Method and system for intelligently collecting frosted granularity on surface of PET (Polyethylene Terephthalate) film

Abstract

The invention discloses an intelligent collection method and system for the frosted granularity of the surface of a PET film, and relates to the field of PET films. The method specifically comprises the following steps: acquiring film acquisition images of a plurality of PET films under different brightness irradiation; identifying the contrast between particles on the surface of the film and the substrate through the film acquisition images; judging whether the contrast ratio of each film acquisition image meets a preset standard contrast ratio threshold value or not under the change of the minimum brightness unit; if yes, taking the corresponding brightness change interval as the optimal brightness range of the film acquisition image; and re-collecting the PET film image according to the optimal brightness range. According to the invention, different PET film surface images can be acquired under proper brightness, so that the accuracy of image recognition of frosted granularity is improved, and high-precision processing control of the PET film is facilitated.

Description

Technical Field

[0001] The invention relates to the field of PET films, and in particular to a method and system for intelligently collecting frosted particle size on the surface of a PET film. Background Art

[0002] As a high-performance polymer material, polyethylene terephthalate (PET) film has been widely used in many fields such as packaging, electronics, optics, and medical treatment due to its excellent mechanical properties, chemical stability, optical transparency, and good barrier properties. In the field of food packaging, PET film can effectively block oxygen, moisture, and microorganisms, extending the shelf life of food. At the same time, its good printing adaptability and glossiness can enhance the appearance of the product. In the electronics industry, PET film is used as a substrate for flexible circuit boards, packaging materials for electronic components, and protective films for display screens. With its insulation properties, high temperature resistance, and flexibility, it meets the manufacturing needs of precision electronic devices. In the field of optics, PET film can be prepared into functional optical films such as polarizers and brightness enhancement films, providing key optical performance support for optoelectronic devices such as liquid crystal displays.

[0003] However, the surface quality of PET film plays a decisive role in its performance and application effect. The presence of surface particles, whether it comes from impurities in the raw materials, agglomeration of additives during processing, or wear debris from production equipment, will cause many problems. These particles may cause the optical properties of the film to deteriorate, such as light scattering and increased haze in optical film applications, affecting display clarity; in the packaging field, the appearance quality of the product is reduced, the printing effect is deteriorated, and the gloss is uneven. At the same time, the particles will also weaken the mechanical properties of the film, become a stress concentration point, reduce the tensile strength and impact resistance of the film, and easily cause crack propagation when subjected to external forces, eventually leading to film rupture.

[0004] Accurately identifying the particle and substrate features on the surface of PET film, and then achieving precise control of the film surface quality, has become a key link in the production and application of PET film. In this identification process, image acquisition and processing technology plays a core role. By acquiring image information on the surface of PET film and analyzing it using image processing algorithms, the microscopic features of the surface can be converted into intuitive quantitative data. Among them, the brightness range of the image, as a key parameter, has a significant impact on the highlighting effect of particle and substrate features. Different brightness settings will change the grayscale distribution of pixels in the image, thereby affecting the contrast between particles and substrates. The appropriate brightness range can enhance the visual difference between particles and substrates, making particles more clearly distinguishable in the image, providing strong support for subsequent feature extraction and quality assessment; on the contrary, if the brightness range is not properly selected, particles may be submerged in the substrate background, resulting in misjudgment or missed detection, seriously affecting the accuracy and reliability of surface quality detection. The contrast between particles and substrates on the surface of PET film is first formed due to their significant differences in materials. PET, as a base material, is a highly crystalline polymer. Its molecular chain is formed by esterification of terephthalic acid and ethylene glycol. The molecules are arranged in a regular and orderly manner to form a tight lattice structure. This structure gives the PET base good mechanical properties, chemical stability and optical transparency. The composition of the surface particles is more complex and diverse, which may come from impurities in the raw materials, agglomeration of additives during processing, and debris generated by equipment wear. Common particulate impurities include silicon dioxide (SiO2), calcium carbonate (CaCO3), metal oxides (such as Fe2O3, TiO2, etc.) and organic polymer particles. Taking silicon dioxide as an example, it has a crystal structure similar to quartz, and silicon atoms and oxygen atoms are covalently bonded in the form of tetrahedrons to form a highly stable three-dimensional network structure. This structure makes the silicon dioxide particles have high hardness and strong chemical stability, which is in sharp contrast to the flexible polymer chain structure of the PET base. In terms of optical properties, the refractive index of silica is about 1.46, while the refractive index of PET is usually between 1.57 - 1.64. The difference in refractive index between the two leads to completely different propagation behaviors of light at the interface between particles and substrates. Calcium carbonate particles usually exist in the form of calcite or aragonite, and their crystal structure is composed of calcium ions and carbonate ions, which have the characteristics of ionic bonds. Compared with the covalently bonded polymer structure of PET, calcium carbonate particles are relatively chemically active and are prone to dissolution reactions in some acidic environments. Optically, the refractive index of calcium carbonate is about 1.58 - 1.66, which is similar to the refractive index of PET, but due to the anisotropy of its crystal structure, the propagation path of light inside the particles is more complicated and the scattering phenomenon is more obvious. Metal oxide particles, such as Fe2O3, have a typical ionic crystal structure, where iron ions and oxygen ions are bound by ionic bonds to form a tightly packed crystal lattice.These metal oxide particles tend to have high density and magnetism, in contrast to the low density and non-magnetic properties of the PET substrate. In terms of optical properties, Fe2O3 has a high refractive index of about 2.94 - 3.22, and has a strong ability to absorb light, especially in the long-wave region of visible light, which makes the light transmittance of the surface of the PET film containing such particles significantly decrease in this band, showing different color and brightness characteristics from the substrate. Organic polymer particles are usually formed by the aggregation of additives (such as antioxidants, lubricants, plasticizers, etc.) added during the processing under specific conditions. The molecular structure of these organic polymer particles is similar to that of PET, but due to the different types of additives and polymerization degrees, their chemical properties and physical properties are different. For example, some low molecular weight organic polymer particles may have a lower glass transition temperature and appear softer at room temperature, in contrast to the rigidity of the PET substrate. In terms of optical properties, the refractive index and absorptivity of organic polymer particles depend on the functional groups in their molecular structure. For example, polymers containing benzene ring structures usually have a higher refractive index, while polymers containing electron-withdrawing functional groups such as carbonyl groups may have a stronger absorption capacity for light of a specific wavelength. These material differences, from chemical composition, crystal structure to intermolecular forces, lead to significant differences in the optical behaviors of particles and substrates in terms of light reflection, refraction, absorption, etc., which in turn lays the foundation for the formation of contrast during image acquisition.

