Blueberry bottle sealing quality detection device and method based on spectral analysis

Abstract

The invention relates to the technical field of sealing function detection, in particular to a blueberry bottle sealing quality detection device and method based on spectral analysis. The technical problems that bottle opening spot check is needed when pressure and vacuum methods are adopted for detecting the sealing performance of existing bottled food, so that products are wasted and cannot be completely checked, the detection rate of micropore leakage is low, internal microscopic cracks cannot be detected, and baseline drift is serious in a high-temperature and high-humidity environment in an existing method; according to the technical scheme, the blueberry bottle sealing quality detection device based on spectral analysis comprises a support, a conveying belt, a limiting groove, a blueberry bottle body, a probe assembly, a window assembly, an environment sensing assembly, a calibration assembly, a cleaning assembly and an edge calculation assembly. According to the invention, through non-contact detection, a self-cleaning mechanism and dynamic environment compensation, nondestructive and high-precision detection of the sealing performance of the blueberry bottle is realized, the product loss is reduced, the detection precision is improved, and the device can work stably under the condition of surface pollution.

Description

Technical Field

[0001] The present invention relates to the technical field of sealing function detection, and particularly to a sealing quality detection device and method for blueberry bottles based on spectral analysis. Background Art

[0002] Currently, the sealing performance detection of bottled foods mainly relies on pressure decay method, vacuum decay method, visual inspection method and spectroscopy; among them, the pressure decay method judges leakage by detecting the pressure change in the bottle under pressure, but it requires destructive sampling inspection and is not sensitive to micron-level leakage; the vacuum decay method uses a vacuum chamber to pump air and detects the pressure recovery speed, but the detection speed is slow and the leakage point cannot be located; the visual inspection method identifies the morphological defects of the bottle mouth through a camera, but is interfered by label reflection and surface contamination, resulting in a high false detection rate; the traditional spectroscopy is based on the detection of a single gas absorption peak, but is easily affected by temperature drift and bottle body material scattering, with insufficient sensitivity;

[0003] When the existing pressure and vacuum methods are used for the sealing performance detection of bottled foods, it is necessary to open the bottle for sampling inspection, resulting in product waste and incomplete inspection. The existing methods have a low detection rate for micro-hole leakage and cannot detect internal micro-cracks. Visual inspection is interfered by light and labels, and the traditional spectroscopy has serious baseline drift in high-temperature and high-humidity environments.

[0004] Therefore, in view of the above problems, a sealing quality detection device and method for blueberry bottles based on spectral analysis are proposed. Through non-contact detection, self-cleaning mechanism and dynamic environment compensation, non-destructive and high-precision detection of the sealing performance of blueberry bottles is realized, product loss is reduced, detection accuracy is improved, and stable operation can be achieved under surface contamination conditions. Summary of the Invention

[0005] In order to overcome the problems that when the existing pressure and vacuum methods are used for the sealing performance detection of bottled foods, it is necessary to open the bottle for sampling inspection, resulting in product waste and incomplete inspection, the existing methods have a low detection rate for micro-hole leakage and cannot detect internal micro-cracks, visual inspection is interfered by light and labels, and the traditional spectroscopy has serious baseline drift in high-temperature and high-humidity environments.

[0006] The technical solution of the present invention is: a blueberry bottle sealing quality detection device based on spectral analysis, which includes a bracket, a conveyor belt, a limiting groove, a blueberry bottle body, a probe assembly, a window assembly, an environmental perception assembly, a calibration assembly, a cleaning assembly and an edge computing assembly. A conveyor belt is arranged above the bracket. The conveyor belt is a horizontal conveying structure. A limiting groove is opened on the upper surface of the conveyor belt. Multiple groups of limiting grooves are provided, and the multiple groups of limiting grooves are evenly distributed on the surface of the conveyor belt. The inner side of the limiting groove is provided with a blueberry bottle body. A probe assembly is arranged above the bracket. A window assembly is arranged at the bottom of the probe assembly. An environmental perception assembly is arranged on the side of the probe assembly. A calibration assembly is arranged below the probe assembly. A cleaning assembly is arranged below the probe assembly. An edge computing assembly is arranged on one side of the bracket. The probe assembly includes a support arm, a probe housing, a near-infrared light source, a programmable optical field modulator, a liquid crystal panel, a spectroscope prism and a spectral camera. A support arm is arranged above the bracket. The support arm is vertically fixed directly above the detection station of the bracket. A probe housing is arranged below the support arm. A near-infrared light source is arranged inside the probe housing. The near-infrared light source is located on the inner top surface of the probe housing. The emission band of the near-infrared light source is 900-2500nm. The light source power of the near-infrared light source is controlled by a pulse modulation circuit. A programmable optical field modulator is arranged inside the probe housing. The programmable optical field modulator is located in the middle position inside the probe housing. The programmable optical field modulator is directly below the near-infrared light source. A liquid crystal panel is arranged inside the programmable optical field modulator. The liquid crystal panel is perpendicular to the optical axis of the light source of the near-infrared light source. A spectroscope prism is arranged inside the probe housing. The spectroscope prism is fixed at the bottom position of the probe housing through optical adhesive. The spectroscope prism is coaxially aligned with the optical path outlet of the programmable optical field modulator. A spectral camera is arranged on the inner side wall of the probe housing. The spectral camera is connected to the spectroscope prism through an optical fiber bundle. The reflection surface of the spectroscope prism and the lens of the spectral camera form a 45° angle. The window assembly includes an optical window and a hydrophobic coating. An optical window is arranged at the bottom of the probe housing. The optical window covers the bottom opening position of the probe housing. A hydrophobic coating is coated on the surface of the optical window. The hydrophobic coating is a graphene superhydrophobic coating, and the coating thickness is 200±10nm.

