A metal intelligent identification method and system based on metal sensor

Abstract

The present invention provides a metal intelligent identification method and system based on a metal sensor, and relates to the field of metal detection technology. The method includes: acquiring historical detection data, extracting frequency response vectors corresponding to multiple electromagnetic response parameters, constructing a first frequency response reference list and a second frequency response reference list for each electromagnetic response parameter; screening out target sample data from the frequency response vectors corresponding to multiple electromagnetic response parameters, constructing a sample data set and determining a first target response matrix and a second target response matrix, and training a target metal combination identification model through the sample data set; inputting the target detection data into the target metal combination identification model to generate a first identification result, generating a second identification result according to the first target response matrix and the second target response matrix, and fusing the first identification result and the second identification result to obtain a target metal identification result. The present invention achieves the improvement of the identification accuracy of similar metals.

Description

Technical Field

[0001] The present invention relates to the technical field of metal detection, and in particular to a metal intelligent identification method and system based on a metal sensor. Background Art

[0002] In the field of metal detection and identification, some detection and identification technologies based on metal sensors identify metals through a single or a small number of electromagnetic response parameters. In practical applications, some metals have similar electromagnetic properties, such as copper and brass, iron and nickel, which are prone to interfere with each other during the detection process, resulting in insufficient recognition accuracy. The main reason is that some of the response characteristics of these metals under the electromagnetic field are relatively similar, making it impossible to accurately capture their differences, thus affecting the accuracy of detection.

[0003] Therefore, in order to address the interference problem existing in this type of metal detection, there is an urgent need to provide a metal intelligent identification method and system based on metal sensors with better detection effect, so as to improve the detection and identification accuracy between similar metals and enhance the system's anti-interference ability. Summary of the invention

[0004] To solve the above technical problems, the present invention proposes a metal intelligent identification method and system based on metal sensors, which can make full use of multiple electromagnetic response parameters and perform comprehensive analysis based on frequency response characteristics, thereby improving the identification accuracy between similar metals.

[0005] To achieve the above object, the present invention provides a metal intelligent identification method based on a metal sensor in a first aspect, comprising:

[0006] Acquire historical detection data for target metal combination identification based on a target metal sensor, where the target metal combination is two metals whose identification interferes with each other, and extract multiple groups of first detection data and second detection data from the historical detection data;

[0007] Extracting frequency response vectors corresponding to the plurality of electromagnetic response parameters of each group of first detection data and second detection data, respectively, and extracting a first frequency response reference list of each electromagnetic response parameter in the plurality of groups of first detection data and a second frequency response reference list of each electromagnetic response parameter in the plurality of groups of second detection data according to the plurality of frequency response vectors corresponding to each electromagnetic response parameter;

[0008] Extracting a first response reference vector for each electromagnetic response parameter based on a first frequency response reference list, extracting a second response reference vector for each electromagnetic response parameter based on a second frequency response reference list, screening target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, constructing a sample data set based on the target sample data and determining a first target response matrix and a second target response matrix, and training a target metal combination recognition model through the sample data set;

[0009] In the process of metal identification based on the target metal sensor, if the pre-identification result corresponding to the collected target detection data belongs to the target metal combination, the target detection data is input into the target metal combination identification model to generate a first identification result corresponding to the target detection data, the target detection data is analyzed according to the first target response matrix and the second target response matrix to generate a second identification result corresponding to the target detection data, and the first identification result and the second identification result are fused to obtain the target metal identification result corresponding to the target detection data.

[0010] Preferably, the extraction of the first frequency response reference list and the second frequency response reference list further includes:

[0011] For each electromagnetic response parameter in the multiple groups of first detection data, the frequency response vectors respectively corresponding to the electromagnetic response parameter are used to determine the response score of each electromagnetic response parameter in each group of first detection data at multiple detection frequencies according to the frequency response vectors, and the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to construct a first frequency response reference list corresponding to each electromagnetic response parameter;

[0012] For the frequency response vectors corresponding to each electromagnetic response parameter in multiple groups of second detection data, the response scores of each electromagnetic response parameter in each group of second detection data at multiple detection frequencies are determined according to the frequency response vectors, and the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to construct a second frequency response reference list corresponding to each electromagnetic response parameter, and the frequency response reference list corresponding to the electromagnetic response parameter includes the total response score of the electromagnetic response parameter at each detection frequency.