[0005] At present, there is a need for an intelligent method and system for collecting the frosted particle size on the surface of a PET film, which can capture images of different PET film surfaces at appropriate brightness, thereby improving the accuracy of image recognition of frosted particle size, and thus facilitating high-precision processing control of the PET film. Summary of the invention

[0006] The technical problem to be solved by the present invention is how to select appropriate ambient brightness to capture images of different PET film surfaces, thereby improving the accuracy of image recognition of frosted particle size, which is beneficial to high-precision processing control of PET films. The purpose is to provide a method and system for intelligently collecting frosted particle size on the surface of PET films to solve the above problems.

[0007] The present invention is achieved through the following technical solutions:

[0008] A method for intelligently collecting frosted granularity of a PET film surface comprises: collecting film collection images of a plurality of PET films under illumination of different brightness;

[0009] Collecting images of each of the films to identify the contrast between particles on the film surface and the substrate;

[0010] Determine whether the contrast of each of the film captured images meets a preset standard contrast threshold under the change of the minimum brightness unit; if so, use the corresponding brightness change interval as the optimal brightness range of the film captured image;

[0011] The PET film image is re-collected according to the optimal brightness range.

[0012] After the PET film image is re-collected according to the optimal brightness range, the method further includes: collecting multiple groups of film image training data; each group of the film image training data includes the film collection image and the PET film image collected by the film collection image under the optimal brightness range;

[0013] A plurality of sets of the film image training data are trained through machine learning to obtain a film image training model; and the film image training model is used to output a plurality of PET film images of the films to be tested.

[0014] After the PET film images are re-collected according to the optimal brightness range, the method further includes: identifying the particle features and substrate features of each output PET film image according to the image recognition technology; obtaining the particle image according to the pre-trained film image segmentation model;

[0015] Identify the particle area uniformity and particle distribution uniformity of the particle image; and adjust the processing parameters of the PET film according to the particle area uniformity and particle distribution uniformity.

[0016] The step of adjusting the processing parameters of the PET film according to the uniformity of the particle area and the uniformity of the particle distribution comprises:

[0017] Determine whether the particle distribution uniformity is lower than the preset standard particle distribution uniformity. If so, reduce the film travel speed during sandblasting of the next PET film particles until the preset standard particle area uniformity is met; determine whether the particle area uniformity is lower than the preset standard particle area uniformity. If so, reduce the sanding roller speed during grinding of the next PET film particles or the film travel speed during grinding until the preset standard particle area uniformity is met.

[0018] The method for intelligently collecting the frosted particle size of the surface of a PET film also includes: collecting multiple groups of film processing training data; each group of the film processing training data includes output images of each PET film, the film travel speed when the PET film particles are sandblasted, and the frosted roller speed when the PET film particles are polished / the film travel speed during polishing; and obtaining a film processing training model through machine learning of the multiple groups of film processing training data.

[0019] The method for intelligently collecting the frosted particle size of the surface of a PET film also includes: obtaining the film travel speed of each film to be tested during sandblasting and the frosted roller speed / film travel speed during polishing through the film processing training model; and processing the film to be tested using the output film travel speed during sandblasting and the frosted roller speed / film travel speed during polishing.

