Spectrum detection model training and application method and electronic device

Abstract

The embodiment of the application provides a spectrum detection model training and application method and an electronic device. The training method comprises the following steps: obtaining spectrum data of multiple samples, constructing a training set and a test set according to the spectrum data of the multiple samples, wherein the multiple samples comprise multiple icing samples and multiple deicing samples, and the spectrum data comprises spectrum data obtained by performing multiple bombardments on each sample in the multiple samples based on a laser-induced breakdown spectroscopy (LIBS) technology; performing at least one iteration update on a pre-constructed basic model based on the training set and the test set until a spectrum detection model meeting a preset requirement is obtained. The application can effectively improve the accuracy and efficiency of spectrum detection by a model according to the similarity between spectrum data.

Description

Technical Field

[0001] This application belongs to the field of low-temperature insulation detection, and relates to spectral detection technology, particularly to a spectral detection model training and application method and electronic equipment. Background Technology

[0002] Silicone rubber composite insulators, with their excellent resistance to flashover, are widely used in power transmission lines. However, long-term outdoor operation leads to aging of the silicone rubber, necessitating testing to replace aging silicone rubber and ensure the normal operation of transmission lines. Current silicone rubber testing methods require insulator sampling, which is cumbersome, time-consuming, and cannot meet the needs of in-situ analysis, lacking efficient and accurate on-site testing technology. Laser-induced breakdown spectroscopy (LAS) uses pulsed lasers to create high-temperature, high-electron-density plasma, collecting optical signals for spectroscopic analysis, and has been widely used in various fields. However, in my country's extensive ultra-high-voltage (UHV) power lines operating in complex environments, silicone rubber may experience icing at low temperatures. This icing layer affects the interaction between the laser and the silicone rubber, reducing the accuracy of LAS detection. Summary of the Invention

[0003] This application provides a method for training and applying a spectral detection model, as well as an electronic device, which can solve the problem of low spectral detection accuracy caused by silicone rubber icing under low temperature conditions.

[0004] The first aspect of this application provides a method for training a spectral detection model, comprising: acquiring spectral data of multiple samples; constructing a training set and a test set based on the spectral data of the multiple samples, wherein the multiple samples include multiple iced samples and multiple de-iced samples; the spectral data includes spectral data obtained from each bombardment of each of the multiple samples by laser-induced breakdown spectroscopy (LIBS) technology; performing at least one iterative update on a pre-constructed basic model based on the training set and the test set until a spectral detection model that meets preset requirements is obtained, wherein any one of the at least one iterative updates includes: inputting two first spectral data corresponding to every two adjacent bombardments of the current training sample in the training set into the current basic model; using the current basic model to determine a first similarity between the two first spectral data; when the first similarity... When the target bombardment number is greater than or equal to a preset threshold, the target bombardment number is determined based on the value of any position corresponding to two adjacent bombardments in the multiple bombardments. When the target bombardment number determines that the current base model meets the first requirement of the preset requirements, the test set is used to determine whether the current base model meets the second requirement of the preset requirements. When the current base model does not simultaneously meet the first and second requirements, the parameters of the current base model are tuned and the next update is performed using the spectral data of the next training sample in the training set and the test set. Or, when the current base model simultaneously meets the first and second requirements, the target bombardment number is used as the de-icing bombardment number of the current training sample, and the current base model is used as the spectral detection model and the update is stopped.

[0005] According to an embodiment of this application, before constructing a training set and a test set based on the spectral data of the plurality of samples, the method further includes preprocessing the spectral data, the preprocessing including: removing the environmental background spectrum of each spectral data in the spectral data; performing noise reduction processing on the spectral data; determining the elemental spectral lines of each element contained in each sample according to the category of each sample, and extracting the data corresponding to the elemental spectral lines from the spectral data of each sample.

[0006] According to an embodiment of this application, the method for obtaining the environmental background spectrum includes: taking the spectrum of a preset time length in the spectral data of each sample as the environmental background spectrum, wherein the preset time length represents the time period within the preset time length after the start time node corresponding to each spectral data.

[0007] According to an embodiment of this application, determining whether the current base model meets the first requirement based on the target number of bombardments includes method one: when the current training sample is a de-iced sample after being bombarded by the target number of bombardments, the current base model is determined to meet the first requirement; or when the current training sample is an iced sample after being bombarded by the target number of bombardments, the current base model is determined not to meet the first requirement.

[0008] According to an embodiment of this application, determining whether the current base model meets the first requirement based on the number of target bombardments includes a second method: inputting two second spectral data corresponding to the number of target bombardments of the current training sample and the next adjacent bombardment into the current base model to obtain a second similarity between the two second spectral data; when the second similarity is greater than or equal to the preset threshold, determining that the current base model meets the first requirement; or when the second similarity is less than the preset threshold, determining that the current base model does not meet the first requirement.

[0009] According to an embodiment of this application, constructing a training set and a test set based on the spectral data of the plurality of samples includes constructing the test set based on the spectral data of the plurality of de-iced samples. Determining whether the current base model meets the second requirement of the preset requirements using the test set includes: inputting two third spectral data points corresponding to every two adjacent bombardments of the current test sample in the test set into the current base model to obtain a third similarity between the two third spectral data points; determining that the current base model meets the second requirement when the third similarity is greater than or equal to the preset threshold; or determining that the current base model does not meet the second requirement when the third similarity is less than the preset threshold.

[0010] According to an embodiment of this application, the formula used to determine the first similarity between the two first spectral data using the current base model includes:

[0011]

[0012] Where B represents the distance between the two first spectral data, μ i Σ represents the average vector of the i-th first spectral data among the two first spectral data. i Let represent the covariance matrix of the i-th first spectral data in the two first spectral data, where i = 1, 2.

[0013] According to an embodiment of this application, the base model includes a backpropagation neural network, and the parameter tuning of the current base model includes updating the number of neurons, activation function, and connection weights between neurons in each network layer of the backpropagation neural network.

