Method for on-line detection of a printing blade damage based on machine vision

Abstract

The application relates to the technical field of image processing, in particular to an online detection method for damage of a printing doctor blade based on machine vision. The method comprises the following steps: collecting a surface image of the doctor blade, performing edge detection on the surface image to obtain an edge image, and obtaining connected domains and corresponding to-be-detected regions of the connected domains; then, analyzing the connected domains and the corresponding to-be-detected regions to obtain edge continuity characteristic values and shape regularity characteristic values of the to-be-detected regions, and then screening all the to-be-detected regions to obtain candidate regions; taking a center point of a candidate region as an origin to determine a main radiation direction, and obtaining pixel point sequences corresponding to the main radiation direction; obtaining a texture periodicity quantization index of the candidate region based on the pixel point sequences corresponding to the main radiation direction of the candidate region; and finally screening all the candidate regions to obtain an abrasion and corrosion region. The application can accurately detect the damage of the printing doctor blade.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an online detection method for damage of a printing doctor blade based on machine vision. BACKGROUND

[0002] As a core component of the printing process, the blade edge morphology and surface integrity of the printing doctor blade directly determine the ink transfer accuracy and printing stability. In the continuous high-speed printing process, the surface of the doctor blade will inevitably form wear and corrosion due to the combined action of mechanical friction and ink chemical medium. This defect is essentially a corrosion of the contact surface caused by vibration or sliding under external load in an aerobic environment. Such damage often appears as microscopic scratches in the early stage, which can easily expand into macro defects under dynamic high pressure. If online identification and intervention are not achieved, it will directly lead to doctor blade cracking, batch printing waste, and even unplanned downtime, significantly affecting production efficiency and resource costs. Therefore, it is extremely important to detect the wear and corrosion of the doctor blade surface. Currently, the printing industry usually uses traditional machine vision technology for detection: the surface image of the doctor blade is acquired by an industrial camera and preprocessed, then morphological and edge detection techniques are used for defect segmentation, and finally defect positioning and recognition are realized based on shape and grayscale features.

[0003] In the detection scene of printing doctor blade wear and corrosion, due to the influence of printing process characteristics and doctor blade working environment, there are often ink residue interferences on the surface. The ink adhesion or dryout in continuous ink scraping operations will form irregular dark areas and broken particle-like residues on the surface. These interference features and real wear and corrosion show high similarity in multiple dimensions: in terms of morphological features, both show irregular non-uniform areas; in terms of optical properties, both show sharp edge grayscale changes and local isolated dark areas. Traditional algorithms based on edge detection and morphological processing can only make judgments based on shallow features such as shape features and grayscale differences, and cannot distinguish the deep differences between wear and corrosion and ink residue, leading to frequent false positives of the detection system. SUMMARY

[0004] To solve the above technical problems, the purpose of the present application is to provide an online detection method for damage of a printing doctor blade based on machine vision, and the technical solution adopted is as follows:

[0005] One embodiment of the present application provides an online detection method for damage of a printing doctor blade based on machine vision, which comprises:

[0006] Collecting the surface image of the doctor blade and performing edge detection on the surface image to obtain an edge image; performing connected component analysis on the edge image to obtain different connected components and edges corresponding to each connected component; and performing closing operation on the edges corresponding to each connected component to obtain a to-be-detected region corresponding to each connected component;

[0007] obtaining an edge continuity characteristic value of each of the to-be-detected regions based on the edges corresponding to the connected domains and the edges of the to-be-detected regions corresponding to the connected domains; obtaining a shape regularity characteristic value based on an area of the to-be-detected region and an area of a convex hull of the to-be-detected region; and screening all the to-be-detected regions based on the edge continuity characteristic value and the shape regularity characteristic value to obtain the candidate regions.

[0008] determining a main radiation direction with a center point of a candidate region as an origin, sequentially taking pixel points along the main radiation direction to form a pixel point sequence corresponding to the main radiation direction, obtaining a texture periodicity quantization index of the candidate region based on distances of the pixel points in the pixel point sequence corresponding to each main radiation direction of the candidate region to the origin and gray values of the pixel points, and screening all the candidate regions based on the texture periodicity quantization index to obtain the wear and corrosion region.

