Soft package leakproofness detection method and system based on image recognition

Abstract

The application discloses a soft package sealing property detection method and system based on image recognition, relates to the field of soft package sealing property detection, and comprises the following steps: placing a soft package to be detected into a vacuum cavity, performing first-stage infrared irradiation under a first pressure condition, and collecting initial image data; performing negative pressure treatment on the vacuum cavity until a set pressure threshold is reached, performing second-stage infrared irradiation under a second pressure condition, and collecting deformation image data; aligning the initial image data and the deformation image data, extracting image feature difference data, and calculating a deformation vector of a sealing area according to the image feature difference data; constructing a soft package surface deformation distribution map according to the deformation vector to perform sealing determination, and determining a leaky package detection result. The application solves the technical problems of low efficiency and accuracy of existing soft package sealing property detection, and achieves the technical effect of improving sealing property detection efficiency and accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of soft package seal detection, and particularly relates to a soft package seal detection method and system based on image recognition. BACKGROUND

[0002] Soft package seal detection is crucial for ensuring product quality and safety, especially in the fields of food, medicine and electronic products. Poor seal can cause product deterioration, contamination or damage, thereby affecting the health of consumers and the service life of products. Currently, soft package seal detection mainly adopts traditional methods such as pressure testing, vacuum testing and manual inspection. These methods are easily disturbed by human factors, have low detection efficiency and are difficult to accurately quantify the small deformation of the sealing area, resulting in inaccurate detection results and low reliability.

[0003] At present, in the related art, there are technical problems of low efficiency and accuracy in soft package seal detection. SUMMARY

[0004] The present application provides a soft package seal detection method and system based on image recognition. The soft package is placed in a vacuum chamber, an initial infrared image is collected under normal pressure, the vacuum is extracted to a set negative pressure, a deformed infrared image is collected, the two images are compared, the seal area deformation is calculated, the sealability is determined according to the deformation distribution map, and the leak package detection result is obtained. Technical means such as, achieve the technical effect of improving the efficiency and accuracy of seal detection.

[0005] The present application provides a soft package seal detection method based on image recognition, comprising: placing a soft package to be tested into a vacuum chamber, performing first-stage infrared radiation under a first pressure condition, and collecting initial image data; performing negative pressure processing on the vacuum chamber to a set pressure threshold, performing second-stage infrared radiation under a second pressure condition, and collecting deformed image data; aligning the initial image data and the deformed image data, extracting image feature difference data, and calculating a deformation vector of a sealing area according to the image feature difference data; constructing a soft package surface deformation distribution map according to the deformation vector for seal determination, and determining a leak package detection result.

[0006] In a possible implementation, the following processing is performed: the collection process of the first-stage infrared irradiation and the second-stage infrared irradiation includes: arranging a plurality of infrared sensor arrays in the vacuum cavity, setting an irradiation intensity of an infrared light source according to a soft package material characteristic, activating the plurality of infrared sensor arrays to collect data at the irradiation intensity, and obtaining a first reflection intensity distribution map; adding the first reflection intensity distribution map to the initial image data; after the negative pressure processing is completed, dynamically adjusting an irradiation intensity of a second-stage infrared light source based on the first reflection intensity distribution map, activating the plurality of infrared sensor arrays to collect data at the irradiation intensity, and obtaining a second reflection intensity distribution map; and adding the second reflection intensity distribution map to the deformed image data.

[0007] In a possible implementation, the initial image data and the deformed image data are registered and aligned, and image feature difference data is extracted, and the following processing is performed: key feature points in the initial image data are extracted based on the first reflection intensity distribution map, and a first feature point set is established; a second feature point set is extracted based on edge enhancement processing of the deformed image data based on the second reflection intensity distribution map; feature point matching is performed based on the first feature point set and the second feature point set, and a two-stage infrared image pair is determined, the two-stage infrared image pair including a spatial transformation matrix; affine transformation is performed on the initial image data and the deformed image data according to the spatial transformation matrix, and the image feature difference data is determined.

[0008] In a possible implementation, a deformation vector of a sealed area is calculated according to the image feature difference data, and the following processing is performed: historical leak package sample data is called, image deformation labeling is performed on the historical leak package sample data, and a deformation abnormal area label of the sealed area is obtained; convolution calculation is performed on the two-stage infrared image pair according to the deformation abnormal area label, and a pixel-level deformation probability map is constructed; deformation calculation is performed on the sealed area according to the pixel-level deformation probability map, and a deformation vector of the sealed area is generated.

[0009] In a possible implementation, the soft package surface deformation distribution map is constructed according to the deformation vector to perform sealing judgment, to determine a leaky package detection result, and the following processing is performed: a soft package surface coordinate system is constructed, the deformation vector is mapped to the soft package surface coordinate system, normalized processing is performed in combination with the soft package material characteristics, and the soft package surface deformation distribution map is constructed; data extraction is performed by traversing the soft package surface deformation distribution map, region segmentation is performed according to a plurality of deformation data sets, and a suspicious leakage region set is generated; morphological optimization is performed on the suspicious leakage region set, deformation analysis is performed according to optimized morphological data, and a deformation vector amplitude is obtained; the deformation vector amplitude is used as an index to perform retrieval matching on the historical leaky package sample data, a leaky package matching degree is generated, and the leaky package matching degree is added to the leaky package detection result.

[0010] In a possible implementation, the deformation vector amplitude is used as an index to perform retrieval matching on the historical leaky package sample data, a leaky package matching degree is generated, and the leaky package matching degree is added to the leaky package detection result, and the following processing is performed: the soft package surface deformation distribution map is grid segmented to obtain a plurality of detection units; the plurality of detection units are calculated in combination with the deformation vector amplitude to generate a multi-dimensional feature vector, the multi-dimensional feature vector is analyzed and reduced in dimension to generate a deformation amplitude feature sequence; the deformation amplitude feature sequence is used as an index to perform similarity retrieval on the historical leaky package sample database to generate a leaky package matching degree; the leaky package matching degree is compared with a dynamic threshold value, a leakage risk level is determined according to a comparison result, and the leaky package matching degree and the leakage risk level are integrated into the leaky package detection result.

