A transparent bottle curved surface printed graphics and text detection system based on AI vision

Abstract

The invention discloses a transparent bottle curved surface printed graphic text detection system based on AI vision, which relates to the field of machine vision detection technology. The main scheme is: pre-processing the collected images, including denoising and enhancement, and being able to calculate the position deviation mean P according to the position accuracy data based on the characteristics of the transparent bottle curved surface, and evaluate whether the position accuracy meets the standard according to the position deviation mean P, and collect images with high unit pixel density and high frame rate to ensure that every subtle detail of the graphic text can be clearly recorded; from image acquisition to result output, the entire system realizes automatic operation without manual intervention, greatly improves the detection speed, and can quickly process a large number of transparent bottles to meet the needs of large-scale production. Each module of the system has strong adaptability and flexibility and can process different types of transparent bottles, graphic text and printing processes.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision detection, and in particular to an AI vision-based transparent bottle curved surface printed graphic text detection system. Background Art

[0002] With the development of science and technology, machine vision technology has been gradually applied to the field of printing quality inspection. Traditional inspection methods have poor adaptability, limited inspection accuracy, low intelligence, and are prone to misjudgment due to inaccurate data.

[0003] However, there are still some technical challenges in detecting printed graphics and text on the curved surface of transparent bottles. On the one hand, the curved surface characteristics of transparent bottles make image acquisition more difficult, and are prone to deformation, uneven lighting and other problems, which affect subsequent image processing and analysis. On the other hand, the diversity and complexity of printed graphics and text, as well as possible subtle defects such as color gradients and tiny stains, place higher requirements on the accuracy and robustness of the detection algorithm. Summary of the invention

[0004] 1. Technical issues to be resolved

[0005] In view of the shortcomings of the prior art, the present invention provides a transparent bottle curved surface printed graphic text detection system based on AI vision, which evaluates whether the text quality meets the standard according to the text quality evaluation score B, evaluates whether the position accuracy meets the standard according to the position deviation mean P, and evaluates whether the color consistency meets the standard according to the color deviation mean K. It can effectively avoid missed detection and false detection, ensure the accuracy and stability of the detection results, and solve the problems of traditional detection methods, such as low efficiency and susceptibility to human factors.

[0006] (II) Technical solution

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A transparent bottle curved surface printed graphic text detection system based on AI vision, comprising:

[0008] An image acquisition module is used to obtain graphic and text image data of the transparent bottle surface;

[0009] An image processing module, capable of adjusting different unit pixel densities according to the area A of the collected object to obtain graphic and text image data of the collected object;

[0010] The data analysis module uses AI vision technology to extract features from the preprocessed graphic text image data, obtains image clarity data, denoising effect data, enhancement effect data, position accuracy data and color data, calculates the image clarity C through the image clarity data, calculates the denoising quality Q through the denoising effect data, and calculates the enhancement effect E through the enhancement effect data; calculates the text quality assessment score B through the image acquisition clarity C, denoising quality Q and enhancement effect E, and evaluates whether the text quality meets the standard based on the text quality assessment score B; calculates the position deviation mean P through the position accuracy data, and evaluates whether the position accuracy meets the standard based on the position deviation mean P; calculates the color deviation mean K through the color data, and evaluates whether the color consistency meets the standard based on the color deviation mean K;

[0011] The result output module determines whether the printing effect is qualified based on the evaluation results of the data analysis module.

[0012] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic and text detection system: the method of adjusting the unit pixel density according to the area A is as follows:

[0013] Obtain parameter data of the acquisition device, the parameter data including the horizontal pixel number M and the vertical pixel number N;

[0014] The unit pixel density R is calculated by the horizontal pixel number M, the vertical pixel number N and the area A, based on the following formula:

[0015] ,

[0016] The value ranges of the horizontal pixel number M and the vertical pixel number N are both [100,5000].

