AI rare earth detection method and system based on millimeter wave radar multi-metal reflection spectrum

Abstract

The invention provides an AI rare earth detection method and system based on a millimeter wave radar multi-metal reflection spectrum, and relates to the technical field of data processing, and the method comprises the steps: obtaining millimeter wave radar original reflection signal data of a to-be-detected sample; performing signal decomposition processing, and separating the reflection signal into a plurality of metal signal components based on metal reflection characteristic parameters; carrying out characteristic ratio calculation, and constructing a characteristic ratio sequence between the rare earth metal component and the non-rare earth metal component by taking the rare earth metal component as a reference; local extremum extraction and segmented weighted stacking processing are executed, a high weight is given to an interval with a rare earth metal signal response mode in the characteristic ratio, and a low weight is given to an interval with a non-rare earth metal signal response mode; extracting a signal response index of the rare earth metal, and performing quantitative analysis to generate a rare earth metal content detection result; according to the invention, the autonomy and accuracy of rare earth detection are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an AI rare earth detection method and system based on a millimeter wave radar multi-metal reflection spectrum. BACKGROUND

[0002] Currently, rare earth detection often uses methods based on traditional X-ray fluorescence analysis or inductively coupled plasma mass spectrometry. These methods generally perform sample pretreatment on ore samples, and then use instruments to quantitatively analyze the signal intensity generated by different elements. In recent years, some detection systems have introduced millimeter wave radar technology, which collects the reflection characteristic curves of different metals on millimeter wave signals, combines data processing algorithms, identifies the types and contents of metals contained in the sample, and thus realizes preliminary detection of rare earth elements.

[0003] However, in actual mine site applications, the existing millimeter wave radar detection system is easily interfered by signal overlap in multi-metal mixed samples. For example, when the sample contains metals such as iron and nickel, the high reflectivity of these metals will mask the weak reflection signal of rare earth metals, resulting in a significant deviation in the detection result of rare earth content. Especially in continuous sampling and real-time monitoring scenarios, the system is difficult to accurately separate and quantify the reflection information of rare earth elements, affecting the detection reliability and ore grading decision. SUMMARY

[0004] The purpose of the present application is to provide an AI rare earth detection method and system based on a millimeter wave radar multi-metal reflection spectrum, aiming to solve the problems mentioned in the background art.

[0005] To solve the above technical problems, the technical solution of the present application is as follows: In a first aspect, an AI rare earth detection method based on a millimeter wave radar multi-metal reflection spectrum, the method comprising: Obtaining millimeter wave radar original reflection signal data of a sample to be tested, containing the reflection response curves of various metals in the sample to be tested on millimeter wave signals; Performing signal decomposition processing on the original reflection signal data, separating the reflection signal into multiple metal signal components based on metal reflection characteristic parameters, and obtaining a metal component data set; Calculating the feature ratio of each metal component data set, respectively taking the rare earth metal component as a reference, constructing the feature ratio sequence between the rare earth metal component and the non-rare earth metal component, and obtaining feature ratio data; According to the feature ratio data, performing local extreme value extraction and segmented weighted superposition processing, giving high weight to the interval with the rare earth metal signal response mode in the feature ratio, and giving low weight to the interval with the non-rare earth metal signal response mode, and obtaining rare earth metal feature enhancement data; According to the rare earth metal characteristic enhancement data, the signal response index of the rare earth metal is extracted, and quantitative analysis is performed to generate a rare earth metal content detection result.

[0006] Preferably, the original reflection signal data is subjected to signal decomposition processing, and the reflection signal is separated into multiple metal signal components based on the metal reflection characteristic parameters to obtain a metal component dataset, including: Based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to frequency spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics; According to the signal of each frequency band, the signal segments matched with different metal characteristic parameters are extracted to obtain a preliminary response signal set of each metal; The preliminary response signal set of each metal is subjected to denoising processing to filter isolated high-frequency noise points, and the signal baseline drift is adaptively corrected to obtain a denoised response signal set; According to the denoised response signal set, the amplitude of each metal signal segment is normalized to a unified amplitude range to obtain a normalized response signal set; The metal signal segments in the normalized response signal set are archived according to the metal categories to generate a metal component dataset.

[0007] Preferably, the characteristic ratio of each metal component dataset is calculated, and the characteristic ratio sequence between the rare earth metal component and the non-rare earth metal component is constructed based on the rare earth metal component to obtain characteristic ratio data, including: All data points of each metal component dataset are traversed, and the signal strength of the rare earth metal component is taken as reference data, and the signal strength of all non-rare earth metal components at the corresponding position is taken as comparison data; According to each group of reference data and comparison data, the ratio operation is performed, and the reference data is divided by the comparison data to obtain multiple ratio subsequences, each ratio subsequence corresponding to a non-rare earth metal component; The sliding average processing is applied to each ratio subsequence to remove abnormal values and smooth the continuous data to obtain a smoothed ratio subsequence; The ratio of each ratio subsequence at each data point is sequentially combined to construct a feature vector, and the feature vectors of all data points are sequentially arranged according to the signal time sequence to generate a feature ratio dataset with a time sequence structure.

[0008] Preferably, according to the characteristic ratio data, local extreme value extraction and segmented weighted superposition processing are performed, high weight is given to the interval with the rare earth metal signal response mode in the characteristic ratio, and low weight is given to the interval with the non-rare earth metal signal response mode to obtain rare earth metal characteristic enhancement data, including: The feature ratio data is segmented according to a preset interval length, extreme value search is performed on each segmented data interval, local maximum and minimum values are sequentially extracted, and a feature sequence of the interval is formed; The similarity between the feature sequence of each interval and a preset rare earth metal signal template is calculated, if the similarity is greater than a preset similarity threshold, the interval is marked as a high weight interval, otherwise it is marked as a low weight interval; For each high weight interval, multiply all ratio data in the interval by a preset high weight factor; for a low weight interval, multiply the ratio data in the interval by a preset low weight factor, and the high weight factor is greater than the low weight factor; All segmented data after weighting processing is accumulated and superimposed to form a rare earth metal feature enhancement data sequence.

[0009] Preferably, according to the rare earth metal feature enhancement data, the signal response index of the rare earth metal is extracted, and quantitative analysis is performed to generate the rare earth metal content detection result, including: According to the rare earth metal feature enhancement data, a main peak search is performed on the sequence to identify the main peak position and its signal strength in the sequence, and a main peak signal response value is obtained; The signal is integrated in a preset continuous interval on the left and right sides of the main peak signal response value to obtain an interval area reflecting the rare earth characteristic energy; The main peak signal response value and the interval area are substituted into a preset rare earth metal standard curve relationship to convert the actual content value of the rare earth metal in the sample; The actual content value is taken as the rare earth metal content detection result, and the result is associated with the unique number of the sample to be measured.

