Scene-adaptive identification method for pattern-connected layout of decorative plate

Abstract

The invention relates to a scene-adaptive decorative plate continuous pattern layout identification method, which relates to a decorative plate layout identification method, and comprises the following steps of: after reading a template picture and an input decorative plate picture, carrying out the same image preprocessing operation, and calling a layout identification model; and comparing the input image with the template image through a normalized correlation coefficient method calculation model, selecting a maximum identification score from all similarity scores, comprehensively considering the maximum similarity scores of the plurality of template images, and determining the connected pattern layout category to which the current decorative plate image belongs. Manual classification is replaced by programmed decorative plate connection pattern layout recognition, and the problems of misjudgment and inaccurate recognition caused by the fact that manual detection of the decorative plate according to the texture degree is easily interfered by various unstable factors are solved. According to the invention, the accuracy of identification of the pattern connection layout of the decorative plate is obviously improved, stable operation in a complex environment can be ensured, and the detection efficiency is effectively improved in actual production.

Description

Technical Field

[0001] The invention relates to a method for recognizing a layout of a decorative plate, and in particular to a method for recognizing a layout of a decorative plate with continuous grains that is adaptive to a scene. Background Art

[0002] In existing factory classification systems, color features and simple texture information are usually used to effectively classify different types of decorative panels.

[0003] However, for continuous texture pages with slight texture differences, manual identification and sorting are still required.

[0004] Since many factories currently produce decorative panels with complex continuous patterns on their surfaces, some decorative panels have more and more colors, patterns, and designs, and their surfaces are becoming more and more complex. However, relying on manual identification and sorting of decorative panels based on image texture is slow, costly, and easily interfered by various unstable factors, leading to misjudgment and misjudgment. Summary of the invention

[0005] The purpose of the present invention is to provide a scene-adaptive decorative plate continuous-grain layout recognition method. The present invention uses a layout recognition model to compare an input image with a template image through a normalized correlation coefficient calculation model, comprehensively considers the similarity scores of multiple template images and the input image, and determines the continuous-grain layout category to which the current decorative plate image belongs. It replaces manual classification, solves the problem that manual detection of decorative plates based on texture levels is easily interfered by various unstable factors, resulting in misjudgment, misjudgment and inaccurate recognition, and improves the accuracy of decorative plate layout recognition.

[0006] The objective of the present invention is achieved through the following technical solutions: A scene-adaptive decorative plate continuous-grain layout recognition method, the method includes input image acquisition, layout library loading, image preprocessing, normalized correlation coefficient layout recognition model, scene-adaptive parameter optimization and configured modules; the input image acquisition is a decorative plate image taken by a camera from a computer acquisition card as an input image; the layout library loads a standard plate image of a decorative plate continuous-grain layout including a texture type as a template image; image preprocessing includes graying, edge clipping and image scaling; the layout recognition model uses a normalized correlation coefficient method to calculate the similarity of the image for recognition; the normalization process eliminates the influence of the difference in image scale and brightness on the recognition result; the scene-adaptive parameter optimization model adaptively optimizes the recognition model parameters according to the data distribution of the highest similarity and the second highest similarity; the configured modules include: a template image loading module, an input image acquisition module, an image preprocessing module, a layout recognition module and a classification output module ; The template image loading module is configured to obtain an input image. When the decorative board passes through the equipment, the PLC signal is triggered by the sensor to control the camera to take pictures and obtain the board image from the image acquisition card; the input image acquisition module is configured to obtain a layout library, which includes a variety of different board continuous grain layouts under the same batch production number; the image preprocessing module is configured to reduce image noise and background interference, improve recognition accuracy and robustness by performing the same preprocessing operation on the template image and the input image, and prepare for automatic recognition; the layout recognition module is configured to call the layout recognition model, compare the input image with the template image through the normalized correlation coefficient method recognition model, and calculate the similarity score between the two; wherein, multi-directional feature recognition is required for the template image; the classification output module is configured to determine the continuous grain layout category to which the current decorative board image belongs by calculating the similarity between the input image and all images in the layout library.