[0011] In a possible implementation, the leaky package matching degree is compared with a dynamic threshold value, and the following processing is performed: a reference matching degree threshold value table is obtained by querying, according to the soft package material characteristics, in combination with a detection environment humidity; a dynamic threshold value is generated by introducing a real-time pressure change rate to perform linear interpolation compensation on the reference matching degree threshold value table; when the leaky package matching degree exceeds the dynamic threshold value, a high-confidence leakage alarm is triggered, and the comparison result is generated.

[0012] The application also provides a soft package leak detection system based on image recognition, comprising: an initial image data acquisition module, configured to place a soft package to be detected into a vacuum cavity, perform first-stage infrared irradiation under a first pressure condition, and acquire initial image data; a deformed image data acquisition module, configured to perform negative pressure treatment on the vacuum cavity to a set pressure threshold, perform second-stage infrared irradiation under a second pressure condition, and acquire deformed image data; an image feature difference data extraction module, configured to register and align the initial image data and the deformed image data, extract image feature difference data, and calculate a deformation vector of a sealing area according to the image feature difference data; and a sealing determination module, configured to construct a soft package surface deformation distribution map according to the deformation vector to determine a sealing result.

[0013] The soft package leak detection method and system based on image recognition provided in the application first place a soft package to be detected into a vacuum cavity, perform first-stage infrared irradiation under a first pressure condition, acquire initial image data, then perform negative pressure treatment on the vacuum cavity to a set pressure threshold, perform second-stage infrared irradiation under a second pressure condition, acquire deformed image data, then register and align the initial image data and the deformed image data, extract image feature difference data, calculate a deformation vector of a sealing area according to the image feature difference data, finally construct a soft package surface deformation distribution map according to the deformation vector to determine a sealing result, thereby improving the sealing detection efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings of the embodiments of the application will be briefly introduced as follows. In the present application, a flowchart is used to illustrate the operations performed by the system according to the embodiments of the application. It should be understood that the foregoing or the following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously according to needs. Meanwhile, other operations can be added to these processes, or one or more steps of operations can be removed from these processes.

[0015] Figure 1 The flowchart of the soft package leak detection method based on image recognition provided in the embodiments of the application.

[0016] Figure 2 The structural schematic diagram of the soft package leak detection system based on image recognition provided in the embodiments of the application.

[0017] Legend: initial image data acquisition module 10, deformed image data acquisition module 20, image feature difference data extraction module 30, and sealing determination module 40. DETAILED DESCRIPTION

[0018] The above description is only a summary of the technical solutions of the present application. In order to make the technical means of the present application more clearly understood, and to be implemented according to the content of the description, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described.

[0019] In order to make the purposes, technical solutions and advantages of the present application more clear, the following will combine the drawings to make a further detailed description of the present application. The described embodiments should not be regarded as a limitation of the present application. All other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] In the following description, "some embodiments" are described, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subset of all possible embodiments, and can be combined with each other without conflict. The term "first\second" referred to is only to distinguish similar objects, and does not represent a specific order of the objects. The terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application.

[0021] The embodiments of the present application provide a soft package leak detection method based on image recognition, as shown in Figure 1 The method comprises the following steps:

[0022] In step S100, the soft package to be tested is placed in a vacuum cavity, and a first stage infrared irradiation is performed under a first pressure condition to collect initial image data.

[0023] Specifically, the soft package to be tested is placed in a sealed vacuum cavity. The cavity is a sealed container for vacuum treatment of the internal object, equipped with a pressure sensor and a vacuum pump for controlling and monitoring the pressure in the cavity. Under the first pressure condition (for example, normal pressure or slight negative pressure), the soft package is irradiated using an infrared irradiation device. The infrared device can be an infrared lamp or an infrared emitter, and the wavelength and power are selected according to the characteristics of the soft package material. High-resolution industrial cameras are used to collect initial image data of the soft package. The parameters of the camera (such as exposure time, resolution, etc.) are optimized according to the size and material of the soft package.

[0024] For example, assuming the soft package is a food packaging bag, it is placed in a vacuum chamber with a diameter of 50 cm and a height of 30 cm. The pressure sensor in the chamber detects that the current pressure is 1 atmosphere (normal pressure). At this time, the infrared lamp with a wavelength of 850 nm and a power of 50 W is started to irradiate the surface of the soft package for 10 seconds. Then, an industrial camera with a resolution of 2048x1536 pixels is used to take an initial image of the soft package with an exposure time of 1 / 100 seconds.

[0025] In step S200, the vacuum chamber is subjected to negative pressure treatment to a set pressure threshold, the second stage infrared irradiation is performed under the second pressure condition, and the deformation image data is collected.

[0026] Specifically, the vacuum chamber is subjected to negative pressure treatment by a vacuum pump, so that the pressure in the chamber is reduced to a set pressure threshold (for example, -0.5 atmosphere). The pressure sensor monitors the pressure change in real time to ensure that the set value is reached. Under the second pressure condition (negative pressure state), the soft package is irradiated again using the infrared irradiation device. The same industrial camera as in step S100 is used to collect the deformation image data of the soft package under the negative pressure condition.

[0027] For example, after step S100 is completed, the vacuum pump is started to reduce the pressure in the chamber from 1 atmosphere to -0.5 atmosphere. When the pressure sensor detects that the pressure reaches the set value, the infrared lamp is started again to irradiate the soft package for 10 seconds with the same parameters (wavelength 850 nm, power 50 W). Then, the industrial camera is used to take a deformation image of the soft package under the negative pressure condition with the same parameters (resolution 2048x1536 pixels, exposure time 1 / 100 seconds).

[0028] In one possible implementation, the collection process of the first stage infrared irradiation and the second stage infrared irradiation includes: arranging a plurality of infrared sensor arrays in the vacuum chamber, setting the irradiation intensity of the infrared light source according to the material characteristics of the soft package, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a first reflection intensity distribution map; adding the first reflection intensity distribution map to the initial image data; after the negative pressure treatment is completed, dynamically adjusting the irradiation intensity of the second stage infrared light source based on the first reflection intensity distribution map, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a second reflection intensity distribution map; and adding the second reflection intensity distribution map to the deformation image data.

