062] The sample to be tested is powered on by a constant current power supply 1 to emit light or introduced into a laser light source 9 to form a surface light source for batch excitation.

[0063] The PC (computer) 5 is provided with a processor for processing batch acquisition, including a model training module and a batch acquisition module, which are respectively used to execute the model training process and the batch acquisition process in the following high-efficiency LED array spectral data batch acquisition method based on AOTF hyperspectral and machine learning.

[0064] Specifically, the high-efficiency LED array spectral data batch acquisition method based on AOTF hyperspectral and machine learning is as follows: Figure 3 As shown:

[0065] The model training process includes the following steps:

[0066] S1-1, set the AOTF hyperspectral imager and camera parameters to obtain LED full spectrum data;

[0067] S1-2, full spectrum data preprocessing: remove noise, normalize data, and obtain training set data;

[0068] S1-3, model architecture: number of iterations, data acquisition range, etc.;

[0069] S1-4, model training: data input, training iteration, model output;

[0070] S1-5. Evaluation Model: R 2 Evaluate the predictive accuracy of the model.

[0071] The batch collection process includes the following steps:

[0072] S2-1, batch acquisition: modify the AOTF hyperspectral imager and camera parameters, collect a small number of wavelengths, and batch collect LED array hyperspectral data;

[0073] S2-2, model mapping: batch collect a small number of wavelengths and input them into the trained SVM model;

[0074] S2-3. Output the complete spectrum, peak wavelength and half-width of the LED array.

[0075] The specific method of step S1-1 is as follows:

[0076] Set the camera acquisition mode to continuous acquisition in the trigger mode. The camera starts acquisition at a certain delay relative to the wavelength setting of the AOTF hyperspectral imager, and the sum of the exposure time and the delay time is less than the interval between the adjacent wavelength settings of the AOTF hyperspectral imager (such as Figure 2 Collect the AOTF hyperspectral data of the LED array to be tested, and obtain the full spectrum data of the chips in each array;

[0077] The specific method of step S1-2 is:

[0078] The spectrum contains noise and other uncertainties, so it is necessary to preprocess the acquired LED full spectrum to remove random noise in the image, smooth the image, and normalize the existing data to highlight the characteristic peaks in the spectrum to improve the training effect of the model and make subsequent machine learning training more effective. The preprocessed data is used as the training set;

[0079] The specific method of step S1-3 is:

[0080] like Figure 4 As shown, use the MLP model, initialize the model parameters: number of iterations, data acquisition range, select an appropriate weight initialization method, etc., and determine the number of hidden layers and neurons after evaluating the data;

[0081] The specific method of step S1-4 is:

[0082] The full spectrum data is used as input data for training, and is input into the one-dimensional vector SVM model classifier for training. The Gaussian kernel function is used as the kernel function of the SVM model. The Gaussian kernel function can handle nonlinear mapping relationships and performs well in reconstructing spectra in high-dimensional space. The activation value of the hidden layer is extracted as a feature vector; the SVM model optimizes the performance and generalization ability of this model by finding the optimal hyperplane and adjusting the bandwidth of the kernel function, and then maps the monochromatic spectrum data of a small amount of wavelength covering the LED emission peak to the data points corresponding to the full spectrum. According to the principle of spectral superposition, the monochromatic LED spectral distribution can be linearly superimposed to generate a new spectral distribution and generate continuous full spectrum data. The mapping relationship of the SVM model can be expressed by the following formula:

[0083]

[0084] Among them, K(x·x i ) is the kernel function, x i are training data points, and b * is a model parameter. The SVM model can reconstruct the complete spectrum curve by learning the mapping relationship between the input data and the target data. LED spectrum data is usually smooth, and the intensity distribution of monochromatic light can often reflect the intensity trend of adjacent wavelengths. SVM captures these nonlinear relationships through this method and realizes the inference of the complete spectrum from partial wavelengths. The SVM model is trained with multiple complete spectrum data to learn the mapping relationship between the input spectrum points and the complete spectrum. In this way, in the inference stage, even if only a few wavelengths of light intensity are input, the model can reconstruct the complete spectrum using the patterns learned in training. The complete spectrum curve is reconstructed by the SVM model, and after eliminating the noise caused by the equipment or environment, the peak wavelength is obtained according to the maximum position of the detected spectrum curve, and the half-peak width of the spectrum curve is obtained by the half-height width. After the SVM model training is completed, the optimized model parameters, including support vectors, kernel function types and their parameters, bias values, and related preprocessing steps, are serialized and saved. The model file is stored using Pickle to ensure its cross-platform compatibility in subsequent applications. The saved model file has the ability to nonlinearly map from the input spectrum data points to the complete spectrum curve, and can be loaded into the online analysis platform to achieve efficient spectrum reconstruction and key parameter extraction.

