Method and device for evaluating purity of interlayer gold plate based on MLPM model, equipment and medium

Abstract

The application discloses a kind of based on MLPM model's pulse eddy current sandwich gold plate purity evaluation method, device, equipment and medium.The method comprises the following steps: collecting original data;Standardize original data, obtain standardized data;Original data and standardized data are input into the two channels of MLPM model for data fusion, then the linearized fusion data is used to train MLPM model, obtain the trained MLPM model;Collect the data of gold to be measured;The collected data are input into the trained MLPM model, and output second prediction vector;Second prediction vector is converted into second one-hot code, and the number of corresponding gold to be measured is obtained by second one-hot code.The application uses double channel of its own input layer, simultaneously trains original data and standardized data, cancels the influence of probe temperature fluctuation on pulse eddy current signal at algorithm level, without changing hardware equipment, and enhances the robustness of model to temperature change.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pulse eddy current metal detection using deep learning methods, and particularly relates to a laminated gold plate purity evaluation method, device, equipment and medium based on an MLPM model. BACKGROUND

[0002] Gold, as a precious noble metal, is widely used in jewelry, electronics, medical and many other fields, playing an important role in financial transactions and the jewelry industry. However, in recent years, there have been a large number of reports of gold adulteration, which not only has a serious impact on the stability and credibility of the gold market, but also causes huge losses to consumers and enterprises. Therefore, detecting the purity of gold, i.e. the gold adulteration rate, has become a very important task.

[0003] Pulsed eddy current detection (PECT) as a non-destructive testing technology has been widely used in many industries and special fields, especially in the detection of defects, thickness measurement, material identification and other scenes of metal components. Pulsed eddy current detection technology can obtain information of different depths of the measured conductive object by transmitting a broadband pulse excitation signal to the detection coil, improving the detection efficiency; at the same time, by increasing the instantaneous magnetic field strength, it can penetrate multiple layers of conductive materials for detection, so it is suitable for gold purity detection of laminated gold plates.

[0004] In view of the problems such as insufficient precision of traditional methods in gold purity detection, machine learning methods provide core technical support for accurate purity determination and play a crucial role. In existing PECT applications, traditional machine learning algorithms have been widely used. By collecting and analyzing a small amount of labeled data, a classification model can be established to determine whether the gold sample is adulterated with other substances. The model can be constructed through feature extraction and ML (such as SVM, RF, etc.) or DL (such as RNN, ResNet). The classification model can help identify whether other substances are adulterated in the gold sample and estimate the purity of the gold sample.

[0005] Usually, these studies adopt a two-step approach. First, the original pulsed eddy current (PEC) signal is subjected to feature transformation or extraction processing, including principal component analysis, Rihaczek distribution time-frequency analysis, Hilbert-Huang transform and other methods. Subsequently, in order to achieve the final task of purity detection and classification, the generated feature representation is input into classification or clustering algorithms, such as support vector machines (SVM), random forests (RF), k-nearest neighbors, decision trees and naive Bayes methods. These traditional machine learning algorithms still have important value in PECT research.

[0006] Compared with traditional machine learning methods, deep learning can automatically extract features and classify through the construction of deep neural network models, thereby achieving more accurate prediction and decision-making. In the field of eddy current testing, due to the large amount of signal data and complex characteristics, the use of deep learning technology can effectively solve these technical problems.

[0007] When detecting by using pulse eddy current technology, there are various interferences, such as the temperature change of the probe of the pulse eddy current device, which will cause differences in the detection results, so that the detection results will be different when the pulse eddy current detection is carried out at different temperatures, thereby reducing the accuracy of the results. Therefore, by improving the deep learning model, the influence of the probe temperature fluctuation on the pulse eddy current detection is solved, which is of great significance for the detection of gold purity. SUMMARY

[0008] The purpose of the present application is to provide a sandwich gold plate purity evaluation method, device, equipment and medium based on MLPM model, to solve the influence caused by probe temperature fluctuation in the process of pulse eddy current detection, and improve the classification accuracy of the MLPM model.