[0007] Preferably, the near-infrared light source emits pulsed light to penetrate the sealing material, exciting the characteristic absorption spectrum of gas molecules inside the blueberry bottle. The programmable optical field modulator and the liquid crystal panel dynamically correct the optical path distortion by loading the Zernike polynomial phase compensation pattern, improving the effective signal intensity. Through the spectroscope prism and the spectral camera: the reflected light is separated by wavelength to generate a hyperspectral image cube, capturing the molecular absorption characteristics of the leakage area. Through photocatalytic decomposition to degrade organic residues, self-cleaning treatment of the detection window is realized, improving the window light transmittance and preventing the increase of false detection rate caused by pollutant residues.

[0008] Preferably, the environmental perception component includes a temperature and humidity sensor and a camera. The temperature and humidity sensor is arranged on one side of the probe housing, and the detection end of the temperature and humidity sensor faces the bottle mouth area of the blueberry bottle. The camera is arranged on the other side of the probe housing. The camera is a 3D ToF camera, and the field of view angle of the camera covers the bottle mouth area of the blueberry bottle body.

[0009] Preferably, the environmental parameters of the bottle mouth area are collected in real time by the temperature and humidity sensor to provide a data basis for spectral compensation. The 3D ToF camera scans the bottle mouth thread and the morphology of the sealing film, and combines the spectral data to locate the leakage area.

[0010] Preferably, the calibration component includes a calibration plate, a quantum dot thin film and a lifting cylinder. The calibration plate is arranged directly below the optical window. The quantum dot thin film is arranged on the surface of the calibration plate. The quantum dot thin film is specifically a CdSe / ZnS quantum dot thin film, and the quantum dot thin film is plated on the surface of the calibration plate. The lifting cylinder is arranged below the calibration plate. The lifting cylinder is located above the bracket, and the lifting stroke of the calibration plate is 0-50 mm.

[0011] Preferably, after detecting a certain number of blueberry bottles, the lifting cylinder is started to raise the calibration plate, and the reflected spectrum of the quantum dot thin film is collected. The current spectrum is compared with the initial reference. If the deviation > 0.5%, an alarm is triggered and recalibration is performed. Thus, a stable reference signal is provided by the quantum dot thin film to eliminate the wavelength drift caused by equipment aging. The position of the calibration plate is adjusted by the lifting cylinder to adapt to different bottle heights.

[0012] Preferably, the cleaning component includes an air flow ring, air injection holes, an air pipe and an air pump. The air flow ring is arranged below the optical window. The air flow ring is of a circular ring structure and surrounds the outer periphery of the optical window. A plurality of air injection holes are opened on the inner side of the air flow ring. The plurality of air injection holes are evenly distributed in a circumferential manner on the inner side of the air flow ring. The axis of the air injection hole forms an angle of 30° with the plane of the optical window. One side of the air flow ring is provided with an air pipe, and one end of the air pipe is provided with an air pump. The air pump is located on one side of the bracket.

[0013] Preferably, when pollution is detected, the air pump starts pulsed air injection, and the air is introduced into the air injection holes through the air pipe to clean the optical window, removing fruit residues and condensed water. The near-infrared light source is controlled to continuously irradiate the optical window to stimulate the photocatalytic decomposition reaction of the graphene hydrophobic coating, avoiding mechanical wiping from damaging the optical window.

[0014] Preferably, the edge computing component includes a heat dissipation box, a main board and a fan. The heat dissipation box is arranged on one side of the bracket. The main board is arranged inside the heat dissipation box. The fan is arranged on the inner side of the heat dissipation box. The GPIO interface of the main board is respectively connected to the spectral camera, the temperature and humidity sensor and the camera through a shielded cable.

[0015] The method for detecting the sealing quality of blueberry bottles based on spectral analysis includes the following steps:

[0016] S1: Start the conveyor belt to move the blueberry bottle body to the detection station. Start the near-infrared light source to emit pulsed light. Load the phase compensation pattern generated by the Zernike polynomial through the liquid crystal panel of the programmable optical field modulator to adjust the incident light field wavefront. The reflected light is introduced into the spectral camera through the beam splitter prism, and a hyperspectral image cube of the bottle mouth area is collected at a rate of 10 ms / frame. The spatial resolution of the hyperspectral image cube is 512×512 pixels, and the spectral resolution is 5 nm;

[0017] S2: Perform Fourier transform analysis on the diffraction fringes of the hyperspectral image to extract spatial frequency features. Among them, if the frequency peak is in the range of 10 - 50 lp / mm, it is determined as fruit residue pollution. If the frequency peak < 10 lp / mm and the spectral reflectivity > 85%, it is determined as condensate water pollution. When organic pollutants are identified, control the near-infrared light source to switch to a wavelength of 1064 nm and continuously irradiate the optical window for 30 seconds at a power of 50 mW to stimulate the photocatalytic decomposition reaction of the graphene hydrophobic coating;

[0018] S3: Read the timing data of the temperature and humidity sensor, input it into the pre-trained Transformer-CNN hybrid network model, and extract the spectral spatial features of the bottle mouth area through 3 convolutional layers, where the convolutional kernel is 3×3 and the stride is 1; capture the correlation between temperature and humidity fluctuations and spectral drift through the Transformer encoder, where the number of attention heads of the Transformer encoder is 8 and the hidden layer dimension is 256; then output a pixel-by-pixel compensation coefficient matrix with a resolution of 256×256, and multiply the compensation coefficient matrix with the original spectral data pixel by pixel to correct the baseline drift;