[0014] A second aspect of this application provides a method for applying a spectral detection model. The method includes: acquiring the spectral data to be detected obtained from each bombardment of an item to be tested using laser-induced breakdown spectroscopy (LIBS) technology; inputting the preprocessed spectral data to be detected into a spectral detection model; using the spectral detection model to determine the similarity between two spectral data points corresponding to each two adjacent bombardments; and determining the number of de-icing bombardments corresponding to the item to be tested based on the similarity between the two spectral data points corresponding to each two adjacent bombardments, wherein the spectral detection model is obtained using a training method for the spectral detection model; and using the spectral data to be detected corresponding to the number of de-icing bombardments as the target spectral data of the item to be tested.

[0015] A third aspect of this application provides a spectral detection model training device, comprising: a construction module for acquiring spectral data of multiple samples and constructing a training set and a test set based on the spectral data of the multiple samples; and an iterative update module for performing at least one iterative update on a pre-constructed basic model based on the training set and the test set until a spectral detection model that meets preset requirements is obtained.

[0016] A fourth aspect of this application provides an electronic device, including: a memory and a processor, wherein the processor executes computer-readable instructions stored in the memory to implement the spectral detection model training method or the spectral detection model application method.

[0017] The spectral detection model training and application method provided in this application embodiment can utilize a laser to generate high-power, high-stability laser pulses to act on the surface of an ice-covered sample, forming a high-temperature, high-electron-density plasma on the surface. During the cooling process of the plasma, specific wavelength spectral lines are emitted. These specific wavelength spectral lines are collected using optical fibers, spectrometers, etc. The similarity between each pair of spectral data is determined by the Bhattacharyya distance. Based on this, a spectral detection model is trained to determine the number of de-icing bombardments required to melt the ice layer on the sample surface. Thus, the influence of surface ice is eliminated based on the number of de-icing bombardments, and the insulator is accurately detected. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram illustrating the application environment of a spectral detection model training method provided in an embodiment of this application.

[0020] Figure 2 This is a flowchart illustrating the spectral detection model training method provided in the embodiments of this application.

[0021] Figure 3 This is an example graph showing the variation of spectral intensity of iced and de-iced samples with the number of bombardments, provided in an embodiment of this application.

[0022] Figure 4 An example diagram illustrating the difference between two spectral data points corresponding to each two adjacent bombardments of an ice-covered sample, provided in an embodiment of this application.

[0023] Figure 5 This is a schematic diagram illustrating the process of any one of the at least one iterative updates to the base model provided in the embodiments of this application.

[0024] Figure 6 This is a flowchart illustrating the application method of the spectral detection model provided in the embodiments of this application.

[0025] Figure 7 This is a schematic diagram of a spectral detection model training device provided in an embodiment of this application.

[0026] Figure 8 This is a schematic diagram of a spectral detection model training device provided in an embodiment of this application.

[0027] Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and drawings of this application are used to distinguish similar objects, not to describe a specific order or sequence.

[0030] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.

[0031] Please see Figure 1 This is a schematic diagram illustrating the application environment of a spectral detection model training method provided in an embodiment of this application. Figure 1 As shown, the electronic device 10 communicates with the laser-induced breakdown spectroscopy system 20 via a network. The network can be a wired network or a wireless network. The wired network can be any of a local area network (LAN), a metropolitan area network (MAN), or a wide area network (WAN), and can be any of the following technologies: Wireless Fidelity (Wi-Fi), ZigBee Wireless Networks (ZigBee), Ultra Wideband (UWB), Universal Serial Bus (USB), etc.

[0032] Electronic device 10 can be a mobile phone, tablet computer, multimedia playback device, personal computer (PC), wearable device, or other electronic device.

[0033] The laser-induced breakdown spectroscopy system 20 includes a laser 201, a reflector 202, a convex lens 203, a delay controller 204, a collimating mirror 205, an optical fiber 206, and a spectrometer 207.

[0034] Laser 201 is a device that emits laser light to an object using the principle of stimulated emission. Specifically, laser 201 can excite an object by emitting pulsed laser light, causing some of the particles in the object to be excited to a higher energy state. When the number of particles in the higher energy state is greater than the number of particles in the lower energy state, the object can amplify the light radiation of a certain wavelength due to stimulated emission. This causes the light radiation of this wavelength to emit light radiation with a greater intensity than the incident light and the same wave position, frequency and direction as the incident light when it passes through the material.

[0035] The reflector 202 can be used to change the propagation path of the laser emitted by the laser 201. When the laser is directed towards the reflector 202, the reflector 202 will reflect the laser to the direction of the lens 203 as needed. The lens 203 can be used to focus the laser to a specified position, such as the surface of a sample (e.g., silicone rubber). The lens 203 can be a convex lens.

[0036] The delay controller 204 can communicate with the laser 201 and / or the spectrometer 207 to provide reliable time delay and trigger signals for the laser 201 and / or the spectrometer 207. The delay controller 204 can be the DG64, a high-precision pulse generator widely used in laboratory, electronic test, and measurement fields at Stanford Research Laboratory.

[0037] The collimating lens 205 and the optical fiber 206 can be used to construct a spectral light receiving device. The collimating lens 205 is used to focus light into a straight line based on the principle of light reflection and refraction, and the optical fiber 206 can transmit the light received by the collimating lens 205 to the spectrometer 207.

[0038] The spectrometer 207 is a device used to measure and analyze the spectrum emitted by a substance. The charge-coupled device (CCD) or intensified charge-coupled device (ICCD) detector in the spectrometer 207 can separate the light emitted by atoms, molecules, etc. in the substance into different wavelengths (colors) and perform photoelectric conversion to obtain the corresponding digital signals, which are then recorded in the form of a spectrum to obtain the spectrum of the substance.