[0029] Specifically, multiple infrared sensor arrays are evenly arranged within the vacuum chamber. These sensor arrays can detect the intensity of infrared reflection and convert it into an electrical signal. The arrangement of the sensor arrays can be grid-like or ring-shaped to ensure full coverage of the soft package surface. For example, in a vacuum chamber with a diameter of 50 cm and a height of 30 cm, the height of the chamber is divided into 5 equal parts, each with a height of 6 cm. At each equal height position, a ring-shaped array is arranged, and 10 sensors are evenly arranged on each ring-shaped array. The angular interval between the sensors is 36 degrees (360 degrees / 10 sensors). These sensors can detect infrared reflection intensity in the wavelength range of 800 nm to 1000 nm.

[0030] According to the characteristics of the soft package material, such as the infrared absorption and reflectivity of the material, the initial wavelength, irradiation intensity, and irradiation angle of the infrared light source are set (the angle between the infrared light source and the soft package surface). These parameters can be determined in advance through experiments to ensure the best image acquisition effect. For example, for a common plastic soft package material, the initial wavelength of the infrared light source is set to 850 nm, the irradiation intensity is 50 watts, and the irradiation angle is 45 degrees. These parameters can ensure that the infrared light can effectively penetrate the soft package material and produce clear reflection signals on the surface.

[0031] The infrared sensor array is activated, and data acquisition is performed according to the set irradiation intensity and parameters. The sensor array converts the detected infrared reflection intensity into a digital signal and generates a first reflection intensity distribution map. The generated first reflection intensity distribution map shows the reflection intensity of different areas of the soft package surface, and these data are added to the initial image data for subsequent image analysis.

[0032] After the negative pressure treatment is completed, the soft package will deform, causing changes in its surface optical properties. In order to ensure that the spectral conditions remain consistent during the deformation image acquisition process, the irradiation intensity, wavelength, and angle parameters of the infrared light source need to be dynamically adjusted. The purpose of adjustment is to improve the accuracy of image registration and feature extraction, thereby more accurately calculating the deformation vector.

[0033] For example, during the first stage of collecting initial image data, the soft package is under normal pressure, and the infrared light source is set to 50 watts, 850 nanometers, and 45 degrees. After the negative pressure treatment is completed, the soft package deforms, and the optical properties of its surface change (e.g., reflectivity, absorbance, etc.). To ensure that the spectral conditions remain consistent during the second stage of collecting deformation images, the parameters of the infrared light source need to be dynamically adjusted based on the comparison between the first and second reflectance intensity distribution maps. Suppose the first reflectance intensity distribution map shows that in certain areas of the soft package, the reflectance intensity has specific distribution characteristics under normal pressure. After the negative pressure treatment, the second reflectance intensity distribution map shows that the reflectance intensity in these areas has changed significantly. This may be due to changes in the optical properties of the soft package surface caused by deformation. Therefore, during the second stage, the infrared light source is adjusted to 60 watts, 880 nanometers, and 50 degrees. These adjustments compensate for the changes in optical properties caused by the deformation of the soft package, ensuring that the spectral conditions remain consistent during image acquisition under different pressure conditions. Then, the infrared sensor array is activated again, and the second reflectance intensity distribution map is collected, which shows the changes in reflectance intensity of the soft package under negative pressure. These data are added to the deformation image data for subsequent image analysis. This implementation dynamically adjusts the parameters of the infrared light source to compensate for changes in optical properties caused by the deformation of the soft package, ensuring that the spectral conditions remain consistent during image acquisition under different pressure conditions. This improves the accuracy of image registration and feature extraction, allowing for more accurate calculation of deformation vectors, ultimately achieving accurate detection of the soft package's sealing performance.

[0034] In step S300, the initial image data and the deformation image data are registered and aligned, the image feature difference data is extracted, and the deformation vector of the sealing area is calculated based on the image feature difference data.

[0035] Specifically, the initial image and the deformation image are registered and aligned using computer vision algorithms (such as feature point-based registration algorithms). For example, the SIFT (Scale-Invariant Feature Transform) algorithm can be used to extract feature points in the image, and the two images can be aligned through these feature points. The image feature difference data, such as gray level changes and edge intensity changes, is extracted. For example, an edge detection algorithm (such as the Canny algorithm) can be used to extract image edges, and then the differences between the two images are calculated. In combination with a pre-set deep learning model (such as a convolutional neural network, CNN), the extracted feature difference data is analyzed, and the deformation vector of the sealing area (a vector representing the degree of deformation of the object surface, used to describe the deformation of the soft package under different pressure conditions) is calculated. The deep learning model learns the deformation patterns of the soft package under different pressure conditions through training data.

[0036] For example, 100 feature points are extracted from the initial image and the deformed image using the SIFT algorithm, and the two images are aligned through these feature points. Then, the Canny algorithm is used to extract the image edges, and the difference in edge intensity between the two images is calculated. These difference data are input into a pre-trained CNN model, and the model outputs the deformation vector of the sealing area. It is assumed that the deformation vector is represented as a two-dimensional array, where each element corresponds to the deformation degree of a pixel point on the surface of the soft package.

[0037] In one possible implementation, the initial image data and the deformed image data are registered and aligned, and the image feature difference data are extracted, and step S300 further includes step S310 of extracting key feature points in the initial image data based on the first reflection intensity distribution map to establish a first feature point set. Specifically, a feature point detection algorithm in computer vision, such as scale-invariant feature transform (SIFT), speeded up robust features (SURF), or oriented FAST and rotated BRIEF (ORB), is used to extract key feature points from the first reflection intensity distribution map. These algorithms can detect significant feature points in the image, such as corner points, edge points, etc. A descriptor is generated for each detected feature point, which is a vector describing the image texture information around the feature point. For example, the SIFT algorithm is used to extract key feature points from the first reflection intensity distribution map. The SIFT algorithm detects extreme points through multi-scale space to find stable feature points. For each feature point, a 128-dimensional descriptor is generated. It is assumed that 100 key feature points are extracted from the first reflection intensity distribution map, and these feature points and their descriptors constitute the first feature point set.

[0038] Step S320, based on the second reflection intensity distribution map, the deformed image data is subjected to edge enhancement processing, and a second feature point set is extracted. Specifically, an edge enhancement algorithm, such as the Canny edge detection algorithm or the Sobel operator, is used to perform edge enhancement processing on the second reflection intensity distribution map. These algorithms can highlight the edge information in the image, making the feature points more obvious. On the edge-enhanced image, the feature point detection algorithm (such as SIFT, SURF, or ORB) is used again to extract key feature points and generate descriptors. For example, the Canny edge detection algorithm is used to perform edge enhancement processing on the second reflection intensity distribution map. The Canny algorithm detects the edges in the image through gradient calculation and non-maximum suppression. Then, the SIFT algorithm is used to extract key feature points from the edge-enhanced image. It is assumed that 120 key feature points are extracted from the edge-enhanced image, and these feature points and their descriptors constitute the second feature point set.

