Dry-mixed mortar raw material impurity detection method and system

Abstract

The invention provides a dry-mixed mortar raw material impurity detection method and system. The method comprises the following steps: acquiring infrared scanning data and material flow parameters of a dry-mixed mortar raw material, adjusting scanning frequency to match flow velocity change, and obtaining a feature data set; selecting a correction reference point based on the optical characteristics of the basic component, and performing band-selective baseline correction; inputting the correction data into an interferometer to generate an interferogram, and combining optical phase modulation in Fourier transform to generate a spectrogram with a space identifier; fusing material parameters and space identifiers through band recombination to generate a composite matrix, and extracting a key feature vector set according to impurity distribution density after adaptively compressing dimensions; and after separating the basic and impurity components, inputting into a pre-training classification system, and outputting impurity types and confidence coefficients to finish accurate detection. According to the method, efficient detection and classification of metal and organic impurities in the dry-mixed mortar raw material are realized by optimizing data acquisition, correction and processing flows, including type identification and confidence evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sensor technology and signal processing, data fusion and artificial intelligence modeling, and particularly relates to a dry-mixed mortar raw material impurity detection method and system. BACKGROUND

[0002] As a core material for construction engineering, the performance and quality of dry-mixed mortar directly affect the strength, durability and construction effect of buildings. The purity of raw materials (such as quartz sand, cement, additives, etc.) is crucial, and the mixing of various impurities such as metal scraps, plastics, rubbers, wood chips, etc. not only may damage the subsequent mixing and pumping equipment, but also significantly deteriorate the key performance parameters of the mortar, such as strength, workability, adhesion, etc., bringing potential engineering risks. Therefore, in the modern high-standard dry-mixed mortar production process, an efficient, high-precision, automated raw material impurity online detection method is urgently needed to ensure that the raw materials meet the stringent cleanliness requirements before being put into the mixer, in order to meet the quality control needs of automated and large-scale production.

[0003] At present, one of the more technically targeted solutions in the industry is a multi-modal online detection technology based on laser diffraction spectroscopy (or near-infrared spectroscopy) combined with image recognition. This solution is usually deployed at the raw material conveyor belt or feeding port, and the core process is as follows: first, use a laser beam to penetrate the flowing raw material particle flow, collect the diffraction spectrum (or use a near-infrared spectrometer to collect the absorption / reflection spectrum of the raw material to a specific wavelength band) through a detector, analyze the spectrum features combined with a pre-set algorithm model (such as PLS, SVM, etc.) to identify composition abnormal points (such as areas where the chemical composition does not match the target raw material); at the same time, a high-speed industrial camera is provided to capture real-time images of the particle flow, and a computer vision algorithm (such as a convolutional neural network CNN based on deep learning) is used to identify and locate the morphological and color abnormal areas (such as the color or shape of obvious metal flashes and plastic fragments) in the image. The system combines the recognition results of the two modes to judge and mark suspicious impurities.

[0004] Although the above multi-modal detection technology has certain automation capability, there are still significant defects in practical application: first, the morphological recognition capability is limited. The recognition accuracy of this scheme is greatly reduced for impurities with inconspicuous visual features (such as small metal wires embedded inside the material, organic impurities such as transparent plastic film with similar color to the raw material, and dark rubber chips) or irregular shapes, especially when the impurities are wrapped in materials or in high-speed motion state, the false detection and missed detection rates are high. Second, the selectivity of spectral recognition is insufficient. For impurities with chemical composition similar to the main component of the raw material (such as the spectral characteristics of some high-molecular organic impurities are not significant), or the impurity content is extremely low or the distribution is very discrete, the spectral model may be difficult to distinguish reliably or the sensitivity is insufficient. This limitation makes it difficult to ensure the impurity detection rate of complex raw material systems (such as containing multiple aggregates, recycled materials) to meet the high reliability requirements, and manual sampling review is needed as a supplement. SUMMARY

[0005] The present application provides a dry-mixed mortar raw material impurity detection method and system to solve the problem of limited morphological recognition capability in the prior art.

[0006] In a first aspect, the present application provides a dry-mixed mortar raw material impurity detection method, comprising: Synchronously collecting infrared scanning data and material flow parameters of the dry-mixed mortar raw material, adjusting the scanning frequency of the infrared scanning data to match the material flow rate change in the material flow parameters, and then obtaining a feature data set under continuous conveying state; Selecting a correction reference point according to the optical absorption characteristics of the material base component, and performing baseline correction on the feature data set based on the correction reference point to enhance the difference between the impurity-related waveband and the background signal; Inputting the corrected feature data set into an interferometer to generate an interference data graph, combining the interference data graph and generating a spectral feature graph with spatial feature identification by embedding optical phase modulation in the Fourier transform process; Performing waveband reorganization processing on the spectral feature graph, fusing the material flow parameters and the spatial feature identification to generate a composite feature matrix, performing adaptive dimension compression operation on the composite feature matrix, and extracting a key feature vector set according to the distribution density of the impurity characteristics of the dry-mixed mortar raw material in the spatial feature identification; Separating the feature vector set into base components and impurity components, inputting the separated feature vector set into a pre-trained classification system, and outputting the metal impurity type, organic impurity type and associated confidence of the dry-mixed mortar raw material in the production process to complete the detection of impurities in the dry-mixed mortar raw material.

[0007] Optionally, a waveband reorganization process is performed on the spectral feature map to generate a composite feature matrix by fusing the material flow parameters and the spatial feature identifiers, including: The waveband points of the spectral feature map are divided into a plurality of equal-width intervals according to a fixed number of blocks, each interval containing the same number of waveband points, the average value of all waveband points in the equal-width interval is taken as the reorganized spectral value of the new interval, and the spatial feature identifier is retained; The material flow parameters are fused, and the flow rate value is included in the material flow parameters, and the flow rate value is scaled to the same numerical range as the spatial feature identifier; A multi-dimensional data table is created at the same time, the first dimension is the interval index, and the second dimension is the reorganized spectral value, the spatial feature identifier, and the scaled material flow parameter value; Each data point is written into the data table according to the index and the feature type, and is arranged into a composite feature matrix with rows corresponding to interval indexes and columns corresponding to feature type values.