[0009] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a sandwich gold plate purity evaluation method based on MLPM model, the method comprising the following steps:

[0010] S1: collecting original data; wherein the original data includes a first pulse eddy current signal and a first probe temperature, the first pulse eddy current signal is the signal returned by the test gold to the pulse eddy current detection device after the pulse eddy current detection device sends a pulse waveform to the test gold, and the first probe temperature is the probe temperature of the pulse eddy current detection device;

[0011] S2: standardizing the original data to obtain standardized data;

[0012] S3: inputting the original data into the addition layer of the MLPM model through the first channel of the MLPM model, inputting the standardized data into the addition layer of the MLPM model through the second channel of the MLPM model, and fusing the original data and the standardized data in the addition layer to obtain fused data;

[0013] S4: linearizing the fused data, training the MLPM model using the linearized fused data, and obtaining a trained MLPM model;

[0014] S5: collecting a second pulse eddy current signal corresponding to the test gold and a second probe temperature of the pulse eddy current detection device; inputting the second pulse eddy current signal and the second probe temperature into the trained MLPM model, and outputting a second prediction vector;

[0015] S6: converting the second prediction vector into a second one-hot code, obtaining a corresponding number of the to-be-tested gold from the second one-hot code, and using the number to reflect the purity of the gold.

[0016] Further, in the S2, the standardization is z-score normalization, and the formula for standardizing the original data is:

[0017]

[0018] is the original data collected, i is the number of dimensions, j is the data amount of each dimension, x ji is the jth sample data on i dimensions, μ i and σ i are the mean and standard deviation of each dimension X i = (x 1i , x 2i , x 3i, , …, x ji ), and z ji represents the normalized data of the jth sample data on i dimensions.

[0019] Further, the original data is time series data.

[0020] Further, in the S3, the fused data is calculated by the following method:

[0021] Fused data = r x standardized data + (1-r) x original data,

[0022] wherein r is a proportion parameter, and is preset weight information.

[0023] Further, the proportion parameter r ranges from 0 to 1, and the optimal proportion parameter r is 0.5.

[0024] Further, in the S4, training the MLPM model includes performing stacked feature extraction on the linearized data to output a first prediction vector; using an argmax function to convert the first prediction vector into a first one-hot code, and obtaining the purity predicted by the MLPM model according to the first one-hot code, thereby evaluating the prediction accuracy of the model.

[0025] Further, the number of stacked layers of the feature extraction is Block number, and setting the Block number includes presetting the Block number to obtain the prediction accuracy of the MLPM model; updating the Block number to obtain the prediction accuracy of the updated MLPM model; comparing the model prediction accuracy to obtain the Block number when the optimal prediction accuracy is obtained.​​

[0026] Based on the same concept, the application also provides a sandwich gold plate purity evaluation device based on an MLPM model, which comprises:

[0027] A collection module is configured to collect original data, wherein the original data comprises a first pulsed eddy current signal and a first probe temperature.

[0028] A standardization module is configured to standardize the original data to obtain standardized data.

[0029] A data fusion module is configured to fuse the original data and the standardized data.

[0030] An MLPM model training module is configured to train the MLPM model to obtain a trained MLPM model.

[0031] A gold purity prediction module is configured to predict the purity of a to-be-tested gold and output a prediction vector.

[0032] A gold purity result identification module is configured to obtain a purity result of the to-be-tested gold according to the prediction vector.

[0033] Based on the same concept, the application also provides a computer device,

[0034] The device comprises:

[0035] A memory is configured to store a computer program.

[0036] A processor is configured to execute the computer program to implement the sandwich gold plate purity evaluation method based on the MLPM model.

[0037] Based on the same concept, the application also provides a computer readable storage medium,

[0038] The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the sandwich gold plate purity evaluation method based on the MLPM model.

