Plastic product detection system based on image recognition

Abstract

The invention discloses a plastic product detection system based on image recognition, which relates to the technical field of plastic product detection and comprises an image acquisition module, a data storage and feature extraction module, a feature analysis and model input module, an intelligent evaluation and classification module, a homogeneous coating acquisition module and a heterogeneous coating adjustment module, first, a scanner is used to irradiate the surface of a plastic product with a preset beam width and focal length, and a high-quality surface image is captured. According to the method, deep learning and image analysis technologies are introduced, so that the detection precision of the non-uniformity of the coating of the plastic product is remarkably improved. The system dynamically analyzes the surface texture and reflection heterogeneity, intelligently evaluates the consistency of the coating, adjusts the width and focal length of the light beam according to the state of the coating, and ensures a high-quality image. According to the intelligent detection and adjustment strategy, the detection efficiency and accuracy are effectively improved, the quality risk and the production cost are reduced, and the limitation of a traditional method is broken through.

Description

Technical Field

[0001] The present invention relates to the technical field of plastic product detection, and in particular to a plastic product detection system based on image recognition. Background Art

[0002] Plastic product inspection based on image recognition is a method of automated quality inspection of plastic products using computer vision technology. Images of plastic products are acquired through a high-resolution camera or scanner and input into an image recognition system, which uses algorithms such as deep learning and convolutional neural networks (CNN) to process, analyze, and extract features from the images. The system can automatically detect defects in plastic products, such as cracks, bubbles, deformation, color difference, or surface contamination. Plastic product inspection based on image recognition not only improves inspection efficiency and reduces manual intervention, but also provides higher accuracy and consistency, helping to promptly detect quality problems in the production process, thereby ensuring high-standard product quality.

[0003] The prior art has the following deficiencies:

[0004] In the prior art, scanners usually illuminate the surface of plastic products with a fixed beam width and focal length, and the light is captured by the sensor after reflection to generate image data. However, when the coating on the plastic surface is uneven, resulting in excessive reflection in certain areas, the sensor may not be able to clearly capture the details of these areas, leading to a series of serious consequences. This excessive reflection may cause the sensor to be overexposed or light saturated, unable to normally capture the details of the reflected light, resulting in distorted or missing image data, and unable to accurately identify potential surface defects such as cracks, bubbles, stains or scratches. As a result, quality inspections may fail, and even if there are serious defects in the plastic products, these problems cannot be discovered in time. If these defects are not identified in time, they may enter the mass production stage and eventually flow into the market, resulting in substandard products, and even product recalls or quality lawsuits, causing huge economic losses and brand reputation crises to the company.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a plastic product detection system based on image recognition. By introducing deep learning models and image analysis technology, this solution significantly improves the detection accuracy of uneven coatings on plastic products. Dynamically analyze the surface texture differentiation reference value and the reflection heterogeneity reference value, intelligently evaluate the coating state, and accurately divide the coating uniformity through the coating consistency index. For products with uneven coatings, the system can adjust the scanner's beam width and focal length in real time to ensure image quality and avoid the limitations of traditional methods. This intelligent detection and adjustment strategy improves detection efficiency and accuracy, reduces quality risks and costs, and solves the problems in the above-mentioned background technology.

[0007] In order to achieve the above-mentioned purpose, the present invention provides the following technical solutions: a plastic product detection system based on image recognition, including an image acquisition module, a data storage and feature extraction module, a feature analysis and model input module, an intelligent evaluation and classification module, a homogeneous coating acquisition module and a heterogeneous coating adjustment module:

[0008] The image acquisition module, first, uses a scanner to illuminate the surface of the plastic product with a preset beam width and focal length to capture high-quality surface images;

[0009] Data storage and feature extraction module, the collected surface images are stored and organized into a structured data set, and key features that help identify coating unevenness are extracted from the data set;

[0010] The feature analysis and model input module analyzes the extracted key features under the detection window and inputs the analyzed key features into the pre-learned deep learning model, and uses the deep learning model to intelligently evaluate the coating status of the plastic product surface;

[0011] Intelligent evaluation and classification module, which divides the surface state of plastic products into homogeneous coating and heterogeneous coating based on the evaluation results of deep learning model;

[0012] Homogeneous coating acquisition module: for plastic products that are judged to be homogeneous coatings, the scanner continues to collect surface state data with the original set beam width and focal length;

[0013] The heterogeneous coating adjustment module dynamically adjusts the scanner's beam width and focal length for plastic products judged to have heterogeneous coatings based on the deep learning model evaluation results.

[0014] Preferably, key features that help identify coating unevenness are extracted from the data set, and the extracted features include texture changes and surface reflected light gradients. Under the detection window, the extracted texture changes and surface reflected light gradients are analyzed to generate surface texture differentiation reference values ​​and surface reflection heterogeneity reference values, respectively. The surface texture differentiation reference values ​​quantify the degree of variation of the surface texture of plastic products in spatial distribution and local areas, and the surface reflection heterogeneity reference values ​​quantify the differences in reflected light intensity in different areas of the surface of plastic products, which is specifically manifested as an uneven distribution of surface reflectivity.

