MLPM model-based sandwich gold plate purity evaluation method, apparatus and device, and medium

Abstract

The invention discloses a pulsed eddy current sandwich gold plate purity evaluation method, device and equipment based on an MLPM model and a medium. The method comprises the following steps: collecting original data; standardizing the original data to obtain standardized data; inputting the original data and the standardized data into two channels of an MLPM model for data fusion, and then training the MLPM model by using linearized fusion data to obtain a trained MLPM model; collecting data of gold to be detected; inputting the collected data into the trained MLPM model, and outputting a second prediction vector; and converting the second prediction vector into a second one-hot code, and obtaining the number of the corresponding gold to be tested through the second one-hot code. According to the method, original data and standardized data are trained at the same time by using double channels of the input layer, the influence of probe temperature fluctuation on a pulsed eddy current signal is counteracted on the algorithm level, hardware equipment does not need to be changed, and the robustness of a model to temperature change is enhanced.

Description

Technical Field

[0001] This invention relates to the technical field of pulsed eddy current metal detection using deep learning methods, and particularly to a method, apparatus, device, and medium for evaluating the purity of sandwich gold plates based on the MLPM model. Background Technology

[0002] Gold, as a precious metal, is widely used in jewelry, electronics, medicine, and many other fields, playing a vital role in financial transactions and the jewelry industry. However, in recent years, numerous reports of gold adulteration have emerged, severely impacting the stability and reputation of the gold market and causing significant losses to consumers and businesses. Therefore, testing the purity of gold, i.e., the percentage of adulteration, has become a crucial task.

[0003] Pulsed eddy current testing (PECT), as a non-destructive testing technique, is now widely used in various industrial and specialized fields, particularly in defect detection, thickness measurement, and material identification of metal components. PECT technology acquires information about different depths of the conductive object being tested by emitting broadband pulse excitation signals to the detection coil, thus improving detection efficiency. Furthermore, by increasing the instantaneous magnetic field strength, it can penetrate multiple layers of conductive materials for detection, making it especially suitable for gold purity testing in laminated gold plates.

[0004] To address the shortcomings of traditional methods in gold purity testing, such as insufficient accuracy, machine learning methods provide crucial technical support for achieving precise purity determination. In existing PECT applications, traditional machine learning algorithms are widely used. By collecting and analyzing small amounts of labeled data, classification models can be built to determine whether gold samples are adulterated with other substances. These models can be constructed using feature extraction and machine learning (ML, such as SVM, RF) or deep learning (such as RNN, ResNet). Classification models can help identify whether gold samples are adulterated with other substances and estimate the purity of the gold samples.

[0005] Typically, these studies employ a two-step approach. First, the original pulsed eddy current (PEC) signal undergoes feature transformation or extraction processing, including methods such as principal component analysis, Rihaczek distribution time-frequency analysis, and Hilbert-Huang transform. Then, to achieve the final tasks 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. These traditional machine learning algorithms remain valuable in PECT research.

[0006] Compared to traditional machine learning methods, deep learning, by constructing deep neural network models, can automatically extract and classify features, thereby achieving more accurate predictions and decisions. In the field of eddy current detection, due to the large volume of signal data and the complexity of its features, deep learning technology can effectively solve these technical challenges.

[0007] When using pulsed eddy current technology for detection, various interferences exist. For example, changes in the probe temperature of the pulsed eddy current device can lead to differences in the detection results. Therefore, pulsed eddy current detection at different temperatures will result in different results, thus reducing the accuracy of the results. Therefore, improving the deep learning model to address the impact of probe temperature fluctuations on pulsed eddy current detection is of great significance for the determination of gold purity. Summary of the Invention

[0008] The purpose of this invention is to provide a method, apparatus, device, and medium for evaluating the purity of sandwich gold plates based on the MLPM model, so as to solve the influence caused by probe temperature fluctuations during pulsed eddy current detection and improve the classification accuracy of the MLPM model.

