Denitration catalyst appearance defect detection method and system based on machine vision

Abstract

The invention discloses a denitration catalyst appearance defect detection method and system based on machine vision, relates to a machine vision detection technology, and is used for solving the problems of low detection efficiency, incomplete image coverage and single defect identification of an existing detection method. The method specifically comprises the following steps: acquiring a denitration catalyst surface image through an industrial camera, segmenting the denitration catalyst surface image, preprocessing the denitration catalyst surface image, detecting whether the preprocessed denitration catalyst surface image has defects or not in advance, performing refined operation on the basis of a gray value of the denitration catalyst surface image with the defects or not to obtain a target defect, and dividing defect types. According to the method, the problems of smooth and reflective surface and fuzzy edge of the denitration catalyst are effectively solved, defect missing detection caused by image quality is avoided, and the detection efficiency is improved.

Description

A Machine Vision-Based Method and System for Detecting Appearance Defects in Denitrification Catalysts Technical Field

[0001] This invention belongs to the field of machine vision inspection technology, specifically relating to a method and system for detecting appearance defects in denitrification catalysts based on machine vision. Background Technology

[0002] Denitrification catalysts are core components of industrial flue gas denitrification systems, commonly used in the purification of process tail gases in chemical plants to reduce air pollution. However, denitrification catalysts are prone to defects such as cracks, breakage, and surface contamination during production due to factors such as raw material ratios, molding processes, and calcination temperatures. Currently, the detection of appearance defects in denitrification catalysts mainly relies on manual visual observation and direct touch. The results depend on the operator's subjective experience, leading to significant differences in judgment among different personnel and making consistency difficult to guarantee. Furthermore, the detection efficiency is low, and the high dust density in denitrification catalyst manufacturing workshops creates a harsh environment that not only affects the accuracy of operator judgment but also easily leads to missed or false detections. Alternatively, laser scanning technology can be used for detection, but high-precision laser scanners are expensive, and dust particles in the workshop can obstruct the laser signal, compromising the integrity and accuracy of the scanned data.

[0003] The invention patent with publication number CN104483320 B discloses a digital defect detection method for industrial denitrification catalysts. The method uses a digital defect detection device and is characterized by the following steps: 1) The denitrification catalyst product (10) on the production line is conveyed by the conveyor belt (7). The position of the denitrification catalyst product (10) is detected by the infrared detector (5). When the product reaches the detection position, a signal is transmitted to stop the conveyor belt (7). The light source (6) illuminates the product. The QR code scanning device (11) scans the QR code information of the denitrification catalyst product (10) and transmits it to the PC host computer (1); 2) The industrial camera group (3) collects images of the surface to be tested of the denitrification catalyst product (10) in blocks. The four industrial cameras of the industrial camera group (3) respectively collect images of the four corners of the surface to be tested of the denitrification catalyst product (10), and the four images are overlapped. Then the images are transmitted to the PC host computer (1); 3) The PC host computer (1) processes the four images in sequence through image preprocessing, image processing and image processing. After registration, transformation model establishment, and transformation to the same coordinate, the overlapping parts of each image are fused to obtain a smooth and seamless panoramic image; 4) Gray-scale processing of the panoramic image is performed using gray-scale transformation method; 5) Feature vector extraction is performed on the gray-scale processed panoramic image; 6) The feature vector is classified using a trained classifier and distinguished as defect feature vector or non-defect feature vector: if the feature vector of any sub-region is a defect feature vector, then the denitrification catalyst product (10) is a defective product; if the feature vectors of all sub-regions are non-defective feature vectors, then the denitrification catalyst product (10) is a qualified product: the defect information obtained after distinction is input into the PC host computer (1); 7) The conveyor belt (7) continues to work, conveying the denitrification catalyst product (10) to the robot arm (9), and the robot arm (9) extracts the defective product by obtaining the QR code information and defect information of the denitrification catalyst product (10) from the PC host computer (1). The existing technology has the following defects: low detection efficiency, incomplete image acquisition coverage, and single defect identification. This is a shortcoming of existing technology.