[0010] Preferably, based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to frequency spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics, including: The original reflection signal data is subjected to time window cutting operation to obtain signal segments in multiple time periods; The time domain signal is converted into a frequency domain signal by using the fast Fourier transform method for each signal segment to obtain the amplitude-frequency characteristic curve of the signal segment; According to the typical frequency response range of the reflection characteristic parameters of various metals, the frequency domain signal is divided according to different frequency intervals, and each frequency interval is taken as an independent frequency band to generate; The data of all frequency intervals are combined in order to form multiple frequency bands covering the entire analysis range.

[0011] Preferably, according to the signal of each frequency band, the signal segment matched with different metal characteristic parameters is extracted respectively to obtain a preliminary response signal set of each metal, including: According to the signal of each frequency band, the amplitude sequence and the phase sequence in the frequency band are extracted as the to-be-compared features; According to each type of metal, the preset peak amplitude parameter and phase feature parameter at a specific frequency are called, and each to-be-compared feature of each frequency band is compared one by one to form a difference sequence of each frequency band; According to the difference sequence, a similarity calculation method is set, and the amplitude difference and the phase difference of the frequency band are compared with the corresponding preset difference threshold value, and if all the differences are lower than the corresponding difference threshold value, it is determined that the signal segment of the frequency band matches the metal feature parameter, and a matching mark is generated; All signal segments determined to match are respectively archived to the corresponding metal category to generate a preliminary response signal set of each metal.

[0012] Preferably, the preliminary response signal set of each metal is subjected to denoising processing, isolated high-frequency noise is filtered, and adaptive correction is performed on signal baseline drift to obtain a denoised response signal set, including: For each preliminary response signal set, a sliding average method is used to smooth the signal curve to generate a preliminary smoothed signal; For each data point of the preliminary smoothed signal, the amplitude difference of the data point and the amplitude of the adjacent data point is judged, and if the amplitude change of the data point exceeds the preset amplitude change threshold value, the data point is regarded as isolated high-frequency noise, and the mean value of the adjacent data points is used to replace the data point to generate a denoised signal; According to the denoised signal, a baseline trend curve is fitted, and the baseline trend curve is deducted from the signal to form a denoised response signal set.

[0013] Preferably, the similarity between the feature sequence of each interval and the preset rare earth metal signal template is calculated, including: For each interval feature sequence, a sliding window method is used for segmentation, and the interval feature sequence is divided into a plurality of preliminary subsequences that overlap with each other; For each preliminary subsequence, a normalization operation is performed respectively, so that the preliminary subsequence has a mean value of zero and a standard deviation of one to obtain a normalized subsequence; Each normalized subsequence is compared with a corresponding length of rare earth metal signal template segment point by point, and the local similarity score of the normalized subsequence and the template segment is obtained by calculating the cumulative sum of the product of each component. The local similarity scores of all normalized subsequences in the same interval are weighted and averaged to obtain the overall similarity score of the interval.

[0014] In a second aspect, an AI rare earth detection system based on a millimeter wave radar multi-metal reflection spectrum is provided, and the system includes: The data acquisition module is configured to acquire millimeter wave radar original reflection signal data of the sample to be tested, and the original reflection signal data includes reflection response curves of various metals in the sample to be tested to millimeter wave signals. The signal decomposition module is configured to perform signal decomposition processing on the original reflection signal data, separate the reflection signals into a plurality of metal signal components based on metal reflection characteristic parameters, and obtain a plurality of metal component data sets. The feature ratio calculation module is configured to calculate feature ratios of the metal component data sets, take the rare earth metal component as a reference, construct a feature ratio sequence between the rare earth metal component and the non-rare earth metal component, and obtain feature ratio data. The weighted reinforcement module is configured to perform local extreme value extraction and segmented weighted superposition processing on the feature ratio data, give a high weight to an interval with a rare earth metal signal response mode in the feature ratio, give a low weight to an interval with a non-rare earth metal signal response mode, and obtain rare earth metal feature reinforcement data. The quantitative analysis module is configured to extract a signal response index of the rare earth metal according to the rare earth metal feature reinforcement data, and perform quantitative analysis to generate a rare earth metal content detection result.

[0015] The above-mentioned scheme of the present application at least includes the following beneficial effects: The present application can effectively overcome the signal overlap and rare earth signal masking problem caused by the mixing of multiple metals in the prior art by acquiring millimeter wave radar original reflection signal data of the sample to be tested and decomposing and extracting features of the signal. By introducing signal decomposition processing based on metal reflection characteristic parameters, the response components of various metals in the original reflection signal can be independently separated, the discrimination between the rare earth metal signal and the high reflectivity metal signal is significantly improved, and the interference of main components such as iron and nickel on the weak rare earth signal is avoided.

[0016] Further, the feature ratio calculation based on the rare earth metal component and the multi-step weighted reinforcement processing can not only effectively amplify the small changes of the rare earth metal signal in the overall data, but also accurately identify the signal interval with the rare earth characteristic by using the extreme value extraction and segmented weighting mechanism, thereby greatly reducing the false detection probability caused by impurity metals and noise. In the continuous sampling and real-time monitoring scene, the present application can stably separate and quantitatively analyze the rare earth signal, realize high-sensitivity and anti-interference detection of the rare earth metal content in the ore, improve the accuracy of detection and the reliability of decision-making, and is helpful for efficient development of rare earth resources and intelligent management of mines. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a flowchart of an AI rare earth detection method based on a millimeter wave radar multi-metal reflection spectrum provided by an embodiment of the present application. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood, and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0019] As Figure 1 shown, the embodiments of the present application propose an AI rare earth detection method based on millimeter wave radar multi-metal reflection spectrum, which comprises: acquiring millimeter wave radar original reflection signal data of the sample to be tested, containing the reflection response curves of various metals in the sample to be tested to millimeter wave signals; performing signal decomposition processing on the original reflection signal data, separating the reflection signal into multiple metal signal components based on metal reflection characteristic parameters to obtain a metal component data set; performing feature ratio calculation on each metal component data set, taking the rare earth metal component as a reference, constructing a feature ratio sequence between the rare earth metal component and the non-rare earth metal component, and obtaining feature ratio data; According to the feature ratio data, perform local extreme value extraction and segmented weighted superposition processing, give high weight to the interval with rare earth metal signal response mode in the feature ratio, and give low weight to the interval with non-rare earth metal signal response mode, to obtain rare earth metal feature enhancement data; According to the rare earth metal feature enhancement data, extract the signal response index of the rare earth metal and perform quantitative analysis to generate the rare earth metal content detection result.