[0015] Preferably, in the detection window, the specific steps of analyzing the texture change and generating the reference value of surface texture differentiation are as follows:

[0016] In the detection window, firstly, the texture features of the image data on the surface of the plastic product are extracted. Through the texture analysis of the local area, the texture pattern data of the area is extracted. According to the extracted texture features, the texture difference of each area in the image is calculated. The calculation expression is as follows:

[0017]

[0018] , where is the value of the kth texture feature dimension of region i in the image, is the value of the kth texture feature dimension of region j in the image, α is the adjustment factor of the difference metric, N is the total number of texture features, represents the difference between region i and region j in the kth texture feature dimension, D(t i ,t j ) is the texture difference;

[0019] Based on the results of texture difference measurement, the texture differences of different regions in the image are calculated, and the differences are weighted by the weight factor. The calculation expression is as follows:

[0020]

[0021] , where Δ i is the regional texture difference, β j is the weight factor of the neighborhood region, is the weight factor of region j, is a set of neighborhood regions, is a set, representing the set of regions adjacent to region i;

[0022] Combine all regional texture differences Δ i , generate the reference value of surface texture differentiation, the calculation expression is as follows:

[0023]

[0024] , where Itexture is the reference value of surface texture differentiation, M is the total number of regions, λ i is the adjustment factor for region i.

[0025] Preferably, the specific steps of analyzing the surface reflected light gradient under the detection window and generating the surface reflectance heterogeneity reference value are as follows:

[0026] Under the detection window, the reflected light data of the surface of the plastic product is collected by the scanner to obtain the reflected light intensity value of each sampling point. The light gradient is obtained by calculating the light intensity change of adjacent pixel points. The calculation expression is as follows:

[0027]

[0028] , where R(x,y) is the intensity of the reflected light on the surface, is the reflected light intensity gradient, is the rate of change of reflected light intensity in the x direction, is the rate of change of reflected light intensity in the y direction;

[0029] By calculating the light gradient of each pixel, the reflection unevenness of each local area is obtained. The calculation expression is as follows:

[0030]

[0031] , where L(x,y) is the measure of local light reflection inhomogeneity, θ is the sensitivity coefficient, is the nonlinear amplification factor;

[0032] After calculating the local reflection heterogeneity metric L(x,y) of each pixel, the local areas within the entire detection window are summarized to obtain the global reflection heterogeneity metric. The calculation expression is as follows:

[0033]

[0034] , where ω(x,y) is the pixel weight, is the gradient attenuation factor, γ is the attenuation coefficient, I reflect It is the reference value of surface reflectance heterogeneity.

[0035] Preferably, after obtaining the surface texture differentiation reference value and the surface reflection heterogeneity reference value generated after analyzing the extracted key features, the surface texture differentiation reference value and the surface reflection heterogeneity reference value are input into a pre-learned deep learning model, and a coating consistency index is generated through the deep learning model. The coating consistency index is used to intelligently evaluate the coating status of the plastic product surface.

[0036] Preferably, the generated coating consistency index is compared and analyzed with a preset coating consistency index reference threshold value to classify the surface state of the plastic product, and the classification steps are as follows:

[0037] If the coating consistency index is greater than or equal to a preset coating consistency index reference threshold, the surface of the current plastic product is in an uneven state, and it is classified as a heterogeneous coating;

[0038] If the coating consistency index is less than a preset coating consistency index reference threshold, the surface of the current plastic product is in a uniform state and is classified as a homogeneous coating.

[0039] Preferably, for plastic products that are judged to be heterogeneous coatings, the specific steps of dynamically adjusting the beam width and focal length of the scanner according to the evaluation results of the deep learning model are as follows:

[0040] If the plastic product being inspected is classified as a heterogeneous coating, the scanner parameters are dynamically adjusted based on the model evaluation results to ensure that more details can be captured in the area of ​​uneven coating. Based on the evaluation results of the deep learning model, the dynamically adjusted beam width and focal length will be refined and optimized according to the degree of uneven coating. The calculation expression is as follows:

[0041] Beam_width_dynamic=f(CCI,Beam_width_init),

[0042] Focus_dynamic=f(CCI,Focus_init)

[0043] , where Beam_width_dynamic is the adjusted beam width, Beam_width_init is the initial beam width, Focus_dynamic is the adjusted focal length value, Focus_init is the initial focal length, and CCI is the coating consistency index;

[0044] Based on the evaluation of coating non-uniformity, the scanner captures surface condition information more precisely by adjusting the beam width and focal length. In this process, the adjustment range will depend on the actual surface coating characteristics and the degree of coating non-uniformity in the target area. The adjustment rule formula is as follows:

[0045] ΔBeam_width=A·(CCI ref -CCI),

[0046] ΔFocus=B·(CCIref-CCI)

[0047] , where ΔBeam_width is the adjustment amount of beam width, A is the beam width adjustment proportional factor, CCIref is the coating consistency index reference threshold, ΔFocus is the adjustment amount of focal length, and B is the focal length adjustment proportional factor;

[0048] To ensure optimal use of the adjusted parameters, when the scanner captures new data, the dynamically adjusted beam width and focal length will be iteratively optimized according to preset rules and feedback mechanisms to ensure that the best image quality can be obtained each time. The calculation expression is as follows:

[0049] Data_quality=f(Beam_width_dynamic,Focus_dynamic)

[0050] , where Data_quality is the data quality.