[0007] The described scene-adaptive decorative plate continuous-grain layout recognition method comprises the image preprocessing, graying, edge cropping and image scaling, wherein the graying is configured to perform a graying operation on the image to convert a color image into a grayscale image; the edge cropping is configured to perform a cropping operation on the template image and the input image respectively to remove interference of irrelevant edge areas on the recognition process; and the image scaling is configured to perform an image scaling operation on the template image and the input image respectively to improve the recognition speed and robustness.

[0008] In the scene-adaptive decorative plate continuous-grain layout recognition method, the layout recognition model uses a multi-directional similarity calculation operation to perform multi-directional recognition on the template image.

[0009] In the scene-adaptive decorative board continuous-grain layout recognition method, the classification output module uses a method of minimizing classification errors to determine the continuous-grain layout category to which a new tile image belongs.

[0010] In the scene-adaptive decorative plate continuous-grain layout recognition method, the image preprocessing module includes: A grayscale submodule is configured to perform a grayscale operation on the image, converting the color image into a grayscale image; The edge cropping submodule is configured to perform cropping operations on the template image and the input image respectively to remove the interference of irrelevant edge areas on the recognition process; The image scaling submodule is configured to perform image scaling operations on the template image and the input image respectively, so as to improve the recognition speed and robustness.

[0011] The advantages and positive effects of the present invention are: 1. After reading the template image and the input decorative board image, the present invention performs the same image preprocessing operation, calls the layout recognition model, and compares the input image with the template image through the normalized correlation coefficient calculation model, wherein the template image is subjected to multi-directional feature recognition. The similarity score between the two is calculated, and the maximum recognition score is selected from all similarity scores. If the score is greater than the preset recognition threshold, the returned result is the category to which the corresponding standard decorative board belongs. The maximum similarity scores of multiple template images are comprehensively considered to determine the continuous grain layout category to which the current decorative board image belongs. Programmed continuous grain layout recognition of decorative board replaces manual classification, solves the problem that manual detection of decorative board according to texture degree is easily interfered by various unstable factors, resulting in misjudgment, misjudgment and inaccurate recognition, and improves the classification accuracy.

[0012] 2. The scene-adaptive threshold optimization model of the present invention: The system automatically detects the scene features of the input image, and adaptively adjusts the recognition threshold according to different detection scenes. During the implementation process, the similarity scores of a fixed number of decorative panels are subjected to unsupervised binary classification through cluster analysis and divided into positive and negative classes. Then, the median between the minimum value of the best similarity score set and the maximum value of the recognition error similarity score set is calculated as the scene-adaptive layout recognition threshold. After the classification of the decorative panel is determined, the corresponding decorative panel picture is used as the standard panel picture of the category to which it belongs, so that the classified standard panel pictures are richer, and there are more reference templates when classifying new decorative panels, so that the recognition and classification of the continuous grain layout of the decorative panel is more detailed. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the method flow of the continuous-grain layout recognition model of the decorative board in an embodiment of the present invention; Figure 2 A schematic diagram of a method flow of a scene-adaptive layout recognition model in an embodiment; Figure 3 It is a schematic diagram of a method flow of a scene-adaptive decorative plate continuous-grain layout recognition system in an embodiment; Figure 4 It is a schematic diagram of the process of a scene-adaptive decorative plate continuous pattern layout recognition module device in the embodiment; Figure 5 Schematic diagram of a module for input image preprocessing in an embodiment; Figure 6 Schematic diagram of the modules of the decorative board continuous pattern layout recognition model in the embodiment; Figure 7 A schematic diagram of a module of a scene-adaptive layout recognition model in an embodiment; Figure 8 It is a schematic diagram of the overall process flow of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the embodiments shown in the accompanying drawings.

[0015] like Figure 1 As shown, the implementation of the present invention includes the following steps: S110: loading template images containing various types of continuous-grain layouts of decorative panels, and acquiring input decorative panel images through a camera; In a specific embodiment, a plurality of standard decorative plate images are obtained in advance, and the sample images can be directly obtained through a camera on a transmission mechanism in a factory, or can be obtained from a database storing decorative plate images. The camera-photographed decorative plate image taken from a computer acquisition card is used as an input image, and the standard plate image of the continuous-grain layout of decorative plates of all texture types is used as a template image.

