Bamboo wood plate online detection method, system, medium and equipment

Abstract

The invention discloses an online detection method, system, medium and equipment for a bamboo board, which can judge whether the roll coating process of the current bamboo board is qualified or not according to the detection result of a neural network model through the steps of image acquisition, image preprocessing, image synthesis, neural network model detection and the like. If qualified, conveying can be continued according to a first conveying strategy; and if the bamboo boards are not qualified, the bamboo boards can be conveyed to a preset marking area and marked. The bamboo board is comprehensively detected through a plurality of links, image information of the bamboo board is better captured, and flaws are accurately detected. In this way, unqualified bamboo boards can be found and treated in time, the product quality management efficiency is improved, the number of unqualified products is reduced, the production efficiency and the product quality are improved, and automatic online detection of the bamboo boards is achieved. Compared with traditional manual detection, the method can improve the detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention relates to the field of bamboo board detection, and in particular to an online detection method, system, medium and equipment for bamboo boards. Background Art

[0002] Bamboo board is a board made of bamboo as raw material, also known as bamboo composite board. It is a man-made board made of bamboo through processes such as cutting, pressing and gluing. The pressed bamboo board may have some unevenness or uneven edges and corners, and needs to be trimmed and modified to make its surface flat and smooth. As needed, the bamboo board is painted to increase its aesthetics and protection. During the painting process of the bamboo board, some defects may occur, such as uneven coating, bubbles, scratches or inconsistent colors, which affect the appearance and quality of the bamboo board. In the existing bamboo board production solutions, most of them rely on manual defect detection and packaging, resulting in low efficiency of the entire defect detection and high labor costs. Summary of the invention

[0003] In view of this, the object of the present invention is to provide an online detection method, system, medium and equipment for bamboo board.

[0004] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:

[0005] In a first aspect, the present invention provides an online detection method for bamboo boards, the method comprising:

[0006] The bamboo boards are sequentially conveyed to the image acquisition area;

[0007] Controlling the image acquisition unit to acquire original images of the bamboo board in the current image acquisition area at a first preset frequency to obtain an original image set;

[0008] Performing image preprocessing on the original images in the original image set one by one to obtain a basic image set, wherein the basic image set includes a plurality of basic images, and each basic image contains image information of the current bamboo board;

[0009] Input the basic image set into the image synthesis algorithm model in sequence to obtain the final synthetic image of the current bamboo board;

[0010] The final synthesized image is input into the neural network model to obtain the defect detection result, which includes the defect annotation image and the defect category;

[0011] Determine whether the current bamboo board rolling coating process is qualified based on the defect detection results;

[0012] If yes, the current bamboo board is transported according to the first transport strategy;

[0013] If not, the current bamboo board is transported to the preset marking area, and the marking component is controlled to mark the bamboo board according to the preset marking label;

[0014] The marked bamboo boards are transported according to the second transport strategy.

[0015] In some embodiments, performing image preprocessing on the original images in the original image set one by one to obtain the basic image set includes:

[0016] Inputting the original images in the original image set into the Gaussian filter model in sequence to obtain a first processed image after removing image noise;

[0017] Performing wavelet transform on the first processed image and marking the bamboo board area to obtain a second processed image;

[0018] The second processed image is cropped to obtain a basic image, which includes the annotation information of the bamboo board area.

[0019] In some embodiments, the basic image set is sequentially input into the image synthesis algorithm model to obtain the final synthetic image of the current bamboo board, including:

[0020] Perform Gabor filtering on the basic images one by one to obtain a first filtered image;

[0021] Extracting feature information of the current bamboo board in each first filtered image according to the annotation information, wherein the feature information includes edge features of the bamboo board and the collected area in each first filtered image;

[0022] Perform feature matching on multiple feature information to obtain the matching similarity between each feature information and the remaining feature information;

[0023] Align the first filtered images one by one according to the matching similarity, generate the position information of each first filtered image, map and store the position information and feature information with the first filtered image, and record the mapped and stored first filtered image as the second filtered image to obtain a second filtered image set;

[0024] Inputting the second filtered image set into the image synthesis algorithm model to obtain an initial synthesized image;

[0025] The initial synthesized image is subjected to image post-processing to obtain a final synthesized image, wherein the image post-processing includes at least one of detail enhancement, artifact removal, and tone adjustment.

[0026] In some embodiments, the final synthesized image is input into a neural network model to obtain defect detection results including:

[0027] Annotating the final synthesized image to generate a plurality of annotation selection boxes, each of which has an annotation area, and recording the annotated final synthesized image as a first annotated image;

[0028] Input the first annotated image into the YOLO-v3 neural network model to obtain at least one annotated area corresponding to an annotated selection box containing the texture of the bamboo board, which is recorded as the second annotated image;

[0029] Inputting the second annotated image into the first classifier to classify whether there is a defect, and obtaining a first classification result, the first classification result including a set of non-defective images and a set of defective images;

[0030] Determine whether the defect image set is an empty set. If not, input the defect image set into a second classifier to classify the defect category to obtain a second classification result, the second classification result includes a first defect image group and a second defect image group, the first defect image group corresponds to a first defect category, the second defect image group corresponds to a second defect category, and the first defect category is different from the second defect category;

[0031] A defect detection result is generated according to the first classification result and the second classification result.