[0051] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0052] The present invention significantly improves the detection accuracy of uneven coatings on plastic products by introducing deep learning models and image analysis technology. In traditional scanner detection methods, due to the fixed light beam and unchanged focal length, the excessive reflection and light saturation problems caused by uneven coatings will affect the sensor's capture of details, resulting in distortion or loss of image data, and thus making it impossible to identify potential defects in a timely manner. However, this solution dynamically analyzes the surface texture differentiation reference value and the surface reflection heterogeneity reference value, two key features that can quantify the uniformity of the coating and the difference in reflected light. Through the deep learning model, these features are intelligently evaluated. The model can accurately divide the coating status according to the coating consistency index, ensuring that more detailed and professional analysis can be performed in areas with uneven coatings. This intelligent detection not only makes up for the limitations of traditional methods, but also can identify and mark potential coating defects in advance, ensuring that unqualified products do not enter the production process, thereby greatly reducing quality risks and reducing the possibility of product recalls.

[0053] For plastic products with uneven coatings, the deep learning model of the present invention can analyze the coating consistency index in real time, and dynamically adjust the beam width and focal length of the scanner according to the model evaluation results. This intelligent adjustment process means that when uneven coating is detected, the scanner can automatically reduce the beam width and adjust the focal length, so as to accurately capture the details of the uneven coating. This strategy effectively avoids the drawbacks of traditional fixed parameter scanning, especially when facing excessive reflection or saturated areas, and can ensure the integrity and high quality of image data with more appropriate optical settings. In addition, the intelligent beam and focal length adjustment strategy also improves the inspection efficiency of the production line, avoids unnecessary repeated inspections, and thus saves time and costs. Through this dynamic adjustment, the system can efficiently respond to the coating status of different plastic products, maintaining high precision and high speed while ensuring the accuracy of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0055] Figure 1 This is a module schematic diagram of a plastic product detection system based on image recognition according to the present invention. DETAILED DESCRIPTION

[0056] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0057] The present invention provides Figure 1 The plastic product inspection system based on image recognition shown in the figure includes an image acquisition module, a data storage and feature extraction module, a feature analysis and model input module, an intelligent evaluation and classification module, a homogeneous coating acquisition module and a heterogeneous coating adjustment module:

[0058] The image acquisition module, first, uses a scanner to illuminate the surface of the plastic product with a preset beam width and focal length to capture high-quality surface images;

[0059] Scanners with preset beam width and focal length mean that during the scanning process, the laser beam width and focal length of the scanner have been fixed by pre-set parameters, which determine the size of the beam when the scanner irradiates the surface of the plastic product and the clarity of the focus. The beam width affects the size of the area covered by the scan, while the focal length determines the clarity of the image and the accuracy of detail capture. By presetting these parameters, the scanner can ensure that the surface image is acquired with consistent accuracy and coverage every time it scans, ensuring the reliability and consistency of the data.

[0060] Data storage and feature extraction module, the collected surface images are stored and organized into a structured data set, and key features that help identify coating unevenness are extracted from the data set;

[0061] To store and organize the collected surface images into a structured data set, the images must first be preprocessed, including denoising, standardization, image segmentation and other steps to ensure the consistency and high quality of the image data. Next, the processed images are organized in a certain format (such as a data table, matrix or feature vector) to form a data set for subsequent analysis. During the structuring process, the image data will be stored together with its corresponding labels (such as coating status) for subsequent model training and evaluation. The purpose of organizing into a structured data set is to facilitate subsequent feature extraction and analysis. The standardized data structure can efficiently screen, retrieve and process data to ensure that the machine learning model can accurately extract valuable features from it and realize intelligent evaluation.

[0062] The feature analysis and model input module analyzes the extracted key features under the detection window and inputs the analyzed key features into the pre-learned deep learning model, and uses the deep learning model to intelligently evaluate the coating status of the plastic product surface;

[0063] Key features that help identify coating unevenness are extracted from the data set. The extracted features include texture changes and surface reflected light gradients. Under the detection window, the extracted texture changes and surface reflected light gradients are analyzed to generate surface texture differentiation reference values ​​and surface reflection heterogeneity reference values, respectively. The surface texture differentiation reference values ​​quantify the degree of variation of the surface texture of plastic products in spatial distribution and local areas. The surface reflection heterogeneity reference values ​​quantify the differences in reflected light intensity in different areas of the plastic product surface, which is specifically manifested as the uneven distribution of surface reflectivity.

[0064] Irregular changes in the surface texture of plastic products usually indicate that the current surface coating of plastic products is uneven. The uniformity of the coating directly affects the texture distribution of the plastic surface, because factors such as the coating method, thickness and adhesion of the coating on the surface will cause different areas to show different gloss, texture and texture structure. When the coating is unevenly applied, some areas may have too thick or too thin coatings, which will cause significant changes in the surface smoothness, roughness or texture direction of these areas. Specifically, uneven coatings may form local protrusions or depressions on the surface, which in turn affects the reflection characteristics of light and causes irregular changes in the surface texture. For example, areas with thicker coatings may show smooth, highly reflective textures, while areas with thinner coatings may have rougher surface textures. This difference in texture will reflect the thickness and uniformity of the coating. Therefore, texture irregularity is often a direct manifestation of uneven coatings, which can help the detection system identify coating quality problems and take corresponding measures.

