An intelligent control system and method for gluing line graphic recognition and automatic waste disposal

Abstract

The application discloses an intelligent control system and method for glue line image-text recognition and automatic waste disposal, and relates to the technical field of computer vision. The system comprises an image acquisition module, an image preprocessing module, a glue line area positioning module, an image-text recognition module, a defect detection module, an intelligent decision module and a waste disposal execution module. The image acquisition module is used for acquiring image information of a printed matter to be detected. The image preprocessing module is used for carrying out denoising and enhancement processing on the acquired image. The glue line area positioning module is used for identifying and positioning a region where the glue line is located. The image-text recognition module is used for recognizing image-text information in the glue line region. The defect detection module is used for detecting printing defects in the glue line region. The intelligent decision module is used for generating a waste disposal decision according to the recognition result and the defect detection result. The waste disposal execution module is used for executing an automatic waste disposal operation according to the waste disposal decision.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, specifically to an intelligent control system and method for perfect binding line graphic recognition and automatic waste removal. Background Technology

[0002] Perfect binding lines are a common technical feature in the printing and binding industry, widely used in the binding process of printed materials such as books, magazines, and brochures. The accuracy and integrity of the position of the perfect binding lines, as well as the printing quality of the text and images within the perfect binding line area, directly affect the overall quality of the printed materials and the user experience. During the printing production process, due to factors such as equipment precision, material characteristics, and environmental factors, various defects such as misalignment, breakage, blurred text and images, missing text and images, and stains may occur in the perfect binding line area.

[0003] In existing technologies, the following methods are mainly used for quality inspection and defect removal in the perfect binding area:

[0004] The first method is manual visual inspection, which relies on operators to visually inspect each printed product and manually remove unqualified products after defects are found. This method has problems such as low inspection efficiency, high labor intensity, inconsistent inspection standards, and easy to miss or falsely detect. Especially on high-speed printing production lines, manual inspection is difficult to meet the production cycle requirements.

[0005] The second method is detection based on simple sensors. This method uses photoelectric sensors to detect the basic features of printed materials, such as size and position. This method can only detect the macroscopic features of printed materials and cannot identify graphic information and minor defects in the perfect binding area, so the detection accuracy is limited.

[0006] The third method is detection based on traditional image processing. This method uses industrial cameras to acquire images and employs traditional image processing algorithms such as edge detection and template matching to detect defects. This method has high requirements for image quality and is sensitive to factors such as changes in lighting and deviations in the position of printed materials. The algorithm is not robust enough and is difficult to adapt to complex production environments.

[0007] The aforementioned existing technologies all have certain limitations, failing to achieve accurate identification of graphic information in the perfect binding area and intelligent detection of printing defects, and are unable to meet the modern printing industry's demand for high-efficiency, high-precision, and intelligent quality control;

[0008] Therefore, there is an urgent need for a technical solution that can automatically identify perfect binding lines and intelligently remove defects, in order to improve the efficiency and accuracy of printed product quality inspection, reduce labor costs, and ensure product quality. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent control system and method for recognizing and automatically removing glue binding lines.

[0010] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control system for perfect binding line graphic recognition and automatic waste removal, comprising: an image acquisition module, an image preprocessing module, a perfect binding line area positioning module, a graphic recognition module, a defect detection module, an intelligent decision-making module, and a waste removal execution module;

[0011] The image acquisition module is used to acquire image information of the printed matter to be inspected on the conveyor belt through an industrial camera, and transmit the image information to the image preprocessing module;

[0012] The image preprocessing module is used to perform grayscale conversion, noise reduction, and contrast enhancement on the image information to generate a preprocessed image;

[0013] The perfect binding line area positioning module is used to extract features from the preprocessed image based on the target detection algorithm, identify and locate the area where the perfect binding line is located, and generate a perfect binding line area image.

[0014] The image recognition module is used to recognize text and pattern information in the image of the perfect binding line area and generate image recognition results;

[0015] The defect detection module is used to extract defect features from the image of the perfect binding line area, detect the type and location of printing defects, and generate defect detection results.

[0016] The intelligent decision-making module is used to calculate a quality score based on the image and text recognition results and the defect detection results, and to generate a waste discharge decision signal based on the comparison result of the quality score and the preset quality threshold.

[0017] The waste removal execution module is used to receive the waste removal decision signal. When the waste removal decision signal is a waste removal instruction, it controls the waste removal mechanism to remove the corresponding printed material from the conveyor belt.