[0004] In view of this, it is very necessary to provide a method and system for detecting appearance defects of denitrification catalysts based on machine vision to solve the above-mentioned defects in the prior art. Summary of the Invention

[0005] To address the technical problems of low detection efficiency, incomplete image acquisition coverage, and limited defect identification in existing technologies, this invention provides a machine vision-based method and system for detecting appearance defects in denitrification catalysts, thereby solving the aforementioned technical problems.

[0006] In a first aspect, the present invention provides a method for detecting appearance defects of denitrification catalysts based on machine vision, comprising: step S1: acquiring surface images of the denitrification catalyst by taking pictures of the denitrification catalyst with an industrial camera; the industrial camera is set at the outlet of a tunnel calcining kiln, and a parallel light source is set at the outlet of the tunnel calcining kiln, the parallel light source being installed at an oblique angle to irradiate the denitrification catalyst.

[0007] Step S2: Preprocessing the surface image of the denitrification catalyst, segmenting the surface image of the denitrification catalyst, and performing line fitting and feature extraction on the segmented surface image of the denitrification catalyst; the surface image of the denitrification catalyst includes the image of the denitrification catalyst, the image of the conveyor shaft, the image of the mechanical frame, and the image of the asbestos rope; the image of the conveyor shaft, the image of the mechanical frame, and the image of the asbestos rope are designated as the background image; segmenting the surface image of the denitrification catalyst includes segmenting the denitrification catalyst image and the background image; using the k-means clustering algorithm to segment the surface image of the denitrification catalyst, clustering the pixels in the surface image of the denitrification catalyst, selecting a representative pixel for each class, and assimilating the pixels of the surface image of the denitrification catalyst to the representative pixel of the corresponding class; using the gradient descent method to process the segmented surface image of the denitrification catalyst. Line fitting is performed on the edges to correct the smoothness of the edge contour of the denitrification catalyst surface image. The denitrification catalyst surface image after line fitting is then labeled using a rectangular labeling box. Each denitrification catalyst is selected and labeled with a tag number to distinguish different lengths of catalyst monomers. The corresponding relationships are as follows: tag number 0 corresponds to 50mm; tag number 1 corresponds to 55mm; tag number 2 corresponds to 60mm; tag number 3 corresponds to 70mm; tag number 4 corresponds to 80mm; tag number 5 corresponds to 90mm; tag number 6 corresponds to 100mm; tag number 7 corresponds to 110mm; and tag number 8 corresponds to 120mm. After image labeling, the pixel coordinate data labels of the denitrification catalyst surface image are exported. The exported data labels are then normalized using the following mathematical expression:

[0008] Where (x,y) represents the center coordinates of the rectangular label box, TL and BL represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, TR and BR represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, and len and wid represent the length and width of the rectangular label box, respectively; a YOLOv5 model is constructed to process the normalized data labels and extract image features.

[0009] Step S3: This step involves classifying the defect types of the denitrification catalyst. The pre-processed surface image of the denitrification catalyst is pre-detected for defects. Based on the grayscale values ​​of the defect-free surface image, a fine-grained operation is performed to obtain the target defects. The minimum bounding rectangle is calculated for each target defect, and the set features of the minimum bounding rectangle are obtained. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, the denitrification catalyst defect types are classified. Denitrification catalyst defect types include cracks, damage, and contamination. Pre-detecting the pre-processed surface image for defects includes binarizing the pre-processed surface image and calculating the pixel threshold T0. Based on the pixel threshold T0, the surface image of the denitrification catalyst is divided into defective and defect-free parts. Images with pixel values ​​greater than T0 are defective parts; images with pixel values ​​less than T0 are defect-free parts. Calculating the pixel threshold T0 includes calculating extreme value normalization and calculating the random perturbation coefficient. The mathematical expression for calculating extreme value normalization is:

[0010] Where S represents the normalized extreme value. and Let g represent the maximum and minimum grayscale values ​​of the target image g, where the target image is an image of the surface of the denitrification catalyst; the mathematical expression for calculating the random perturbation coefficient is:

[0011] Where K is the random disturbance coefficient. The pixel threshold T0 represents the grayscale value of a random pixel in the target image g; the mathematical expression for T0 is: After the target image is segmented for the first time based on polarization processing, it is divided into two parts, f and p, where f is the defective part and p is the defect-free part. The target defects are obtained by refining the gray values ​​of the surface images of the denitrification catalyst with and without defects. This includes updating the pixel threshold based on the gray values ​​of the two parts. The target defects are obtained after n segmentations of the target image. Updating the pixel threshold includes calculating a dynamic correction coefficient and iteratively updating the threshold. The mathematical expression for the dynamic correction coefficient is:

[0012] in, This is a dynamic correction factor. and These represent the maximum and minimum grayscale values ​​of the segment f, respectively. and Let p represent the maximum and minimum grayscale values ​​of the segmented part p, respectively, and n represent the number of iterations; the mathematical expression for threshold iteration update is:

[0013] in, Let be the pixel threshold after n iterations. After n iterations of the target image g, the target defect is obtained. At this point, the target defect is represented by discrete pixels. The defect features are defined by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle of the discrete pixels. Assuming the pixel coordinates of the four points of the minimum bounding rectangle are P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4), the mathematical expression for the length of the minimum bounding rectangle is: The mathematical expression for the minimum bounding rectangle width is: The mathematical expression for the area of ​​the minimum circumscribed rectangle is: The mathematical expression for the minimum circumscribed aspect ratio is: The mathematical expression for roundness is: The defect characteristics were classified by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle, resulting in a corresponding table of defect characteristics. Specifically: when A > 2893, the surface of the denitrification catalyst is considered cracked; when A > 6389, the surface is considered damaged; when A > 9877, the surface is considered dirty. When L > 291, the surface is considered cracked; when L > 289, the surface is considered damaged; when L > 194, the surface is considered dirty. When W > 23, the surface is considered... The catalyst surface is considered cracked when W>34, damaged when W>69, and dirty when W>69; the catalyst surface is considered cracked when B>9.215, damaged when B>6.392, and dirty when B>2.482; the catalyst surface is considered cracked when C>0.094, damaged when C>0.726, and dirty when C>0.923.

[0014] Secondly, the present invention also provides a machine vision-based system for detecting defects in the appearance of denitrification catalysts, including a module for acquiring surface images of the denitrification catalyst, a module for preprocessing surface images of the denitrification catalyst, and a module for classifying the types of defects in the denitrification catalyst. The module for acquiring surface images of the denitrification catalyst uses an industrial camera to capture images of the denitrification catalyst surface. The industrial camera is located at the outlet of a tunnel-type calcining kiln. The industrial camera is a CMOS area array camera capable of high-speed imaging. A parallel light source is installed at the outlet of the tunnel-type calcining kiln, and the parallel light source is installed obliquely to irradiate the denitrification catalyst.

[0015] The preprocessing module for the denitrification catalyst surface image segments the denitrification catalyst surface image and performs linear fitting and feature extraction on the segmented denitrification catalyst surface image. The denitrification catalyst surface image includes the denitrification catalyst image, the conveyor shaft image, the mechanical frame image, and the asbestos rope image. The conveyor shaft image, the mechanical frame image, and the asbestos rope image are designated as the background image. The segmentation of the denitrification catalyst surface image includes segmenting the denitrification catalyst image and the background image.

[0016] The module for classifying denitrification catalyst defect types pre-detects the presence of defects in the pre-processed surface image of the denitrification catalyst. Based on the grayscale values ​​of the surface image of the denitrification catalyst with and without defects, a fine-grained operation is performed to obtain the target defects. The minimum bounding rectangle of the target defects is calculated, and the set features of the minimum bounding rectangle are obtained. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, the denitrification catalyst defect types are classified.