[0019] S4: Perform partial least squares discriminant analysis on the compensated spectral data, extract the absorption peak intensity of O2 at 760 nm, and calculate the O2 permeability based on Fick's law:

[0020]

[0021] Among them, D is the O2 diffusion coefficient, A is the leakage area, P in and P out are the air pressures inside and outside the bottle respectively, L is the thickness of the sealing film. The diffusion coefficient is obtained through the pre-stored bottle mouth material database, and the leakage area is calculated by scanning the bottle mouth morphology through the camera; if the calculated permeability value is greater than the preset threshold, it is determined that the seal has failed;

[0022] S5: Generate a seal quality report, send it to the production line PLC through the Modbus protocol, and at the same time link the external robotic arm device to grab the unqualified bottle body and laser engrave the leakage coordinates on the bottle body label.

[0023] Preferably, in step S1, the method for generating the Zernike polynomial phase compensation pattern is as follows:

[0024] S101: Calculate the wavefront distortion function W(x, y) caused by pollutant scattering

[0025] S102: Expand it into 36-term Zernike polynomials according to the following formula:

[0026]

[0027] where, Z k (x, y) is the Zernike basis function, and a k is the fitting coefficient;

[0028] S103: Load the coefficient matrix into the programmable optical field modulator.

[0029] Preferably, in step S3, the training method of the Transformer-CNN hybrid network model specifically includes:

[0030] S301: Construct an adversarial training dataset to simulate extreme environmental scenarios, specifically:

[0031] Temperature step change: -15°C to +15°C, gradient 2°C / s;

[0032] Humidity step change: 30%RH to 90%RH, gradient 20%RH / s;

[0033] S302: Adopt a weighted loss function, specifically:

[0034] L = α·MSE(y pred , y true ) + β·CrossEntropy(c pred , c true );

[0035] where, α = 0.7, β = 0.3, y is the spectral intensity, and c is the seal status classification label.

[0036] Preferably, in step S4, the calculation method of the leakage area A is as follows:

[0037] S401: Obtain the bottle mouth point cloud data through a 3D ToF camera;

[0038] S402: Fit the bottle mouth plane using the random sample consensus algorithm;

[0039] S403: Calculate the deviation between the point cloud and the fitted plane. If the local deviation > 0.05 mm, it is determined as the leakage area, and the total area A is calculated cumulatively.

[0040] Preferably, in step S5, the coordinate positioning method for laser engraving is as follows:

[0041] S501: Convert the pixel coordinates with the highest leakage signal in the hyperspectral image into the mechanical coordinate system;

[0042] S502: Map it to the robotic arm operating space through the hand-eye calibration matrix.

[0043] Advantages of the present invention:

[0044] 1. The present invention emits broadband light through a near-infrared light source, penetrates the bottle mouth sealing material, dynamically adjusts the phase of the incident light field through a light field modulator to eliminate label reflection interference; collects the reflected light signal through a spectral camera to generate a hyperspectral image cube, and collects temperature and humidity data in real time through an environmental perception component, inputs it into a Transformer-CNN hybrid network model, outputs a pixel-by-pixel compensation coefficient matrix, extracts the O2 absorption peak intensity from the compensated spectral data through the PLS-DA algorithm, and calculates the permeability in combination with Fick's law, avoiding physical contact with the bottle body; thus realizing non-contact non-destructive detection, without opening the bottle or sampling inspection to reduce product loss;

[0045] 2. The present invention loads a Zernike polynomial phase compensation pattern through a light field modulator to enhance the intensity of the effective signal light; uses a quantum dot thin film as an embedded reference benchmark to calibrate the wavelength drift of the light source in real time; constructs an adversarial training dataset to train the Transformer-CNN model to compensate for non-linear environmental interference; thus improving the detection accuracy, increasing the detection rate of microporous leakage, and at the same time reducing the spectral baseline drift caused by temperature and humidity fluctuations;

[0046] 3. The present invention synergistically removes surface contaminants through the air flow ring and the hydrophobic coating, decomposes and degrades organic residues through photocatalysis, realizes self-cleaning treatment of the detection window, improves the light transmittance of the window, and prevents the increase of false detection rate caused by pollutant residues. Description of the drawings

[0047] Figure 1 Shown is a three-dimensional structural schematic diagram of the blueberry bottle sealing quality detection device based on spectral analysis of the present invention;

[0048] Figure 2 Shown is a sectional three-dimensional structural schematic diagram of the blueberry bottle sealing quality detection device based on spectral analysis of the present invention;

[0049] Figure 3 Shown is a partial three-dimensional structural schematic diagram of the blueberry bottle sealing quality detection device based on spectral analysis of the present invention;

[0050] Figure 4The figure shows a schematic diagram of the planar sectional three-dimensional structure of the blueberry bottle sealing quality detection device based on spectral analysis according to the present invention;

[0051] Figure 5 The figure shows a schematic diagram of the step flow of the blueberry bottle sealing quality detection method based on spectral analysis according to the present invention