[0018] As a further embodiment of the present invention: the perfect binding line area localization module includes: a feature extraction unit, a region candidate unit, and a boundary regression unit;

[0019] The feature extraction unit is used to perform multi-scale feature extraction on the preprocessed image using a convolutional neural network to generate a feature map.

[0020] The region candidate unit is used to generate glue binding line candidate regions based on the feature map, and to calculate a confidence score for each candidate region;

[0021] The boundary regression unit is used to perform boundary box regression on candidate regions with confidence scores greater than a preset confidence threshold to determine the precise coordinates of the binding line region.

[0022] As a further aspect of the present invention: the image and text recognition module includes: a text recognition unit and a pattern recognition unit;

[0023] The text recognition unit is used to perform text detection and text recognition on the text region in the perfect binding line area image, and output the text content and text position information.

[0024] The pattern recognition unit is used to extract pattern features and match patterns in the pattern area of ​​the glued seam area image, and output pattern category and pattern completeness information.

[0025] As a further aspect of the present invention: the types of printing defects detected by the defect detection module include: perfect binding line misalignment defects, perfect binding line breakage defects, blurred graphics defects, missing graphics defects, and stain defects;

[0026] The defect detection module calculates a defect severity value for each defect type, and the defect severity value is an integer ranging from 0 to 100.

[0027] As a further aspect of the present invention: the intelligent decision-making module calculates the quality score in the following way:

[0028] The image recognition accuracy score is multiplied by the first weighting coefficient to obtain the first weighted score, the defect detection score is multiplied by the second weighting coefficient to obtain the second weighted score, and the first weighted score and the second weighted score are added together to obtain the quality score.

[0029] The first weighting coefficient is 0.4, and the second weighting coefficient is 0.6.

[0030] This invention also provides an intelligent control method for perfect binding line graphic recognition and automatic waste removal, comprising the following steps:

[0031] S1. Acquire image information of the printed materials to be inspected on the conveyor belt using an industrial camera;

[0032] S2. Perform grayscale conversion, noise reduction, and contrast enhancement on the image information to generate a preprocessed image;

[0033] S3. Based on the target detection algorithm, feature extraction is performed on the preprocessed image to identify and locate the area where the binding line is located, and an image of the binding line area is generated.

[0034] S4. Recognize the text and pattern information in the image of the glued seam area and generate a text and image recognition result;

[0035] S5. Extract defect features from the image of the perfect binding line area, detect the printing defect type and defect location, and generate defect detection results.

[0036] S6. Calculate a quality score based on the image recognition result and the defect detection result, and generate a waste discharge decision signal based on the comparison result of the quality score and the preset quality threshold.

[0037] S7. When the waste discharge decision signal is a waste discharge command, the waste discharge mechanism is controlled to remove the corresponding printed material from the conveyor belt.

[0038] As a further aspect of the present invention: the specific process of identifying and locating the area where the perfect binding line is located in step S3 includes:

[0039] S31. A convolutional neural network is used to extract multi-scale features from the preprocessed image to generate a feature map;

[0040] S32. Generate candidate regions for perfect binding lines based on the feature map, and calculate a confidence score for each candidate region;

[0041] S33. Select candidate regions with confidence scores greater than the preset confidence threshold as valid candidate regions;

[0042] S34. Perform non-maximum suppression processing on the effective candidate regions to remove overlapping regions;

[0043] S35. Perform bounding box regression on the candidate regions after non-maximum suppression to determine the precise coordinates of the binding line region.

[0044] As a further aspect of the present invention: the specific process of recognizing text information in step S4 includes:

[0045] S41. Perform text region detection on the image of the perfect binding line area and locate the sub-region where the text is located;

[0046] S42. Perform character segmentation on each text sub-region to obtain a single character image;

[0047] S43. Use a character recognition model to recognize each individual character image to obtain the character recognition result;

[0048] S44. Concatenate all character recognition results according to their position order to generate a complete text recognition result.

[0049] As a further aspect of the present invention: the specific process for detecting the type of printing defects in step S5 includes:

[0050] S51. Register and align the image of the perfect binding line area with the pre-stored standard template image.

[0051] S52. Calculate the difference between the registered perfect binding line area image and the standard template image;

[0052] S53. Perform threshold segmentation on the difference image to extract the difference region;

[0053] S54. Perform morphological feature analysis on each differential region to determine the corresponding defect type;

[0054] S55. Calculate the severity value of the defect based on the area and location information of the difference area.