[0013] Preferably, selecting target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters respectively based on the first response reference vector and the second response reference vector includes:

[0014] Calculating a first similarity between a frequency response vector corresponding to each electromagnetic response parameter of the first detection data and a first response reference vector, and calculating a second similarity between a frequency response vector corresponding to each electromagnetic response parameter of the second detection data and a second response reference vector;

[0015] A similarity weight of each electromagnetic response parameter is calculated according to the first response reference vector and the second response reference vector of each electromagnetic response parameter, a feature score of each group of first detection data and second detection data is calculated based on multiple similarity weights, and multiple groups of first detection data and second detection data with feature scores higher than a preset feature score threshold are screened out as target sample data.

[0016] Preferably, the target sample data further includes:

[0017] Determine a sample label for each group of first detection data and second detection data in the target sample data, construct a sample data set including multiple groups of detection data and a sample label for each group of detection data, and train a target metal combination recognition model based on the sample data set;

[0018] Extracting a plurality of groups of first detection data from the target sample data with respect to a response vector set of each electromagnetic response parameter, constructing a target optimization function for the response vector set of each electromagnetic response parameter, minimizing the score error between each frequency response vector included in the response vector set of the electromagnetic response parameter and the first response reference vector of the electromagnetic response parameter as the optimization target of the target optimization function, iteratively optimizing the first response reference vector of each electromagnetic response parameter based on a stochastic gradient descent algorithm, generating a first response parameter vector for each electromagnetic response parameter, and constructing a first target response matrix based on the plurality of first response parameter vectors;

[0019] A response vector set of multiple groups of second detection data for each electromagnetic response parameter in the target sample data is extracted, and a target optimization function for the response vector set of each electromagnetic response parameter is constructed. The optimization target of the target optimization function is to minimize the score error between each frequency response vector contained in the response vector set of the electromagnetic response parameter and the second response reference vector of the electromagnetic response parameter. The second response reference vector of each electromagnetic response parameter is iteratively optimized based on the stochastic gradient descent algorithm to generate a second response parameter vector for each electromagnetic response parameter, and a second target response matrix is ​​constructed based on multiple second response parameter vectors.

[0020] Preferably, the first recognition result and the second recognition result further include:

[0021] The first recognition result includes the first classification score of each sample label, and the second recognition result includes the second classification score of each sample label. The first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label.

[0022] Preferably, the target metal combination identification model is a SVM model.

[0023] A second aspect of the present invention provides a metal intelligent identification system based on a metal sensor, which is used to implement the above-mentioned metal intelligent identification method based on a metal sensor, comprising:

[0024] A data acquisition module is used to obtain historical detection data for target metal combination identification based on a target metal sensor, where the target metal combination is two metals that interfere with each other in identification, and extract multiple groups of first detection data and second detection data from the historical detection data;

[0025] A frequency response analysis module, used to extract frequency response vectors corresponding to each group of first detection data and second detection data with respect to a plurality of electromagnetic response parameters, and extract a first frequency response reference list of each electromagnetic response parameter in the plurality of groups of first detection data and a second frequency response reference list of each electromagnetic response parameter in the plurality of groups of second detection data according to the plurality of frequency response vectors corresponding to each electromagnetic response parameter;

[0026] A sample data extraction and analysis module is used to extract a first response reference vector for each electromagnetic response parameter based on a first frequency response reference list, extract a second response reference vector for each electromagnetic response parameter based on a second frequency response reference list, filter out target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, construct a sample data set based on the target sample data and determine a first target response matrix and a second target response matrix, and train a target metal combination recognition model through the sample data set;

[0027] A metal detection and identification module is used to, during the process of metal identification based on a target metal sensor, if the pre-identification result corresponding to the collected target detection data belongs to a target metal combination, input the target detection data into a target metal combination identification model to generate a first identification result corresponding to the target detection data, analyze the target detection data according to a first target response matrix and a second target response matrix to generate a second identification result corresponding to the target detection data, and fuse the first identification result and the second identification result to obtain a target metal identification result corresponding to the target detection data.

[0028] Preferably, the metal detection and recognition module fuses the first recognition result and the second recognition result to obtain a target metal recognition result corresponding to the target detection data, including:

[0029] The first recognition result includes the first classification score of each sample label, and the second recognition result includes the second classification score of each sample label. The first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label.

[0030] The present invention has the following beneficial effects:

[0031] The present invention effectively solves the technical problem that similar metals are difficult to distinguish during the detection process based on frequency response feature analysis and machine learning models. By extracting and optimizing the response reference vectors of multiple electromagnetic response parameters, a representative target response matrix is ​​constructed, and then combined with the target metal combination recognition model for intelligent recognition of similar metals. The present invention can significantly improve the recognition accuracy of similar metals. Compared with the traditional single-parameter analysis method, the present invention realizes comprehensive analysis of multi-dimensional parameters, improves the anti-interference ability to noise and abnormal data, and enhances the robustness and reliability of similar metal recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of a metal intelligent identification method based on a metal sensor provided in one of the embodiments of the present invention.