[0052] Description of reference numerals: 1, bracket; 2, conveyor belt; 3, blueberry bottle body; 201, limit groove; 103, support arm; 402, probe housing; 401, near-infrared light source; 403, programmable optical field modulator; 406, liquid crystal panel; 404, beam splitter prism; 405, spectral camera; 501, optical window; 502, hydrophobic coating; 601, temperature and humidity sensor; 602, camera; 701, calibration plate; 702, quantum dot film; 703, lifting cylinder; 801, air flow ring; 802, air jet hole; 803, air pipe; 804, air pump; 901, heat dissipation box; 902, main board; 903, fan. Detailed implementation manners

[0053] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0054] Please refer to Figure 1 、 Figure 2 、 Figure 3 and Figure 4, the present invention provides an embodiment: a blueberry bottle sealing quality detection device based on spectral analysis, including a bracket 1, a conveyor belt 2, a limiting groove 201, a blueberry bottle body 3, a probe assembly, a window assembly, an environmental perception assembly, a calibration assembly, a cleaning assembly, and an edge computing assembly. Above the bracket 1, there is a conveyor belt 2. The conveyor belt 2 is a horizontal conveying structure. The upper surface of the conveyor belt 2 is provided with a limiting groove 201. There are multiple groups of limiting grooves 201, and the multiple groups of limiting grooves 201 are evenly distributed on the surface of the conveyor belt 2. Inside the limiting groove 201, there is a blueberry bottle body 3. Above the bracket 1, there is a probe assembly. At the bottom of the probe assembly, there is a window assembly. On the side of the probe assembly, there is an environmental perception assembly. Below the probe assembly, there is a calibration assembly. Below the probe assembly, there is a cleaning assembly. On one side of the bracket 1, there is an edge computing assembly. The probe assembly includes a support arm 103, a probe housing 402, a near-infrared light source 401, a programmable optical field modulator 403, a liquid crystal panel 406, a beam splitter prism 404, and a spectral camera 405. Above the bracket 1, there is a support arm 103. The support arm 103 is vertically fixed directly above the detection station of the bracket 1. Below the support arm 103, there is a probe housing 402. Inside the probe housing 402, there is a near-infrared light source 401. The near-infrared light source 401 is located on the inner top surface of the probe housing 402. The emission band of the near-infrared light source 401 is 900 - 2500 nm. The light source power of the near-infrared light source 401 is controlled by a pulse modulation circuit. Inside the probe housing 402, there is a programmable optical field modulator 403. The programmable optical field modulator 403 is located in the middle position inside the probe housing 402. The programmable optical field modulator 403 is directly below the near-infrared light source 401. Inside the programmable optical field modulator 403, there is a liquid crystal panel 406. The liquid crystal panel 406 is perpendicular to the optical axis of the light source of the near-infrared light source 401. Inside the probe housing 402, there is a beam splitter prism 404. The beam splitter prism 404 is fixed to the bottom position of the probe housing 402 by optical adhesive. The beam splitter prism 404 is coaxially aligned with the optical path exit of the programmable optical field modulator 403. On the inner side wall of the probe housing 402, there is a spectral camera 405. The spectral camera 405 is connected to the beam splitter prism 404 through an optical fiber bundle. The reflection surface of the beam splitter prism 404 and the lens of the spectral camera 405 form a 45° angle. The window assembly includes an optical window 501 and a hydrophobic coating 502. At the bottom of the probe housing 402, there is an optical window 501. The optical window 501 covers the bottom opening position of the probe housing 402. The surface of the optical window 501 is coated with a hydrophobic coating 502. The hydrophobic coating 502 is a graphene superhydrophobic coating 502, and the coating thickness is 200 ± 10 nm.

[0055] Preferably, a pulsed light is emitted by the near-infrared light source 401 to penetrate the sealing material, exciting the characteristic absorption spectrum of the gas molecules inside the blueberry bottle. The programmable optical field modulator 403 and the liquid crystal panel 406 dynamically correct the optical path distortion by loading the Zernike polynomial phase compensation pattern, enhancing the effective signal intensity. Through the beam splitting prism 404 and the spectral camera 405: the reflected light is separated by wavelength to generate a hyperspectral image cube, capturing the molecular absorption characteristics of the leakage area, and realizing self-cleaning treatment of the detection window by photocatalytic decomposition and degradation of organic residues, improving the window light transmittance and preventing the increase of false detection rate caused by pollutant residues.

[0056] Preferably, the environmental perception component includes a temperature and humidity sensor 601 and a camera 602. The temperature and humidity sensor 601 is arranged on one side of the probe housing 402, and the detection end of the temperature and humidity sensor 601 faces the bottle mouth area of the blueberry bottle body 3. The camera 602 is arranged on the other side of the probe housing 402. The camera 602 is a 3D ToF camera 602, and the field of view angle of the camera 602 covers the bottle mouth area of the blueberry bottle body.

[0057] Preferably, the environmental parameters of the bottle mouth area are collected in real time by the temperature and humidity sensor 601, providing a data basis for spectral compensation. The 3D ToF camera 602 scans the bottle mouth thread and the sealing film morphology, and locates the leakage area in combination with the spectral data.

[0058] Preferably, the calibration component includes a calibration plate 701, a quantum dot thin film 702, and a lifting cylinder 703. The calibration plate 701 is arranged directly below the optical window 501. The surface of the calibration plate 701 is provided with the quantum dot thin film 702. The quantum dot thin film 702 is specifically a CdSe / ZnS quantum dot thin film 702, and the quantum dot thin film 702 is plated on the surface of the calibration plate 701. A lifting cylinder 703 is arranged below the calibration plate 701. The lifting cylinder 703 is located above the bracket 1, and the lifting stroke of the calibration plate 701 is 0 - 50 mm.