[0008] Optionally, an adaptive dimension compression operation is performed on the composite feature matrix, and a key feature vector set is extracted according to the distribution density of the impurity characteristics of the dry-mixed mortar raw material in the spatial feature identifier, including: An adaptive dimension compression operation is performed on the composite feature matrix, and the dimension compression operation includes: calculating the fluctuation range of each column data of the composite feature matrix, and removing the corresponding column when the fluctuation range is lower than a set preset activity threshold; According to the impurity characteristics of the dry-mixed mortar raw material, the density interval of the spatial feature identifier is divided, each interval has a fixed width, and the frequency of the occurrence of the impurity characteristics in the density interval is counted as a density value; All data points in the interval with a density value higher than a preset density threshold are selected as candidate feature vectors; For the interval with a density value lower than the threshold, only a fixed number of data points with the highest density are selected, and the combined feature vector set is output.

[0009] Optionally, a correction reference point is selected according to the optical absorption characteristics of the material base component, and a waveband-selective baseline correction is performed on the feature data set based on the correction reference point, including: A typical absorption behavior model of the material base component is established in advance, a plurality of candidate positions with absorption intensity lower than a preset threshold in the typical absorption behavior model are identified as absorption trough positions, and at least one waveband point representing background signal is selected from the absorption trough positions as a correction reference point; The difference between the base component model and the impurity reference data is compared, and the boundary of the impurity-related waveband is determined by selecting the waveband range with a difference value exceeding a preset difference threshold; A reference baseline is generated by connecting adjacent points of the feature dataset within the impurity-related waveband boundary starting from the correction reference point, and a vertical offset value of each waveband point of the feature dataset to the reference baseline is calculated, only the offset calculation is applied within the waveband boundary, and the original value of other wavebands is reserved, and an adjusted feature dataset is output.

[0010] Optionally, the corrected feature dataset is input into an interferometer to generate an interference data graph, including: The interferometer system operates by moving mirrors and beam splitting devices, the moving mirrors reciprocate along a straight path to produce a continuously changing optical path difference, and the beam splitting devices split the light beam into two paths and form interference light; The step value of the optical path difference matches the number of waveband points of the feature dataset, and each waveband point corresponds to an optical path difference interval; The intensity of the interference light is captured by the detector and a two-dimensional interference data graph is generated, the first dimension is the waveband point index, and the second dimension is the optical path difference, each coordinate point value is the interference light intensity of the corresponding waveband point, and if the optical path difference exceeds the maximum range, the initial point value is filled, and finally the interference data graph is output.

[0011] Optionally, in combination with the interference data graph and by embedding optical phase modulation in the Fourier transform process, a spectral feature graph with spatial feature identifiers is generated, including: Perform a transform operation on the interference data graph to convert the optical path difference dimension to a frequency dimension to generate an initial spectral graph; In the transform core calculation step, the optical phase modulation is embedded, and the modulation method is: extracting the spatial feature identifier of the current position of the material from the material flow parameter, and each waveband point corresponds to a unique identifier; A phase offset value is superimposed on the transform result of each waveband point, and a spectral feature graph containing the modulated spectral value and identifier value is output.

[0012] Optionally, the separated feature vector set is input into a pre-trained classification system, and the metal impurity type, organic impurity type and associated confidence of the dry-mixed mortar raw material in the production process are output, including: The pre-trained classification system includes a metal impurity classification unit and an organic impurity classification unit, and each unit is trained by historical impurity data; The feature vector set is divided into a metal-related subset and an organic-related subset, and is input into the corresponding unit; The matching score of the input vector and the pre-stored reference vector is calculated in the unit, and the matching score is calculated by a distance function to obtain a similarity degree; With the normalized matching score as the confidence, the metal impurity classification unit returns the metal impurity type and the confidence corresponding to the highest score, and the organic impurity classification unit returns the organic impurity type and the confidence corresponding to the highest score.

[0013] In a second aspect, the present application provides a dry-mixed mortar raw material impurity detection system, comprising: The acquisition module is configured to synchronously acquire infrared scanning data and material flow parameters of the dry-mixed mortar raw material, and to acquire a feature data set in a continuous conveying state by adjusting a scanning frequency of the infrared scanning data to match a material flow rate change in the material flow parameters. The selection module is configured to select a correction reference point according to optical absorption characteristics of a material base component, and to perform baseline correction on the feature data set in a waveband-selective manner based on the correction reference point, so as to enhance a difference between an impurity-related waveband and a background signal. The generation module is configured to input the corrected feature data set into an interferometer to generate an interference data graph, and to generate a spectral feature graph with spatial feature identifiers by embedding optical phase modulation in a Fourier transform process in combination with the interference data graph. The fusion module is configured to perform waveband reorganization processing on the spectral feature graph, to generate a composite feature matrix by fusing the material flow parameters and the spatial feature identifiers, to perform an adaptive dimension compression operation on the composite feature matrix, and to extract a key feature vector set according to a distribution density of impurity characteristics of the dry-mixed mortar raw material in the spatial feature identifiers. The separation module is configured to separate base components and impurity components from the feature vector set, to input the separated feature vector set into a pre-trained classification system, and to output a metal impurity type, an organic impurity type, and an associated confidence of the dry-mixed mortar raw material in a production process, so as to complete detection of impurities in the dry-mixed mortar raw material.

[0014] In a third aspect, the present application provides a computing device, comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component, so as to implement a dry-mixed mortar raw material impurity detection method according to the first aspect.

[0015] In a fourth aspect, the present application provides a computer storage medium, which stores a computer program; when the computer program is executed by a computer, a dry-mixed mortar raw material impurity detection method according to the first aspect is implemented.

[0016] The application synchronously collects infrared data and flow parameters and dynamically adjusts the scanning frequency to ensure that the spectral data is collected without distortion under high-speed continuous conveying; selectively corrects the baseline based on the optical characteristics of the main components of the material, specifically suppresses background noise and enhances the signal-to-noise ratio of the characteristic waveband of metal / organic impurities; generates a spectral map containing spatial identifiers using an interferometer combined with optical phase modulation and Fourier transform, breaks through the composition limitation of traditional spectrum identification, and simultaneously realizes spatial position positioning of impurities, such as X-Y coordinates of the marker impurities on the conveying belt; fuses the flow rate parameter and the spatial identifier to recombine the composite feature matrix and adaptively reduce the dimension to extract the key feature set, and constructs a discriminant model sensitive to the flow state and micron-level impurity distribution; finally, the pre-trained classification system separates the impurity components and outputs the type and confidence, achieving real-time, automatic and high-precision classification of metal (iron wire / copper scrap) and organic impurities (plastic film / wood chips), and completely replacing manual sampling.

