A multi-sensing near-infrared spectrum online sorting detection method and system

Abstract

The application discloses a kind of multi-sensing near infrared spectrum online sorting detection method and system, method includes: setting multiple near infrared spectrum detection equipment in sequence on conveying belt;Collect multiple conveying belt background spectrum data;Multiple equipment online and gradually on conveying belt uniform speed acquisition;When the difference of certain spectrum data and conveying belt background spectrum data exceeds threshold value, then determine that detection area appears to be detected object;Each near infrared spectrum detection equipment extracts N spectrum to the same to-be-detected object, after mean processing to N spectrum data, prediction result is output through model;Multiple prediction results make final determination result based on voting mechanism, and according to determination result, send to conveying belt and branch push-out signal instruction, so as to realize the sorting of to-be-detected object.The application proposes multi-sensing space division multiplexing method to improve sampling efficiency, combined with multispectral average, multiple equipment result voting determination method to improve the stability and recognition accuracy of spectrum detection.

Description

Technical Field

[0001] This invention relates to the field of near-infrared spectroscopy online detection technology, and in particular to a multi-sensor near-infrared spectroscopy online sorting and detection method and system. Background Technology

[0002] Near-infrared spectroscopy is characterized by its ease of operation, lack of pollution, and non-destructive nature to sample structure, leading to its widespread application in various fields such as food, chemical, and pharmaceutical industries in recent years. With the development of microelectromechanical systems (MEMS) and 5G technology, near-infrared spectral chips based on MEMS Fabry-Perot cavities have been successfully applied to spectral detection equipment, promoting further miniaturization and cost reduction. Furthermore, due to their lack of moving parts and good shock resistance, near-infrared spectroscopy is rapidly developing towards portable and online applications.

[0003] Currently, the main near-infrared spectroscopy detection equipment on the market includes laboratory near-infrared spectrometers, portable near-infrared spectrometers, and online near-infrared spectrometers installed on production lines. Among them, online spectrometers can monitor batches of materials in real time, and the application market is huge, but the requirements are also very high. These instruments are easily affected by complex environmental factors in production line applications, resulting in unstable detection spectra, low recognition accuracy, and detection speed that is difficult to keep up with the production line's output pace. Therefore, they have not yet been widely used. How to improve the stability, detection speed, and recognition accuracy of the instruments through hardware upgrades and algorithm optimization has become the core issue in solving the large-scale application of online near-infrared spectrometers. Summary of the Invention

[0004] This invention provides a multi-sensor near-infrared spectroscopy online sorting and detection method and system to solve the problems of unstable detection spectrum, slow detection speed and low recognition accuracy of existing online spectrometers.

[0005] The technical solution adopted in this invention is: to provide a multi-sensor near-infrared spectroscopy online sorting and detection method, comprising:

[0006] Multiple near-infrared spectroscopy detection devices are arranged in sequence directly above the conveyor belt;

[0007] The conveyor belt is allowed to idle for a specific time. Each near-infrared spectroscopy detection device detects a specified number of background spectral data of the conveyor belt, uploads them to the cloud platform for storage, and issues a start detection command.

[0008] The object to be inspected is laid flat on the center line of the conveyor belt, and multiple near-infrared spectral detection devices are used to collect spectral data of the conveyor belt online and one by one at a uniform speed.

[0009] When the difference between a certain spectral data collected at a constant speed and the background spectral data of the conveyor belt exceeds a threshold, it is determined that an object to be detected has appeared in the detection area, and thus that spectral data is used as the starting spectral signal.

[0010] After the initial spectral signal, each near-infrared spectral detection device extracts N spectra of the same analyte as valid spectral data. After averaging the N valid spectral data, the data are input into the model for category prediction, and the prediction results are uploaded to the cloud platform for storage.

[0011] Multiple near-infrared spectroscopy detection devices generate multiple prediction results, which are then used to make a final judgment based on a voting mechanism. Based on the judgment result, a pass-through or branch push-out signal is sent to the conveyor belt to achieve the sorting of the items to be inspected.

[0012] Furthermore, each of the near-infrared spectroscopy detection devices integrates multiple MEMS sensors in the same spectral band; the sampling task in a specific spectral band is evenly distributed through spatial multiplexing; after sampling is completed, the complete spectral information of the spectral band is spliced ​​and output through a spectral splicing algorithm.

[0013] Furthermore, the spectral stitching method includes:

[0014] After normalizing the light intensity information of different bands, interpolation is performed at the splicing point to achieve complete splicing of the bands.

[0015] Furthermore, the method for normalizing the light intensity information includes: maximum value normalization, mean value normalization, or peak value normalization.