[0020] The film image segmentation model is obtained by the following steps:

[0021] Collect multiple sets of film image segmentation data; each set of the film image segmentation data includes a particle image and its particle features, a substrate image and its substrate features; and obtain the film image segmentation model through training of the multiple sets of the film image segmentation data.

[0022] The particle characteristics include any one or more characteristics of particle area, shape, brightness, and color; the substrate characteristics include any one or more characteristics of substrate area, shape, brightness, and color.

[0023] A system applied to any one of the methods for intelligently collecting the frosted particle size of a PET film surface, comprising:

[0024] Image acquisition module: used to acquire film acquisition images of multiple PET films under different brightness;

[0025] Image recognition module: used to collect images of each of the films to identify the contrast between particles on the surface of the film and the substrate;

[0026] Optimal brightness module: used to determine whether the contrast of each of the film captured images meets the preset standard contrast threshold under the change of the minimum brightness unit; if it does, the corresponding brightness change interval is used as the optimal brightness range of the film captured image;

[0027] Image updating module: used for re-collecting the PET film image according to the optimal brightness range.

[0028] An electronic device comprises a memory, a processor and a computer program running on the processor, wherein when the processor executes the computer program, the steps of any one of the methods for intelligently collecting the frosted particle size of a PET film surface are implemented.

[0029] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0030] The present invention collects film collection images of multiple PET films under different brightness, identifies the contrast between the particles on the film surface and the substrate, and thus determines whether the contrast of each film collection image meets the standard contrast threshold under the change of the minimum brightness unit; when it meets the standard contrast threshold, the corresponding brightness change interval that can highlight the particle characteristics is used as the optimal brightness range of the film collection image, and is applied to the collection of PET film images. The present invention can collect different PET film surface images at appropriate brightness, thereby improving the accuracy of image recognition of frosted particle size, and thus facilitating high-precision processing of PET films. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0032] Figure 1 Schematic diagram of the intelligent collection method of the frosted particle size on the surface of the PET film according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0034] Example 1

[0035] like Figure 1 As shown, the embodiment of the present application provides a method for intelligently collecting the frosted particle size of a PET film surface, comprising: collecting film collection images of a plurality of PET films under illumination of different brightness;

[0036] Collecting images of each of the films to identify the contrast between particles on the film surface and the substrate;

[0037] Determine whether the contrast of each of the film captured images meets a preset standard contrast threshold under the change of the minimum brightness unit; if so, use the corresponding brightness change interval as the optimal brightness range of the film captured image;

[0038] The PET film image is re-collected according to the optimal brightness range.

[0039] The above-mentioned preset standard contrast threshold can be manually set according to the target contrast. Optionally, the preset standard contrast threshold is obtained by performing multiple experiments on film acquisition images that meet the standard quality requirements at brightnesses that meet the standard quality requirements, wherein the brightness and the standard quality requirements of the film acquisition images can be set according to the image clarity, which is a known method for those skilled in the art and does not need to be specifically limited here. The minimum brightness unit is a brightness value of 1 candela / square meter (cd / m2), and the brightness variation range that meets the contrast requirements of the PET film image recognition particles is obtained experimentally at different brightnesses, so that the PET film image is acquired using the corresponding brightness variation range.

[0040] After the PET film image is re-collected according to the optimal brightness range, the method further includes: collecting multiple groups of film image training data; each group of the film image training data includes the film collection image and the PET film image collected by the film collection image under the optimal brightness range;

[0041] A plurality of sets of the film image training data are trained through machine learning to obtain a film image training model; the film image training model is used to input images of a plurality of films to be tested to obtain re-collected PET film images.

[0042] After the PET film images are re-collected according to the optimal brightness range, the method further includes: identifying the particle features and substrate features of each output PET film image according to the image recognition technology; obtaining the particle image according to the pre-trained film image segmentation model;

[0043] Identify the particle area uniformity and particle distribution uniformity of the particle image; and adjust the processing parameters of the PET film according to the particle area uniformity and particle distribution uniformity.

[0044] Among them, the image recognition technology identifies the particle characteristics and substrate characteristics of the PET film image, and uses the existing image segmentation model to analyze the particle characteristics and substrate characteristics, and directly distinguishes the particle image by image segmentation. This belongs to the existing technology and does not need to be specifically limited here.

[0045] The step of adjusting the processing parameters of the PET film according to the uniformity of the particle area and the uniformity of the particle distribution comprises:

[0046] Determine whether the particle distribution uniformity is lower than the preset standard particle distribution uniformity. If so, reduce the film travel speed during sandblasting of the next PET film particles until the preset standard particle area uniformity is met; determine whether the particle area uniformity is lower than the preset standard particle area uniformity. If so, reduce the sanding roller speed during grinding of the next PET film particles or the film travel speed during grinding until the preset standard particle area uniformity is met.