[0009] Preferably, the edges corresponding to the connected domains are subjected to a closing operation to obtain the to-be-detected regions corresponding to the connected domains, including:

[0010] performing a morphological closing operation on the edges of a connected domain to obtain edges of the connected domain after morphological connection processing, and a region surrounded by the edges of the connected domain after the morphological connection processing being a to-be-detected region corresponding to the connected domain.

[0011] Preferably, the edge continuity characteristic value of each of the to-be-detected regions is obtained based on the edges corresponding to the connected domains and the edges of the to-be-detected regions corresponding to the connected domains, including:

[0012] obtaining a difference between a number of edge pixels on the edges of a connected domain and a number of edge pixels on the edges of a to-be-detected region corresponding to the connected domain, and comparing the difference with the number of edge pixels on the edges of the to-be-detected region corresponding to the connected domain to obtain an edge continuity characteristic value of the to-be-detected region.

[0013] Preferably, the shape regularity characteristic value is obtained based on an area of the to-be-detected region and an area of a convex hull of the to-be-detected region, including:

[0014] comparing the area of the to-be-detected region with the area of the convex hull of the to-be-detected region to obtain an area ratio, and taking a difference between a first preset value and the area ratio as the shape regularity characteristic value of the to-be-detected region.

[0015] Preferably, the candidate regions are obtained by screening all the to-be-detected regions based on the edge continuity characteristic value and the shape regularity characteristic value, including:

[0016] The normalized value of the edge continuity feature value and the normalized value of the shape regularity feature value of a to-be-detected region are weighted and summed to obtain a geometric morphology comprehensive index of the to-be-detected region; if the geometric morphology comprehensive index of a to-be-detected region is greater than or equal to a first reference threshold, the to-be-detected region is a candidate region.

[0017] Preferably, the main radiation direction is determined with the center point of a candidate region as the origin, comprising:

[0018] In a plane rectangular coordinate system with the center point of a candidate region as the origin, 0°, 45°, 90° and 135° are determined as the main axis radiation directions.

[0019] Preferably, the texture periodicity quantization index of a candidate region is obtained based on the distance of each pixel point in the pixel point sequence corresponding to each main radiation direction of the candidate region from the origin and the gray value of each pixel point, comprising:

[0020] The distance of each pixel point in the pixel point sequence corresponding to each main radiation direction of a candidate region from the origin is calculated, and a scatter plot is constructed with the distance of each pixel point from the origin as the horizontal axis and the gray value of each pixel point as the vertical axis; adjacent points in the scatter plot are connected to obtain a connected image, and the connected image is smoothed to obtain a smoothed image; extreme points in the smoothed image are taken in the order of the horizontal axis to form an extreme point sequence corresponding to the main radiation direction; the difference between the horizontal coordinates of every two adjacent extreme points in the extreme point sequence is calculated, and the average is obtained as the texture periodicity feature value of the main radiation direction; the difference between the horizontal coordinates of every two adjacent extreme points is the difference between the horizontal coordinates of the latter extreme point and the former extreme point; the main radiation direction with the largest texture periodicity feature value in the candidate region is removed, and the average of the texture periodicity feature values of the other main radiation directions is taken as the texture periodicity quantization index of the candidate region.

[0021] Preferably, all candidate regions are screened based on the texture periodicity quantization index to obtain a wear and corrosion region, comprising:

[0022] A candidate region with a texture periodicity quantization index less than or equal to a second reference threshold is recorded as a wear and corrosion region.

[0023] The application has at least the following beneficial effects: the application obtains the connected domain and the edge of the connected domain by analyzing the surface image of the doctor blade, and then obtains the to-be-detected region corresponding to each connected domain by performing a closing operation on the edge corresponding to each connected domain, and then obtains the edge continuity characteristic value and the shape regularity characteristic value of the to-be-detected region by analyzing the connected domain, the edge corresponding to each connected domain, the edge of the to-be-detected region corresponding to each connected domain, and the area of the to-be-detected region, and finally obtains the candidate region by screening all the to-be-detected regions based on the edge continuity characteristic value and the shape regularity characteristic value, so that the to-be-detected regions are preliminarily screened based on the difference between the wear and corrosion and the ink in the macroscopic geometric shape, the analysis range is reduced, and the overall calculation load is reduced;