[0020] In the embodiment of the present application, by collecting and multi-stage processing the millimeter wave radar original reflection signal data of the sample to be tested, high-precision and intelligent detection of rare earth metal components in the sample can be realized. The method first collects original signals containing the reflection response curves of various metals, which can completely cover the weak and strong signals of all metal components in the sample, ensuring the comprehensiveness of the detection basis. Then, the signal decomposition processing is used to effectively distinguish different metal signals, so that the small response signal of rare earth metal can be accurately separated from the complex background and no longer be covered by the high-intensity main metal signal. The characteristic ratio calculation step further uses the relative change of the rare earth signal and other metal signals to strengthen the recognition degree of the rare earth component in the overall data. After the extreme value extraction and segmented weighting processing of the ratio data, the system can recognize and enhance the rare earth signal pattern in the interval, thereby suppressing the noise and interference of irrelevant metals. The finally extracted rare earth signal response index can be accurately converted into the content of rare earth metal through quantitative analysis, realizing high-sensitivity, anti-interference and high-accuracy detection of rare earth elements in complex ore samples. For example, in a mixed sample with iron, nickel and other metal contents much higher than that of rare earth elements, the rare earth signal can be accurately distinguished and its concentration can be quantified, which significantly improves the reliability and efficiency of on-site detection of rare earth ore and reduces the demand for laboratory analysis and the probability of misjudgment.

[0021] The millimeter wave radar original reflection signal data of the sample to be tested is obtained, which contains the reflection response curves of various metals in the sample to be tested, and specifically includes: Firstly, the sample to be tested is placed in the effective detection area of the millimeter wave radar detection device, and the radar emission module is started according to the preset parameter configuration, so that it continuously emits millimeter wave signals in a specific frequency range to the sample. At the same time of signal emission, the receiving module synchronously collects the echo signals reflected from the surface and inside of the sample. The system performs preliminary pretreatment on the received original echo signals, including amplification, filtering and analog-to-digital conversion of analog signals, etc., to generate digitized original reflection signal data. Then, the digitized signal is stored as a processable data stream, and each sampling point in the data contains the instantaneous reflection response of each metal element to the millimeter wave signal at a specific time. This data stream completely records the reflection characteristics of various metals in the sample to different frequency millimeter waves, providing basic data resources for subsequent decomposition, identification and analysis of different metal components.

[0022] In a preferred embodiment of the present application, the original reflection signal data is subjected to signal decomposition processing, and the reflection signal is separated into multiple metal signal components based on metal reflection characteristic parameters, to obtain a metal component data set, including: Based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to frequency spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics; According to the signal of each frequency band, signal segments matched with different metal characteristic parameters are extracted respectively to obtain a preliminary response signal set of each metal; The preliminary response signal set of each metal is subjected to denoising processing, isolated high-frequency noise points are filtered, and the signal baseline drift is adaptively corrected to obtain a denoised response signal set; According to the denoised response signal set, the amplitude of each metal signal segment is normalized to standardize all segments to a unified amplitude range to obtain a normalized response signal set; Each metal signal segment in the normalized response signal set is archived according to the metal category to generate a metal component data set.

[0023] In the embodiments of the present application, by performing signal decomposition based on metal reflection characteristic parameters on the original reflection signal data, signals of different metal categories can be extracted and processed in a targeted manner, greatly enhancing the detection capability of trace components such as rare earth metals. In the spectral analysis stage, the original time domain signal is first decomposed into respective frequency response intervals, so that the response frequency bands of different metals can be accurately divided. For example, the high-intensity reflection signals of metals such as iron and copper can be separated in their typical frequency bands, and even if the amplitude of the rare earth metal signal is low, it can also be captured separately through parameter comparison. In the feature segment extraction link, each frequency band signal is compared with the known metal parameters, and only the part that truly belongs to the response of a specific metal is retained, greatly reducing the confusion of impurity signals. Through denoising and baseline adaptive correction, high-frequency noise points and signal drift are effectively eliminated, and the signal quality is improved. Amplitude normalization processing eliminates systematic differences in signal amplitude under different samples or different test environments, making the final metal component data set more uniform and comparable. Taking on-site detection as an example, after the above decomposition and normalization processing, even in a complex environment with large changes in impurity components, the stability and traceability of each metal signal component can be ensured, facilitating subsequent feature analysis and quantitative calculation.

[0024] In a preferred embodiment of the present application, feature ratio calculation is performed on each metal component data set, and a feature ratio sequence between the rare earth metal component and the non-rare earth metal component is constructed based on the rare earth metal component to obtain feature ratio data, including: All data points of each metal component data set are traversed, the signal intensity of the rare earth metal component is taken as reference data, and the signal intensity of all non-rare earth metal components at the corresponding position is taken as comparison data; According to each group of reference data and comparison data, a ratio operation is performed, the reference data is divided by the comparison data to obtain multiple ratio subsequences, each ratio subsequence corresponding to a non-rare earth metal component; A sliding average processing is applied to each ratio subsequence to remove abnormal values and smooth continuous data to obtain a smoothed ratio subsequence; The ratios of each ratio sub-sequence at each data point are sequentially combined to form a feature vector, and the feature vectors of all data points are sequentially arranged according to the signal time sequence to generate a feature ratio data set with a time sequence structure.

[0025] In the embodiment of the present application, by performing feature ratio calculation on each metal component data set, the difference characteristics between the rare earth metal signal and other metal components can be highlighted in the signal environment of a mixed metal sample. In specific implementation, the signal intensity of the rare earth metal component is sequentially compared with the signal of other metals at the same time, and the sliding average is used to smooth the abnormal or mutant values, so that the result sequence is more continuous and true, and the interference of single-point extreme value on analysis is avoided. The ratio results of all data points are combined into a multi-dimensional feature vector sequence according to the time sequence. This structure not only highlights the relative intensity of rare earth metals and other metals at each time, but also reflects the trend of signal change with sample collection time. For example, in the process of detecting a continuous section of ore, if the rare earth metal content fluctuates or locally increases, the above-mentioned ratio feature will form a clear wave peak in the sequence, which is convenient for automatic detection system identification and alarm. The feature ratio data set provides a solid data foundation for subsequent extreme value interval extraction, signal enhancement and final content quantitative analysis, significantly improves the sensitivity and automation level of the system, and is conducive to the realization of high-throughput online monitoring and intelligent mine construction.

[0026] The ratio operation is performed according to each group of reference data and comparison data, the reference data is divided by the comparison data, and a plurality of ratio sub-sequences are obtained, and the specific operations include: First, from the processed metal component data sets, data points at each time are sequentially selected, the signal intensity of the rare earth metal component is taken as the reference data at this time, and the signal intensity of other metal components at the same time is taken as the comparison data. For each group of reference data and comparison data, the ratio calculation is performed, that is, the value of the signal intensity of the rare earth metal at this time is taken as the numerator, and the signal intensity value of each non-rare earth metal component at the same time is taken as the denominator, and the quotient of the two is calculated. The calculation result of each quotient represents the relative amplitude relationship of the rare earth signal and the signal of a certain non-rare earth metal at the same sampling time. The above process is repeated for all sampling times and all non-rare earth metal components to obtain a plurality of ratio sub-sequences arranged in time sequence, and each sub-sequence reflects the dynamic relationship between the rare earth signal and a certain non-rare earth metal signal.

