Near-infrared nondestructive testing system and method based on vector-subspace discrimination

Abstract

The invention relates to a near-infrared spectrum detection technology, discloses a near-infrared nondestructive detection system and method based on vector-subspace discrimination, and solves the problems of low accuracy and instability when a current near-infrared spectrum model is used for detecting a complex component sample. According to the invention, a corresponding sampling tool piece is designed for a solid or solid-liquid mixed sample, so that standard sample preparation with uniform and consistent distribution can be realized; the method comprises the following steps: controlling a near-infrared spectrometer to collect spectral data of a sample through a control terminal, uploading the spectral data to a cloud platform, performing component detection on a sample spectrum by the cloud platform by adopting a sample spectrum model established based on a vector-subspace discrimination method, and returning a detection result to the control terminal; wherein during establishment of the spectrum model, each component in the mixed spectrum is separated through a vector-subspace discrimination method, and modeling is performed independently for each component, so that mutual interference between the same components is reduced, and the accuracy and stability of the model are improved. The method is suitable for nondestructive detection of components of a solid or solid-liquid mixed sample.

Description

Technical Field

[0001] The present invention relates to near-infrared spectroscopy detection technology, and particularly to a near-infrared non-destructive detection system and method based on vector-subspace discrimination. Background Art

[0002] In modern industrial production, the control of product quality is becoming increasingly strict. Traditional detection methods often take a long time, are costly, and sometimes damage the samples. For example, chemical analysis methods require a large amount of reagents, and the samples may be damaged during the detection process and cannot be used again. These methods are not only inefficient but also difficult to meet the requirements of rapid, efficient, and non-destructive detection in modern industrial production.

[0003] In recent years, near-infrared spectroscopy technology, as a rapid and non-destructive detection means, has been widely applied. Near-infrared spectroscopy technology can quickly obtain the component information of samples by measuring the spectral reflection or transmission characteristics of samples in the near-infrared band. This method has the advantages of high speed, non-destructiveness, and strong repeatability, and is applicable to the detection of various samples, including solids, liquids, and powders.

[0004] However, existing near-infrared spectroscopy detection methods may be affected by background interference when dealing with complex samples, resulting in inaccurate detection results. For example, when there are multiple components in a sample, the spectral characteristics of different components may overlap, making it difficult to detect a single component. In addition, traditional near-infrared spectroscopy models are usually based on single spectral data for modeling, and the stability and accuracy of this method are often insufficient when dealing with complex samples. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to propose a near-infrared non-destructive detection system and method based on vector-subspace discrimination to solve the problems of low accuracy and instability existing in current near-infrared spectroscopy models when detecting complex component samples.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows: On the one hand, the present invention provides a near-infrared non-destructive detection system based on vector-subspace discrimination, including: A sampling tooling part for carrying the sample to be detected and compressing and leveling it; A spectral sampling module for collecting spectral data of the sample to be detected; A control terminal communicatively connected to the spectral sampling module, for controlling the spectral sampling module to collect data, remotely communicating with the cloud platform, uploading the spectral data of the sample to be detected collected to the cloud platform, and receiving the detection results returned by the cloud platform for display; A cloud platform is used to perform component detection on the spectral data of a sample to be detected uploaded by a control terminal by using a established sample spectral model, and feed back the detection result to the control terminal; the sample spectral model includes a plurality of spectral sub-models established separately for each component of the sample to be detected based on the vector-subspace discrimination method.

[0007] Further, the sampling tooling includes a compaction hammer, a loading dish and a clamping member; The compaction hammer includes a handle and an end surface connected to the handle, and is used to compact and level the sample to be detected in the loading dish; The loading dish is used to load the sample to be detected, and a detachable quartz window piece is provided at the top thereof for the spectral sampling module to collect the spectral data of the sample to be detected from the quartz window piece; The clamping member is used to clamp and fix the loaded sample to be detected from the bottom of the loading dish.

[0008] Further, the compaction hammer is made of a solid metal structural member, and the diameter of the end surface is smaller than the inner diameter of the loading dish.

[0009] Further, the spectral sampling module uses a near-infrared spectrometer.

[0010] Further, the control terminal uses a mobile terminal and establishes a communication connection with the spectral sampling module in a wired or wireless manner.