[0024] Then, the center point of a candidate region is taken as the origin to determine the main radiation direction, and the pixel points are sequentially taken along a main radiation direction to form a pixel point sequence corresponding to the main radiation direction; the texture periodicity quantitative index of the candidate region is obtained based on the distance of each pixel point in the pixel point sequence corresponding to each main radiation direction of the candidate region from the origin and the gray value of each pixel point; and the wear and corrosion region is obtained by screening all the candidate regions based on the texture periodicity quantitative index, so that the consistency feature of the radial texture distribution is evaluated, the texture diffusion regularity from the center to the periphery is quantified, the subtle difference between the wear and corrosion and the ink residue is accurately captured, and the accuracy of the wear and corrosion region detection of the doctor blade is improved. BRIEF DESCRIPTION OF DRAWINGS

[0025] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 A method flowchart of an online detection method for damage of a doctor blade for printing based on machine vision provided by the embodiment of the application;

[0027] Figure 2 A to-be-detected region schematic diagram of an online detection method for damage of a doctor blade for printing based on machine vision provided by the embodiment of the application;

[0028] Figure 3 A main radiation direction schematic diagram of an online detection method for damage of a doctor blade for printing based on machine vision provided by the embodiment of the application. DETAILED DESCRIPTION

[0029] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an online detection method for printing blade damage based on machine vision proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0031] The following description, in conjunction with the accompanying drawings, details a specific scheme for an online detection method for damage to printing squeegees based on machine vision, provided by the present invention.

[0032] Example: The main application scenario of this invention is as follows: This application proposes a processing idea and method based on "macro-geometric screening-micro-texture subdivision" by combining image processing related algorithms. By deeply exploring the essential differences between wear corrosion and ink in macro-geometric morphology and micro-texture features, it specifically solves the problems and shortcomings of traditional methods for detecting wear corrosion on the surface of the scraper, and achieves efficient and accurate defect detection processing.

[0033] Please see Figure 1 The diagram illustrates a flowchart of an online detection method for printing blade damage based on machine vision, according to an embodiment of the present invention. The method includes the following steps:

[0034] Step S1: Acquire a surface image of the scraper and perform edge detection on the surface image to obtain an edge image; perform connected component analysis on the edge image to obtain different connected components and the edges corresponding to each connected component; perform a closing operation on the edges corresponding to each connected component to obtain the detection area corresponding to each connected component.

[0035] First, an image of the scraper surface needs to be acquired. Specifically, a light source system needs to be set up: a highly uniform coaxial LED light source (wavelength 450-650nm visible light band, power 20-40W) should be used, along with an adjustable polarizer to eliminate ink reflection on the scraper surface; a diffuser should be equipped simultaneously to weaken the problem of overexposure of the sharp edges of the scraper blade; specifically, for the reflection of dried ink on the metal scraper surface (reflectivity 30%-40%), a low-frequency pulsed coaxial light source (frequency 200-500Hz) should be used to enhance the grayscale contrast between the dark areas of wear and corrosion and the ink residue, so that the grayscale difference between the wear and corrosion area and the background is increased by 40-50%, and the ink reflection is prevented from covering the tiny corrosion points.

[0036] Further, select an industrial camera, parameter requirements: resolution ≥ 16MP (pixel size ≤ 1.6μm), frame rate ≥ 50fps, support global shutter and HDR function (dynamic range ≥ 85dB).

[0037] Its lens configuration is: when the flatness of the scraper blade is ±0.2mm, a low-distortion industrial telecentric lens (magnification 0.5X-2.0X, distortion rate ≤0.1%) is selected, the working distance is 150-300mm, the depth of field is ≥1.5mm, and a laser focusing module (accuracy ±0.01mm) is provided.

[0038] The shooting environment is controlled, the ambient light intensity is ≤30lux, a closed light shielding box is installed, the equipment vibration amplitude is <0.01mm to avoid image ghosting, and the working temperature is 20-28℃ to reduce the influence of temperature on the imaging quality.

[0039] Thus, the image of the scraper surface is collected, the entire working surface of the scraper needs to be covered during the collection, and the collected image is verified for clarity (gray scale contrast ≥30% is qualified), so that the image of the scraper surface can be obtained.

