Mining industry all-element intelligent online detection and analysis method based on laser-induced breakdown spectroscopy

Abstract

The invention relates to the technical field of spectrum on-line detection, and particularly discloses a mining industry all-element intelligent on-line detection analysis method based on laser-induced breakdown spectroscopy, which comprises the following steps: acquiring mining industry on-line mineral aggregate flow, initial laser parameters, real-time environment parameters and mineral aggregate preprocessing information, performing pulse coding laser-induced plasma optimization, and performing laser-induced plasma optimization; self-adaptive spectrum collection is carried out, background suppression is carried out on a mining industry scene, a net characteristic spectrum after background removal is obtained, a spectrum band attention-element category attention double-branch network is constructed, a high-dimensional compact characteristic vector, a characteristic attention weight map and an element preliminary identification result are output, and the high-dimensional compact characteristic vector and the element preliminary identification result are obtained; and carrying out feature weighting-plasma parameter correction fusion analysis, carrying out mining all-element detection correction, and outputting corrected final element content and a mineral identification result. The method solves the problems of complex sample pretreatment, long time consumption, low light element detection precision and incapability of online detection in the traditional detection technology.

Description

Technical Field

[0001] This invention relates to the field of online spectral detection technology, specifically to an intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy. Background Technology

[0002] Currently, modern mining is developing towards refinement and efficiency, requiring real-time monitoring of core indicators such as ore grade to support dynamic adjustments in mining and beneficiation processes, thereby improving resource utilization and economic benefits. The construction of intelligent mining systems necessitates real-time and comprehensive elemental data to support intelligent control and digital twins, promoting automation and reduced manpower in production processes.

[0003] Currently, there are still some shortcomings in the research on intelligent online detection and analysis of all elements in the mining industry. Specifically, traditional detection technologies have inherent shortcomings such as complex sample pretreatment, long processing time, low accuracy of light element detection, and inability to perform online detection, making it difficult to meet the industry's demand for rapid, non-destructive, all-element, and low-pollution detection. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an intelligent online detection and analysis method for all elements in the mining industry based on laser-induced breakdown spectroscopy, which can effectively solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, this invention employs the following technical solution: a smart online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy, comprising the following steps: collecting online ore flow data, initial laser parameters, real-time environmental parameters, and ore preprocessing information; performing pulse-coded laser-induced plasma optimization; and outputting plasma time-resolved spectra and key plasma parameters; performing adaptive spectral acquisition based on plasma time-resolved spectra and key plasma parameters, and suppressing background in the mining scene to obtain a background-free net feature spectrum; constructing a spectral band attention-element category attention dual-branch network based on the net feature spectrum and ore matrix type, and outputting a high-dimensional compact feature vector, feature attention weight map, and preliminary element identification results; performing feature weighting-plasma parameter correction fusion analysis on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain initial values ​​of quantitative content for each element and confidence levels of quantitative results; and combining the confidence levels of quantitative results, high-dimensional compact feature vectors, and net feature spectra to perform full-element detection correction in mining, and outputting the corrected final element content and ore type identification results.

[0006] As a further method, online mining material flow, initial laser parameters, real-time environmental parameters, and material pretreatment information are collected. The specific analysis process is as follows: Online mining material flow, initial laser parameters, real-time environmental parameters, and material pretreatment information are collected: Online mining material flow specifically includes material belt conveyor speed, material particle size, and material moisture content; Initial laser parameters specifically include fundamental frequency, pulse energy range, pulse width, and repetition frequency; Real-time environmental parameters specifically include real-time ambient temperature T, real-time ambient humidity, and real-time ambient dust concentration; Material pretreatment information specifically includes preliminary screening particle size and material surface roughness.

[0007] As a further method, pulse-coded laser-induced plasma optimization is performed, outputting the plasma time-resolved spectrum and key plasma parameters. The specific analysis process is as follows: Using a genetic algorithm, online ore flow data, initial laser parameters, real-time environmental parameters, and ore preprocessing information are transformed into a specific three-pulse coded sequence with the goal of maximizing the signal-to-noise ratio and optimizing the stability of the laser-induced plasma excitation temperature. The optimized coded sequence is then obtained. Real-time imaging of the ore surface is performed using a high-speed linear array camera, combined with a laser ranging module to calculate the surface undulation height. Simultaneously, visual positioning is used to monitor the ore flow's offset on the conveyor belt. A piezoelectric ceramic-driven dynamic lens group is used to adjust the surface undulation height according to the... The laser was positioned to ensure the laser focus always falls on the ore surface, based on the surface undulation height stored in the database. The lens position was then adjusted. Following the optimized coding sequence, the laser emitted three pulses: a preheating pulse bombarded the ore surface, the main excitation pulse penetrated the ore core to form high-temperature plasma, and a supplementary pulse replenished energy during the initial plasma expansion to suppress rapid cooling. A spectrometer simultaneously acquired the plasma's continuous radiation spectrum, selecting two spectral lines for iron at 259.94 nm and 260.07 nm. The excitation temperature of the laser-induced plasma was calculated using the Boltzmann diagram method, with a target range of 8000-15000 K. The Balmer series of hydrogen atoms was measured. The Stark broadening of the line is used to calculate the plasma electron density induced by the laser; the plasma excitation temperature and plasma electron density are recorded as key plasma parameters; the plasma time-resolved spectrum and key plasma parameters are output.