[0039] In this embodiment, the laser-induced breakdown spectroscopy system 20 can be used to acquire spectral data of iced silicone rubber composite insulators using laser-induced breakdown spectroscopy (LIBS). Specifically, the process of acquiring spectral data of the sample using the laser-induced breakdown spectroscopy system 20 includes: a laser 201 generates a pulsed laser, which is focused onto the sample surface by a mirror 202 and a lens 203. The sample is ablated, excited, evaporated, and dissociated by the focused laser, forming a high-temperature, high-electron-density plasma on the sample surface. During the cooling process, the high-temperature, high-density plasma generates elemental electrical de-electrons through bremsstrahlung and recombination radiation, forming a continuous background spectrum. This process may last for several hundred nanoseconds. The delay controller 204 can be used to wait for this process to end before triggering the spectrometer 207, or to record the start and end times of this process for the spectrometer 207. The time nodes are defined as follows: bremsstrahlung refers to the reduction of kinetic energy of free electrons colliding with ions and emitting photons; recombination radiation refers to the capture and combination of free electrons by ions to form neutral ions and emitting photons; after significant decay of continuous radiation, the sample begins to emit a large number of atomic and ionic line spectra. Electrons in excited atoms and molecules transition between discrete bound energy levels and emit line spectra of corresponding wavelengths, i.e., atomic emission spectra; the spectral receiving device (collimating lens 205 and optical fiber 206) collects the light signals emitted by the plasma and transmits them to the spectrometer 207 for spectral dispersion. The CCD or ICCD detector in the spectrometer 207 completes photoelectric conversion and transmission to obtain the spectral data of the sample.

[0040] In this embodiment, the principle of performing spectral detection on a sample based on its spectral data includes: the wavelength position and intensity of the spectral lines corresponding to the type and content of each element in the sample are different. Therefore, by determining the spectral lines in the sample's spectral data, the elements in the sample can be identified and determined. Specifically, the spectral lines in the sample's spectral data can be compared with the standard spectral lines corresponding to each element in a preset atomic spectral database, and the elemental analysis results in the sample can be obtained based on the comparison results.

[0041] In one embodiment of this application, the training of the spectral detection model can be performed in the electronic device 10. Specifically, the electronic device 10 can acquire the spectral data of the sample from the laser-induced breakdown spectroscopy system 20 via a network, and train the spectral detection model based on the spectral data. After the spectral detection model is trained, the electronic device 10 acquires the spectral data of the item to be tested from the laser-induced breakdown spectroscopy system 20, uses the spectral detection model to determine the number of de-icing bombardments required for the item to be tested based on the spectral data, and uses the spectral data corresponding to the number of de-icing bombardments as the target spectral data of the item to be tested.

[0042] The following describes a computer program product that implements the spectral detection model training method of the embodiments of this application, running on an electronic device (such as...). Figure 1 The following explanation uses electronic device 10 as an example. Please refer to [link / reference]. Figure 2 The diagram shown is a flowchart illustrating the spectral detection model training method provided in an embodiment of this application. In one embodiment of this application, the method includes the following steps:

[0043] Step S301: Obtain spectral data of multiple samples, and construct a training set and a test set based on the spectral data of the multiple samples.

[0044] In some embodiments of this application, the plurality of samples includes a plurality of iced samples and a plurality of de-iced samples, and the spectral data includes spectral data obtained from each bombardment of each of the plurality of samples during multiple bombardments based on laser-induced breakdown spectroscopy (LIBS) technology.

[0045] In some embodiments of this application, to enhance the generalization ability of the trained spectral detection model, the multiple samples may include various categories of articles, such as insulating varnish, insulating adhesive, fiber products, rubber, plastics and their products, glass, ceramic products, mica, asbestos and their products, etc. The iced sample refers to an article with an ice layer of varying thickness on its surface, while the de-iced sample refers to an exposed article without an ice layer on its surface.

[0046] In some embodiments of this application, when using LIBS technology to bombard an iced sample with laser pulses (hereinafter referred to as bombardment) for spectral data acquisition, due to factors such as water vapor evaporating during the dissolution process of the ice layer under bombardment, the spectral data acquired between each two adjacent bombardments will change drastically before the ice layer completely dissolves and disappears. However, once the ice layer of the iced sample has completely disappeared, each subsequent bombardment hits the exposed surface of the sample, which is equivalent to bombarding the de-iced sample. At this time, the changes between the spectral data obtained between each two adjacent bombardments are smaller and tend to stabilize.

[0047] For further explanation of the above content, please refer to... Figure 3 and Figure 4 . Figure 3 This is an example graph showing the variation of spectral intensity of iced and uniced (or de-iced) samples with the number of bombardments, provided in the embodiments of this application. The curve in the upper left corner corresponds to the temperature at which the spectral data was acquired. It can be seen that the spectral intensity of the iced sample changes drastically between each two adjacent bombardments at low temperatures, while the change in spectral intensity between each two adjacent bombardments is smaller for the uniced (or de-iced) sample at room temperature.

[0048] refer to Figure 4 As shown, Figure 4An example diagram illustrating the difference between two spectral data points corresponding to every two adjacent bombardments of an icy sample, provided in an embodiment of this application, wherein the difference uses the Bhattacharyya distance between the two spectral data points corresponding to every two adjacent bombardments. Figure 4 It can be seen that in the first few bombardments of the ice-covered sample, due to the presence of the ice layer, the Bhattacharyya distance between the two spectral data corresponding to each two adjacent bombardments is large, that is, the difference is large; however, after bombardment to the point of melting the ice layer (hereinafter referred to as the de-icing bombardment number), subsequent bombardments all reach the bare surface of the sample, and the Bhattacharyya distance between the two spectral data corresponding to each two adjacent bombardments is small and the distance value tends to be stable.