[0085] The specific method of step S1-5 is:

[0086] Calculate the R2 value to evaluate the performance of the SVM model, obtain the successfully trained SVM model, and enter the batch acquisition module stage.

[0087] In the batch collection process, the specific method of step S2-1 is:

[0088] Implement batch acquisition of LED arrays and adjust the driving frequency: modify the AOTF hyperspectral imager frequency to collect a small number of wavelengths, and modify the camera exposure time accordingly to collect a small number of wavelengths covering the LED emission peak. Move the mechanical stage to perform batch spectral data and image acquisition of the LED array.

[0089] The specific method of step S2-2 is:

[0090] The monochromatic spectral data of a small amount of wavelengths in the LED array obtained by batch collection are input into the trained SVM model. The mapping relationship learned in the model can reconstruct a complete spectral curve from the monochromatic spectral data of a small amount of wavelengths.

[0091] The specific method of step S2-3 is:

[0092] The model outputs the complete spectrum of the LED array and simultaneously extracts the peak wavelength and half-width.

[0093] In the specific implementation process, Figure 4 As shown, multiple calculations and iterations are performed between the input layer, hidden layer, and output layer, and finally the model is capable of generating a complete spectrum given the input. In the example, R2 is used to evaluate the model performance. The training set R2 is 0.9999781, which indicates that the model training results are good. In terms of acquisition parameters, the switching time of the AOTF hyperspectral imager linked camera is 0.05s / wavelength.

[0094] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0095] In addition, the method of the present invention is not limited to being executed in the time sequence described in the specification, and may also be executed in other time sequences, in parallel or independently. Therefore, the execution order of the method described in this specification does not limit the technical scope of the present invention.

[0096] Although the present invention has been disclosed above by describing specific embodiments of the present invention, it should be understood that all the above embodiments and examples are exemplary rather than restrictive. Those skilled in the art may design various modifications, improvements or equivalents of the present invention within the spirit and scope of the appended claims. These modifications, improvements or equivalents should also be considered to be included in the protection scope of the present invention.

Claims

1. A LED array spectrum data batch acquisition system, characterized by: The system comprises a current source, a laser source, a microscope, an AOTF hyperspectral imager, a camera and a processor for batch acquisition; the microscope comprises a plurality of objective lenses and a mechanical stage; the AOTF hyperspectral imager is respectively connected with interfaces of the camera and the microscope; a plurality of sample fixtures to be tested are placed on the mechanical stage of the microscope, and the knob of the mechanical stage is turned so that the sample to be tested faces the objective lens; the sample to be tested is excited in batches by the current source energizing and emitting light or by introducing a laser source to form a surface light source; the AOTF hyperspectral imager is pre-set with a specific driving frequency, and only light of a specific wavelength can be emitted from the light outlet of the AOTF hyperspectral imager; the emitted light is imaged on the camera through a lens; The processor is used to perform: adjusting the driving frequency of the AOTF hyperspectral imager, collecting images at different wavelengths, and outputting a complete spectrum of the LED array; The processor includes a model training module and a batch acquisition module. The model training module is used to build an SVM model. The batch acquisition module is used to acquire the LED array spectrum under partial wavelengths and input the LED array spectrum into the SVM model to obtain the complete LED array spectrum.

2. The LED array spectrum data batch acquisition system according to claim 1, characterized in that: The model training module includes the following units: LED array full spectrum data acquisition unit: set the parameters of AOTF hyperspectral imager and camera respectively to acquire the full spectrum data of LED array; Preprocessing unit: preprocess the acquired LED array full spectrum data to obtain training set data; preprocessing operations include noise removal, data standardization or normalization to ensure that all features are on the same scale; SVM model building unit: set the number of iterations and data range; Model training unit: the pre-processed LED array full spectrum data is divided into a feature matrix and a target vector, and the SVM model is used for training. According to the set number of iterations and data range, the SVM model parameters are iteratively updated to minimize the loss function. After the training is completed, the SVM model outputs the complete spectrum of the LED array and the peak wavelength and half-height width; Model evaluation unit: evaluates the prediction accuracy of the SVM model according to the R2 value; if the R2 value is low, returns to step S1-3 to adjust the number of iterations and the data range, and retrains the model until the prediction accuracy of the SVM model meets the requirements.

3. The LED array spectrum data batch acquisition system according to claim 1, characterized in that: The batch acquisition module includes the following units: Batch acquisition unit: Modify the parameters of AOTF hyperspectral imager and camera to collect LED array hyperspectral data at a small number of wavelengths; Model mapping unit: inputs a small amount of collected wavelengths into the SVM model obtained in the model training process; Model output unit: the SVM model outputs the complete spectrum, peak wavelength, and half-width of the LED array.