[0040] Then, the image of the scraper surface needs to be preprocessed. First, a conventional gray scale method is used to convert the image of the scraper surface into a gray scale image, the quality of the gray scale image is optimized through median filtering to reduce the interference of environmental noise on the printing scraper surface on feature extraction, the gray scale difference between the wear and corrosion area and the normal scraper surface is intensified through global histogram equalization to weaken the visual redundancy of the background uniform area; then, combined with morphological operations such as bottom hat transformation, the surface depression and irregular texture features caused by wear and corrosion are highlighted, and the invalid information of the flat area is weakened, and then the surface image of the scraper is obtained. Then, the Canny operator is used to perform edge detection on the preprocessed surface image of the scraper to obtain an edge image.

[0041] The surface image obtained by the edge detection algorithm has a high similarity in appearance between the real wear and corrosion and the ink residue, and the screening accuracy is often insufficient. In view of this problem, the core of the present application is to construct a two-level discrimination strategy of “macroscopic geometric screening-microscopic texture subdivision”: first, the geometric and morphological index is constructed by combining the differences in edge continuity and shape regularity between the two: the index reflects the edge integrity by quantifying the edge breakage degree, and the geometric regularity of the region is evaluated by combining the shape regularity degree, which constitutes the comprehensive discrimination basis at the macroscopic level; according to the typical characteristics of wear and corrosion, the threshold interval of the geometric and morphological index is set to preliminarily screen the suspicious area, directly exclude the ink area with continuous edge and regular shape, significantly reduce the analysis range, and reduce the overall calculation load.

[0042] Traditional doctor blade wear and corrosion detection methods mostly rely on single visual feature for preliminary screening, and the core defect is that it is difficult to distinguish highly similar ink residue interference in macroscopic morphology, resulting in high false detection rate. In view of this problem, the application proposes a fusion discrimination strategy based on edge-shape double geometric features, the core of which lies in that the physical causes of the two types of defects are essentially different: wear and corrosion as a loss of base material damage, showing high edge fracture degree and irregular shape; while ink residue as a surface attachment, generally showing edge continuity and relatively regular shape. By quantifying edge continuity and shape regularity, a discrimination index of double geometric features is constructed, so as to exclude typical ink interference area by using geometric shape features, and greatly reduce the data size of subsequent processing.

[0043] Specifically, the edge image is subjected to connected domain analysis to obtain different connected domains in the edge image and edges of different connected domains, and a connected domain is denoted as E j , representing the jth connected domain, and the edge pixel points of the connected domain E j are traversed to count the number of edge pixel points in the original edge of the connected domain.

[0044] Further, the edge of a connected domain is subjected to morphological closing operation to obtain the edge of the connected domain after morphological connection processing, and the area surrounded by the edge after morphological connection processing is the detection area corresponding to the connected domain. Thus, the edge of each connected domain, the detection area corresponding to each connected domain, and the edge of the detection area corresponding to each connected domain can be obtained, and then subsequent edge continuity analysis and area shape analysis can be performed.

[0045] In step S2, the edge continuity feature value of each detection area is obtained based on the edge of each connected domain and the edge of the detection area corresponding to each connected domain; the shape regularity feature value is obtained based on the area of the detection area and the area of the convex hull of the detection area; and the candidate area is obtained by screening all detection areas based on the edge continuity feature value and the shape regularity feature value.

[0046] The above step S1 obtains the edge of each connected domain, the detection area corresponding to each connected domain, and the edge of the detection area corresponding to each connected domain. In this step, the edge pixel points before and after the closing operation according to the edge of the connected domain are analyzed, and the shape of the detection area obtained after the closing operation is analyzed.

[0047] An edge continuity characteristic value of each to-be-detected region is obtained based on the edge corresponding to each connected domain and the edge of the to-be-detected region corresponding to each connected domain. Specifically, a difference value between the number of edge pixels on the edge of a connected domain and the number of edge pixels on the edge of the to-be-detected region corresponding to the connected domain is obtained, and the difference value is compared with the number of edge pixels on the edge of the to-be-detected region corresponding to the connected domain to obtain the edge continuity characteristic value of the to-be-detected region.