[0047] The particle area uniformity of the particle image can be determined by identifying different particle area features in the particle image, such as diameter, shape, color, etc. The particle distribution uniformity of the particle image can be determined by identifying whether all particles in the particle array arrangement of the particle image are missing or redundant, whether the distances between the particles are consistent, etc. Therefore, it is determined whether the preset requirements of the corresponding features are met based on the above-mentioned characteristics of particle area uniformity and particle distribution uniformity. Therefore, the preset standard particle area uniformity and the corresponding feature size of the preset standard particle area uniformity are set according to actual needs. The setting of the specific numerical value belongs to the means known to those skilled in the art based on conventional experience, and no specific limitation is made here.

[0048] Since the PET film frosting is to spray the frosting particles evenly onto the film and then solidify it, and then grind the size of the frosting particles to make the size uniform. In addition, the film moves at a uniform speed during the sandblasting process. When the frosting particles are unevenly distributed after sandblasting, it is determined that the speed of sandblasting is too fast. Therefore, by reducing the speed of the film during sandblasting, the uniformity of the particle distribution on the next film to be sandblasted is improved to meet the uniformity standard. Then, depending on whether the size of the frosting particles formed after grinding is uniform, the grinding speed of the roller or the speed of the film during grinding is reduced to improve the uniformity of the particle distribution on the next film to be sandblasted so that the uniformity standard is met. The above-mentioned reduction in travel speed is used to improve the uniformity of particle distribution of the next film, thereby obtaining a film that meets the standard of uniform distribution, and the grinding speed of the roller during grinding or the travel speed of the film during grinding is reduced to improve the uniformity of particle area of ​​the next film, thereby obtaining a film that meets the uniformity of size. The specific size settings are obtained through experiments, and the configuration for evaluating the uniformity can be based on the routine experience of technicians in this field and is not limited here.

[0049] The method for intelligently collecting the frosted particle size of the surface of a PET film also includes: collecting multiple groups of film processing training data; each group of the film processing training data includes output images of each PET film, the film travel speed when the PET film particles are sandblasted, and the frosted roller speed when the PET film particles are polished / the film travel speed during polishing; and obtaining a film processing training model through machine learning of the multiple groups of film processing training data.

[0050] The method for intelligently collecting the frosted particle size of the surface of a PET film also includes: obtaining the film travel speed of each film to be tested during sandblasting and the frosted roller speed / film travel speed during polishing through the film processing training model; and processing the film to be tested using the output film travel speed during sandblasting and the frosted roller speed / film travel speed during polishing.

[0051] Through the particle quality of the PET film image, the sandblasting film travel speed that needs to be optimized during processing, the sandblasting roller speed and the film travel speed during grinding are obtained, thereby obtaining a film processing training model.

[0052] The film image segmentation model is obtained by the following steps:

[0053] Collect multiple sets of film image segmentation data; each set of the film image segmentation data includes a particle image and its particle features, a substrate image and its substrate features; and obtain the film image segmentation model through training of the multiple sets of the film image segmentation data.

[0054] The particle characteristics include any one or more characteristics of particle area, shape, brightness, and color; the substrate characteristics include any one or more characteristics of substrate area, shape, brightness, and color. They may also include characteristics such as particle area distribution and average roundness.

[0055] Among them, the relevant data of the area size, shape, brightness and color of the particles and the substrate can be obtained by image recognition technology, and the particles and the substrate are obtained by feature comparison, so as to distinguish them, which belongs to the prior art and is not specifically limited here. For example, the relevant data of the particle shape can be circularity, and when evaluating the distribution uniformity, the average circularity obtained after statistical analysis of the extracted circularity can be used to directly judge whether the average circularity meets the requirements, and when evaluating the size uniformity, the average area obtained after statistical analysis of the extracted particle area and the average circularity obtained after statistical analysis of the circularity can be used, so as to judge whether the corresponding requirements are met by using the size of the average area and the average circularity. The process of statistical analysis can adopt Gaussian mixture model (GMM), and the algorithm of statistical analysis average value can adopt the prior art. The specific target data set is directly obtained according to the film processing parameters required in actual application, which does not belong to the limited content of the present invention.