[0016] Furthermore, the methods for determining whether the difference exceeds the threshold include Mahalanobis distance, Euclidean distance, or Manhattan distance.

[0017] Furthermore, the method for extracting the effective spectral data includes:

[0018] The amount of effective spectral data that the online near-infrared spectrometer can extract from the object during inspection is determined by considering the conveyor belt, the near-infrared spectroscopy equipment, and the state of the object to be inspected. The formula is as follows:

[0019]

[0020] In the formula: N is the number of effective spectral lines, L is the length of the object to be tested in the direction of the conveyor belt, V is the speed of the conveyor belt, and Δt is the sampling interval.

[0021] Furthermore, the method for determining the final result based on the multiple prediction results using a voting mechanism includes:

[0022] When the predicted results are of different categories, the result with the higher vote rate is selected as the final output.

[0023] When the predicted results are of the same category, the result with the higher confidence level is selected as the final output.

[0024] The present invention also provides a multi-sensor near-infrared spectroscopy online sorting and detection system, comprising:

[0025] A conveyor belt, including at least one branch conveyor belt, for transporting items to be inspected;

[0026] Multiple near-infrared spectroscopy detection devices are arranged in sequence directly above the conveyor belt to collect spectral data of the objects to be inspected on the conveyor belt and predict their categories.

[0027] The sorting module is used to push the item out of the branch when multiple near-infrared spectroscopy detection devices detect that the item belongs to a set category.

[0028] The spectral cloud platform is used for storing spectral data and processing results. Based on the prediction results generated by each near-infrared spectral detection device, it makes a final judgment based on a voting mechanism and sends the judgment result to the system control module for sorting instructions.

[0029] The system control module is used to control the power supply of the detection system, the conveyor speed of the conveyor belt, and to control the branch push of the sorting module.

[0030] Furthermore, each of the near-infrared spectroscopy detection devices integrates multiple MEMS sensors in the same spectral band; the sampling task in a specific spectral band is evenly distributed through spatial multiplexing; after sampling is completed, the complete spectral information of the spectral band is spliced ​​and output through a spectral splicing algorithm.

[0031] Furthermore, the near-infrared spectroscopy detection device integrates a spectral stitching algorithm and a pre-built spectral model.

[0032] The beneficial effects of this invention are as follows: Addressing the current issues of unstable spectral information, long detection time, and low recognition accuracy in online near-infrared spectroscopy detection, this invention proposes a multi-sensor spatial multiplexing method to improve sampling efficiency, introduces a spectral initiation signal recognition scheme, and combines multi-spectral averaging and multi-device result voting methods to improve the stability and recognition accuracy of spectral detection. This has played a positive role in promoting near-infrared spectroscopy detection technology and is conducive to driving greater industrial value of near-infrared spectroscopy detection technology in the field of online detection. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating the multi-sensor near-infrared spectroscopy online sorting and detection method disclosed in this invention. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be described in further detail below with reference to the accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0035] Example 1:

[0036] In this embodiment, taking near-infrared spectroscopy online sorting of textiles as an example, the wavelength range is 1750nm~2150nm, the number of sampling points is 51, the number of near-infrared spectroscopy detection devices mounted on the conveyor belt is 3, and the number of sensors in each device is 3, to determine whether the textiles are pure cotton or not.

[0037] See Figure 1 This embodiment discloses a multi-sensor near-infrared spectroscopy online sorting and detection method, including the following steps:

[0038] S1: First, install three near-infrared spectroscopy detection devices arranged in sequence directly above the conveyor belt.

[0039] Specifically, each near-infrared spectroscopy detection device integrates three MEMS sensors in the same spectral band. The sampling task for a specific spectral band is evenly distributed using a spatial division multiplexing method. After sampling, a spectral stitching algorithm is used to stitch together and output the complete spectral information for that band. By evenly distributing the spectral band across each sensor, multiple sensors simultaneously acquire spectral information for the corresponding band, followed by spectral stitching and output. This trades space for time, thereby reducing spectral sampling time and meeting the high sampling time requirements of the production line.

[0040] Specifically, the spectral splicing method includes: normalizing the light intensity information of different bands, and then performing interpolation calculations at the splicing point to achieve complete splicing of the bands. The method for normalizing the light intensity information includes: maximum value normalization, mean value normalization, or peak value normalization.

[0041] S2: Allow the conveyor belt to idle for a specific time. After the near-infrared detection equipment has fully warmed up, each near-infrared spectral detection device detects 10 background spectral data points of the conveyor belt, uploads them to the cloud platform for storage, and issues a start detection command.