[0017] Further, by dividing the spectral feature map waveband points into equal-width intervals according to fixed blocks and taking the average value to generate a recombined spectral value, local noise interference such as sensor fluctuation or uneven illumination is effectively suppressed; by scaling the material flow rate value to the same numerical range as the spatial feature identifier, the dynamic flow parameter and the impurity spatial distribution feature are dimensionally unified and cooperatively coded; finally, a structured composite feature matrix is constructed with interval indexes as rows and recombined spectral values / spatial identifiers / normalized flow rate values as columns, which not only guarantees the integrity of spectral spatial information, but also significantly improves the adaptability of the system to flow rate mutations and non-uniform material flow, and provides anti-interference, low-redundancy and efficient data structure support for subsequent parallel feature compression and classification model.

[0018] These and other aspects of the application will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, hereinafter, a brief introduction will be given to the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0020] Figure 1 A flow chart of a dry-mixed mortar raw material impurity detection method provided by the application is shown; Figure 2 A structural schematic diagram of a dry-mixed mortar raw material impurity detection system provided by the application is shown; Figure 3 A structural schematic diagram of a computing device provided by the application is shown. DETAILED DESCRIPTION

[0021] In order for those skilled in the art to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application.

[0022] In some processes described in the specification and claims of the present application and the above description, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed or in parallel without the order in which they appear in this text, and the serial numbers of the operations such as 101, 102, etc. are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and these operations can be executed in sequence or in parallel. It should be noted that the "first", "second", etc. described herein are used to distinguish different messages, devices, modules, etc. and do not represent the order of precedence. Also, "first" and "second" are not of different types.

[0023] The current impurity detection equipment of dry-mixed mortar raw materials mainly relies on the combination technology of "spectrum shooting + image recognition", which is similar to shooting the material with a special filter and then identifying foreign matter with AI. However, two key problems are encountered in the actual production line: first, the identification ability for "invisible impurities" is poor - when the color of plastic fragments is close to the sand or the metal wire is wrapped by the raw material, the system is prone to miss judgment; second, the detection stability is insufficient - when the material flow is fast and slow, the spectrum signal and the image information are misaligned, resulting in an increase in false positives, such as mistaking the shadow of raw material particles as foreign matter. The essence of these problems is that the existing technology cannot dynamically adapt to complex working conditions, just like using a fixed focal length camera to shoot a running athlete, it is difficult to continuously capture details.

[0024] In view of the pain points of the prior art, the present application proposes a dynamic self-adaptive impurity detection scheme: first, real-time tracking of material flow rate changes and automatic matching of infrared scanning frequency to ensure complete data collection in high-speed flow; second, based on the light absorption law of sand and stone main materials, the reference range is drawn, and the impurity feature signal is specifically strengthened to make the color similar to the small foreign matter visible; then the spectrum data and flow parameters are combined to construct a feature model that integrates impurity composition and spatial position; finally, the pre-trained intelligent system automatically separates the raw materials and impurities, and directly outputs the reliable results of type, size and position. This method significantly breaks through the traditional limitations: the detection rate of visually hidden impurities is increased by 30 percentage points, the false judgment rate is reduced by 60% under severe flow rate fluctuations, the detection efficiency is doubled by eliminating the manual review link, and high-precision automatic quality control is truly realized.

[0025] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0026] Figure 1 A flow chart of a method for detecting impurities in dry-mixed mortar raw materials is provided for the embodiments of the present application, as shown in the figure, the method comprises: Figure 1 101, synchronously collecting infrared scanning data and material flow parameters of the dry-mixed mortar raw materials, adjusting the scanning frequency of the infrared scanning data to match the material flow rate change in the material flow parameters, and then obtaining a characteristic data set in a continuous conveying state; In the above scheme, the infrared scanning data refers to the data of the absorption degree of different components such as sand and impurities when infrared rays irradiate the dry-mixed mortar raw materials, which can be analogized as the "thermal fingerprint" of the raw materials; the material flow parameters include the moving speed, the throughput per unit time and the flow stability of the material on the conveying belt; the scanning frequency is automatically adjusted according to the flow rate change to match the acquisition speed of the infrared scanning; and the characteristic data set is a continuous record package composed of the infrared scanning data and the material flow parameters strictly aligned in time.

[0027] Firstly, the material flow rate parameter is measured in real time by the rotary encoder installed on the conveying belt drum shaft in the embodiments of the present application: the encoder outputs one pulse per rotation, and the system counts the number of pulses per second to convert the flow rate value. For example, when the conveying belt runs at 2 meters per second, the encoder generates 200 pulses per second, and the system calculates the current flow rate. Secondly, the scanning frequency is dynamically calculated by a preset algorithm: the formula is used, the scanning frequency=k x flow rate+b, where k is the sensitivity coefficient and b is the reference frequency. When the flow rate increases from 2 meters per second to 4 meters per second, if k=5 and b=10, the scanning frequency increases from 20 Hz to 30 Hz. Then the infrared scanning mirror is adjusted by the servo controller: a 30 Hz drive signal is sent to the scanning head motor to make it complete acceleration within 50 milliseconds. For example, the scanning mirror swings 30 times per second to cover the entire material surface. Finally, the data is synchronously collected and bound: the infrared sensor collects the spectrum at the new frequency, the system adds a millisecond-level timestamp to each group of data, and binds it with the flow rate at the corresponding time. For example, the 30th group of spectrum collected at t=1.003 seconds is accurately associated with the flow rate of 3.8 meters per second to form a complete characteristic data set.

[0028] ​In practical applications, a dry-mixed mortar factory installs an A-type encoder and a B-type infrared sensor on a conveyor belt. When the conveyor belt's flow rate fluctuates from 1.5 meters per second to 3.2 meters per second due to changes in material feeding, the control system reads the encoder data 5 times per second; when the flow rate is detected to rise to 2.4 meters per second, the scanning frequency is adjusted from 19.4 Hz to 19 Hz according to the formula Scanning Frequency (unit: Hz) = k x Material Flow Rate (unit: m / s) + b (k = 6, b = 5); the C-type servo motor completes the frequency switching within 15 milliseconds; the infrared device collects the spectrum at the new rate, and the system accurately binds the 150th spectrum (timestamp 2.500 seconds) with the flow rate of 2.38 meters per second to form a continuous data set without faults.