[0059] Preferably, after detecting a certain number of blueberry bottles, the lifting cylinder 703 is started to raise the calibration plate 701, and the reflected spectrum of the quantum dot thin film 702 is collected. The current spectrum is compared with the initial reference. If the deviation > 0.5%, an alarm is triggered and recalibration is performed, so as to provide a stable reference signal through the quantum dot thin film 702, eliminate the wavelength drift caused by equipment aging, and adjust the position of the calibration plate 701 by the lifting cylinder 703 to adapt to different bottle heights.

[0060] Preferably, the cleaning component includes an air flow ring 801, air jet holes 802, an air pipe 803, and an air pump 804. An air flow ring 801 is provided below the optical window 501. The air flow ring 801 is in a circular ring structure and surrounds the outer periphery of the optical window 501. Air jet holes 802 are formed inside the air flow ring 801. There are multiple groups of air jet holes 802, and the multiple groups of air jet holes 802 are evenly distributed in a circumferential direction inside the air flow ring 801. The axis of the air jet hole 802 forms a 30° angle with the plane of the optical window 501. An air pipe 803 is provided on one side of the air flow ring 801, and one end of the air pipe 803 is provided with an air pump 804. The air pump 804 is located on one side of the bracket 1.

[0061] Preferably, when pollution is detected, the air pump 804 starts pulsed jetting, and the air is introduced into the air jet holes 802 along the air pipe 803 to clean the optical window 501, removing fruit residues and condensed water. The near-infrared light source 401 is controlled to continuously irradiate the optical window 501 to stimulate the photocatalytic decomposition reaction of the graphene hydrophobic coating 502, avoiding mechanical wiping from damaging the optical window 501.

[0062] Preferably, the edge computing component includes a heat dissipation box 901, a main board 902, and a fan 903. A heat dissipation box 901 is provided on one side of the bracket 1. The main board 902 is arranged inside the heat dissipation box 901, and a fan 903 is provided inside the heat dissipation box 901. The GPIO interfaces of the main board 902 are respectively connected to the spectral camera 405, the temperature and humidity sensor 601, and the camera 602 through shielded cables.

[0063] Please refer to Figure 5 , a method for detecting the sealing quality of blueberry bottles based on spectral analysis, includes the following steps:

[0064] S1: Start the conveyor belt 2 to move the blueberry bottle body 3 to the detection station. Start the near-infrared light source 401 to emit pulsed light. Load the phase compensation pattern generated by the Zernike polynomial through the liquid crystal panel 406 of the programmable optical field modulator 403 to adjust the wavefront of the incident light field. The reflected light is introduced into the spectral camera 405 through the beam splitter prism 404, and a hyperspectral image cube of the bottle mouth area is collected at a rate of 10 ms / frame. The spatial resolution of the hyperspectral image cube is 512×512 pixels, and the spectral resolution is 5 nm;

[0065] S2: Perform Fourier transform analysis on the diffraction fringes of the hyperspectral image to extract spatial frequency characteristics. Among them, if the frequency peak is in the range of 10 - 50 lp / mm, it is determined as fruit residue pollution. If the frequency peak < 10 lp / mm and the spectral reflectance > 85%, it is determined as condensed water pollution. When organic pollutants are identified, control the near-infrared light source 401 to switch to a wavelength of 1064 nm and continuously irradiate the optical window 501 for 30 seconds at a power of 50 mW to stimulate the photocatalytic decomposition reaction of the graphene hydrophobic coating 502;

[0066] S3: Read the timing data of the temperature and humidity sensor 601, input it into the pre-trained Transformer-CNN hybrid network model, and extract the spectral spatial features of the bottle mouth area through 3 convolutional layers, where the convolutional kernel is 3×3 and the stride is 1; capture the correlation between temperature and humidity fluctuations and spectral drift through the Transformer encoder, where the number of attention heads of the Transformer encoder is 8 and the hidden layer dimension is 256; then output a per-pixel compensation coefficient matrix with a resolution of 256×256, multiply the compensation coefficient matrix with the original spectral data pixel by pixel to correct the baseline drift;

[0067] S4: Perform partial least squares discriminant analysis on the compensated spectral data, extract the absorption peak intensity of O2 at 760 nm, and calculate the O2 permeability based on Fick's law:

[0068]

[0069] where D is the O2 diffusion coefficient, A is the leakage area, P in and P out are the air pressures inside and outside the bottle respectively, L is the thickness of the sealing film, the diffusion coefficient is obtained through the pre-stored bottle mouth material database, and the leakage area is calculated by scanning the bottle mouth morphology with the camera 602; if the calculated permeability value is obtained, it is determined that the seal has failed;

[0070] S5: Generate a seal quality report, send it to the production line PLC through the Modbus protocol, and at the same time link the external robotic arm device to grab the unqualified bottles and laser engrave the leakage coordinates on the bottle body label.

[0071] Preferably, in step S1, the generation method of the Zernike polynomial phase compensation pattern is as follows:

[0072] S101: Calculate the wavefront distortion function W(x, y) caused by pollutant scattering

[0073] S102: Expand it into 36 terms of Zernike polynomials according to the following formula:

[0074]

[0075] where Z k (x, y) is the Zernike basis function, and a k is the fitting coefficient;

[0076] S103: Load the coefficient matrix into the programmable optical field modulator 403.