[0042] S3: Lay the textile flat on the center line of the conveyor belt. After receiving the start detection command, collect spectral data of the conveyor belt online and one by one through three near-infrared spectral detection devices.

[0043] Traditional methods often employ proximity or vision sensors to detect the object's proximity and then send commands to a spectral sensor to begin spectral acquisition. This not only increases equipment costs but also results in time-consuming signal transmission between different devices, which is detrimental to the high timeliness requirements of online detection. Uniformly acquiring spectral data effectively ensures sufficient sampling time while reducing hardware costs.

[0044] S4: When the difference between a certain spectral data collected at a constant speed and the background spectral data of the conveyor belt exceeds the threshold, it is determined that textiles are present in the detection area, and thus the spectrum is used as the starting spectral signal.

[0045] The methods for determining if the difference exceeds the threshold include Mahalanobis distance, Euclidean distance, or Manhattan distance. When the distance between two spectral data points exceeds the set threshold, it is determined that textile material has entered the spectral recognition area, and this data point is used as the starting signal for valid spectral data extraction. This step utilizes spectral information for starting signal determination, which is low-cost and highly reliable.

[0046] S5: After the initial spectral signal, each near-infrared spectral detection device extracts N spectra of the same analyte as valid spectral data. After averaging the N valid spectral data, the data are input into the model for category prediction, and the prediction results are uploaded to the cloud platform for storage.

[0047] By considering the conveyor belt, near-infrared spectroscopy equipment, and the condition of the textile to be inspected, the amount of effective spectral data that can be extracted by the online near-infrared spectroscopy equipment during the inspection process can be determined. The specific formula is as follows:

[0048]

[0049] In the formula: N is the number of effective spectral lines, L is the length of the textile in the direction of the conveyor belt, V is the speed of the conveyor belt, and Δt is the sampling interval.

[0050] By averaging multiple spectral data points from the same textile fabric, the influence of the surrounding environment on spectral acquisition can be minimized, thus improving the stability of the spectral data.

[0051] S6: After the textiles pass through 3 near-infrared spectroscopy detection devices, 3 sets of pure cotton / non-pure cotton prediction results are generated. These 3 sets of prediction results are uploaded to the cloud platform, and a final judgment result is made based on the voting mechanism. Based on the judgment result, a pass or branch push signal instruction is sent to the conveyor belt to realize the sorting of the items to be inspected. Specifically, pure cotton textiles pass the inspection, while non-pure cotton textiles are pushed to the branch conveyor belt for further processing.

[0052] Specifically, the method for determining the final result based on the multiple prediction results using a voting mechanism includes:

[0053] When the predicted results are of different categories, the result with the higher vote rate is selected as the final output.

[0054] When the predicted results are of the same category, the result with the higher confidence level is selected as the final output.

[0055] The online sorting and detection method proposed in this embodiment utilizes three near-infrared spectroscopy detection devices, introduces a spectral initiation signal recognition scheme, and employs a result determination method based on a voting mechanism using multiple devices. This enables the online detection application of near-infrared spectroscopy in the field of textile sorting, which can greatly promote the industrialization and application of near-infrared online detection technology.

[0056] Example 2:

[0057] In this embodiment, taking near-infrared spectroscopy online sorting of textiles as an example, the wavelength range is 1750nm~2150nm, the number of sampling points is 51, the number of near-infrared spectroscopy detection devices mounted on the conveyor belt is 3, and the number of sensors on each device is 3.

[0058] This embodiment discloses a multi-sensor near-infrared spectroscopy online sorting and detection system, including:

[0059] The conveyor belt has two branches. Pure cotton textiles are inspected and transported on the main line, while non-pure cotton textiles are pushed to the branches for transport. The conveyor belt speed and production line branch configuration can be flexibly adjusted according to the actual production line needs.

[0060] Three near-infrared spectroscopy detection devices are arranged in sequence above the conveyor belt. Each detection device integrates three MEMS sensors in the 1750nm-2150nm band. With the help of spatial multiplexing, each sensor is assigned an average of 17 sampling points. After sampling, the spectral information of a total of 51 sampling points in the complete band is spliced ​​and output using a spectral stitching algorithm.

[0061] The sorting module is used to push non-pure cotton textiles out of the branch when multiple near-infrared spectroscopy detection devices detect them.

[0062] The spectral cloud platform is used for storing spectral data and processing results. Based on the prediction results generated by each near-infrared spectral detection device, it makes a final judgment based on a voting mechanism and sends the judgment result to the system control module for sorting instructions.