[0049] Therefore, the difference between the spectral data of the iced sample obtained based on LIBS technology and the spectral data of the bare sample can be used to build a basic model based on the similarity of the spectral data under two adjacent bombardments. Then, a large amount of data can be used to update the basic model to obtain a spectral detection model that can accurately analyze the number of de-icing bombardments required to remove the ice layer, thereby eliminating the influence of the ice layer on the sample detection.

[0050] In some embodiments of this application, when acquiring the spectral data of each sample (e.g., iced sample and de-iced sample), the spectral data obtained from each bombardment and the corresponding position value can be recorded after each bombardment. For example, the spectral data obtained from the 10th bombardment can be acquired and the corresponding position value can be recorded as 10.

[0051] In some embodiments of this application, before constructing a training set and a test set based on the spectral data of the plurality of samples, the method further includes preprocessing the spectral data, the preprocessing including: removing the environmental background spectrum of each spectral data in the spectral data; performing noise reduction processing on the spectral data; determining the elemental spectral lines of each element contained in each sample according to the category of each sample, and extracting the data corresponding to the elemental spectral lines from the spectral data of each sample.

[0052] In some embodiments of this application, during each bombardment of the sample, the high-temperature, high-density plasma generated during cooling produces elemental electrical disconnections via bremsstrahlung and recombination radiation, forming a continuous background spectrum. This background spectrum formation process may last for hundreds of nanoseconds. To remove the influence of the background spectrum on the sample's spectral data, the background spectrum can be acquired first, and then removed from the sample's spectral data, thereby removing the background spectrum from each spectral data point.

[0053] In some embodiments of this application, the method for obtaining the environmental background spectrum includes: using the spectrum of a preset time length in the spectral data of each sample as the environmental background spectrum, wherein the preset time length represents the time period within the preset time length after the start time node corresponding to each spectral data. A delay controller can be used to record the start and end time nodes of the environmental background spectrum formation process, wherein the start time node of the environmental background spectrum formation process is the same as the start time node of the sample's spectral data; the duration of the environmental background formation process is determined based on the start and end time nodes as the preset time length.

[0054] In other embodiments of this application, an electronic device can be used to control a delay controller to trigger a spectrometer to collect spectral data of the sample at the end of the formation process of the ambient background spectrum. The spectral data of the sample collected in this way will not contain the ambient background spectrum and can be directly applied to subsequent steps.

[0055] In some embodiments of this application, the spectral data of the sample may contain a lot of noise. In order to improve the signal-to-noise ratio of the spectral data, the spectral data of the sample can be denoised. Specifically, the methods used to denoise the spectral data of the sample include, but are not limited to, a combination of one or more of the following methods: filter denoising, such as using a low-pass filter to remove high-frequency noise from the spectral data; statistical denoising, such as denoising by averaging multiple spectral data or using an autocorrelation function; moving average method, such as taking the arithmetic mean of the data within a preset range of preset sampling points in the spectral data as the smoothed value of that sampling point; wavelet threshold denoising, such as using wavelet decomposition to update the high-frequency coefficients in the spectral data to zero, thereby removing high-frequency noise, and then performing post-processing to obtain the denoised spectral data.

[0056] In some embodiments of this application, since the wavelength position and intensity of the spectral lines corresponding to the type and content of each element in the sample are different, the elemental spectral lines of each element contained in each sample can be determined according to the category of each sample in a preset atomic spectral database, and the data corresponding to the elemental spectral lines can be extracted from the spectral data of each sample, thereby further improving the signal-to-noise ratio and signal-to-background ratio of the spectral data.

[0057] In some embodiments of this application, the step of constructing a training set and a test set based on the spectral data of the plurality of samples includes: constructing the training set based on the spectral data of the plurality of iced samples, for example, selecting seven-tenths of the data from the spectral data of the plurality of iced samples to construct the training set; and constructing the test set based on the spectral data of the plurality of de-iced samples, for example, selecting seven-tenths of the data from the spectral data of the plurality of de-iced samples to construct the test set.

[0058] In other embodiments, the spectral data of the iced samples and the spectral data of the de-iced samples can be mixed together to construct the training set and the test set. For example, based on the training set and test set constructed using the above method, the remaining three-tenths of the spectral data of the multiple de-iced samples are added to the training set, and the remaining three-tenths of the spectral data of the multiple iced samples are added to the test set.

[0059] In some embodiments of this application, to facilitate the distinction between the spectral data of samples in the training set and the spectral data of samples in the test set, the samples corresponding to the training set can be defined as training samples, and the samples corresponding to the test set can be defined as test samples. To simplify the model training process, the training samples in subsequent embodiments only include icing samples.

[0060] Step S302: Based on the training set and the test set, the pre-built basic model is iterated and updated at least once until a spectral detection model that meets the preset requirements is obtained.

[0061] In some embodiments of this application, the base model includes a backpropagation neural network (BPNN). Specifically, the base model consists of an input layer, a hidden layer, and an output layer. Each layer consists of multiple neuron nodes, and each node is connected to all nodes in the next layer with corresponding connection weights. The output layer receives two spectral data points corresponding to every two adjacent bombardments of the current sample and passes these spectral data points to the hidden layer. The hidden layer performs nonlinear transformations and feature extraction on the spectral data output by the output layer. The output layer determines the distance or similarity between the two spectral data points and determines the number of de-icing bombardments for the current sample based on this distance or similarity.

[0062] In some embodiments of this application, since the model parameters of the basic model (e.g., the number of neurons in each network layer, activation function, connection weights between neurons, bias of neurons, etc.) are obtained during initialization, the model performance (e.g., prediction accuracy, etc.) of the basic model is low. The basic model can be iterated and updated at least once based on the training set and the test set. The parameters of the basic model are tuned through the iterative update process until a spectral detection model that meets the preset requirements is obtained.

[0063] In some embodiments of this application, the parameter tuning of the current base model includes updating the number of neurons, activation function, and connection weights between neurons in each network layer of the backpropagation neural network.