[0039] Compared with the prior art, the application has the following beneficial effects:

[0040] The application introduces a double-input channel structure, deforms an MLP-Mixer neural network model to obtain an MLPM model, the first channel of the MLPM model inputs original data, and the second channel inputs standardized data, so that the standardized data and the original data can be processed at the same time, the influence of probe temperature fluctuation on the pulsed eddy current signal is offset at the algorithm level, without changing the hardware device, and the robustness of the model to temperature changes is enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the following embodiment description will be briefly introduced. Obviously, the drawings in the following description are only one embodiment of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0042] Figure 1 is a flowchart of the interlayer gold plate purity evaluation method based on the MLPM model in the embodiment of the present application;

[0043] Figure 2 is a PEC signal diagram of the same gold sample measured by the pulsed eddy current device under the conditions of low temperature and high temperature of the probe in the embodiment of the present application;

[0044] Figure 3 is an architecture diagram of the MLPM neural network in the embodiment of the present application;

[0045] Figure 4 is a typical image of the gold PEC signal detected by the device in the embodiment of the present application;

[0046] Figure 5 is a confusion matrix diagram of the prediction result of MLPM-60 in the embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the present application will be described clearly and completely in the following description of the drawings in the embodiment of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0048] The technical solutions of the present application will be described in detail in the following specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0049] Embodiment 1

[0050] Figure 1 The flowchart of the interlayer gold plate purity evaluation method based on the Multi-input Channel MLP-Mixer (MLPM) model provided by the embodiment of the present application is shown. As shown in Figure 1 The interlayer gold plate purity evaluation method includes the following steps:

[0051] S1: Collecting original data.

[0052] The probe temperature of the pulsed eddy current detection device has an impact on the PEC amplitude, which means that the probe temperature should be kept in a stable range during the acquisition process to avoid too much deviation of the PEC signal of the same sample measured each time, so that the model trained at the previous probe temperature fails to predict accurately at different probe temperatures or even fails.

[0053] Since it is not easy to suppress the impact of the probe temperature at the hardware level, the embodiment of the present application considers improving it at the algorithm level. To this end, the deep learning technology is used to realize the training and prediction of the model. By constructing a deep neural network model, a large amount of sample data is used to train the model, so that it can accurately predict the eddy current detection signal at different temperatures. At the same time, appropriate data processing and feature extraction methods can further improve the performance and accuracy of the model. With the help of deep learning technology, high-precision and high-efficiency nondestructive detection of pulsed eddy current detection is finally realized, which provides more reliable protection for industrial production.

[0054] The MLPM model of the embodiment of the present application indirectly eliminates the impact of the probe temperature on the data collected by the device during pulsed eddy current detection through a double-input channel mode, directly eliminates the impact of the probe temperature fluctuation from the algorithm level, and does not need to change the hardware device.

[0055] As shown in Figure 2 It can be seen that the amplitude size and the probe temperature show correlation during the acquisition of the same gold sample. Therefore, the probe temperature is also used as the data for model training.

[0056] A large number of PEC signals and probe temperatures of various gold samples are collected by the pulsed eddy current detection device as input data of the MLPM model, wherein the PEC signal is the signal returned by the gold sample to the pulsed eddy current detection device after the pulsed eddy current detection device sends a pulse waveform to the gold sample, and the probe temperature is the probe temperature of the pulsed eddy current detection device, which is obtained through software monitoring of the temperature control module of the pulsed eddy current detection device. The PEC signal is a 1000-dimensional time series, the probe temperature is a 1-dimensional time series, and the input feature is a 1001-dimensional feature vector.

[0057] The purity and other properties of each gold are shown in Table 1. The PEC data of 13 kinds of gold samples with different purity and thickness are collected by using the pulsed eddy current detection device. The training set and the test set both include the data of 13 kinds of gold samples. In the training set, the number of sample data of each gold sample is 800. In the test set, the number of sample data of each gold sample is 400.

