Data Analysis-Based Glass Non-Destructive Testing System

Abstract

This invention discloses a glass non-destructive testing system based on data analysis. The invention relates to the field of glass testing technology. It obtains a set of test images of a glass product by repeatedly capturing images of the bottle body during a step-rotating process using an image acquisition module. A test data interaction module crops several images within the test image set, reducing its size. This accelerates the defect detection speed and efficiency of a single glass product. A defect analysis unit analyzes stored defect analysis data to determine redundant image numbers that can be quickly identified. This ensures that the overlapping areas retained in the test image set are the overlapping areas with the optimal recognition rate, reducing the transmission rate of the test image set and saving resources for the glass defect detection model in recognizing image data from a single test image set.

Description

Technical Field

[0001] This invention relates to the field of glass testing technology, specifically to a glass non-destructive testing system based on data analysis. Background Technology

[0002] In modern glass manufacturing, non-destructive testing is a crucial step in ensuring product quality. To comprehensively inspect surface defects in glass bottles, glassware, and other products, existing technologies often employ a rotating imaging inspection scheme. This involves using a turntable-type bottle-picking device to rotate the glass product step-by-step within a horizontal plane. After each rotation at a certain angle, a pair of cameras, illuminated by a light source, capture images of the bottle, thus obtaining a complete 360° image. While this scheme effectively covers the inspection area, it inevitably produces a large amount of image overlap—the overlapping area between adjacent shooting angles typically accounts for 20% to 50% of a single image.

[0003] In the subsequent defect identification stage, repeatedly detecting all overlapping regions would lead to a waste of computational resources and a decrease in detection efficiency. To address this issue, existing methods typically employ an "overlapping region deduplication" strategy, which involves retaining only one copy of the data for analysis of repeatedly captured image portions. However, due to significant differences in lighting conditions, image clarity, and feature representation within the same overlapping region under different shooting angles, the model's recognition rate for each overlapping region varies. In large-scale detection scenarios, this rate difference can lead to an increase in overall processing time, severely limiting the throughput of the detection system. Randomly selecting and retaining data from a particular overlapping region may further reduce detection speed due to the accidental selection of inefficient recognition data. Existing deduplication schemes do not consider the differences in recognition efficiency among overlapping regions, which may result in further reductions in detection speed due to the accidental selection of inefficient recognition data.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] The purpose of this invention is to provide a glass non-destructive testing system based on data analysis, in order to solve the problems mentioned in the background art.

[0006] This invention provides a data analysis-based nondestructive testing system for glass, comprising:

[0007] The detection data interaction module is used to transmit the detection image set of each glass product to the model detection unit when there are no redundant image numbers stored. The detection image set contains a number of image data, and each image data corresponds to a number number, which starts from 1 and continues sequentially.

[0008] The detection data interaction module is also used to, after storing a number of redundant image numbers, for each detection image set of a glass product received, for any image data in the detection image set, first search whether there is a redundant image number stored in the detection data interaction module that matches the numerical number of the image data, and select whether to update the image data in the detection image set based on the search result.

[0009] The model detection unit is used to input the set of detection images of the glass product into the glass defect detection model after receiving the set of images, and the glass defect detection model outputs the detection result data of the glass product, which contains the result signal.

[0010] The result signal is selected from the numbers 1 and 0. When the result signal is 1, the detection result data also includes several defect image data of identified defects and the corresponding identification rate. The defects identified in the defect image data are marked with red boxes, and the corresponding defect type is marked on the red boxes.

[0011] For any glass product, if the result signal contained in the test result data of the glass product is 1, then the defect analysis data of the glass product is generated based on the test result data and the test image set of the glass product.

[0012] The defect analysis unit is used to analyze all the defect analysis data stored in it to obtain a number of redundant image numbers after the amount of defect analysis data stored in it reaches a fixed amount.