[0016] S120: performing consistent image preprocessing on the template image and the input image, including graying, edge cropping, and image scaling; In a specific embodiment, the same image preprocessing operation is performed on the template image and the input image, including graying, edge cropping and image scaling, wherein: graying is configured to grayscale the image and convert the color image into a grayscale image; edge cropping is configured to crop the template image and the input image respectively to remove the interference of the irrelevant edge area on the recognition process, wherein the template image is cropped by 400 pixels at the edge and the input image is cropped by 200 pixels at the edge; image scaling is configured to scale the template image and the input image respectively to improve the recognition speed. Image preprocessing can reduce image noise and background interference, improve matching accuracy and robustness, and the output of image preprocessing is used as the input of layout recognition.

[0017] S130: using a normalized correlation coefficient method to calculate the similarity between the input image and the template image, and performing multi-directional feature recognition to optimize the similarity calculation; In a specific embodiment, the normalized correlation coefficient method evaluates the degree of recognition by calculating the correlation coefficient between the template pixel and the corresponding pixel of the image, thereby calculating the similarity of the image for recognition. The correlation coefficient ranges from 0 to 1, and a value of 1 indicates a perfect match. It is suitable for identifying objects with complex textures and lighting changes to eliminate the impact of differences in image scale and brightness on the recognition results.

[0018] In a specific embodiment, the decorative plate input image is compared with the template image, and the template image is rotated clockwise by 0°, 90°, 180° and 270° respectively using a multi-directional feature recognition operation to optimize the calculation result of the similarity score.

[0019] S140: Classify the input image into the corresponding continuous pattern layout category according to the similarity score, and output the finally identified continuous pattern layout category.

[0020] In a specific embodiment, for each rotated template image, the similarity result with the decorative plate image is calculated, and the maximum similarity score is saved in a list. By comparing these maximum similarity scores, the most matching template image is finally determined, thereby determining the category to which the image belongs.

[0021] In a general specific embodiment, after reading the input image and template image of the decorative plate, the same preprocessing operation is performed on them respectively, and the normalized correlation coefficient is called to calculate the recognition model. By calculating the similarity score between the current decorative plate image and the template image of each category, the maximum similarity score is selected from all similarity scores. If the maximum similarity score is greater than the preset recognition threshold, the category of the template image corresponding to the maximum similarity score is determined as the category of the current decorative plate. If the maximum similarity score is less than the preset recognition threshold, a new category is added to the current decorative plate to complete the classification of the current decorative plate. Automated classification and recognition replaces manual classification, solves the problem that manual detection of decorative plates based on continuous lines is easily interfered by various unstable factors, resulting in misjudgment, misjudgment and inaccurate classification, and improves the accuracy of decorative plate classification.

[0022] Figure 2 A schematic diagram of a method flow of a scene-adaptive layout recognition model in an embodiment of the present invention is shown, and the method specifically comprises the following steps: S210: Data acquisition, capturing the image data of the decorative board through an industrial camera to ensure that the entire area of ​​the decorative board is covered; In a specific embodiment, the image of the decorative plate currently on the transmission mechanism is acquired by a line array camera or an area array camera to ensure that all valid information of the current decorative plate is acquired.

[0023] S220: Application of layout recognition model, normalized correlation coefficient combined with multi-directional feature recognition technology to increase recognition accuracy; In a specific embodiment, a normalized correlation coefficient method is used to calculate the similarity between the input image and the template image, and multi-directional feature recognition is performed on the template image to further improve accuracy.

[0024] S230: Outputting accurate continuous pattern layout recognition results by comparing the best similarity score with a preset recognition threshold; In a specific embodiment, based on the similarity scores between the current decorative board image and the template image of each category, the maximum similarity score between the current decorative board image and the standard template image of each category is selected; the maximum similarity score is compared with a preset threshold, and if the maximum similarity score is greater than the preset threshold, the category of the standard template image corresponding to the maximum similarity score is determined as the category of the current input decorative board; if the maximum similarity score is less than the preset threshold, a new category is added to the current input decorative board image, and the current input decorative board image is used as the standard template image corresponding to the category.