[0064] In some embodiments of this application, such as Figure 5 As shown, any one of the at least one iterative updates performed on the base model includes the following steps:

[0065] Step S401: Input the two first spectral data corresponding to each two adjacent bombardments of the current training sample in the training set into the current base model, and use the current base model to determine the first similarity between the two first spectral data.

[0066] In some embodiments of this application, the current model used during the first update is the base model, and the current model used in each subsequent update represents the model obtained after the last update of the base model. The current training samples in the training set used in any update include any training sample or any batch of training samples that have not been used in previous updates.

[0067] In some embodiments of this application, the principle upon which the spectral detection model determines the number of de-icing bombardments includes the similarity between two spectral data points corresponding to every two adjacent bombardments. The formula used to determine the first similarity between the two first spectral data points using the current base model includes:

[0068]

[0069] Where B represents the distance between the two first spectral data, μ i Σ represents the average vector of the i-th first spectral data among the two first spectral data. i Let represent the covariance matrix of the i-th first spectral data in the two first spectral data, where i = 1, 2.

[0070] Specifically, the above formula calculates the Bhattacharyya distance between two spectral data. Since the Bhattacharyya distance can be used to measure the similarity between two discrete or continuous distributions, the above formula can be used to determine the similarity between two spectral data regardless of whether the spectral data obtained after preprocessing is discrete or continuous.

[0071] In some embodiments of this application, a larger Bhattacharyya distance indicates a lower similarity between two spectral data points, and vice versa. The first similarity can be inversely proportional to the distance between two first spectral data points; for example, the first similarity can be set to 1 - B.

[0072] In other embodiments of this application, other distances can also be used to determine the similarity between two spectral data, such as Kullback-Leibler divergence, Jensen-Shannon divergence, Earth Mover's distance, TotalVariation, etc.

[0073] Step S402: When the first similarity is greater than or equal to a preset threshold, the number of target bombardments is determined based on the value of any position corresponding to the two adjacent bombardments in the multiple bombardments.

[0074] In some embodiments of this application, the two first spectral data corresponding to each two adjacent bombardments of the current training sample are continuously input into the current base model. The current base model is used to determine the first similarity between the two first spectral data corresponding to each two adjacent bombardments until a first similarity greater than or equal to a preset threshold (e.g., 0.7) is obtained. The target bombardment number is determined based on the value of any position corresponding to the two adjacent bombardments of the two first spectral data corresponding to the first similarity greater than or equal to the preset threshold in the multiple bombardments.

[0075] For example, the current base model determines that the first similarity between the two first spectral data corresponding to the first and second bombardments of the current training sample is 0.51, which is less than the preset threshold; the current base model is then used to determine that the first similarity between the two first spectral data corresponding to the second and third bombardments of the current training sample is 0.62, which is still less than the preset threshold; this process is repeated until the current base model determines that the first similarity between the two first spectral data corresponding to the 12th and 13th bombardments of the current training sample is 0.73, which is greater than the preset threshold; at this point, the target bombardment number is determined based on the value of either the 12th or 13th bombardment, for example, the target bombardment number is 12 or 13.

[0076] In some embodiments of this application, the method for determining the preset threshold may include: determining the average similarity among multiple spectral data obtained after bombarding the de-iced sample multiple times as the preset threshold.

[0077] In some embodiments of this application, since the similarity between the two spectral data of two adjacent bombardments corresponding to the target bombardment number is high, the target bombardment number is used as the number of bombardments at the interface between the ice layer and the sample (i.e., the number of de-icing bombardments). However, the number determined at this time may not be accurate enough, so it needs to be further judged in subsequent processes.

[0078] Step S403: When it is determined that the current base model meets the first requirement of the preset requirements based on the number of target bombardments, the test set is used to determine whether the current base model meets the second requirement of the preset requirements.

[0079] In some embodiments of this application, determining whether the current base model meets the first requirement based on the target number of bombardments includes method one: when the current training sample is a de-iced sample after being bombarded by the target number of bombardments, the current base model is determined to meet the first requirement; or when the current training sample is an iced sample after being bombarded by the target number of bombardments, the current base model is determined not to meet the first requirement.

[0080] In some embodiments of this application, the system can determine whether the current base model meets the first requirement by receiving the user's judgment on whether the current training sample is de-iced or iced after being bombarded the target number of times. The first requirement is essentially determining the accuracy of the current base model's prediction of the number of de-icing bombardments for the current training sample.

[0081] Specifically, if the current training sample is indeed de-iced after being bombarded according to the target bombardment number obtained by the current base model, then the current base model can be considered to have high accuracy in predicting the number of de-icing bombardments for the current training sample, thus satisfying the first requirement. If the current training sample is still iced after being bombarded according to the target bombardment number obtained by the current base model, then the current base model can be considered to have low accuracy in predicting the number of de-icing bombardments for the current training sample, thus not satisfying the first requirement.

[0082] In some embodiments of this application, determining whether the current base model meets the first requirement based on the number of target bombardments includes method two: inputting two second spectral data corresponding to the number of target bombardments of the current training sample and the next adjacent bombardment into the current base model to obtain a second similarity between the two second spectral data; when the second similarity is greater than or equal to the preset threshold, determining that the current base model meets the first requirement; or when the second similarity is less than the preset threshold, determining that the current base model does not meet the first requirement.

[0083] In some embodiments of this application, if the second similarity between two second spectral data is greater than or equal to the preset threshold, it can be considered that the current training sample is indeed a de-iced sample after bombardment corresponding to the target bombardment number obtained by the current base model, and the current base model can be considered to have high prediction accuracy for the de-icing bombardment number of the current training sample, thus satisfying the first requirement. If the second similarity between two second spectral data is less than the preset threshold, it can be considered that the current training sample is still an iced sample after bombardment corresponding to the target bombardment number obtained by the current base model, and the current base model can be considered to have low prediction accuracy for the de-icing bombardment number of the current training sample, thus not satisfying the first requirement.