[0058] Table 1: Gold purity attribute table

[0059]

[0060] During network training, non-overlapping training and test sets were selected from the dataset, with a 2:1 ratio of training to test set data sizes. Their dimensions were (10400, 1001) and (5200, 1001), respectively. The data size for each gold sample number was the same in both the training and test sets. The time-varying amplitude of the induced pulse eddy currents generated in the tested samples is shown in the image below. Figure 4 As shown.

[0061] S2: Standardize the original data to obtain standardized data.

[0062] By utilizing the dual-channel feature of its input layer, the MLPM model is trained by inputting standardized data and original data respectively, enabling it to acquire a wider range of data feature information, thereby solving the problem of offset error caused by probe temperature.

[0063] The training and test sets were standardized using Z-score normalization, ensuring that the mean was zero and the standard deviation was 1 across all dimensions. The resulting training data was then used for network training, and the test set was used for model performance evaluation.

[0064] The standardization method uses Z-score normalization, which is calculated as follows:

[0065] ,

[0066] in, The original data collected, where i is the number of dimensions, j is the amount of data in each dimension, and x is the number of dimensions. ji It is the j-th sample data in the i-th dimension, μ i and σ i Each dimension X is a separate dimension. i = (x 1i ,x 2i ,x 3i, …,x ji The mean and standard deviation of z on the z-axis. ji This represents the normalized data of the j-th sample across i dimensions. Where X... i = (x i1 ,x i2 ,x i3 …x ij ), is the tensor flattened when the dimension is i, where the number of dimensions i is 1001, j is 10400 in the training set, and j is 5200 in the test set. i and σ i They are X j = (x 1j ,x2j ,x 3j, …,x ij The mean and standard deviation of z ij This represents the normalized data of the j-th sample data in the i-th dimension.

[0067] Where, μ i and σ i The results are obtained using the following formulas:

[0068] ,

[0069] ,

[0070] In the training set, N=10400, and in the test set, N=5200.

[0071] Among them, such as Figure 4 As shown, dimension i is obtained by dividing the first 100 milliseconds of the pulsed eddy current decay signal into 1000 nodes with an interval of 0.1 milliseconds. The amplitude signal data corresponding to the 1000 nodes are used as the first 1000 dimensions, and the temperature signal data is used as the 1001st dimension. j represents the sample data of the gold sample.

[0072] S3: Input the raw data into the addition layer of the MLPM model through the first channel of the MLPM model, and input the standardized data into the addition layer of the MLPM model through the second channel of the MLPM model. In the addition layer, the raw data and the standardized data are fused to obtain fused data.

[0073] The MLP-Mixer model is an innovative neural network architecture proposed in 2021 that performs exceptionally well in image classification tasks. Because each operation in the MLP-Mixer model is an interpretable linear and nonlinear transformation, the network's decision-making process and internal representation are easier to understand. Its innovative architecture is also considered to have potential applications in other fields.

[0074] The MLP-Mixer model is designed for image classification tasks and processes two-dimensional data. The sandwich gold purity evaluation aims to directly classify the PEC signals of different purities of gold obtained by the detection equipment. Based on the concept of classifying time series data, the existing MLP-Mixer model architecture is modified. Unlike the MLP-Mixer model used to process two-dimensional image data, the MLP-M model is set to be dedicated to processing time series data, so that the model architecture matches the essential characteristics of the data. The modified model is called the MLP-M model, which removes the encoding layer used to disassemble two-dimensional data and replaces it with a linear layer for double-channel input. The two channels input normalized data and original data respectively. Some hidden layers that are not suitable for sequence processing (originally used to process two-dimensional data) are deleted to make them suitable for processing time series similar to PEC signals.

[0075] Compared with the single input channel of MLP-Mixer, the MLP-M double input channel can make full use of the effective information in the original data and the standardized data as much as possible, so that the model can learn the statistical characteristics and original information of the data at the same time, thereby better capturing the key features of the data.