[0009] To achieve the above objectives, the technical solution adopted by this invention is: a method for evaluating the purity of sandwich gold plates based on the MLPM model, the method comprising the following steps:

[0010] S1: Collect raw data; wherein, the raw data includes a first pulse eddy current signal and a first probe temperature, the first pulse eddy current signal is the signal returned to the pulse eddy current detection device after the pulse eddy current detection device sends a pulse waveform to the test gold and the test gold induces eddy currents, and the first probe temperature is the probe temperature of the pulse eddy current detection device.

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

[0012] 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.

[0013] S4: Linearize the fused data, and use the linearized fused data to train the MLPM model to obtain the trained MLPM model;

[0014] S5: Collect the second pulse eddy current signal corresponding to the gold to be tested and the temperature of the second probe of the pulse eddy current detection device; input the second pulse eddy current signal and the second probe temperature into the trained MLPM model, and output the second prediction vector;

[0015] S6: Convert the second prediction vector into a second one-hot code, and obtain the corresponding number of the gold to be tested through the second one-hot code. The number is used to reflect the purity of the gold.

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

[0017] ,

[0018] in, , 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 data in the i-th dimension.

[0019] Furthermore, the original data is time-series data.

[0020] Furthermore, in S3, the fused data is calculated in the following manner:

[0021] Fusion data = r × standardized data + (1-r) × original data

[0022] Where r is the scaling parameter, which is the preset weight information.

[0023] Furthermore, the range of the scaling parameter r is 0 to 1, and the optimal scaling parameter r is 0.5.

[0024] Furthermore, in step S4, training the MLPM model includes performing stacked feature extraction on the linearized data to output a first prediction vector; using the argmax function, converting the first prediction vector into a first one-hot code; obtaining the purity of the MLPM model's prediction based on the first one-hot code; and thus evaluating the model's prediction accuracy.

[0025] Furthermore, the number of stacked layers for feature extraction is the number of blocks. The number of blocks is set to include a preset number of blocks to obtain the prediction accuracy of the MLPM model; the number of blocks is updated to obtain the prediction accuracy of the updated MLPM model; the prediction accuracy of the models is compared to obtain the number of blocks at the optimal prediction accuracy.

[0026] Based on the same concept, the present invention also provides a purity evaluation device for sandwich gold plates based on the MLPM model, the device comprising:

[0027] The acquisition module is used to acquire raw data, wherein the raw data includes a first pulse eddy current signal and a first probe temperature;

[0028] The standardization module is used to standardize the original data to obtain standardized data;

[0029] The data fusion module is used to merge raw data and standardized data;

[0030] The MLPM model training module is used to train the MLPM model and obtain the trained MLPM model.

[0031] The gold purity prediction module is used to predict the purity of the gold to be tested and output a prediction vector.

[0032] The gold purity determination module is used to obtain the purity result of the gold to be tested based on the prediction vector.

[0033] Based on the same concept, the present invention also provides a computer device.

[0034] The device includes:

[0035] Memory, used to store computer programs;

[0036] A processor is used to execute the computer program to implement the MLPM-based method for evaluating the purity of sandwich gold plates as described above.

[0037] Based on the same concept, the present invention also provides a computer-readable storage medium.

[0038] The computer-readable storage medium stores a computer program that, when executed by a processor, implements the MLPM-based method for evaluating the purity of sandwich gold plates as described above.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] This invention introduces a dual-input channel structure to modify the MLP-Mixer neural network model, resulting in the MLPM model. The first channel of the MLPM model inputs the original data, and the second channel inputs the standardized data, enabling the simultaneous processing of both standardized and original data. This algorithmically offsets the influence of probe temperature fluctuations on the pulsed eddy current signal without requiring changes to the hardware, thus enhancing the model's robustness to temperature changes. Attached Figure Description

[0041] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of the purity evaluation method for sandwich gold plates based on the MLPM model in an embodiment of the present invention;

[0043] Figure 2 This is a PEC signal diagram obtained by the pulsed eddy current device in this embodiment of the invention, when the probe is in both low-temperature and high-temperature states, on the same gold sample.