[0032] In some embodiments, the second conveying strategy includes the first adjustment strategy and the second adjustment strategy, and judging whether the current rolling coating process of the bamboo board is qualified according to the defect detection result also includes:

[0033] A first defect image group is selected from the defect categories, the first defect image group includes defect annotated images of the first defect category, recorded as first defect images, and the first defect category is one of a film uniformity defect, a film flatness defect, and a film gloss defect;

[0034] Calculate the defect distribution area on the bamboo board according to the first defect image, and record it as the first defect distribution area;

[0035] Determine whether the first defect distribution area is within a preset qualified threshold range;

[0036] If yes, the rolling coating process of the current bamboo board is determined to be qualified, and a second control instruction is generated, the second control instruction is used to make the current bamboo board be transported according to the first transport strategy;

[0037] If not, determining whether the first defect distribution area is within a preset repair threshold range;

[0038] If the first defect distribution area is within the preset repair threshold range, a first marking label and a second control instruction are generated, and the second control instruction is used to process the bamboo board according to the first adjustment strategy;

[0039] If the first defect distribution area is not within the preset repair threshold range, a second marking label and a third control instruction are generated, and the third control instruction is used to process the bamboo board according to the second adjustment strategy.

[0040] In some embodiments, judging whether the current rolling coating process of the bamboo board is qualified according to the defect detection result further includes:

[0041] A second defect image group is selected from the defect categories, the second defect image group includes defect annotated images of the second defect category, recorded as second defect images, and the second defect category is one of a film peeling defect, a film peeling defect, and a film fault defect;

[0042] It is determined whether the second defect image group is an empty set. If not, a third marking label and a fourth control instruction are generated. The fourth control instruction is used to scrap the bamboo board.

[0043] In some embodiments, the first conveying strategy is configured to convey the bamboo board to a qualified area;

[0044] The first adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a first marking label, and transport the marked bamboo board to a repair area;

[0045] The second adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a second marking label, and transport the marked bamboo board to the first scrapping area.

[0046] In a second aspect, the present invention provides an online detection system for bamboo boards, which is applicable to the method described in the first aspect, and the system includes a frame, a transmission component, a shaping component, an image acquisition unit and a marking component; the transmission component includes a conveyor belt, the conveyor belt is arranged on the frame, and a plurality of bamboo boards are placed on the conveyor belt; the shaping component includes a first shaping plate, a second shaping plate, a first adjustment rod and a second adjustment rod, the first shaping plate and the second shaping plate are arranged opposite to each other, and the first shaping plate and the second shaping plate are extended along the transmission direction of the conveyor belt, the first adjustment rod is connected to the first shaping plate, the first adjustment rod is used to adjust the position of the first shaping plate on the conveyor belt, the second adjustment rod is connected to the second shaping plate, and the second adjustment rod is used to adjust the position of the second shaping plate on the conveyor belt; the image acquisition unit includes a camera and a light source, the light source is arranged above the conveyor belt, and the camera is arranged above the light source; the marking component includes a seal device and a first cylinder, the seal device is arranged above the conveyor belt, and the seal device and the camera are arranged successively along the transmission direction of the conveyor belt, and the output end of the first cylinder is connected to the seal device, which is used to drive the seal device to move back and forth in the vertical direction.

[0047] In a third aspect, the present invention further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0048] In a fourth aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect.

[0049] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0050] This technical solution uses steps such as image acquisition, image preprocessing, image synthesis, and neural network model detection to determine whether the current rolling coating process of the bamboo board is qualified based on the detection results of the neural network model. If qualified, the bamboo board can continue to be transported according to the first conveying strategy; if unqualified, the bamboo board can be transported to the preset marking area and marked. Through multiple links, the bamboo board is comprehensively inspected to better capture the image information of the bamboo board and accurately detect defects. In this way, unqualified bamboo boards can be discovered and handled in a timely manner, the efficiency of product quality management can be improved, the number of unqualified products can be reduced, production efficiency and product quality can be improved, and automated online detection of bamboo boards can be realized. Compared with traditional manual detection, this method can improve detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 It is a step diagram of the online detection method of bamboo board;

[0053] Figure 2 It is the first schematic diagram of the online inspection system for bamboo boards;

[0054] Figure 3 It is the second schematic diagram of the bamboo board online detection system.

[0055] Reference numerals:

[0056] 1. Framework;

[0057] 11. The first bracket;

[0058] 12. The second bracket;

[0059] 2. Transmission components;

[0060] 21. A first driving unit;

[0061] 22. Conveyor belt;

[0062] 23. Button;

[0063] 24. First sensor;

[0064] 3. Plastic components;

[0065] 31. The first shaping plate;

[0066] 32. Second shaping plate;

[0067] 33. First adjustment lever;

[0068] 34. Second adjustment lever;

[0069] 35. The third adjustment lever;

[0070] 36. Fourth adjustment lever;

[0071] 37. A first fixing block;

[0072] 38. Second fixing block;

[0073] 4. Image acquisition unit;

[0074] 41. Camera;

[0075] 42. Light source;

[0076] 5. Marking components;

[0077] 51. Seal equipment;

[0078] 52. First cylinder;

[0079] 6. Protective cover;

[0080] 7. Warning light;

[0081] 8. Display;

[0082] 9. Touch screen. DETAILED DESCRIPTION

[0083] The present invention will be further described in detail below in conjunction with the accompanying drawings and examples. It is particularly noted that the following examples are only used to illustrate the present invention, but are not intended to limit the scope of the present invention. Similarly, the following examples are only partial embodiments of the present invention rather than all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0084] See also Figure 1 In a first aspect, this embodiment provides an online detection method for bamboo boards, the method comprising:

[0085] S101, conveying the bamboo boards to the image acquisition area in sequence;

[0086] S102, controlling the image acquisition unit to acquire original images of the bamboo board in the current image acquisition area at a first preset frequency to obtain an original image set;

[0087] S103, performing image preprocessing on the original images in the original image set one by one to obtain a basic image set, where the basic image set includes a plurality of basic images, each of which contains image information of the current bamboo board;

[0088] S104, inputting the basic image set into the image synthesis algorithm model in sequence to obtain the final synthesized image of the current bamboo board;

[0089] S105, inputting the final synthesized image into the neural network model to obtain a defect detection result, where the defect detection result includes a defect annotated image and a defect category;

[0090] S106, judging whether the current roller coating process of the bamboo board is qualified according to the defect detection result;

[0091] S107: If yes, transport the current bamboo board according to the first transport strategy;

[0092] S108, if not, transport the current bamboo board to the preset marking area, and control the marking component to mark the bamboo board according to the preset marking label;

[0093] S109, transporting the marked bamboo board according to the second transport strategy.