[0027] The sliding average processing is applied to each ratio sub-sequence to remove mutant abnormal values and smooth the continuous data to obtain a smoothed ratio sub-sequence, and the specific operations include: For each ratio sub-sequence, a sliding window is set (such as a fixed number of consecutive data points), and the window is slid point by point over the entire ratio sub-sequence. At each sliding step, the average value of all data points in the window is taken, and the average value is taken as a new data point at the center of the current window, which replaces the corresponding data point in the original ratio sub-sequence. In this way, the value of each data point is not only affected by itself, but also affected by the neighboring data points before and after it, thereby effectively reducing the influence of sudden noise or abnormal extreme values on the overall trend of the data sequence. After the moving average processing, the generated smoothed ratio sub-sequence greatly weakens the data jumps caused by single-point anomalies or environmental disturbances while maintaining the main change trend, facilitating subsequent pattern recognition and interval analysis.

[0028] wherein the ratio of each data point in each ratio sub-sequence is combined in sequence to form a feature vector, and the feature vectors of all data points are arranged in sequence according to the signal timing to generate a feature ratio data set with a time sequence structure, specifically including: For all smoothed ratio sub-sequences, at each same sampling time, the ratio of each sub-sequence at that time is read, and these ratios are arranged in sequence according to a fixed order to form a multi-dimensional feature vector. This feature vector completely reflects the relative relationship between the rare earth metal signal and all other non-rare earth metal signals at that sampling time. The feature vectors corresponding to each sampling time are arranged in sequence according to the order of collection time to form a set of multi-dimensional data sequences that change over time, i.e., a feature ratio data set. This data set not only records the relative change trend between the metal signals, but also retains the signal evolution details at each time during the detection process, providing a rich information base for subsequent interval division, extreme value analysis, and rare earth signal recognition.

[0029] In a preferred embodiment of the present application, according to the feature ratio data, local extreme value extraction and segmented weighted superposition processing are performed, high weight is given to the interval with the response mode of the rare earth metal signal in the feature ratio, and low weight is given to the interval with the response mode of the non-rare earth metal signal, to obtain rare earth metal feature enhancement data, including: The feature ratio data is segmented according to a predetermined interval length, the extreme values of each segmented data interval are searched, the local maximum and minimum values are extracted in sequence, and a feature sequence of the interval is formed; The similarity between the feature sequence of each interval and a predetermined rare earth metal signal template is calculated, and if the similarity is greater than a predetermined similarity threshold, the interval is marked as a high weight interval, otherwise it is marked as a low weight interval; For each high weight interval, all ratio data in the interval is multiplied by a predetermined high weight factor; for a low weight interval, the ratio data in the interval is multiplied by a predetermined low weight factor, and the high weight factor is greater than the low weight factor; The all weighted processed segmented data are accumulated and superimposed to form a rare earth metal characteristic enhancement data sequence.

[0030] In the embodiment of the present application, by performing local extremum extraction and segmented weighting superposition processing on the characteristic ratio data, the recognition degree of the rare earth metal signal in the complex background can be significantly improved. The characteristic ratio data is segmented according to a preset interval, and the extremum search and characteristic sequence extraction are performed on each segmented interval, which can effectively identify the abnormal peak and valley values existing in the data. These extreme values often correspond to the local enhancement or weakening of the rare earth metal signal. Then, the interval characteristic sequence is calculated with the similarity of the rare earth metal signal template, and different weights are given to the interval according to the calculation result. The highly matched interval is weighted and enhanced, and the unmatched interval is weighted and weakened. Through this weighting accumulation process, not only the expression of the rare earth metal characteristic signal in the global data is strengthened, but also the false positive risk caused by impurity metals or noise is greatly reduced. For example, when the signals of iron or copper in the ore sample fluctuate sharply, the system can automatically reduce the weight of the ratio of these intervals, and the sensitive interval of the rare earth metal signal is weighted and amplified, so as to ensure that the finally extracted data more truly reflects the actual change of the rare earth content. The processing method is suitable for various actual detection scenes, especially under the conditions of complex ore, coexistence of multiple metals, strong background interference and the like, the reliability and high precision of the detection result can still be ensured, and the feasibility of online automatic analysis and grading decision is improved.

[0031] The preset interval length specifically includes: Before the segmented processing of the characteristic ratio data, a fixed interval length is first set according to the actual application scene and signal characteristics. The interval length can be set according to historical detection data, target signal duration, sampling rate and the like. For example, the typical appearance width of the rare earth metal signal response in the sample can be counted, and the width is converted into the number of corresponding sampling points as the length reference of a single interval. After the interval length is set, the entire characteristic ratio data set is divided into several continuous and equal-length data intervals from the beginning. Each interval contains the same number of sampling points, ensuring that the subsequent extremum search, characteristic sequence extraction and weight distribution operations are completed in a standardized data window. Through this preset length segmentation method, the statistical characteristics of the signal in the local interval can be kept consistent, which helps to improve the accuracy of interval characteristic extraction and the stability of interval similarity comparison. In actual operation, if there are multiple signal duration lengths in the detection task, multiple candidate interval lengths can be set and compared according to the actual target, and finally the interval length that best reflects the characteristics of the rare earth metal signal is selected for segmentation.

[0032] The preset high weight factor and low weight factor specifically include: Before the weight distribution of each segmented interval of the characteristic ratio data, the values of the high weight factor and the low weight factor are set according to the signal analysis task requirements and historical experimental data. The high weight factor is usually set as a positive number greater than one, which is used to significantly amplify the influence of the interval data similar to the signal pattern of rare earth metals in the final characteristic enhancement data sequence. The low weight factor is set as a positive number not greater than one (for example, one or a smaller fraction), which is used to weaken the data contribution of the interval similar to the signal pattern of rare earth metals. When setting, the weight combination that can effectively improve the recognition rate and reduce the false positive rate can be selected by analyzing the statistical distribution characteristics of the rare earth signal and the interference signal interval in multiple groups of samples. In the distribution process, the system traverses all the segmented intervals, multiplies the data points of the high similarity interval by the high weight factor, multiplies the data points of the low similarity interval by the low weight factor, and finally outputs the weighted and accumulated characteristic enhancement data. Through the amplification and weakening effect of the weight factor, the signal characteristics of rare earth metals can be more prominently expressed in a complex background, effectively improving the robustness of signal recognition and the accuracy of automatic recognition of the system.

[0033] In a preferred embodiment of the present application, according to the rare earth metal characteristic enhancement data, the signal response index of the rare earth metal is extracted and quantitatively analyzed to generate the rare earth metal content detection result, including: According to the rare earth metal characteristic enhancement data, the main peak search is performed on the sequence to identify the main peak position and its signal strength in the sequence, and the main peak signal response value is obtained; The signal is integrated in the preset continuous interval on both sides of the main peak signal response value to obtain the interval area reflecting the rare earth characteristic energy; The main peak signal response value and the interval area are substituted into the preset rare earth metal standard curve relationship to convert the actual content value of the rare earth metal in the sample; The actual content value is taken as the rare earth metal content detection result, and the result is associated with the unique number of the sample to be tested.