[0025] In a specific embodiment, when the number of template images contained in a category reaches a preset threshold, no template images are added to the layout library. In order to prevent excessive storage of template image information and affect the running speed of the program, a threshold for the number of template images contained in the category is set.

[0026] S240: The system detects scene features of the input image and automatically adjusts the recognition threshold to adapt to different detection scenarios.

[0027] In a specific embodiment, cluster analysis is used to divide the similarity scores into positive and negative categories according to the distribution of the recognition similarity scores of a fixed number of decorative panels, and the boundary points of the similarity scores of correct and incorrect recognition are calculated to determine the scene-adaptive layout recognition threshold. The cluster analysis performs unsupervised binary classification on the similarity score set to obtain the best similarity score set and the recognition error similarity score set, and the median between the minimum value of the best similarity score set and the maximum value of the recognition error similarity score set is the scene-adaptive threshold.

[0028] In a general specific embodiment, an input decorative sheet image is obtained through an industrial camera, a template image in a layout library is loaded, image preprocessing operations are performed respectively, an adaptive decorative sheet continuous pattern layout recognition model is called, the similarity score between the current input decorative sheet image and the standard template image of each category is calculated, and the maximum similarity score is selected from all similarity scores. If the maximum similarity score is greater than a preset threshold, the category of the template image corresponding to the maximum similarity score is determined as the category of the current decorative sheet. If the maximum similarity score is less than the preset threshold, a new category is added to the current decorative sheet to complete the classification of the current decorative sheet. The system detects the scene features of the input image and automatically adjusts the recognition threshold to adapt to different detection scenes. From reading the decorative sheet image to completing the classification, it is automatically carried out according to a pre-set program, and the system reasonably completes the classification task. Automated classification replaces manual classification, and also solves the problem that manual detection of decorative sheets according to the degree of texture is easily interfered by various unstable factors, resulting in misjudgment, misjudgment and inaccurate classification, thereby improving the accuracy of decorative sheet classification.

[0029] Figure 3 A schematic diagram of a method flow of a scene-adaptive decorative plate continuous pattern recognition system according to an embodiment of the present specification is shown. Figure 3 As shown, after each category of template image and input decorative sheet image are successfully read, corresponding image preprocessing operations are performed to reduce image noise and background interference, and improve recognition accuracy and robustness. The recognition model is calculated based on the normalized correlation coefficient method, and multi-directional feature recognition is performed. The current input decorative sheet image is compared with the template image of each category to obtain the similarity score between the current decorative sheet image and the template image of each category, and the maximum value is found. The maximum similarity score is compared with the preset threshold. If the maximum similarity score is greater than the preset threshold, the category of the template image corresponding to the maximum similarity score is determined as the category of the currently input decorative sheet image; if the maximum similarity score is less than the preset threshold, a new category is added to the current decorative sheet image. The system realizes automation from reading the decorative sheet image to completing the classification through a preset program, successfully replacing manual classification. This automated classification not only effectively avoids the misjudgment and misjudgment problems that may be caused by unstable factors such as texture differences during manual detection, but also significantly improves the accuracy of decorative sheet classification.

[0030] Figure 4 A schematic diagram of a process of a scene-adaptive decorative plate continuous pattern recognition device according to an embodiment of the present specification is shown. Figure 4As shown, after the decorative board flows into the transmission mechanism, the camera shoots the tile currently located on the transmission mechanism to obtain the current image, ensuring that it contains complete information of the decorative board image. After the image processing unit, the acquired image data is processed, such as grayscale, edge cropping and image scaling, to ensure high-quality data input, and GPU can be optionally used for accelerated processing. The layout recognition model based on the normalized correlation coefficient method is called to perform multi-directional feature recognition, and the currently input decorative board image is compared with the template image of each category, the similarity score between them is calculated, and the maximum similarity value is found. Then, the maximum similarity score is compared with the preset threshold. If the maximum similarity score is greater than the preset threshold, the template image category corresponding to the maximum similarity score is used as the category of the current decorative board image; if the maximum similarity score is less than the preset threshold, a new category is assigned to the current decorative board image. The system detects the scene features of the input image and automatically adjusts the recognition threshold to adapt to different detection scenarios. In automated production, the device can directly issue instructions to the control system of the production line to make actual adjustments. The device outputs the results of recognition and adjustment through the display or control panel, and the operator can view and intervene in real time. At the same time, the system can also output data to a database, control system or cloud for recording and analysis.