[0011] On the other hand, the present invention also provides a near-infrared non-destructive detection method based on vector-subspace discrimination, which is applied to the above system, and the method includes the following steps: S1. Load the sample to be detected into the sampling tooling and compact and level it; S2. Control the spectral sampling module to collect the spectral information of the sample to be detected through the control terminal and upload it to the cloud platform; S3. The cloud platform performs component detection on the spectral data of the sample to be detected uploaded by the control terminal by using the established sample spectral model, and feeds back the detection result to the control terminal; the sample spectral model includes a plurality of spectral sub-models established separately for each component of the sample to be detected based on the vector-subspace discrimination method; S4. The control terminal receives the detection result returned by the cloud platform and displays it.

[0012] Further, in step S3, the method for establishing the sample spectral model includes: a. Place the samples one by one in the sampling tooling, control the spectral sampling module to start, collect the spectral data of each sample, and upload the spectral data to the cloud platform; b. Detect the indicators of each sample based on physical and chemical analysis methods, and establish a corresponding relationship between the indicator detection values of the samples and the spectral data of the corresponding samples, and complete the calibration one by one on the cloud platform; c. Based on the vector-subspace discrimination method, create a sample spectral model for the spectral data calibrated on the cloud platform, and upload the model to the cloud platform for binding with the control terminal for backup.

[0013] Further, in step a, when collecting spectral data for each sample, by rotating the spectral sampling module at different angles, multiple spectral data are collected and the average value is taken as the spectral data of the sample.

[0014] Further, in step c, the method for creating a sample spectral model based on the vector-subspace discrimination method includes: c1. Select one component from all the components to be detected in the sample as the current component; c2. Based on the calibration value distribution interval corresponding to the current component in all samples, uniformly sample m spectral data to form a spectral matrix Y; c3. Based on the calibration values corresponding to the current component corresponding to the m sampled spectral data, construct a concentration matrix C; c4. Based on the formula Y = C×T, obtain the spectral matrix T of the substance per unit concentration. The row vector of the first row of T is the spectral of the substance per unit concentration of the current component, denoted as s; the remaining row vectors of T form a background subspace N without the current component; c5. For the spectral data v of each sample, perform a subtraction operation respectively to extract the spectrum S of the current component contained in the spectral data v of the corresponding sample u ; The subtraction operation includes: according to the preset subtraction step , gradually calculate the angle between the vector d i after each subtraction and the background subspace N, where d i =v - i× ×s, until the angle between the vector after subtraction and the background subspace N is the smallest. Assume that the subtraction times at this time is i = p, then the spectrum S of the current component contained in the spectral data v u =p× ×s; c6. Repeat steps c1 - c5 to extract the spectral data of each component in the sample respectively; c7. For each component, based on the spectral data of the corresponding component, use a machine learning algorithm to build a separate model to obtain a spectral sub-model for each component.

[0015] Further, in step c2, the number of spectral data m sampled uniformly is greater than or equal to the number of components in the sample.

[0016] The beneficial effects of the present invention are as follows: The sampling tooling part designed for solid or solid-liquid mixed samples can achieve standard sample preparation with uniform distribution, improving the efficiency and quality of sample processing. In addition, during the modeling process, through the vector-subspace discrimination method, each component in the mixed spectrum can be effectively separated, and separate models are built for each component, reducing the mutual interference between different components and improving the accuracy and stability of the model. When dealing with complex samples, compared with the traditional modeling method based on single spectral data, the detection accuracy can be greatly improved. Description of the Drawings

[0017] Figure 1 It is a schematic structural diagram of the sampling tooling part in the embodiment of the present invention; Figure 2 It is a flowchart of the near-infrared non-destructive detection method in the embodiment of the present invention; Figure 3 It is a flowchart of the method for creating a spectral model based on the vector-subspace discrimination method. Detailed Embodiments

[0018] The present invention aims to provide a near-infrared non-destructive detection system and method based on vector-subspace discrimination to solve the problems of low accuracy and instability in the detection of complex-component samples by current near-infrared spectral models. Its core idea is: for solid or solid-liquid mixed samples, a corresponding sampling tooling part is designed to achieve standard sample preparation with uniform distribution; the control terminal controls the near-infrared spectrometer to collect the spectral data of the sample to be detected and uploads it to the cloud platform, and the cloud platform uses the sample spectral model established based on the vector-subspace discrimination method to detect the components of the sample spectrum and transmits the detection result back to the control terminal; among them, during the process of establishing the spectral model, through the vector-subspace discrimination method, each component in the mixed spectrum is separated, and separate models are built for each component, thereby reducing the mutual interference between the same components and improving the accuracy and stability of the model.

[0019] The solution of the present invention will be described in detail below with reference to the drawings and embodiments.

[0020] In this embodiment, the component detection of solid-liquid mixed fermented grains samples is taken as an example. First, this embodiment provides a near-infrared non-destructive detection system, which includes a sampling tooling part, a spectral sampling module, a control terminal, and a cloud platform.