[0017] In the preferred embodiment of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method for evaluating whether the text quality meets the standard is as follows:

[0018] The calculation of the text quality evaluation score B includes at least: image acquisition clarity C, denoising quality Q and enhancement effect E. The text quality evaluation score B is calculated by the image acquisition clarity C, denoising quality Q and enhancement effect E, and the formula is as follows:

[0019] ,

[0020] in: α is the weight coefficient of image acquisition clarity C, ranging from 0.2 to 0.3. β is the weight coefficient of denoising quality Q, ranging from 0.3 to 0.4. c is the weight coefficient of the enhancement effect E, ranging from 0.3 to 0.5.α + β + c =1;

[0021] A qualified text quality assessment score threshold DR is preset, and the text quality assessment score B is compared with the preset text quality assessment score threshold DR. If B≥DR, the text quality meets the standard; if B<DR, the text quality does not meet the standard.

[0022] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method for calculating the image acquisition clarity C is as follows:

[0023] Image resolution data includes sampling frequency f s and optical system aberration L; through the horizontal pixel number M, vertical pixel number N, sampling frequency f s The image acquisition clarity C is calculated based on the optical system aberration L, and the formula is as follows:

[0024] ,

[0025] Where: sampling frequency f s , the value range is [0,100], and the value range of the optical system aberration L is [0,10].

[0026] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method for calculating the denoising quality Q is as follows:

[0027] The denoising effect data includes the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H; the denoising quality Q is calculated by the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H, and the formula is as follows:

[0028] ,

[0029] Where: f(i, v) is the image pixel value before denoising; i represents the horizontal position coordinate of the pixel point in the image, and the value range of i is from 0 to the number of image rows T-1; v represents the vertical position coordinate of the pixel point in the image, and the value range of v is from 0 to the number of image columns H-1; g(i, v) is the image pixel value after denoising.

[0030] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method for calculating the enhancement effect E is as follows:

[0031] The enhancement effect data includes the original image pixel value z t and the enhanced image pixel value yt , through the original image pixel value z t and the enhanced image pixel value y t The enhancement effect E is calculated based on the following formula:

[0032] ,

[0033] Among them, the enhancement effect E The value range of z is [1,10]; t is the pixel of the tth original image, y t is the pixel after the tth enhancement, t represents the sequence number of the image, and its value is [1,n], and n is the number of specific samples in the image, and its value is a positive integer.

[0034] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method of position accuracy evaluation is as follows:

[0035] The location accuracy data includes at least: standard location coordinates r u and the detected position coordinates g u , through the standard position coordinates r u and the detected position coordinates g u The position deviation mean P is calculated based on the following formula:

[0036] ,

[0037] Where: r u is the position point of the u-th standard position coordinate, g u is the u-th detected position coordinate, u is the serial number of the position point, the value range is [1, s], s is the total number of position points involved in the position accuracy evaluation, and the value is a positive integer;

[0038] The preset qualified position deviation mean range is [DS, DH], and the position deviation mean P is compared with the qualified position deviation mean range [DS, DH]. If DS≤P≤DH, the position accuracy meets the standard; if P<DS or P>DH, the position accuracy does not meet the standard.

[0039] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printed graphic text detection system: the method for calculating the color consistency evaluation is as follows:

[0040] Color data at least includes: standard color red R value a j 、Standard color green G value c j 、Standard color blue B value e j , detected color red R value b j , detected color green G value d jand the detected color blue B value w j ;

[0041] The color deviation mean K is calculated from the color data according to the following formula:

[0042] ,

[0043] Among them: a j is the jth standard color sample, b j is the jth detected color sample, j is the serial number of the color sample, and its value range is [1, m], and m is the number of detected color samples, and its value range is [0, 255];

[0044] The preset qualified color deviation mean range is [DA, DB], and the color deviation mean K is compared with the qualified color deviation mean range. If DA≤K≤DB, the color consistency meets the standard; if K<DA or K>DB, the color consistency does not meet the standard.