[0056] When the embodiment of the present application is used, a high-resolution camera can be used to collect images of multiple PET films under different brightness. The light source used for illumination with different brightness provides uniform and stable lighting conditions, reduces shadows and reflections during image acquisition, and improves image processing accuracy. The system can use a custom-designed LED lighting array, which is evenly distributed across the entire film width according to the characteristics of the PET film and the speed of the production line. The brightness and color temperature of the light source are adjustable to accommodate different frosted particle sizes and film colors to ensure that high-quality images can be obtained under various conditions. To further achieve uniform illumination, the following methods can be selected: 1. Multi-angle illumination: By setting multiple LED modules in the imaging area, the PET film is illuminated from different angles to reduce shadows caused by particle shape and surface irregularities; 2. Optical diffuser: An optical diffuser is installed in front of the LED light source to evenly diffuse the light and avoid hot spots and uneven brightness. 3. Reflection suppression technology: A special anti-reflection coating or structure is set between the light source and the imaging device to reduce image distortion caused by reflection on the film surface. When the PET film passes through the production line at a speed of 100 meters per minute, the light source system on the production line automatically adjusts to the optimal brightness range and color temperature to ensure the real-time and accuracy of image acquisition.

[0057] Optionally, the particle image and base image used in the film image segmentation model training can be preprocessed by grayscale conversion, filtering denoising and edge enhancement of the film acquisition image to eliminate irrelevant information and noise interference in the image; then, the frosted particles in the film acquisition image are segmented to obtain the segmented particle image. The image segmentation algorithm can use algorithms such as region growing algorithm and threshold segmentation.

[0058] Optionally, edge enhancement can use the Canny operator to perform edge detection to obtain edge information of frosted particles. The Canny edge detection algorithm can be expressed as:

[0059] ;

[0060] ;

[0061] ;

[0062] .

[0063] is the gradient operator, the film acquisition image about Partial derivatives of direction; is the gradient operator, the film acquisition image about Partial derivatives of direction; Represents the edge information of frosted particles; express The angle of the vector in the two-dimensional coordinate system.

[0064] When the region growing algorithm is used to segment the frosted particles, a set of points on the segmented image is obtained. Contains thin film acquisition images All points (x, y) on the surface that satisfy the similarity condition of frosted particles can be expressed as:

[0065] ;

[0066] in, is the set after region growing; Indicates the film acquisition image Point on ; Indicate point In two-dimensional space The value of the direction.

[0067] According to the area of ​​the segmented particle image, the particle area feature is extracted, which is expressed as: ;

[0068] in, is the area of ​​the segmented particle image, N is the length of the segmented particle image in the x-axis direction; M is the length of the segmented particle image in the y-axis direction; Indicates the film acquisition image Frosted particles on .

[0069] The extracted circularity feature can be expressed as:

[0070] ;

[0071] Where C is the circularity and P is the particle circumference.

[0072] Optionally, before extracting features such as particle area and circularity in the segmented particle image, the shape of the frosted particles obtained from the image can be optimized by morphological operations, such as using dilation and erosion operations, to optimize the obtained particle area, circularity and other features.

[0073] Among them, the connected region labeling algorithm is an image processing algorithm used to identify and label connected regions in an image. It is mainly used to find and label each connected region in an image for subsequent analysis and processing. The connected region labeling algorithm is usually implemented based on the traversal of the image, and determines the pixels belonging to the same connected region by judging the connectivity between pixels, and assigns a unique label to each connected region.

[0074] Optionally, a U-Net network is used to segment the frosted particles in the film acquisition image to obtain a segmented particle image. The identified particle image is used to calculate the geometric features of the image using a U-Net network structure based on deep learning, wherein the core formula or structure of the U-Net network can be summarized as follows:

[0075] Encoder (contraction path): Extract image features through a series of convolutional layers and pooling layers, and gradually reduce the spatial resolution of the image. This process can be expressed as:

[0076] The convolution operation is expressed as: ;

[0077] The pooling operation is expressed as: ;

[0078] in, is the input particle image, It is The convolutional feature map of the layer, It is Pooling result of the layer.

[0079] Decoder (expansion path): The decoder part restores the spatial resolution of the image through a series of upsampling layers and convolutional layers, and merges the corresponding feature maps from the encoder. This process can be expressed as:

[0080] Upsampling operation: ;

[0081] Convolution operation and feature merging: ;

[0082] in, It is The upsampling result of the layer, is the feature map skipped from the encoder, is the merged feature map.

[0083] For example, the following is the complex operation structure of the U-Net network:

[0084] Skip connection: directly connect the feature map in the encoder to the corresponding upsampling layer in the decoder , j is the decoder layer corresponding to i, and the process of skip connection can be expressed as:

[0085] ;

[0086] Final segmentation output: The merged feature map is processed through a series of convolutional layers to obtain the final convolution result .

[0087] ;

[0088] in, is the output image, which contains the particle image of the segmentation output.

[0089] Optionally, a Gaussian mixture model (GMM) is used to perform statistical analysis on characteristic data such as size, distribution, and density of frosted particles detected in the particle image to obtain frosted particle features. The specific process includes:

[0090] For a data point x, its probability density function p(x) is expressed by the following formula:

[0091] ;

[0092] in, is the mixing coefficient of the i-th Gaussian distribution; is a single Gaussian component of the probability density function, which represents the probability density function of the Gaussian distribution and describes the data point The probability distribution under the i-th component; where the mixing coefficient represents the contribution ratio of the i-th distribution to the overall distribution among all Gaussian distributions; Represents the number of Gaussian distributions.