[0077] Preferably, in step S3, the training method of the Transformer-CNN hybrid network model specifically includes:

[0078] S301: Construct an adversarial training dataset to simulate extreme environmental scenarios, specifically:

[0079] Temperature step change: from -15°C to +15°C, gradient 2°C / s;

[0080] Humidity step change: from 30%RH to 90%RH, gradient 20%RH / s;

[0081] S302: Adopt a weighted loss function, specifically:

[0082] L = α·MSE(y pred , y true ) + β·CrossEntropy(c pred , c true );

[0083] Among them, α = 0.7, β = 0.3, y is the spectral intensity, and c is the classification label of the sealing state.

[0084] Preferably, in step S4, the calculation method of the leakage area A is:

[0085] S401: Obtain the bottle mouth point cloud data through the 3D ToF camera 602;

[0086] S402: Fit the bottle mouth plane using the random sample consensus algorithm;

[0087] S403: Calculate the deviation between the point cloud and the fitted plane. If the local deviation > 0.05mm, it is determined as the leakage area, and the total area A is calculated cumulatively.

[0088] Preferably, in step S5, the coordinate positioning method for laser engraving is:

[0089] S501: Convert the pixel coordinates with the highest leakage signal in the hyperspectral image into the mechanical coordinate system;

[0090] S502: Map it to the robotic arm operating space through the hand-eye calibration matrix.

[0091] Example 1

[0092] Optionally, it is set that a certain blueberry processing production line needs to conduct 100% online sealing detection on 250mL cylindrical PET bottles, with a bottle mouth diameter of 28mm, a heat-sealed aluminum film thickness of 0.1mm, and a production line speed of 300 bottles per minute. The traditional vacuum detection method cannot meet the requirements due to a speed < 100 bottles per minute and is insensitive to micron-level leaks; therefore, the online detection of standard blueberry bottles is carried out using the blueberry bottle sealing quality detection device and method based on spectral analysis of the present invention; the specific steps are:

[0093] Device Configuration: The spectral probe is installed 20 cm above the conveyor belt 2. The power of the near-infrared light source 401 is set to 80 mW. The frame rate of the spectral camera 405 is 60 fps, the resolution is 512×512 pixels, and the spectral resolution is 5 nm; the edge computing unit loads a pre-trained model, and the pre-trained model is trained based on 5000 sets of standard bottle spectral data + environmental parameters;

[0094] Detection Process:

[0095] Step 1: After the bottle body is in place, the optical field modulator loads the Zernike 36-term compensation pattern, where the fitting coefficients a1 to a36 are obtained through on-line calibration, and a hyperspectral image is collected;

[0096] Step 2: When fruit residue contamination is detected, with a frequency peak of 35 lp / mm, the 1064 nm light source cleaning window is activated for 30 seconds, and the photocatalytic decomposition efficiency is ≥85%;

[0097] Step 3: The environmental sensor measures a temperature of 25°C ± 2°C and a humidity of 60% RH. The Transformer-CNN model outputs the compensation coefficient, where the maximum correction amount is 0.8%;

[0098] Step 4: Calculate the O2 permeability and determine it to be qualified;

[0099] Through the above steps, the specific detection results are as follows: 0.05 mm micro-hole leakage is detected, and the detection rate is 99.1%; the single-bottle detection time is 180 ms, supporting a production speed of 300 bottles per minute; the misdetection rate of label reflection is 0.4%.

[0100] Example 2

[0101] Optionally, when detecting a 500 mL square HDPE bottle, the bottle mouth thread is asymmetric, the sealing film is made of biodegradable PLA material with a thickness of 0.15 mm; traditional spectral detection requires manual re-calibration of the bottle type, which takes up to 8 hours; therefore, the spectral analysis-based blueberry bottle sealing quality detection device and method of the present invention are used for detecting the sealing of special-shaped bottle mouths; the specific steps are as follows:

[0102] Step 1: Update the bottle mouth material database, and the O2 diffusion coefficient D of PLA is 1.2*10 -13 m 2 / s;

[0103] Step 2: The 3D ToF camera 602 scans the bottle mouth morphology, with a point cloud resolution of 0.1 mm, and generates a three-dimensional model of the thread area;

[0104] Step 3: The model inputs 50 sets of new bottle type data online, and the compensation accuracy is restored from 71% to 97%;

[0105] Step 4: The RANSAC algorithm fits the bottle mouth plane, and the calculation error of the leakage area is ≤5%;

[0106] Through the above steps, the specific detection results are as follows: The adaptation time for the new bottle type is 23 minutes; the deformation detection accuracy is 0.08 mm.

[0107] Example 3

[0108] Optionally, set the environmental humidity fluctuation in the production line in the tropical region to 70 - 95% RH. Condensate water on the bottle surface causes the misdetection rate of traditional spectral detection to be as high as 12.3%. Use the blueberry bottle sealing quality detection device and method based on spectral analysis of the present invention to conduct high-humidity environment stability tests; the specific steps are as follows:

[0109] Step 1: Activate the pulsed air flow ring 801 to remove condensate water, where the air pressure is 0.5 MPa and the jetting frequency is 10 Hz;

[0110] Step 2: Reduce water droplet adhesion through the contact angle of the superhydrophobic coating 502 > 150°;

[0111] Step 3: The Transformer encoder captures the humidity step change, where the gradient is 20% RH / s, and the spectral baseline drift after compensation is ≤0.5%;

[0112] Through the above steps, the specific detection results are as follows: The misdetection rate at a humidity of 90% RH is 0.8%; the pulsed air flow removes condensate water, avoiding mechanical contact pollution.