[0008] As a further method, adaptive spectral acquisition is performed based on plasma time-resolved spectroscopy and key plasma parameters, and background suppression is applied to the mining scene to obtain the net feature spectrum after background removal. The specific analysis process is as follows: obtain the mineral matrix type; determine the optimal spectral acquisition window based on the plasma excitation temperature; obtain the original spectral intensity based on plasma time-resolved spectroscopy and acquire the instrument noise spectrum; construct a dynamic gradient background suppression formula with adaptive weights to suppress the background of the mining scene and obtain the net feature spectrum after background removal.

[0009] As a further method, a dynamic gradient background suppression formula with adaptive weights is constructed to suppress the background in the mining scene, obtaining the net feature spectrum after background removal. The specific analysis process is as follows: Constructing a dynamic gradient background suppression formula with adaptive weights, the specific calculation formula is as follows: ; in, Net characteristic spectrum, The original spectral intensity, The background spectrum is based on nonnegative matrix factorization. The first gradient of the background spectrum. For the instrument noise spectrum, As the matrix background suppression weight, As gradient background suppression weights, For noise suppression weights; ; ; ; in, The plasma excitation temperature, This represents the real-time ambient dust concentration. The baseline dust concentration is stored in the database. Real-time ambient humidity; Output the net feature spectrum after background removal.

[0010] As a further method, based on the net feature spectrum and the mineral matrix type, a dual-branch network of spectral band attention and element category attention is constructed. This network outputs a high-dimensional compact feature vector, a feature attention weight map, and preliminary element identification results. The specific analysis process is as follows: Obtain the labeled mineral sample spectral dataset; construct the dual-branch network of spectral band attention and element category attention; input the net feature spectrum, standardize the input net feature spectrum (including mean subtraction, variance normalization, and smoothing); and then... The module learns the weights of each spectral band, maps the preprocessed net feature spectrum to a feature map, compresses the global features of the net feature spectrum through global average pooling, learns the band weights through a fully connected layer, and outputs the attention weights of each band through a sigmoid function. Four convolutional blocks are set, each containing one convolutional layer, one batch normalization (BN) layer, one ReLU activation function, and one max pooling layer. The kernel sizes of the convolutional layers are 3×3, 5×5, 3×3, and 5×5, with output channels of 32, 64, 128, and 256, respectively. The kernel size of the max pooling layer is 2×2 with a stride of 2, extracting local peak shapes and intensities of the spectrum. Based on the mineral matrix type, an attention matrix corresponding to the element category is loaded. After fusing the convolutional features and attention weights, dimensionality reduction is performed through a fully connected layer, outputting a feature vector F, denoted as a high-dimensional compact feature vector. Simultaneously, the feature attention weight map and preliminary element identification results are output.

[0011] As a further method, a feature-weighted plasma parameter correction fusion analysis is performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain the initial values ​​of quantitative content for each element and the confidence level of the quantitative results. The specific analysis process is as follows: A feature-weighted plasma parameter correction fusion analysis is performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to construct a feature-weighted plasma parameter correction quantitative model and calculate the initial values ​​of quantitative content for each element; the similarity between the net feature spectrum and the standard spectral template of the same mineral in the training set is calculated using cosine similarity and recorded as the confidence similarity; the stability of the key plasma parameters is calculated based on the weighted average of the relative standard deviations of the key plasma parameters; the confidence similarity and the stability of the key plasma parameters are weighted and summed to obtain the confidence factor of the quantitative results; the quantitative result confidence factor-quantitative result confidence level mapping table stored in the database is obtained, and the matching quantitative result confidence level is determined based on the current quantitative result confidence factor; the initial values ​​of quantitative content for each element and the confidence level of the quantitative results are output.