[0017] The beneficial effects of this invention are as follows: The present invention provides a machine vision-based method and system for detecting appearance defects in denitrification catalysts. It acquires images in real time using an industrial camera, allowing integration into continuous production processes without interrupting the production line, thus improving detection efficiency and adapting to large-scale mass production needs. It also solves the problem of incomplete image coverage. Through linear fitting and feature extraction in image preprocessing, it effectively addresses the issues of smooth, reflective surfaces and blurred edges on denitrification catalysts, avoiding missed defects due to image quality. By pre-detecting the presence of defects, it first roughly divides defective and non-defective parts, and then calculates the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle for the target defect, accurately classifying cracks, breaks, and dirt defects. This solves the problem of single defect identification and meets the needs of defect cause tracing and classification control in production.

[0018] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 is a flowchart of a machine vision-based method for detecting appearance defects in denitrification catalysts provided by the present invention.

[0021] Figure 2 is a schematic diagram of a machine vision-based denitrification catalyst appearance defect detection system provided by the present invention. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0023] 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. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0024] Example 1: As shown in Figure 1, this embodiment of the invention provides a method for detecting appearance defects of denitrification catalysts based on machine vision, including the following steps: Step S1: The step of acquiring images of the surface of the denitrification catalyst. The denitrification catalyst is photographed by an industrial camera to acquire images of its surface. The industrial camera is set at the outlet of the tunnel calcining kiln. A parallel light source is set at the outlet of the tunnel calcining kiln. The parallel light source is installed at an oblique angle to irradiate the denitrification catalyst and eliminate the influence of light reflection on the surface of the denitrification catalyst.

[0025] Step S2: Preprocessing the surface image of the denitrification catalyst, segmenting the surface image of the denitrification catalyst, and performing linear fitting and feature extraction on the segmented surface image of the denitrification catalyst; the surface image of the denitrification catalyst includes the image of the denitrification catalyst, the image of the conveyor shaft, the image of the mechanical frame, and the image of the asbestos rope; the image of the conveyor shaft, the image of the mechanical frame, and the image of the asbestos rope are designated as the background image; segmenting the surface image of the denitrification catalyst includes segmenting the denitrification catalyst image and the background image; using the k-means clustering algorithm to segment the surface image of the denitrification catalyst, clustering the pixels in the surface image of the denitrification catalyst, selecting a representative pixel for each class, and assimilating the pixels of the surface image of the denitrification catalyst into the representative pixels of the corresponding class; using the gradient descent method to perform linear fitting and feature extraction on the segmented surface image of the denitrification catalyst. Linear fitting is performed on the edges of the catalyst surface image to correct the smoothness of the edge contour. The linearly fitted catalyst surface image is then labeled using a rectangular labeling box to select each catalyst element. Different lengths of catalyst elements are distinguished by label numbers, with the following correspondences: label number 0 corresponds to 50mm; label number 1 corresponds to 55mm; label number 2 corresponds to 60mm; label number 3 corresponds to 70mm; label number 4 corresponds to 80mm; label number 5 corresponds to 90mm; label number 6 corresponds to 100mm; label number 7 corresponds to 110mm; and label number 8 corresponds to 120mm. Preferably, the web-based image labeling tool LabelMe is used to label the acquired catalyst surface images.

[0026] After image annotation, the pixel coordinate data labels of the denitrification catalyst surface image are exported. The exported data labels are then normalized, and the mathematical expression is as follows:

[0027] Where (x, y) represents the center coordinates of the rectangular label box, TL and BL represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, TR and BR represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, and len and wid represent the length and width of the rectangular label box, respectively. A YOLOv5 model is constructed to process the normalized data labels and extract image features. For example, in this embodiment, a four-layer YOLOv5 model is used to process the dataset. The Input layer scales and enhances the image, the Backbone extracts image features, the Neck performs multi-scale fusion of features, and the Head extracts image features.