[0039] At step S330, feature point matching is performed based on the first set of feature points and the second set of feature points to determine a two-stage infrared image pair, which includes a spatial transformation matrix. Specifically, a feature point matching algorithm, such as nearest neighbor matching (NN) or bidirectional nearest neighbor matching (NN+NNR), is used to match feature points in the first set of feature points with feature points in the second set of feature points. The matching algorithm finds the most similar feature point pairs by comparing the descriptors of the feature points. Using the matched feature point pairs, a spatial transformation matrix is calculated using a least squares method or other optimization algorithm. The spatial transformation matrix is used to describe the geometric relationship between the two images. For example, 100 feature points in the first set of feature points are matched with 120 feature points in the second set of feature points using the bidirectional nearest neighbor matching algorithm. The matching algorithm finds the matched feature point pairs by calculating the Euclidean distance between the descriptors. Assuming that 80 pairs of feature points are matched, the spatial transformation matrix is calculated using the least squares method using these matched point pairs. The spatial transformation matrix is a 3x3 matrix that describes the geometric transformation relationship between the two images.

[0040] At step S340, affine transformation is performed on the initial image data and the deformed image data according to the spatial transformation matrix to determine the image feature difference data. Specifically, affine transformation is performed on the initial image data and the deformed image data using the spatial transformation matrix. Affine transformation is a geometric transformation that can eliminate image offset errors caused by pressure changes, making the two images spatially aligned. The aligned images are then pixel-by-pixel difference calculated to obtain the image feature difference data. The difference data can be represented as the gray value difference or feature vector difference of the corresponding pixel points of the two images. For example, affine transformation is performed on the initial image data using the calculated spatial transformation matrix to make it spatially aligned with the deformed image data. Then, pixel-by-pixel difference calculation is performed on the two aligned images. Assuming that the resolution of the initial image and the deformed image is 2048x1536 pixels, the gray value difference of the two images is calculated pixel by pixel to obtain a 2048x1536 difference image. This difference image reflects the deformation characteristics of the soft package under different pressure conditions. This implementation method realizes accurate alignment of the initial image and the deformed image and accurate extraction of the feature difference data through feature point extraction, edge enhancement, feature point matching, and affine transformation, thereby improving the reliability and accuracy of the soft package sealing performance detection.

[0041] In one possible implementation, the step S300 of calculating the deformation vector of the sealing area according to the image feature difference data further includes a step S350 of retrieving historical leak package sample data, traversing the historical leak package sample data to perform image deformation labeling, and obtaining a deformation abnormal area label of the sealing area. Specifically, historical leak package sample data is retrieved from a database, and the data includes images of known leak packages and corresponding deformation information thereof. The historical leak package sample data is labeled using an image labeling tool (such as LabelImg, VGG Image Annotator, etc.), and the labeling content includes the position and range of the deformation abnormal area. The historical leak package sample data is automatically labeled using a machine learning or deep learning algorithm (such as a convolutional neural network, CNN), and a deformation abnormal area label is generated. For example, 1000 images of known leak packages and corresponding deformation information thereof are retrieved from a database. The images are manually labeled using a LabelImg tool, and the deformation abnormal area in each leak package image is labeled. At the same time, the images are automatically labeled using a pre-trained CNN model, and a deformation abnormal area label is generated. It is assumed that the labeling result shows that there are obvious deformation abnormalities in the edge area and the center area of some images, and these areas are marked as the deformation abnormal area label.

[0042] In step S360, the two-stage infrared image pair is convolved according to the deformation abnormal area label to construct a pixel-level deformation probability map. Specifically, the two-stage infrared image pair is convolved using a pre-trained CNN model. The CNN model extracts features in the image through convolution layers, pooling layers, and fully connected layers, and calculates the deformation probability of each pixel. According to the output of the CNN model, a pixel-level deformation probability map is generated. The deformation probability value of each pixel represents the likelihood of deformation of the pixel point. For example, a pre-trained CNN model (such as U-Net) is used to convolve the two-stage infrared image pair. It is assumed that the resolution of the input two-stage infrared image pair is 2048x1536 pixels, and the CNN model outputs a pixel-level deformation probability map with the same resolution. In the deformation probability map, the value of each pixel ranges from 0 to 1, indicating the probability of deformation of the pixel point. For example, the deformation probability value of some pixel points is 0.9, indicating that these pixel points have a high likelihood of deformation.

[0043] At step S370, the deformation of the sealed area is calculated according to the pixel-level deformation probability map to generate a deformation vector of the sealed area. Specifically, the deformation of the sealed area is calculated using a deformation calculation algorithm (such as an optical flow method, a block matching method, etc.) according to the pixel-level deformation probability map. These algorithms calculate the displacement vector of each pixel point by analyzing the pixel-level deformation probability map. The calculated pixel point displacement vectors are summarized to generate the deformation vector of the sealed area. The deformation vector represents the overall deformation of the soft package under different pressure conditions. For example, the optical flow method is used to calculate the deformation of the pixel-level deformation probability map. Assuming that the deformation probability values of some areas in the deformation probability map are relatively high, the optical flow method calculates the displacement vectors of the pixel points in these areas. These displacement vectors are summarized to generate the deformation vector of the sealed area. Assuming that the deformation vector shows that the maximum deformation of the soft package under negative pressure occurs in the central area, and the deformation degree is 10%. This implementation method realizes accurate analysis of the deformation of the soft package and generation of the deformation vector by calling historical leaky package sample data, image labeling, convolution calculation, and deformation calculation, which can improve the accuracy and reliability of detection and help more accurately determine whether the soft package has a leak.

[0044] At step S400, a soft package surface deformation distribution map is constructed according to the deformation vector to perform sealing judgment and determine a leaky package detection result.