[0012] As a further method, a characteristic-weighted plasma parameter-corrected quantitative model was constructed to calculate the initial values ​​of elemental quantitative content. The specific analysis process is as follows: ; in, This provides initial values ​​for the quantitative content of elements. The feature-weighted regression coefficient matrix stored in the database. It is a high-dimensional compact feature vector. The bias vector is stored in the database. This is the difference between the current plasma excitation temperature and the reference plasma excitation temperature stored in the database. This is the difference between the current plasma electron density and the baseline plasma electron density stored in the database. The plasma excitation temperature correction coefficients are stored in the database. The plasma electron density correction coefficients are stored in the database; Output the initial values ​​of the quantitative content of the elements.

[0013] As a further method, combining quantitative result confidence, high-dimensional compact feature vectors, and net feature spectra, we perform full-element detection correction for minerals, and output the corrected final element content and mineral type identification results. The specific analysis process is as follows: Construct a mineral type fingerprint database, which specifically includes spectral fingerprints and elemental proportion fingerprints, and perform automatic mineral type identification and mixed ore ratio calculation: A dual verification strategy of fingerprint matching + feature vector classification is adopted: The net feature spectrum is correlated with the spectral fingerprints in the fingerprint database. If the Pearson correlation coefficient is not less than 0.85, it is considered a successful match. At the same time, the high-dimensional compact feature vector is input into the trained SVM mineral type classifier to obtain the classification result. If the two are consistent, the mineral type name is output; If the net feature spectrum matches multiple mineral types represented by spectral fingerprints in the mineral type fingerprint database, the mixed ore ratio is calculated based on the elemental proportion fingerprint; Output the corrected final element content and mineral type identification results.

[0014] As a further method, a mineral fingerprint database is constructed. The specific analysis process is as follows: a standardized fingerprint database is established, and spectral fingerprints and elemental ratio fingerprints of minerals detected over the past five years are collected and recorded as the mineral fingerprint database. Among them, the spectral fingerprint is the characteristic spectral template of the minerals detected over the past five years, containing the characteristic peak positions and intensity ratios of key elements, and the elemental ratio fingerprint is the content ratio of typical elements of the minerals detected over the past five years.

[0015] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) This invention provides a method for intelligent online detection and analysis of all elements in mining based on laser-induced breakdown spectroscopy. By deeply integrating laser-induced breakdown spectroscopy with pulse-coded laser optimization, adaptive spectral acquisition, background suppression and dual-branch attention network, it achieves rapid, accurate and full-element detection of online mineral flow in mining. Pulse-coded laser can dynamically optimize plasma generation according to mineral characteristics and environmental conditions, significantly improving spectral quality and stability. Adaptive spectral acquisition and background suppression technology effectively reduce noise and interference in complex mining scenarios, making the net feature spectrum more representative, laying the foundation for subsequent element identification and quantitative analysis.

[0016] (2) This invention automatically focuses on key spectral features and element-related information through a dual-branch network structure of spectral band attention and element category attention, thereby improving the model's adaptability to complex mineral matrices. The output high-dimensional compact feature vector and attention weight map not only improve the accuracy of element identification but also enhance the interpretability of the model. Furthermore, through the fusion analysis of feature weighting and plasma parameter correction, it can effectively compensate for matrix effects and measurement fluctuations, improve the accuracy and confidence of quantitative results, and provide real-time and high-precision data support for online sorting, intelligent ore blending and production process optimization in the mining industry, thereby promoting the development of the mining industry towards intelligence, refinement and greening. Attached Figure Description

[0017] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 As shown, this invention provides an intelligent online detection and analysis method for all elements in the mining industry based on laser-induced breakdown spectroscopy, including: collecting online ore flow data, initial laser parameters, real-time environmental parameters, and ore pretreatment information; performing pulse-coded laser-induced plasma optimization; and outputting plasma time-resolved spectra and key plasma parameters.

[0021] The process involves collecting online ore flow data, initial laser parameters, real-time environmental parameters, and ore pretreatment information. Specifically, the online ore flow data includes the ore belt conveyor speed, ore particle size, and ore moisture content. Initial laser parameters include the fundamental frequency, pulse energy range, pulse width, and repetition frequency. Real-time environmental parameters include real-time ambient temperature (T), real-time ambient humidity, and real-time ambient dust concentration. Ore pretreatment information includes the initial screening particle size and ore surface roughness.