[0028] Step S3: This step involves classifying the defect types of the denitrification catalyst. The pre-processed surface image of the denitrification catalyst is pre-detected for defects. Based on the grayscale values ​​of the defect-free surface image, a fine-grained operation is performed to obtain the target defects. The minimum bounding rectangle is calculated for each target defect, and the set features of the minimum bounding rectangle are obtained. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, the denitrification catalyst defect types are classified. Denitrification catalyst defect types include cracks, breakage, and contamination. Pre-detecting the pre-processed surface image for defects includes binarizing the pre-processed surface image, calculating the pixel threshold T0, and dividing the surface image into defective and defect-free parts based on the pixel threshold. Images with pixel values ​​greater than T0 are considered defective parts; images with pixel values ​​less than T0 are considered defect-free parts. Calculating the pixel threshold T0 includes calculating extreme value normalization and calculating the random perturbation coefficient. The mathematical expression for calculating extreme value normalization is:

[0029] Where S represents the normalized extreme value. and Let g represent the maximum and minimum grayscale values ​​of the target image g, where the target image is an image of the surface of the denitrification catalyst; the mathematical expression for calculating the random perturbation coefficient is:

[0030] Where K is the random disturbance coefficient. The pixel threshold T0 represents the grayscale value of a random pixel in the target image g; the mathematical expression for T0 is: After the target image is segmented for the first time based on polarization processing, it is divided into two parts, f and p, where f is the defective part and p is the defect-free part. The target defects are obtained by refining the gray values ​​of the surface images of the denitrification catalyst with and without defects. This includes updating the pixel threshold based on the gray values ​​of the two parts. The target defects are obtained after n segmentations of the target image. Updating the pixel threshold includes calculating a dynamic correction coefficient and iteratively updating the threshold. The mathematical expression for the dynamic correction coefficient is:

[0031] in, This is a dynamic correction factor. and These represent the maximum and minimum grayscale values ​​of the segment f, respectively. and Let p represent the maximum and minimum grayscale values ​​of the segmented part p, respectively, and n represent the number of iterations; the mathematical expression for threshold iteration update is:

[0032] in, Let be the pixel threshold after n iterations. After n iterations of the target image g, the target defect is obtained. At this point, the target defect is represented by discrete pixels. The defect features are defined by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle of the discrete pixels. Assuming the pixel coordinates of the four points of the minimum bounding rectangle are P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4), the mathematical expression for the length of the minimum bounding rectangle is: The mathematical expression for the minimum bounding rectangle width is: The mathematical expression for the area of ​​the minimum circumscribed rectangle is: The mathematical expression for the minimum circumscribed aspect ratio is: The mathematical expression for roundness is: .

[0033] For example, in this embodiment, the defect features are classified by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle, resulting in a defect feature correspondence table. Specifically: when A>2893, the surface of the denitrification catalyst is identified as cracked; when A>6389, the surface of the denitrification catalyst is identified as damaged; when A>9877, the surface of the denitrification catalyst is identified as dirty; when L>291, the surface of the denitrification catalyst is identified as cracked; when L>289, the surface of the denitrification catalyst is identified as damaged; when L>194, the surface of the denitrification catalyst is identified as dirty; when W>23, the surface of the denitrification catalyst is identified as... The surface of the denitrification catalyst is considered cracked when W>34, damaged when W>69, and dirty when W>69; the surface of the denitrification catalyst is considered cracked when B>9.215, damaged when B>6.392, and dirty when B>2.482; the surface of the denitrification catalyst is considered cracked when C>0.094, damaged when C>0.726, and dirty when C>0.923.

[0034] Example 2: As shown in Figure 2, this example also provides a machine vision-based denitrification catalyst appearance defect detection system, including a denitrification catalyst surface image acquisition module 1, a denitrification catalyst surface image preprocessing module 2, and a denitrification catalyst defect type classification module 3; the denitrification catalyst surface image acquisition module 1 captures images of the denitrification catalyst surface using an industrial camera; the industrial camera is located at the outlet of the tunnel calcining kiln; wherein, the industrial camera is a CMOS area array camera capable of high-speed imaging; a parallel light source is set at the outlet of the tunnel calcining kiln, and the parallel light source is installed obliquely to irradiate the denitrification catalyst.