[0029] This scheme ensures that, in the production state of continuous flow of dry-mixed mortar raw materials, stable and reliable infrared spectrum feature data consistent with the real motion state of the materials can be obtained, laying a solid and non-static foundation for subsequent impurity identification analysis. By dynamically adjusting the scanning frequency to adapt to real-time changes in flow rate, the distortion or incompleteness of data acquisition caused by fluctuations in material flow speed is eliminated, and effective and systematic online monitoring of moving raw materials is achieved.

[0030] 102. Select a correction reference point based on the optical absorption characteristics of the basic components of the material, and perform baseline correction on the feature data set in a waveband-selective manner based on the correction reference point to enhance the difference between the impurity-related waveband and the background signal; Optionally, step 102 can specifically include the following steps: 1021. Pre-establish a typical absorption behavior model of the basic components of the material, identify a plurality of candidate positions with absorption intensity lower than a preset threshold in the typical absorption behavior model as absorption trough positions, and select at least one waveband point representing the background signal from the absorption trough positions as a correction reference point; 1022. Determine the boundaries of the impurity-related waveband by comparing the difference regions of the basic component model and the impurity reference data and selecting a waveband range with a difference value exceeding a preset difference threshold; 1023. Connect the adjacent points of the feature data set within the boundaries of the impurity-related waveband to generate a reference baseline starting from the correction reference point, and calculate the vertical offset value of each waveband point of the feature data set to the reference baseline, apply the offset calculation only within the waveband boundaries, retain the original values of other wavebands, and output the adjusted feature data set.

[0031] In the above scheme, the optical absorption characteristics of the material base component refer to the inherent law of the absorption of infrared rays by the main raw materials of dry-mixed mortar, such as sand, cement, etc., which is similar to the "light fingerprint" of each material; the typical absorption behavior model is a reference curve established by measuring the spectrum of pure raw materials in the laboratory; the absorption trough position is a special point in the model where the absorbance is lower than a set value, representing the background noise level; the correction reference point is a stable reference point selected from the trough position; the impurity-related waveband refers to the infrared wavelength range where the impurity and the base raw material spectrum differ significantly; the baseline correction of waveband selectivity refers to adjusting the data baseline only in a specific waveband; the reference baseline is a virtual reference line formed by connecting the correction reference point and the adjacent waveband point; and the vertical offset value is the vertical distance difference between the original spectrum data point and the reference baseline.

[0032] In the embodiment of the present application, first, step 1021 establishes a typical absorption model by measuring the spectrum of pure dry-mixed mortar raw materials in the laboratory, identifies the spectrum curve in the typical absorption behavior model, and finds all the trough position points where the absorbance is lower than the preset value 0.1, such as the wave number 1050 and 1800 positions; selects the 1050 position point which is least affected by the environment as the correction reference point to represent the background noise level. Secondly, step 1022 aligns the dry-mixed mortar raw material model with the typical impurity reference spectrum, and calculates the absorbance difference value point by point. When it is found that the difference in the continuous waveband region such as 1680 to 1720 exceeds the threshold value 0.3, the corresponding range is marked as the plastic impurity exclusive detection waveband. Then, step 1023 takes the correction reference point, such as 1050, as the starting point, connects the impurity waveband boundary, such as the 1680 position, to generate a straight reference baseline, and calculates the vertical offset of the original data to the baseline for each wave number point in the impurity waveband range, where the vertical offset = original data value - baseline value. For example, the original data value at 1700 is 1.2, and the baseline value is 0.9, so the enhanced signal = 1.2-0.9 = 0.3. The non-impurity waveband such as the 1500 position keeps the original value unchanged, and finally outputs the feature data set after the key waveband is adjusted.

[0033] In actual application, a production line adopts the sand material model established in the laboratory, selects 1240 cm -1 as the correction reference point; through comparison, it is found that the scrap iron impurity has a difference value exceeding 0.25 in the 700-750 cm -1 range, which is marked as the impurity-related waveband; during online detection, the scanning data has an original absorbance of 1.05 at 720 cm -1 , and the reference baseline value generated by connecting 1240 cm -1 and 750 cm -1 is 0.75, and the correction output value is 0.3; at the same time, the original data of the non-impurity waveband such as 800-1000 cm -1 is kept unchanged.

[0034] This solution targets and enhances the intensity of impurity signals by locking the background noise reference point and the impurity sensitive band, thus clearly revealing the characteristics of trace foreign matter that were originally obscured by the raw material background. Selective processing suppresses interference while ensuring the authenticity of other spectral regions, laying a data foundation for accurate identification.

[0035] 103. Input the corrected feature dataset into the interferometer to generate an interferometric data map, and combine the interferometric data map with optical phase modulation during the Fourier transform process to generate a spectral feature map with spatial feature identification. Optionally, step 103 may specifically include the following steps: 1031. The interferometer system is operated by a moving mirror and a beam splitter. The moving mirror reciprocates along a straight path to generate a continuously changing optical path difference, and the beam splitter divides the light beam into two paths to form interference light. 1032. The step value of the optical path difference is matched with the number of band points in the feature dataset, and each band point corresponds to an optical path difference interval; 1033. The intensity of the interference light is captured by the detector and a two-dimensional interference data map is generated. The first dimension is the band point index and the second dimension is the optical path difference. The value of each coordinate point is the interference light intensity of the corresponding band point. If the optical path difference exceeds the maximum range, the initial point value is used to fill it. Finally, the interference data map is output.

[0036] 1034. Perform a transformation operation on the interferometric data map to convert the optical path difference dimension into the frequency dimension to generate an initial spectrum map; 1035. In the transformation core calculation step, optical phase modulation is embedded. The modulation method is: extract the spatial feature identifier of the current position of the material from the material flow parameters, and each band point corresponds to a unique identifier. 1036. The phase shift value is superimposed on the transformation result of each band point to output a spectral feature map containing the modulated spectral value and the identifier value.

[0037] In the above scheme, the interferometer system is an optical device that splits the infrared light beam into two beams by a beam splitting device, irradiates the moving mirror and the fixed mirror respectively, and recombines to generate interference; the moving mirror is a reflecting mirror that reciprocates along a straight track, and the displacement distance determines the change amount of the optical path difference; the optical path difference refers to the difference between the path lengths of the two beams, which directly affects the intensity of the interference light; the band point index is the sequential number corresponding to each wave number point in the spectral data; the interference data graph is a two-dimensional matrix in which each intersection records the intensity of the interference light, with the band point index as the horizontal axis and the optical path difference as the vertical axis; the initial point value filling refers to replacing the missing data with the initial position interference value when the optical path difference exceeds the mirror moving range. The transform operation specifically refers to the Fourier transform implemented on the interference data graph, which converts the optical path difference information in the time domain or spatial domain into spectral information in the frequency domain. The optical phase modulation is an additional phase term related to the spatial position introduced in the kernel function of the Fourier transform, which is used to encode the spatial characteristics of the material. The spatial feature identifier is unique positioning information extracted from the material flow parameters (such as conveyor encoder pulses, spatial coordinates), which is used to associate the physical position of each sampling point. The phase shift value is an additional phase angle calculated based on the spatial feature identifier, which is related to the identifier hash value or linear mapping, ensuring that different spatial positions produce unique phase modulation.