[0084] In some embodiments of this application, in order to determine the prediction accuracy and generalization ability of the model, after determining that the current base model meets the first requirement, the test set can also be used to determine whether the current base model meets the second requirement in the preset requirements. This includes: inputting two third spectral data corresponding to each two adjacent bombardments of the current test sample in the test set into the current base model to obtain a third similarity between the two third spectral data; when the third similarity is greater than or equal to the preset threshold, determining that the current base model meets the second requirement; or when the third similarity is less than the preset threshold, determining that the current base model does not meet the second requirement.

[0085] In some embodiments of this application, the current test sample used in the above embodiments is a de-iced sample. The two third spectral data corresponding to each two adjacent bombardments of the de-iced sample are input into the current base model. If the third similarity between the two third spectral data is greater than or equal to the preset threshold, it can be considered that the current model has a high accuracy in the similarity test of the de-iced sample, thus meeting the second requirement. If the third similarity is less than the preset threshold, it can be considered that the current model has a low accuracy in the similarity test of the de-iced sample, thus not meeting the second requirement.

[0086] In other embodiments, an iced sample can be used as the current test sample. In this case, verifying whether the current base model meets the second requirement can include: inputting two fourth spectral data points corresponding to every two adjacent bombardments of the iced sample in the test set into the current base model to obtain a fourth similarity between the two fourth spectral data points; when the fourth similarity is greater than or equal to the preset threshold, determining the target bombardment number based on the value of any position corresponding to the two adjacent bombardments in the multiple bombardments; when the current base model is determined to meet the first requirement based on the target bombardment number, it is determined that the current base model meets the second requirement. In this case, the second requirement is actually to verify the accuracy of the current base model's prediction of the number of de-icing bombardments for the iced sample in the test sample, thereby improving the model's generalization ability.

[0087] Step S404: Determine whether the current basic model simultaneously meets the first requirement and the second requirement. If the current basic model does not simultaneously meet the first requirement and the second requirement, proceed to step S405; if the current basic model simultaneously meets the first requirement and the second requirement, proceed to step S406.

[0088] Step S405: Parameter tuning is performed on the current base model. Proceed to step S401: The next update is performed using the spectral data of the next training sample in the training set and the test set.

[0089] In some embodiments of this application, if the current base model does not simultaneously meet the first and second requirements, it indicates that the prediction accuracy and / or generalization ability of the current base model are insufficient, and parameter tuning of the current base model is required. The current base model is trained by using the spectral data of the next training sample in the training set and the test set, thereby continuously updating the parameters of the current base model to improve the prediction accuracy and / or generalization ability of the model.

[0090] Step S406: The number of target bombardments is taken as the number of de-icing bombardments of the current training sample, and the current base model is taken as the spectral detection model and updates are stopped.

[0091] In some embodiments of this application, if the current base model simultaneously meets both the first and second requirements, it indicates that the prediction accuracy and / or generalization ability of the current base model have met the requirements. The number of target bombardments can be used as the number of de-icing bombardments of the current training sample, and the current base model can be used as the spectral detection model, and updates can be stopped.

[0092] In some embodiments of this application, in order to further optimize the model, an iteration number can be set. If the set iteration number has not been reached after the current basic model simultaneously meets the first requirement and the second requirement, the current basic model can be updated again until the current basic model simultaneously meets the first requirement, the second requirement and the set iteration number.

[0093] The spectral detection model training method provided in this application utilizes a laser to generate high-power, high-stability laser pulses that act on the surface of an ice-covered sample, forming a high-temperature, high-electron-density plasma on the surface. During the plasma cooling process, specific wavelength spectral lines are emitted. These specific wavelength spectral lines are collected using optical fibers, spectrometers, etc. The similarity between each pair of spectral data is determined by the Bhattacharyya distance. Based on this, a spectral detection model is trained to determine the number of de-icing bombardments required to melt the ice layer on the sample surface. Thus, the influence of surface ice is eliminated based on the number of de-icing bombardments, enabling accurate detection of insulators.

[0094] In other embodiments of this application, in addition to constructing a model for predicting the number of de-icing bombardments based on the similarity between the spectral data corresponding to each two adjacent bombardments of an iced sample, a model can also be constructed for determining the difference between the spectral data of an iced sample and its corresponding de-iced sample. By training the model using two spectral data sets of each iced sample and its corresponding de-iced sample, the model can accurately predict the difference in spectral data between the iced sample and its corresponding de-iced sample based on the difference in their spectral data. Thus, the difference in spectral data can be removed from the spectral data of the iced sample to obtain the spectral data of the de-iced sample corresponding to the iced sample. Here, the de-iced sample corresponding to the iced sample refers to the sample of the exposed surface after the ice layer of the iced sample has melted and disappeared.

[0095] In some embodiments of this application, the training process of the spectral detection model has been described in detail above. The application method of using the spectral detection model for spectral detection will be described next. For example... Figure 6 As shown, the application method of the spectral detection model includes the following steps:

[0096] Step S501: Obtain the spectral data to be detected from each bombardment of the object to be tested during multiple bombardments based on laser-induced breakdown spectroscopy (LIBS) technology.

[0097] In some embodiments of this application, the electronic device 10 can acquire the spectral data of the item to be tested from the laser-induced breakdown spectroscopy system 20.

[0098] Step S502: Input the preprocessed spectral data to be detected into the spectral detection model, use the spectral detection model to determine the similarity between the two spectral data corresponding to each two adjacent bombardments, and determine the number of de-icing bombardments corresponding to the item to be detected based on the similarity between the two spectral data corresponding to each two adjacent bombardments.