[0055] As a further aspect of the present invention: the specific process of calculating the quality score in step S6 includes:

[0056] S61. Calculate the image recognition accuracy score based on the degree of matching between the image recognition result and the standard image information. The image recognition accuracy score is an integer ranging from 0 to 100.

[0057] S62. Calculate the defect deduction value based on the defect detection results. The defect deduction value is the sum of the defect severity values ​​of all detected defects.

[0058] S63. Calculate the defect detection score, where the defect detection score is equal to 100 minus the defect deduction value. When the calculation result is less than 0, the defect detection score is 0.

[0059] S64. Multiply the image recognition accuracy score by 0.4 to obtain the first weighted score, and multiply the defect detection score by 0.6 to obtain the second weighted score;

[0060] S65. The first weighted score and the second weighted score are added together to obtain the quality score;

[0061] S66. When the quality score is less than the preset quality threshold, a waste discharge instruction is generated;

[0062] When the quality score is greater than the preset quality threshold, a pass instruction is generated;

[0063] The preset quality threshold value is 60.

[0064] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0065] First, this invention achieves automatic recognition of images and text in the binding line area and intelligent waste removal of defective products through the coordinated work of an image acquisition module, an image preprocessing module, a binding line area positioning module, an image and text recognition module, a defect detection module, an intelligent decision-making module, and a waste removal execution module. Compared with manual inspection methods, this invention significantly improves inspection efficiency and accuracy.

[0066] Second, the present invention employs a perfect binding line area localization module based on target detection algorithm, which performs multi-scale feature extraction through convolutional neural network, and can accurately identify and locate the area where the perfect binding line is located, and has strong robustness to factors such as changes in lighting and deviations in the position of printed matter.

[0067] Third, the graphic recognition module of the present invention includes a text recognition unit and a pattern recognition unit, which can simultaneously recognize text information and pattern information in the perfect binding line area, and the recognition results are comprehensive and accurate.

[0068] Fourth, the defect detection module of the present invention can detect five types of defects: glue binding thread misalignment defect, glue binding thread breakage defect, image and text blurring defect, image and text missing defect, and stain defect, and calculate the defect severity value for each defect, realizing refined defect detection and quantitative evaluation.

[0069] Fifth, the intelligent decision-making module of this invention uses a weighted scoring method to calculate the quality score, taking into account both the accuracy of image and text recognition and the results of defect detection. The decision result is scientific and reasonable, and can effectively distinguish between qualified and unqualified products.

[0070] Sixth, the waste discharge execution module of the present invention can automatically control the waste discharge mechanism to remove unqualified products from the conveyor belt according to the waste discharge decision signal, realizing a fully automated quality control process and reducing labor costs. Attached Figure Description

[0071] Figure 1 This is an overall architecture diagram of an intelligent control system for graphic recognition and automatic waste removal of glue binding lines, provided in an embodiment of the present invention.

[0072] Figure 2 This is a schematic diagram of the structure of the perfect binding line area positioning module provided in an embodiment of the present invention.

[0073] Figure 3 This is a schematic diagram of the image recognition module provided in an embodiment of the present invention.

[0074] Figure 4 The flowchart illustrates an intelligent control method for identifying and automatically removing patterns from glue binding lines, as provided in an embodiment of the present invention.

[0075] Figure 5 A detailed flowchart of the positioning of the perfect binding line area provided in an embodiment of the present invention.

[0076] Figure 6 A detailed flowchart of defect detection provided for embodiments of the present invention.

[0077] Figure 7 A flowchart for calculating the quality score provided in an embodiment of the present invention. Detailed Implementation

[0078] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0079] Example 1

[0080] like Figure 1 As shown, this embodiment provides an intelligent control system for perfect binding line graphic recognition and automatic waste removal. This system is applied to the quality inspection stage of the printing production line to realize graphic recognition of the perfect binding line area of ​​printed materials and automatic waste removal of defects. The system includes: an image acquisition module, an image preprocessing module, a perfect binding line area positioning module, a graphic recognition module, a defect detection module, an intelligent decision-making module, and a waste removal execution module.

[0081] The image acquisition module is located above the inspection station on the printing production line. It is used to acquire image information of the printed materials to be inspected on the conveyor belt through an industrial camera and transmit the image information to the image preprocessing module.

[0082] In this embodiment, the industrial camera is a high-resolution industrial area scan camera with a resolution of 2048 pixels by 1536 pixels and a frame rate of 60 frames per second. The industrial camera is used in conjunction with a ring light source, which is installed around the camera lens to provide uniform and stable lighting conditions. The industrial camera is connected to the system's main control computer via a gigabit Ethernet interface to achieve high-speed transmission of image data.