[0035] The preprocessing module 2 for the denitrification catalyst surface image segmentation performs line fitting and feature extraction on the segmented image. The denitrification catalyst surface image includes images of the denitrification catalyst, conveyor shaft, mechanical frame, and asbestos rope. The conveyor shaft, mechanical frame, and asbestos rope images are designated as background images. Segmentation of the denitrification catalyst surface image involves dividing the denitrification catalyst image and the background image. A k-means clustering algorithm is used for image segmentation, clustering pixels in the denitrification catalyst surface image, selecting a representative pixel for each cluster, and assimilating the pixels of the denitrification catalyst surface image into the representative pixels of the corresponding clusters. Gradient descent is used to refine the edges of the segmented denitrification catalyst surface image. Linear fitting was performed to correct the smoothness of the edge contours of the denitrification catalyst surface image. The surface image of the denitrification catalyst after linear fitting was then labeled using a rectangular labeling box. Each denitrification catalyst was selected and labeled with a tag number to distinguish catalyst monomers of different lengths. The corresponding relationships are as follows: tag number 0 corresponds to 50mm; tag number 1 corresponds to 55mm; tag number 2 corresponds to 60mm; tag number 3 corresponds to 70mm; tag number 4 corresponds to 80mm; tag number 5 corresponds to 90mm; tag number 6 corresponds to 100mm; tag number 7 corresponds to 110mm; and tag number 8 corresponds to 120mm. After image labeling, the pixel coordinate data labels of the denitrification catalyst surface image were exported. The exported data labels were then normalized using the following mathematical expression:

[0036] Where (x,y) represents the center coordinates of the rectangular label box, TL and BL represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, TR and BR represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, and len and wid represent the length and width of the rectangular label box, respectively; a YOLOv5 model is constructed to process the normalized data labels and extract image features.

[0037] Module 3, which categorizes denitrification catalyst defects, pre-detects the presence of defects in the pre-processed surface image of the denitrification catalyst. Based on the grayscale values ​​of the defect-free surface image, it performs refinement operations to obtain target defects. It calculates the minimum bounding rectangle for each target defect, obtaining the set features of the minimum bounding rectangle. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, it categorizes the denitrification catalyst defects into types. Denitrification catalyst defect types include cracks, damage, and contamination. Pre-detection of defects in the pre-processed surface image of the denitrification catalyst includes binarization processing, calculating a pixel threshold T0, and dividing the surface image into defective and defect-free parts based on the pixel threshold. Images with pixel values ​​greater than T0 are considered defective parts; images with pixel values ​​less than T0 are considered defect-free parts. Calculating the pixel threshold T0 includes calculating extreme value normalization and calculating the random perturbation coefficient. The mathematical expression for calculating extreme value normalization is:

[0038] Where S represents the normalized extreme value. and Let g represent the maximum and minimum grayscale values ​​of the target image g, where the target image is an image of the surface of the denitrification catalyst; the mathematical expression for calculating the random perturbation coefficient is:

[0039] Where K is the random disturbance coefficient. The pixel threshold T0 represents the grayscale value of a random pixel in the target image g; the mathematical expression for T0 is: After the target image is segmented for the first time based on polarization processing, it is divided into two parts, f and p, where f is the defective part and p is the defect-free part. The target defects are obtained by refining the gray values ​​of the surface images of the denitrification catalyst with and without defects. This includes updating the pixel threshold based on the gray values ​​of the two parts. The target defects are obtained after n segmentations of the target image. Updating the pixel threshold includes calculating a dynamic correction coefficient and iteratively updating the threshold. The mathematical expression for the dynamic correction coefficient is:

[0040] in, This is a dynamic correction factor. and These represent the maximum and minimum grayscale values ​​of the segment f, respectively. and Let p represent the maximum and minimum grayscale values ​​of the segmented part p, respectively, and n represent the number of iterations; the mathematical expression for threshold iteration update is:

[0041] in, Let be the pixel threshold after n iterations. After n iterations of the target image g, the target defect is obtained. At this point, the target defect is represented by discrete pixels. The defect features are defined by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle of the discrete pixels. Assuming the pixel coordinates of the four points of the minimum bounding rectangle are P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4), the mathematical expression for the length of the minimum bounding rectangle is: The mathematical expression for the minimum bounding rectangle width is: The mathematical expression for the area of ​​the minimum circumscribed rectangle is: The mathematical expression for the minimum circumscribed aspect ratio is: The mathematical expression for roundness is: .