[0113] Example 4

[0114] Optionally, the contamination rate of residual pectin and fruit residue at the bottle mouth of the existing blueberry jam filling line is 15%. Traditional detection requires frequent shutdowns for manual cleaning. Therefore, use the blueberry bottle sealing quality detection device and method based on spectral analysis of the present invention to conduct standard fruit residue pollution self-cleaning detection; the specific steps are as follows:

[0115] Step 1: Identify pectin contamination, where the spectral reflectance is 92%, start photocatalytic cleaning, and the infrared light source parameters are 1064 nm, 60 mW, and 40 seconds;

[0116] Step 2: The graphene coating catalyzes the decomposition of pectin, and the decomposition rate is 87%;

[0117] Step 3: After cleaning, the measured light transmittance of the window returns to 98.6%;

[0118] Through the above steps, the specific detection results are as follows: The pollution misjudgment rate is 0.2%; no manual intervention is required for cleaning; photocatalytic decomposition does not consume chemical reagents.

[0119] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A blueberry bottle sealing quality detection device based on spectral analysis, comprising a bracket (1) and a conveyor belt (2), characterized in that: It also includes a limit groove (201), a blueberry bottle body (3), a probe assembly, a window assembly, an environmental perception assembly, a calibration assembly, a cleaning assembly and an edge computing assembly. Above the bracket (1), there is a conveyor belt (2). The conveyor belt (2) is a horizontal conveying structure. The upper surface of the conveyor belt (2) is provided with limit grooves (201). There are multiple groups of limit grooves (201), and the multiple groups of limit grooves (201) are evenly distributed on the surface of the conveyor belt (2). Inside the limit groove (201), there is a blueberry bottle body (3). Above the bracket (1), there is a probe assembly. At the bottom of the probe assembly, there is a window assembly. On the side of the probe assembly, there is an environmental perception assembly. Below the probe assembly, there is a calibration assembly. Below the probe assembly, there is a cleaning assembly. On one side of the bracket (1), there is an edge computing assembly. The probe assembly includes a support arm (103), a probe housing (402), a near-infrared light source (401), a programmable optical field modulator (403), a liquid crystal panel (406), a beam splitting prism (404) and a spectral camera (405). Above the bracket (1), there is a support arm (103). The support arm (103) is vertically fixed directly above the detection station of the bracket (1). Below the support arm (103), there is a probe housing (402). Inside the probe housing (402), there is a near-infrared light source (401). The near-infrared light source (401) is located on the inner top surface of the probe housing (402). The emission band of the near-infrared light source (401) is 900 - 2500 nm. The light source power of the near-infrared light source (401) is controlled by a pulse modulation circuit. Inside the probe housing (402), there is a programmable optical field modulator (403). The programmable optical field modulator (403) is located in the middle position inside the probe housing (402). The programmable optical field modulator (403) is directly below the near-infrared light source (401). Inside the programmable optical field modulator (403), there is a liquid crystal panel (406). The liquid crystal panel (406) is perpendicular to the optical axis of the light source of the near-infrared light source (401). Inside the probe housing (402), there is a beam splitting prism (404). The beam splitting prism (404) is fixed at the bottom position of the probe housing (402) through optical adhesive. The beam splitting prism (404) is coaxially aligned with the optical path outlet of the programmable optical field modulator (403). On the inner side wall of the probe housing (402), there is a spectral camera (405). The spectral camera (405) is connected to the beam splitting prism (404) through an optical fiber bundle. The reflection surface of the beam splitting prism (404) and the lens of the spectral camera (405) form a 45° angle. The window assembly includes an optical window (501) and a hydrophobic coating (502). At the bottom of the probe housing (402), there is an optical window (501). The optical window (501) covers the bottom opening position of the probe housing (402). The surface of the optical window (501) is coated with a hydrophobic coating (502). The hydrophobic coating (502) is a graphene superhydrophobic coating (502), and the coating thickness is 200 ± 10 nm.

2. The blueberry bottle sealing quality detection device based on spectral analysis according to claim 1, characterized in that: The environmental perception component includes a temperature and humidity sensor (601) and a camera (602). The temperature and humidity sensor (601) is provided on one side of the probe housing (402), and the detection end of the temperature and humidity sensor (601) faces the bottle mouth area of the blueberry bottle body (3). The camera (602) is provided on the other side of the probe housing (402). The camera (602) is a 3D ToF camera (602), and the field of view angle of the camera (602) covers the bottle mouth area of the blueberry bottle body.

3. The blueberry bottle sealing quality detection device based on spectral analysis according to claim 1, characterized in that: The calibration component includes a calibration plate (701), a quantum dot thin film (702), and a lifting cylinder (703). The calibration plate (701) is provided directly below the optical window (501). The quantum dot thin film (702) is provided on the surface of the calibration plate (701). The quantum dot thin film (702) is specifically a CdSe / ZnS quantum dot thin film (702), and the quantum dot thin film (702) is plated on the surface of the calibration plate (701). The lifting cylinder (703) is provided below the calibration plate (701). The lifting cylinder (703) is located above the bracket (1), and the lifting stroke of the calibration plate (701) is 0 - 50 mm.