[0021] Sampling tooling part: It is used for standard sample preparation of solid-liquid mixed fermented grains samples. See Figure 1, which consists of three parts: a compaction hammer 101, a loading dish 102, and a clamping piece 103. Among them, the compaction hammer 101 is made of a solid metal structural part, and the diameter of the end face is slightly smaller than the inner diameter of the loading dish 102, which is used to compact the fermented grains sample to ensure the uniformity of the sample; the loading dish 102 is used to load the fermented grains sample, and a quartz window piece is designed on its top for the spectral sampling module to perform spectral sampling; the clamping piece 103 is used to clamp and fix the sample from the bottom of the loading dish 102 to ensure the stability of the sample form.

[0022] Spectral sampling module: The sampling window is closely attached to the quartz window piece on the loading dish 102, placed directly above the sampling tooling piece, and collects the near-infrared spectral information of the sample. In an exemplary solution, a near-infrared spectrometer with 6 built-in near-infrared band sensors and indium gallium arsenide spectral detectors can be selected to collect the near-infrared spectral information reflected diffusely by the fermented grains sample, and convert the optical signal into an electrical signal for data informatization processing.

[0023] Control terminal: Communicates with the spectral sampling module, used to control the spectral sampling module to collect data, communicates remotely with the cloud platform, used to upload the spectral information collected by the spectral sampling module to the cloud platform, and display the detection results returned by the cloud platform. In specific implementation, the control terminal can use a mobile terminal, such as a mobile phone, a tablet computer, etc., to establish a communication connection with the spectral sampling module through wired or wireless means to control the spectral sampling module to collect data; through remote communication with the cloud platform, such as Internet communication, to upload the collected data to the cloud platform and receive the detection results returned by the cloud platform for display.

[0024] Cloud platform: Used to perform component detection on the spectral information of the fermented grains sample to be detected uploaded by the control terminal based on the pre-established fermented grains component spectral model, and feedback the detection results to the control terminal. Among them, the fermented grains component spectral model includes four sub-models, that is, independent modeling is obtained for the four components (moisture, acidity, starch, residual sugar) of the fermented grains by the over-vector subspace discrimination method.

[0025] Based on the above detection system, the process of the near-infrared non-destructive detection method for the fermented grains sample provided in this embodiment is shown in Figure 2 , which includes the following implementation processes: S1. Place the fermented grains sample in the tooling piece for sample preparation; In this step, due to the physical properties of the fermented grains in a solid-liquid mixed state, it is necessary to compact and level the fermented grains sample in the tooling piece to ensure the uniform distribution of the sample, and thus ensure the stability of the sampling spectral data.

[0026] S2. Collect the spectral information of the fermented grains sample and upload it to the cloud platform; In this step, the control terminal controls the start of the spectral sampling module, and then collects the spectral information of the wine grains sample to be detected. To improve the stability of the spectral data, the sampling module can be rotated at different angles, and multiple spectral data are collected and averaged as the spectral data of this sample, which is uploaded to the cloud platform through the network.

[0027] S3. The cloud platform uses the established spectral model of wine grains components to detect components and feedback to the control terminal; In this step, based on the spectral information of the wine grains to be detected uploaded by the control terminal, the cloud platform detects the component content based on the pre-established spectral model of wine grains components, and feeds back the detection results to the control terminal.

[0028] S4. The control terminal displays the detection results returned by the cloud platform.

[0029] In this step, after obtaining the detection results, the control terminal can visually display the detection results in the form of a bar chart.

[0030] In the above solution, the spectral model of wine grains components is created based on the vector-subspace discrimination method, and the creation process is as follows: 1. Select samples; In this step, 200 samples of out-cellar wine grains are selected for spectral data collection and modeling. The components to be analyzed are 4 types: moisture, acidity, starch, and residual sugar.

[0031] 2. Place the samples in the sampling tooling for sample preparation; In this step, the samples are placed one by one in the sampling tooling, compacted and fastened, and the quartz window is placed facing up.

[0032] 3. Collect the near-infrared spectral data of the wine grains; In this step, the control terminal controls the start of the near-infrared spectral detector to scan the wine grains samples one by one. To improve the stability of the spectral data, the spectral acquisition module can be rotated at different angles, and multiple spectral data are collected and averaged as the spectral data of the current sample.

[0033] 4. Upload the spectral data; In this step, the average spectral data obtained in the previous step is uploaded to the cloud platform for storage, so that the spectral data can be viewed in waveform, preprocessed and modeled on the platform side in the future.