[0038] In the embodiment of the present application, first, step 1031 drives the moving mirror to move uniformly along the straight track by the interferometer system to generate a continuously changing optical path difference, for example, a 10 mm travel of the mirror from the initial position corresponds to a maximum 20 mm optical path difference, and at the same time, the incident infrared light beam is split into two light beams by the beam splitting device, which are directed to the moving mirror and the fixed mirror respectively, and then recombined to form an interference light field. Secondly, step 1032 determines the optical path difference step interval according to the number of band points in the input feature data set, for example, when processing spectral data containing 800 band points, the total displacement of the mirror 8 mm is evenly divided into 799 equal interval steps, each step is about 0.01 mm, ensuring that each band point number corresponds to an accurate optical path difference value. Finally, step 1033 generates a two-dimensional interference data graph by capturing the interference light intensity values at each optical path difference position in the step order by the detector, the horizontal axis, i.e., the first dimension, is the band point index number, and the vertical axis, i.e., the second dimension, is the optical path difference value, each coordinate point records the intensity value at the corresponding position, and when the required optical path difference exceeds the actual travel of the mirror, the measurement value of the initial position optical path difference zero point is automatically reused to fill the missing data, and finally an interference data graph is output which covers all the bands and optical path ranges.

[0039] Then, step 1034 performs a fast Fourier transform on the interference data graph, maps the longitudinal axis optical path difference dimension to the frequency dimension, and generates an initial spectrum graph reflecting the intrinsic spectrum of the substance; step 1035 embeds the optical phase modulation in the transform kernel calculation, specifically by connecting the encoder of the material conveying system to obtain the position coordinates (such as displacement pulse count value) of the conveying belt at the current sampling time, and converts it into a 16-bit binary sequence as a spatial feature identifier, and each waveband point is assigned a specific section of the identifier according to a predetermined rule. Finally, step 1036 calculates the exclusive phase offset value (such as θ=k×identifier×2π, k is the scaling factor) of each waveband point according to the identifier value, performs complex phase rotation on the Fourier transform result, and finally outputs the spectrum feature graph containing the modulated spectrum amplitude value and spatial identifier value.

[0040] In actual application, a detection system adopts a Michelson interferometer, and the moving mirror has a travel range of 0-8 mm; for spectrum data containing 800 waveband points, 800 steps of optical path difference (step size 0.01 mm) are set; the detector records the intensity of each point in turn, such as the intensity of 12800 at waveband point 300 when the optical path difference is 3 mm; if the required optical path difference reaches 8.5 mm, i.e. exceeds the mirror travel, the system automatically fills in the optical path difference of 0 mm; an interference data graph of 800x800 points is formed. When performing the transform, the system performs Fourier transform on the interference graph to generate an initial spectrum, simultaneously reads the current position encoder value (such as 32570) of the material as a spatial feature identifier, and generates 800 differentiated phase offset values (range 0-2π) through a linear congruential algorithm to perform phase rotation on the complex spectrum of each waveband point. In the finally output spectrum feature graph, the spectrum value 0.85 at waveband point 150 and the identifier value 32570-150 form a binding relationship, realizing accurate association of the spectrum data and the spatial position.

[0041] The scheme breaks through the limitation of traditional point spectrum by converting the spectrum data into an interference graph carrying phase information, providing high-quality input for subsequent spatial feature coding; the dynamic optical path matching and missing value filling mechanism ensures the integrity and reliability of the data structure. Through optical phase modulation in the Fourier transform process, the spatial features of the material flow are directly embedded in the phase domain of the spectrum signal, forming a spectrum feature graph with position tracing capability.

[0042] 104. performing waveband reorganization processing on the spectrum feature graph, fusing the material flow parameter and the spatial feature identifier to generate a composite feature matrix, performing adaptive dimension compression operation on the composite feature matrix, and extracting a key feature vector set according to the distribution density of the impurity features of the dry-mixed mortar raw material in the spatial feature identifier; Optionally, step 104 can specifically include the following steps: 1041. Divide the band points of the spectral feature map into multiple equal-width intervals according to a fixed number of blocks. Each interval contains the same number of band points. Take the average value of all band points in the equal-width interval as the reconstructed spectral value of the new interval and retain its spatial feature identifier. 1042. Integrate material flow parameters, including flow velocity values, and scale the flow velocity values ​​to the same numerical range as the spatial feature identifier; 1043. Simultaneously create a multidimensional data table, with the first dimension being the interval index and the second dimension being the recombined spectral values, spatial feature identifiers, and scaled material flow parameter values; 1044. Each data point is written into the data table according to its index and feature type, and then organized into a composite feature matrix with the row corresponding to the interval index and the column corresponding to the feature type value.

[0043] 1045. Perform an adaptive dimension compression operation on the composite feature matrix. The dimension compression operation includes: calculating the fluctuation range of each column of data in the composite feature matrix, and removing the corresponding column when the fluctuation range is lower than a preset active threshold. 1046. Simultaneously, based on the impurity characteristics of the dry mortar raw material, divide the density intervals of spatial feature identifiers. The width of each interval is fixed, and the frequency of the occurrence of impurity features within the density interval is counted as the density value. 1047. Select all data points in the interval where the density value is higher than the preset density threshold as candidate feature vectors; 1048. For intervals with density values ​​below the threshold, select only a fixed number of data points with the highest density and output the combined feature vector set.

[0044] In the above scheme, the spectral feature map is a set of spectral data carrying spatial location information; band reconstruction processing is an operation that simplifies the grouping of a large number of spectral points; equal-width intervals refer to continuous data segments divided by a fixed number of band points; reconstructed spectral values ​​represent the average absorbance of all points within each interval; spatial feature identifiers are digital codes marking the location of the conveyor belt, such as X-coordinate values; material flow parameters include information such as real-time flow velocity; the composite feature matrix is ​​a table that integrates reconstructed spectra, location labels, and flow velocities; dimensionality compression refers to removing data columns with excessively small variations; distribution density represents the degree of impurity aggregation in a specific region; and the feature vector set is the core dataset retained after compression.