[0093] in:

[0094] ;

[0095] is the mean vector of the i-th Gaussian distribution; is the covariance matrix of the i-th Gaussian distribution; d is the dimension of the data point; express The vector value of .

[0096] Use the complex operations provided by Gaussian mixture models:

[0097] Expectation step (E-step): In the E-step, the posterior probability of each data point belonging to each Gaussian distribution needs to be calculated ;

[0098] ;

[0099] represents the probability that the jth data point belongs to the i-th Gaussian distribution.

[0100] Maximization step (M-step): In the M-step, the parameters of each Gaussian distribution are updated:

[0101] The updated mixing coefficient can be expressed as: ;

[0102] Where N is the number of iterations.

[0103] The updated mean can be expressed as: ;

[0104] The updated covariance can be expressed as: ;

[0105] Iterative process: Repeat E-step and M-step until the likelihood function L converges, where the likelihood function is usually expressed as:

[0106] ;

[0107] Indicates the number of iterations of the convergence process;

[0108] Through the above steps, GMM can adapt to multimodal characteristics and provide more accurate statistical information for complex operations.

[0109] In order to further explore the contrast variation between the particles and the substrate on the surface of the PET film under different brightness conditions, a large amount of image data collected was carefully analyzed. With the increase of brightness, the grayscale values ​​of the particle area and the substrate area both show an upward trend, which is consistent with theoretical expectations, that is, the increase in brightness will increase the intensity of reflected light on the surface of the object, which will lead to an increase in the grayscale value of the pixels collected by the camera. However, the rising rates of the two are not completely consistent. In the low brightness stage, the grayscale value of the particle area increases relatively slowly, while the grayscale value of the substrate area increases slightly faster; as the brightness further increases, the growth rate of the grayscale value of the particle area gradually accelerates, and in the higher brightness range, the rising slope of the grayscale value of the particle area is significantly greater than that of the substrate area.

[0110] This difference in the rate of change of grayscale values ​​is directly reflected in the change of contrast. When the initial brightness is low, the contrast is relatively low. As the brightness gradually increases, the contrast shows a trend of first slowly rising and then rapidly increasing within a certain brightness range. When the brightness exceeds a certain threshold, the growth of the contrast tends to be gentle, and even slightly decreases in some samples. This trend shows that within the appropriate brightness range, the increase in brightness can effectively enhance the contrast between particles and the substrate, making the particle features more prominent in the image; but when the brightness is too high, the image quality may be reduced due to factors such as overexposure and light scattering, and the difference between particles and the substrate will decrease, resulting in a decrease in contrast.

[0111] Further experiments were conducted on PET film samples with different materials and different particle distribution characteristics. It was found that although the overall trend of contrast changing with brightness was similar, there were differences in the specific details of the changes. For film samples with larger particle sizes and sparser distribution, the contrast improvement was relatively gentle, and the brightness value required to achieve the maximum contrast was relatively high; while for films with fine particles and dense distribution, the contrast change was more sensitive at lower brightness, and a larger contrast improvement could be achieved within a relatively small brightness adjustment range. These differences provide an important basis for determining the optimal brightness range based on different film characteristics.

[0112] The contrast between the particles on the surface of PET film and the substrate changes with brightness, which is affected by a combination of factors. Particle size is an important factor. Larger particles occupy a larger pixel area in the image, and their scattering and reflection effects on light are more significant. At low brightness, due to the limited light energy, the scattered light intensity of large particles is relatively weak, the brightness difference with the substrate is not obvious, and the contrast is low; as the brightness increases, large particles can reflect light more effectively, so that the brightness of the particle area rises rapidly, and the contrast with the substrate increases accordingly. However, when the brightness is too high, the reflected light on the surface of large particles may form a halo effect in the surrounding area, blurring the boundary between the particles and the substrate, resulting in a decrease in contrast. The shape of the particles also has an important influence on the contrast change. Spherical particles reflect light more evenly in all directions, and their contrast changes are relatively smooth; while irregularly shaped particles, such as flake and needle-shaped particles, have more complex reflection and refraction paths on the particle surface due to the diversity of their surface normal directions. At different brightness, different parts of these irregular particles will show different brightness characteristics, making the overall brightness distribution of the particles uneven, and the contrast changes with the substrate also show complex fluctuations. At certain brightness, the protruding parts of the particles may form bright spots due to strong light reflection, enhancing the contrast; at other brightness, the shadows of the concave parts may be confused with the brightness of the substrate, reducing the contrast. The distribution density of the particles will also change the change pattern of contrast with brightness. When the particles are sparsely distributed, the mutual influence between the particles is small, and the contrast of each particle mainly depends on its own characteristics, the substrate and the brightness conditions. At this time, as the brightness increases, the particles stand out one by one, and the contrast gradually increases, but due to the small number of particles, the overall contrast increases relatively slowly. For a group of particles with high density distribution, at a lower brightness, the shadows between the particles overlap each other, making the particle area appear darker as a whole, and the contrast with the substrate is low; as the brightness increases, the shadows are gradually illuminated, the overall brightness of the particle group rises rapidly, and the contrast also increases sharply. However, when the brightness continues to increase, the particle group may form a bright area due to overexposure, and the distinction from the substrate is reduced again, and the contrast decreases. The roughness of the substrate is also one of the key factors affecting the change of contrast. The rough substrate surface causes multiple scattering of the incident light, forming diffuse reflection, resulting in a relatively uniform brightness distribution in the substrate area but a relatively low overall brightness. At low brightness, the contrast between the particles and the rough substrate is relatively small; as the brightness increases, the reflected light intensity of the particles increases more significantly, and the contrast with the substrate gradually increases. However, if the brightness is too high, the diffuse reflected light intensity of the rough substrate will also increase significantly, making the brightness of the substrate close to or even exceeding the brightness of the particles, resulting in a sharp drop in contrast.