[0042] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0043] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0044] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0045] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0046] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0047] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0048] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0049] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0050] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for detecting appearance defects in denitrification catalysts based on machine vision, characterized in that, The steps include: Step S1: Acquiring images of the surface of the denitrification catalyst by taking pictures of the denitrification catalyst with an industrial camera; Step S2: Preprocessing the surface images of the denitrification catalyst by segmenting the surface images of the denitrification catalyst and performing linear fitting and feature extraction on the segmented surface images of the denitrification catalyst. Step S3: The step of classifying the defect types of the denitrification catalyst. The surface image of the pre-processed denitrification catalyst is pre-detected for defects. Based on the gray value of the surface image of the denitrification catalyst with and without defects, the target defects are obtained by fine-tuning. The minimum bounding rectangle of the target defects is calculated, and the set features of the minimum bounding rectangle are obtained. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, the denitrification catalyst defect types are classified.

2. The method for detecting appearance defects in denitrification catalysts based on machine vision according to claim 1, characterized in that, The industrial camera is installed at the outlet of the tunnel calcining kiln, and a parallel light source is installed at the outlet of the tunnel calcining kiln. The parallel light source is installed at an angle to irradiate the denitrification catalyst.

3. The method for detecting appearance defects in denitrification catalysts based on machine vision according to claim 1, characterized in that, The surface image of the denitrification catalyst includes an image of the denitrification catalyst, an image of the conveyor shaft, an image of the mechanical frame, and an image of the asbestos rope; the images of the conveyor shaft, the mechanical frame, and the asbestos rope are designated as the background image; segmenting the surface image of the denitrification catalyst includes segmenting the denitrification catalyst image and the background image.

4. The method for detecting appearance defects in denitrification catalysts based on machine vision according to claim 3, characterized in that, The k-means clustering algorithm is used to segment the surface image of the denitrification catalyst. The pixels in the surface image of the denitrification catalyst are clustered, and a representative pixel is selected for each class. The pixels of the surface image of the denitrification catalyst are assimilated into the representative pixel of the corresponding class.

5. The method for detecting appearance defects in denitrification catalysts based on machine vision according to claim 4, characterized in that, The gradient descent method was used to fit straight lines to the edges of the segmented denitrification catalyst surface images, correcting the smoothness of the edge contours. The surface images of the denitrification catalysts after straight line fitting were then labeled. A rectangular label box was used to select each denitrification catalyst unit, and labels were used to distinguish different lengths of the catalyst monomers. The corresponding relationships are as follows: label number 0 corresponds to 50mm; label number 1 corresponds to 55mm; label number 2 corresponds to 60mm; label number 3 corresponds to 70mm; label number 4 corresponds to 80mm; label number 5 corresponds to 90mm; label number 6 corresponds to 100mm; label number 7 corresponds to 110mm; and label number 8 corresponds to 120mm. After image annotation, the pixel coordinate data labels of the denitrification catalyst surface image are exported. The exported data labels are then normalized, and the mathematical expression is as follows: Where (x,y) represents the center coordinates of the rectangular label box, TL and BL represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, TR and BR represent the pixel coordinates of the top left and bottom left of the rectangular label box, respectively, and len and wid represent the length and width of the rectangular label box, respectively; a YOLOv5 model is constructed to process the normalized data labels and extract image features.