[0065] In the detection window, the specific steps for analyzing texture changes and generating reference values ​​for surface texture differentiation are as follows:

[0066] In the detection window, the texture features of the image data on the surface of the plastic product are first extracted. In order to quantify the changes in surface texture, the gray level co-occurrence matrix (GLCM) is used to capture the changes in surface details. Texture features usually include contrast, homogeneity, energy, entropy, etc., which are used to represent the changes in gray values ​​in the local area of ​​the image. Through the texture analysis of the local area, the texture pattern data of the area is extracted. These data reflect the differences in coating quality in different areas. According to the extracted texture features, the texture difference of each area in the image is calculated. The calculation expression is as follows:

[0067]

[0068] , where is the value of the kth texture feature dimension of region i in the image, is the value of the kth texture feature dimension of region j in the image, α is the adjustment factor of the difference metric, which is used to control the contribution of the texture feature difference to the total difference metric, N is the total number of texture features, represents the difference between region i and region j in the kth texture feature dimension, D(t i ,t j ) is the texture difference, is the texture feature vector t i and t j Differences in texture between

[0069] This step captures subtle texture changes by calculating the nonlinear difference in texture between local regions.

[0070] Based on the results of texture difference measurement, the texture differences of different regions in the image are calculated, and the differences are weighted by the weight factor. The calculation expression is as follows:

[0071]

[0072] , where Δ i is the regional texture difference, indicating the texture difference of region i in the image, β j is the weight factor of the neighborhood area, is the weight factor of area j, and the weight factor is used to indicate the influence of neighborhood area j on the texture difference calculation of the current area i. Different weights can be assigned to different areas according to the relative importance of the neighborhood areas. is a set of neighborhood regions, is a set, representing the set of regions adjacent to region i;

[0073] This step further enhances the representation capability of local texture differences by weighting.

[0074] Combine all regional texture differences Δ i , generate the reference value of surface texture differentiation, the calculation expression is as follows:

[0075]

[0076] , where I texture is the reference value of surface texture differentiation, M is the total number of regions, λ i is the adjustment factor of region i, which adjusts the contribution of the texture difference of this region to the overall texture differentiation index;

[0077] Under the detection window, the larger the performance value of the surface texture differentiation reference value generated after analyzing the texture changes, it usually indicates that the coating on the surface of the current plastic product is uneven. The surface texture differentiation index reflects the consistency of the coating on the plastic surface by quantifying the degree of change in the spatial distribution and local area of ​​the texture. When the coating is uneven, irregular texture changes usually form on the surface, causing the texture characteristics of some areas to be significantly different from other areas. This irregular texture change will cause the differentiation index to increase. On the contrary, when the coating is uniform, the surface texture changes less, the texture distribution of the entire surface tends to be consistent, and the differentiation index value is lower.

[0078] A rapid increase in the gradient of reflected light on the surface can usually indicate that the surface coating of the plastic product is uneven. The unevenness of the coating may cause the reflected light intensity in certain areas of the surface to increase abnormally, especially in areas where the coating is weak or unevenly distributed. This is because the optical properties of the coating (such as refractive index, reflectivity, etc.) will cause changes in the reflected light on the surface due to inconsistent thickness. In places with uneven coating, when light contacts the surface, some areas may produce different degrees of light reflection due to thin coating or partial loss, while the coating in other places may be thicker or uniform, and the reflected light is relatively stable. When the light gradient increases rapidly, it means that in some local areas, the intensity of the reflected light changes very drastically, which may be caused by uneven coating thickness, bubbles, cracks or other defects. This phenomenon indicates that the surface has uneven coating characteristics, which in turn affects image capture and subsequent quality inspection. Therefore, the rapidly rising reflected light gradient is an important indicator for evaluating the uneven coating of plastic products.

[0079] Under the detection window, the specific steps for analyzing the surface reflected light gradient and generating the surface reflectance heterogeneity reference value are as follows:

[0080] Under the detection window, the reflected light data of the surface of the plastic product is collected by the scanner to obtain the reflected light intensity value of each sampling point. The light gradient is obtained by calculating the light intensity change of adjacent pixel points. The calculation expression is as follows:

[0081]

[0082] , where R(x,y) is the surface reflected light intensity value, which represents the reflected light intensity value at the image coordinate (x,y). is the reflected light intensity gradient, which represents the reflected light intensity value of the reflected light intensity R(x,y) in the x and y directions, and is a measure of the change in local light intensity. is the rate of change of the reflected light intensity in the x direction (horizontal direction), is the rate of change of the reflected light intensity in the y direction (vertical direction);

[0083] By calculating the reflected light gradient, the changes in light reflectance in local areas of the surface can be revealed, thereby providing a data basis for subsequent reflection heterogeneity analysis.

[0084] By calculating the light gradient of each pixel, the reflection unevenness of each local area is obtained. The calculation expression is as follows:

[0085]

[0086] , where L(x,y) is the local light reflection inhomogeneity measure, θ is the sensitivity coefficient, which is an adjustment coefficient used to control the sensitivity of the gradient to the inhomogeneity measure, is the nonlinear amplification factor, which controls the nonlinear degree of light gradient change;

[0087] This step can more accurately capture the uneven changes in the surface coating by exponentially amplifying the changes in the reflectance gradient.