[0034] In the embodiment of the present application, by performing spectral analysis on the original reflection signal data based on the pre-stored reflection characteristic parameters of various metals, and dividing the signal into multiple frequency bands, the accuracy of signal decomposition and target metal identification is greatly improved. The use of time window interception operation can ensure that the local changes of the signal are sampled in detail, and the resolution of subsequent frequency domain analysis is improved. After each signal segment is accurately converted into a frequency domain signal by fast Fourier transform, the typical frequency responses of different metals are clearly separated. The system can accurately segment the original signal into several representative frequency intervals according to the parameters of each metal, and each interval can reflect the typical response characteristics of the corresponding metal. For example, for a composite ore sample containing multiple metals, the system can automatically separate the frequency intervals of rare earth metals, iron, copper and other metals, effectively avoiding the cross interference of different metal signals, and laying a solid foundation for subsequent signal matching and feature extraction. This method significantly improves the accuracy of target metal detection in complex environments, shortens the analysis time, and helps to realize high-throughput continuous detection.

[0035] wherein, according to the rare earth metal characteristic enhancement data, a main peak search is performed on the sequence to identify the main peak position and its signal strength in the sequence, and a main peak signal response value is obtained, specifically including: First, starting from the starting point of the rare earth metal characteristic enhancement data sequence, each data point is traversed in turn, and its signal strength is compared with the signal strength of adjacent data points. In the traversal process, if the signal strength of a data point is greater than that of its left and right adjacent points, and the intensity value exceeds the preset background noise threshold, the point is determined as a candidate main peak. For the case where there are multiple candidate main peaks, the candidate point with the highest signal strength can be selected as the main peak position, and the index of the point in the sequence and its signal strength are recorded. Finally, the specific position (i.e. the sampling sequence number or the corresponding time point) of the main peak and the signal response value of the main peak are output as the main characteristic parameter representing the significance of the rare earth metal signal.

[0036] wherein, for the preset continuous interval on the left and right sides of the main peak signal response value, an integral operation is performed on the signal to obtain an interval area reflecting the energy of the rare earth characteristic, specifically including: After determining the main peak position, a preset number of sampling points are expanded to the left and right of the main peak according to the sequence index corresponding to the main peak, to determine a continuous interval symmetrically centered on the main peak. The length of the interval can be set according to experience or the peak width characteristics of the rare earth signal, to ensure that the main peak response and the main signal part in the surrounding area are covered. Within the continuous interval, the signal strength of each sampling point is added in turn, and multiplied by the sampling interval time or distance represented by each point. The accumulation result of all sampling points is taken as the area value of the interval. The area value can comprehensively reflect the overall energy of the rare earth signal in the main peak and its neighborhood, and eliminate accidental errors caused by single-point noise fluctuations, which is one of the key parameters for determining the actual content of rare earth metals.

[0037] The main peak signal response value and the interval area are substituted into the preset rare earth metal standard curve relationship formula, and the actual content value of the rare earth metal in the sample is converted, and the specific operation includes: According to the rare earth metal standard curve calibrated in advance by the detection system, the main peak signal response value and the interval area are respectively taken as input parameters. The standard curve is obtained by fitting the actual content of a series of samples with known rare earth metal content and the corresponding main peak response value and area value, and the linear fitting, piecewise linear fitting or polynomial fitting method can be used. According to the standard curve relationship, the detected main peak signal response value and interval area are input, the corresponding rare earth metal content is found or interpolated, and the quantitative conversion between signal strength and actual content is realized. The system finally outputs the actual content value of the rare earth metal in the detected sample, which is used for subsequent ore classification, resource evaluation or production control.

[0038] In a preferred embodiment of the present application, based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to frequency spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics, including: The time window cutting operation is performed on the original reflection signal data to obtain signal segments in multiple time periods; The time domain signal is converted into a frequency domain signal by using the fast Fourier transform method for each signal segment, and the amplitude-frequency characteristic curve of the signal segment is obtained; According to the typical frequency response range of the reflection characteristic parameters of various metals, the frequency domain signal is divided according to different frequency intervals, and each frequency interval is taken as an independent frequency band to generate; The data of all frequency intervals are combined in order to form multiple frequency bands covering the entire analysis range.

[0039] In the embodiment of the present application, by accurately matching each frequency band signal with different metal characteristic parameters, and archiving the preliminary response signal set of each metal, the accuracy of target metal identification can be effectively improved. In specific implementation, the system extracts the amplitude sequence and phase sequence from each frequency band signal, and then compares these features with the standard parameters (such as peak amplitude, specific phase characteristics, etc.) of each metal one by one. Using the set similarity algorithm and threshold determination standard, the system can filter out the signal segment that best matches the specific metal such as rare earth metal from all frequency bands. In this way, each archived response signal set has extremely high metal specificity, significantly reducing signal confusion caused by metal coexistence or noise influence. For example, when detecting a sample containing multiple metals, even if the amplitude of some metal signals is weak, as long as its characteristics are highly consistent with the preset parameters, it can be accurately identified and stored. This process lays the foundation for subsequent denoising and normalization operations, further improving the specificity, anti-interference ability and detection reliability of the entire detection system.

[0040] Among them, the time domain signal is converted into a frequency domain signal by using the fast Fourier transform method for each signal segment, and the amplitude-frequency characteristic curve of the signal segment is obtained, which specifically includes: For each signal segment obtained by time window interception operation, the original time domain data is first input into the signal processing unit. The system uses the fast Fourier transform algorithm to perform frequency spectrum analysis on the time domain signal, and converts the amplitude information of continuous time sampling points into amplitude and phase information of different frequency components. After processing, the system outputs a sequence of data points with frequency as the independent variable and amplitude as the dependent variable, which is the amplitude-frequency characteristic curve of the signal segment. This curve directly reflects the energy distribution of the segment at each frequency component, providing a basis for subsequent metal component identification and frequency band analysis.

[0041] Among them, according to the typical frequency response range of the reflection characteristic parameters of various metals, the frequency domain signal is divided according to different frequency intervals, and each frequency interval is generated as an independent frequency band, which specifically includes: According to the reflection characteristic parameters of various metals stored in the system in advance, the system can retrieve the corresponding typical response interval of the target metal in the millimeter wave frequency range. Subsequently, the system groups all data points on the amplitude-frequency characteristic curve according to their frequency size, and divides them into different frequency intervals. Each interval corresponds to the specific response range of one or more metals, and a data set of the interval is generated separately, containing all amplitude and phase data within the frequency range. In this way, the original frequency domain signal can be divided into several frequency intervals covering the typical response of all target metals, creating conditions for feature extraction and identification of each metal.

[0042] The data of all frequency intervals are combined in sequence to form a plurality of frequency bands covering the entire analysis range, specifically including: The system sequentially arranges all the separately generated frequency interval data according to the sampling or detection order of the original signal, and combines them into a set of ordered frequency band data. The data of each frequency band not only contains the amplitude, phase and other characteristic information within it, but also retains the time information and frequency label corresponding to the original signal segment. The data set of all frequency bands collectively covers the response performance of the signal segment under test in the entire analysis frequency range. This combination process ensures that the subsequent metal feature matching and comparison steps can be successfully carried out on the basis of complete and non-missing data, improving the accuracy and system compatibility of the overall signal analysis.