[0076] As shown in Figure 3 , the input data is first processed by the double-channel input layer. The input data is copied into two parts and input into two channels. One of the two parts of data is standardized, and the other part is not processed. Two linear layers are used to linearize the standardized data and the original data respectively. Then the data of the two channels is fused through an addition layer, and then processed by a linear layer. This allows the MLP-M to process two input data of different dimensions simultaneously and fuse their information.

[0077] A proportion parameter r is provided in the addition layer to allocate the weights of the data in the two channels, so as to flexibly control the influence of each input on the output. This allows the model to fine-tune according to the importance of different features, improving the model's expression ability and adaptability. Through weighted summation, input features can be automatically selected. If the data in one of the two channels has a smaller contribution to the task, its weight can be adjusted to a smaller value, thereby reducing its impact on the model. This operation helps to reduce the influence of redundant features and improve the generalization ability of the model. The main purpose of the MLP-M double input channel is to fuse the relevant information between the two input channels so that the neural network can better understand and process the features of the input data.

[0078] As shown in Figure 3As shown, the proportion parameter r is used to allocate the addition proportion of two-channel data: addition layer output = r x normalized data + (1-r) x original data, that is, weighted summation, in this way flexibility: by assigning different weights to different input features, the influence of each input on the output can be flexibly controlled to improve the prediction accuracy of the model.

[0079] S4: linearize the fused data, train the MLPM model using the linearized fused data to obtain a trained MLPM model.

[0080] The data fused by the addition layer is linearized by another linear layer, and the linearized data is stacked for feature extraction. Through the stacking feature extraction process, automatic feature extraction and classification of time series data can be realized, improving the model's ability to solve complex problems and enhancing the model's classification accuracy.

[0081] Stacked feature extraction includes first layer standardization of data, then linearization of layer-standardized data through a linear layer, introduction of nonlinearity using GeLU function as activation function, and finally linearization through another linear layer to obtain processed data. After stacking feature extraction of data, the processed data is layer standardized, and then the layer-standardized data is input into a linear layer to output a prediction vector, which is used for classification of each gold sample.

[0082] The number of stacked layers of stacked feature extraction is controlled by the number of blocks. That is, the depth and expression ability of the model are mainly controlled by the number of blocks, so the number of blocks determines the complexity and learning ability of the model. A larger number of blocks may increase the expression ability of the model, but it may also increase the risk of overfitting and computational complexity. The appropriate number of blocks needs to be selected according to the characteristics of the specific task and data set. Generally, more complex tasks and data sets may require more blocks to extract richer feature representations, while simpler tasks and data sets may be able to use fewer blocks. Therefore, selecting the appropriate number of blocks is important for the final performance and training efficiency of the model.

[0083] Setting the number of blocks includes presetting the number of blocks to obtain the prediction accuracy of the MLPM model, updating the number of blocks to obtain the prediction accuracy of the updated MLPM model, and comparing the prediction accuracy of the model to obtain the number of blocks when the optimal prediction accuracy is obtained. For ease of description in the following text, the network is named MLPM-N, as shown in Figure 3 , where N represents the number of blocks, for example, MLPM-16 represents an MLPM with 16 blocks.

[0084] In the embodiments of the present application, before the MLPM model is trained, setting various hyperparameters of the MLPM model and configuring a training environment of the MLPM model are further included. By setting appropriate parameters, the convergence and stability of the model can be guaranteed, and the generalization ability of the model can be improved. Specifically, the number of blocks is set to 60, the learning rate is initialized to 1 × 10 −3 , the number of training iterations epoch is 200. The Adam optimizer is used to update all weights and biases of the model, and the L2 regularization parameter is 1 × 10 −3 . The cross-entropy loss function is used. The gradient function is used to update the model parameters. The batch size is 120. When the change of Loss tends to be stable during the training process, it is considered that the model has converged. In order to prevent overfitting, early stopping operation is performed at this time to obtain the trained MLPM model.