[0088] After calculating the local reflection heterogeneity metric L(x,y) of each pixel, the local areas within the entire detection window are summarized to obtain the global reflection heterogeneity metric. The calculation expression is as follows:

[0089]

[0090] , where ω(x,y) is the pixel weight, which is the weight of each pixel, indicating the importance of the pixel in calculating the reflectance heterogeneity metric. is the gradient attenuation factor, γ is the attenuation coefficient, which controls the influence of the light gradient on the reflectance heterogeneity metric, and I reflect It is the reference value of surface reflectance heterogeneity.

[0091] Under the detection window, the larger the surface reflection heterogeneity reference value generated after analyzing the surface reflection light gradient, the more obvious the unevenness of the surface coating of the plastic product is. When the coating on the plastic surface is uneven, the intensity and distribution of the reflected light in different areas will change significantly. By analyzing the reflected light gradient and calculating the surface reflection heterogeneity index, the unevenness of the reflected light can be quantified. Areas with large surface reflection light gradients are usually caused by changes in coating thickness or uneven coating distribution, resulting in large differences in light reflection in space. This difference is reflected as a higher value in the surface reflection heterogeneity index. Therefore, when the surface reflection heterogeneity index has a large value, it means that the coating has obvious uniformity problems on the surface; conversely, if the value of the index is low, it means that the coating on the surface of the plastic product is more uniform and the change in reflected light is small.

[0092] After obtaining the surface texture differentiation reference value and the surface reflection heterogeneity reference value generated after analyzing the extracted key features, the surface texture differentiation reference value and the surface reflection heterogeneity reference value are input into a pre-learned deep learning model, and a coating consistency index is generated through the deep learning model. The coating consistency index is used to intelligently evaluate the coating status of the plastic product surface.

[0093] The deep learning model is not limited here, and can achieve the surface texture differentiation reference value I texture and the surface reflectance heterogeneity reference value I reflect Any deep learning model that performs comprehensive analysis to generate the coating consistency index CCI can be used. To implement the technical solution of the present invention, the present invention provides a specific implementation method:

[0094] The coating consistency index CCI generation formula is as follows:

[0095]

[0096] , where b1 and b2 are the reference values ​​of surface texture differentiation I texture and the surface reflectance heterogeneity reference value I reflect The preset proportional coefficient, and b1 and b2 are both greater than 0.

[0097] The preset proportionality coefficient refers to the parameter used to adjust the relative influence weight of different variables in the formula on the final coating consistency index CCI. In the formula, b1 and b2 are the reference values ​​of surface texture differentiation I texture and the surface reflectance heterogeneity reference value I reflect The proportional coefficients of b1 and b2 are used to indicate the importance or contribution of these two parameters in the comprehensive generation of the coating consistency index. By adjusting these two coefficients, different application scenarios or detection requirements can be flexibly adapted. For example, if the change of surface texture in the detection is more important for the coating consistency assessment, the value of b1 can be increased to give the surface texture differentiation reference value I texture A greater weight is given to the coating consistency index CCI, which focuses more on the influence of texture features. The reasonable selection of the preset proportional coefficient is optimized based on experimental data or actual experience to ensure that the formula can adapt to specific technical requirements and testing environment.

[0098] It can be seen from the coating consistency index that, under the detection window, the larger the performance value of the surface texture differentiation reference value generated after analyzing the texture change, and the larger the performance value of the surface reflection heterogeneity reference value generated after analyzing the surface reflected light gradient, the larger the performance value of the coating consistency index generated by comprehensive analysis of the extracted key features under the detection window, indicating that the coating on the surface of the current plastic product is uneven, otherwise, it indicates that the coating on the surface of the plastic product is relatively uniform.

[0099] A pre-learned deep learning model usually refers to a neural network model that has been trained on a large amount of labeled data and has adjusted weights and biases through the back-propagation algorithm. During the training process, this model learns a large amount of labeled data (for example, surface image data of plastic products with known good and bad coating consistency), can extract effective features from these data and learn how to associate the relationship between features and surface coating consistency. During the training process, the model continuously optimizes parameters (such as the weight matrix) to reduce the error between the predicted value and the actual value until the model can handle similar new data well. Therefore, the pre-learned deep learning model already has the ability to automatically analyze and process the input feature data. In practical applications, only new feature data needs to be input to make fast and efficient predictions.

[0100] For coating consistency detection, the pre-learned deep learning model will learn the complex relationship between texture changes and spectral absorption characteristics and coating consistency based on known annotated data. These relationships may be difficult to capture through manually set rules in traditional regular models. Deep learning models, especially models with advantages in image processing and sequence data analysis such as convolutional neural networks (CNN) and recurrent neural networks (RNN), can automatically extract key features reflecting coating consistency from surface images and make intelligent evaluations. The surface texture differentiation reference value and surface reflection heterogeneity reference value input into the model will be processed layer by layer by multiple neurons of the model, and finally generate a coating consistency index, which quantifies the uniformity of the coating. The generation of the coating consistency index is an intelligent result obtained by the deep learning model after multiple training of the original data. It can make efficient and accurate judgments on the coating status and provide support for subsequent production decisions or quality control.