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

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

[0046] Figure 5 This is a confusion matrix diagram of the prediction results of MLPM-60 in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

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

[0049] Example 1

[0050] Figure 1 The flowchart illustrates a method for evaluating the purity of sandwich gold plates based on a multi-input channel multilayer perceptron (MLPM) model, as provided in an embodiment of the present invention. Figure 1 As shown, the purity evaluation method for the sandwich gold plate includes the following steps:

[0051] S1: Collect raw data.

[0052] The probe temperature of a pulsed eddy current (PEC) testing device affects the PEC amplitude. This means that it is best to keep the probe temperature within a stable range during the acquisition process to avoid excessive deviations in the PEC signal measured from the same sample each time. This would cause the model trained at a previous probe temperature to experience a decrease in prediction accuracy or even failure at different probe temperatures. Solving this problem at the equipment level is complex and costly.

[0053] Given the difficulty of suppressing the impact of probe temperature at the hardware level, this invention considers improvements at the algorithm level. To this end, deep learning technology is employed for model training and prediction. By constructing a deep neural network model and training it with a large amount of sample data, the model can accurately predict eddy current detection signals at different temperatures. Simultaneously, appropriate data processing and feature extraction methods can further improve the model's performance and accuracy. With the help of deep learning technology, high-precision and high-efficiency non-destructive testing of pulsed eddy currents is ultimately achieved, providing more reliable assurance for industrial production.

[0054] The MLPM model in this embodiment of the invention indirectly eliminates the influence of probe temperature on the data acquired by the device during pulse eddy current detection through a dual-input channel mode, and directly eliminates the influence caused by probe temperature fluctuations at the algorithm level without changing the hardware device.

[0055] like Figure 2 As shown, the amplitude of the same gold sample during acquisition is correlated with the probe temperature. Therefore, the probe temperature is also used as data for model training.

[0056] A large number of PEC signals and probe temperatures from various gold samples were collected using a pulsed eddy current (PEC) testing device. These were used as input data for the MLPM model. The PEC signal is the signal returned to the PEC testing device after the gold sample generates eddy currents following the transmission of a pulse waveform. The probe temperature is the temperature of the PEC probe, which is monitored by the temperature control module of the PEC testing device via software. The PEC signal is a 1000-dimensional time series, the probe temperature is a 1-dimensional time series, and the input features are 1001-dimensional feature vectors.

[0057] The purity and other properties of each gold sample are shown in Table 1. PEC data of 13 gold samples with different purities and thicknesses were collected using a pulsed eddy current testing (PDC) device. Both the training and test sets included data for the 13 gold samples. In the training set, there were 800 data points for each type of gold sample. In the test set, there were 400 data points for each type of gold sample.

[0058] Table 1 Gold Purity Properties

[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, processing two-dimensional data. The purity assessment of sandwich gold plates aims to directly classify the PEC signals of gold with different purities obtained by the detection equipment. Based on the concept of classifying time-series data, this invention modifies the architecture of the existing MLP-Mixer model. Unlike the MLP-Mixer model used for processing two-dimensional image data, the MLPM model is specifically designed for processing time-series data, matching the model architecture to the inherent characteristics of the data. The modified model, called the MLPM model, removes the encoding layer used to decompose two-dimensional data and replaces it with a linear layer with dual-channel input, where the two channels respectively input standardized data and the original data. Some hidden layers (originally used for processing two-dimensional data) that are unsuitable for sequence processing are removed, making it suitable for processing time-series signals similar to PEC signals.

[0075] Compared to the single-input channel of MLP-Mixer, MLPM's dual-input channel can make full use of the effective information in both the raw and standardized data, allowing the model to learn both the statistical characteristics and the raw information of the data simultaneously, thus better capturing the key features of the data.