[0083] The image acquisition module also includes a trigger control unit, which is connected to the photoelectric encoder of the conveyor belt. When the printed material moves to the inspection station, the trigger control unit sends a trigger signal to the industrial camera, triggering the camera to acquire the current frame image. This external trigger acquisition method can ensure that each printed material is accurately photographed at a fixed position, avoiding motion blur.

[0084] The image preprocessing module is used to perform grayscale conversion, noise reduction, and contrast enhancement on image information to generate a preprocessed image.

[0085] Grayscale conversion converts a color image into a grayscale image, reducing the computational load of subsequent processing. Grayscale conversion uses a weighted average method, which assigns different weights to the red, green, and blue channels based on the differences in human eye sensitivity to different colors, and then sums them up.

[0086] The noise reduction process uses Gaussian filtering, which smooths the image by using a Gaussian convolution kernel to remove random noise. The size of the Gaussian convolution kernel is 5 pixels by 5 pixels, and the standard deviation of the Gaussian kernel is 1.2.

[0087] The contrast enhancement process employs histogram equalization, which redistributes the grayscale values ​​of the image to enhance the overall contrast and make image details clearer. For images with uneven local contrast, an adaptive histogram equalization method is further used for local enhancement.

[0088] like Figure 2 As shown, the glue binding line region localization module is used to extract features from the preprocessed image based on the target detection algorithm, identify and locate the region where the glue binding line is located, and generate a glue binding line region image. The glue binding line region localization module includes: a feature extraction unit, a region candidate unit, and a boundary regression unit.

[0089] The feature extraction unit is used to extract multi-scale features from the preprocessed image using a convolutional neural network to generate a feature map.

[0090] In this embodiment, the convolutional neural network adopts a residual network structure, including an input layer, multiple residual blocks, and an output layer. The residual blocks contain convolutional layers, batch normalization layers, and activation function layers. The convolutional neural network outputs feature maps of multiple different scales to achieve effective detection of targets of different sizes.

[0091] Region candidate units are used to generate glue binding line candidate regions based on feature maps and to calculate a confidence score for each candidate region.

[0092] In this embodiment, the region candidate unit uses an anchor frame mechanism to generate candidate regions. Multiple predefined anchor frames are set at each position of the feature map. The size and aspect ratio of the anchor frames are designed according to the typical size of the perfect binding line. For each anchor frame, the degree of matching with the actual perfect binding line region is calculated, and the confidence score is output.

[0093] The boundary regression unit is used to perform bounding box regression on candidate regions with confidence scores greater than a preset confidence threshold, thereby determining the precise coordinates of the binding line region.

[0094] In this embodiment, the preset confidence threshold is set to 0.7. Bounding box regression predicts the center point offset and width-height scaling ratio of the bounding box through a regression network, and finely adjusts the bounding box of the candidate region to make it more accurately fit the actual binding line region boundary.

[0095] like Figure 3 As shown, the image recognition module is used to recognize text and pattern information in the image of the binding line area and generate image recognition results. The image recognition module includes a text recognition unit and a pattern recognition unit.

[0096] The text recognition unit is used to detect and recognize text in the text region of the perfect binding line area image, and output the text content and text position information.

[0097] In this embodiment, the text recognition unit first uses a text detection network to locate the text region in the binding line area image. The text detection network outputs the bounding box coordinates of the text region. Then, each text region is segmented into individual characters. Finally, a character recognition network is used to recognize each character and output the character category. The character recognition network uses a convolutional neural network structure and can recognize Chinese characters, English letters, and numbers.

[0098] The pattern recognition unit is used to extract pattern features and match patterns in the pattern region of the perfect binding line image, and output pattern category and pattern completeness information.

[0099] In this embodiment, the pattern recognition unit uses a feature extraction network to extract the depth feature vector of the pattern region. Then, it performs similarity matching between the extracted feature vector and the pre-stored standard pattern feature library to determine the pattern category. At the same time, by comparing the similarity between the current pattern features and the standard pattern features, the pattern completeness information is calculated. The pattern completeness information represents the degree of matching between the current pattern and the standard pattern.

[0100] The defect detection module is used to extract defect features from the image of the perfect binding line area, detect the type and location of printing defects, and generate defect detection results.

[0101] In this embodiment, the printing defect types detected by the defect detection module include: perfect binding line misalignment defect, perfect binding line breakage defect, blurred image and text defect, missing image and text defect, and stain defect.