[0101] Intelligent evaluation and classification module, which divides the surface state of plastic products into homogeneous coating and heterogeneous coating based on the evaluation results of deep learning model;

[0102] The generated coating consistency index is compared and analyzed with the preset coating consistency index reference threshold, and the surface state of the plastic product is divided. The division steps are as follows:

[0103] If the coating consistency index is greater than or equal to a preset coating consistency index reference threshold, the surface of the current plastic product is in an uneven state, and it is classified as a heterogeneous coating;

[0104] If the coating consistency index is less than a preset coating consistency index reference threshold, the surface of the current plastic product is in a uniform state and is classified as a homogeneous coating.

[0105] Homogeneous coating means that the coating on the surface of plastic products has a high degree of consistency and uniformity, which is manifested by the uniform distribution of the material in terms of texture, thickness, color, gloss and reflective characteristics, without obvious regional differences or abnormal characteristics; heterogeneous coating means that the coating on the surface of plastic products has obvious non-uniformity or abnormal characteristics, which is manifested by significant changes in the texture, thickness, color or gloss of the material in local areas.

[0106] Homogeneous coating acquisition module: for plastic products that are judged to be homogeneous coatings, the scanner continues to collect surface state data with the original set beam width and focal length;

[0107] For plastic products that are judged to be homogeneous coatings, the original set beam width and focal length continue to be used to collect surface state data, in order to ensure that a high scanning efficiency is maintained when the coating is uniform. In this case, the coating consistency is good, and the scanner does not need to make complex beam width and focal length adjustments, and can directly collect data according to the initially set parameters. This not only avoids unnecessary parameter adjustments, saves time and computing resources during the scanning process, but also improves overall work efficiency. By maintaining constant scanning parameters, the surface information of the homogeneous coating can be quickly and accurately obtained, providing reliable real-time data for subsequent quality monitoring and production processes, while reducing the system burden and optimizing the operation of the overall production line.

[0108] The heterogeneous coating adjustment module dynamically adjusts the scanner’s beam width and focal length for plastic products that are judged to have heterogeneous coatings based on the deep learning model evaluation results;

[0109] For plastic products that are judged to have heterogeneous coatings, the scanner’s beam width and focal length are dynamically adjusted based on the deep learning model evaluation results. The specific steps are as follows:

[0110] If the plastic product being inspected is classified as a heterogeneous coating, the scanner parameters, especially the beam width and focal length, are dynamically adjusted based on the model evaluation results to ensure that more details can be captured in the area of ​​uneven coating. Based on the evaluation results of the deep learning model, the dynamically adjusted beam width and focal length will be refined and optimized according to the degree of uneven coating. The calculation expression is as follows:

[0111] Beam_width_dynamic=f(CCI,Beam_width_init),

[0112] Focus_dynamic=f(CCI,Focus_init)

[0113] , where Beam_width_dynamic is the adjusted beam width, which ensures that the scanner can finely capture the image data of the uneven coating area when processing heterogeneous coatings, Beam_width_init is the initial beam width, Focus_dynamic is the adjusted focal length value, Focus_init is the initial focal length, and CCI is the coating consistency index;

[0114] Based on the evaluation of coating non-uniformity, the scanner captures surface condition information more precisely by adjusting the beam width and focal length. In this process, the adjustment range will depend on the actual surface coating characteristics and the degree of coating non-uniformity in the target area. If the coating non-uniformity is high, the scanner's beam width may be reduced and the focal length may be shortened to obtain more details and surface texture information. The adjustment rule formula is as follows:

[0115] ΔBeam_width=A·(CCI ref -CCI),

[0116] ΔFocus=B·(CCI ref -CCI)

[0117] , where ΔBeam_width is the adjustment amount of the beam width, that is, the beam width value that needs to be increased or decreased based on the original set beam width, A is the beam width adjustment proportional factor, which indicates the strength of the relationship between the coating consistency index CCI difference and the beam width adjustment amount, CCI ref is the coating consistency index reference threshold, ΔFocus is the focal length adjustment amount, that is, the focal length value that needs to be changed to ensure that the scanner can focus clearly on the coating uneven area, and B is the focus adjustment scale factor, which indicates the strength of the relationship between the coating consistency index CCI difference and the focus adjustment amount;

[0118] Dynamic adjustments based on coating non-uniformity assessment help to accurately capture surface details and ensure high quality of scan data.

[0119] The purpose of calculating the adjusted beam width and focal length values ​​first and then the adjustment amount is to ensure that the scanner settings can be optimized according to the coating non-uniformity. The results are evaluated by the deep learning model, the target beam width and focal length are determined, and then the adjustment amount is calculated to ensure accurate adaptation to different coating states and avoid affecting other acquisition parameters. This can improve the accuracy of the scan results and the efficiency of adjustments.

[0120] To ensure optimal use of the adjusted parameters, when the scanner captures new data, the dynamically adjusted beam width and focal length will be iteratively optimized according to preset rules and feedback mechanisms to ensure that the best image quality can be obtained each time. The calculation expression is as follows:

[0121] Data_quality=f(Beam_width_dynamic,Focus_dynamic

[0122] , where Data_quality is the data quality, which is the final output variable and represents the quality of the image data obtained by dynamically adjusting the beam width and focal length.