[0045] Specifically, a soft package surface deformation distribution map is constructed according to the calculated deformation vector. The areas with a larger deformation degree can be represented by red, and the areas with a smaller deformation degree can be represented by blue in the form of a heat map. Sealing judgment is performed based on a preset deformation threshold rule. The deformation threshold rule is a rule for judging the sealing of the soft package, and when the deformation degree exceeds the threshold, it is determined that the soft package has a leak. For example, if the deformation degree of a certain area exceeds the preset threshold (such as 10%), it is determined that the area may have a leak. For example, the heat map constructed according to the deformation vector shows that the deformation degree of most areas on the surface of the soft package is below 5% (blue area), but in a corner of the soft package, the deformation degree reaches 15% (red area). According to the preset deformation threshold (10%), it is determined that the corner area has a leak, and finally it is determined that the soft package is a leaky package. The embodiments of the present application use the technical means of packaging the soft package into a vacuum chamber, first collecting an initial infrared image under normal pressure, then vacuumizing to a set negative pressure, collecting an infrared image after deformation, comparing the two images, calculating the deformation of the sealed area, judging the sealing according to the deformation distribution map, and obtaining a leaky package detection result, which achieves the technical effects of improving the sealing detection efficiency and accuracy.

[0046] In a possible implementation, the soft package surface deformation distribution map is constructed according to the deformation vector to determine the sealing and the leakage detection result, and step S400 further includes step S410 of constructing a soft package surface coordinate system, mapping the deformation vector to the soft package surface coordinate system, performing normalization processing in combination with the soft package material characteristics, and constructing the soft package surface deformation distribution map. Specifically, a two-dimensional or three-dimensional coordinate system is established on the soft package surface, which is used to accurately locate the position of the deformation vector. The calculated deformation vector is mapped to the soft package surface coordinate system, and each deformation vector corresponds to a point on the soft package surface. According to the characteristics (such as the elastic modulus and the Poisson's ratio) of the soft package material, the deformation vector is normalized to make the deformation data comparable. The normalized deformation vector data is visualized to construct the soft package surface deformation distribution map, and the deformation degree is represented in the form of a heat map.

[0047] For example, it is assumed that the soft package is a rectangular package with a length of 20 cm and a width of 10 cm. A two-dimensional coordinate system is established on the soft package surface, and the coordinate origin is located at the lower left corner of the soft package. The calculated deformation vector is mapped to the coordinate system, and the starting point of each deformation vector corresponds to a point on the soft package surface. According to the elastic modulus (assuming 1.5 GPa) and the Poisson's ratio (assuming 0.3) of the soft package material, the deformation vector is normalized. The normalized deformation vector data is visualized to generate a soft package surface deformation distribution map in the form of a heat map, in which red represents a higher deformation degree and blue represents a lower deformation degree.

[0048] Step S420, data extraction is performed by traversing the soft package surface deformation distribution map, region segmentation is performed according to a plurality of deformation data sets, and a suspicious leakage region set is generated. Specifically, the deformation data of each pixel point is extracted by traversing the soft package surface deformation distribution map. An image segmentation algorithm (such as threshold segmentation, region growing, and dam algorithm) is used to perform region segmentation on the deformation distribution map to identify regions with a higher deformation degree. The segmented high-deformation regions are marked as suspicious leakage regions to generate a suspicious leakage region set. For example, a threshold segmentation algorithm is used to perform region segmentation on the soft package surface deformation distribution map. It is assumed that the set threshold is 0.5 (normalized deformation degree), and all regions with a deformation degree greater than 0.5 are identified as suspicious leakage regions. By traversing the deformation distribution map, three regions with a higher deformation degree are extracted, which are marked as suspicious leakage regions to generate a suspicious leakage region set.

[0049] Step S430, the set of suspicious leakage regions is morphologically optimized, morphological data is used for morphing analysis to obtain morphing vector amplitude. Specifically, morphological image processing techniques (such as dilation, erosion, opening operation, closing operation, etc.) are used to optimize the suspicious leakage region, and remove noise and small interference regions. Morphing analysis is performed on the optimized suspicious leakage region to calculate the morphing vector amplitude of each region. For example, the set of suspicious leakage regions is morphologically optimized, and the closing operation is used to remove small holes in the region, and the opening operation is used to smooth the boundary of the region. After optimization, the morphology of each suspicious leakage region is more clear. Morphing analysis is performed on the optimized region to calculate the morphing vector amplitude of each region. Assuming that the morphing vector amplitude of the first suspicious leakage region after optimization is 0.8, the second region is 0.6, and the third region is 0.7.

[0050] Step S440, the morphing vector amplitude is used as an index to search and match the historical leak sample data, generate a leak matching degree, and add the leak matching degree to the leak detection result. Specifically, a feature-based search and matching algorithm (such as nearest neighbor matching, cosine similarity matching, etc.) is used to match the morphing vector amplitude with the historical leak sample data. According to the matching result, a leak matching degree is generated to represent the similarity between the current soft package and the historical leak sample. The leak matching degree is added to the leak detection result to provide a reference for the final sealing judgment. For example, the morphing vector amplitude (0.8, 0.6, 0.7) of the optimized suspicious leakage region is used as an index to search and match the historical leak sample data. Assuming that in the historical leak sample data, the sample matching degree of the morphing vector amplitude of 0.8 is 0.9, the sample matching degree of the morphing vector amplitude of 0.6 is 0.7, and the sample matching degree of the morphing vector amplitude of 0.7 is 0.8. These matching degrees are integrated, and the highest matching degree 0.9 is taken as the final result. This implementation realizes accurate detection of soft package sealing by constructing a morphing distribution map, identifying suspicious leakage regions, optimizing the morphology of the leakage region, quantifying the morphing degree, and searching and matching historical sample data.

[0051] In a possible implementation, the step S440 further includes a step S441 of grid segmentation on the soft package surface deformation distribution map to obtain a plurality of detection units. Specifically, the soft package surface deformation distribution map is divided into a plurality of grid units of equal size using an image segmentation technique. Each grid unit is an independent detection unit for subsequent deformation analysis. According to a preset grid size, the deformation distribution map is segmented into a plurality of detection units, and each detection unit contains a certain number of pixel points. For example, assuming that the resolution of the soft package surface deformation distribution map is 2048*1536 pixels. The image is segmented into 128*128 pixel grid units, and each grid unit is a detection unit. In this way, the entire deformation distribution map is segmented into 16*12 detection units.