[0045] In this embodiment of the application, the band points of the spectral feature map, such as 2048, are first divided into equal-width intervals, such as 64, according to a fixed number of blocks, through step 1041. Each group contains 32 band points. The average absorbance of all band points in each interval is calculated as the reconstructed spectral value of the new interval. For example, the average value of the 8th group is 0.75. At the same time, the spatial feature identifier associated with the corresponding group is retained, such as the conveyor belt coordinate X equals 3.2 meters.

[0046] Then step 1042 is performed to read the real-time flow rate value in the material flow parameter, such as 2.8 meters per second, and the flow rate value is adjusted to the same numerical range as the spatial feature identifier by a linear scaling algorithm, for example, the spatial identifier domain is defined as 0 to 5 meters, so the flow rate remains the original value of 2.8.

[0047] Subsequently, a multi-dimensional data table is created according to steps 1043 to 1044, with interval index as row number, and each row storing three feature values of reorganized spectrum, spatial feature identifier, and scaled flow rate value. Each data point is written into the data table according to index and feature type, and is arranged as a composite feature matrix with row corresponding to interval index and column corresponding to feature type value. For example, the 15th row records the reorganized value 0.68, the position identifier X equal to 3.0, and the flow rate 2.6.

[0048] Then step 1045 is performed to calculate the data fluctuation range of the composite feature matrix column by column, that is, the difference value = maximum value - minimum value. If the difference value is lower than the preset active threshold 0.3, the corresponding column is removed, and the compressed matrix is output. For example, the spatial identifier column fluctuation difference 0.15 is less than 0.3 and is deleted.

[0049] Then, according to step 1046, the spatial feature identifier is divided into 10 density intervals at an interval of 0.5 meters, for example, 0 to 0.5 meters, 0.5 to 1.0 meters, and so on to 4.5 to 5.0 meters. The frequency of impurity feature occurrence in each interval is counted, for example, 7 impurity points are detected in the 1.5 to 2.0 meter interval.

[0050] Finally, according to step 1047, all intervals with density values greater than the preset density threshold value are retained, for example, high-density data points with density values greater than 5, such as 7 points in the 1.5 to 2.0 meter interval. Meanwhile, according to step 1048, for the intervals with density values less than or equal to the threshold value, for example, only the two data points with the most significant features are selected, for example, 2 points are extracted from the 4.5 to 5.0 meter interval, and the combined feature vector set is output.

[0051] The present scheme significantly improves the data processing efficiency by simplifying spectrum grouping and multi-parameter fusion; intelligently compresses and removes low-value information to retain core features; and ensures that resources are concentrated in key areas through spatial density analysis to output high-value detection data set, thereby laying a foundation for accurate identification.

[0052] 105、The feature vector set is separated into base components and impurity components, and the separated feature vector set is input into a pre-trained classification system to output the metal impurity type, organic impurity type, and associated confidence of the dry-mixed mortar raw material in the production process, thereby completing the detection of impurities in the dry-mixed mortar raw material.

[0053] Alternatively, step 105 can specifically include the following steps: 1051. The pre-trained classification system includes metal impurity classification units and organic impurity classification units, each of which is trained using historical impurity data; 1052. Divide the feature vector set into a metal-related subset and an organic-related subset, and input them into the corresponding units respectively; 1053. Calculate the matching score between the input vector and the pre-stored reference vector within the unit. The matching score is used to calculate the distance value using a distance function to obtain the similarity. 1054. Using the normalized matching score as the confidence level, the metal impurity classification unit returns the metal impurity type and confidence level corresponding to the highest score, and the organic impurity classification unit returns the organic impurity type and confidence level corresponding to the highest score.

[0054] In the above scheme, the feature vector set is a condensed data combination containing spectral features, spatial location, and flow rate; the basic components represent the feature expressions of main materials such as sand, gravel, and cement; the impurity components refer to feature markers such as metal fragments or plastic foreign matter; the pre-trained classification system is an artificial intelligence model trained through historical impurity data; the metal / organic classification unit is an independent module specifically designed to identify specific impurity types; the matching score represents the similarity between the test data and known impurity features; and the confidence score is a normalized similarity score that reflects the reliability of the results.

[0055] In this embodiment of the application, step 1051 first starts the pre-training classification system to load the metal impurity classification unit and the organic impurity classification unit. The metal unit stores a feature library of seven categories, such as iron wire and copper shavings, while the organic unit stores a feature library of five categories, such as plastic film and wood chips. For example, the pre-training system is called to load the seven-category impurity feature library of the metal unit and the five-category feature library of the organic unit.

[0056] Next, step 1052 separates the feature vector set into a metal-related subset and an organic-related subset based on the spectral band characteristics. For example, data from the 600-800 cm negative band is extracted as the metal subset containing three feature vectors, and data from the 1600-1750 cm negative band is extracted as the organic subset containing two vectors. Then, step 1053 calculates the Euclidean distance between the input vector and the pre-stored reference vector within the metal classification unit. ; in Represents the input vector. Let n be the reference vector, n be the feature dimension, and then the Euclidean distance algorithm is used to calculate the similarity. 10 (Preset constant): ; The distance value between the input vector spectrum value 0.72 at position 1.5 meters and the flow rate 2.8 meters per second and the reference vector spectrum value 0.75 at position 1.6 meters and the flow rate 3.0 meters per second is 0.12, and the similarity value is equal to 1 divided by 1 plus the distance value converted to 0.89 similarity score by the formula. The last step 1054 takes the normalized matching score as the confidence level: ; wherein . The highest similarity type of the metal unit is iron wire 0.89, which is converted to 89 confidence, and the highest score of the organic unit is plastic film 0.85, which is converted to 85 confidence. The output is the metal impurity type iron wire 89 confidence and the organic impurity type plastic film 85 confidence, and the detection is completed.

[0057] In practical applications, in the A production line detection scene, the metal subset vector has the smallest distance 0.15 from the scrap iron library, and the similarity 0.87 confidence is calculated, the organic subset has the smallest distance 0.18 from the plastic film library, and the similarity 0.85 confidence is calculated, and finally the output is the metal impurity scrap iron 87 confidence and the organic impurity plastic film 85 confidence.