[0076] like Figure 3 As shown, the input data is first processed through a dual-channel input layer. The input data is copied twice and fed into two separate channels. One copy is standardized, while the other remains unprocessed. Two linear layers are used to linearize the standardized data and the original data, respectively. Then, an additive layer fuses the data from both channels, followed by further processing through a linear layer. This allows MLPM to simultaneously process input data from two different dimensions and fuse their information.

[0077] The additive layer includes a scaling parameter *r* to allocate weights between the two channels, allowing for flexible control over the impact of each input on the output. This enables fine-tuning of the model based on the importance of different features, improving its expressive power and adaptability. Weighted summation automatically selects input features. If one channel's data contributes less to the task, its weight can be adjusted to a smaller value, reducing its impact on the model. This helps reduce the influence of redundant features and improves the model's generalization ability. The main purpose of MLPM's dual-input channel approach is to fuse the relevant information between the two input channels, enabling the neural network to better understand and process the features of the input data.

[0078] like Figure 3As shown, the scaling parameter r is used to allocate the weight of the summation of the two channels: the output of the additive layer = r × standardized data + (1-r) × original data, which is a weighted summation. This method provides 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, and use the linearized fused data to train the MLPM model to obtain the trained MLPM model.

[0080] The data fused by the additive layers is linearized through another linear layer, and then stacked feature extraction is performed on the linearized data. This stacked feature extraction process enables automatic feature extraction and classification of time-series data, improving the model's ability to solve complex problems and enhancing its classification accuracy.

[0081] Stacked feature extraction involves first performing layer standardization on the data, then linearizing the standardized data through a linear layer using the GeLU function as the activation function to introduce non-linearity, and finally linearizing it through another linear layer to obtain the processed data. After stacked feature extraction, the processed data undergoes layer standardization, and then the standardized data is input into a linear layer to output a prediction vector, which is used to classify each gold sample.

[0082] In this context, the number of stacked layers for stacked feature extraction is controlled by the number of blocks. In other words, the model's depth and expressive power are primarily determined by the number of blocks; therefore, the number of blocks determines the model's complexity and learning ability. A larger number of blocks may increase the model's expressive power, but it may also increase the risk of overfitting and computational complexity. The appropriate number of blocks needs to be selected based on the specific task and dataset characteristics. Generally, more complex tasks and datasets may require more blocks to extract richer feature representations, while simpler tasks and datasets may use fewer blocks. Therefore, selecting an appropriate number of blocks is crucial to the model's final performance and training efficiency.

[0083] Setting the number of blocks involves setting a preset number of blocks to obtain the prediction accuracy of the MLPM model; updating the number of blocks to obtain the updated prediction accuracy of the MLPM model; and comparing the prediction accuracies of the models to obtain the optimal number of blocks for prediction accuracy. For ease of description later, the network will be named MLPM-N, such as... Figure 3 As shown in the figure, N represents the number of blocks, for example, MLPM-16 represents an MLPM with 16 blocks.

[0084] In embodiments of the present invention, before training the MLPM model, the method further includes setting various hyperparameters of the MLPM model and configuring the training environment of the MLPM model. 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, this includes setting the number of blocks to 60 and initializing the learning rate to 1 × 10⁻⁶. −3 The training iterations consisted of 200 epochs. All weights and biases of the model were updated using the Adam optimizer, with an L2 regularization parameter of 1 × 10⁻⁶. −3 The cross-entropy loss function is used. Model parameters are updated using the gradient function. The batch size is 120. When the loss value stabilizes during training, the model is considered converged. To prevent overfitting, early stopping is performed at this point to obtain the trained MLPM model.

[0085] The training process was coded using the PyTorch machine learning library and performed on an Nvidia GeForce RTX 4070Ti GPU. The MLPM-60 training iteration with 200 iterations took 22 minutes and 16.1 seconds. As the number of blocks increases, the model's computation becomes more complex, and the training time also increases.