[0052] In the step S442, the plurality of detection units are calculated in combination with the deformation vector amplitudes to generate a multi-dimensional feature vector, and the multi-dimensional feature vector is analyzed and reduced in dimension to generate a deformation amplitude feature sequence. Specifically, for each detection unit, the average amplitude, the maximum amplitude, and the amplitude standard deviation of the deformation vector are calculated. These statistics can reflect the deformation characteristics in the detection unit. The calculated statistics are combined into a multi-dimensional feature vector. A dimension reduction algorithm (such as principal component analysis PCA, t-SNE, etc.) is used to analyze and reduce the multi-dimensional feature vector in dimension to generate a deformation amplitude feature sequence. The reduced feature sequence can be more effectively used for subsequent similarity retrieval. For example, for each detection unit, the average amplitude of the deformation vector is 0.5, the maximum amplitude is 0.8, and the amplitude standard deviation is 0.2. These values are combined into a three-dimensional feature vector (0.5, 0.8, 0.2). The multi-dimensional feature vectors of all detection units are reduced in dimension by PCA to generate a two-dimensional deformation amplitude feature sequence. Assuming that the reduced feature sequence is (0.6, 0.7).

[0053] Step S443, using the deformation amplitude feature sequence as an index, traversing the historical leak sample database to perform similarity search, and generating a leak matching degree. Specifically, using a feature-based similarity search algorithm (such as nearest neighbor matching, cosine similarity, etc.), the deformation amplitude feature sequence of the current detection unit is matched with the feature sequence in the historical leak sample database. According to the matching result, the leak matching degree of each detection unit is calculated. The higher the matching degree, the more similar the current detection unit is to the historical leak sample, and the higher the possibility of leak. For example, assuming that there are 1000 samples in the historical leak sample database, each sample has a deformation amplitude feature sequence. Using the cosine similarity algorithm, the deformation amplitude feature sequence (0.6, 0.7) of the current detection unit is matched with the historical sample. Assuming that the matching degree of the sample with the highest matching degree is 0.85, this matching degree is taken as the leak matching degree of the current detection unit.

[0054] Step S444, comparing the leak matching degree with a dynamic threshold, determining the leakage risk level according to the comparison result, integrating the leak matching degree and the leakage risk level into the leak detection result. Specifically, according to historical data and detection requirements, a dynamic threshold is set. The dynamic threshold can be adjusted according to the detection environment and the characteristics of the soft package material. The leak matching degree is compared with the dynamic threshold, and the leakage risk level is determined according to the comparison result. For example, the detection unit with a matching degree higher than the threshold is determined as high risk, the detection unit with a matching degree lower than the threshold but close to the threshold is determined as medium risk, and the detection unit with a matching degree far lower than the threshold is determined as low risk. The leak matching degree and the leakage risk level are integrated into the leak detection result, and a detailed detection report is output. For example, assuming that the dynamic threshold is set to 0.8. For the detection unit with a matching degree of 0.85, it is determined as high risk; for the detection unit with a matching degree of 0.75, it is determined as medium risk; for the detection unit with a matching degree of 0.6, it is determined as low risk. These matching degrees and risk levels are integrated into the leak detection result, and a detailed detection report is output, including the leakage position, the matching degree and the risk level. This implementation mode realizes the fine detection and risk assessment of the soft package surface deformation distribution map through grid segmentation, feature extraction and dimension reduction, similarity search and risk assessment.

[0055] In a possible implementation, the leak package matching degree is compared with a dynamic threshold, and step S444 further includes step S4441 of querying a reference matching degree threshold table according to the soft package material characteristics in combination with detection of environmental humidity. Specifically, a database containing different soft package material characteristics (such as elastic modulus, Poisson's ratio, material thickness, and the like) is established. A humidity sensor is installed in the detection environment to monitor the environmental humidity in real time during detection. According to the soft package material characteristics and the detection environmental humidity, a corresponding reference matching degree threshold table is queried from the database. The table stores the matching degree threshold values under different materials and humidity conditions. For example, it is assumed that the soft package material is polyethylene, the elastic modulus is 1.5 GPa, the Poisson's ratio is 0.35, and the material thickness is 0.1 mm. The detection environmental humidity is 60% RH. By querying the material characteristic database, the reference matching degree threshold table of the polyethylene material under the 60% RH humidity condition is found, and the table shows that the reference matching degree threshold value is 0.75.

[0056] Step S4442, the real-time pressure change rate is introduced to linearly interpolate and compensate the reference matching degree threshold table to generate a dynamic threshold. Specifically, a pressure sensor is installed in the vacuumizing cavity to monitor the pressure change in real time. The real-time pressure change rate, that is, the change amount of the pressure per unit time, is calculated. According to the pressure change rate, the reference matching degree threshold is linearly interpolated and compensated to generate a dynamic threshold. The compensation formula can be expressed as: dynamic threshold = reference matching degree threshold + k x pressure change rate, wherein k is a compensation coefficient determined according to experimental data. For example, it is assumed that the reference matching degree threshold is 0.75, the real-time pressure change rate is 0.1 kPa / s, and the compensation coefficient k is 0.05. According to the compensation formula: dynamic threshold = 0.75 + 0.05 x 0.1 = 0.755, therefore, the generated dynamic threshold is 0.755.

[0057] Step S4443, when the leak package matching degree exceeds the dynamic threshold, a high-confidence leakage alarm is triggered, and the comparison result is generated. Specifically, the leak package matching degree is compared with the dynamic threshold to determine whether the threshold is exceeded. If the leak package matching degree exceeds the dynamic threshold, a high-confidence leakage alarm is triggered. On the basis of the high-confidence leakage alarm, a detailed report is generated through further comprehensive determination, which is used to confirm the specific position, matching degree, and risk level of the leak package. The comparison result generation step is as follows: according to historical data, the average deformation vector amplitude of a normally sealed soft package under negative pressure conditions is counted and set as a first determination threshold. According to the deformation distribution characteristics of the leakage sample, the maximum curvature parameter of the abnormal deformation region is extracted and set as a second determination threshold. If there is a continuous region in the deformation distribution diagram that exceeds both the first determination threshold and the second determination threshold, the leak package is determined.