[0013] Furthermore, the steps for updating several image data within a set of inspected glass artifacts are as follows:

[0014] S11: Label all image data in the detected image set as N1, N2, ..., NP5 in ascending order of their numerical designations;

[0015] S12: In the detection data interaction module, search for whether redundant image numbers 1-1 and 1-2 are stored. Based on the search result, select whether to update the image data N1 in the detection image set, as follows:

[0016] If the detection data interaction module stores redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will not be updated. If the detection data interaction module stores only redundant image number 1-1, then the image data N1 in the detection image set will be updated according to the preset first truncation update rule. If the detection data interaction module stores only redundant image number 1-2, then the image data N1 in the detection image set will be updated according to the preset second truncation update rule. If the detection data interaction module does not store redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will be updated according to the preset third truncation update rule.

[0017] Furthermore, the first interception update rules are as follows:

[0018] SS21: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is to the right horizontally, and the positive y-axis is to the bottom vertically. Pixel coordinates are represented by integers.

[0019] SS22: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1];

[0020] SS23: Determine the right sub-coordinate (B1-1-3*B1*P2,0) of image data N1, extract the right sub-redundant image of image data N1 from the image data N1 according to the right sub-coordinate (B1-1-3*B1*P2,0), and replace the image data N1 in the detection image set with the remaining image data N1 after extraction.

[0021] Furthermore, the second interception and update rules are as follows:

[0022] SS31: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is horizontal to the right, and the positive y-axis is vertical downward.

[0023] SS32: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1];

[0024] SS33: Determine the left sub-coordinates (R1*P2, R2) of image data N1, extract the left sub-redundant image of image data N1 from the image data N1 according to the left sub-coordinates (R1*P2, R2), and replace the image data N1 in the detection image set with the remaining image data N1 after extraction.

[0025] Compared with existing technologies, it has the following advantages:

[0026] This invention obtains a set of inspection images of a glass product by taking multiple photos of the bottle body during step-rotation using an image acquisition module. Before the inspection image set is transmitted to the cloud inspection platform and input into the glass defect detection model, a detection data interaction module is set up to crop several image data in the inspection image set, reducing the size of the image data in the inspection image set. In this way, the speed and efficiency of defect detection of a single glass product are accelerated.

[0027] This invention analyzes several defect analysis data stored within a defect analysis unit. By analyzing the recognition rate of different defect types in overlapping areas of adjacent image data and the number of different defect types identified in adjacent image data, redundant image numbers that can be quickly identified are determined. In this way, the overlapping area images retained in the image data of the detection image set are the overlapping area images with the optimal recognition rate. This not only reduces the transmission rate of the detection image set but also saves the recognition resources of the glass defect detection model for the image data in a single detection image set, while ensuring the recognition efficiency of the glass defect detection model for defect types in the detection image set. Attached Figure Description

[0028] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Please see Figure 1 This application provides a glass non-destructive testing system based on data analysis, including an image acquisition module, a test data interaction module, a cloud-based testing platform, and a defect alarm module;

[0031] The image acquisition module is used to acquire images of several glass products on the annealing furnace conveyor line. The process of acquiring images of glass products is accomplished by a detection device and a bottle-picking device. The detection device consists of a camera and a light source, and the bottle-picking device is a turntable-type bottle-picking device that can drive the glass products to rotate.

[0032] For a glass product, a rotary bottle-picking device is used to rotate the glass product stepwise in a horizontal plane, with each rotation angle being P1. Under the illumination of a light source, a camera captures images of the glass product to obtain several image data of the glass product. The value of P1 is preset by the management personnel according to the properties of the glass product, the manufacturing process, the position of the light source, and the field of view of the camera. In this application, the value of P1 is in the range [15, 22.5].

[0033] The camera takes a picture of each rotation.

[0034] In this application, based on several image data of any glass product obtained by shooting, each image data is assigned a numerical number according to the shooting order. The numerical number starts from 1 and proceeds sequentially. The earlier the shooting order, the smaller the corresponding numerical number.

[0035] A detection image set of the glass product is generated based on several image data of the glass product. At this time, the detection image set contains several image data of the glass product. In this application, the glass product is circularly symmetrical.