[0086] After training a convergent MLPM model using the training set, the test set was input into the model for testing. Based on the top-1 prediction accuracy of MLPM with different block counts shown in Table 2, it can be observed that as the number of blocks increases, the accuracy of the MLPM model initially improves, indicating a stronger classification ability. However, when the number of blocks reaches 60, further increases in the block count cause the model's accuracy to decrease, signifying a weakening of classification ability. Therefore, MLPM-60 was selected as the optimal model, achieving an accuracy of 99.9%, nearly 100%. The confusion matrix of the prediction results is shown below. Figure 5 As shown, except for three samples of number 4 that were incorrectly identified as sample 1, all other samples were accurately predicted.

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

[0088]

[0089] After determining the optimal number of blocks for the model, to determine the proportion parameter r for achieving the best classification effect between the two channels of MLPM, further testing was conducted on r in the interval [0,1]. The test accuracy of the model was used to judge the importance of each channel. For example, if the model has a higher test accuracy at r=0.7 than at r=0.8, it indicates that the actual contribution of standardized data may be more important than under the current weights. The proportion parameter r was set to 0, 0.1, 0.3, 0.5, 0.7, 0.9, and 1. Using the optimal MLPM-60 as the test model, the test accuracy results of the proportion parameter r are shown in Table 3. r=0 and r=1 represent the results of inputting only raw data and only standardized data when MLPM has only a single input channel. The results show that the model accuracy decreases significantly when only standardized data or raw data is used as input data; this indicates 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 the PEC signal. The results show that the MLPM-60 model has the highest prediction accuracy when r=0.5.

[0090] Table 3 Classification Results

[0091]

[0092] In general models, the temperature difference between the probe and the training / test sets affects the results because the change in probe temperature causes a directional shift in the amplitude measured by the device. For example, it cannot distinguish between 99.90% and 99.99% gold with extremely similar purity. MLPM has overcome this bottleneck, which is the decisive factor that allows MLPM to achieve higher accuracy than other neural networks. MLPM's dual input channels simultaneously extract feature information from both standardized and raw data, mitigating the impact of this shift and ensuring consistent feature space regression between the two datasets.

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

[0094] A pulsed eddy current testing device is used to collect the pulsed eddy current signal corresponding to the gold under test and the probe temperature of the device. In the specific testing process, the pulsed eddy current signal is collected by placing the gold under test above the sensor of the pulsed eddy current testing device. The device then generates a rectangular wave, which is amplified by a power amplifier module and applied to the excitation coil in the sensor. The excitation coil thus generates a primary magnetic field whose intensity varies with time, forming a rectangle. When the primary magnetic field undergoes a sudden change, eddy currents are induced in the gold plate. The induction and dissipation of eddy currents are affected by the purity of the gold plate. During the eddy current generation and dissipation process, the eddy currents generate a secondary magnetic field that varies with time. The receiving coil in the sensor converts the time-varying secondary magnetic field into a voltage signal, which is amplified by a signal amplification module. Data is acquired by a dual-function module for signal generation and acquisition. The probe temperature of the pulsed eddy current testing device is obtained by monitoring the probe temperature through software.

[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 the prediction vector of the gold to be tested.

[0097] S6: Convert the prediction vector of the gold to be tested into a one-hot code using the argmax function. Obtain the corresponding number of the gold to be tested using the one-hot code. Look up the corresponding purity value in Table 1 based on the number to assess the purity of the gold.

[0098] Example 2

[0099] Based on the same concept, embodiments of the present invention also provide a purity evaluation device for sandwich gold plates based on the MLPM model, the device comprising:

[0100] The acquisition module is used to acquire raw data, wherein the raw data includes a first pulse eddy current signal and a first probe temperature;

[0101] The standardization module is used to standardize the original data to obtain standardized data;

[0102] The data fusion module is used to merge raw data and standardized data;

[0103] The MLPM model training module is used to train the MLPM model and obtain the trained MLPM model.

[0104] The gold purity prediction module is used to predict the purity of the gold to be tested and output a prediction vector.

[0105] The gold purity determination module is used to obtain the purity result of the gold to be tested based on the prediction vector.