[0094] In step S101, the image acquisition area can be understood as the direction of the shooting end of the camera described later. The bamboo boards can be manually loaded, or the bamboo boards can be automatically loaded using a mechanical arm, a mechanical clamp, etc. It should be noted that during the loading process, there is a certain interval between two adjacent bamboo boards.

[0095] In step S102, the image acquisition unit acquires original images of the bamboo board in the current image acquisition area at a first preset frequency. That is, when it is detected that a bamboo board has entered the image acquisition area, image acquisition at a first preset frequency is performed on the current bamboo board to obtain an original image set, which includes a group of original images of the current bamboo board.

[0096] In step S103, image preprocessing is performed on the original images in the current original image set to remove image noise such as ghost images and areas with too dark light in the original images to obtain basic images. Each basic image contains image information of the current bamboo board. It can be understood that each basic image contains a partial or complete image of the current bamboo board.

[0097] In step S104, the basic images in the basic image set are input into the image synthesis algorithm model at one time to obtain the final synthetic image of the current bamboo board. This step can perform a comprehensive image restoration of the bamboo board based on the image information of the bamboo board to highlight the texture characteristics of the current bamboo board. In the prior art, a logic of performing a separate defect recognition on each basic image and then sorting and outputting the defect results is often used. This method is prone to repeated defect recognition and the inability to accurately estimate the defect distribution area. In the field of roller coating of bamboo boards, roller coating defects include repairable defects and unrepairable defects. This step facilitates the subsequent precise control of different defect categories and defect distribution.

[0098] In step S105, the final synthesized image is input into the neural network model to obtain a defect detection result. The neural network model can perform defect recognition on the final synthesized image. Specifically, the neural network model can be obtained based on deep learning model training.

[0099] In step S106, it is determined whether the rolling coating process of the current bamboo board is qualified based on the defect detection result. Furthermore, in step S107, if the rolling coating process of the current bamboo board is qualified, the current bamboo board is transported according to the first conveying strategy. The first conveying strategy may be to transport the bamboo board to the next processing link, or it may be to transport the bamboo board to a qualified area of ​​the product for stacking and storage.

[0100] In step S108, if the current bamboo board's roller coating process is unqualified, the bamboo board is transported to a preset marking area, and the marking component is controlled to mark the bamboo board according to the preset marking label, so that the bamboo board with unqualified roller coating process is distinguished from the qualified bamboo board. In step S109, the marked bamboo board is transported according to the second transport strategy, and the second transport strategy can be to transport the marked bamboo board to the scrap area, or to transport the marked bamboo board to the repair area for repair.

[0101] This embodiment uses steps such as image acquisition, image preprocessing, image synthesis, and neural network model detection, and according to the detection results of the neural network model, it can be determined whether the current rolling coating process of the bamboo board is qualified. If qualified, the bamboo board can continue to be transported according to the first conveying strategy; if unqualified, the bamboo board can be transported to the preset marking area and marked. The bamboo board is comprehensively inspected through multiple links to better capture the image information of the bamboo board and accurately detect defects. In this way, unqualified bamboo boards can be discovered and handled in a timely manner, the efficiency of product quality management can be improved, the number of unqualified products can be reduced, production efficiency and product quality can be improved, and automated online detection of bamboo boards can be realized. Compared with traditional manual detection, this method can improve detection efficiency and accuracy.

[0102] In some embodiments, performing image preprocessing on the original images in the original image set one by one to obtain the basic image set includes:

[0103] Inputting the original images in the original image set into the Gaussian filter model in sequence to obtain a first processed image after removing image noise;

[0104] Performing wavelet transform on the first processed image and marking the bamboo board area to obtain a second processed image;

[0105] The second processed image is cropped to obtain a basic image, which includes the annotation information of the bamboo board area.

[0106] In this embodiment, the image preprocessing method includes Gaussian filtering and wavelet transform. Among them, the Gaussian filtering model is a smoothing filtering model based on the Gaussian function, and the filtering operation is realized by weighted averaging the pixels around each pixel. Specifically, for each pixel, the Gaussian filtering model calculates the weighted average of the pixels in the neighborhood around the pixel, wherein the weight of each pixel is calculated by the Gaussian function. The Gaussian function determines the weight according to the distance from the center pixel. The closer the distance, the greater the weight of the pixel, and the farther the distance, the smaller the weight of the pixel. By inputting the original images in the original image set into the Gaussian filtering model in sequence, the first processed image with the image noise removed can be obtained. The function of the Gaussian filtering model is to smooth the image, remove the high-frequency noise in the original image, make the original image smoother and clearer, and improve the effect of the subsequent processing steps.