[0033] Figure 2 A structural block diagram of a metal intelligent identification system based on a metal sensor provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make those skilled in the art better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0035] When metal detection and identification is performed based on metal sensors, electromagnetic waves, eddy currents or magnetic fields are emitted to metal objects through metal sensors. When these waves or fields interact with metal objects, responses are generated on the metal surface. These response signals depend on the physical properties of the metal, such as electrical conductivity, magnetic permeability, material density, surface structure, etc. Different metals will cause different sensor responses. Taking electromagnetic sensors as an example, electromagnetic sensors are one of the most common metal sensors. They use the principle of electromagnetic induction to detect metals. When metal detection is performed through electromagnetic sensors, a series of relevant data at different detection frequencies are usually collected. This technology is called multi-frequency electromagnetic detection or spectrum analysis. Its principle is to collect corresponding electromagnetic response data from low frequency to high frequency in sequence by changing the frequency of the excitation signal, and to obtain impedance change, electrical conductivity and magnetic permeability data by analyzing these electromagnetic response data, and then perform metal detection and identification based on these data.

[0036] However, in actual applications, the electromagnetic properties of many metal materials are very similar, such as aluminum and magnesium, copper and brass, etc. These materials have little difference in conductivity and magnetic permeability, which makes it difficult to distinguish these metal materials simply based on the detected parameters. In view of this situation, an embodiment of the present invention provides a metal intelligent identification method based on a metal sensor, which further analyzes and mines the data collected by the metal sensor, makes full use of the data related to multiple electromagnetic response parameters collected by the metal sensor, and performs a comprehensive analysis based on the frequency response characteristics to provide a solution that can more accurately identify similar metals.

[0037] like Figure 1 As shown, one embodiment of the present invention provides a metal intelligent identification method based on a metal sensor, which specifically includes the following steps:

[0038] Step S1, obtaining historical detection data for target metal combination identification based on target metal sensors, extracting multiple groups of first detection data and second detection data from the historical detection data, and extracting frequency response vectors corresponding to each group of first detection data and second detection data with respect to multiple electromagnetic response parameters.

[0039] In this embodiment, the historical detection data for target metal combination identification based on the target metal sensor can be, for example, detailed electromagnetic response record data and final detection result data when the electromagnetic sensor detects the target metal combination in different environments and detection frequencies. The target metal combination can be a metal combination that is easy to interfere with each other in the actual detection process, such as the metal combination of copper and brass. In the process of detecting and identifying copper or brass, it is easy to misidentify, such as misdetecting copper as brass or misdetecting brass as copper. After obtaining multiple groups of historical detection data related to copper and brass, they are marked as, for example, first detection data representing copper and second detection data representing brass. Each group of detection data contains multiple electromagnetic response parameters under a series of detection frequencies, such as impedance, amplitude, phase angle, conductivity and other parameters collected at different detection frequencies. These electromagnetic response parameters will show different response characteristics as the detection frequency changes. The frequency response vectors corresponding to these electromagnetic response parameters are extracted from each group of first detection data and second detection data. For example, for a certain detection data of copper, the frequency response vector of the extracted impedance parameter is used to represent the impedance change characteristics at different frequencies.

[0040] Step S2: extracting a first frequency response reference list of each electromagnetic response parameter in the multiple sets of first detection data and a second frequency response reference list of each electromagnetic response parameter in the multiple sets of second detection data according to the multiple frequency response vectors corresponding to each electromagnetic response parameter.

[0041] In this embodiment, for each electromagnetic response parameter, by analyzing multiple frequency response vectors corresponding to the electromagnetic response parameter, characteristic information in multiple vectors is calculated and extracted, thereby forming a frequency response reference list, which is used to reflect the performance of the electromagnetic response parameter at different detection frequencies. For example, for the electromagnetic response parameter of impedance in multiple groups of first detection data, the corresponding first frequency response reference list can be used to characterize the response trend of impedance at different detection frequencies during the detection of copper. Multiple first frequency response reference lists and second frequency response reference lists respectively represent the electromagnetic response characteristics of typical electromagnetic response parameters of copper and brass at different detection frequencies.

[0042] Step S3, extracting the first response reference vector of each electromagnetic response parameter based on the first frequency response reference list, extracting the second response reference vector of each electromagnetic response parameter based on the second frequency response reference list, screening out target sample data from the frequency response vectors corresponding to the multiple electromagnetic response parameters based on the first response reference vector and the second response reference vector, constructing a sample data set based on the target sample data and determining the first target response matrix and the second target response matrix, and training the target metal combination recognition model through the sample data set.