[0022] Furthermore, pulse-coded laser-induced plasma optimization was performed, outputting the plasma time-resolved spectrum and key plasma parameters. The specific analysis process involved using a genetic algorithm to transform online ore flow data, initial laser parameters, real-time environmental parameters, and ore preprocessing information into a specific three-pulse coded sequence, aiming to maximize the signal-to-noise ratio and optimize the stability of the laser-induced plasma excitation temperature. The optimized coded sequence was then obtained by: real-time imaging of the ore surface using a high-speed linear array camera, combined with a laser ranging module to calculate the surface undulation height, and simultaneously monitoring the ore flow offset on the conveyor belt using visual positioning; and using piezoelectric ceramics to drive a dynamic lens group based on the surface undulation height. The system determines the lens position corresponding to the surface undulation height stored in the database, ensuring the laser focus always falls on the ore surface, and adjusts the lens position. Following the optimized coding sequence, the laser emits three pulses: a preheating pulse bombards the ore surface, the main excitation pulse core penetrates the ore to form high-temperature plasma, and a supplementary pulse replenishes energy during the initial plasma expansion to suppress rapid cooling. A spectrometer simultaneously acquires the plasma continuous radiation spectrum, selecting two spectral lines for iron at 259.94 nm and 260.07 nm. The excitation temperature of the laser-induced plasma is calculated using the Boltzmann diagram method, with a target range of 8000-15000 K. The Balmer series of hydrogen atoms is measured. lines Expansion degree, combined with The broadening factor is used to calculate the plasma electron density induced by the laser; the plasma excitation temperature and plasma electron density are recorded as key plasma parameters; the plasma time-resolved spectrum and key plasma parameters are output.

[0023] In one specific embodiment, by comprehensively collecting online ore flow data, initial laser parameters, real-time environmental parameters, and ore preprocessing information, and utilizing a genetic algorithm for pulse-coded laser-induced plasma optimization, the laser pulse sequence and focus position can be dynamically adjusted according to ore characteristics and environmental changes, significantly improving the stability and spectral quality of plasma generation. Multi-source information fusion makes laser parameters more adaptable to complex and variable mining scenarios. The three-pulse coding sequence, through the synergistic effects of preheating, main excitation, and energy replenishment, effectively enhances plasma intensity, suppresses cooling effects, and improves the spectral signal-to-noise ratio. The dynamic lens group, combined with visual positioning and laser ranging, achieves real-time compensation for ore surface undulations and belt offset, ensuring the laser focus remains stably on the ore surface and reducing measurement deviations. By calculating the plasma excitation temperature and electron density and outputting them as key parameters, a more accurate physical basis is provided for subsequent spectral analysis and quantitative models, thereby comprehensively improving the stability, repeatability, and accuracy of detection, laying the foundation for online full-element detection in the mining industry.

[0024] Adaptive spectral acquisition is performed based on plasma time-resolved spectroscopy and key plasma parameters, and background suppression is applied to the mining scene to obtain the net feature spectrum after background removal.

[0025] Specifically, adaptive spectral acquisition is performed based on plasma time-resolved spectroscopy and key plasma parameters, and background suppression is applied to the mining scene to obtain the net feature spectrum after background removal. The specific analysis process is as follows: the type of mineral matrix is ​​obtained; based on the plasma excitation temperature, the optimal spectral acquisition window is determined; when the plasma excitation temperature is higher than 12000K, the acquisition window is set to... When the plasma excitation temperature is below 12000K but not below 10000K, the acquisition window is set to... When the plasma excitation temperature is below 10000K, the acquisition window is set to... The original spectral intensity was obtained based on plasma time-resolved spectroscopy, and the instrument noise spectrum was collected. A dynamic gradient background suppression formula with adaptive weights was constructed to suppress the background of the mining scene and obtain the net feature spectrum after background removal.

[0026] Furthermore, a dynamic gradient background suppression formula with adaptive weights is constructed to suppress the background in the mining scene, obtaining the net feature spectrum after background removal. The specific analysis process is as follows: A dynamic gradient background suppression formula with adaptive weights is constructed, and the specific calculation formula is as follows: ; in, Net characteristic spectrum, The original spectral intensity, The background spectrum is based on nonnegative matrix factorization. The first gradient of the background spectrum. For the instrument noise spectrum, As the matrix background suppression weight, As gradient background suppression weights, For noise suppression weights; ; ; ; in, The plasma excitation temperature, This represents the real-time ambient dust concentration. The baseline dust concentration is stored in the database. Real-time ambient humidity; Output the net feature spectrum after background removal.

[0027] In one specific embodiment, adaptive spectral acquisition based on plasma time-resolved spectroscopy and key parameters, combined with a background suppression formula containing dynamic weights, automatically adjusts the acquisition time interval according to the plasma excitation temperature. This enables the capture of the most representative characteristic spectra at different plasma evolution stages, effectively improving signal strength and the clarity of elemental characteristic peaks. The dynamic gradient background suppression formula introduces three suppression terms: matrix background, background gradient, and instrument noise. It also uses plasma temperature, ambient dust concentration, and humidity to adjust the weights in real time, allowing background subtraction to adapt to changes in the mineral matrix and environmental interference. This significantly reduces the noise impact caused by continuous background, dust scattering, and humidity, resulting in a purer and more stable net characteristic spectrum. This improves the signal-to-noise ratio and reliability of the spectral data, provides high-quality input for subsequent elemental identification and quantitative analysis, and enhances the robustness and accuracy of the entire online detection system in complex mining environments.