[0107] Wavelet transform decomposes the first processed image into frequency components of different scales. Through wavelet transform, the first processed image can be decomposed into low-frequency and high-frequency parts, where the low-frequency part represents the general structure of the first processed image, and the high-frequency part represents the detailed information of the first processed image. In this process, the first processed image will be input into the wavelet transform model, and after processing, the bamboo board area will be marked for the first processed image to obtain the second processed image. The second processed image contains more information about the bamboo board, including the structure, texture and other characteristics of the bamboo board, which facilitates further analysis and processing of the second processed image to achieve more accurate bamboo board detection operations.

[0108] Furthermore, this embodiment also crops the second processed image to obtain a basic image. Specifically, the cropping condition may be to retain the annotation information of the bamboo board area and to crop and delete the remaining irrelevant annotation information. For example, the second processed image may include the edge of the frame. When annotating, the edge of the frame also has an annotation information. Then, the edge of the frame may be cropped and the annotation information of the edge of the frame may be deleted. It is understandable that a uniform cropping size may be set to make the size of the obtained basic image uniform.

[0109] In some embodiments, the basic image set is sequentially input into the image synthesis algorithm model to obtain the final synthetic image of the current bamboo board, including:

[0110] Perform Gabor filtering on the basic images one by one to obtain a first filtered image;

[0111] Extracting feature information of the current bamboo board in each first filtered image according to the annotation information, wherein the feature information includes edge features of the bamboo board and the collected area in each first filtered image;

[0112] Perform feature matching on multiple feature information to obtain the matching similarity between each feature information and the remaining feature information;

[0113] Align the first filtered images one by one according to the matching similarity, generate the position information of each first filtered image, map and store the position information and feature information with the first filtered image, and record the mapped and stored first filtered image as the second filtered image to obtain a second filtered image set;

[0114] Inputting the second filtered image set into the image synthesis algorithm model to obtain an initial synthesized image;

[0115] The initial synthesized image is subjected to image post-processing to obtain a final synthesized image, wherein the image post-processing includes at least one of detail enhancement, artifact removal, and tone adjustment.

[0116] The Gabor filter model is a multi-scale and multi-directional filter model that can be used to extract texture and edge features in images. The core of the Gabor filter model is the Gabor function, which has good properties between the spatial domain and the frequency domain. By applying the Gabor filter model, the response of the image at different scales and directions can be obtained, thereby extracting features with local texture structure.

[0117] In this embodiment, the original image is smoothed using a Gaussian filter model to remove noise and details. After obtaining a basic image, the Gabor filter model is used to extract texture features from the smoothed image. This method can improve the stability and reliability of the features of the roller coating on the bamboo board while reducing the impact of other image noise.

[0118] In each first filtered image, the annotation information containing the bamboo board area is screened out, and feature information is extracted based on the annotation information. The feature information includes edge features and the collected area. The edge features can be understood as the edge features of the bamboo board, and the collected area is the area of ​​the bamboo board in the current first filtered image. It should be noted that the first filtered image obtained after the bamboo board enters the image acquisition area and moves out of the image acquisition area at a uniform speed through the first preset frequency acquisition and processing includes the image of the bamboo board just moving into the image acquisition area and the image of the bamboo board just moving out of the image acquisition area. In this state, the bamboo board in the first filtered image is incomplete. Based on this, the identification of the bamboo board can be achieved by extracting feature information from multiple first filtered images.

[0119] Based on the feature information, feature matching is performed to obtain the matching similarity between each feature information and the remaining feature information. Further, multiple first filtered images are aligned according to the matching similarity, and the corresponding position information is obtained. It should be noted that the position information can be obtained as follows: a basic canvas is constructed, the upper right corner of the canvas is set as the reference origin, and the upper right corner of the bamboo board is set to coincide with the reference origin. On this basis, multiple first filtered images are spliced ​​and aligned, and finally the corresponding position information of each first filtered image on the basic canvas is obtained. The position information, feature information and first filtered image are mapped and stored, and for easy distinction, they are recorded as second filtered images.

[0120] The second filtered image set is input into the image synthesis algorithm model, and the image synthesis algorithm performs image fusion on the position information and feature information in the second filtered image to obtain an initial synthesized image.

[0121] The initial composite image is subjected to image post-processing, such as detail enhancement, artifact removal, and color tone adjustment, to obtain a final composite image. It should be noted that the final composite image shown in this embodiment is a defect detection step in the next step. In the rolling coating process of the bamboo board, there is a defect problem in the color uniformity of the film layer after rolling coating. Based on this, a color adjustment feedback step can be added to the image post-processing of the initial composite image. Specifically, the actual color rendering parameters of the conveyor belt or the exclusive color mark under the current light intensity can be obtained, and the color of the current bamboo board can be adjusted according to the actual color rendering parameters. It should be noted that the color adjustment step is to make the appearance of the bamboo board close to the actual color rendering state, not to uniformly process the color of the film layer on the bamboo board. On the contrary, this step can avoid excessive adjustment of the rolled film layer of the bamboo board in the initial composite image by image post-processing, and make the area where the bamboo board is located keep the actual color dimension as much as possible, which is convenient for subsequent defect detection.

[0122] In some embodiments, the final synthesized image is input into a neural network model to obtain defect detection results including:

[0123] Annotating the final synthesized image to generate a plurality of annotation selection boxes, each of which has an annotation area, and recording the annotated final synthesized image as a first annotated image;

[0124] Input the first annotated image into the YOLO-v3 neural network model to obtain at least one annotated area corresponding to an annotated selection box containing the texture of the bamboo board, which is recorded as the second annotated image;

[0125] Inputting the second annotated image into the first classifier to classify whether there is a defect, and obtaining a first classification result, the first classification result including a set of non-defective images and a set of defective images;

[0126] Determine whether the defect image set is an empty set. If not, input the defect image set into a second classifier to classify the defect category to obtain a second classification result, the second classification result includes a first defect image group and a second defect image group, the first defect image group corresponds to a first defect category, the second defect image group corresponds to a second defect category, and the first defect category is different from the second defect category;

[0127] A defect detection result is generated according to the first classification result and the second classification result.