[0058] For example, assuming that the packet leakage matching degree is 0.8 and the dynamic threshold is 0.755. Since 0.8>0.755, a high-confidence leakage alarm is triggered. Further analysis of the deformation distribution map: the average deformation vector amplitude of a normal sealed soft packet under negative pressure conditions is 0.5 (the first determination threshold). The maximum curvature parameter of the abnormal deformation region is 0.9 (the second determination threshold). In the deformation distribution map, it is found that the deformation vector amplitude of a continuous region is 0.6 (exceeding the first determination threshold 0.5), and the maximum curvature parameter of the region is 1.0 (exceeding the second determination threshold 0.9). The comprehensive determination result is a leaky packet, and a detailed comparison result report is generated, including information such as the leakage position (the region with a deformation vector amplitude of 0.6 in the deformation distribution map), the matching degree (0.8), the risk level (high), and the determination result (leaky packet). This implementation mode realizes accurate detection of soft packet sealing by dynamically adjusting the threshold, triggering a high-confidence leakage alarm, and comprehensive determination, which can improve the adaptability, reliability and accuracy of detection, help to more accurately judge whether the soft packet has a gas leakage problem, and provide a detailed detection report.

[0059] In the foregoing, reference is made to Figure 1 The soft packet sealing detection method based on image recognition according to the embodiments of the present application is described in detail. Next, the soft packet sealing detection system based on image recognition according to the embodiments of the present application will be described with reference to Figure 2 The soft packet sealing detection system based on image recognition according to the embodiments of the present application is described in detail. Next, the soft packet sealing detection system based on image recognition according to the embodiments of the present application will be described with reference to

[0060] The soft packet sealing detection system based on image recognition according to the embodiments of the present application is used to solve the technical problems of low efficiency and accuracy of existing soft packet sealing detection, and achieves the technical effect of improving the sealing detection efficiency and accuracy. The soft packet sealing detection system based on image recognition comprises an initial image data acquisition module 10, a deformation image data acquisition module 20, an image feature difference data extraction module 30, and a sealing determination module 40.

[0061] The initial image data acquisition module 10 is used to place a soft packet to be tested into a vacuum cavity, perform a first-stage infrared radiation under a first pressure condition, and acquire initial image data; the deformation image data acquisition module 20 is used to perform negative pressure processing on the vacuum cavity to a set pressure threshold, perform a second-stage infrared radiation under a second pressure condition, and acquire deformation image data; the image feature difference data extraction module 30 is used to register and align the initial image data and the deformation image data, extract image feature difference data, and calculate a deformation vector of a sealing region according to the image feature difference data; and the sealing determination module 40 is used to construct a soft packet surface deformation distribution map according to the deformation vector for sealing determination, and determine a leaky packet detection result.

[0062] The system further comprises an image acquisition unit used when performing the first-stage infrared irradiation and the second-stage infrared irradiation. The image acquisition unit can further comprise: an infrared light source irradiation intensity setting subunit for arranging a plurality of infrared sensor arrays in the vacuum cavity, setting the irradiation intensity of the infrared light source according to the soft package material characteristics, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a first reflection intensity distribution map; a data adding subunit for adding the first reflection intensity distribution map to the initial image data; a second-stage infrared light source irradiation intensity adjustment subunit for dynamically adjusting the irradiation intensity of the second-stage infrared light source based on the first reflection intensity distribution map after the negative pressure treatment is completed, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a second reflection intensity distribution map; and a data adding subunit for adding the second reflection intensity distribution map to the deformed image data.

[0063] In the following, the specific configuration of the image feature difference data extraction module 30 will be described in detail. As described above, the initial image data and the deformed image data are registered and aligned, and the image feature difference data is extracted. The image feature difference data extraction module 30 can further comprise: a first feature point set establishment unit for extracting key feature points in the initial image data based on the first reflection intensity distribution map, and establishing a first feature point set; a second feature point set extraction unit for performing edge enhancement processing on the deformed image data based on the second reflection intensity distribution map, and extracting a second feature point set; a feature point matching unit for performing feature point matching based on the first feature point set and the second feature point set, determining a two-stage infrared image pair, and the two-stage infrared image pair containing a spatial transformation matrix; and an affine transformation unit for performing affine transformation on the initial image data and the deformed image data according to the spatial transformation matrix, and determining the image feature difference data.

[0064] The image feature difference data is used to calculate the deformation vector of the sealed area. The image feature difference data extraction module 30 can further comprise: a historical leak package sample data calling unit for calling historical leak package sample data, traversing the historical leak package sample data to perform image deformation labeling, and obtaining a deformation abnormal area label of the sealed area; a pixel-level deformation probability map construction unit for performing convolution calculation on the two-stage infrared image pair according to the deformation abnormal area label, and constructing a pixel-level deformation probability map; and a deformation calculation unit for performing deformation calculation on the sealed area according to the pixel-level deformation probability map, and generating a deformation vector of the sealed area.

[0065] The specific configuration of the sealing determination module 40 will be described in detail below. As described above, the sealing determination is performed by constructing a soft package surface deformation distribution map according to the deformation vector, the leak detection result is determined, and the sealing determination module 40 can further include: a soft package surface deformation distribution map construction unit for constructing a soft package surface coordinate system, mapping the deformation vector to the soft package surface coordinate system, performing normalization processing in combination with the soft package material characteristics, and constructing the soft package surface deformation distribution map; a region segmentation unit for traversing the soft package surface deformation distribution map to extract data, performing region segmentation according to a plurality of deformation data sets, and generating a suspicious leakage region set; a morphology optimization unit for performing morphology optimization on the suspicious leakage region set, performing deformation analysis according to optimized morphology data, and obtaining a deformation vector amplitude; and a retrieval matching unit for taking the deformation vector amplitude as an index, performing retrieval matching on the historical leak sample data, generating a leak matching degree, and adding the leak matching degree to the leak detection result.

[0066] The retrieval matching unit can further include: a gridding segmentation subunit for performing gridding segmentation on the soft package surface deformation distribution map to obtain a plurality of detection units; an analysis dimension reduction subunit for calculating the plurality of detection units in combination with the deformation vector amplitude to generate a multi-dimensional feature vector, performing analysis dimension reduction on the multi-dimensional feature vector, and generating a deformation amplitude feature sequence; a similarity retrieval subunit for taking the deformation amplitude feature sequence as an index, traversing the historical leak sample database to perform similarity retrieval, and generating a leak matching degree; and a leakage risk level determination subunit for comparing the leak matching degree with a dynamic threshold, determining a leakage risk level according to a comparison result, and integrating the leak matching degree and the leakage risk level into the leak detection result.