[0036] The image acquisition module transmits the set of test images of the glass product to the test data interaction module;

[0037] The detection data interaction module is used to process and interact with the detection image set of glass products, and the steps are as follows:

[0038] S11: Label all image data in the detected image set as N1, N2, ..., NP5 in ascending order of their numerical designations;

[0039] S12: In the detection data interaction module, search for whether redundant image numbers 1-1 and 1-2 are stored. Based on the search result, select whether to update the image data N1 in the detection image set, as follows:

[0040] If the detection data interaction module stores redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will not be updated. If the detection data interaction module only stores redundant image number 1-1, then the image data N1 in the detection image set will be updated according to the preset first truncation update rule. If the detection data interaction module only stores redundant image number 1-2, then the image data N1 in the detection image set will be updated according to the preset second truncation update rule. If the detection data interaction module does not store redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will be updated according to the preset third truncation update rule.

[0041] The first truncation and update rules are as follows:

[0042] SS21: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is to the right horizontally, and the positive y-axis is to the bottom vertically. Pixel coordinates are represented by integers.

[0043] SS22: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1], and the units of R1 and R2 are pixels;

[0044] SS23: Determine the right sub-coordinates (B1-1-3*B1*P2,0) of image data N1, specifically:

[0045] B1-1-3*B1*P2 is used as the right width of image data N1, R2 is used as the right height of image data N1, and P2 is the preset width ratio. The value of P2 is set by the administrator according to the value of P1.

[0046] SS24: Extract the right sub-redundant image of the image data N1 from the image data N1 according to the right sub-coordinates (B1-1-3*B1*P2,0). The right sub-redundant image is a rectangle with a width of B1-1-3*B1*P2 pixels and a height of R2 pixels. The coordinates from the upper left corner to the lower right corner are (B1-1-3*B1*P2,0) to (B1, B2).

[0047] Replace the image data N1 in the detection image set with the remaining image data N1 after cropping;

[0048] The second truncation and update rules are as follows:

[0049] SS31: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is to the right horizontally, and the positive y-axis is downward vertically. Pixel coordinates are represented by integers.

[0050] SS32: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1], and the units of R1 and R2 are pixels;

[0051] SS33: Determine the left child coordinates (R1*P2, R2) of image data N1, specifically:

[0052] R1*P2 is used as the left width of image data N1, R2 is used as the left height of image data N1, and P2 is the preset width ratio. The value of P2 is set by the administrator according to the value of P1. It should be noted that, based on the rotation angle of the step rotation, the image data captured by the camera after each rotation will have a fixed size of repeated area.

[0053] SS34: Extract the left redundant image of the image data N1 from the image data N1 according to the left child coordinates (R1*P2, R2). The left redundant image is a rectangle with a width of R1*P2 pixels and a height of R2 pixels, and the coordinates from the upper left corner to the lower right corner are (0,0) to (R1*P2, R2).

[0054] Replace the image data N1 in the detection image set with the remaining image data N1 after cropping;

[0055] The third interception and update rules are as follows:

[0056] SS41: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is to the right horizontally, and the positive y-axis is to the bottom vertically. Pixel coordinates are represented by integers.

[0057] SS42: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1], and the units of R1 and R2 are pixels;

[0058] SS43: Determine the right sub-coordinates (B1-1-3*B1*P2,0) and left sub-coordinates (R1*P2,R2) of image data N1;

[0059] SS44: Extract the right sub-redundant image and the left sub-redundant image of the image data N1 sequentially from the image data N1 according to S24 and S34;

[0060] Replace the image data N1 in the detection image set with the remaining image data N1 after cropping;

[0061] S13: Following S12, sequentially search within the detection data interaction module to see if redundant image numbers 2-1 and 2-2, 3-1 and 3-2, ..., P5-1 and P5-2 are stored. Based on the search results, select and update the image data N2, N3, ..., NP5 in the detection image set accordingly.

[0062] S14: After selecting and updating the image data NP5 in the detection image set, the current detection image set is obtained and transmitted to the cloud detection platform;

[0063] A cloud-based inspection platform is used to perform defect inspection on glass products on an annealing furnace conveyor line in the cloud. The cloud-based inspection platform includes a model inspection unit and a defect analysis unit.

[0064] The model detection unit has a pre-trained glass defect detection model. The glass defect detection model is designed to accurately identify various defects on or inside the surface of glass products. Defect types include, but are not limited to, bubbles, various impurities, cracks, tin spots, and surface roughness.