[0031] Figure 5 FIG. 4 shows a schematic diagram of a module for preprocessing an input image in an embodiment of the present invention. Figure 5 As shown, it may include: Grayscale module S510: is configured to perform grayscale operation on the image and convert the color image into a grayscale image.

[0032] The edge cropping module S520 is configured to perform cropping operations on the template image and the input image respectively to remove the interference of the edge irrelevant area on the recognition process.

[0033] Image scaling module S530: configured to perform image scaling operations on the template image and the input image respectively, so as to improve the recognition speed and robustness.

[0034] As can be seen from the above, the image preprocessing module optimizes the image quality through a series of operations to improve the accuracy and efficiency of subsequent image recognition. The grayscale module converts the color image into a grayscale image to simplify the processing process; the edge cropping module removes irrelevant areas at the edge of the image to reduce interference; and the image scaling module adjusts the image size to speed up the recognition process and enhance robustness. These operations work together to ensure that the input image can be processed more accurately and efficiently in the subsequent recognition process.

[0035] Figure 6 FIG. 4 shows a schematic diagram of a module of a decorative board continuous pattern recognition model according to an embodiment of the present invention. Figure 6As shown, a device for identifying a continuous-grained pattern surface model of a decorative plate provided in an embodiment of this specification may include: Template image reading module S610: is configured to load various types of decorative board continuous pattern layout template images from the layout library. The template images correspond to board layouts of different texture categories for subsequent matching and recognition.

[0036] Input image loading module S620: is configured to obtain an input decorative plate image from an industrial camera. This module can automatically collect images when the sensor is triggered and pass them to the subsequent processing module.

[0037] Image preprocessing module S630: is configured to perform unified preprocessing operations on the template image and the input image, including steps such as grayscale, edge cropping and image scaling, so as to eliminate noise, remove irrelevant areas and improve the speed and robustness of image recognition.

[0038] Similarity score calculation module S640: is configured to calculate the similarity score between the input image and the template image using a normalized correlation coefficient method, and compare all template images according to the similarity score to select the best match.

[0039] The classification output module S650 is configured to determine the continuous pattern layout category to which the input decorative sheet image belongs based on the similarity scores between the multiple template images and the input image. If the similarity score is higher than a set threshold, the classification module will output the continuous pattern layout category of the decorative sheet image.

[0040] Figure 7 FIG. 4 shows a schematic diagram of a module of a scene-adaptive layout recognition model in an embodiment of the present invention. Figure 7 As shown, a device for a scene-adaptive layout recognition model provided by an embodiment of this specification may include: Data acquisition module S710: is configured to acquire decorative board image data through an image acquisition device, including template images of each texture category in the layout library and decorative board images on the current input mechanism, to provide raw data for subsequent recognition model training.

[0041] Image processing module S720: is configured to perform a series of processing operations on the input image, such as grayscale, edge cropping and image scaling, to improve image quality and reduce interference, ensuring that the model can accurately perform layout recognition.

[0042] Calculation module S730: configured to perform normalized correlation coefficient method to calculate the similarity between the processed input image and the template image. This module is responsible for calculating the similarity score between the template image and the input image based on multiple directional feature recognitions and outputting a matching score.

[0043] Scene adaptive threshold module S740: After obtaining the similarity score, it is configured to adaptively adjust the classification threshold according to the actual situation of the current input image. By analyzing the quality, texture complexity and other characteristics of the input image, the threshold for determining the similarity of the template image is dynamically adjusted to avoid inaccurate classification caused by factors such as lighting changes, texture blur or noise.