[0128] In this embodiment, the YOLO-v3 neural network model is used to detect defects in the final synthetic image. Specifically, the defects involved in the coating on the bamboo board are prone to cause changes in the texture. The multi-scale detection and deeper network structure of YOLO v3 can better capture and represent texture features. In addition, the introduced FPN can fuse features at different scales to provide more comprehensive texture information. Therefore, YOLO v3 has better performance and accuracy in the field of texture recognition.

[0129] On this basis, the final composite image is first annotated to obtain multiple annotated selection boxes, each of which has an annotated area. The annotated area may include peeling, bubbles, and uneven coloring on the bamboo board, and may also include the annotation of the inherent pattern on the bamboo board. Further, the first annotated image is input into the YOLO-v3 neural network model to obtain at least one annotated area corresponding to the annotated selection box containing the texture of the bamboo board, which is recorded as the second annotated image; this step can identify and distinguish the annotated area, and screen the annotated area containing defects or suspected defects, that is, the annotated selection box containing the texture of the bamboo board in this step can be understood as an annotated area with obvious texture features on the outer surface of the bamboo board, such as the defective texture or non-defective texture (such as the inherent pattern) of the bamboo board. It should be noted that the second annotated image contains at least one annotated area of ​​the texture of the bamboo board, and the second annotated image may also include multiple annotated areas of the texture of the bamboo board.

[0130] The second annotated image is input into the first classifier to classify whether there are defects, and a first classification result is obtained. The first classification result includes a set of defect-free images and a set of defective images. The first classifier can perform preliminary classification of defects and non-defects for multiple annotated areas in the second annotated image. This step does not involve specific classification of defects. If there are no defects in the current bamboo board, the second annotated image is assigned to the set of defect-free images. If there are defects in the annotated area of ​​the current bamboo board, the second annotated image is assigned to the set of defective images.

[0131] Determine whether the defect image set is an empty set. If so, it means that the current bamboo board does not have any rolling defects. If not, the defect image set is input into the second classifier to classify the defect categories to obtain the second classification result. The second classification result includes the first defect image group and the second defect image group. The first defect image group corresponds to the first defect category, and the second defect image group corresponds to the second defect category. The first defect category is different from the second defect category. It should be noted that the second classifier can classify the defects in the annotated area in the second annotated image. The rolling defects of the bamboo board include the film uniformity defects, film flatness defects, film gloss defects, film peeling defects, film peeling defects, film fault defects, etc. described later. Among them, the film uniformity defects, film flatness defects and film gloss defects can be repaired by returning to the factory for repeated rolling when the defect range is small, so as to reduce the scrap rate. Finally, the defect detection result is generated according to the first classification result and the second classification result.

[0132] In some embodiments, the second conveying strategy includes the first adjustment strategy and the second adjustment strategy, and judging whether the current rolling coating process of the bamboo board is qualified according to the defect detection result also includes:

[0133] A first defect image group is selected from the defect categories, the first defect image group includes defect annotated images of the first defect category, recorded as first defect images, and the first defect category is one of a film uniformity defect, a film flatness defect, and a film gloss defect;

[0134] Calculate the defect distribution area on the bamboo board according to the first defect image, and record it as the first defect distribution area;

[0135] Determine whether the first defect distribution area is within a preset qualified threshold range;

[0136] If yes, the rolling coating process of the current bamboo board is determined to be qualified, and a second control instruction is generated, the second control instruction is used to make the current bamboo board be transported according to the first transport strategy;

[0137] If not, determining whether the first defect distribution area is within a preset repair threshold range;

[0138] If the first defect distribution area is within the preset repair threshold range, a first marking label and a second control instruction are generated, and the second control instruction is used to process the bamboo board according to the first adjustment strategy;

[0139] If the first defect distribution area is not within the preset repair threshold range, a second marking label and a third control instruction are generated, and the third control instruction is used to process the bamboo board according to the second adjustment strategy.

[0140] In this embodiment, if the second annotated image of the current bamboo board is finally detected as a defect image, there is a further subdivision of the defect image. Under this premise, the first defect image group is screened out, the first defect image group contains the first defect image, and the first defect image contains multiple defect annotated areas under the first defect category. The first defect category can be film uniformity defects, film flatness defects, and film gloss defects. It should be noted that the defect annotated images under the first defect category can be defects that can be determined based on actual conditions whether they can be returned to the factory for repair.

[0141] Specifically, the defect distribution area on the bamboo board is calculated according to the first defect image and recorded as the first defect distribution area. The specific calculation method can be: accumulating the defect marked areas of the first defect category in the first defect image to obtain the total defect area, and the position of the defect marked area on the bamboo board can also be identified and counted to obtain the distribution position of the defect marked area on the bamboo board.

[0142] It is determined whether the first defect distribution area is within a preset qualified threshold range, and the preset qualified threshold range can be pre-set. If so, the roller coating process of the current bamboo board is determined to be qualified, and a second control instruction is generated. The second control instruction is used to make the current bamboo board be transported according to the first transport strategy. For example, the preset qualified threshold range is set to 0.8%. If the current first defect distribution area is less than 0.8%, it means that the roller coating process of the current bamboo board is qualified, and the current bamboo board can be transported to the next operation area, or the qualified products can be stored.