[0099] In some embodiments of this application, the method used for preprocessing the spectral data to be detected can be referred to the description in step S301. When determining the similarity between two spectral data points corresponding to each two adjacent bombardments, the Bhattacharyya distance between the two spectral data points can be calculated. When determining the number of de-icing bombardments corresponding to the item to be detected, the value of any position corresponding to two adjacent bombardments with a similarity greater than or equal to the preset threshold can be used as the number of de-icing bombardments, as specifically described in step S402.

[0100] Step S503: Use the spectral data to be detected corresponding to the number of de-icing bombardments as the target spectral data of the item to be detected.

[0101] In some embodiments of this application, after being bombarded with the number of de-icing bombardments, the item to be tested is a de-iced item. That is, the spectral data to be tested corresponding to the number of de-icing bombardments is the spectral data of the exposed surface of the item to be tested. Therefore, it can be used as the target spectral data of the item to be tested, so that elemental analysis can be performed on the target spectral data to determine the elemental composition of the item to be tested.

[0102] In some embodiments of this application, if the type of the article to be tested is known, that is, the standard elemental composition of the article to be tested is known, but the elemental composition obtained by analysis of the target spectral data is not any of the standard elemental compositions, it can be considered that the article to be tested has undergone a qualitative change. For example, silicone rubber is affected by a combination of factors such as light, moisture, dirt, discharge and mechanical forces, and exhibits aging problems such as decreased hydrophobicity, increased hardness, fading, and powdering.

[0103] The spectral detection method provided in this application can be applied to the spectral detection of icy items to determine the state of the items.

[0104] In one example, for example Figure 7The diagram shown is an example of the flowchart of the spectral detection model training and application method provided in this application embodiment. LIBS detection is performed on different iced samples to obtain spectral data for each sample. The spectral data is then preprocessed, and the Bhattacharyya distance between the preprocessed spectral data is used to measure the similarity between two spectral data points corresponding to each two adjacent bombardments. Based on this, a discriminant model is trained to determine the number of de-icing bombardments corresponding to the interface between the ice layer and the sample. LIBS detection is performed on items in the test set to obtain spectral data for different items. This spectral data is then preprocessed and input into the trained discriminant model to obtain the number of de-icing bombardments, thereby removing the influence of icing on spectral detection.

[0105] Please see Figure 8 This is a schematic diagram of a spectral detection model training device provided in an embodiment of this application. It is provided to meet one of the purposes of this application and is a functional embodiment of the spectral detection model training method of this application. The spectral detection model training device includes: a construction module 61, used to acquire spectral data of multiple samples and construct a training set and a test set based on the spectral data of the multiple samples; and an iterative update module 62, used to perform at least one iterative update on the pre-constructed basic model based on the training set and the test set until a spectral detection model that meets preset requirements is obtained.

[0106] Another embodiment of this application also provides an electronic device. The computer program product implementing the spectral detection model training or application method of the embodiments of this application can run on any electronic device with sufficient computing power (such as...). Figure 9 The electronic device shown executes the various steps of the spectral detection model training or application method, thereby providing the function of spectral detection model training or application.

[0107] Please see Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 9 As shown, in one embodiment of this application, the electronic device 10 can be a mobile phone, tablet computer, smart wearable device, augmented reality (AR) / virtual reality (VR) device, laptop computer, netbook, etc. This application embodiment does not impose any restrictions on the specific type of electronic device 10.

[0108] like Figure 9As shown, the electronic device 10 may include, but is not limited to, a communication module 81, a memory 82, a processor 83, an input / output (I / O) interface 84, and a bus 85. The processor 83 is coupled to the communication module 81, the memory 82, and the I / O interface 84 via the bus 85.

[0109] Those skilled in the art will understand that the schematic diagram is merely an example of the electronic device 10 and does not constitute a limitation on the electronic device 10. It may include more or fewer components than shown, or combine certain components, or different components. For example, the electronic device 10 may also include network access devices, etc.

[0110] The communication module 81 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as Universal Serial Bus (USB) and Controller Area Network (CAN). The wireless communication module may provide one or more wireless communication solutions such as Wireless Fidelity (Wi-Fi), Bluetooth (BT), mobile communication networks, Frequency Modulation (FM), Near Field Communication (NFC), and Infrared (IR) technologies.

[0111] The memory 82 can be used to store computer-readable instructions and / or modules. The processor 83 implements various functions of the electronic device 10 by running or executing the computer-readable instructions and / or modules stored in the memory 82 and by calling the data stored in the memory 82. The memory 82 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device 10, etc. The memory 82 may include non-volatile and volatile memory, such as: hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other storage devices.

[0112] The memory 82 can be the external memory and / or internal memory of the electronic device 10. Furthermore, the memory 82 can be a memory in physical form, such as a memory stick, a TF card (Trans-flash Card), etc.

[0113] Processor 83 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. Processor 83 is the computational core and control center of electronic device 10, connecting various parts of electronic device 10 through various interfaces and lines, and executing the operating system of electronic device 10, as well as various installed application programs and program code.

[0114] For example, computer-readable instructions can be divided into one or more modules / submodules / units, which are stored in memory 82 and executed by processor 83 to complete this application. One or more modules / submodules / units can be a series of computer-readable instruction segments capable of performing a specific function, describing the execution process of the computer-readable instructions in electronic device 10. For example, computer-readable instructions can be divided into a building module 61 and an iterative update module 62.

[0115] If the modules / units integrated in the electronic device 10 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium, and when executed by a processor, they can implement the steps of the various method embodiments described above.

[0116] Computer-readable instructions include computer-readable instruction code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer-readable instruction code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), and random access memory (RAM).

[0117] Combination Figures 2 to 7 The memory 82 in the electronic device 10 stores computer-readable instructions, and the processor 83 can execute the computer-readable instructions stored in the memory 82 to achieve, for example, Figures 2 to 7 The training or application method of the spectral detection model is shown.