[0043] In this embodiment, the response reference vector of the electromagnetic response parameter represents the group response characteristics of the electromagnetic response parameter at different detection frequencies in the historical data. For example, the first response reference vector of the amplitude represents the performance of the amplitude parameter at different detection frequencies in the historical detection data of copper. By analyzing the association and difference between the group characteristics and the individual characteristics, some representative detection data are finally selected from multiple groups of first detection data and second detection data as target sample data. The target sample data includes multiple groups of detection data related to copper detection and brass detection. After adding corresponding sample labels to these data, a sample data set is constructed to train a target metal combination recognition model that can detect and identify copper and brass. For the target sample data, the association between the multiple groups of first detection data contained therein is also analyzed, and the common characteristics represented by the first response reference vector are combined to construct a first target response matrix representing the copper detection characteristics in the historical detection data, which includes the response characteristics of different electromagnetic response parameters at different detection frequencies. Similarly, a second target response matrix representing the brass detection characteristics in the historical detection data is constructed.

[0044] Step S4: In the process of metal identification based on the target metal sensor, if the pre-identification result corresponding to the collected target detection data belongs to the target metal combination, the target detection data is input into the target metal combination identification model to generate a first identification result corresponding to the target detection data, the target detection data is analyzed according to the first target response matrix and the second target response matrix to generate a second identification result corresponding to the target detection data, and the first identification result and the second identification result are fused to obtain the target metal identification result corresponding to the target detection data.

[0045] In this embodiment, in the process of real-time metal detection based on the target metal sensor, the target detection data collected by the metal sensor includes the response characteristics of multiple electromagnetic response parameters at different detection frequencies. The target detection data can be well analyzed by conventional technology to obtain a pre-identification result to determine whether the metal to be detected belongs to one of the metals in the target metal combination. For example, it can be determined that the metal to be detected is most likely one of copper and brass rather than aluminum, iron, and other metals with greatly different properties from the metal combination of copper and brass.

[0046] In this case, in order to obtain a more accurate recognition result, the collected target detection data is input into the pre-trained target metal combination recognition model to generate a first recognition result corresponding to the target detection data, which is used to determine whether the metal to be detected is copper or brass. And further based on the first target response matrix and the second target response matrix, the electromagnetic response characteristics of the target detection data are analyzed. By calculating the correlation between the detection data and the typical response matrix of copper and brass, a second recognition result is generated, and finally the first recognition result and the second recognition result are fused. For example, a weighted fusion method can be used to combine the output of the model with the matrix analysis result to finally obtain the target metal recognition result, so as to determine whether the metal to be detected is copper or brass, obtain an accurate metal recognition result, and improve the accuracy of identifying similar metals.

[0047] As an optional implementation process, the extraction of the first frequency response reference list and the second frequency response reference list in step S2 specifically includes:

[0048] For the frequency response vectors corresponding to each electromagnetic response parameter in multiple groups of first detection data, the response score of each electromagnetic response parameter in each group of first detection data at multiple detection frequencies is determined according to the frequency response vectors. Since each frequency response vector represents the response of one of the electromagnetic response parameters at different detection frequencies, after normalizing the frequency response vectors, the normalized results of the electromagnetic response parameters at different detection frequencies are directly used as the response scores, that is, the normalized results of the parameters at each detection frequency are used as the response scores at the inspection frequency, representing the typical response intensity of the electromagnetic response parameter at a specific frequency. Finally, the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to obtain the total response score of each detection frequency, so that the overall response intensity of each electromagnetic response parameter at each detection frequency can be obtained, thereby constructing a first frequency response reference list to obtain the total response score corresponding to each electromagnetic response parameter.

[0049] Similarly, for the frequency response vectors corresponding to each electromagnetic response parameter in multiple groups of second detection data, after normalizing the frequency response vectors, the response scores of each electromagnetic response parameter in each group of second detection data at multiple detection frequencies are determined according to the frequency response vectors, and the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to construct a second frequency response reference list corresponding to each electromagnetic response parameter.

[0050] As an optional implementation process, in step S3, target sample data is screened out from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, which specifically includes:

[0051] A first similarity between a frequency response vector corresponding to each electromagnetic response parameter of the first detection data and the first response reference vector is calculated, and a second similarity between a frequency response vector corresponding to each electromagnetic response parameter of the second detection data and the second response reference vector is calculated, wherein the similarity may be calculated using a Euclidean distance, cosine similarity, or other calculation methods, and cosine similarity is taken as an example in this embodiment.