[0045] In the preferred solution of the above-mentioned AI vision-based transparent bottle curved surface printing graphic text detection system: according to the evaluation results of the data analysis module, the method for judging whether the printing effect is qualified is as follows: when the text quality, position accuracy and color consistency all meet the following three conditions:

[0046] , it can be judged that the printing effect is qualified; otherwise it is unqualified.

[0047] (III) Beneficial effects

[0048] The present invention provides an AI vision-based transparent bottle curved surface printed graphic text detection system, which has the following beneficial effects:

[0049] (1) The image acquisition module can accurately obtain the graphic and text image data of the transparent bottle surface, ensuring the integrity and authenticity of the original information; this provides a reliable basis for subsequent analysis and detection, and avoids misjudgment caused by inaccurate data;

[0050] (2) The denoising and enhancement operations of the image processing module can effectively improve the image quality. The denoising operation removes the noise interference in the image, making the graphics and text clearer and more legible; the enhancement operation highlights the key features of the image and improves the contrast and brightness. These preprocessing operations greatly improve the readability and analyzability of the image, thereby improving the accuracy and reliability of detection.

[0051] (3) The data analysis module uses AI vision technology to conduct in-depth analysis of pre-processed images, obtain multi-dimensional data and perform precise calculations; by calculating indicators such as image clarity, denoising quality, enhancement effect, position deviation mean and color deviation mean, it can comprehensively evaluate the quality and printing effect of graphics and text; this scientific analysis method reduces the influence of human factors and makes the evaluation results more objective and accurate;

[0052] (4) The result output module makes accurate judgments based on the evaluation results of the data analysis module and can quickly determine whether the printing effect is qualified; this helps companies to discover problems in a timely manner and take corresponding measures to prevent unqualified products from entering the market and improve the overall quality level of products. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 The present invention is a schematic diagram of the working steps in a transparent bottle curved surface printed graphics and text detection system based on AI vision. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 , the present invention provides a transparent bottle curved surface printed graphic text detection system based on AI vision, including: an image acquisition module, which can adjust different unit pixel densities according to the area A of the collected object to obtain the graphic text image data of the collected object;

[0056] The method for calculating the pixel density of the acquisition unit is as follows:

[0057] The image definition data includes the horizontal pixel number M and the vertical pixel number N. The unit pixel density R is calculated by the horizontal pixel number M, the vertical pixel number N and the area A of the captured object. The formula is as follows:

[0058] ,

[0059] The horizontal pixel number M and the vertical pixel number N depend on the hardware parameters of the camera and have a value range of [100, 5000]. A is obtained by measuring the surface area of ​​the transparent bottle and has a value range of [100, 5000] pixels / square centimeter. If the surface area of ​​the object being photographed is small, a higher unit pixel density is required to ensure the image's detail expression capability. If the surface area of ​​the object being photographed is large, the unit pixel density can be reduced within a certain range.

[0060] It should be noted that it is derived from the basic principles of image acquisition and processing. In the image acquisition process, the number of pixels of the camera and the size of the object being photographed are two key factors that affect the unit pixel density of the image. By combining these two factors, a calculation formula that can reflect the unit pixel density of the image on a specific object can be obtained;

[0061] The image acquisition module can accurately obtain the graphic and text image data of the transparent bottle surface, ensuring the integrity and authenticity of the original information. This provides a reliable basis for subsequent analysis and detection, and avoids misjudgment caused by inaccurate data.