[0028] Based on the net feature spectrum and mineral matrix type, a dual-branch network of spectral band attention and element category attention is constructed to output a high-dimensional compact feature vector, a feature attention weight map, and preliminary element identification results.

[0029] Specifically, based on the net feature spectrum and mineral matrix type, a dual-branch network of spectral band attention and element category attention is constructed, outputting a high-dimensional compact feature vector, a feature attention weight map, and preliminary element identification results. The specific analysis process is as follows: Obtain the labeled mineral sample spectral dataset; construct the dual-branch network of spectral band attention and element category attention; input the net feature spectrum, standardize the input net feature spectrum, including mean subtraction and variance normalization, and perform smoothing; through... The module learns the weights of each spectral band, mapping the preprocessed net feature spectrum to a 1×6000×16 feature map. Global average pooling compresses the net feature spectrum into a 1×1×16 global feature map. Then, two fully connected layers learn the band weights, with 8 and 16 neurons respectively. The function outputs the attention weights for each band; four convolutional blocks are set, each containing one convolutional layer, one batch normalization (BN) layer, one ReLU activation function, and one max pooling layer. The kernel sizes of the convolutional layers are 3×3, 5×5, 3×3, and 5×5, and the number of output channels are 32, 64, 128, and 256, respectively. The kernel size of the max pooling layer is 2×2, with a stride of 2, extracting the local peak shape and intensity of the spectrum; based on the mineral matrix type, the corresponding element category attention matrix is ​​loaded; after fusing the convolutional features and attention weights, dimensionality reduction is achieved through two fully connected layers with 512 and 128 neurons, respectively, outputting a 128-dimensional feature vector F, denoted as the high-dimensional compact feature vector, and simultaneously outputting the feature attention weight map and preliminary element identification results.

[0030] It should be explained that the aforementioned dual-branch network of spectral band attention and element category attention achieves deep, intelligent, and interpretable feature extraction of net feature spectra, significantly improving the accuracy of element identification and the model's adaptability to complex mineral matrices. The module automatically learns the importance weights of different spectral bands, effectively highlighting key bands related to elemental characteristic peaks and suppressing irrelevant or noisy bands, thereby improving feature representation capabilities. Multi-scale convolutional blocks can capture local peak shapes, intensity variations, and fine structures of the spectrum at different scales, making the model more sensitive to the features of different elements. Based on the mineral matrix type, corresponding element category attention matrices are loaded, enabling the model to automatically adjust the degree of attention to various elements according to the matrix characteristics of different minerals, effectively mitigating the interference caused by matrix effects. The fully connected layer reduces the dimensionality of the fused features into high-dimensional compact feature vectors, which not only preserves key information but also improves computational efficiency. At the same time, the output attention weight map enhances the interpretability of the model. The dual-branch attention network can extract spectral features more accurately, improve the reliability of preliminary element identification, and lay a high-quality feature foundation for subsequent quantitative analysis.

[0031] Feature-weighted plasma parameter correction fusion analysis was performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain the initial values ​​of quantitative content of each element and the confidence level of the quantitative results.

[0032] Specifically, a feature-weighted plasma parameter correction fusion analysis is performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain the initial values ​​of quantitative content for each element and the confidence level of the quantitative results. The specific analysis process is as follows: A feature-weighted plasma parameter correction fusion analysis is performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to construct a feature-weighted plasma parameter correction quantitative model and calculate the initial values ​​of quantitative content for each element; the similarity between the net feature spectrum and the standard spectral template of the same mineral in the training set is calculated using cosine similarity and recorded as the confidence similarity; the stability of the key plasma parameters is calculated based on the weighted average of the relative standard deviations of the key plasma parameters; the confidence similarity and the stability of the key plasma parameters are weighted and summed to obtain the confidence factor of the quantitative results; the quantitative result confidence factor-quantitative result confidence level mapping table stored in the database is obtained, and the matching quantitative result confidence level is determined based on the current quantitative result confidence factor; the initial values ​​of quantitative content for each element and the confidence level of the quantitative results are output.