[0043] In a preferred embodiment of the present application, according to the signal of each frequency band, the signal segment matched with different metal feature parameters is extracted respectively to obtain a preliminary response signal set of each metal, including: According to the signal of each frequency band, the amplitude sequence and phase sequence within the frequency band are extracted as the features to be compared; According to each type of metal, the preset peak amplitude parameter and phase feature parameter at a specific frequency are called and compared one by one with the features to be compared of each frequency band to form a difference sequence for each frequency band; According to the difference sequence, a similarity calculation method is set to compare the amplitude difference and phase difference of the frequency band with the corresponding preset difference threshold value. If all the differences are below the respective difference threshold value, it is determined that the signal segment of the frequency band is matched with the metal feature parameter, and a matching flag is generated; All signal segments determined to be matched are filed into the corresponding metal category to generate a preliminary response signal set of each metal.

[0044] In the embodiments of the present application, by denoising the preliminary response signal set of each metal, the purity of the signal and the accuracy of subsequent analysis can be significantly improved. Specifically, the sliding average method is used to smooth each signal set, effectively weakening the random high-frequency noise introduced by the measurement environment, equipment jitter or transient interference. For isolated high-amplitude mutation points appearing in the signal, by detecting the amplitude difference between the mutation point and the adjacent data points, the mutation point is determined as a high-frequency noise point and replaced by the mean value of the adjacent data points, thereby eliminating the distortion of the abnormal point to the overall data. In addition, the system automatically fits the baseline trend of each signal and deducts the baseline drift part, realizing adaptive correction of slow drift or low-frequency interference. After the above multi-level denoising and correction processing, the denoised response signal set not only retains the effective components of the original signal, but also greatly reduces the irrelevant interference. For example, when detecting complex ore samples in the field, even if the signal has strong background noise and instrument baseline drift, the system can automatically recover high-quality metal response characteristics, providing a reliable data basis for further normalization and quantitative analysis, effectively improving the accuracy and stability of the detection.

[0045] The preset specific frequency specifically includes: Before performing metal characteristic parameter matching, the system first sets one or more typical response frequencies for each target metal according to the historical detection data of various metals or the physical properties of the metals described in public literature. Generally, these preset frequencies correspond to the frequency points at which the target metal can produce significant reflection or absorption characteristics under millimeter wave irradiation. In implementation, the system locates these specific frequencies in the signal spectrum and focuses on the amplitude, phase, and other parameter characteristics of the signals at these frequencies. For example, if the reflectivity of rare earth metals shows a significant peak in a specific frequency range, the signal characteristics at this frequency point will be focused on in subsequent data comparison and analysis. The preset specific frequency can be obtained by experiment calibration, or can be flexibly adjusted according to different ore types to ensure sensitive response and high-precision identification of the system to various metal signals.

[0046] The similarity calculation method is set according to the difference sequence, and the amplitude difference and the phase difference of the frequency band are compared with the corresponding preset difference threshold, specifically including: First, for the frequency band signals to be compared and the characteristic parameters of the target metal, the corresponding amplitude sequence and phase sequence are extracted. The amplitude sequence of the frequency band signal is subtracted from the amplitude sequence of the target metal parameter one by one to form an amplitude difference sequence; the phase sequence is also subtracted in the same way to obtain a phase difference sequence. Subsequently, the system calculates the amplitude difference sequence and the phase difference sequence according to the established similarity calculation method, such as the maximum absolute difference or the root mean square difference, to obtain the amplitude difference index and the phase difference index representing the overall difference.

[0047] The calculated each difference index is compared with the difference threshold value preset by the system. If the amplitude difference index is less than or equal to the amplitude threshold value, and the phase difference index is less than or equal to the phase threshold value, it is determined that the frequency band signal matches the target metal characteristic parameter, otherwise it is determined as not matching. The threshold value can be set according to historical comparison data or actual engineering requirements in different metals and different scenes, so as to ensure that the system can effectively suppress false positives and false negatives, and improve the accuracy of metal matching.

[0048] In a preferred embodiment of the present application, the preliminary response signal set of each metal is denoised, isolated high-frequency noise is filtered, and the signal baseline drift is adaptively corrected to obtain a denoised response signal set, comprising: For each preliminary response signal set, the signal curve is smoothed using a moving average method to generate a preliminary smoothed signal; For each data point of the preliminary smoothed signal, the amplitude difference of the data point and the adjacent data points is judged, and if the amplitude change of the data point exceeds the preset amplitude change threshold, the data point is replaced with the mean value of the adjacent data points to generate a denoised signal; According to the denoised signal, the baseline trend curve is fitted and subtracted from the signal to form a denoised response signal set.

[0049] In the embodiment of the present application, the sliding window method is used to segment and process each interval feature sequence, and the normalized and point-to-point comparison mechanism is combined to realize accurate similarity calculation between the interval features and the rare earth metal signal template. The sliding window segmentation method can fully capture the details of the signal in the interval, and improve the sensitivity of pattern recognition to local features. The normalization operation eliminates the amplitude and scale differences between different subsequences, ensuring that the comparison result only reflects the consistency of the shape and change rule. By one-to-one correspondence between the normalized subsequence and the signal template segment, the local similarity score is calculated point by point to obtain the local similarity score. The system can sensitively identify the interval that matches the template. After weighted averaging of the local similarity scores of all normalized subsequences, the interval overall similarity score reflects not only the matching degree of the signal main component, but also avoids the influence of individual abnormal points. For example, in the complex scene of automatically identifying rare earth ore, the system can accurately determine whether there is a rare earth metal response feature in the signal interval based on this method, greatly enhancing the detection sensitivity and accuracy of rare earth signals, and providing data basis for subsequent weighting processing and final result output.

[0050] The preliminary response signal set of each metal is denoised, isolated high-frequency noise is filtered, and the signal baseline drift is adaptively corrected to obtain a denoised response signal set, comprising: For the set of preliminary response signals obtained under each metal category, a suitable sliding window length is first selected for the signal sequence. Then, the window is slid point by point from the start of the signal sequence, and the arithmetic mean of all data points in the window is calculated each time, and the mean is taken as the new signal value at the center position of the window. In this way, the entire signal sequence is sequentially covered, and each point in the original signal is replaced by the average of several points around it, obtaining the smoothed signal sequence. Through the moving average method, random high-frequency noise and sudden abnormal fluctuations in the measurement process can be effectively eliminated, making the signal curve more continuous and truly reflecting the changing trend of the metal response, laying a data foundation for subsequent noise point identification and baseline fitting.