[0123] By dynamically adjusting the scanner's beam width and focal length, the scanner will re-collect image data on the surface of the plastic product. The goal of this step is to optimize scanning efficiency and data quality under the adjusted parameters, ensure that details of uneven coating areas can be clearly captured, and avoid time waste and computing burden caused by excessive adjustments. The new scanning parameters ensure accurate image acquisition through intelligent control while maintaining high production efficiency.

[0124] For plastic products that are judged to have heterogeneous coatings, after the deep learning model evaluates the coating non-uniformity, the scanner needs to dynamically adjust its beam width and focal length based on the model's evaluation results. The main purpose of this step is to ensure that sufficient details and accurate data can be obtained in areas with uneven coatings to better identify surface defects such as bubbles, cracks, scratches, etc. Since areas with uneven coatings usually present large visual texture changes and reflection differences, fixed beam width and focal length may not be able to capture these details, so they need to be optimized based on the actual degree of coating non-uniformity.

[0125] Specifically, by dynamically adjusting the beam width, the scanner can focus more on local areas with uneven coatings, avoid overexposure or light saturation, and ensure that details in highly reflective areas remain clearly visible. Adjusting the focal length allows the scanner to maintain high image clarity at different surface heights and curvatures, thereby improving the accuracy and reliability of detection. Through this dynamic adjustment, the system can adapt to different coating states in real time, which not only improves the sensitivity of detection, but also improves scanning efficiency, avoids unnecessary repetitive operations and waste of resources. The implementation of this process can more accurately identify potential surface defects, ensure product quality, and reduce the flow of defective products into the market during the production process, thereby reducing quality risks and corporate losses.

[0126] The present invention significantly improves the detection accuracy of uneven coatings on plastic products by introducing deep learning models and image analysis technology. In traditional scanner detection methods, due to the fixed light beam and unchanged focal length, the excessive reflection and light saturation problems caused by uneven coatings will affect the sensor's capture of details, resulting in distortion or loss of image data, and thus making it impossible to identify potential defects in a timely manner. However, this solution dynamically analyzes the surface texture differentiation reference value and the surface reflection heterogeneity reference value, two key features that can quantify the uniformity of the coating and the difference in reflected light. Through the deep learning model, these features are intelligently evaluated. The model can accurately divide the coating status according to the coating consistency index, ensuring that more detailed and professional analysis can be performed in areas with uneven coatings. This intelligent detection not only makes up for the limitations of traditional methods, but also can identify and mark potential coating defects in advance, ensuring that unqualified products do not enter the production process, thereby greatly reducing quality risks and reducing the possibility of product recalls.

[0127] For plastic products with uneven coatings, the deep learning model of the present invention can analyze the coating consistency index in real time, and dynamically adjust the beam width and focal length of the scanner according to the model evaluation results. This intelligent adjustment process means that when uneven coating is detected, the scanner can automatically reduce the beam width and adjust the focal length, so as to accurately capture the details of the uneven coating. This strategy effectively avoids the drawbacks of traditional fixed parameter scanning, especially when facing excessive reflection or saturated areas, and can ensure the integrity and high quality of image data with more appropriate optical settings. In addition, the intelligent beam and focal length adjustment strategy also improves the inspection efficiency of the production line, avoids unnecessary repeated inspections, and thus saves time and costs. Through this dynamic adjustment, the system can efficiently respond to the coating status of different plastic products, maintaining high precision and high speed while ensuring the accuracy of detection.

[0128] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A plastic product detection system based on image recognition, characterized in that: It includes image acquisition module, data storage and feature extraction module, feature analysis and model input module, intelligent evaluation and classification module, homogeneous coating acquisition module and heterogeneous coating adjustment module: The image acquisition module, first, uses a scanner to illuminate the surface of the plastic product with a preset beam width and focal length to capture high-quality surface images; Data storage and feature extraction module, the collected surface images are stored and organized into a structured data set, and key features that help identify coating unevenness are extracted from the data set; The feature analysis and model input module analyzes the extracted key features under the detection window and inputs the analyzed key features into the pre-learned deep learning model, and uses the deep learning model to intelligently evaluate the coating status of the plastic product surface; Intelligent evaluation and classification module, which divides the surface state of plastic products into homogeneous coating and heterogeneous coating based on the evaluation results of deep learning model; Homogeneous coating acquisition module: for plastic products that are judged to be homogeneous coatings, the scanner continues to collect surface state data with the original set beam width and focal length; The heterogeneous coating adjustment module dynamically adjusts the scanner's beam width and focal length for plastic products judged to have heterogeneous coatings based on the deep learning model evaluation results.

2. A plastic product detection system based on image recognition according to claim 1, characterized in that: Key features that help identify coating unevenness are extracted from the data set. The extracted features include texture changes and surface reflected light gradients. Under the detection window, the extracted texture changes and surface reflected light gradients are analyzed to generate surface texture differentiation reference values ​​and surface reflection heterogeneity reference values, respectively. The surface texture differentiation reference values ​​quantify the degree of variation of the surface texture of plastic products in spatial distribution and local areas. The surface reflection heterogeneity reference values ​​quantify the differences in reflected light intensity in different areas of the plastic product surface, which is specifically manifested as the uneven distribution of surface reflectivity.