[0048] The calculation model of the edge continuity characteristic value is specifically as follows:

[0049] ,

[0050] wherein, denotes the edge continuity characteristic value of the to-be-detected region corresponding to the jth connected domain; denotes the number of edge pixels on the edge of the to-be-detected region corresponding to the jth connected domain, that is, the number of edge pixels of the edge of the jth connected domain after morphological connection, denotes the number of edge pixels on the edge of the jth connected domain, the numerator reflects the number of edge breakpoint pixels filled in the closing operation process; a continuous and smooth edge will not change significantly in the number of edge pixels after morphological processing; and a high degree of broken edge needs to add a large number of pixels to connect it into a continuous whole. Finally the value of the edge continuity characteristic value is closer to 1, indicating that the continuity of the edge of the to-be-detected region is poorer, and is more in line with the edge characteristics of the wear and corrosion region.

[0051] Then, the shape regularity characteristic value is obtained based on the area of the to-be-detected region and the area of the convex hull of the to-be-detected region. Specifically, an area ratio is obtained by comparing the area of a to-be-detected region with the area of the convex hull of the to-be-detected region; and a difference value between a first preset value and the area ratio is taken as the shape regularity characteristic value of the to-be-detected region.

[0052] The calculation model of the shape regularity characteristic value is specifically as follows:

[0053] ,

[0054] wherein, denotes the shape regularity characteristic value of the to-be-detected region corresponding to the jth connected domain; denotes the area of the to-be-detected region corresponding to the jth connected domain, denotes the area of the convex hull of the to-be-detected region corresponding to the jth connected domain; reflects the fitting degree of the actual shape of the to-be-detected region and the convex hull shape, and the core idea is that: a region with a regular shape is full, and the actual area is almost equal to the area of the outermost convex hull; and a region with an irregular shape, a recess or a defect, the actual area is significantly smaller than the convex hull area. Finally The value closer to 0 indicates that the to-be-detected region (the region obtained after the edge closing operation of the connected domain) is more regular in shape and more consistent with the shape characteristics of the ink residue region.

[0055] It should be noted that the schematic diagram of the to-be-detected region and the convex hull of the to-be-detected region is as shown in Figure 2 , Figure 2 The left side is a schematic diagram of the to-be-detected region, and the right side is a schematic diagram of the convex hull wrapping the to-be-detected region, wherein the acquisition of the convex hull is a known technology and will not be described in detail here.

[0056] Finally, the edge continuity feature value and the shape regularity feature value of the to-be-detected region need to be fused to obtain a geometric morphology comprehensive index. Specifically, the normalized value of the edge continuity feature value and the normalized value of the shape regularity feature value of a to-be-detected region are weighted and summed to obtain the geometric morphology comprehensive index of the to-be-detected region.

[0057] The specific calculation model of the geometric morphology comprehensive index is:

[0058]

[0059] wherein Norm[·] represents a maximum-minimum normalization function, which constrains the calculation result in the interval [0, 1] and eliminates the dimensional difference. Alpha is a weight parameter for measuring the contribution degree of edge continuity and shape regularity. In order to match the physical differences of the doctor blade defects, the contributions of edge and shape features should be balanced to ensure stable discrimination ability for different morphological defects and disturbances. Therefore, the present application gives a reference value of alpha=0.5, which can be adjusted according to the scene requirements in actual application. and respectively represent the edge continuity feature value and the shape regularity feature value of a to-be-detected region; the geometric morphology comprehensive index has a value range of [0, 1], and the numerical size directly reflects the possibility of the region belonging to wear corrosion. When the value tends to 0, it represents a region with continuous edge and regular shape, which matches the typical characteristics of ink residue; when the value tends to 1, it represents a region with broken edge and irregular shape, which matches the geometric properties of wear corrosion.

[0060] ​In summary, the first reference threshold T1 = 0.4 is set, and the actual can be adjusted on demand: when the geometric morphology comprehensive index of a to-be-detected region is greater than or equal to the first reference threshold, the to-be-detected region is determined as a region meeting the wear corrosion feature and is reserved as a candidate region; when the geometric morphology comprehensive index is less than the first reference threshold, it is determined as an ink residue interference region and is excluded. Each to-be-detected region is processed in turn in combination with the sequential traversal strategy, and the discrimination is executed in a loop until all regions are processed, and finally a set of candidate regions is obtained, thereby greatly reducing the data scale of subsequent processing and providing high-quality input for subsequent accurate classification.