[0085] The training process is coded using the Pytorch machine learning library and trained on an Nvidia GeForce RTX 4070Ti GPU. Among them, the training of MLPM-60 for 200 iterations takes 22 minutes and 16.1 seconds. As the number of blocks increases, the calculation of the model will be more complex, and the training time will also become longer.

[0086] After the converged MLPM model is trained using the training set, the test set is input into the model for testing. According to the top-1 prediction accuracy of the MLPM under different block numbers shown in Table 2, it can be found that as the number of blocks increases, the accuracy of the MLPM model first starts to improve, and the classification ability of the model will become stronger. However, when the number of blocks reaches 60, the accuracy of the model starts to decrease by continuing to increase the number of blocks, which means that the classification ability starts to weaken. Therefore, MLPM-60 is selected as the optimal model, and its accuracy reaches 99.9%, close to 100%. The confusion matrix of the prediction result is shown in Figure 5 , except that 3 pieces of No. 4 samples are incorrectly identified as No. 1 samples, the remaining samples are accurately predicted.

[0087] Table 2 Prediction results of MLPM models with different block numbers

[0088]

[0089] After determining the optimal Block number of the model, in order to determine the proportion parameter of the two channels of the MLPM to achieve the best classification effect, the proportion parameter r is continued to be tested, r on the interval [0, 1] is experimented, and the test accuracy of the model is used to judge the importance of the two channels respectively, for example, the test accuracy of the model when r = 0.7 is higher than that when r = 0.8, which indicates that the actual contribution of the standardized data may be more important than the current weight. Set the proportion parameter r to be 0, 0.1, 0.3, 0.5, 0.7, 0.9, and 1. Take the optimal MLPM-60 as the test model, and the test results of the proportion parameter r are shown in Table 3. Among them, r = 0 and r = 1 represent the results of only inputting the original data and only inputting the standardized data when the MLPM has only a single input channel. The results show that when only standardized data or original data is used as input data, the model accuracy decreases significantly; this shows that the dual-channel structure plays a key role in improving the robustness of the model and can effectively offset the influence of probe temperature fluctuations on PEC signals. According to the results, when r = 0.5, the prediction accuracy of the MLPM-60 model is the highest.

[0090] Table 3 Classification results

[0091]

[0092] Generally, the influence of the temperature difference of the probe on the results when collecting the training set and the test set is caused by the change of the probe temperature, which causes the directional deviation of the amplitude size measured by the device, such as the inability to distinguish between 99.90% and 99.99% gold with extremely similar purity. The MLPM has broken through this bottleneck, which is also a decisive factor for the MLPM to obtain higher accuracy than other neural networks. The dual-input channel of the MLPM extracts feature information from standardized data and original data at the same time, which can weaken the influence caused by the deviation, and make the feature spaces of the two data sets consistent.

[0093] S5: Collect the pulse eddy current signal corresponding to the gold to be tested and the probe temperature of the pulse eddy current detection device, and input it into the trained MLPM model to obtain a prediction vector.

[0094] The pulse eddy current detection equipment is used to collect the pulse eddy current signal corresponding to the to-be-detected gold and the probe temperature of the pulse eddy current detection equipment. In the specific detection process, the pulse eddy current signal is collected by placing the to-be-detected gold above the sensor of the pulse eddy current detection equipment, and then the pulse eddy current detection equipment generates a rectangular wave, which is amplified by a power amplification module and then loaded into an excitation coil in the sensor. The excitation coil thus generates a primary magnetic field with a rectangular time-varying intensity. When the primary magnetic field jumps, eddy current is induced in the gold plate. The induction and dissipation of the eddy current are affected by the purity of the gold plate. During the induction and dissipation of the eddy current, the eddy current generates a time-varying secondary magnetic field. The time-varying secondary magnetic field is converted into a voltage signal by a receiving coil in the sensor, amplified by a signal amplification module, and collected by a signal generation and collection dual-function module. The probe temperature of the pulse eddy current detection equipment is obtained by software monitoring the probe temperature.