[0102] A perfect binding line offset defect refers to a defect in which the actual position of the perfect binding line deviates from the standard position. The defect detection module determines whether an offset defect exists by calculating the distance between the actual position of the perfect binding line and the standard position of the perfect binding line. When the offset distance is greater than a preset offset threshold, a perfect binding line offset defect is determined to exist. The preset offset threshold is set to 2 mm.

[0103] A broken binding thread defect refers to a defect in which the binding thread is discontinuous or interrupted. The defect detection module analyzes the connectivity of the binding thread area to detect whether the binding thread is broken. When the number of connected areas of the binding thread is greater than 1, it is determined that a broken binding thread defect exists.

[0104] Image and text blurring defect refers to the defect where the text and patterns in the perfect binding area are not clearly printed. The defect detection module determines whether blurring defect exists by calculating the clarity index of the image and text area. When the clarity index is lower than the preset clarity threshold, it is determined that image and text blurring defect exists.

[0105] The text and image missing defect refers to the absence of text and images that should be present within the perfect binding area. The defect detection module compares the current text and image area with a standard template to detect whether there is missing content. When the proportion of missing area is greater than the preset missing threshold, it is determined that there is a text and image missing defect.

[0106] Stain defects refer to defects in the area of ​​perfect binding lines where there are abnormal stains and impurities. The defect detection module determines whether there are stain defects by detecting abnormal areas in the image. When an abnormal area with an area larger than the preset stain area threshold is detected, it is determined that there are stain defects.

[0107] The defect detection module calculates a defect severity value for each defect type. The defect severity value is an integer ranging from 0 to 100. A defect severity value of 0 indicates that there is no defect of that type, while a defect severity value of 100 indicates that the defect of that type is the most severe.

[0108] The intelligent decision-making module is used to calculate the quality score based on the image and text recognition results and the defect detection results, and to generate a waste discharge decision signal based on the comparison result of the quality score and the preset quality threshold.

[0109] In this embodiment, the intelligent decision-making module calculates the quality score by multiplying the image recognition accuracy score by the first weight coefficient to obtain the first weighted score, multiplying the defect detection score by the second weight coefficient to obtain the second weighted score, and adding the first weighted score and the second weighted score to obtain the quality score. The first weight coefficient is 0.4 and the second weight coefficient is 0.6.

[0110] The text recognition accuracy score is calculated based on the degree of matching between the text recognition result and the standard text information. The specific calculation method is as follows: count the ratio of the number of correctly recognized characters to the total number of characters, and multiply it by 100 to obtain the text recognition accuracy.

[0111] The ratio of the number of correctly matched patterns to the total number of patterns is calculated and multiplied by 100 to obtain the pattern recognition accuracy.

[0112] The average of the text recognition accuracy and the pattern recognition accuracy is used to obtain the image and text recognition accuracy score.

[0113] The defect detection score is calculated based on the defect detection results. The specific calculation method is as follows: add up the defect severity values ​​of all detected defects to obtain the defect deduction value.

[0114] Subtract the defect deduction value from 100 to get the defect detection score;

[0115] When the calculation result is less than 0, the defect detection score is 0.

[0116] The preset quality threshold is set to 60. When the quality score is less than the preset quality threshold, the intelligent decision-making module generates a waste discharge instruction.

[0117] When the quality score is greater than the preset quality threshold, the intelligent decision-making module generates a qualified instruction.

[0118] The formula for calculating the quality score is as follows:

[0119] ;

[0120] in, To rate the quality, The image recognition accuracy score is given. The score is the defect detection score. This is the first weighting coefficient, with a value of 0.4. This is the second weighting coefficient, with a value of 0.6.

[0121] The formula for calculating the defect detection score is as follows:

[0122] ;

[0123] in, The score is the defect detection score. For the first The severity value of each defect type This represents the total number of defect types, with a value of 5.

[0124] The waste removal execution module is used to receive waste removal decision signals. When the waste removal decision signal is a waste removal command, it controls the waste removal mechanism to remove the corresponding printed material from the conveyor belt.

[0125] In this embodiment, the waste removal execution module includes a waste removal controller and a waste removal mechanism. The waste removal controller is communicatively connected to the intelligent decision module and receives waste removal decision signals. The waste removal mechanism is installed at the waste removal station of the conveyor belt and includes a pneumatic push rod and a waste removal channel. When the waste removal controller receives a waste removal command, it controls the pneumatic push rod to extend and push the corresponding unqualified printed product from the conveyor belt into the waste removal channel. When the waste removal controller receives a qualified command, the pneumatic push rod remains in the retracted state, and the qualified printed product continues to move along the conveyor belt to the next process.