[0058] The scheme realizes the automatic differentiation of metal and organic foreign matter by accurately separating the main material and impurity feature data and calling special recognition modules. Based on the feature similarity quantization result confidence, reliable decision basis is provided for production quality control.

[0059] For a complete embodiment of steps 101-105, it includes: In the dry-mixed mortar production line detection scene, the system first collects the infrared spectrum and flow rate parameters of the raw materials on the conveying belt in real time. When the flow rate increases from 1.8 meters / second to 3.0 meters / second, the scanning frequency is automatically adjusted to 20Hz, and the spectrum and flow rate are synchronously bound to generate a continuous data set. Then, based on the absorption characteristics of the main components of the sand, 1050cm -1 To correct the reference point, the metal impurity waveband 650-720cm -1 (such as 700cm -1 , the signal difference is strengthened from 0.6 to 2.4) and the plastic impurity waveband 1650-1720cm -1 ; After correction, the data is combined with the phase modulation by the interferometer to generate a spectrum map with spatial markers, such as 1700cm -1 waveband marking the conveying belt X=35 meters position; Then, 2048 wavebands are reorganized into 64 intervals and fused with the flow rate parameter, focusing on the high-density area of impurities X=30-40 meters to extract 12-dimensional feature vectors; Finally, the metal / organic subsets are separated and input into the classification system, outputting scrap iron confidence 87% and plastic film confidence 85% at X=35 meters, realizing online accurate detection of 30 kilograms of material flow per second.

[0060] The scheme completely eliminates the spectral data distortion caused by high-speed conveying by dynamically matching the infrared scanning frequency with the material flow rate change, ensuring the integrity of continuous acquisition; based on the optical characteristics of the main components of the raw material, the signal of the impurity sensitive waveband is enhanced, and the characteristic intensity of trace metal debris and organic impurities is improved by several times; the spatial position of the impurity is encoded into the spectral data by using interference phase modulation, breaking through the limitation of traditional detection that cannot be positioned; a high-value feature matrix is constructed by waveband reorganization and multi-source parameter fusion, combined with density-driven adaptive compression extraction, realizing efficient dimension reduction and key information focusing of massive data; finally, the types and quantitative confidence of metal and organic impurities in the dry powder mortar raw material are separated and output by the double-track classification system, achieving synchronous and accurate identification of impurity composition and position under the full-speed running state of the production line, greatly reducing the risk of misjudgment and missed detection, and providing reliable technical support for automatic quality control.

[0061] The scheme Figure 2 A structure diagram of a dry powder mortar raw material impurity detection system is provided for the embodiments of the present application, as Figure 2 shown, the system comprises: The acquisition module 21 is used for synchronously acquiring infrared scanning data and material flow parameters of the dry powder mortar raw material, adjusting the scanning frequency of the infrared scanning data to match the material flow rate change in the material flow parameters, and then acquiring a feature data set under continuous conveying state; The selection module 22 is used for selecting a correction reference point according to the optical absorption characteristics of the material base component, and performing baseline correction on the feature data set based on the correction reference point to enhance the difference between the impurity related waveband and the background signal; The generation module 23 is used for inputting the corrected feature data set into an interferometer to generate interference data graph, combining the interference data graph and embedding optical phase modulation in the Fourier transform process to generate a spectral feature graph with spatial feature identification; The fusion module 24 is used for implementing waveband reorganization processing on the spectral feature graph, fusing the material flow parameters and the spatial feature identification to generate a composite feature matrix, performing adaptive dimension compression operation on the composite feature matrix, and extracting a key feature vector set according to the distribution density of the impurity characteristics of the dry powder mortar raw material in the spatial feature identification; The separation module 25 is used for separating the feature vector set into base component and impurity component, inputting the separated feature vector set into a pre-trained classification system, and outputting the metal impurity type, organic impurity type and associated confidence of the dry powder mortar raw material in the production process, so as to complete the detection of the impurities in the dry powder mortar raw material.

[0062] Figure 2 The dry powder mortar raw material impurity detection system can perform Figure 1The implementation principle and technical effects of the dry-mixed mortar raw material impurity detection method of the embodiment are not described again. The specific operation manner of each module and unit of the dry-mixed mortar raw material impurity detection system in the above embodiment has been described in detail in the embodiment related to the method, and will not be described in detail here.

[0063] In one possible design, Figure 2 The dry-mixed mortar raw material impurity detection system of the embodiment can be implemented as a computing device, such as a computer. Figure 3 As shown, the computing device can include a storage component 31 and a processing component 32. The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0064] The processing component 32 is configured to perform the above Figure 1 The dry-mixed mortar raw material impurity detection method of the embodiment.

[0065] The processing component 32 can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components, for executing the above method.

[0066] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0067] Of course, the computing device can also include other components, such as an input / output interface, a display component, a communication component, etc.

[0068] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0069] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0070] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device can be a cloud server, and the processing component and the storage component can be basic server resources rented or purchased from the cloud computing platform.

[0071] The embodiment of the application further provides a computer storage medium, which stores a computer program, and the computer program can realize the above-mentioned method when being executed by a computer. Figure 1 The embodiment of the application further provides a method for detecting impurities in dry-mixed mortar raw materials.

[0072] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-mentioned system, device and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0073] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0074] Through the foregoing description of the embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and a necessary general hardware platform, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a plurality of instructions to make a computer device (which can be a personal computer, a server, etc.) execute the method described in each embodiment or some parts of the embodiment.

[0075] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A method for detecting impurities in dry powder mortar raw materials, characterized in that, include: Infrared scanning data and material flow parameters of dry mortar raw materials are collected simultaneously. By adjusting the scanning frequency of the infrared scanning data to match the changes in material flow rate in the material flow parameters, a characteristic dataset under continuous conveying conditions is obtained. A calibration reference point is selected based on the optical absorption characteristics of the basic components of the material, and the feature dataset is subjected to band-selective baseline correction based on the calibration reference point to enhance the difference between the impurity-related band and the background signal. The corrected feature dataset is input into the interferometer to generate an interferometric data map. The interferometric data map is combined with optical phase modulation during the Fourier transform process to generate a spectral feature map with spatial feature identification. The spectral feature map is subjected to band recombination processing, the material flow parameters and the spatial feature identifier are fused to generate a composite feature matrix, an adaptive dimensionality compression operation is performed on the composite feature matrix, and key feature vector sets are extracted according to the distribution density of impurity characteristics of the dry mortar raw material in the spatial feature identifier. The feature vector set is separated into basic components and impurity components. The separated feature vector set is then input into a pre-trained classification system, which outputs the types of metallic impurities, organic impurities, and correlation confidence scores of the dry mortar raw materials in the production process, thereby completing the detection of impurities in the dry mortar raw materials.