[0052] After calculating and obtaining multiple first similarities and second similarities, a feature score for each group of first detection data is calculated based on the multiple first similarities for each group of first detection data, and a feature score for each group of second detection data is calculated based on the multiple second similarities for each group of second detection data.

[0053] Specifically, the similarity weight of each electromagnetic response parameter is calculated based on the first response reference vector and the second response reference vector of each electromagnetic response parameter. It is worth noting that in the process of metal detection and identification, the degree of difference between different metals in different electromagnetic response parameters is often different. For example, although the difference between copper and brass in some electromagnetic response parameters is small, when it comes to the difference to a smaller degree, the degree of difference of some parameters in multiple electromagnetic response parameters will be greater. By calculating the degree of difference between the first response reference vector and the second response reference vector of each electromagnetic response parameter, the degree of difference is used as the similarity weight. The greater the degree of difference, the greater the difference in the performance of copper and brass at different detection frequencies on the electromagnetic response parameter. When judging the reference value of a certain group of first detection data compared with the group characteristics, the similarity weight of the first similarity corresponding to the parameter is greater. After determining the similarity weight of each electromagnetic response parameter in this way, based on multiple similarity weights and multiple first similarities, the feature score of each group of first detection data can be calculated based on weighted analysis. This scheme is also adopted for the analysis of multiple groups of second detection data, and finally the feature scores of the first detection data and the second detection data can be calculated.

[0054] Finally, the feature score of each group of first detection data and second detection data is calculated based on multiple similarity weights, and multiple groups of first detection data and second detection data with feature scores higher than a preset feature score threshold are selected as target sample data. In this way, more representative individual data can be selected, especially avoiding the use of data that does not match the group characteristics or has a lot of noise for training and recognition.

[0055] In this embodiment, the weight of each electromagnetic response parameter is determined based on the difference in group characteristics, and the reference value of the individual data is determined by combining the similarity between the individual data and the group characteristics. This method can effectively improve the recognition accuracy in similar metal detection.

[0056] After the target sample data is determined by the above steps, the sample labels of each group of first detection data and second detection data in the target sample data are determined to characterize whether the corresponding data belongs to copper or brass, and a sample data set containing multiple groups of detection data and sample labels of each group of detection data is constructed. The target metal combination recognition model is trained based on the constructed sample data set. The model is specifically a classification model (such as a support vector machine, a decision tree, a neural network, etc.). In this embodiment, the target metal combination recognition model is preferably a support vector machine (SVM model). The SVM model is a commonly used two-classification model that can process multi-dimensional electromagnetic response parameter data and is used to learn the differences between copper and brass in different electromagnetic response parameters, so that metals can be classified in real-time detection. The goal of the target metal combination recognition model training is to find the pattern and relationship between each electromagnetic response parameter and the metal type, so that the metal type can be automatically determined by the detection data in the future. In this embodiment, the type of the target metal combination recognition model is not specifically limited, and those skilled in the art can select a suitable classification model according to actual needs.

[0057] It is worth mentioning that for the construction of the sample data set, since the process of selecting the target sample data selects some of the first and second test data with higher reference value, and excludes data that does not match the group characteristics or has high noise, this method can effectively reduce the time and computing resources required for model training, because the model can converge faster and reduce the complexity of processing confusing samples. Selecting data with clear features can reduce interference and reduce noise in the model training process, thereby improving the classification accuracy of the final model.

[0058] For the target sample data, the following processing is also performed to extract the response vector sets of multiple groups of first detection data in the target sample data with respect to each electromagnetic response parameter. The response vector set of the electromagnetic response parameter includes the frequency response vector of each group of first detection data in the target sample data with respect to the electromagnetic response parameter. And a target optimization function is constructed for each response vector set of the electromagnetic response parameter. The optimization goal of the target optimization function of each response vector set is to minimize the score error between the response vector of the individual data in the set and the corresponding group characteristic reference vector, that is, the score error between each frequency response vector in the set and the first response reference vector of the electromagnetic response parameter corresponding to the set. The purpose is to optimize the first response reference vector of each electromagnetic response parameter so that the first response reference vector of each electromagnetic response parameter can better approximate the optimal feature representation of each electromagnetic response parameter, that is, it can more accurately reflect the group characteristics. The optimization process can reduce the deviation between individual data and group reference, so that the final model is more accurate in classification.