4. The blueberry bottle sealing quality detection device based on spectral analysis according to claim 1, characterized in that: The cleaning component includes an air flow ring (801), air jet holes (802), an air pipe (803), and an air pump (804). The air flow ring (801) is provided below the optical window (501). The air flow ring (801) is of a circular ring structure and surrounds the outer periphery of the optical window (501). A plurality of air jet holes (802) are provided on the inner side of the air flow ring (801). The plurality of air jet holes (802) are evenly distributed in a circumferential direction on the inner side of the air flow ring (801). The axis of the air jet hole (802) forms an angle of 30° with the plane of the optical window (501). The air pipe (803) is provided on one side of the air flow ring (801), and one end of the air pipe (803) is provided with the air pump (804). The air pump (804) is located on one side of the bracket (1).

5. The blueberry bottle sealing quality detection device based on spectral analysis according to claim 1, characterized in that: The edge computing component includes a heat dissipation box (901), a main board (902), and a fan (903). The heat dissipation box (901) is provided on one side of the bracket (1). The main board (902) is provided inside the heat dissipation box (901). The fan (903) is provided on the inner side of the heat dissipation box (901). The GPIO interface of the main board (902) is respectively connected to the spectral camera (405), the temperature and humidity sensor (601), and the camera (602) through shielded cables.

6. A method for detecting the sealing quality of blueberry bottles based on spectral analysis, characterized in that: It includes the following steps: S1: Start the conveyor belt (2) to move the blueberry bottle body (3) to the detection station. Start the near-infrared light source (401) to emit pulsed light. Load the phase compensation pattern generated by the Zernike polynomial through the liquid crystal panel (406) of the programmable optical field modulator (403) to adjust the incident light field wavefront. The reflected light is introduced into the spectral camera (405) through the beam splitter prism (404), and a hyperspectral image cube of the bottle mouth area is collected at a rate of 10 ms / frame. The spatial resolution of the hyperspectral image cube is 512×512 pixels, and the spectral resolution is 5 nm. S2: Perform Fourier transform analysis on the diffraction patterns of the hyperspectral image to extract spatial rate features, wherein if the frequency peak is in the range of 10-50lp / mm, it is determined to be pomace pollution; if the frequency peak is <10lp / mm and the spectral reflectance is >85%, it is determined to be condensed water pollution. When organic pollutants are identified, the near-infrared light source (401) is controlled to switch to a wavelength of 1064nm, and the optical window (501) is continuously irradiated with a power of 50mW for 30 seconds to stimulate the photocatalytic decomposition reaction of the graphene hydrophobic coating (502); S3: read the time series data of the temperature and humidity sensor (601), input the pre-trained Transformer-CNN hybrid network model, and extract the spectral spatial features of the bottle mouth area through 3 convolution layers, where the convolution kernel is 3×3 and the step size is 1; capture the correlation between temperature and humidity fluctuations and spectral drift through the Transformer encoder, where the Transformer encoder has 8 attention heads and a hidden layer dimension of 256; then output a pixel-by-pixel compensation coefficient matrix with a resolution of 256×256, multiply the compensation coefficient matrix by the original spectral data pixel by pixel, and correct the baseline drift; S4: Perform partial least squares discriminant analysis on the compensated spectral data, extract the absorption peak intensity of O2 at 760nm, and calculate the O2 permeability based on Fick's law: where D is the O2 diffusion coefficient, A is the leakage area, P in and P out are the air pressures inside and outside the bottle respectively, L is the thickness of the sealing film, the diffusion coefficient is obtained from the pre-stored database of bottle mouth materials, and the leakage area is calculated by scanning the morphology of the bottle mouth with a camera (602); if the calculated permeability value is greater than the preset threshold, it is determined that the seal has failed; S5: Generate a sealing quality report and send it to the production line PLC via the Modbus protocol. At the same time, an external robotic arm device is linked to grab unqualified bottles and laser engrave the leakage coordinates on the bottle label.

7. The method for detecting the sealing quality of blueberry bottles based on spectral analysis according to claim 6, characterized in that: In step S1, the method for generating the Zernike polynomial phase compensation pattern is: S101: Calculate the wavefront distortion function W(x, y) caused by contaminant scattering S102: Expand into 36 Zernike polynomials according to the following formula: Among them, Z k (x, y) is the Zernike basis function, and a k is the fitting coefficient; S103: Load the coefficient matrix into the programmable light field modulator (403).

8. The method for detecting the sealing quality of blueberry bottles based on spectral analysis according to claim 6, characterized in that: In step S3, the training method of the Transformer-CNN hybrid network model specifically includes: S301: Construct an adversarial training dataset to simulate extreme environmental scenarios, specifically: Temperature step change: -15℃ to +15℃, gradient 2℃ / s; Humidity step change: 30%RH to 90%RH, gradient 20%RH / s; S302: Using a weighted loss function, specifically: L = α·MSE(y pred , y true ) + β·CrossEntropy(c pred , c true ); Among them, α=0.7, β=0.3, y is the spectral intensity, and c is the sealing state classification label.

9. The method for detecting the sealing quality of blueberry bottles based on spectral analysis according to claim 6, wherein: In step S4, the calculation method of the leakage area A is: S401: Obtaining bottle mouth point cloud data through a 3D ToF camera (602); S402: fitting the bottle mouth plane using a random sampling consistency algorithm; S403: Calculate the deviation between the point cloud and the fitting plane. If the local deviation is greater than 0.05 mm, it is determined to be a leakage area, and the total area A is cumulatively calculated.

10. The method for detecting the sealing quality of blueberry bottles based on spectral analysis according to claim 6, characterized in that: In step S5, the coordinate positioning method of laser marking is: S501: converting the pixel coordinates with the highest leakage signal in the hyperspectral image into a mechanical coordinate system; S502: Mapping to the robot arm operation space through the hand-eye calibration matrix.

Images