[0061] In step S3, the main radiation direction is determined with the center point of a candidate region as the origin, and the pixel points are sequentially taken along a main radiation direction to form a pixel point sequence corresponding to the main radiation direction; the texture periodicity quantitative index of the candidate region is obtained based on the distance of each pixel point in the pixel point sequence corresponding to each main radiation direction of the candidate region from the origin and the gray value of each pixel point; and the wear corrosion region is obtained by screening all candidate regions based on the texture periodicity quantitative index.

[0062] In step S2, the geometric morphology index is constructed according to the differences in edge continuity and shape regularity characteristics, the to-be-detected regions are preliminarily screened, and the candidate regions are obtained. For the candidate regions, micro-texture analysis is further introduced, the consistency feature of radial texture distribution is evaluated, the texture diffusion regularity from the center to the periphery is quantified, the subtle differences between wear corrosion and ink residue are accurately captured, and a discrimination function is constructed to provide a quantitative basis for accurate classification of the two types of regions.

[0063] In the candidate regions obtained by macro-geometric feature screening, there are still some stubborn ink interference that is highly similar to the real wear corrosion in edge and shape features. Therefore, the core of this step is to introduce the consistency analysis of micro-texture structure: since wear corrosion is a missing damage of the base material, the exposed material texture usually presents obvious directionality, and linear wear texture along the material crystal direction or stress direction often appears inside the region; and the ink residue as a surface adherend has randomness in the film forming or accumulation process, resulting in uniform and disordered texture distribution. This step aims to quantize the texture change pattern in multiple directions radiating from the region center to the periphery, effectively capture the essential difference between the two types of defects in micro-texture structure, and construct a discrimination function to provide a reliable quantitative basis for the final accurate classification of wear corrosion and ink residue.

[0064] Thus, the geometric centroid algorithm is used to determine the center point of the candidate region, and its coordinates are marked as (x0, y0). Further, in the Cartesian coordinate system with the point (x0, y0) as the origin, eight radial directions can be defined based on the eight-neighborhood. However, due to the axial coincidence of the direction vectors (for example, the 0° and 180° directions are collinear), these directions are not all independent. Therefore, the four directions of 0°, 45°, 90°, and 135° are determined as the principal axis radial directions, which are linearly independent on the plane and can effectively represent the neighborhood structure, as shown in the red arrow directions in the following figure. Figure 3 The scene core contradiction is to distinguish the directionality and periodicity of the texture of the wear and corrosion from the disordered and aperiodic texture of the ink residue. The periodic texture of wear and corrosion can appear in any linear direction on the plane, and the four directions of 0°, 45°, 90°, and 135° satisfy the linear independence on the plane, that is, the periodic characteristics of any direction cannot be derived from other directions. Therefore, these four directions can completely cover all possible linear periodic directions on the plane, ensuring that the periodic texture along any principal direction can be captured, and the periodic texture will not be missed due to the lack of direction, and the periodicity misjudgment caused by incomplete direction coverage can be avoided.

[0065] Further, the pixel points are collected along a principal radial direction from the center point of a candidate region to the boundary of the candidate region, and the collected pixel points are arranged in a sequence according to the collection order, serving as the pixel point sequence corresponding to the principal radial direction. Thus, the pixel point sequence corresponding to each principal radial direction of each candidate region can be obtained, ensuring that the texture information of each direction is completely covered.

[0066] Then, the texture periodicity quantization index of the candidate region is obtained based on the distance of each pixel point from the origin and the gray value of each pixel point in the pixel point sequence corresponding to each principal radial direction of the candidate region.

[0067] Specifically, distances of each pixel point in a pixel point sequence corresponding to a main radiation direction of a candidate region from an origin are calculated, and a scatter plot is constructed with the distances of each pixel point from the origin as the horizontal axis and the gray values of each pixel point as the vertical axis; adjacent points in the scatter plot are connected to obtain a connected image, the connected image is smoothed to obtain a smoothed image, and extreme value points in the smoothed image are taken in the order of the horizontal axis to form an extreme value point sequence corresponding to the main radiation direction; a difference value of horizontal coordinates of each two adjacent extreme value points in the extreme value point sequence is calculated, and an average value of the difference values is obtained as a texture periodicity quantitative index of the candidate region.