[0065] After receiving the transmitted set of inspection images of the glass product, the cloud inspection platform transmits it to the model inspection unit. After receiving the transmitted set of inspection images of the glass product, the model inspection unit uses it as input to the glass defect detection model. The glass defect detection model outputs the inspection result data of the glass product, which includes the result signal.

[0066] The result signal is selected from the numbers 1 and 0. When the result signal is 1, the detection result data also includes several defect image data of identified defects and the corresponding identification rate. The defects identified in the defect image data are marked with red boxes, and the corresponding defect type is marked on the red boxes.

[0067] For any glass product, if the result signal contained in the test result data of the glass product is 1, then the test result data of the glass product is transmitted to the defect alarm module.

[0068] The defect alarm module is used to issue defect alarms to management personnel. After receiving the transmitted inspection result data of the glass product, the defect alarm module extracts all the defect image data contained therein, displays all the extracted defect image data to the management personnel for viewing, and issues alarms to the management personnel.

[0069] For any glass product, if the result signal contained in the test result data of the glass product is 1, then the defect analysis data of the glass product is selected and generated according to the test result data and the test image set, and the defect analysis data is transmitted to the defect analysis unit. It should be noted that steps S21-S25 are only executed when the defect analysis unit has not analyzed and generated several redundant image numbers.

[0070] The generation steps are as follows:

[0071] S21: For any defect image data in the test result data of the glass product, obtain the numerical code of the defect image data when it was used as image data before defect labeling, and establish an association between the numerical code and the defect image data;

[0072] S22: After establishing a correlation between all defect image data in the detection result data and their numerical identifiers, the defect image data are sequentially labeled as A1, A2, ..., Aa, where a ≥ 1, according to the numerical identifiers in ascending order.

[0073] S23: Extract the left and right sub-redundant images and their hierarchical numbers from the defect image data A1 according to the preset extraction rules. The extraction rules are as follows:

[0074] S231: The origin Z1(0,0) is the top left corner of the defect image data A1, the positive x-axis is to the right horizontally, and the positive y-axis is to the bottom vertically. Pixel coordinates are represented by integers.

[0075] S232: Obtain the width B1 and height B2 of the defect image data A1, where x∈[0,B1−1], y∈[0,B2−1], and the units of B1 and B2 are pixels;

[0076] S233: Determine the left sub-coordinates (B1*P2, B2) of defect image data A1, specifically:

[0077] B1*P2 is used as the left width of defect image data A1, B2 is used as the left height of defect image data A1, P2 is the preset width ratio, and the value of P2 is set by the management personnel according to the value of P1. It should be noted that, based on the rotation angle of the step-rotation, the image data captured by the camera after each rotation will have a fixed size of repeated area.

[0078] S234: Extract the left redundant image of the defect image data A1 from the defect image data A1 according to the left child coordinates (B1*P2, B2). The left redundant image is a rectangle with a width of B1*P2 pixels and a height of B2 pixels, and the coordinates from the upper left corner to the lower right corner are (0,0) to (B1*P2, B2).

[0079] Synchronously, based on the numerical identifier of the defective image data A1, the left sub-redundant image is associated with the hierarchical identifier P3-1, where P3 is the numerical identifier of the acquired defective image data A1.

[0080] In the hierarchical numbering, the number before the "-" is the primary level, and the number after the "-" is the secondary level.

[0081] S235: According to S231 to S234, extract the right sub-redundant image of the defect image data A1. The right sub-redundant image is a rectangle with a width of B1*P2 pixels and a height of B2 pixels. The coordinates from the upper left corner to the lower right corner are (B1-1-3*B1*P2,0) to (B1, B2). Simultaneously, establish an association between the right sub-redundant image and the layer number P3-2 according to the numerical number of the defect image data A1.

[0082] S24: According to S23, extract the corresponding left sub-redundant image and right sub-redundant image from the defect image data A2, A3, ..., Aa respectively;

[0083] S25: The left and right sub-redundant images of the defect image data A1, A2, ..., Aa are traversed sequentially. For each left or right sub-redundant image that is marked with a red box, the left or right sub-redundant image is extracted. At the same time, the recognition rate of the defect image data of the extracted left or right sub-redundant image is obtained, and the recognition rate is used as the feature rate of the extracted left or right sub-redundant image.