[0051] wherein the preset amplitude change threshold value specifically includes: When detecting abnormal points in the smoothed signal, the system needs to set a preset amplitude change threshold value in advance. The threshold value can be determined according to the statistical analysis results of historical sample signal amplitudes or actual test experience, and is usually taken as the maximum expected change amplitude in the normal fluctuation range of the signal. In actual operation, when judging whether a sampling point is a noise point, the system calculates the amplitude difference between the point and its adjacent sampling points (such as one point before and after). If the difference is greater than the preset amplitude change threshold value, it is determined that the point is an abnormal mutation point or an isolated high-frequency noise point. At this time, the system replaces the value of the point with the mean of its adjacent sampling points to ensure the continuity and smoothness of the signal curve. Through this setting, accidental abnormal signals can be effectively filtered out, and normal metal response characteristics will not be damaged.

[0052] wherein a baseline trend curve is fitted according to the denoised signal, and the baseline trend curve is subtracted from the signal to form a set of denoised response signals, specifically including: After completing noise removal and signal smoothing, the system uses a baseline fitting algorithm (such as least squares polynomial fitting) to analyze the global trend of the remaining signal sequence. The system takes the amplitudes and corresponding serial numbers of all sampling points as input and fits a baseline curve that can represent the overall trend of the signal. Then, the system traverses each sampling point of the signal sequence and subtracts the value of the fitted baseline curve at that point from the amplitude of the sampling point to automatically correct the signal baseline drift. After this processing, the signal obtained is the response signal set after removing the baseline drift. This processing can eliminate the slow background influence caused by instrument zero point change, environmental temperature drift, etc., so that the final signal more accurately reflects the real-time response of the metal composition, greatly improving the accuracy of subsequent feature analysis and quantitative detection.

[0053] In a preferred embodiment of the present application, the similarity between the feature sequence of each interval and the preset rare earth metal signal template is calculated, including: For each interval of the characteristic sequence, a sliding window method is used for segmentation, and the characteristic sequence in the interval is divided into a plurality of preliminary subsequences which overlap with each other; For each preliminary subsequence, a normalization operation is respectively performed, so that the preliminary subsequence has a mean value of zero and a standard deviation of one, and a normalized subsequence is obtained; Each normalized subsequence is compared with a rare earth metal signal template fragment of a corresponding length in a point-by-point manner, and a local similarity score of the normalized subsequence and the template fragment is obtained by calculating the accumulation sum of the product of each component. The local similarity scores of all normalized subsequences in the same interval are weighted and averaged to obtain a similarity score of the interval as a whole.

[0054] In the embodiments of the present application, by segmenting the characteristic ratio data according to the preset interval length, performing extreme value search and characteristic sequence extraction on each segmented interval, and combining similarity calculation and weighted processing, the expression and differentiation degree of the rare earth metal signal in the overall data can be further enhanced. Specifically, the interval segmentation and extreme value search process can effectively find the main change characteristics of the signal in different intervals, and distinguish the rare earth signal from the interference signal. For high-similarity intervals, the prominence of the rare earth signal is significantly improved by assigning a high weight and accumulating its data; and for low-similarity intervals, the background interference is effectively suppressed by reducing the weight. Finally, all the segmented data after weighted processing are superimposed to form a rare earth metal characteristic enhanced data sequence, which can truly reflect the spatial distribution and content change trend of the rare earth metal in the sample. For example, in the process of ore continuous sampling or real-time monitoring, the processing mechanism can dynamically capture the appearance, enhancement or weakening process of the rare earth signal, providing technical support for efficient development and intelligent grading of rare earth mineral resources, and improving the automation level and adaptability of the system.

[0055] The embodiments of the present application also provide an AI rare earth detection system based on a millimeter wave radar multi-metal reflection spectrum, which comprises: A data acquisition module is configured to acquire millimeter wave radar original reflection signal data of a sample to be measured, which contains reflection response curves of various metals in the sample to be measured to millimeter wave signals; A signal decomposition module is configured to perform signal decomposition processing on the original reflection signal data, separate the reflection signal into a plurality of metal signal components based on metal reflection characteristic parameters, and obtain a metal component dataset; A characteristic ratio calculation module is configured to calculate a characteristic ratio of each metal component dataset, and construct a characteristic ratio sequence between a rare earth metal component and a non-rare earth metal component based on the rare earth metal component, to obtain characteristic ratio data. The weighted reinforcement module is used for performing local extreme value extraction and segmented weighted superposition processing according to the characteristic ratio data, giving a high weight to an interval with a rare earth metal signal response mode in the characteristic ratio and giving a low weight to an interval with a non-rare earth metal signal response mode, and obtaining rare earth metal characteristic reinforcement data; The quantitative analysis module is used for extracting a signal response index of the rare earth metal according to the rare earth metal characteristic reinforcement data, performing quantitative analysis, and generating a rare earth metal content detection result.

[0056] It should be noted that the system is a system corresponding to the above method, and all the implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0057] Embodiments of the present application also provide a computing device, comprising a processor and a memory storing a computer program, wherein the computer program is executed by the processor to perform the method described above. All the implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0058] Embodiments of the present application also provide a computer readable storage medium storing instructions, wherein the instructions are executed on a computer to make the computer perform the method described above. All the implementation manners in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0059] The above is the preferred embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. An AI rare earth detection method based on millimeter wave radar multi-metal reflection patterns is characterized by: The method comprises: Obtain the original millimeter-wave radar reflection signal data of the sample to be tested, including the reflection response curves of various metals in the sample to the millimeter-wave signal; Perform signal decomposition processing on the original reflection signal data, separate the reflection signal into multiple metal signal components based on the metal reflection characteristic parameters, and obtain the data sets of each metal component; Calculate the characteristic ratio of each metal component data set, and construct a characteristic ratio sequence between the rare earth metal component and the non-rare earth metal component based on the rare earth metal component to obtain the characteristic ratio data; Based on the characteristic ratio data, local extreme value extraction and segmented weighted superposition processing are performed to assign high weights to intervals with rare earth metal signal response patterns in the characteristic ratios and assign low weights to intervals with non-rare earth metal signal response patterns, thereby obtaining rare earth metal feature enhancement data; Based on the rare earth metal characteristic enhancement data, the signal response indicators of rare earth metals are extracted and quantitatively analyzed to generate rare earth metal content detection results.

2. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection spectrum according to claim 1 is characterized in that: The original reflection signal data is decomposed and separated into multiple metal signal components based on the metal reflection characteristic parameters, and the metal component data sets are obtained, including: Based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics; According to the signal of each frequency band, the signal segments matching the characteristic parameters of different metals are extracted to obtain the preliminary response signal set of each metal; De-noising the initial response signal set of each metal, filtering out isolated high-frequency noise points, and adaptively correcting the signal baseline drift to obtain a de-noised response signal set; According to the denoised response signal set, each metal signal segment is amplitude normalized, and all segments are standardized to a uniform amplitude range to obtain a normalized response signal set; Each metal signal segment in the normalized response signal set is filed according to metal category to generate a data set of each metal component.

3. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection pattern according to claim 1 is characterized in that: The characteristic ratios of each metal component data set were calculated. Taking the rare earth metal component as the benchmark, the characteristic ratio sequences between the rare earth metal component and the non-rare earth metal component were constructed to obtain the characteristic ratio data, including: Traverse all data points of each metal component data set, use the signal strength of the rare earth metal component as reference data, and use the signal strength of all non-rare earth metal components at the corresponding position as comparison data; Performing a ratio operation on each set of reference data and comparison data, dividing the reference data by the comparison data to obtain a plurality of ratio subsequences, each ratio subsequence corresponding to a non-rare earth metal component; Apply sliding average processing to each ratio subsequence to remove sudden outliers and smooth continuous data to obtain a smoothed ratio subsequence; The ratios of the ratio subsequences at each data point are sequentially combined to construct a feature vector, and the feature vectors of all data points are arranged in sequence according to the signal time sequence to generate a feature ratio data set with a time series structure.

4. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection pattern according to claim 1 is characterized in that: Based on the characteristic ratio data, local extreme value extraction and segmented weighted superposition processing are performed, and high weights are assigned to intervals with rare earth metal signal response patterns in the characteristic ratios, and low weights are assigned to intervals with non-rare earth metal signal response patterns, thereby obtaining rare earth metal feature enhancement data, including: The characteristic ratio data is segmented according to the preset interval length, and the extreme value search is performed on the interval of each segmented data, and the local maximum and minimum values ​​are extracted in sequence to form the characteristic sequence of the interval; Calculate the similarity between the characteristic sequence of each interval and the preset rare earth metal signal template. If the similarity is greater than the preset similarity threshold, mark the interval as a high-weight interval; otherwise, mark it as a low-weight interval. For each high-weight interval, all ratio data within the interval are multiplied by a preset high-weight factor; for a low-weight interval, all ratio data within the interval are multiplied by a preset low-weight factor, and the high-weight factor is greater than the low-weight factor; All weighted segmented data are accumulated and superimposed to form a rare earth metal feature enhanced data sequence.

5. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection spectrum according to claim 1 is characterized in that: Based on the rare earth metal characteristic enhancement data, the signal response indicators of rare earth metals are extracted and quantitatively analyzed to generate rare earth metal content detection results, including: According to the rare earth metal feature enhancement data, a main peak search is performed on the sequence to identify the main peak position and signal intensity in the sequence and obtain the main peak signal response value; Integrate the signal within the preset continuous intervals on both sides of the main peak signal response value to obtain the interval area reflecting the rare earth characteristic energy; Substitute the main peak signal response value and the interval area into the preset rare earth metal standard curve relationship to calculate the actual content of rare earth metals in the sample; The actual content value is used as the rare earth metal content test result, and the result is associated with the unique number of the sample to be tested.

6. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection spectrum according to claim 2 is characterized in that: Based on the pre-stored reflection characteristic parameters of various metals, the original reflection signal data is subjected to spectrum analysis, and the original signal is divided into multiple frequency bands according to the frequency response characteristics, including: Performing a time window interception operation on the original reflection signal data to obtain signal segments within multiple time periods; For each signal segment, the fast Fourier transform method is used to convert the time domain signal into a frequency domain signal to obtain the amplitude-frequency characteristic curve of the signal segment; According to the typical frequency response range of various metal reflection characteristic parameters, the frequency domain signal is divided into different frequency intervals, and each frequency interval is generated as an independent frequency band; The data of all frequency intervals are combined in sequence to form multiple frequency bands covering the entire analysis range.

7. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection spectrum according to claim 2 is characterized in that: Based on the signal of each frequency band, the signal segments that match the characteristic parameters of different metals are extracted to obtain the preliminary response signal set of each metal, including: According to the signal of each frequency band, the amplitude sequence and phase sequence within the frequency band are extracted as the features to be compared; According to each type of metal, the peak amplitude parameter and phase characteristic parameter at the preset specific frequency are called, and compared one by one with the characteristics to be compared in each frequency band in turn to form a difference sequence for each frequency band; According to the difference sequence, a similarity calculation method is set to compare the amplitude difference and phase difference of the frequency band with the corresponding preset difference thresholds. If all the differences are lower than the corresponding difference thresholds, it is determined that the signal segment of the frequency band matches the metal characteristic parameters and a matching mark is generated; All signal segments that are judged to be matched are filed into corresponding metal categories to generate a preliminary response signal set for each metal.

8. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection pattern according to claim 2 is characterized in that: The initial response signal set of each metal is denoised, isolated high-frequency noise points are filtered, and the signal baseline drift is adaptively corrected to obtain a denoised response signal set, including: For each set of preliminary response signals, the signal curve is smoothed using the sliding average method to generate a preliminary smoothed signal; For each data point of the preliminary smoothed signal, the difference between its amplitude and the amplitude of the adjacent data points before and after is determined. If the amplitude change of the data point exceeds the preset amplitude change threshold, it is regarded as an isolated high-frequency noise point and replaced with the mean of the adjacent data points to generate a denoised signal. According to the denoised signal, a baseline trend curve is fitted, and the baseline trend curve is subtracted from the signal to form a denoised response signal set.

9. The AI ​​rare earth detection method based on millimeter wave radar multi-metal reflection pattern according to claim 4 is characterized in that: Calculate the similarity between the characteristic sequence of each interval and the preset rare earth metal signal template, including: For the characteristic sequence of each interval, the sliding window method is used to segment the characteristic sequence in the interval into several overlapping preliminary subsequences; For each preliminary subsequence, perform normalization operation respectively, so that the mean of the preliminary subsequence is zero and the standard deviation is one, and obtain a normalized subsequence; Performing point-to-point comparison of each normalized subsequence with a rare earth metal signal template fragment of corresponding length, and calculating the cumulative sum of the products of their respective components to obtain a local similarity score between the normalized subsequence and the template fragment; The weighted average of the local similarity scores of all normalized subsequences in the same interval is performed to obtain the overall similarity score of the interval.

10. AI rare earth detection system based on millimeter wave radar multi-metal reflection spectrum, characterized by: Applied to the method according to any one of claims 1 to 9, the system comprises: The data acquisition module is used to obtain the original millimeter-wave radar reflection signal data of the sample to be tested, including the reflection response curves of various metals in the sample to be tested to the millimeter-wave signal; A signal decomposition module is used to perform signal decomposition processing on the original reflection signal data, separate the reflection signal into multiple metal signal components based on the metal reflection characteristic parameters, and obtain a data set of each metal component; The characteristic ratio calculation module is used to calculate the characteristic ratio of each metal component data set, and construct a characteristic ratio sequence between the rare earth metal component and the non-rare earth metal component based on the rare earth metal component to obtain the characteristic ratio data; A weighted enhancement module is used to perform local extreme value extraction and segmented weighted superposition processing based on the characteristic ratio data, assigning high weights to intervals with rare earth metal signal response patterns in the characteristic ratios and assigning low weights to intervals with non-rare earth metal signal response patterns, thereby obtaining rare earth metal characteristic enhancement data; The quantitative analysis module is used to extract the signal response index of rare earth metals based on the rare earth metal characteristic enhancement data, and perform quantitative analysis to generate rare earth metal content detection results.

Images