[0033] Furthermore, a feature-weighted plasma parameter-corrected quantitative model was constructed to calculate initial values ​​for elemental quantitative content. The specific analysis process is as follows: ; in, This provides initial values ​​for the quantitative content of elements. The feature-weighted regression coefficient matrix stored in the database. It is a high-dimensional compact feature vector. The bias vector is stored in the database. This is the difference between the current plasma excitation temperature and the reference plasma excitation temperature stored in the database. This is the difference between the current plasma electron density and the baseline plasma electron density stored in the database. The plasma excitation temperature correction coefficient is stored in the database. The plasma electron density correction coefficients are stored in the database; Output the initial values ​​of the quantitative content of the elements.

[0034] In one specific embodiment, by fusing high-dimensional compact feature vectors, feature attention weight maps, preliminary element identification results, and key plasma parameters through feature weighting and plasma parameter correction, the accuracy, stability, and reliability of elemental quantitative analysis are significantly improved. This approach not only utilizes advanced spectral features extracted by neural networks but also combines key band and element category information revealed by the attention weight map, enabling the quantitative model to focus on the most discriminative spectral features and improving its adaptability to complex mineral matrices. By introducing correction terms for plasma excitation temperature and electron density, it effectively compensates for plasma instability caused by laser energy fluctuations, changes in mineral surface state, and environmental interference, significantly reducing matrix effects and measurement biases, making the quantitative results more robust. By evaluating the degree of spectral matching through cosine similarity and calculating the confidence factor of the quantitative results in conjunction with plasma parameter stability, the quantitative accuracy can be evaluated in real time, providing a basis for subsequent correction. This fusion analysis method achieves synergistic optimization of spectral features, attention mechanisms, and plasma physical parameters, not only improving the accuracy of initial elemental quantitative values ​​but also providing reliable confidence assessments, offering high-quality, traceable foundational data for the final full-element detection correction.

[0035] By combining the confidence level of quantitative results, high-dimensional compact feature vectors, and net feature spectra, we perform full-element detection and correction for minerals, and output the final element content and mineral type identification results after correction.

[0036] Specifically, combining quantitative result confidence, high-dimensional compact feature vectors, and net feature spectra, we perform full-element detection correction for minerals, and output the corrected final element content and mineral type identification results. The specific analysis process is as follows: Construct a mineral type fingerprint database, which includes spectral fingerprints and elemental proportion fingerprints, and perform automatic mineral type identification and mixed ore ratio calculation. A dual verification strategy of fingerprint matching + feature vector classification is adopted: The net feature spectrum is correlated with the spectral fingerprints in the fingerprint database. If the Pearson correlation coefficient is not less than 0.85, it is considered a successful match. At the same time, the high-dimensional compact feature vector is input into the trained SVM mineral type classifier to obtain the classification result. If the two are consistent, the mineral type name is output. If the net feature spectrum matches multiple mineral types represented by spectral fingerprints in the mineral type fingerprint database, the mixed ore ratio is calculated based on the elemental proportion fingerprint. The corrected final element content and mineral type identification results are output.

[0037] The specific analysis process for constructing a mineral fingerprint database is as follows: a standardized fingerprint database is established, and spectral fingerprints and elemental proportion fingerprints of minerals detected over the past five years are collected and recorded as the mineral fingerprint database. Among them, the spectral fingerprint is the characteristic spectral template of the minerals detected over the past five years, containing the characteristic peak positions and intensity ratios of key elements, and the elemental proportion fingerprint is the content ratio of typical elements of the minerals detected over the past five years.

[0038] In one specific embodiment, by combining the confidence level of quantitative results, high-dimensional compact feature vectors, and net feature spectra for full-element detection and correction, and introducing a mineral fingerprint database to achieve automatic mineral identification and mixed ore ratio calculation, the accuracy, stability, and reliability of the final elemental content and mineral identification are significantly improved. The confidence level of quantitative results is used to judge the reliability of the previous quantification. Further optimization using high-dimensional compact feature vectors and net feature spectra can effectively correct deviations caused by matrix effects, environmental interference, or plasma fluctuations, making the final elemental content more accurate. By constructing a mineral fingerprint database containing spectral fingerprints and elemental ratio fingerprints, and employing a dual verification strategy of fingerprint matching and feature vector classification, not only is the accuracy of mineral identification improved, but the mixed ore ratio can also be automatically calculated when multiple minerals are matched, enhancing the adaptability to complex ores. The fingerprint database, built based on years of historical data, has good representativeness and robustness, and can continuously optimize the identification and correction effects, achieving accurate correction from initial quantitative values ​​to final results, while simultaneously completing reliable mineral identification. This provides a high-precision, high-reliability full-element analysis solution for online mining detection.