[0062] The image processing module is used to preprocess the graphic text image data, including denoising and enhancement operations; the data analysis module uses AI visual technology to extract features from the preprocessed graphic text image data, obtain image clarity data, denoising effect data, enhancement effect data, position accuracy data and color data, calculate the image clarity C through the image clarity data, calculate the denoising quality Q through the denoising effect data, and calculate the enhancement effect E through the enhancement effect data; calculate the text quality assessment score B through the image acquisition clarity C, denoising quality Q and enhancement effect E, and evaluate whether the text quality meets the standard based on the text quality assessment score B, calculate the position deviation mean P through the position accuracy data, and evaluate whether the position accuracy meets the standard based on the position deviation mean P, calculate the color deviation mean K through the color data, and evaluate whether the color consistency meets the standard based on the color deviation mean K;

[0063] The method for evaluating whether the text quality meets the standards is as follows:

[0064] The text quality assessment score B is calculated by the image acquisition clarity C, denoising quality Q and enhancement effect E, based on the following formula:

[0065] ,

[0066] in: α is the weight coefficient of image acquisition clarity C, ranging from 0.2 to 0.3. β is the weight coefficient of denoising quality Q, ranging from 0.3 to 0.4. c is the weight coefficient of the enhancement effect E, ranging from 0.3 to 0.5. α + β + c =1;

[0067] A qualified text quality assessment score threshold DR is preset, and the text quality assessment score B is compared with the preset text quality assessment score threshold DR. If B≥DR, the text quality meets the standard; if B<DR, the text quality does not meet the standard;

[0068] This formula is based on the different contributions of different image features to the overall quality. It adjusts the weight coefficient to highlight the impact of important features, thereby comprehensively evaluating the quality of the image. By taking the weighted sum of important features such as clarity, denoising effect, and enhancement effect, a more comprehensive quality assessment indicator can be obtained.

[0069] The method for calculating the image acquisition clarity C is as follows:

[0070] Image resolution data includes sampling frequency f s and optical system aberration L; through the horizontal pixel number M, vertical pixel number N, sampling frequency f s The image acquisition clarity C is calculated based on the optical system aberration L, and the formula is as follows:

[0071] ,

[0072] Among them: the number of horizontal pixels M and the number of vertical pixels N depend on the parameters of the camera, the sampling frequency fs is determined by the settings of the camera, and a higher sampling frequency can capture more image information, and the optical system aberration L is obtained by the design and measurement of the optical system;

[0073] It should be noted that the calculation principle of the image acquisition clarity C is to consider multiple factors comprehensively; the horizontal pixel number M and the vertical pixel number N reflect the image unit pixel density and spatial detail richness, plus the sampling frequency fs determined by the camera settings, to comprehensively reflect the information richness of the image; take the logarithm of the product of the three and multiply it by 20, compress the value range to make it more consistent with human eye perception, and adjust the value range and weight at the same time; finally divide it by the optical system aberration L to offset its negative impact on clarity, so as to accurately evaluate the image acquisition clarity; because the product of the horizontal pixel number M, the vertical pixel number N and the sampling frequency fs is a large value, by taking the logarithm, its range can be narrowed to an interval that is easier to handle and compare, making the clarity evaluation more reasonable, log10 multiplied by 20 is to further adjust the value range and weight, so that the final calculated clarity C is within a suitable range, which is convenient for feature extraction and comparison;

[0074] Preprocess the collected graphic and text images, including denoising and enhancement operations;

[0075] The method for calculating the denoising quality Q is as follows:

[0076] The denoising effect data includes the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H; the denoising quality Q is calculated by the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H, and the formula is as follows:

[0077] ,

[0078] Where: i in the original image pixel value f(i, v) represents the horizontal position coordinate of the pixel in the image, and the value range of i is from 0 to the number of image rows T-1; v in the original image pixel value f(i, v) represents the vertical position coordinate of the pixel in the image, and the value range of v is from 0 to the number of image columns H-1; this is because the coordinates of the image usually start counting from 0, and the maximum coordinate value is 1 less than the number of rows and columns of the image, respectively, to ensure that all pixels of the entire image can be traversed; the original image pixel value f(i, v) and the denoised image pixel value g(i, v) are obtained from the images before and after processing, and the number of image rows T and the number of image columns H are determined by the image size, and the value range is [0,255];

[0079] This formula measures the effect of the denoising operation by calculating the sum of the squares of the differences between the corresponding pixel values ​​of the original image and the denoised image, and then taking the square root of the average value. The principle is to compare the difference between the image before and after denoising. The smaller the difference, the better the denoising effect.