[0063] The system control module is used to control the power supply of the detection system, the conveyor speed of the conveyor belt, and to control the branch push of the sorting module.

[0064] Specifically, the near-infrared spectroscopy detection device integrates a spectral stitching algorithm and a pre-built spectral model to perform local spectral stitching and model prediction calculations, thereby further reducing online detection time.

[0065] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-sensor near-infrared spectroscopy online sorting and detection method, characterized in that, include: Multiple near-infrared spectroscopy detection devices are arranged in sequence directly above the conveyor belt; The conveyor belt is allowed to idle for a specific time. Each near-infrared spectroscopy detection device detects a specified number of background spectral data of the conveyor belt, uploads them to the cloud platform for storage, and issues a start detection command. The object to be inspected is laid flat on the center line of the conveyor belt, and multiple near-infrared spectral detection devices are used to collect spectral data of the conveyor belt online and one by one at a uniform speed. When the difference between a certain spectral data collected at a constant speed and the background spectral data of the conveyor belt exceeds a threshold, it is determined that an object to be detected has appeared in the detection area, and thus that spectral data is used as the starting spectral signal. After the initial spectral signal, each near-infrared spectral detection device extracts N spectra of the same analyte as valid spectral data. After averaging the N valid spectral data, the data are input into the model for category prediction, and the prediction results are uploaded to the cloud platform for storage. Multiple near-infrared spectroscopy detection devices generate multiple prediction results, which are then used to make a final judgment based on a voting mechanism. Based on the judgment result, a pass-through or branch push-out signal is sent to the conveyor belt to achieve the sorting of the items to be inspected.

2. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 1, characterized in that, Each of the near-infrared spectroscopy detection devices integrates multiple MEMS sensors in the same spectral band; the sampling task in a specific spectral band is evenly distributed through spatial multiplexing; after sampling is completed, the complete spectral information of the spectral band is spliced ​​and output through a spectral splicing algorithm.

3. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 2, characterized in that, The spectral stitching method includes: After normalizing the light intensity information of different bands, interpolation is performed at the splicing point to achieve complete splicing of the bands.

4. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 3, characterized in that, The methods for normalizing the light intensity information include: maximum value normalization, mean value normalization, or peak value normalization.

5. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 1, characterized in that, The methods for determining if the difference exceeds the threshold include: Mahalanobis distance, Euclidean distance, or Manhattan distance.

6. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 1, characterized in that, The method for extracting the effective spectral data includes: The amount of effective spectral data that the online near-infrared spectrometer can extract from the object during inspection is determined by considering the conveyor belt, the near-infrared spectroscopy equipment, and the state of the object to be inspected. The formula is as follows: -1； In the formula: N is the number of effective spectral lines, L is the length of the object to be tested along the conveyor belt, and V is the speed of the conveyor belt. The sampling interval is denoted as .

7. The multi-sensor near-infrared spectroscopy online sorting and detection method according to claim 1, characterized in that, The method for determining the final result based on multiple prediction results using a voting mechanism includes: When the predicted results are of different categories, the result with the higher vote rate is selected as the final output. When the predicted results are of the same category, the result with the higher confidence level is selected as the final output.

8. A multi-sensor near-infrared spectroscopy online sorting and detection system, used to implement the multi-sensor near-infrared spectroscopy online sorting and detection method as described in any one of claims 1-7, characterized in that, include: A conveyor belt, including at least one branch conveyor belt, for transporting items to be inspected; Multiple near-infrared spectroscopy detection devices are arranged in sequence directly above the conveyor belt to collect spectral data of the objects to be inspected on the conveyor belt and predict their categories. The sorting module is used to push the item out of the branch when multiple near-infrared spectroscopy detection devices detect that the item belongs to a set category. The spectral cloud platform is used for storing spectral data and processing results. Based on the prediction results generated by each near-infrared spectral detection device, it makes a final judgment based on a voting mechanism and sends the judgment result to the system control module for sorting instructions. The system control module is used to control the power supply of the detection system, the conveyor speed of the conveyor belt, and to control the branch push of the sorting module.

9. The multi-sensor near-infrared spectroscopy online sorting and detection system according to claim 8, characterized in that, Each of the near-infrared spectroscopy detection devices integrates multiple MEMS sensors in the same spectral band; the sampling task in a specific spectral band is evenly distributed through spatial multiplexing; after sampling is completed, the complete spectral information of the spectral band is spliced ​​and output through a spectral splicing algorithm.

10. The multi-sensor near-infrared spectroscopy online sorting and detection system according to claim 9, characterized in that, The near-infrared spectroscopy detection device integrates a spectral stitching algorithm and a pre-built spectral model.

Images