[0126] The waste removal execution module also includes a position tracking unit, which is connected to the photoelectric encoder of the conveyor belt to track the position of each printed product on the conveyor belt in real time. When the printed product moves to the waste removal station, the position tracking unit sends a position arrival signal to the waste removal controller, triggering the waste removal controller to perform the corresponding waste removal operation.

[0127] Example 2

[0128] like Figure 4 As shown, this embodiment provides an intelligent control method for perfect binding thread graphic recognition and automatic waste removal. This method is applied to the intelligent control system of Embodiment 1 and includes the following steps:

[0129] S1. Acquire image information of the printed materials to be inspected on the conveyor belt using an industrial camera.

[0130] In this step, when the printed material moves to the inspection station with the conveyor belt, the photoelectric encoder detects the arrival of the printed material and sends a trigger signal to the industrial camera. The industrial camera acquires the current frame image as the image information to be inspected. The acquired image information is an RGB color image with a resolution of 2048 pixels by 1536 pixels.

[0131] S2. Perform grayscale conversion, noise reduction, and contrast enhancement on the image information to generate a preprocessed image.

[0132] In this step, the RGB color image is first converted to a grayscale image to reduce the computational load of subsequent processing. Then, Gaussian filtering is used to denoise the grayscale image and remove random noise. Finally, histogram equalization is used to enhance the contrast of the image, improving its clarity and contrast, and generating a preprocessed image.

[0133] S3. Based on the target detection algorithm, feature extraction is performed on the preprocessed image to identify and locate the area where the binding line is located, and an image of the binding line area is generated.

[0134] like Figure 5 As shown, the specific process of this step includes:

[0135] S31. Use a convolutional neural network to extract multi-scale features from the preprocessed image and generate a feature map.

[0136] S32. Generate candidate regions for glue binding lines based on the feature map, and calculate the confidence score for each candidate region.

[0137] S33. Select candidate regions with confidence scores greater than the preset confidence threshold as valid candidate regions. The preset confidence threshold is set to 0.7.

[0138] S34. Perform non-maximum suppression processing on the effective candidate regions to remove overlapping regions. Non-maximum suppression processing is used to remove duplicate detections of the same target and retain the candidate regions with the highest confidence.

[0139] S35. Perform bounding box regression on the candidate regions after non-maximum suppression to determine the precise coordinates of the binding line region. Bounding box regression finely adjusts the bounding boxes of the candidate regions by predicting the center point offset and the width and height scaling ratio of the bounding boxes.

[0140] Based on the determined coordinates of the binding line area, the binding line area image is cropped from the preprocessed image for subsequent image recognition and defect detection.

[0141] S4. Recognize the text and pattern information in the image of the glue binding line area and generate the image and text recognition results.

[0142] The specific process of recognizing text information in this step includes:

[0143] S41. Perform text region detection on the image of the perfect binding line area and locate the sub-region where the text is located.

[0144] S42. Perform character segmentation on each text sub-region to obtain a single character image.

[0145] S43. Use a character recognition model to recognize each individual character image to obtain the character recognition result.

[0146] S44. Concatenate all character recognition results according to their position order to generate a complete text recognition result.

[0147] In this step, the specific process of identifying pattern information includes: extracting pattern features from the pattern area in the perfect binding line area image and generating a pattern feature vector;

[0148] The pattern feature vector is matched with a pre-stored standard pattern feature library to determine the pattern category;

[0149] Calculate the similarity between the current pattern features and the standard pattern features to generate pattern integrity information.

[0150] S5. Extract defect features from the image of the perfect binding line area, detect the type and location of printing defects, and generate defect detection results.

[0151] like Figure 6 As shown, the specific process of this step includes:

[0152] S51. Register and align the image of the glue binding area with the pre-stored standard template image. The registration and alignment adopts the feature point matching method. First, feature points are detected in the current image and the standard template image respectively. Then, the transformation matrix between the images is calculated by feature point matching. Finally, the current image is registered to the coordinate system of the standard template image according to the transformation matrix.

[0153] S52. Calculate the difference image between the registered glue binding line area image and the standard template image. The difference image is obtained by calculating the difference in gray values ​​at corresponding pixel positions in the two images.

[0154] S53. Perform threshold segmentation on the difference image to extract the difference region. The threshold segmentation marks pixels with difference values ​​greater than the preset difference threshold as difference regions and pixels with difference values ​​less than the preset difference threshold as normal regions.