[0080] The method for calculating the enhancement effect E is as follows:

[0081] The enhancement effect data includes the original image pixel value z t The enhancement effect E is calculated by the original image pixel value zt and the enhanced image pixel value yt. The formula is as follows:

[0082] ,

[0083] Among them: the value range of the enhancement effect E is [1,10]. When the enhancement effect is weak, the ratio is close to 1, and the value of the enhancement effect E is close to 1; as the enhancement effect increases, the ratio increases, and the value of also increases accordingly, but generally the enhancement effect will not increase indefinitely, so there is an upper limit to the value range, which is in the range of [1,10]. If the value of the enhancement effect E is less than 1, it means that the contrast of the enhanced image is reduced, which rarely occurs in general image enhancement operations;

[0084] This formula measures the effect of image enhancement by calculating the average value of the ratio of the corresponding pixel values ​​of the enhanced image to the original image. The principle is to compare the degree of change of the image pixel values ​​before and after enhancement. The closer the ratio is to 1, the less obvious the enhancement effect is, and the larger the ratio is, the more significant the enhancement effect is.

[0085] The location accuracy assessment method is as follows:

[0086] The position accuracy data at least includes: standard position coordinates ru and detection position coordinates gu. The position deviation mean P is calculated by the standard position coordinates ru and detection position coordinates gu. The formula is as follows:

[0087] ,

[0088] Among them: the standard position coordinates ru are obtained through pre-set standards or measurements, and have no fixed value range, depending on the specific measurement standards and scenarios; the detected position coordinates gu are obtained based on actual detection and calculation, and have no fixed value range, depending on the actual detection results;

[0089] The preset qualified position deviation mean range is [DS, DH]. The position deviation mean P is compared with the qualified position deviation mean range [DS, DH]. If DS≤P≤DH, the position accuracy meets the standard. If P<DS or P>DH, the position accuracy does not meet the standard. The qualified position deviation mean range [DS, DH] is obtained according to industry standards, product requirements or customer requirements, and can be adjusted according to different application scenarios, printing requirements and industry standards.

[0090] It should be noted that the formula calculates the sum of the absolute values ​​of the deviations between each detected position and the standard position, and then divides it by the total number of position points involved in the evaluation to obtain the mean position deviation. The principle is to measure the overall deviation between the detected position and the standard position by the average deviation degree. The smaller the mean deviation, the higher the position accuracy.

[0091] The method to calculate the color consistency assessment is as follows:

[0092] The color data at least includes: a standard color red R value aj, a standard color green G value cj, a standard color blue B value ej, a detected color red R value bj, a detected color green G value dj, and a detected color blue B value wj;

[0093] The color deviation mean K is calculated from the color data according to the following formula:

[0094] ,

[0095] aj is the jth standard color sample, bj is the jth detected color sample, cj is the jth standard color sample, dj is the jth detected color sample, ej is the jth standard color sample, wj is the jth detected color sample, j is the sequence number of the color sample, and its value range is [1, m], m is the number of detected color samples, and its value range is [0, 255]; aj, cj, ej are determined according to product design requirements, industry standards or specific color specifications, and bj, dj, wj are obtained through actual color detection equipment or algorithms;

[0096] The qualified color deviation mean range is preset as [DA, DB], and the color deviation mean K is compared with the qualified color deviation mean range. If DA≤K≤DB, the color consistency meets the standard; if K<DA or K>DB, the color consistency does not meet the standard;

[0097] The formula calculates the square root of the sum of the squares of the differences between each test color sample and the standard color in the three RGB channels, and then takes the average of all samples to obtain the color deviation mean. The principle is to use the Euclidean distance in three-dimensional space to measure the difference between the test color and the standard color. The smaller the deviation mean, the higher the color consistency.