[0084] Defect analysis data for the glass product is generated based on all extracted left and right sub-redundant images, their hierarchical numbers, and feature rates.

[0085] It should be noted here that if after traversing all the left and right sub-redundant images of the defect image data A1, A2, ..., Aa, there is no defect marked with a red box in any of the left or right sub-redundant images, then no processing is performed.

[0086] The defect analysis unit receives and stores the test result data of the glass product after receiving it.

[0087] Once the defect analysis unit has stored a fixed amount of defect analysis data for all glass products, it analyzes all the defect analysis data. The analysis steps are as follows:

[0088] S31: Randomly select one defect type as the type to be analyzed from all the defect types that the glass defect detection model can identify. Obtain all the defect analysis data of the type to be analyzed from several left sub-redundant images or right sub-redundant images contained in the defect analysis unit, and label them as C1, C2, ..., Cc, where c≥1;

[0089] S32: Create a right sub-variable F1 and a left sub-variable F2 related to the number 1, with the initial values ​​of F1 and F2 being 0;

[0090] S33: Calculate and obtain the redundancy evaluation indices M1 and M2 for the lower-level numbers 1-2 and 2-1 of the type to be analyzed according to the preset calculation rules. The calculation rules are as follows:

[0091] S331: Extract all left and right sub-redundant images from defect analysis data C1 that only indicate the type to be analyzed;

[0092] S332: Extract the feature rates G1 and H2 of the right sub-redundant image and left sub-redundant image with level numbers 1-2 and 2-1 from the level numbers corresponding to all extracted left and right sub-redundant images. If only the right sub-redundant image with level number 1-2 can be extracted from the level numbers corresponding to all extracted left and right sub-redundant images, the value of the right sub-variable F1 is incremented by 1. If only the left sub-redundant image with level number 2-1 can be extracted from the level numbers corresponding to all extracted left and right sub-redundant images, the value of the left sub-variable F2 is incremented by 1. It should be noted that if there are no left and right sub-redundant images with level numbers 1-2 and 2-1 among the level numbers corresponding to all extracted left and right sub-redundant images, the value of the feature rates F1 and F2 is 0.

[0093] S333: Extract the feature rates of the right sub-redundancy image and the left sub-redundancy image with level numbers 1-2 and 2-1 from the defect analysis data C2, C3, ..., Cc in sequence according to S331 to S332. After extracting the feature rates of the right sub-redundancy image and the left sub-redundancy image with level numbers 1-2 and 2-1 from the defect analysis data Cc, obtain the values ​​L1 and L2 of the right sub-variable F1 and the left sub-variable F2 at this time.

[0094] The feature rates of the right sub-redundant images with level numbers 1-2 in the defect analysis data C2, C3, ..., Cc are labeled as G2, G3, ..., Gc, and the feature rates of the left sub-redundant images with level numbers 2-1 in the defect analysis data C2, C3, ..., Cc are labeled as H2, H3, ..., Hc.

[0095] S334: Calculate and obtain the rate representation quantities J1 and J2 for levels 1-2 and 2-1 sequentially according to the preset calculation rules:

[0096] The calculation rules for the rate characterization quantity J1 are as follows: using the formula Calculate the rate deviation I1 of the characteristic rates G1, G2, ..., Gc, and compare I1 with I. Here, I is the preset rate deviation threshold of level number 1-2, and Gi represents each of the characteristic rates G1, G2, ..., Gc.