[0034] 5. Calibrate the spectral data of the wine grains; In this step, the moisture content, acidity, starch content, and residual sugar content of the fermented grains samples are detected based on traditional laboratory methods (such as drying method, titration method, enzymatic hydrolysis method, etc.). The test results are uploaded to the cloud platform, and the corresponding relationship between the test values and the spectral data of the fermented grains is created, and one-by-one calibration is completed on the cloud platform to form a complete sample data set.

[0035] 6. Establishment of the spectral model for the components of fermented grains; In this step, modeling is carried out for each component based on the method of vector-subspace discrimination, and the process is as Figure 3 shown: (1) Select moisture as the current component; In this step, the current component is the component to be extracted from the original spectrum.

[0036] (2) Sampling of spectral data; In this step, based on the moisture calibration value, 200 spectral data of 200 samples are arranged in sequence, and 10 spectral data are evenly extracted for standby.

[0037] (3) Construction of the concentration matrix and the spectral matrix; In this step, based on the moisture calibration values of the 10 extracted spectral data, the concentration matrix C is constructed, and the corresponding 10 spectral data values form the spectral matrix Y.

[0038] (4) Obtaining the spectral of unit concentration of moisture and the background subspace; In this step, based on the formula Y = C×T, the spectral matrix T of the substance of unit concentration is obtained. The row vector of the first row of T is the spectral of unit concentration of moisture, denoted as s; the remaining row vectors of T form the background subspace N without moisture.

[0039] (5) Perform subtraction operation on the spectral data of the samples to extract the moisture spectrum in the samples; In this step, for the spectral data v of 200 samples, subtraction operations are respectively performed to extract the moisture spectrum S contained in the spectral data v of the corresponding samples u .

[0040] The subtraction operation includes: according to the preset subtraction step (such as set to 1 / 1000), gradually calculate the angle between the vector d i after each subtraction and the background subspace N, where d i = v - i× ×s, until the angle between the vector after subtraction and the background subspace N is the smallest. Assume that the subtraction times at this time is i = p, then the moisture spectrum S contained in the spectral data v u = p× ×s.

[0041] (6)Select acidity, starch, and residual sugar as the current components in sequence, and extract the spectra of the corresponding components in the sample; In this step, select acidity, starch, and residual sugar as the current components in sequence, and perform the operations similar to those in steps (1)-(5). The spectral data of acidity, starch, and residual sugar in the sample can be obtained respectively.

[0042] (7)Build separate models for moisture, acidity, starch, and residual sugar respectively; In this step, for moisture, acidity, starch, and residual sugar respectively, based on the spectral data of the corresponding components, use machine learning algorithms to build separate models, and finally obtain the moisture spectral sub-model, acidity spectral sub-model, starch spectral sub-model, and residual sugar spectral sub-model.

[0043] 7. Upload the spectral model; In this step, package the moisture spectral sub-model, acidity spectral sub-model, starch spectral sub-model, and residual sugar spectral sub-model into a spectral model of the fermented grains sample, and upload it to the cloud platform after packaging. The cloud platform binds the spectral model of the fermented grains sample to the control terminal. During the application process, the cloud platform can directly call the spectral model according to the information of the control terminal to complete the detection of the spectral information of the fermented grains to be detected, so as to obtain the content result values of different components of the fermented grains.

[0044] The above embodiments are based on the near-infrared spectroscopy detection technology. Using the vector-subspace discrimination method, the collected mixed spectral data is subtracted and separated, and then separate sub-models for each component are established to achieve the non-destructive, portable, and high-accuracy identification effect of the fermented grains sample. Compared with the existing detection solutions, the stability and identification accuracy of this method are comparable to those of large laboratory equipment, and the sample preparation is simple, the system is portable, which can strongly promote the wide application of near-infrared detection equipment in the production site.

[0045] Although the embodiments of the present invention have been described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention, and all are within the protection scope of the present invention.

Claims

1. A near-infrared non-destructive testing system based on vector-subspace discrimination, characterized in that, Including: A sampling tooling part for carrying the sample to be detected and compressing and leveling it; A spectral sampling module for collecting spectral data of the sample to be detected; A control terminal, communicatively connected to the spectral sampling module, for controlling the spectral sampling module to collect data, remotely communicating with a cloud platform, uploading the spectral data of the sample to be detected collected to the cloud platform, and receiving and displaying the detection results returned by the cloud platform; A cloud platform for performing component detection on the spectral data of the sample to be detected uploaded by the control terminal using the established sample spectral model and feeding back the detection results to the control terminal; The sample spectral model includes multiple spectral sub-models established separately for each component of the sample to be detected based on the vector-subspace discrimination method.