[0106] Example 3

[0107] Based on the same concept, embodiments of the present invention also provide a computer device, the device comprising:

[0108] Memory, used to store computer programs;

[0109] A processor is used to implement the MLPM-based method for evaluating the purity of sandwich gold plates described in Embodiment 1 when executing the computer program.

[0110] Example 4

[0111] Based on the same concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the MLPM-based method for evaluating the purity of sandwich gold plates described in Embodiment 1 above.

[0112] This invention utilizes the dual-channel feature of its input layer, inputting standardized data and original data into each channel for training, thereby enabling the model to acquire a wider range of data feature information. This improves the classification accuracy of sandwich gold plates and eliminates the offset error caused by probe temperature in pulse detection equipment.

[0113] The above embodiments should be understood as being used only to illustrate the present invention more clearly, and not to limit the scope of the present invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art fall within the scope defined by the appended claims.

Claims

1. A method for evaluating the purity of sandwich gold plates based on the MLPM model, characterized in that, The method includes the following steps: S1: Collect raw data; wherein, the raw data includes a first pulse eddy current signal and a first probe temperature, the first pulse eddy current signal is the signal returned to the pulse eddy current detection device after the pulse eddy current detection device sends a pulse waveform to the test gold and the test gold induces eddy currents, and the first probe temperature is the probe temperature of the pulse eddy current detection device. S2: Standardize the original data to obtain standardized data; 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. S4: Linearize the fused data, and use the linearized fused data to train the MLPM model to obtain the trained MLPM model; S5: Collect the second pulse eddy current signal corresponding to the gold to be tested and the temperature of the second probe of the pulse eddy current detection device; input the second pulse eddy current signal and the second probe temperature into the trained MLPM model, and output the second prediction vector; S6: Convert the second prediction vector into a second one-hot code, and obtain the corresponding number of the gold to be tested through the second one-hot code. The number is used to reflect the purity of the gold.

2. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 1, characterized in that, In S2, the standardization is z-score normalization, and the formula for standardizing the original data is: ， 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 data in the i-th dimension.

3. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 1, characterized in that, The original data is time series data.

4. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 1, characterized in that, In step S3, the fused data is calculated in the following manner: Fusion data = r × standardized data + (1-r) × original data Where r is the scaling parameter, which is the preset weight information.

5. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 4, characterized in that, The range of the scaling parameter r is 0 to 1, and the optimal scaling parameter r is 0.

5.

6. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 5, characterized in that, In step S4, training the MLPM model includes performing stacked feature extraction on the linearized data to output a first prediction vector; using the argmax function, converting the first prediction vector into a first one-hot code; obtaining the purity of the MLPM model's prediction based on the first one-hot code; and thus evaluating the model's prediction accuracy.

7. The method for evaluating the purity of sandwich gold plates based on the MLPM model according to claim 6, characterized in that, The number of stacked layers for feature extraction is the number of blocks. The number of blocks is set to include a preset number of blocks to obtain the prediction accuracy of the MLPM model. The number of blocks is updated to obtain the prediction accuracy of the updated MLPM model. The prediction accuracy of the models is compared to obtain the number of blocks at the optimal prediction accuracy.

8. A purity evaluation device for sandwich gold plates based on the MLPM model, characterized in that, The device includes: The acquisition module is used to acquire raw data, wherein the raw data includes a first pulse eddy current signal and a first probe temperature; The standardization module is used to standardize the original data to obtain standardized data; The data fusion module is used to merge raw data and standardized data; The MLPM model training module is used to train the MLPM model and obtain the trained MLPM model. The gold purity prediction module is used to predict the purity of the gold to be tested and output a prediction vector. The gold purity determination module is used to obtain the purity result of the gold to be tested based on the prediction vector.

9. A computer device, characterized in that, The device includes: Memory, used to store computer programs; A processor, configured to implement the method for evaluating the purity of sandwich gold plates based on the MLPM model as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for evaluating the purity of sandwich gold plates based on the MLPM model as described in any one of claims 1 to 7.

Images