[0113] In addition, the stability of the light source has a crucial impact on the accuracy of the experimental results and the stability of the contrast. If the light source has unstable phenomena such as light intensity fluctuations and spectral drift during the brightness adjustment process, the lighting conditions for image acquisition will be inconsistent, resulting in random errors in the brightness changes of the particles and the substrate, which in turn affects the measurement accuracy of the contrast. During long-term experiments, the heat generated by the light source may also cause changes in its performance. For example, as the temperature of an LED light source increases, the luminous efficiency and spectral characteristics may change, making the actual brightness output under the same driving current different from the initial state, interfering with the study of the law of contrast changes with brightness. Therefore, ensuring the high stability of the light source and equipping it with a high-precision light intensity monitoring and feedback control system are necessary prerequisites for obtaining reliable experimental data and in-depth exploration of the contrast change mechanism.

[0114] In the field of electronic component packaging, experiments were conducted on PET films used for chip protection. Hundreds of film samples from different batches and under different production process conditions were collected. When the brightness was gradually increased from 50 cd / m² to 500 cd / m² with a step size of 50 cd / m², the surface images of the film were collected and the increase in contrast between the particles and the substrate was calculated. The experimental results show that when the contrast increase is less than 30%, the recognition of tiny particles with a size of 10-20μm in the image is extremely low, and they are easily confused with the substrate noise, resulting in a misjudgment rate of more than 30%; when the contrast increase reaches 40% or more, these tiny particles can be clearly separated from the substrate background, and the misjudgment rate can be controlled within 5%. Taking into account the extremely high requirements for the reliability of electronic components in this field, 40% is used as the lower limit of the contrast increase threshold of such PET films in electronic component packaging applications to ensure that in the subsequent production inspection process, film products that may pose risks can be effectively identified and eliminated.

[0115] In the food packaging application scenario, similar experiments are conducted on PET films used for beverage bottle labels, food cling film and other products. Due to the focus on hygiene and safety in food packaging, more emphasis is placed on detecting the impact of larger particles on packaging appearance and barrier properties. Experimental data show that within the brightness variation range, for particles larger than 50μm, when the contrast increase is less than 20%, the contours of these particles in the image are blurred, and it is difficult to accurately judge their shape, size and distribution, which may cause defective films to be misjudged as qualified and enter the market; when the contrast increase is increased to 25% or above, the particle characteristics are significantly enhanced, and the inspectors or automated inspection systems can quickly and accurately identify particle defects to ensure the quality of food packaging. Based on this, 25% is determined as the contrast increase threshold standard for PET films for food packaging, which provides food packaging manufacturers with clear quantitative indicators in the quality control link.

[0116] After acquiring the images of the PET film under different brightness settings, the contrast calculation method described above is first used to calculate the contrast between the particles and the substrate under each brightness condition. Taking a PET film sample as an example, at the initial brightness (L_0), the average pixel value of the particle area is (I_p^0), and the average pixel value of the substrate area is (I_b^0) through grayscale value contrast or spectral reflectance contrast, and the initial contrast (C_0) is calculated; then the brightness is gradually adjusted, such as adding a minimum brightness unit (\Delta L), and the particle and substrate pixel values ​​(I_p^1) and (I_b^1) under the new brightness (L_1 = L_0 + \Delta L) are obtained, and then the contrast (C_1) is calculated. Similarly, a series of contrast data sequences ({C_n}) under brightness changes are obtained, and the contrast increase (\Delta C_n = C_n - C_{n - 1}) is calculated. Compare the calculated contrast increase (\Delta C_n) with the preset standard contrast change threshold (T). If (\Delta C_n \geq T), it means that within this brightness change range, the image contrast improvement has reached the expected standard, and the particle characteristics can be more effectively highlighted. At this time, the brightness range corresponding to this brightness change range is recorded as one of the potential optimal brightness ranges; if (\Delta C_n < T), it means that under this brightness adjustment, the contrast enhancement effect does not meet the requirements, and the distinction between particles and the substrate is still not ideal. Continue to adjust and judge the next brightness unit until all the set brightness change ranges are traversed.