3. A plastic product detection system based on image recognition according to claim 2, characterized in that: In the detection window, the specific steps for analyzing texture changes and generating reference values ​​for surface texture differentiation are as follows: In the detection window, firstly, the texture features of the image data on the surface of the plastic product are extracted. Through the texture analysis of the local area, the texture pattern data of the area is extracted. According to the extracted texture features, the texture difference of each area in the image is calculated. The calculation expression is as follows: ， In the formula, is the value of the kth texture feature dimension of region i in the image, is the value of the kth texture feature dimension of region j in the image, α is the adjustment factor of the difference metric, N is the total number of texture features, represents the difference between region i and region j in the kth texture feature dimension, D(t i ,t j ) is the texture difference; Based on the results of texture difference measurement, the texture differences of different regions in the image are calculated, and the differences are weighted by the weight factor. The calculation expression is as follows: ， In the formula, Δ i is the regional texture difference, β j is the weight factor of the neighborhood region, is the weight factor of region j, is a set of neighborhood regions, is a set, representing the set of regions adjacent to region i; Combine all regional texture differences Δ i , generate the reference value of surface texture differentiation, the calculation expression is as follows: ， In the formula, I texture is the reference value of surface texture differentiation, M is the total number of regions, λ i is the adjustment factor for region i.

4. The plastic product detection system based on image recognition according to claim 2, characterized in that: Under the detection window, the specific steps for analyzing the surface reflected light gradient and generating the surface reflectance heterogeneity reference value are as follows: Under the detection window, the reflected light data of the surface of the plastic product is collected by the scanner to obtain the reflected light intensity value of each sampling point. The light gradient is obtained by calculating the light intensity change of adjacent pixel points. The calculation expression is as follows: ， Where R(x,y) is the intensity of the reflected light on the surface. is the reflected light intensity gradient, is the rate of change of reflected light intensity in the x direction, is the rate of change of reflected light intensity in the y direction; By calculating the light gradient of each pixel, the reflection unevenness of each local area is obtained. The calculation expression is as follows: ， Where L(x,y) is the measure of local light reflection inhomogeneity, θ is the sensitivity coefficient, is the nonlinear amplification factor; After calculating the local reflection heterogeneity metric L(x,y) of each pixel, the local areas within the entire detection window are summarized to obtain the global reflection heterogeneity metric. The calculation expression is as follows: ， Where ω(x,y) is the pixel weight, is the gradient attenuation factor, γ is the attenuation coefficient, I reflect It is the reference value of surface reflectance heterogeneity.

5. The plastic product detection system based on image recognition according to claim 2, characterized in that: After obtaining the surface texture differentiation reference value and the surface reflection heterogeneity reference value generated after analyzing the extracted key features, the surface texture differentiation reference value and the surface reflection heterogeneity reference value are input into a pre-learned deep learning model, and a coating consistency index is generated through the deep learning model. The coating consistency index is used to intelligently evaluate the coating status of the plastic product surface.

6. The plastic product detection system based on image recognition according to claim 5, characterized in that: The generated coating consistency index is compared and analyzed with the preset coating consistency index reference threshold, and the surface state of the plastic product is divided. The division steps are as follows: If the coating consistency index is greater than or equal to a preset coating consistency index reference threshold, the surface of the current plastic product is in an uneven state, and it is classified as a heterogeneous coating; If the coating consistency index is less than a preset coating consistency index reference threshold, the surface of the current plastic product is in a uniform state and is classified as a homogeneous coating.

7. The plastic product detection system based on image recognition according to claim 6, characterized in that: For plastic products that are judged to have heterogeneous coatings, the scanner’s beam width and focal length are dynamically adjusted based on the deep learning model evaluation results. The specific steps are as follows: If the plastic product being inspected is classified as a heterogeneous coating, the scanner parameters are dynamically adjusted based on the model evaluation results to ensure that more details can be captured in the area of ​​uneven coating. Based on the evaluation results of the deep learning model, the dynamically adjusted beam width and focal length will be refined and optimized according to the degree of uneven coating. The calculation expression is as follows: Beam_width_dynamic=f(CCI,Beam_width_init), Focus_dynamic=f(CCI,Focus_init), Where Beam_width_dynamic is the adjusted beam width, Beam_width_init is the initial beam width, Focus_dynamic is the adjusted focal length value, Focus_init is the initial focal length, and CCI is the coating consistency index; Based on the evaluation of coating non-uniformity, the scanner captures surface condition information more precisely by adjusting the beam width and focal length. In this process, the adjustment range will depend on the actual surface coating characteristics and the degree of coating non-uniformity in the target area. The adjustment rule formula is as follows: ΔBeam_width=A·(CCI ref -CCI), ΔFocus=B·(CCI ref -CCI), Where ΔBeam_width is the adjustment amount of the beam width, A is the beam width adjustment proportional factor, and CCI ref is the coating consistency index reference threshold, ΔFocus is the focal length adjustment amount, and B is the focal length adjustment proportional factor; To ensure optimal use of the adjusted parameters, when the scanner captures new data, the dynamically adjusted beam width and focal length will be iteratively optimized according to preset rules and feedback mechanisms to ensure that the best image quality can be obtained each time. The calculation expression is as follows: Data_quality=f(Beam_width_dynamic,Focus_dynamic) Where Data_quality is the data quality.

Images