[0068] The connected image obtained by connecting adjacent points in the scatter plot presents significant high-frequency jagged fluctuations due to the influence of imaging noise, material microscopic unevenness and other factors on the original data. In order to intuitively reflect the overall trend of the gray scale change, the original data on the connected image is smoothed by using Gaussian filtering to suppress high-frequency noise and highlight macroscopic change rules to obtain the smoothed image. In addition, the distance of the pixel point from the origin is the Euclidean distance.

[0069] The calculation model of the texture periodicity characteristic value of one main radiation direction is specifically:

[0070] ,

[0071] Wherein, represents the texture periodicity characteristic value of the dth main radiation direction of a candidate region, and M represents the number of extreme value points in the extreme value point sequence corresponding to the main radiation direction. and respectively represent the horizontal coordinates of the i+1th extreme value point and the ith extreme value point in the extreme value point sequence, which can also be said to be the distances of the pixel points corresponding to the two extreme value points from the origin. can represent the average interval of the periodic change of the gray scale, which is the interval between the two extreme value points, and the core logic is to distinguish the microscopic texture essence difference between wear corrosion and ink residue by quantifying the regularity of texture repetition.

[0072] The texture periodicity characteristic values of the four main radiation directions are sorted, and the largest texture periodicity characteristic value is identified and removed. The largest texture periodicity characteristic value usually corresponds to a main radiation direction parallel to the main direction of the linear texture of wear corrosion, for example Figure 3When the 0° main radiation direction is parallel to the linear wear texture represented by the dashed line inside the region, the texture period characteristic value will be significantly larger due to the consistency of the texture direction. Such components cannot effectively represent the real distribution of the texture in the candidate region, and therefore need to be removed. After removal, the texture period characteristic values of the three main radiation directions remaining in a candidate region are averaged to obtain the texture periodicity quantization index of the candidate region. That is, the texture period characteristic values of the effective main radiation directions are integrated into a single quantization index. In the effective main radiation direction, the pixel sequence gray change periodicity is strong, the period value is concentrated, and the arithmetic mean of the texture period characteristic value is small. The ink residue has no fixed period, and the period value is randomly dispersed in the effective main radiation direction, and the arithmetic mean of the texture period characteristic value is large.

[0073] Thus, the texture periodicity quantization index of each candidate region can be obtained. The value range of the texture periodicity quantization index is (0, 1], and the value quantizes the periodicity distribution of the texture in the candidate region: the wear corrosion region has parallel linear wear texture inside, and the gray sequence periodicity is strong along the wear vertical direction, and the value of the texture periodicity quantization index is generally low; the ink residue region has random film forming process, and the texture has no fixed period, and the value of the texture periodicity quantization index is generally high. Based on the above, the second reference threshold T2 is set to 0.5, which is the middle value of the value distribution, maximizes the distinguishing ability of the micro-texture characteristics, and balances the risk of false and missed judgment. In practical applications, the specific imaging conditions and detection requirements can be dynamically adjusted.

[0074] Thus, all candidate regions are screened according to the texture periodicity quantization index to obtain the wear corrosion region. Specifically, the candidate region with a texture periodicity quantization index less than or equal to the second reference threshold is recorded as a wear corrosion region. The candidate region with a texture periodicity quantization index equal to the second reference threshold is determined as an ink residue interference region and is removed. The verified wear corrosion defect region is marked and the detection result is output, completing the complete online detection process from the surface image of the doctor blade to the doctor blade defect.

[0075] In summary, the present application combines image processing related algorithms to propose a processing idea and method based on "macro-geometric screening-micro-texture subdivision". By deeply mining the essential differences between wear corrosion and ink in macro-geometric shape and micro-texture characteristics, the problems and deficiencies of traditional methods for detecting wear corrosion on the surface of the doctor blade are solved, and efficient and accurate defect detection processing is realized.