[0059] In the actual detection process, individual data may contain noise or random errors. By iteratively optimizing the reference vector, these noises can be smoothed out so that the optimized vector can accurately reflect the true characteristics of the metal without being disturbed by accidental fluctuations in the individual data or measurement noise. The differences in electromagnetic response parameters of different metals may be subtle, and these features can be further refined through iterative optimization, especially for parameters with smaller differences, the optimization can better highlight the characteristic differences between metals. For example, the optimized first response reference vector that characterizes the impedance characteristics can more clearly show the impedance response difference between copper and brass. This refined distinction can help the system show a stronger ability to distinguish when facing similar metals, thereby improving the accuracy of detection and identification.

[0060] In this embodiment, during the optimization process based on the stochastic gradient descent algorithm, the iterative process calculates the gradient of the loss function, i.e., the target optimization function, and updates the parameters of the response vector according to the reverse of the gradient to gradually approach the global optimal solution. The optimization progress can be controlled by setting the maximum number of iterations. After generating the first response parameter vector after each electromagnetic response parameter optimization, multiple first response parameter vectors are combined to construct a first target response matrix, which is used to characterize the final characteristic response of copper under different electromagnetic response parameters in historical data.

[0061] For multiple groups of second sample data in the target sample data, a similar scheme is adopted for processing, and the response vector sets of multiple groups of second detection data in the target sample data for each electromagnetic response parameter are extracted, and a target optimization function for the response vector set of each electromagnetic response parameter is constructed. The optimization target of the target optimization function is to minimize the score error between each frequency response vector contained in the response vector set of the electromagnetic response parameter and the second response reference vector of the electromagnetic response parameter. The second response reference vector of each electromagnetic response parameter is iteratively optimized based on the stochastic gradient descent algorithm to generate the second response parameter vector of each electromagnetic response parameter, and the second target response matrix is ​​constructed based on multiple second response parameter vectors.

[0062] The first target response matrix and the second target response matrix finally constructed can characterize the common characteristics between two similar metals under different electromagnetic response parameters, and sort the noise information in the data, which can more accurately represent the performance differences between similar metals at different detection frequencies.

[0063] In the metal identification of the target metal combination, the first target response matrix and the second target response matrix can accurately describe the typical response characteristics of similar metals at different frequencies. Accurate identification results can be obtained by analyzing the difference between real-time data and feature matrices. The target metal combination identification model integrates multi-dimensional data and complex nonlinear features for classification, which is suitable for processing complex and diverse data scenarios. The construction of the response matrix is ​​based on the typical characteristics in historical data, providing a robust comparison benchmark that is not easily affected by short-term noise or data anomalies. Although the model can identify complex patterns and trends, it may be sensitive to abnormal points or noise in specific data. The fusion of the two can reduce the interference caused by noise or abnormal data, reduce misjudgment, and improve the identification accuracy of similar metals.

[0064] As an optional implementation process, in step S4, the first recognition result and the second recognition result are merged to obtain the target metal recognition result corresponding to the target detection data, which specifically includes:

[0065] The first recognition result of the target metal combination recognition model includes the first classification score of each sample label. For example, the probability that the target detection data is copper is recorded as the first classification score of the sample label copper, and the probability that it is brass is recorded as the first classification score of the sample label brass.

[0066] The second recognition result generated by analyzing the target detection data according to the first target response matrix and the second target response matrix includes a second classification score for each sample label. For example, after extracting the target detection response matrix corresponding to the target detection data, the similarity between the target detection response matrix and the first target response matrix is ​​calculated to obtain the second classification score of the sample label copper, and the similarity between the target detection response matrix and the second target response matrix is ​​calculated to obtain the second classification score of the sample label brass.

[0067] Finally, the first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label. Among them, the feature fusion method can be a weighted fusion method, and the fusion weight can be based on a pre-set weight value, or it can be an adaptive weight value determined after analyzing the first recognition result and the second recognition result. Exemplarily, the maximum value of the first classification score in the first recognition result is 0.95, and the maximum value of the second classification score in the second recognition result is 0.84. Then, the fusion weight of the first classification score can be preferably 0.95, and the fusion weight of the second classification score can be preferably 0.84, so that the final fusion result is more biased towards the actual detection result and more biased towards the historical data, avoiding the presence of data anomalies or short-term noise in the real-time detected data, which affects the recognition accuracy of similar metals and improves the anti-interference ability and robustness of the recognition system.