[0098] The denoising and enhancement operations of the image processing module can effectively improve the image quality; the denoising operation removes the noise interference in the image, making the graphics and text clearer and more legible; the enhancement operation highlights the key features of the image and improves the contrast and brightness; the correction operation corrects the deformation and distortion of the image to ensure that the shape of the graphics and text is accurate; these preprocessing operations greatly improve the readability and analyzability of the image, thereby improving the accuracy and reliability of detection.

[0099] The data analysis module uses AI vision technology to conduct in-depth analysis of pre-processed images, obtain multi-dimensional data and perform precise calculations; by calculating indicators such as image clarity, denoising quality, enhancement effect, position deviation mean and color deviation mean, it can comprehensively evaluate the quality and printing effect of graphics and text; this scientific analysis method reduces the influence of human factors and makes the evaluation results more objective and accurate.

[0100] The result output module determines whether the printing effect is qualified according to the evaluation results of the data analysis module;

[0101] According to the evaluation results of the data analysis module, the method for judging whether the printing effect is qualified is as follows: when the text quality, position accuracy and color consistency all meet the following three conditions:

[0102] ,

[0103] The printing effect can be judged to be qualified; otherwise, it is unqualified;

[0104] The result output module makes accurate judgments based on the evaluation results of the data analysis module and can quickly determine whether the printing effect is qualified; this helps companies to discover problems in a timely manner and take corresponding measures to avoid unqualified products from entering the market and improve the overall quality level of products.

[0105] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware or in combination with computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0106] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0107] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A transparent bottle curved surface printed graphics and text detection system based on AI vision, characterized by: include: The image acquisition module can adjust different resolutions according to the area A of the acquired object to acquire the graphic and text image data of the acquired object; Image processing module, used for preprocessing graphic and text image data, including denoising, enhancement and correction operations; The data analysis module uses AI vision technology to analyze the pre-processed graphic and text image data to obtain image clarity data, denoising effect data, enhancement effect data, position accuracy data and color data. The image clarity C is calculated by the image clarity data, the denoising quality Q is calculated by the denoising effect data, and the enhancement effect E is calculated by the enhancement effect data. The text quality evaluation score B is calculated through the image acquisition clarity C, denoising quality Q and enhancement effect E. Whether the text quality meets the standard is evaluated based on the text quality evaluation score B. The position deviation mean P is calculated through the position accuracy data. Whether the position accuracy meets the standard is evaluated based on the position deviation mean P. The color deviation mean K is calculated through the color data. Whether the color consistency meets the standard is evaluated based on the color deviation mean K. The result output module determines whether the printing effect is qualified according to the evaluation results of the data analysis module; The method to adjust the resolution according to area A is as follows: Obtain parameter data of the acquisition device, the parameter data including the horizontal pixel number M and the vertical pixel number N; The resolution R is calculated by the horizontal pixel number M, the vertical pixel number N and the area A, based on the following formula: ; Among them, the value range of the horizontal pixel number M and the vertical pixel number N is [100,5000]; The method for evaluating whether the text quality meets the standards is as follows: The text quality assessment score B is calculated by the image acquisition clarity C, denoising quality Q and enhancement effect E, based on the following formula: ; Among them: α is the weight coefficient of image acquisition clarity C, ranging from 0.2 to 0.3, β is the weight coefficient of denoising quality Q, ranging from 0.3 to 0.4, γ is the weight coefficient of enhancement effect E, ranging from 0.3 to 0.5, α+β+γ=1; A qualified text quality assessment score threshold DR is preset, and the text quality assessment score B is compared with the preset text quality assessment score threshold DR. If B≥DR, the text quality meets the standard; if B<DR, the text quality does not meet the standard; The method to calculate the color consistency assessment is as follows: Color data at least includes: standard color red R value a j 、Standard color green G value c j 、Standard color blue B value e j , detected color red R value b j , detected color green G value d j and the detected color blue B value w j ; The color deviation mean K is calculated from the color data according to the following formula: ; Among them: a j is the jth standard color sample, b j is the jth detected color sample, j is the serial number of the color sample, and its value range is 1, 2, 3...m, m is the number of detected color samples, and its value range is [0,255]; The preset qualified color deviation mean range is [DA, DB], and the color deviation mean K is compared with the qualified color deviation mean range. If DA≤K≤DB, the color consistency meets the standard; if K<DA or K>DB, the color consistency does not meet the standard.