[0076] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0077] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for on-line detection of damage to a printing blade based on machine vision, characterized by, The method comprises: Collecting a surface image of the scraper and performing edge detection on the surface image to obtain an edge image; performing connected domain analysis on the edge image to obtain different connected domains and edges corresponding to each connected domain; performing a closing operation on the edges corresponding to each connected domain to obtain a to-be-detected region corresponding to each connected domain; Based on the edges corresponding to each connected domain and the edges of the to-be-detected region corresponding to each connected domain, an edge continuity feature value of each to-be-detected region is obtained; based on the area of the to-be-detected region and the area of the convex hull of the to-be-detected region, a shape regularity feature value is obtained; based on the edge continuity feature value and the shape regularity feature value, all to-be-detected regions are screened to obtain a candidate region; A center point of a candidate region is taken as an origin to determine a main radiation direction, and pixel points are sequentially taken along one main radiation direction to form a pixel point sequence corresponding to the main radiation direction. Based on the distances of the pixel points in the pixel point sequence corresponding to each main radiation direction of a candidate region from the origin and the gray values of the pixel points, a texture periodicity quantization index of the candidate region is obtained, including: calculating the distances of the pixel points in the pixel point sequence corresponding to one main radiation direction of a candidate region from the origin, and constructing a scatter plot with the distances of the pixel points from the origin as the horizontal axis and the gray values of the pixel points as the vertical axis; connecting adjacent points in the scatter plot to obtain a connected image, smoothing the connected image to obtain a smoothed image, and sequentially taking extreme points in the smoothed image as extreme point sequences corresponding to the main radiation direction; calculating the difference between the horizontal coordinates of every two adjacent extreme points in the extreme point sequence, and averaging to obtain a texture period feature value of the main radiation direction; the difference between the horizontal coordinates of every two adjacent extreme points is the difference between the horizontal coordinates of the latter extreme point and the former extreme point; removing the main radiation direction with the largest texture period feature value in the candidate region, and calculating the average of the texture period feature values of the other main radiation directions as the texture periodicity quantization index of the candidate region; Based on the texture periodicity quantization index, all candidate regions are screened to obtain a wear and corrosion region.

2. The method for online detection of damage of a printing blade based on machine vision according to claim 1, characterized in that, The closing operation on the edges corresponding to each connected domain to obtain the to-be-detected region corresponding to each connected domain comprises: Performing a morphological closing operation on the edges of one connected domain to obtain edges after morphological connection processing corresponding to the connected domain, and the region surrounded by the edges after morphological connection processing is the to-be-detected region corresponding to the connected domain.

3. The method of claim 1, wherein the method is characterized by: The edge continuity feature value of each to-be-detected region is obtained based on the edges corresponding to each connected domain and the edges of the to-be-detected region corresponding to each connected domain, comprising: The difference between the number of edge pixels on the edges of one connected domain and the number of edge pixels on the edges of the to-be-detected region corresponding to the connected domain is obtained, and compared with the number of edge pixels on the edges of the to-be-detected region corresponding to the connected domain to obtain the edge continuity feature value of the to-be-detected region.

4. The method for online detection of damage of a printing doctor blade based on machine vision according to claim 1, characterized in that, The shape regularity feature value is obtained based on the area of the to-be-detected region and the area of the convex hull of the to-be-detected region, comprising: The area ratio is obtained by comparing the area of a to-be-detected region with the area of a convex hull of the to-be-detected region; and a difference between a first preset value and the area ratio is taken as a shape regularity feature value of the to-be-detected region.

5. The method of claim 1, wherein the method is characterized by: The screening of all the to-be-detected regions based on the edge continuity feature value and the shape regularity feature value to obtain candidate regions comprises: The geometric morphology comprehensive index of a to-be-detected region is obtained by weighted summation of a normalized value of the edge continuity feature value and a normalized value of the shape regularity feature value of the to-be-detected region; if the geometric morphology comprehensive index of a to-be-detected region is greater than or equal to a first reference threshold, the to-be-detected region is a candidate region.

6. The method of claim 1, wherein the method is characterized by: The main radiation direction is determined with a center point of a candidate region as an origin, comprising: In a plane rectangular coordinate system with the center point of the candidate region as an origin, 0°, 45°, 90° and 135° are determined as main axis radiation directions.

7. The method for online detection of damage of a printing blade based on machine vision according to claim 1, characterized in that, The screening of all the candidate regions based on the texture periodicity quantization index to obtain a wear and corrosion region comprises: A candidate region with a texture periodicity quantization index less than or equal to a second reference threshold is recorded as a wear and corrosion region.

Images