[0143] If not, determine whether the first defect distribution area is within the preset repair threshold range. This step can further subdivide and confirm the first defect distribution area to determine whether it needs to be returned to the factory for repair. Specifically, if the first defect distribution area is within the preset repair threshold range, a first marking label and a second control instruction are generated. The second control instruction is used to process the bamboo board according to the first adjustment strategy. It should be noted that the first compliance label can be the words "return to the factory for repair" or a graphic symbol to clearly indicate that the current rolling coating process of the bamboo board needs to be re-executed. If the first defect distribution area is not within the preset repair threshold range, a second marking label and a third control instruction are generated. The third control instruction is used to process the bamboo board according to the second adjustment strategy. It should be noted that the second compliance label can be the words "scrap" or a graphic symbol to clearly indicate that the current rolling coating process of the bamboo board is unqualified and needs to be scrapped.

[0144] In some embodiments, the second conveying strategy further includes a third adjustment strategy, and judging whether the current rolling coating process of the bamboo board is qualified according to the defect detection result also includes:

[0145] A second defect image group is selected from the defect categories, the second defect image group includes defect annotated images of the second defect category, recorded as second defect images, and the second defect category is one of a film peeling defect, a film peeling defect, and a film fault defect;

[0146] It is determined whether the second defect image group is an empty set. If not, a third marking label and a fourth control instruction are generated. The fourth control instruction is used to scrap the bamboo board according to the third adjustment strategy.

[0147] This method is another judgment logic for the defective images of bamboo boards. The second defect category is an irreparable defect category and can only be scrapped. If the second defect image group is not an empty set, it means that the current bamboo board has the second defect category and cannot be repaired.

[0148] Further, in some embodiments, the first conveying strategy is configured to convey the bamboo board to a qualified area;

[0149] The first adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a first marking label, and transport the marked bamboo board to a repair area;

[0150] The second adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a second marking label, and transport the marked bamboo board to the first scrapping area;

[0151] The third adjustment strategy is configured to transport the bamboo board to the preset marking area, control the marking component to mark the bamboo board with the third marking label, and transport the marked bamboo board to the second scrapping area.

[0152] This method can achieve detailed classification and processing of bamboo boards with different test results to improve the accuracy of bamboo board testing, while achieving the storage of qualified bamboo boards to facilitate the next operation.

[0153] See also Figure 2 and Figure 3In the second aspect, the present embodiment provides an online detection system for bamboo boards, which is applicable to the method described in the first aspect. The system includes a frame 1, a conveying component 2, a shaping component 3, an image acquisition unit 4 and a marking component 5; the conveying component 2 includes a conveyor belt 22, the conveyor belt 22 is arranged on the frame 1, and a plurality of bamboo boards are placed on the conveyor belt 22; the shaping component 3 includes a first shaping plate 31, a second shaping plate 32, a first adjusting rod 33 and a second adjusting rod 34, the first shaping plate 31 and the second shaping plate 32 are arranged opposite to each other, and the first shaping plate 31 and the second shaping plate 32 are extended along the transmission direction of the conveyor belt 22, the first adjusting rod 33 is connected to the first shaping plate 31, and the first adjusting rod 34 is connected to the first shaping plate 31. The rod 33 is used to adjust the position of the first shaping plate 31 on the conveyor belt 22, the second adjustment rod 34 is connected to the second shaping plate 32, and the second adjustment rod 34 is used to adjust the position of the second shaping plate 32 on the conveyor belt 22; the image acquisition unit 4 includes a camera 41 and a light source 42, the light source 42 is arranged above the conveyor belt 22, and the camera 41 is arranged above the light source 42; the marking component 5 includes a seal device 51 and a first cylinder 52, the seal device 51 is arranged above the conveyor belt 22, and the seal device 51 and the camera 41 are arranged successively along the transmission direction of the conveyor belt 22, and the output end of the first cylinder 52 is connected to the seal device 51, which is used to drive the seal device 51 to move back and forth in the vertical direction.

[0154] In this embodiment, the frame 1 can be constructed by materials such as aluminum alloy profiles, square tubes, rectangular tubes and round tubes. The frame 1 is provided with a first bracket 11 and a second bracket 12. The first bracket 11 is fixed with an image acquisition unit 4, and the second bracket 12 is fixed with a marking component 5. Further, a shield 6 is provided above the frame 1, and a display screen and a touch screen 9 can be integrated on the shield 6, wherein the display screen can be used to display the detection status of the bamboo board in the shield 6 in real time, and the touch screen 9 can be used to input corresponding instructions, such as the retrieval and selection of the bamboo board detection data. An alarm light 7 is provided above the shield 6. When there are problems such as abnormal detection and bamboo board jamming, the alarm of the entire detection system can be realized through the alarm light 7, so that the staff can conduct investigations and improve the safety performance of the entire system.

[0155] In this embodiment, a conveying assembly 2 is provided on the frame 1, and the conveying assembly 2 includes a first driving unit 21, a conveyor belt 22, a button 23 and a first sensor 24, wherein the first driving unit 21 is a servo motor, which is arranged on the frame 1, and the first driving unit 21 is connected to the end roller of the conveyor belt 22 by transmission to drive the conveyor belt 22 to rotate. Bamboo boards can be placed on the conveyor belt 22 in sequence, and an image acquisition unit 4 and a marking assembly 5 are arranged above the conveyor belt 22 to realize the image acquisition of the bamboo board and the marking operation of the bamboo board. Specifically, a button 23 is also provided next to the frame 1, and the button 23 includes an emergency stop button and a start button. The emergency stop button can stop the conveyor belt 22 from transporting, and the start button can start the conveyor belt 22 from transporting. The first sensor 24 can be a photoelectric counter-beam sensor, that is, the number of the first sensors 24 is two, which are relatively arranged on both sides of the conveyor belt 22, so as to detect the presence or absence of the bamboo board of the current conveyor belt 22 and avoid the conveyor belt 22 from being empty.