[0068] like Figure 2 As shown, one embodiment of the present invention provides a metal intelligent identification system based on a metal sensor, which applies any of the above-mentioned metal intelligent identification methods based on a metal sensor to perform metal intelligent identification, and includes the following structure:

[0069] A data acquisition module is used to obtain historical detection data for target metal combination identification based on a target metal sensor, where the target metal combination is two metals that interfere with each other in identification, and extract multiple groups of first detection data and second detection data from the historical detection data;

[0070] A frequency response analysis module, used to extract frequency response vectors corresponding to each group of first detection data and second detection data with respect to a plurality of electromagnetic response parameters, and extract a first frequency response reference list of each electromagnetic response parameter in the plurality of groups of first detection data and a second frequency response reference list of each electromagnetic response parameter in the plurality of groups of second detection data according to the plurality of frequency response vectors corresponding to each electromagnetic response parameter;

[0071] A sample data extraction and analysis module is used to extract a first response reference vector for each electromagnetic response parameter based on a first frequency response reference list, extract a second response reference vector for each electromagnetic response parameter based on a second frequency response reference list, filter out target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, construct a sample data set based on the target sample data and determine a first target response matrix and a second target response matrix, and train a target metal combination recognition model through the sample data set;

[0072] A metal detection and recognition module is used for, in the process of metal recognition based on the target metal sensor, if the pre-recognition result corresponding to the collected target detection data belongs to the target metal combination, inputting the target detection data into the target metal combination recognition model to generate a first recognition result corresponding to the target detection data, analyzing the target detection data according to the first target response matrix and the second target response matrix to generate a second recognition result corresponding to the target detection data, and fusing the first recognition result and the second recognition result to obtain a target metal recognition result corresponding to the target detection data;

[0073] The first recognition result and the second recognition result are fused to obtain the target metal recognition result corresponding to the target detection data, including:

[0074] The first recognition result includes the first classification score of each sample label, and the second recognition result includes the second classification score of each sample label. The first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label.

[0075] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A metal intelligent identification method based on metal sensor, characterized in that: include: Acquire historical detection data for target metal combination identification based on a target metal sensor, where the target metal combination is two metals whose identification interferes with each other, and extract multiple groups of first detection data and second detection data from the historical detection data, where the first detection data and the second detection data are detection data corresponding to one of the metals in the target metal combination; Extracting frequency response vectors corresponding to the plurality of electromagnetic response parameters of each group of first detection data and second detection data, respectively, and extracting a first frequency response reference list of each electromagnetic response parameter in the plurality of groups of first detection data and a second frequency response reference list of each electromagnetic response parameter in the plurality of groups of second detection data according to the plurality of frequency response vectors corresponding to each electromagnetic response parameter; Extracting a first response reference vector for each electromagnetic response parameter based on a first frequency response reference list, extracting a second response reference vector for each electromagnetic response parameter based on a second frequency response reference list, screening target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, constructing a sample data set based on the target sample data and determining a first target response matrix and a second target response matrix, the first target response matrix and the second target response matrix are respectively used to describe typical response characteristics of one of the metals in the target metal combination at different detection frequencies, and training a target metal combination recognition model through the sample data set; In the process of metal identification based on the target metal sensor, if the pre-identification result corresponding to the collected target detection data belongs to the target metal combination, the target detection data is input into the target metal combination identification model to generate a first identification result corresponding to the target detection data, and the target detection data is analyzed according to the first target response matrix and the second target response matrix to generate a second identification result corresponding to the target detection data, including extracting the target detection response matrix corresponding to the target detection data, calculating the similarity between the target detection response matrix and the first target response matrix and the second target response matrix respectively, to obtain the second identification result, and fusing the first identification result and the second identification result to obtain the target metal identification result corresponding to the target detection data.

2. The metal intelligent identification method based on metal sensor according to claim 1 is characterized in that: The extraction of the first frequency response reference list and the second frequency response reference list further includes: For each electromagnetic response parameter in the multiple groups of first detection data, the frequency response vectors respectively corresponding to the electromagnetic response parameter are used to determine the response score of each electromagnetic response parameter in each group of first detection data at multiple detection frequencies according to the frequency response vectors, and the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to construct a first frequency response reference list corresponding to each electromagnetic response parameter; For the frequency response vectors corresponding to each electromagnetic response parameter in multiple groups of second detection data, the response scores of each electromagnetic response parameter in each group of second detection data at multiple detection frequencies are determined according to the frequency response vectors, and the multiple response scores of each electromagnetic response parameter at each detection frequency are summarized to construct a second frequency response reference list corresponding to each electromagnetic response parameter, and the frequency response reference list corresponding to the electromagnetic response parameter includes the total response score of the electromagnetic response parameter at each detection frequency.