[0117] In summary, the embodiment of the present application provides a method and system for intelligently collecting frosted particle size on the surface of a PET film: by collecting film collection images of multiple PET films under different brightness, identifying the contrast between particles on the surface of the film and the substrate, thereby judging whether the contrast of each film collection image meets the standard contrast threshold under the change of the minimum brightness unit; when it meets the standard contrast threshold, the corresponding brightness change interval that can highlight the particle characteristics is used as the optimal brightness range of the film collection image, and is applied to the collection of PET film images. The present invention can collect different PET film surface images at appropriate brightness, thereby improving the accuracy of image recognition of frosted particle size, and thus facilitating high-precision processing of PET films.

[0118] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for intelligently collecting the frosted particle size of a PET film surface, characterized in that: include: Collect film collection images of multiple PET films under different brightness; Collecting images of each of the films to identify the contrast between particles on the film surface and the substrate; Determine whether the contrast of each of the film captured images meets a preset standard contrast threshold under the change of the minimum brightness unit; if so, use the corresponding brightness change interval as the optimal brightness range of the film captured image; The PET film image is re-collected according to the optimal brightness range.

2. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 1, characterized in that: After the PET film image is re-collected according to the optimal brightness range, the method further includes: Collecting multiple groups of film image training data; each group of the film image training data includes the film acquisition image, and the PET film image acquired by the film acquisition image under the optimal brightness range; A plurality of sets of the film image training data are trained through machine learning to obtain a film image training model; and the film image training model is used to output a plurality of PET film images of the films to be tested.

3. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 2, characterized in that: After the PET film images are re-collected according to the optimal brightness range, the method further includes: identifying the particle features and substrate features of each output PET film image according to the image recognition technology; obtaining the particle image according to the pre-trained film image segmentation model; Identify the particle area uniformity and particle distribution uniformity of the particle image; and adjust the processing parameters of the PET film according to the particle area uniformity and particle distribution uniformity.

4. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 3, characterized in that: The step of adjusting the processing parameters of the PET film according to the uniformity of the particle area and the uniformity of the particle distribution comprises: Determine whether the particle distribution uniformity is lower than the preset standard particle distribution uniformity. If so, reduce the film travel speed during sandblasting of the next PET film particles until the preset standard particle area uniformity is met; determine whether the particle area uniformity is lower than the preset standard particle area uniformity. If so, reduce the sanding roller speed during grinding of the next PET film particles or the film travel speed during grinding until the preset standard particle area uniformity is met.

5. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 4, characterized in that: Also includes: Collect multiple sets of film processing training data; each set of the film processing training data includes the output PET film images, the film travel speed when the PET film particles are sandblasted, and the sanding roller speed when the PET film particles are polished / the film travel speed when polishing; and obtain a film processing training model through machine learning of the multiple sets of film processing training data.

6. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 5, characterized in that: Also includes: The film processing training model is used to obtain the film travel speed of each film to be tested during sandblasting, and the sanding roller speed / film travel speed during grinding; The film to be tested is processed using the output film travel speed during sandblasting and the sanding roller speed / film travel speed during grinding.

7. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 3, characterized in that: The film image segmentation model is obtained by the following steps: Collecting multiple sets of film image segmentation data; each set of the film image segmentation data includes a particle image and its particle features, a substrate image and its substrate features; The film image segmentation model is obtained by training multiple sets of the film image segmentation data.

8. The method for intelligently collecting the frosted particle size of a PET film surface according to claim 3, characterized in that: The particle characteristics include any one or more characteristics of particle area, shape, brightness, and color; The substrate characteristics include any one or more characteristics of substrate area, shape, brightness, and color.

9. A system for intelligently collecting the frosted particle size of a PET film surface according to any one of claims 1 to 8, characterized in that: include: Image acquisition module: used to acquire film acquisition images of multiple PET films under different brightness; Image recognition module: used to collect images of each of the films to identify the contrast between particles on the surface of the film and the substrate; Optimal brightness module: used to determine whether the contrast of each of the film captured images meets the preset standard contrast threshold under the change of the minimum brightness unit; if it does, the corresponding brightness change interval is used as the optimal brightness range of the film captured image; Image updating module: used for re-collecting the PET film image according to the optimal brightness range.

10. An electronic device comprising a memory, a processor, and a computer program running on the processor, characterized in that: When the processor executes the computer program, the steps of the intelligent collection method of frosted particle size on the surface of a PET film as described in any one of claims 1 to 8 are implemented.

Images