[0039] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for intelligent online detection and analysis of all elements in the mining industry based on laser-induced breakdown spectroscopy, characterized in that, Includes the following steps: Collect online ore flow, initial laser parameters, real-time environmental parameters and ore pretreatment information in the mining industry, perform pulse-code laser-induced plasma optimization, and output plasma time-resolved spectrum and key plasma parameters; Adaptive spectral acquisition is performed based on plasma time-resolved spectroscopy and key plasma parameters, and background suppression is applied to the mining scene to obtain the net feature spectrum after background removal. Based on the net feature spectrum and mineral matrix type, a dual-branch network of spectral band attention and element category attention is constructed to output a high-dimensional compact feature vector, feature attention weight map and preliminary element identification results. A feature-weighted plasma parameter correction fusion analysis was performed on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain the initial values ​​of quantitative content of each element and the confidence level of the quantitative results. By combining the confidence level of quantitative results, high-dimensional compact feature vectors, and net feature spectra, we perform full-element detection and correction for minerals, and output the final element content and mineral type identification results after correction.

2. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 1, characterized in that: The specific analysis process for collecting online ore flow data, initial laser parameters, real-time environmental parameters, and ore pretreatment information in the mining industry is as follows: Collect online ore flow data, initial laser parameters, real-time environmental parameters, and ore pretreatment information from the mining industry. The online mineral flow in the mining industry specifically includes the belt conveyor speed, particle size, and moisture content of the minerals. The initial laser parameters specifically include the fundamental frequency, pulse energy range, pulse width, and repetition frequency; Real-time environmental parameters specifically include real-time ambient temperature (T), real-time ambient humidity, and real-time ambient dust concentration. The specific information on mineral pretreatment includes the initial screening particle size and the surface roughness of the mineral.

3. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 2, characterized in that: The process of optimizing pulse-coded laser-induced plasma, outputting plasma time-resolved spectra and key plasma parameters, is as follows: Using a genetic algorithm, online ore flow data, initial laser parameters, real-time environmental parameters, and ore preprocessing information are transformed into a specific three-pulse coding sequence with the goals of maximizing the signal-to-noise ratio and optimizing the stability of the laser-induced plasma excitation temperature. The resulting optimized coding sequence is then obtained. The surface of the ore is captured in real time by a high-speed linear array camera, and the height of the undulation of the ore surface is calculated by combining it with a laser ranging module. At the same time, the offset of the ore flow on the conveyor belt is monitored by visual positioning. The piezoelectric ceramic drives the dynamic lens group to determine the lens position corresponding to the surface undulation height stored in the database, which ensures that the laser focus always falls on the surface of the ore, and then adjusts the lens position. According to the optimized coding sequence, the laser is controlled to emit three pulses: the preheating pulse bombards the surface of the ore, the main excitation pulse core breaks through the ore to form high-temperature plasma, and the energy replenishment pulse replenishes energy in the early stage of plasma expansion to suppress rapid cooling. The spectrometer synchronously acquires the continuous radiation spectrum of the plasma, and selects two spectral lines of iron at 259.94 nm and 260.07 nm. The excitation temperature of the laser-induced plasma is calculated by the Boltzmann diagram method, with a target range of 8000-15000 K. Measuring the Balmer series of hydrogen atoms The Stark broadening of the line is used to calculate the laser-induced plasma electron density in combination with the Stark broadening factor. Plasma excitation temperature and plasma electron density are denoted as key plasma parameters. Output plasma time-resolved spectrum and key plasma parameters.

4. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 1, characterized in that: The adaptive spectral acquisition based on plasma time-resolved spectroscopy and key plasma parameters, combined with background suppression of the mining scene, yields a net feature spectrum after background removal. The specific analysis process is as follows: Obtain the mineral matrix type; The optimal spectral acquisition window is determined based on the plasma excitation temperature; The raw spectral intensity is obtained based on plasma time-resolved spectroscopy, and the instrument noise spectrum is acquired. A dynamic gradient background suppression formula with adaptive weights is constructed to suppress the background of mining scenes and obtain the net feature spectrum after background removal.

5. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 4, characterized in that: The constructed dynamic gradient background suppression formula with adaptive weights is used to suppress the background in the mining scene, obtaining the net feature spectrum after background removal. The specific analysis process is as follows: A dynamic gradient background suppression formula with adaptive weights is constructed, and the specific calculation formula is as follows: ； in, Net characteristic spectrum, The original spectral intensity, The background spectrum is based on nonnegative matrix factorization. The first gradient of the background spectrum. For the instrument noise spectrum, As the matrix background suppression weight, As gradient background suppression weights, For noise suppression weights; ； ； ； in, The plasma excitation temperature, This represents the real-time ambient dust concentration. The baseline dust concentration is stored in the database. Real-time ambient humidity; Output the net feature spectrum after background removal.

6. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 1, characterized in that: Based on the net feature spectrum and mineral matrix type, a dual-branch network of spectral band attention and element category attention is constructed, which outputs a high-dimensional compact feature vector, a feature attention weight map, and preliminary element identification results. The specific analysis process is as follows: Obtain the labeled mineral sample spectral dataset; Construct a dual-branch network for spectral band attention and element category attention: Input the net feature spectrum, standardize the input net feature spectrum, including mean subtraction, variance normalization, and smoothing; The weights of each spectral band are learned through the squeeze-excitation module, the preprocessed net feature spectrum is mapped to a feature map, the global features of the net feature spectrum are compressed through global average pooling, the band weights are learned through a fully connected layer, and the attention weights of each band are output through the sigmoid function. Four convolutional blocks are set up, each containing one convolutional layer, one batch normalization (BN) layer, one ReLU activation function, and one max pooling layer. The kernel sizes of the convolutional layers are 3×3, 5×5, 3×3, and 5×5, and the number of output channels are 32, 64, 128, and 256, respectively. The kernel size of the max pooling layer is 2×2, with a stride of 2. The local peak shape and intensity of the spectrum are extracted. Based on the mineral matrix type, load the corresponding element category attention matrix; After fusing convolutional features with attention weights, the dimensionality is reduced through a fully connected layer, and the output feature vector F is denoted as a high-dimensional compact feature vector. At the same time, the feature attention weight map and the preliminary element recognition results are also output.

7. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 1, characterized in that: The analysis involves a feature-weighted, plasma parameter-corrected fusion analysis of the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to obtain initial values ​​of quantitative content for each element and the confidence level of the quantitative results. The specific analysis process is as follows: We conducted a feature-weighted plasma parameter correction fusion analysis on the high-dimensional compact feature vector, feature attention weight map, preliminary element identification results, and key plasma parameters to construct a feature-weighted plasma parameter correction quantitative model and calculate the initial values ​​of element quantitative content. The similarity between the net feature spectrum and the standard spectral template of the same mineral in the training set is calculated using cosine similarity and denoted as confidence similarity. The stability of key plasma parameters is calculated based on the weighted average of the relative standard deviations of the key plasma parameters. The confidence factor of the quantitative result is obtained by weighted summation of confidence similarity and stability of key plasma parameters. Retrieve the quantitative result confidence factor-quantitative result confidence level mapping table stored in the database, and determine the matching quantitative result confidence level based on the current quantitative result confidence factor; Output the initial values ​​of the quantitative content of each element and the confidence level of the quantitative results.

8. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 7, characterized in that: The specific analysis process for constructing the feature-weighted plasma parameter-corrected quantitative model and calculating the initial values ​​of elemental quantitative content is as follows: ； in, This provides initial values ​​for the quantitative content of elements. The feature-weighted regression coefficient matrix stored in the database. It is a high-dimensional compact feature vector. The bias vector is stored in the database. This is the difference between the current plasma excitation temperature and the reference plasma excitation temperature stored in the database. This is the difference between the current plasma electron density and the baseline plasma electron density stored in the database. The plasma excitation temperature correction coefficient is stored in the database. The plasma electron density correction coefficients are stored in the database; Output the initial values ​​of the quantitative content of the elements.

9. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 1, characterized in that: The method combines quantitative result confidence, high-dimensional compact feature vector, and net feature spectrum to perform full-element detection and correction for minerals, and outputs the final elemental content and mineral type identification results after correction. The specific analysis process is as follows: Construct a mineral fingerprint database, specifically including spectral fingerprints and elemental proportion fingerprints, to automatically identify minerals and calculate blending ratios. A dual verification strategy of fingerprint matching and feature vector classification is adopted: The net feature spectrum is correlated with the spectral fingerprints in the fingerprint database. If the Pearson correlation coefficient is not less than 0.85, the match is considered successful. At the same time, the high-dimensional compact feature vector is input into the trained SVM mineral classifier to obtain the classification result. If the two are consistent, the mineral name is output. If the net feature spectrum matches multiple minerals represented by spectral fingerprints in the mineral fingerprint database, the blending ratio is calculated based on the elemental ratio fingerprint. Output the corrected final elemental content and mineral type identification results.

10. The intelligent online detection and analysis method for all elements in mining based on laser-induced breakdown spectroscopy according to claim 9, characterized in that: The specific analysis process for constructing the mineral fingerprint database is as follows: Establish a standardized fingerprint database, collect the spectral fingerprints and elemental ratio fingerprints of mineral types detected over the past five years, and record them as a mineral type fingerprint database; Among them, the spectral fingerprint is the characteristic spectral template of the mineral type detected over the past five years, containing the characteristic peak positions and intensity ratios of key elements, and the elemental proportion fingerprint is the typical element content ratio of the mineral type detected over the past five years.

Images