4. A method for batch collecting spectrum data of LED arrays, applied to the above-mentioned batch collecting system for spectrum data of LED arrays, characterized in that: It includes a model training process and a batch acquisition process. The model training process includes the following steps: Step S1-1, respectively setting the parameters of the AOTF hyperspectral imager and the camera to obtain full spectrum data of the LED array; Step S1-2, preprocessing the acquired LED array full spectrum data to obtain training set data; the preprocessing operation includes removing noise, data standardization or normalization processing to ensure that all features are on the same scale; Step S1-3, construct the SVM model, set the number of iterations and data range; Step S1-4, model training: the full spectrum data of the LED array preprocessed in step S1-2 is divided into a feature matrix and a target vector, and the SVM model constructed in step S1-3 is used for training. According to the set number of iterations and data range, the SVM model parameters are iteratively updated to minimize the loss function; after the training is completed, the SVM model outputs the complete spectrum of the LED array and the peak wavelength and half-height width; Step S1-5, evaluate the model: evaluate the prediction accuracy of the SVM model according to the R2 value; if the R2 value is low, return to step S1-3 to adjust the number of iterations and the data range, and re-execute the model training of step S1-4 until the prediction accuracy of the SVM model meets the requirements; The batch collection process includes the following steps: Step S2-1, batch acquisition: modify the parameters of the AOTF hyperspectral imager and the camera to collect LED array hyperspectral data at a small number of wavelengths; Step S2-2, model mapping: inputting a small amount of collected wavelengths into the SVM model obtained in the model training process; Step S2-3: The SVM model outputs the complete spectrum, peak wavelength, and half-width of the LED array.

5. The method for batch collecting spectrum data of LED array according to claim 4, characterized in that: The step S1-1 specifically includes: Set the camera acquisition mode to continuous acquisition in trigger mode. The camera starts acquisition at a certain delay relative to the wavelength setting of the AOTF hyperspectral imager, and the sum of the exposure time and the delay time is less than the setting interval of adjacent wavelengths of the AOTF hyperspectral imager. Collect the AOTF hyperspectral data of the samples to be tested, and obtain the full spectrum data of the samples to be tested in each array.

6. The method for batch collecting spectrum data of LED arrays according to claim 4, characterized in that: The step S1-2 specifically includes: The spectrum contains noise and other uncertainties, so the acquired full spectrum data needs to be preprocessed to remove random noise in the image, smooth the image, and normalize the existing data to highlight the characteristic peaks in the spectrum to improve the training effect of the model and make subsequent machine learning training more effective. The preprocessed full spectrum data is used as the training set.

7. The method for batch collecting spectrum data of LED arrays according to claim 4, characterized in that: The step S1-3 specifically includes: Use the multi-layer perceptron MLP model to initialize the model parameters, including the number of iterations, the data range, and the selection of an appropriate weight initialization method. After evaluating the data, determine the number of hidden layers and the number of neurons.

8. The method for batch collecting spectrum data of LED arrays according to claim 4, characterized in that: The steps S1-4 specifically include: The full spectrum data is used as input data for training and input into the one-dimensional vector support vector machine SVM classifier for training, and the activation value of the hidden layer is extracted as the feature vector; the SVM model optimizes the performance and generalization ability of the model by finding the optimal hyperplane and adjusting the bandwidth of the kernel function, and then maps the monochromatic spectrum data of a small amount of wavelength covering the LED light-emitting peak to the data points corresponding to the full spectrum to generate continuous full spectrum data; after the training is completed, the SVM model is generated. At this time, the SVM model has learned to extract feature values ​​from the input data and make high-precision full-spectrum restoration of the input monochromatic spectrum data of a small amount of wavelength.

9. The method for batch collecting spectrum data of LED arrays according to claim 4, characterized in that: The steps S1-5 specifically include: Calculate the R2 value to evaluate the performance of the SVM model, obtain the SVM model of the training results, and enter the batch acquisition module stage.

10. The method for batch collecting spectrum data of LED arrays according to claim 4, characterized in that: The step S2-1 specifically includes: Implement batch acquisition of LED arrays and adjust the driving frequency: modify the frequency of the AOTF hyperspectral imager to collect a small number of wavelengths, and modify the exposure time of the camera accordingly to collect monochromatic spectral data of a small number of wavelengths covering the LED emission peak; move the mechanical stage to collect batch spectral data and images of the LED array; The step S2-2 specifically includes: The monochromatic spectrum data of a small amount of wavelengths in the LED array obtained by batch collection are input into the trained SVM model; the SVM model reconstructs a complete spectrum curve from the monochromatic spectrum data of a small amount of wavelengths according to the learned mapping relationship; The step S2-3 specifically includes: The SVM model outputs the complete spectrum of the LED array and simultaneously extracts the peak wavelength and half-width.

Images