6. The method for detecting appearance defects in denitrification catalysts based on machine vision according to claim 1, characterized in that, Defect types in denitrification catalysts include cracks, breakage, and contamination. Pre-detection of defects in the pre-treated catalyst surface image includes binarization, calculation of a pixel threshold T0, and division of the catalyst surface image into defective and defect-free portions based on the pixel threshold. Surface images with pixel values ​​greater than T0 are considered defective, while those with pixel values ​​less than T0 are considered defect-free. Calculating the pixel threshold T0 includes calculating extremum normalization and random perturbation coefficients. The mathematical expression for calculating extremum normalization is: Where S represents the normalized extreme value. and Let g represent the maximum and minimum grayscale values ​​of the target image g, where the target image is an image of the surface of the denitrification catalyst; the mathematical expression for calculating the random perturbation coefficient is: Where K is the random disturbance coefficient. The pixel threshold T0 represents the grayscale value of a random pixel in the target image g; the mathematical expression for T0 is: 。 7. The method for detecting appearance defects of denitrification catalysts based on machine vision according to claim 6, wherein after the target image is segmented for the first time based on polarization processing, the target image is divided into two parts, f and p, wherein, f represents the defective portion, and p represents the defect-free portion. The target defect is obtained by refining the grayscale values ​​of the surface image of the denitrification catalyst with and without defects. This includes updating the pixel threshold based on the grayscale values ​​of both portions, and segmenting the target image n times. Updating the pixel threshold involves calculating a dynamic correction coefficient and iteratively updating the threshold. The mathematical expression for the dynamic correction coefficient is: in, For dynamic correction coefficients, and These represent the maximum and minimum grayscale values ​​of the segment f, respectively. and Let p represent the maximum and minimum grayscale values ​​of the segmented part p, respectively, and n represent the number of iterations; the mathematical expression for threshold iteration update is: in, This is the pixel threshold after n iterations.

8. According to claim 7, a method for detecting appearance defects of denitrification catalysts based on machine vision is used to obtain target defects after n iterations of the target image g. At this time, the target defects are represented as discrete pixels. The defect features are defined by calculating the length, width, area, aspect ratio, and circularity of the minimum bounding rectangle of the discrete pixels. Assuming the pixel coordinates of the four points of the minimum bounding rectangle are P1(x1,y1), P2(x2,y2), P3(x3,y3), and P4(x4,y4), the mathematical expression for the length of the minimum bounding rectangle is: The mathematical expression for the minimum bounding rectangle width is: The mathematical expression for the area of ​​the minimum circumscribed rectangle is: The mathematical expression for the minimum circumscribed aspect ratio is: The mathematical expression for the roundness is: 。 9. A machine vision-based system for detecting appearance defects in denitrification catalysts, characterized in that, The system includes a module for acquiring images of the surface of the denitrification catalyst, a module for preprocessing images of the surface of the denitrification catalyst, and a module for classifying the defect types of the denitrification catalyst. The module for acquiring images of the surface of the denitrification catalyst uses an industrial camera to capture images of the surface of the denitrification catalyst. The module for preprocessing images of the surface of the denitrification catalyst segments the surface images of the denitrification catalyst and performs linear fitting and feature extraction on the segmented surface images of the denitrification catalyst. The module for classifying the defect types of denitrification catalysts pre-detects whether there are defects in the pre-processed surface image of the denitrification catalyst. Based on the grayscale values ​​of the surface image of the denitrification catalyst with and without defects, a fine-grained operation is performed to obtain the target defects. The minimum bounding rectangle of the target defects is calculated to obtain the set features of the minimum bounding rectangle. Based on the set features of the minimum bounding rectangle and the correspondence table of defect features, the denitrification catalyst defect types are classified.

10. The machine vision-based denitrification catalyst appearance defect detection system according to claim 9, characterized in that, The industrial camera is installed at the outlet of the tunnel-type calcining kiln; the industrial camera is a CMOS area array camera capable of high-speed imaging; a parallel light source is installed at the outlet of the tunnel-type calcining kiln, and the parallel light source is installed obliquely to irradiate the denitrification catalyst; the surface image of the denitrification catalyst includes an image of the denitrification catalyst, an image of the conveyor shaft, an image of the mechanical frame, and an image of the asbestos rope; the image of the conveyor shaft, the image of the mechanical frame, and the image of the asbestos rope are designated as the background image; segmenting the surface image of the denitrification catalyst includes segmenting the denitrification catalyst image and the background image.

Images