2. According to claim 1, a transparent bottle curved surface printed graphics and text detection system based on AI vision is characterized by: The method for calculating the image acquisition clarity C is as follows: Image resolution data includes sampling frequency f s and optical system aberration L; through the horizontal pixel number M, vertical pixel number N, sampling frequency f s The image acquisition clarity C is calculated based on the optical system aberration L, and the formula is as follows: ; Where: sampling frequency f s , the value range is [10,100], and the value range of the optical system aberration L is [0,10].

3. The AI ​​vision-based transparent bottle curved surface printed graphics and text detection system according to claim 2, characterized in that: The method for calculating the denoising quality Q is as follows: The denoising effect data includes the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H; the denoising quality Q is calculated by the original image pixel value f(i, v), the denoised image pixel value g(i, v), the number of image rows T and the number of image columns H, and the formula is as follows: ; Where: f(i, v) is the image pixel value before denoising; i represents the horizontal position coordinate of the pixel point in the image, and the value range of i is from 0 to the number of image rows T-1; v represents the vertical position coordinate of the pixel point in the image, and the value range of v is from 0 to the number of image columns H-1; g(i, v) is the image pixel value after denoising.

4. According to claim 3, the AI ​​vision-based transparent bottle curved surface printed graphics and text detection system is characterized by: The method for calculating the enhancement effect E is as follows: The enhancement effect data includes the original image pixel value z t and the enhanced image pixel value y t , through the original image pixel value z t and the enhanced image pixel value y t The enhancement effect E is calculated based on the following formula: ; The value range of the enhancement effect E is [1,10]; t is the pixel of the tth original image, y t is the pixel after the tth enhancement, t represents the sequence number of the image, and its value is 1, 2, 3...n, and n is the number of specific samples in the image, and its value is a positive integer.

5. The AI ​​vision-based transparent bottle curved surface printed graphics and text detection system according to claim 4, characterized in that: The method for calculating the correction error J is as follows: The corrected error data includes the corrected ideal position p r and the actual corrected position q r , the ideal position after correction p r and the actual corrected position q r The correction error J is calculated based on the following formula: ; Where: p r is the pixel of the rth corrected pixel, q r is the pixel after the actual correction of the r-th pixel, r represents the serial number of the pixel, the value of r is 1, 2, 3, ... k, k is the total number of pixels involved in the correction error calculation, and the value is a positive integer.

6. The AI ​​vision-based transparent bottle curved surface printed graphics and text detection system according to claim 5, characterized in that: The location accuracy assessment method is as follows: The location accuracy data includes at least: standard location coordinates r u and the detected position coordinates g u , through the standard position coordinates r u and the detected position coordinates g u The position deviation mean P is calculated based on the following formula: ; Where: r u is the position point of the u-th standard position coordinate, g u is the u-th detection position coordinate, u is the serial number of the position point, and its value range is 1, 2, 3...s, s is the total number of position points involved in the position accuracy evaluation, and its value is a positive integer; The preset qualified position deviation mean range is [DS, DH], and the position deviation mean P is compared with the qualified position deviation mean range [DS, DH]. If DS≤P≤DH, the position accuracy meets the standard; if P<DS or P<DH, the position accuracy does not meet the standard.

7. The AI ​​vision-based transparent bottle curved surface printed graphics and text detection system according to claim 6, characterized in that: According to the evaluation results of the data analysis module, the method for judging whether the printing effect is qualified is as follows: when the text quality, position accuracy and color consistency all meet the following three conditions: ; The printing effect can be judged to be qualified; otherwise, it is unqualified.

Images