[0156] In this embodiment, a shaping component 3 is also included. The shaping component 3 can adjust the bamboo board so that the bamboo board can be completely placed in the image acquisition area. Specifically, the shaping component 3 includes a first shaping plate 31, a second shaping plate 32, a first adjustment rod 33, a second adjustment rod 34, a third adjustment rod 35, a fourth adjustment rod 36, a first fixed block 37 and a second fixed block 38. The first shaping plate 31 and the second shaping plate 32 are relatively arranged on the conveyor belt 22. The structure of the first shaping plate 31 and the second shaping plate 32 can be the same. The first shaping plate 31 and the second shaping plate 32 can be formed by bending sheet metal. Optionally, the ends of the first shaping plate 31 and the second shaping plate 32 can be bent to form a V-shaped guide port to facilitate guiding the bamboo board.

[0157] In this embodiment, the first adjusting rod 33 is connected to the first shaping plate 31, the first adjusting rod 33 is connected to the third adjusting rod 35 through the first fixing block 37, the third adjusting rod 35 is connected to the frame 1 through the second fixing block 38, and the first adjusting rod 33 and the third adjusting rod 35 are arranged vertically. This method can realize the adjustment of the first shaping plate 31 in two directions to meet the actual use requirements; similarly, the second adjusting rod 34 is connected to the second shaping plate 32, the second adjusting rod 34 and the fourth adjusting rod 36 are connected to the first fixing block 37, the fourth adjusting rod 36 is connected to the frame 1 through the second fixing block 38, and the second adjusting rod 34 and the fourth adjusting rod 36 are arranged vertically. This method can realize the adjustment of the second shaping plate 32 in two directions to meet the actual use requirements, that is, the number of the first fixing block 37 and the second fixing block 38 are both two, and one first fixing block 37 and one second fixing block 38 are used in combination to achieve the fixing requirements of the first shaping plate 31 and the second shaping plate 32.

[0158] In this embodiment, an image acquisition unit 4 is also included. The image acquisition unit 4 includes a camera 41 and a light source 42. Figure 3 As shown, the light source 42 is arranged below the camera 41 to provide brightness to the bamboo board so that the camera 41 can collect the original image. The camera 41 can be a CCD camera 41 or a line laser camera 41.

[0159] In this embodiment, a marking assembly 5 is also included, and the marking assembly 5 specifically includes a first cylinder 52 and a stamp device 51. The first cylinder 52 is arranged on the second bracket 12, and the stamp device 51 is pushed downward by the first cylinder 52 to press on the surface of the bamboo board to complete the marking operation. In some embodiments, the stamp device 51 can be replaced with a laser marking machine to facilitate adaptation to different marking requirements.

[0160] In a third aspect, the present invention further provides a computer-readable storage medium storing computer program instructions, which implement the method described in the first aspect when executed by a processor.

[0161] In a fourth aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method described in the first aspect.

[0162] The storage medium / memory includes but is not limited to: RAM, ROM, disk, tape, CD, flash memory, USB flash drive, mobile hard disk, memory card, memory stick, network server storage, network cloud storage, etc. The processor includes but is not limited to CPU (central processing unit), GPU (graphics processing unit), MCU (microprocessor), etc.

[0163] Through the steps of image acquisition, image preprocessing, image synthesis and neural network model detection, according to the detection results of the neural network model, it can be judged whether the current rolling coating process of the bamboo board is qualified. If qualified, it can continue to be transported according to the first conveying strategy; if unqualified, the bamboo board can be transported to the preset marking area and marked. Through multiple links, the bamboo board is comprehensively inspected to better capture the image information of the bamboo board and accurately detect defects. In this way, unqualified bamboo boards can be discovered and handled in time, the efficiency of product quality management can be improved, the number of unqualified products can be reduced, the production efficiency and product quality can be improved, and the automated online detection of bamboo boards can be realized. Compared with traditional manual detection, this method can improve detection efficiency and accuracy.

[0164] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0165] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0166] The above descriptions are only some embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Any equivalent device or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A bamboo board online detection method, characterized in that: The method comprises: The bamboo boards are sequentially conveyed to the image acquisition area; Controlling the image acquisition unit to acquire original images of the bamboo board in the current image acquisition area at a first preset frequency to obtain an original image set; Performing image preprocessing on the original images in the original image set one by one to obtain a basic image set, wherein the basic image set includes a plurality of basic images, each of which contains image information of the current bamboo board; Inputting the basic image set into the image synthesis algorithm model in sequence to obtain the final synthesized image of the current bamboo board; Inputting the final synthesized image into a neural network model to obtain a defect detection result, wherein the defect detection result includes a defect annotated image and a defect category; Determine whether the current roller coating process of the bamboo board is qualified according to the defect detection result; If yes, the bamboo board is transported according to the first transport strategy; If not, the current bamboo board is transported to a preset marking area, and the marking component is controlled to mark the bamboo board according to a preset marking label; The marked bamboo board is transported according to the second transport strategy.

2. The online detection method for bamboo board according to claim 1, characterized in that: The original images in the original image set are subjected to image preprocessing one by one, and the basic image set obtained includes: Inputting the original images in the original image set into the Gaussian filter model in sequence to obtain a first processed image after removing image noise; Performing wavelet transform on the first processed image and marking the bamboo board area to obtain a second processed image; The second processed image is cropped to obtain the basic image, wherein the basic image includes the annotation information of the bamboo board area.