[0097] If I1≥I, then delete the corresponding Gi in descending order of |Gi - G| and calculate the rate deviation I1 of the remaining Gi. Compare the magnitudes of I1 and I again until I1 < I. Obtain the average value of all the characteristic rates involved in the calculation of I1 at this time, and calibrate this average value as the rate characterization quantity J1 of level number 1 - 2;

[0098] The calculation rule of the rate characterization quantity J2 is as follows: Use the formula to calculate and obtain the rate deviation K1 of the characteristic rates H1, H2,..., Hc. Compare the magnitudes of K1 and K. Here, K is the preset rate deviation threshold of level number 2 - 1, and Hi represents each of the characteristic rates H1, H2,..., Hc;

[0099] If K1≥K, then delete the corresponding Hi in descending order of |Hi - H| and calculate the rate deviation K1 of the remaining Hi. Compare the magnitudes of K1 and K again until K1 < K. Obtain the average value of all the characteristic rates involved in the calculation of K1 at this time, and calibrate this average value as the rate characterization quantity J2 of level number 2 - 1;

[0100] S335: Use the formula M1 = L1×ɑ1 + J1×P4×ɑ2 to calculate and obtain the redundancy evaluation index M1 of level number 1 - 2 under the type to be analyzed. In the formula, ɑ1 and ɑ2 are the preset first and second proportion weights respectively, and P4 is the preset dimension factor, whose function is to perform a standardized conversion on the calculation dimension of the rate characterization quantity to ensure that features with different dimensions can perform numerical operations on a unified scale;

[0101] Use the formula M2 = L2×ɑ1 + J2×P4×ɑ2 to calculate and obtain the redundancy evaluation index M2 of level number 2 - 1 under the type to be analyzed;

[0102] S34: Select successively all the defect types that the glass defect detection model can identify as the types to be analyzed, and calculate and obtain the redundancy evaluation indexes of level number 1 - 2 and 2 - 1 under all these defect types in sequence according to S33 to S34. Calculate the average value Z1 of the sum of the redundancy evaluation indexes of level number 1 - 2 under all these defect types. Similarly, calculate the average value Z2 of the sum of the redundancy evaluation indexes of level number 2 - 1 under all these defect types;

[0103] S35: Compare the magnitudes of Z1 and Z2. If Z1≥Z2, then determine that level number 1 - 2 is the redundant image number; otherwise, determine that level number 2 - 1 is the redundant image number;

[0104] S36: Redundant image numbering is determined sequentially for layer numbers 2-2 and 3-1, 3-2 and 4-1, ..., P5-2 and 1-1 according to S33-S35. Here, P5 is the total number of image data contained in the detection image set of any glass product. It should be noted that the process of shooting the glass product is to rotate the glass product step by step in the horizontal plane. Therefore, the total number of image data contained in the detection image set of each glass product is consistent.

[0105] S37: After step S36 is completed, obtain all redundant image numbers and transmit them to the detection data interaction module for storage;

[0106] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0107] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A glass non-destructive testing system based on data analysis, characterized in that, include: The detection data interaction module is used to transmit the detection image set of each glass product to the model detection unit when there are no redundant image numbers stored. The detection image set contains a number of image data, and each image data corresponds to a number number, which starts from 1 and continues sequentially. The detection data interaction module is also used to, after storing a number of redundant image numbers, for each detection image set of a glass product received, for any image data in the detection image set, first search whether there is a redundant image number stored in the detection data interaction module that matches the numerical number of the image data, and select whether to update the image data in the detection image set based on the search result. The model detection unit is used to input the set of detection images of the glass product into the glass defect detection model after receiving the set of images, and the glass defect detection model outputs the detection result data of the glass product, which contains the result signal. The result signal is selected from the numbers 1 and 0. When the result signal is 1, the detection result data also includes several defect image data of identified defects and the corresponding identification rate. The defects identified in the defect image data are marked with red boxes, and the corresponding defect type is marked on the red boxes. For any glass product, if the result signal contained in the test result data of the glass product is 1, then the defect analysis data of the glass product is generated based on the test result data and the test image set of the glass product. The defect analysis unit is used to analyze all the defect analysis data stored in it to obtain a number of redundant image numbers after the amount of defect analysis data stored in it reaches a fixed amount.

2. The glass non-destructive testing system based on data analysis according to claim 1, characterized in that, It also includes an image acquisition module, which is used to take several pictures of the glass product bottle body during step-rotation to obtain a set of detection images of the glass product.