2. The near-infrared non-destructive detection system based on vector-subspace discrimination according to claim 1, characterized in that The sampling tooling part includes a compaction hammer, a loading dish and a clamping part; The compaction hammer includes a handle and an end surface connected to the handle, for compressing and leveling the sample to be detected in the loading dish; The loading dish is used for loading the sample to be detected, and a detachable quartz window piece is provided at the top thereof for the spectral sampling module to collect the spectral data of the sample to be detected from the quartz window piece; The clamping part is used for clamping and fixing the loaded sample to be detected from the bottom of the loading dish.

3. The near-infrared non-destructive detection system based on vector-subspace discrimination according to claim 2, characterized in that The compaction hammer is made of a solid metal structural part, and the diameter of the end surface is smaller than the inner diameter of the loading dish.

4. The near-infrared non-destructive detection system based on vector-subspace discrimination according to claim 1, characterized in that The spectral sampling module uses a near-infrared spectrometer.

5. The near-infrared non-destructive detection system based on vector-subspace discrimination according to claim 1, characterized in that The control terminal uses a mobile terminal and establishes a communication connection with the spectral sampling module by wire or wirelessly.

6. A near-infrared non-destructive testing method based on vector-subspace discrimination, applied to the near-infrared non-destructive testing system based on vector-subspace discrimination according to any one of claims 1-5, characterized in that, The method includes the following steps: S1. Load the sample to be detected into the sampling tooling part and compress and level it; S2. Control the spectral sampling module to collect the spectral information of the sample to be detected through the control terminal and upload it to the cloud platform; S3. The cloud platform performs component detection on the spectral data of the sample to be detected uploaded by the control terminal using the established sample spectral model and feeds back the detection results to the control terminal; the sample spectral model includes multiple spectral sub-models established separately for each component of the sample to be detected based on the vector-subspace discrimination method; S4. The control terminal receives the detection results returned by the cloud platform and displays them.

7. The near-infrared non-destructive detection method based on vector-subspace discrimination according to claim 6, characterized in that In step S3, the method for establishing the sample spectral model includes: a. Place the samples one by one in the sampling tooling part, control the spectral sampling module to start, collect spectral data of each sample, and upload the spectral data to the cloud platform; b. Detect the indicators of each sample based on physical and chemical analysis methods, establish a corresponding relationship between the indicator detection values of the samples and the spectral data of the corresponding samples, and complete the calibration one by one on the cloud platform; c. Based on the spectral data calibrated on the cloud platform, create a sample spectral model using the vector-subspace discrimination method, and upload the model to the cloud platform for binding with the control terminal for backup.

8. A near-infrared non-destructive detection method based on vector-subspace discrimination according to claim 7, characterized in that In step a, when collecting spectral data of each sample, by rotating the spectral sampling module at different angles, after collecting multiple spectral data, the average value is taken as the spectral data of the sample.

9. A near-infrared non-destructive detection method based on vector-subspace discrimination according to claim 7, characterized in that In step c, the method for creating a sample spectral model using the vector-subspace discrimination method includes: c1. Select one component from all the components to be detected in the sample as the current component; c2. Uniformly sample m spectral data from the calibration value distribution range corresponding to the current component in all samples to form a spectral matrix Y; c3. Based on the calibration values corresponding to the current component corresponding to the m spectral data sampled, construct a concentration matrix C; c4. Based on the formula Y = C×T, obtain the spectral matrix T of the unit concentration substance. The row vector of the first row of T is the spectral of the unit concentration of the current component, denoted as s; the remaining row vectors of T form a background subspace N without the current component; c5. For the spectral data v of each sample, perform a subtraction operation respectively to extract the spectrum S of the current component contained in the spectral data v of the corresponding sample u ; The deduction operation includes: according to a preset deduction step size , gradually calculate the angle between the vector d after each deduction i and the background subspace N, where i d = v - i × u × s, until the angle between the vector after deduction and the background subspace N is the smallest. Assume that the deduction times i = p at this time, then the spectrum S of the current component contained in the spectral data v = p × c6. Repeat steps c1-c5 to extract the spectral data of each component in the sample respectively; c7. For each component, according to the spectral data of the corresponding component, use a machine learning algorithm to build a separate model to obtain a spectral sub-model for each component.

10. A near-infrared non-destructive detection method based on vector-subspace discrimination according to claim 9, characterized in that In step c2, the number m of the uniformly sampled spectral data is greater than or equal to the number of components in the sample.

Images