3. The online detection method for bamboo board according to claim 2, characterized in that: The basic image set is sequentially input into the image synthesis algorithm model to obtain the final synthetic image of the current bamboo board, including: Performing Gabor filtering on the basic images one by one to obtain a first filtered image; Extracting characteristic information of the current bamboo board in each of the first filtered images according to the annotation information, wherein the characteristic information includes edge features and a collected area of ​​the bamboo board in each of the first filtered images; Perform feature matching on the plurality of feature information to obtain a matching similarity between each feature information and the remaining feature information; Performing image alignment on the first filtered images one by one according to the matching similarity, generating position information of each first filtered image, mapping and storing the position information, feature information and the first filtered image, and recording the mapped and stored first filtered images as second filtered images, to obtain a second filtered image set; Inputting the second filtered image set into an image synthesis algorithm model to obtain an initial synthesized image; The initial synthesized image is subjected to image post-processing to obtain the final synthesized image, wherein the image post-processing includes at least one of detail enhancement, artifact removal, and tone adjustment.

4. The bamboo board online detection method according to claim 3, characterized in that: The final synthesized image is input into the neural network model to obtain defect detection results including: Annotating the final synthesized image to generate a plurality of annotation selection boxes, each of which has an annotation area, and recording the annotated final synthesized image as a first annotated image; Input the first annotated image into a YOLO-v3 neural network model to obtain at least one annotated area corresponding to an annotated selection box containing a bamboo board texture, which is recorded as a second annotated image; Inputting the second annotated image into a first classifier to classify whether there is a defect, and obtaining a first classification result, wherein the first classification result includes a set of non-defective images and a set of defective images; Determine whether the defect image set is an empty set, if not, input the defect image set into a second classifier to classify the defect category, and obtain a second classification result, the second classification result includes a first defect image group and a second defect image group, the first defect image group corresponds to a first defect category, the second defect image group corresponds to a second defect category, and the first defect category is different from the second defect category; The defect detection result is generated according to the first classification result and the second classification result.

5. The online detection method for bamboo board according to claim 4, characterized in that: The second conveying strategy includes a first adjustment strategy and a second adjustment strategy, and judging whether the current rolling coating process of the bamboo board is qualified according to the defect detection result also includes: A first defect image group is selected from the defect categories, the first defect image group includes defect-annotated images of a first defect category, recorded as first defect images, and the first defect category is one of a film uniformity defect, a film flatness defect, and a film gloss defect; Calculate the defect distribution area on the bamboo board according to the first defect image, and record it as the first defect distribution area; Determining whether the first defect distribution area is within a preset qualified threshold range; If yes, the rolling coating process of the current bamboo board is determined to be qualified, and a second control instruction is generated, wherein the second control instruction is used to make the current bamboo board be transported according to the first transport strategy; If not, determining whether the first defect distribution area is within a preset repair threshold range; If the first defect distribution area is within a preset repair threshold range, a first marking label and a second control instruction are generated, and the second control instruction is used to process the bamboo board according to the first adjustment strategy; If the first defect distribution area is not within the preset repair threshold range, a second marking label and a third control instruction are generated, and the third control instruction is used to process the bamboo board according to the second adjustment strategy.

6. The bamboo board online detection method according to claim 5, characterized in that: Judging whether the current roller coating process of the bamboo board is qualified according to the defect detection result also includes: A second defect image group is selected from the defect categories, the second defect image group includes defect-annotated images of the second defect category, recorded as second defect images, and the second defect category is one of a film peeling defect, a film peeling defect, and a film fault defect; It is determined whether the second defect image group is an empty set. If not, a third marking label and a fourth control instruction are generated. The fourth control instruction is used to scrap the bamboo board.

7. The online detection method for bamboo board according to claim 6, characterized in that: The first conveying strategy is configured to convey the bamboo board to a qualified area; The first adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a first marking label, and transport the marked bamboo board to a repair area; The second adjustment strategy is configured to transport the bamboo board to a preset marking area, control the marking component to mark the bamboo board with a second marking label, and transport the marked bamboo board to a first scrapping area.

8. An online detection system for bamboo boards, characterized in that: The method according to any one of claims 1 to 7, wherein the system comprises: frame; A conveying assembly, comprising a conveyor belt, wherein the conveyor belt is arranged on the frame, and a plurality of bamboo boards are placed on the conveyor belt; A shaping assembly, comprising a first shaping plate, a second shaping plate, a first adjusting rod and a second adjusting rod, wherein the first shaping plate and the second shaping plate are arranged opposite to each other, and the first shaping plate and the second shaping plate are extended along the transmission direction of the conveyor belt, the first adjusting rod is connected to the first shaping plate, and the first adjusting rod is used to adjust the position of the first shaping plate on the conveyor belt, the second adjusting rod is connected to the second shaping plate, and the second adjusting rod is used to adjust the position of the second shaping plate on the conveyor belt; An image acquisition unit, comprising a camera and a light source, wherein the light source is arranged above the conveyor belt, and the camera is arranged above the light source; The marking component includes a stamp device and a first cylinder. The stamp device is arranged above the conveyor belt, and the stamp device and the camera are arranged successively along the transmission direction of the conveyor belt. The output end of the first cylinder is connected to the stamp device to drive the stamp device to move back and forth in the vertical direction.

9. A computer-readable storage medium storing computer program instructions, characterized in that: The computer program instructions implement the method of any one of claims 1 to 7 when executed by a processor.

10. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store one or more computer program instructions, wherein the one or more computer program instructions are executed by the processor to implement the method according to any one of claims 1 to 7.

Images