3. The glass non-destructive testing system based on data analysis according to claim 1, characterized in that, The steps for generating defect analysis data for the glass product are as follows: S21: For any defect image data in the test result data of the glass product, obtain the numerical code of the defect image data when it was used as image data before defect labeling, and establish an association between the numerical code and the defect image data; S22: After establishing a correlation between all defect image data in the detection result data and their numerical identifiers, the defect image data are sequentially labeled as A1, A2, ..., Aa, where a ≥ 1, according to the numerical identifiers in ascending order. S23: Extract the left and right sub-redundant images and their hierarchical numbers from the defect image data A1 according to the preset extraction rules. The extraction rules are as follows: S231: The origin of the coordinate system is Z1(0,0), with the upper left corner of the defect image data A1 as the origin. The positive x-axis is horizontal to the right and the positive y-axis is vertical downward. S232: Obtain the width B1 and height B2 of the defect image data A1, where x∈[0,B1−1], y∈[0,B2−1], and the units of B1 and B2 are pixels; S233: Determine the left sub-coordinate (B1*P2, B2) of the defect image data A1, where P2 is the preset width ratio; S234: Extract the left redundant image of the defect image data A1 from the defect image data A1 according to the left child coordinates (B1*P2, B2), and establish an association between the left redundant image and the level number P3-1 according to the numerical number of the defect image data A1, where P3 is the numerical number of the obtained defect image data A1, and the number before "-" in the level number is the primary level and the number after "-" is the secondary level; S235: Extract the right sub-redundant image of the defective image data A1 from the defective image data A1 according to S231 to S234, and establish an association between the right sub-redundant image and the level number P3-2 according to the numerical number of the defective image data A1. S24: Extract the corresponding left and right sub-redundant images from the defect image data A2, A3, ..., Aa respectively, according to S23; S25: The left and right sub-redundant images of the defect image data A1, A2, ..., Aa are traversed sequentially. For each left or right sub-redundant image that is marked with a red box, the left or right sub-redundant image is extracted. At the same time, the recognition rate of the defect image data of the extracted left or right sub-redundant image is obtained, and the recognition rate is used as the feature rate of the extracted left or right sub-redundant image. Defect analysis data for the glass product is generated based on all extracted left and right sub-redundant images, their hierarchical numbers, and characteristic rates.

4. The glass non-destructive testing system based on data analysis according to claim 3, characterized in that, The steps to obtain several redundant image numbers through analysis are as follows: S31: Randomly select one defect type as the type to be analyzed from all the defect types that the glass defect detection model can identify. Obtain all the defect analysis data of the type to be analyzed from several left sub-redundant images or right sub-redundant images contained in the defect analysis unit, and label them as C1, C2, ..., Cc, where c≥1; S32: Create right sub-variable F1 and left sub-variable F2 for number 1, with the initial values ​​of F1 and F2 being 0; S33: Calculate and obtain the redundancy evaluation indices M1 and M2 for the lower-level numbers 1-2 and 2-1 of the type to be analyzed according to the preset calculation rules. The calculation rules are as follows: S331: Extract all left and right sub-redundancy images from defect analysis data C1, only those labeled with the type to be analyzed; S332: Extract the feature rates G1 and H2 of the right sub-redundant image and the left sub-redundant image with level numbers 1-2 and 2-1 from the level numbers corresponding to all the extracted left and right sub-redundant images. If only the right sub-redundant image with level number 1-2 can be extracted from the level numbers corresponding to all the extracted left and right sub-redundant images, then the value of the right sub-variable F1 is incremented by 1. If only the left sub-redundant image with level number 2-1 can be extracted from the level numbers corresponding to all the extracted left and right sub-redundant images, then the value of the left sub-variable F2 is incremented by 1. S333: Extract the feature rates of the right sub-redundancy image and the left sub-redundancy image with level numbers 1-2 and 2-1 from the defect analysis data C2, C3, ..., Cc in sequence according to S331 to S332. After extracting the feature rates of the right sub-redundancy image and the left sub-redundancy image with level numbers 1-2 and 2-1 from the defect analysis data Cc, obtain the values ​​L1 and L2 of the right sub-variable F1 and the left sub-variable F2 at this time. The feature rates of the right sub-redundant images with level numbers 1-2 in the defect analysis data C2, C3, ..., Cc are labeled as G2, G3, ..., Gc, and the feature rates of the left sub-redundant images with level numbers 2-1 in the defect analysis data C2, C3, ..., Cc are labeled as H2, H3, ..., Hc. S334: Calculate and obtain the rate representation quantities J1 and J2 for levels 1-2 and 2-1 sequentially according to the preset calculation rules: S335: Use the formula M1=L1×ɑ1+J1×P4×ɑ2 to calculate the redundancy evaluation index M1 for the lower level number 1-2 of the type to be analyzed, where ɑ1 and ɑ2 are the preset first and second proportion weights, respectively, and P4 is the preset dimension factor. Use the formula M2=L2×ɑ1+J2×P4×ɑ2 to calculate the redundancy evaluation index M2 for the lower level number 2-1 of the type to be analyzed. S34: Select all defect types that the glass defect detection model can identify as the types to be analyzed in sequence. Calculate the redundancy evaluation indexes of the lower-level numbers 1-2 and 2-1 of all defect types in sequence according to S33 to S34. Calculate the average value Z1 of the sum of the redundancy evaluation indexes of the lower-level numbers 1-2 of all defect types. Calculate the average value Z2 of the sum of the redundancy evaluation indexes of the lower-level numbers 2-1 of all defect types. S35: Compare the size of Z1 and Z2. If Z1≥Z2, then determine that level number 1-2 is a redundant image number; otherwise, determine that level number 2-1 is a redundant image number. S36: Redundant image numbering is determined sequentially for layer numbers 2-2 and 3-1, 3-2 and 4-1, ..., P5-2 and 1-1 according to S33-S35, where P5 is the total number of image data contained in the detection image set of glass products; S37: After step S36 is completed, obtain all redundant image numbers and transmit them to the detection data interaction module for storage.