[0155] S54. Perform morphological feature analysis on each differential region to determine the corresponding defect type. Morphological features include the area, shape, location, and grayscale distribution characteristics of the differential region. Based on these morphological features, the differential regions are classified into five defect types: glue binding offset defect, glue binding breakage defect, blurred image and text defect, missing image and text defect, and stain defect.

[0156] S55. Calculate the defect severity value based on the area and location information of the difference area. The defect severity value is directly proportional to the area of ​​the difference area and inversely proportional to the distance of the difference area from the key graphic area. The defect severity value ranges from 0 to 100.

[0157] S6. Calculate the quality score based on the image recognition results and defect detection results, and generate a waste discharge decision signal based on the comparison between the quality score and the preset quality threshold.

[0158] like Figure 7 As shown, the specific process of this step includes:

[0159] S61. Calculate the image recognition accuracy score based on the degree of matching between the image recognition result and the standard image information. The image recognition accuracy score ranges from 0 to 100.

[0160] Specifically, the ratio of the number of correctly recognized characters to the total number of characters is multiplied by 100 to obtain the text recognition accuracy. The ratio of the number of correctly matched patterns to the total number of patterns is multiplied by 100 to obtain the pattern recognition accuracy. The average of the text recognition accuracy and the pattern recognition accuracy is then taken to obtain the image-text recognition accuracy score.

[0161] S62. Calculate the defect deduction value based on the defect detection results. The defect deduction value is the sum of the defect severity values ​​of all detected defects.

[0162] S63. Calculate the defect detection score. The defect detection score is equal to 100 minus the defect deduction value. When the calculation result is less than 0, the defect detection score is 0.

[0163] S64. Multiply the image recognition accuracy score by 0.4 to obtain the first weighted score, and multiply the defect detection score by 0.6 to obtain the second weighted score.

[0164] S65. Add the first weighted score to the second weighted score to obtain the quality score.

[0165] S66. When the quality score is less than the preset quality threshold, a waste discharge instruction is generated.

[0166] When the quality score is greater than the preset quality threshold, a qualified instruction is generated. The preset quality threshold is 60.

[0167] S7. When the waste discharge decision signal is a waste discharge command, the waste discharge mechanism will remove the corresponding printed material from the conveyor belt.

[0168] In this step, the waste discharge controller performs corresponding operations based on the received waste discharge decision signal. When the waste discharge decision signal is a waste discharge instruction, the waste discharge controller waits for the position arrival signal sent by the position tracking unit. When the corresponding printed product arrives at the waste discharge station, it controls the pneumatic push rod to extend and push the unqualified printed product from the conveyor belt into the waste discharge channel. When the waste discharge decision signal is a qualified instruction, the pneumatic push rod remains in the retracted state, and the qualified printed product continues to move along the conveyor belt to the next process.

[0169] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent control system for perfect binding thread graphic recognition and automatic waste removal, characterized in that: It includes an image acquisition module, an image preprocessing module, a perfect binding line area positioning module, an image and text recognition module, a defect detection module, an intelligent decision-making module, and a waste disposal execution module; The image acquisition module is used to acquire image information of the printed matter to be inspected on the conveyor belt through an industrial camera, and transmit the image information to the image preprocessing module; The image preprocessing module is used to perform grayscale conversion, noise reduction, and contrast enhancement on the image information to generate a preprocessed image; The adhesive binding line area localization module is used to perform multi-scale feature extraction on the preprocessed image using a convolutional neural network to generate a feature map. Based on the feature map, candidate regions for perfect binding lines are generated, and a confidence score is calculated for each candidate region; candidate regions with confidence scores greater than a preset confidence threshold are selected as valid candidate regions. Non-maximum suppression is applied to the effective candidate regions to remove overlapping regions; Then, bounding box regression is performed on the candidate regions after non-maximum suppression to determine the precise coordinates of the binding line region and generate the binding line region image. The image recognition module is used to recognize text and pattern information in the image of the perfect binding line area and generate image recognition results; The defect detection module is used to register and align the image of the glue binding line area with a pre-stored standard template image; The process involves calculating the difference between the registered perfect binding line area image and the standard template image; performing threshold segmentation on the difference image to extract the difference regions; performing morphological feature analysis on each difference region to determine the corresponding defect type, which includes perfect binding line offset defects, perfect binding line breakage defects, image and text blurring defects, image and text missing defects, and stain defects; and calculating the defect severity value based on the area and location information of the difference regions to generate defect detection results. The intelligent decision-making module is used to multiply the image and text recognition accuracy score by a first weighting coefficient to obtain a first weighted score, multiply the defect detection score by a second weighting coefficient to obtain a second weighted score, add the first weighted score and the second weighted score to obtain a quality score, and generate a waste discharge decision signal based on the comparison result of the quality score and a preset quality threshold; wherein, the first weighting coefficient is 0.4 and the second weighting coefficient is 0.6; The waste removal execution module is used to receive the waste removal decision signal. When the waste removal decision signal is a waste removal instruction, it controls the waste removal mechanism to remove the corresponding printed material from the conveyor belt.