[0095] The collected data is preprocessed to obtain time series data.

[0096] The preprocessed time series data is input into the MLPM model to obtain a prediction vector of the to-be-detected gold.

[0097] S6: The prediction vector of the to-be-detected gold is converted into a one-hot code by an argmax function, and the corresponding number of the to-be-detected gold is obtained through the one-hot code. According to the number, the corresponding purity value is obtained by querying Table 1 to evaluate the purity of the gold.

[0098] Embodiment 2

[0099] Based on the same concept, the present application also provides a sandwich gold plate purity evaluation device based on an MLPM model, which comprises:

[0100] The acquisition module is configured to acquire raw data, wherein the raw data comprises a first pulse eddy current signal and a first probe temperature.

[0101] The standardization module is configured to standardize the raw data to obtain standardized data.

[0102] The data fusion module is configured to fuse the raw data and the standardized data.

[0103] The MLPM model training module is configured to train the MLPM model to obtain a trained MLPM model.

[0104] The gold purity prediction module is configured to predict the purity of the to-be-detected gold and output a prediction vector.

[0105] The gold purity result identification module is configured to obtain the purity result of the to-be-detected gold according to the prediction vector.

[0106] Embodiment 3

[0107] Based on the same concept, the embodiment of the present application also provides a computer device, which comprises:

[0108] a memory for storing a computer program;

[0109] a processor for implementing the MLPM model-based interlayer gold sheet purity evaluation method of the above-mentioned embodiment 1 when executing the computer program.

[0110] Embodiment 4

[0111] Based on the same concept, the embodiment of the present application also provides a computer readable storage medium, which has a computer program stored thereon, and the computer program is executed by a processor to implement the MLPM model-based interlayer gold sheet purity evaluation method of the above-mentioned embodiment 1.

[0112] The present application uses the characteristics of the double channels of the input layer itself, inputs the standardized data and the original data into the two channels respectively for training, so that the model can obtain more extensive data feature information. The classification accuracy of the interlayer gold sheet is improved, and the offset error caused by the probe temperature of the pulse detection device is solved.

[0113] The content illustrated in the above-mentioned embodiments should be understood as the embodiments only for more clearly illustrating the present application, and should not be used for limiting the scope of the present application. After reading the present application, the modification of various equivalent forms of the present application by the person skilled in the art all falls within the scope defined by the claims attached to the present application.

Claims

1. A method for evaluating the purity of a sandwich gold plate based on an MLPM model, characterized by, The method comprises the following steps: S1: collecting original data; wherein the original data comprises a first pulsed eddy current signal and a first probe temperature, the first pulsed eddy current signal is a signal returned by a test gold to a pulsed eddy current detection device after the test gold generates eddy current in response to a pulsed waveform sent by the pulsed eddy current detection device, and the first probe temperature is a probe temperature of the pulsed eddy current detection device; S2: standardizing the original data to obtain standardized data; S3: inputting the original data into an addition layer of an MLPM model through a first channel of the MLPM model, inputting the standardized data into the addition layer of the MLPM model through a second channel of the MLPM model, and fusing the original data and the standardized data in the addition layer to obtain fused data; wherein the fused data is calculated in the following manner: fused data = r x standardized data + (1-r) x original data, r is a proportion parameter, and is preset weight information; S4: linearizing the fused data, training the MLPM model using the linearized fused data, and obtaining a trained MLPM model; S5: collecting a second pulsed eddy current signal corresponding to a test gold and a second probe temperature of the pulsed eddy current detection device, inputting the second pulsed eddy current signal and the second probe temperature into the trained MLPM model, and outputting a second prediction vector; S6: converting the second prediction vector into a second one-hot code, obtaining a corresponding number of the test gold through the second one-hot code, and using the number to reflect the purity of the gold; wherein the MLPM model comprises a double-channel input layer, a mixer module and a fourth linear layer connected in sequence; the double-channel input layer comprises a first channel, a second channel, an addition layer and a third linear layer; the first channel comprises a first linear layer, and the second channel comprises a second linear layer; the first linear layer is used to process the original data, and the second linear layer is used to process the standardized data; the output ends of the first linear layer and the second linear layer are connected to two input ends of the addition layer respectively, and the output end of the addition layer is connected to an input end of the third linear layer.