[0118] Specifically, the specific implementation method of the processor 83 for the above-mentioned computer-readable instructions can be found in [reference]. Figures 2 to 7 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0119] I / O interface 84 is used to provide a channel for user input or output. For example, I / O interface 84 can be used to connect various input / output devices, such as a mouse, keyboard, touch device, display screen, etc., so that users can enter information or visualize information. I / O interface 84 can also be used to connect to laser-induced breakdown spectroscopy system 20.

[0120] Bus 85 is used at least to provide a channel for communication between communication modules 81, memory 82, processor 83, and I / O interface 84 in electronic device 10.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0122] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0123] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0124] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this application. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0125] Furthermore, it is clear that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices can also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of this application without departing from the spirit and scope of the technical solutions of this application.

Claims

1. A method for training a spectral detection model, characterized in that, The method includes: Acquire spectral data of multiple samples, and construct training and testing sets based on the spectral data of the multiple samples. The multiple samples include multiple iced samples and multiple de-iced samples. The spectral data includes spectral data obtained from each bombardment of each of the multiple samples by laser-induced breakdown spectroscopy (LIBS) technology. The pre-built basic model is iteratively updated at least once based on the training set and the test set until a spectral detection model that meets the preset requirements is obtained, wherein any one of the at least one iterative updates includes: Input the two first spectral data corresponding to each two adjacent bombardments of the current training sample in the training set into the current base model, and use the current base model to determine the first similarity between the two first spectral data. When the first similarity is greater than or equal to a preset threshold, the target bombardment number is determined based on the value of any position corresponding to two adjacent bombardments in the multiple bombardments; and When it is determined based on the number of target bombardments that the current base model meets the first requirement of the preset requirements, the test set is used to determine whether the current base model meets the second requirement of the preset requirements; When the current base model does not simultaneously meet the first and second requirements, the parameters of the current base model are tuned, and the next update is performed using the spectral data of the next training sample in the training set and the test set; or When the current base model simultaneously satisfies both the first and second requirements, the number of target bombardments is taken as the number of de-icing bombardments for the current training sample, and the current base model is taken as the spectral detection model and updates are stopped.

2. The spectral detection model training method according to claim 1, characterized in that, Before constructing training and testing sets based on the spectral data of the multiple samples, the method further includes preprocessing the spectral data, the preprocessing including: Remove the ambient background spectrum from each spectral data point in the spectral data; The spectral data is then denoised. The elemental spectral lines of each element contained in each sample are determined according to the category of each sample, and the data corresponding to the elemental spectral lines are extracted from the spectral data of each sample.

3. The spectral detection model training method according to claim 2, characterized in that, The method for obtaining the environmental background spectrum includes: The spectrum of a preset time length in the spectral data of each sample is used as the environmental background spectrum, wherein the preset time length represents the time period within the preset time length after the start time node corresponding to each spectral data.

4. The spectral detection model training method according to claim 1, characterized in that, Determining whether the current base model meets the first requirement based on the number of target bombardments includes method one: When the current training sample is de-iced after being bombarded the target number of times, it is determined that the current base model meets the first requirement; or If the current training sample becomes an icy sample after being bombarded by the target a certain number of times, it is determined that the current base model does not meet the first requirement.

5. The spectral detection model training method according to claim 1, characterized in that, Determining whether the current base model meets the first requirement based on the number of target bombardments includes method two: The target bombardment number of the current training sample and the two second spectral data corresponding to the next adjacent bombardment are input into the current base model to obtain the second similarity between the two second spectral data. When the second similarity is greater than or equal to the preset threshold, it is determined that the current base model meets the first requirement; or When the second similarity is less than the preset threshold, it is determined that the current base model does not meet the first requirement.

6. The spectral detection model training method according to claim 1, characterized in that, The step of constructing the training set and test set based on the spectral data of the multiple samples includes constructing the test set based on the spectral data of the multiple de-iced samples, and the step of using the test set to determine whether the current base model meets the second requirement in the preset requirements includes: Input the two third spectral data corresponding to each two adjacent bombardments of the current test sample in the test set into the current base model to obtain the third similarity between the two third spectral data. When the third similarity is greater than or equal to the preset threshold, it is determined that the current base model meets the second requirement; or When the third similarity is less than the preset threshold, it is determined that the current base model does not meet the second requirement.

7. The spectral detection model training method according to claim 1, characterized in that, The formula used to determine the first similarity between the two first spectral data using the current base model includes: Where B represents the distance between the two first spectral data, μ i Σ represents the average vector of the i-th first spectral data in the two first spectral data. i Let represent the covariance matrix of the i-th first spectral data in the two first spectral data, where i = 1, 2.

8. The spectral detection model training method according to claim 1, characterized in that, The base model includes a backpropagation neural network, and the parameter tuning of the current base model includes updating the number of neurons, activation function, and connection weights between neurons in each network layer of the backpropagation neural network.

9. A method for applying a spectral detection model, characterized in that, The method includes: Acquire the spectral data of the test object obtained from each bombardment during multiple bombardments of the test object based on laser-induced breakdown spectroscopy (LIBS) technology. The preprocessed spectral data to be detected is input into the spectral detection model. The spectral detection model is used to determine the similarity between the two spectral data corresponding to each two adjacent bombardments. The number of de-icing bombardments corresponding to the item to be detected is determined based on the similarity between the two spectral data corresponding to each two adjacent bombardments. The spectral detection model is obtained using the spectral detection model training method as described in any one of claims 1 to 8. The spectral data to be detected corresponding to the number of de-icing bombardments is used as the target spectral data of the item to be detected.

10. An electronic device, characterized in that, The electronic device includes a memory and at least one processor, the memory storing at least one instruction, the at least one instruction being controlled by the at least one processor. The processor executes the spectral detection model training method as described in any one of claims 1 to 8. Alternatively, it can implement the spectral detection model application method as described in claim 9.

Images