2. The intelligent control system for perfect binding thread graphic recognition and automatic waste removal according to claim 1, characterized in that: The image and text recognition module includes: a text recognition unit and a pattern recognition unit; The text recognition unit is used to perform text detection and text recognition on the text region in the perfect binding line area image, and output the text content and text position information. The pattern recognition unit is used to extract pattern features and match patterns in the pattern area of ​​the glued seam area image, and output pattern category and pattern completeness information.

3. The intelligent control system for perfect binding thread graphic recognition and automatic waste removal according to claim 1, characterized in that: The defect detection module calculates a defect severity value for each defect type, with the value ranging from 0 to 100. The defect detection score is equal to 100 minus the sum of the defect severity values ​​of all detected defects. When the calculation result is less than 0, the defect detection score is 0.

4. A smart control method for perfect binding thread graphic recognition and automatic waste removal according to any one of claims 1-3, characterized in that: Includes the following steps: S1. Acquire image information of the printed materials to be inspected on the conveyor belt using an industrial camera; S2. Perform grayscale conversion, noise reduction, and contrast enhancement on the image information to generate a preprocessed image; S3. Use a convolutional neural network to perform multi-scale feature extraction on the preprocessed image to generate a feature map; Based on the feature map, candidate regions for perfect binding lines are generated, and a confidence score is calculated for each candidate region; candidate regions with confidence scores greater than a preset confidence threshold are selected as valid candidate regions. Non-maximum suppression is applied to the effective candidate regions to remove overlapping regions; Boundary box regression is performed on the candidate regions after non-maximum suppression to determine the precise coordinates of the binding line region and generate the binding line region image. S4. Recognize the text and pattern information in the image of the glued seam area and generate a text and image recognition result; S5. Register and align the image of the perfect binding line area with the pre-stored standard template image. The difference image between the registered perfect binding line area image and the standard template image is calculated; the difference image is segmented by thresholding to extract the difference region; morphological feature analysis is performed on each difference region to determine the corresponding defect type, which includes perfect binding line offset defect, perfect binding line breakage defect, image and text blurring defect, image and text missing defect, and stain defect; the defect severity value is calculated based on the area and location information of the difference region to generate the defect detection result. S6. Calculate the image and text recognition accuracy score based on the degree of matching between the image and text recognition results and the standard image and text information; The defect deduction value is calculated based on the defect detection results. The defect deduction value is the sum of the defect severity values ​​of all detected defects. Calculate the defect detection score, which equals 100 minus the defect deduction value. A value of 0 is taken when the result is less than 0. Multiply the image recognition accuracy score by 0.4 to obtain a first weighted score, and multiply the defect detection score by 0.6 to obtain a second weighted score. Add the first weighted score and the second weighted score to obtain a quality score. When the quality score is less than a preset quality threshold of 60, a rejection instruction is generated; when the quality score is greater than the preset quality threshold of 60, a pass instruction is generated. S7. When the waste discharge decision signal is a waste discharge command, the waste discharge mechanism is controlled to remove the corresponding printed material from the conveyor belt.

5. The intelligent control method for perfect binding thread graphic recognition and automatic waste removal according to claim 4, characterized in that: In step S4, the specific process of recognizing text information includes: S41. Perform text region detection on the image of the perfect binding line area and locate the sub-region where the text is located; S42. Perform character segmentation on each text sub-region to obtain a single character image; S43. Use a character recognition model to recognize each individual character image to obtain the character recognition result; S44. Concatenate all character recognition results according to their position order to generate a complete text recognition result.

6. The intelligent control method for perfect binding thread graphic recognition and automatic waste removal according to claim 4, characterized in that: In step S5, the registration and alignment adopts a feature point matching method. First, feature points are detected in the current image and the standard template image respectively. Then, the transformation matrix between the images is calculated through feature point matching. Finally, the current image is registered to the coordinate system of the standard template image according to the transformation matrix. The morphological features include the area, shape, position and gray-level distribution features of the difference region.

Images