5. The glass non-destructive testing system based on data analysis according to claim 4, characterized in that, In S332, if there are no left and right sub-redundant images with level numbers 1-2 and 2-1 in the level numbers corresponding to all extracted left and right sub-redundant images, the values ​​of feature rates F1 and F2 are 0.

6. The glass non-destructive testing system based on data analysis according to claim 5, characterized in that, The steps for updating several image data within an image set of a glass product inspection are as follows: S11: Label all image data in the detected image set as N1, N2, ..., NP5 in ascending order of their numerical designations; S12: In the detection data interaction module, search for whether redundant image numbers 1-1 and 1-2 are stored. Based on the search result, select whether to update the image data N1 in the detection image set, as follows: If the detection data interaction module stores redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will not be updated. If the detection data interaction module stores only redundant image number 1-1, then the image data N1 in the detection image set will be updated according to the preset first truncation update rule. If the detection data interaction module stores only redundant image number 1-2, then the image data N1 in the detection image set will be updated according to the preset second truncation update rule. If the detection data interaction module does not store redundant image numbers 1-1 and 1-2, then the image data N1 in the detection image set will be updated according to the preset third truncation update rule.

7. The glass non-destructive testing system based on data analysis according to claim 6, characterized in that, The first truncation and update rules are as follows: SS21: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is to the right horizontally, and the positive y-axis is to the bottom vertically. Pixel coordinates are represented by integers. SS22: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1]; SS23: Determine the right sub-coordinate (B1-1-3*B1*P2,0) of image data N1, extract the right sub-redundant image of image data N1 from the image data N1 according to the right sub-coordinate (B1-1-3*B1*P2,0), and replace the image data N1 in the detection image set with the remaining image data N1 after extraction.

8. The glass non-destructive testing system based on data analysis according to claim 6, characterized in that, The second truncation and update rules are as follows: SS31: The origin Q1(0,0) is the top left corner of the image data N1, the positive x-axis is horizontal to the right, and the positive y-axis is vertical downward. SS32: Obtain the width R1 and height R2 of image data N1, where x∈[0,R1−1], y∈[0,R2−1]; SS33: Determine the left sub-coordinates (R1*P2, R2) of image data N1, extract the left sub-redundant image of image data N1 from the image data N1 according to the left sub-coordinates (R1*P2, R2), and replace the image data N1 in the detection image set with the remaining image data N1 after extraction.

Images