3. The metal intelligent identification method based on metal sensor according to claim 2 is characterized in that: Filtering target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, including: Calculating a first similarity between a frequency response vector corresponding to each electromagnetic response parameter of the first detection data and a first response reference vector, and calculating a second similarity between a frequency response vector corresponding to each electromagnetic response parameter of the second detection data and a second response reference vector; A similarity weight of each electromagnetic response parameter is calculated according to the first response reference vector and the second response reference vector of each electromagnetic response parameter, a feature score of each group of first detection data and second detection data is calculated based on multiple similarity weights, and multiple groups of first detection data and second detection data with feature scores higher than a preset feature score threshold are screened out as target sample data.

4. The metal intelligent identification method based on metal sensor according to claim 3 is characterized in that: For the target sample data, it also includes: Determine a sample label for each group of first detection data and second detection data in the target sample data, construct a sample data set including multiple groups of detection data and a sample label for each group of detection data, and train a target metal combination recognition model based on the sample data set; Extracting a plurality of groups of first detection data from the target sample data with respect to a response vector set of each electromagnetic response parameter, constructing a target optimization function for the response vector set of each electromagnetic response parameter, minimizing the score error between each frequency response vector included in the response vector set of the electromagnetic response parameter and the first response reference vector of the electromagnetic response parameter as the optimization target of the target optimization function, iteratively optimizing the first response reference vector of each electromagnetic response parameter based on a stochastic gradient descent algorithm, generating a first response parameter vector for each electromagnetic response parameter, and constructing a first target response matrix based on the plurality of first response parameter vectors; A response vector set of multiple groups of second detection data for each electromagnetic response parameter in the target sample data is extracted, and a target optimization function for the response vector set of each electromagnetic response parameter is constructed. The optimization target of the target optimization function is to minimize the score error between each frequency response vector contained in the response vector set of the electromagnetic response parameter and the second response reference vector of the electromagnetic response parameter. The second response reference vector of each electromagnetic response parameter is iteratively optimized based on the stochastic gradient descent algorithm to generate a second response parameter vector for each electromagnetic response parameter, and a second target response matrix is ​​constructed based on multiple second response parameter vectors.

5. The metal intelligent identification method based on metal sensor according to claim 1 is characterized in that: For the first recognition result and the second recognition result, it also includes: The first recognition result includes the first classification score of each sample label, and the second recognition result includes the second classification score of each sample label. The first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label.

6. The metal intelligent identification method based on metal sensor according to claim 5 is characterized in that: The target metal combination recognition model is the SVM model.

7. A metal intelligent identification system based on metal sensor, characterized in that: The system is used to implement a metal intelligent identification method based on a metal sensor as described in any one of claims 1 to 6, comprising: A data acquisition module is used to obtain historical detection data for target metal combination identification based on a target metal sensor, where the target metal combination is two metals that interfere with each other in identification, and extract multiple groups of first detection data and second detection data from the historical detection data; A frequency response analysis module, used to extract frequency response vectors corresponding to each group of first detection data and second detection data with respect to a plurality of electromagnetic response parameters, and extract a first frequency response reference list of each electromagnetic response parameter in the plurality of groups of first detection data and a second frequency response reference list of each electromagnetic response parameter in the plurality of groups of second detection data according to the plurality of frequency response vectors corresponding to each electromagnetic response parameter; A sample data extraction and analysis module is used to extract a first response reference vector for each electromagnetic response parameter based on a first frequency response reference list, extract a second response reference vector for each electromagnetic response parameter based on a second frequency response reference list, filter out target sample data from frequency response vectors corresponding to a plurality of electromagnetic response parameters based on the first response reference vector and the second response reference vector, construct a sample data set based on the target sample data and determine a first target response matrix and a second target response matrix, and train a target metal combination recognition model through the sample data set; A metal detection and identification module is used to, during the process of metal identification based on a target metal sensor, if the pre-identification result corresponding to the collected target detection data belongs to a target metal combination, input the target detection data into a target metal combination identification model to generate a first identification result corresponding to the target detection data, analyze the target detection data according to a first target response matrix and a second target response matrix to generate a second identification result corresponding to the target detection data, and fuse the first identification result and the second identification result to obtain a target metal identification result corresponding to the target detection data.

8. The metal intelligent identification system based on metal sensor according to claim 7 is characterized in that: The metal detection and recognition module fuses the first recognition result and the second recognition result to obtain a target metal recognition result corresponding to the target detection data, including: The first recognition result includes the first classification score of each sample label, and the second recognition result includes the second classification score of each sample label. The first classification score and the second classification score of each sample label are fused to obtain the target metal recognition result corresponding to the target detection data, which includes the total sample score of each sample label.

Images