[0044] Category determination module S750: is configured to determine the continuous pattern layout category to which the input image belongs based on the comprehensive similarity score and output the final classification result. This module automatically classifies the decorative board based on the maximum similarity score of the match and the preset threshold.

[0045] The above content shows that after reading the current decorative board image, the decorative board classification system calculates the matching score between the current image and the template images of each category by applying the template matching model of the normalized correlation coefficient matching method. The system selects the maximum value from all the calculated matching values ​​and compares it with the preset matching threshold. If the maximum matching value exceeds the threshold, the template image category corresponding to the maximum value is determined as the category of the current decorative board; if the maximum matching value is lower than the preset detection threshold, the current decorative board is classified into a new category, thereby completing the classification process. The whole process is automatically executed according to the preset program, and the system can efficiently and accurately complete the classification task of decorative boards. Automated classification not only replaces manual operation, but also effectively avoids the problems of misjudgment, misjudgment and inaccurate classification caused by external interference factors during manual classification, thereby significantly improving the accuracy of decorative board classification.

Claims

1. A scene-adaptive decorative plate continuous-grain layout recognition method, characterized in that: The method includes input image acquisition, layout library loading, image preprocessing, normalized correlation coefficient layout recognition model, scene adaptive parameter optimization and configured modules; the input image acquisition is a decorative plate image taken by a camera from a computer acquisition card as an input image; the layout library loads a standard plate image of a decorative plate with a texture type and a continuous grain layout as a template image; image preprocessing includes graying, edge clipping and image scaling; the layout recognition model uses a normalized correlation coefficient method to calculate the similarity of the image for recognition; the normalization process eliminates the influence of the difference in image scale and brightness on the recognition result; the scene adaptive parameter optimization model adaptively optimizes the recognition model parameters according to the data distribution of the highest similarity and the second highest similarity; the configured modules include: a template image loading module, an input image acquisition module, an image preprocessing module, a layout recognition module and a classification output module; the template image loading module The module is configured to obtain an input image. When the decorative board passes through the equipment, the PLC signal is triggered by the sensor to control the camera to take pictures and obtain the board image from the image acquisition card; the input image acquisition module is configured to obtain a layout library, which includes a variety of different board continuous grain layouts under the same batch production number; the image preprocessing module is configured to reduce image noise and background interference by performing the same preprocessing operation on the template image and the input image, improve recognition accuracy and robustness, and prepare for automatic recognition; the layout recognition module is configured to call the layout recognition model, compare the input image with the template image through the normalized correlation coefficient method recognition model, and calculate the similarity score between the two; wherein, the template image needs to be recognized in multiple directions; the classification output module is configured to determine the continuous grain layout category to which the current decorative board image belongs by calculating the similarity between the input image and all images in the layout library.

2. The scene-adaptive decorative plate continuous-grain layout recognition method according to claim 1, characterized in that: The image preprocessing includes graying, edge cropping and image scaling, wherein the graying is configured to perform a graying operation on the image to convert the color image into a grayscale image; Edge cropping is configured to perform cropping operations on the template image and the input image respectively to remove interference of irrelevant edge areas on the recognition process; image scaling is configured to perform image scaling operations on the template image and the input image respectively to improve the recognition speed and robustness.

3. The scene-adaptive decorative plate continuous-grain layout recognition method according to claim 1, characterized in that: The layout recognition model uses a multi-directional similarity calculation operation to perform multi-directional recognition on the template image.

4. The scene-adaptive decorative plate continuous-grain layout recognition method according to claim 1, characterized in that: The classification output module uses a method of minimizing classification errors to determine the continuous-grain layout category to which the new tile image belongs.

5. The scene-adaptive decorative plate continuous-grain layout recognition method according to claim 1, characterized in that: The image preprocessing module includes: A grayscale submodule is configured to perform a grayscale operation on the image, converting the color image into a grayscale image; The edge cropping submodule is configured to perform cropping operations on the template image and the input image respectively to remove the interference of irrelevant edge areas on the recognition process; The image scaling submodule is configured to perform image scaling operations on the template image and the input image respectively, so as to improve the recognition speed and robustness.

Images