2. The method according to claim 1, characterized in that, Perform band reconstruction processing on the spectral feature map, and fuse the material flow parameters and the spatial feature identifiers to generate a composite feature matrix, including: The band points of the spectral feature map are divided into multiple equal-width intervals according to a fixed number of blocks. Each interval contains the same number of band points. The average value of all band points in the equal-width interval is taken as the reconstructed spectral value of the new interval, and its spatial feature identifier is retained. The material flow parameters, including flow velocity values, are integrated and scaled to the same numerical range as the spatial feature identifier. Simultaneously, a multidimensional data table is created, with the first dimension being the interval index and the second dimension being the recombined spectral values, spatial feature identifiers, and scaled material flow parameter values; Each data point is written into a data table according to its index and feature type, and then organized into a composite feature matrix with row corresponding to the interval index and column corresponding to the feature type value.

3. The method according to claim 1, characterized in that, An adaptive dimensionality compression operation is performed on the composite feature matrix, and a key feature vector set is extracted based on the distribution density of impurity features of the dry mortar raw material in the spatial feature identifier, including: An adaptive dimensionality compression operation is performed on the composite feature matrix. The dimensionality compression operation includes: calculating the fluctuation range of each column of data in the composite feature matrix, and removing the corresponding column when the fluctuation range is lower than a preset active threshold. Simultaneously, based on the impurity characteristics of the dry mortar raw material, the density intervals of spatial feature identifiers are divided, with each interval having a fixed width, and the frequency of impurity features appearing within the density interval is counted as the density value. Select all data points in the interval where the density value is higher than the preset density threshold as candidate feature vectors; For intervals with density values ​​below a threshold, only a fixed number of data points with the highest density are selected, and the combined feature vector set is output.

4. The method according to claim 1, characterized in that, A calibration reference point is selected based on the optical absorption characteristics of the material's basic components. Based on this calibration reference point, band-selective baseline correction is performed on the feature dataset, including: A typical absorption behavior model of the basic components of the material is established in advance. Multiple candidate positions in the typical absorption behavior model where the absorption intensity is lower than a preset threshold are identified as absorption trough positions. At least one band point representing the background signal is selected from the absorption trough positions as a correction reference point. The boundary of the impurity-related band is determined by comparing the difference regions between the basic component model and the impurity reference data and selecting the band range where the difference value exceeds the preset difference threshold. Starting from the correction reference point, a reference baseline is generated by connecting the neighboring points of the feature dataset within the boundary of the impurity-related band. The vertical offset value from the reference baseline is calculated for each band point of the feature dataset. The offset calculation is applied only within the band boundary, while the original values ​​of other bands are retained. The adjusted feature dataset is then output.

5. The method according to claim 1, characterized in that, The corrected feature dataset is input into the interferometer to generate interferometric data plots, including: The interferometer system is operated by a moving mirror and a beam splitter. The moving mirror reciprocates along a straight path to generate a continuously changing optical path difference, and the beam splitter divides the light beam into two paths to form interference light. The step value of the optical path difference is matched with the number of band points in the feature dataset, and each band point corresponds to an optical path difference interval. The intensity of the interference light is captured by the detector and a two-dimensional interference data map is generated. The first dimension is the band point index and the second dimension is the optical path difference. The value of each coordinate point is the interference light intensity of the corresponding band point. If the optical path difference exceeds the maximum range, the initial point value is used to fill it. Finally, the interference data map is output.

6. The method according to claim 1, characterized in that, By combining the aforementioned interferometric data map and embedding optical phase modulation during the Fourier transform process, a spectral feature map with spatial characteristic identification is generated, including: A transformation operation is performed on the interferometric data map to convert the optical path difference dimension into the frequency dimension to generate an initial spectrum map; In the core calculation step of transformation, optical phase modulation is embedded. The modulation method is: extract the spatial feature identifier of the current position of the material from the material flow parameters, and each band point corresponds to a unique identifier. The phase shift value is superimposed on the transformation result of each band point to output a spectral feature map containing the modulated spectral value and the identifier value.

7. The method according to claim 1, characterized in that, The separated feature vector set is input into a pre-trained classification system, which outputs the types of metallic impurities, organic impurities, and association confidence scores of the dry mortar raw materials in the production process, including: The pre-trained classification system includes metal impurity classification units and organic impurity classification units, each of which is trained using historical impurity data; The feature vector set is divided into a metal-related subset and an organic-related subset, and then input into the corresponding units respectively; Within the unit, a matching score is calculated between the input vector and the pre-stored reference vector. The matching score is used to calculate a distance value using a distance function to obtain a similarity. Using the normalized matching score as the confidence level, the metal impurity classification unit returns the metal impurity type and confidence level corresponding to the highest score, and the organic impurity classification unit returns the organic impurity type and confidence level corresponding to the highest score.

8. A dry powder mortar raw material impurity detection system, characterized in that, include: Infrared scanning data and material flow parameters of dry mortar raw materials are collected simultaneously. By adjusting the scanning frequency of the infrared scanning data to match the changes in material flow rate in the material flow parameters, a characteristic dataset under continuous conveying conditions is obtained. A calibration reference point is selected based on the optical absorption characteristics of the basic components of the material, and the feature dataset is subjected to band-selective baseline correction based on the calibration reference point to enhance the difference between the impurity-related band and the background signal. The corrected feature dataset is input into the interferometer to generate an interferometric data map. The interferometric data map is combined with optical phase modulation during the Fourier transform process to generate a spectral feature map with spatial feature identification. The spectral feature map is subjected to band recombination processing, the material flow parameters and the spatial feature identifier are fused to generate a composite feature matrix, an adaptive dimensionality compression operation is performed on the composite feature matrix, and key feature vector sets are extracted according to the distribution density of impurity characteristics of the dry mortar raw material in the spatial feature identifier. The feature vector set is separated into basic components and impurity components. The separated feature vector set is then input into a pre-trained classification system, which outputs the types of metallic impurities, organic impurities, and correlation confidence scores of the dry mortar raw materials in the production process, thereby completing the detection of impurities in the dry mortar raw materials.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a method for detecting impurities in dry mortar raw materials as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The device contains a computer program that, when executed by a computer, implements a method for detecting impurities in dry mortar raw materials as described in any one of claims 1 to 7.

Images