2. The MLPM model based interleaf gold sheet purity evaluation method according to claim 1, wherein, In the S2, the standardization is z-score normalization, and the formula for standardizing the original data is: ， where, is the raw data collected, i is the number of dimensions, j is the amount of data per dimension, x ji is the i th sample data in the j th dimension, μ i and σ i are the mean and standard deviation of each dimension X i = (x 1i , x 2i , x 3i, ,..., x ji ), respectively, and z ji represents the normalized data of the i th sample data in the j th dimension. 3.The MLPM model based interleaf gold sheet purity evaluation method according to claim 1, wherein, The original data is time series data. 4.The MLPM model based interleaf gold sheet purity evaluation method of claim 1, wherein, The range of the proportion parameter r is 0 to 1, and the optimal proportion parameter r is 0.

5. 5.The MLPM model based interleaf gold sheet purity evaluation method according to claim 1, wherein, In the S4, training the MLPM model comprises performing stacked feature extraction on the linearized data to output a first prediction vector; using an argmax function to convert the first prediction vector into a first one-hot code, obtaining a purity predicted by the MLPM model according to the first one-hot code, and thus evaluating the prediction accuracy of the model. 6.The MLPM model based interleaf gold sheet purity evaluation method according to claim 5, wherein, The number of stacked layers of the feature extraction is Block number, setting the Block number includes preset Block number, obtaining the prediction accuracy of the MLPM model; updating the Block number, obtaining the prediction accuracy of the updated MLPM model; comparing the model prediction accuracy, obtaining the Block number when the optimal prediction accuracy is obtained.

7. A sandwich gold sheet purity evaluation device based on an MLPM model, characterized by, The device comprises: The acquisition module is configured to acquire raw data, wherein the raw data comprises a first pulsed eddy current signal and a first probe temperature; The standardization module is configured to standardize the raw data to obtain standardized data; The data fusion module is configured to fuse the raw data and the standardized data, The data fusion module inputs the raw data into an addition layer of the MLPM model through a first channel of the MLPM model, inputs the standardized data into the addition layer of the MLPM model through a second channel of the MLPM model, and fuses the raw data and the standardized data in the addition layer to obtain fused data; The fused data is calculated in the following manner: fused data = r×standardized data + (1-r)×raw data, r is a proportion parameter, and is preset weight information; The MLPM model training module is configured to train the MLPM model to obtain a trained MLPM model; The gold purity prediction module is configured to predict the purity of the to-be-tested gold and output a prediction vector; The gold purity result identification module is configured to obtain a purity result of the to-be-tested gold according to the prediction vector, The MLPM model comprises a double-channel input layer, a mixer module, and a fourth linear layer connected in sequence; the double-channel input layer comprises a first channel, a second channel, an addition layer, and a third linear layer; The first channel comprises a first linear layer, and the second channel comprises a second linear layer; the first linear layer is configured to process the raw data, and the second linear layer is configured to process the standardized data; The output ends of the first linear layer and the second linear layer are connected to two input ends of the addition layer respectively, and the output end of the addition layer is connected to an input end of the third linear layer. 8.A computer device, comprising: The device comprises: A memory for storing a computer program; A processor for executing the computer program to implement the MLPM model-based interlayer gold plate purity evaluation method of any one of claims 1-6. 9.A computer-readable storage medium, comprising: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the MLPM model-based interlayer gold plate purity evaluation method of any one of claims 1-6.

Images