[0067] The leakage risk level determination subunit can further include: a reference matching degree threshold table acquisition component for querying a reference matching degree threshold table according to the soft package material characteristics in combination with a detection environment humidity to obtain the reference matching degree threshold table; a linear interpolation compensation component for introducing a real-time pressure change rate to perform linear interpolation compensation on the reference matching degree threshold table to generate a dynamic threshold; and a comparison result generation component for triggering a high-confidence leakage alarm and generating the comparison result when the leak matching degree exceeds the dynamic threshold.

[0068] The soft package sealing detection system based on image recognition provided in the embodiments of the present application can perform the soft package sealing detection method based on image recognition provided in any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method.

[0069] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or the server, the various units and modules included are only divided according to the functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific name of each functional unit is only for the convenience of mutual differentiation, and does not serve to limit the protection scope of the present application.

[0070] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application. In some cases, the actions or steps described in the present application can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

Claims

1. A soft pack tightness detection method based on image recognition, characterized in that, The method comprises: placing a soft package to be tested into a vacuum cavity, performing first-stage infrared irradiation under a first pressure condition, and collecting initial image data; performing negative pressure treatment on the vacuum cavity to a set pressure threshold, performing second-stage infrared irradiation under a second pressure condition, and collecting deformation image data; aligning the initial image data and the deformation image data, extracting image feature difference data, and calculating a deformation vector of a sealed area according to the image feature difference data; constructing a soft package surface deformation distribution map according to the deformation vector to determine a sealed area, and determining a leaky package detection result; the collection process of the first-stage infrared irradiation and the second-stage infrared irradiation comprises: arranging a plurality of infrared sensor arrays in the vacuum cavity, setting the irradiation intensity of an infrared light source according to the material characteristics of the soft package, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a first reflection intensity distribution map; adding the first reflection intensity distribution map to the initial image data; after the negative pressure treatment is completed, dynamically adjusting the irradiation intensity of a second-stage infrared light source based on the first reflection intensity distribution map, activating the plurality of infrared sensor arrays to collect data according to the irradiation intensity, and obtaining a second reflection intensity distribution map; adding the second reflection intensity distribution map to the deformation image data.

2. The image recognition-based soft pack tightness detection method of claim 1, wherein, aligning the initial image data and the deformation image data, extracting image feature difference data, and the method comprises: extracting key feature points in the initial image data based on the first reflection intensity distribution map, and establishing a first feature point set; performing edge enhancement processing on the deformation image data based on the second reflection intensity distribution map, and extracting a second feature point set; performing feature point matching based on the first feature point set and the second feature point set, determining a two-stage infrared image pair, and the two-stage infrared image pair contains a spatial transformation matrix; performing affine transformation on the initial image data and the deformation image data according to the spatial transformation matrix, and determining the image feature difference data.

3. The image recognition-based soft pack tightness detection method of claim 2, wherein, calculating a deformation vector of a sealed area according to the image feature difference data, and the method comprises: calling historical leaky package sample data, traversing the historical leaky package sample data to perform image deformation labeling, and obtaining a deformation abnormal area label of a sealed area; performing convolution calculation on the two-stage infrared image pair according to the deformation abnormal area label, and constructing a pixel-level deformation probability map; performing deformation calculation on the sealed area according to the pixel-level deformation probability map, and generating a deformation vector of the sealed area.

4. The image recognition-based soft pack tightness detection method of claim 3, wherein, constructing a soft package surface deformation distribution map according to the deformation vector to determine a sealed area, and determining a leaky package detection result, and the method comprises: constructing a soft package surface coordinate system, mapping the deformation vector to the soft package surface coordinate system, normalizing the deformation vector in combination with the material characteristics of the soft package, and constructing the soft package surface deformation distribution map; traversing the soft package surface deformation distribution map to extract data, performing region segmentation according to a plurality of deformation data sets, and generating a suspicious leakage area set; Performing morphological optimization on the suspicious leakage area set, performing morphing analysis according to the optimized morphological data to obtain a morphing vector amplitude; Taking the morphing vector amplitude as an index, searching and matching the historical leakage sample data to generate a leakage matching degree, and adding the leakage matching degree to the leakage detection result.

5. The image recognition-based soft pack tightness detection method of claim 4, wherein, Taking the morphing vector amplitude as an index, searching and matching the historical leakage sample data to generate a leakage matching degree, and adding the leakage matching degree to the leakage detection result, the method comprising: Grid segmentation is performed on the soft package surface morphing distribution map to obtain a plurality of detection units; A multi-dimensional feature vector is generated by calculating the plurality of detection units in combination with the morphing vector amplitude, and a morphing amplitude feature sequence is generated by analyzing and reducing the multi-dimensional feature vector; Taking the morphing amplitude feature sequence as an index, the historical leakage sample data is traversed for similarity search to generate a leakage matching degree; The leakage matching degree is compared with a dynamic threshold, and a leakage risk level is determined according to the comparison result, and the leakage matching degree and the leakage risk level are integrated into the leakage detection result.

6. The image recognition-based soft pack tightness detection method of claim 5, wherein, The method for comparing the leakage matching degree with a dynamic threshold comprises: According to the characteristics of the soft package material, the detection environment humidity is queried to obtain a reference matching degree threshold table; A real-time pressure change rate is introduced to perform linear interpolation compensation on the reference matching degree threshold table to generate a dynamic threshold; When the leakage matching degree exceeds the dynamic threshold, a high-confidence leakage alarm is triggered, and the comparison result is generated.

7. The soft pack tightness detection system based on image recognition, characterized in that, The system is used to implement the image recognition-based soft package sealing detection method of any one of claims 1-6, and the system comprises: An initial image data acquisition module is configured to place a soft package to be tested in a vacuum chamber, perform first-stage infrared irradiation under a first pressure condition, and acquire initial image data; A deformation image data acquisition module is configured to perform negative pressure processing on the vacuum chamber to a set pressure threshold, perform second-stage infrared irradiation under a second pressure condition, and acquire deformation image data; An image feature difference data extraction module is configured to align the initial image data and the deformation image data, extract image feature difference data, and calculate a morphing vector of a sealing area according to the image feature difference data; A sealing determination module is configured to construct a soft package surface morphing distribution map according to the morphing vector to determine a leakage detection result.

Images