An intelligent detection method for blockage of honeycomb catalyst module

Abstract

The present invention relates to the field of image data processing, and specifically to a method for intelligent detection of blockage of a honeycomb catalyst module, comprising: obtaining an edge image of a module; inputting any edge in the edge image into a trained neural network model, and outputting the confidence that the corresponding channel is blocked; calculating the degree of similarity between any edge and other edges; calculating the degree of non-uniformity between any edge and multiple adjacent edges; and then calculating the blockage index of each edge; connecting suspected blockage edges that are less than a distance threshold and performing convex hull detection to obtain a suspected blockage area, wherein the blockage index of the suspected blockage edge is greater than the blockage threshold; if the total number of pixels in any suspected blockage area is greater than the number threshold, then the suspected blockage area is judged to be a blockage area. The present invention improves the accuracy of blockage detection.

Description

Technical Field

[0001] The present invention relates to the field of image data processing. The present invention relates to an intelligent detection method for blockage of a honeycomb catalyst module. Background Art

[0002] The honeycomb catalyst module is a catalyst carrier with a honeycomb structure, which is mainly used in the fields of industrial waste gas treatment, etc. It can promote the conversion and removal of harmful substances, improve the efficiency of waste gas treatment, and reduce environmental pollution. For example, coal-fired power plants are one of the main sources of nitrogen oxide emissions. Honeycomb catalysts use selective catalytic reduction technology to inject ammonia or urea as a reducing agent into the catalyst, which reacts with nitrogen oxides in the waste gas to generate nitrogen and water vapor, effectively reducing the emission concentration of nitrogen oxides.

[0003] During the use of honeycomb catalysts, blockage may occur. For example, in the chemical production process, the gas often contains impurities or particles, which are easy to deposit in the catalyst channel and cause blockage; fly ash in the flue gas will also cover the surface of the catalyst monomer, causing blockage; in the low-temperature SCR reaction, With reactants ( and ) will react to form ammonium sulfate (such as , ) and other sulfates, which will be adsorbed on the active sites of the catalyst, aggravating the deactivation and clogging of the catalyst.

[0004] When the catalyst is clogged, the flue gas in the clogged area can only pass through the unblocked catalyst area, which speeds up the flow of the flue gas. The wear of the catalyst near the clogged area is several times the normal wear rate. This eventually leads to a large amount of damage and collapse of the local catalyst. Therefore, it is of great significance to detect whether the catalyst module is clogged.

[0005] The patent document with the authorization announcement number CN115311291B is named as a cleaning method for hole blockage used in hole processing, which includes: obtaining a grayscale image of the workpiece surface, performing edge detection on the image to obtain an edge image, and then determining the degree of blockage in the hole area based on the edge image. When detecting whether the honeycomb catalyst module is blocked, it can also be determined based on the edge detection algorithm. Specifically, it can be determined whether the catalyst module is blocked by comparing the edge to be detected with the standard edge (the edge corresponding to the unblocked area). However, due to the influence of factors such as lighting, the edge image of the unblocked honeycomb catalyst module includes non-module edges such as shadow edges, which leads to poor accuracy in detecting whether the honeycomb catalyst module is blocked.

[0006] Based on this, it is particularly important to improve the accuracy of detecting whether the honeycomb catalyst module is blocked. Summary of the invention

[0007] In order to solve the above-mentioned technical problem of poor accuracy in detecting whether a honeycomb catalyst module is blocked, the present invention proposes an intelligent detection method for blocking a honeycomb catalyst module.

[0008] A method for intelligently detecting blockage of a honeycomb catalyst module, comprising:

[0009] Acquire an edge image of the honeycomb-type catalyst module to be detected; input any edge in the edge image into a trained neural network model, and output the confidence level that the channel corresponding to the edge is blocked;

[0010] Calculating the similarity between any one edge and other edges in the edge image; calculating the non-uniformity between any one edge and multiple adjacent edges;

[0011] The calculation formula of the blocking index of each edge is:

[0012] ,in, is the clogging index of edge i in the edge image, is the similarity corresponding to the edge i, is the non-uniformity degree corresponding to the edge i, is the confidence corresponding to the edge i, is the positive correlation normalization function;

[0013] The suspected blocked edges that are less than the distance threshold are connected and convex hull detection is performed to obtain the suspected blocked area of ​​the honeycomb catalyst module to be detected, wherein the blocking index of the suspected blocked edge is greater than the blocking threshold; if the total number of pixels in any suspected blocked area is greater than the number threshold, then the any suspected blocked area is judged to be a blocked area.

[0014] The present invention Correction for weight , so that the blockage index of edge i can better reflect the possibility of blockage in the channel corresponding to edge i. Since the blockage area is large and has a certain aggregation, the above method can improve the accuracy of catalyst module blockage detection.

[0015] Optionally, the similarity is calculated as:

[0016] ;

[0017] in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i, is the total number of pixels contained in the convex hull of edge i, is the edge in the edge image except the edge i The total number of pixels, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull of is the central abscissa of edge i, is the center ordinate of edge i, For the edge The central abscissa of For the edge The center vertical coordinate of .

[0018] The larger it is, the greater the possibility that edge i is similar to its adjacent edges.

[0019] Optionally, the similarity is calculated as:

[0020] ;

[0021] in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i, is the total number of pixels contained in the convex hull of edge i, is the edge in the edge image except the edge i The total number of pixels, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull.

[0022] The larger it is, the more similar edge i is to other edges, which means that the channel corresponding to edge i is less likely to be blocked.

[0023] Optionally, the degree of non-uniformity is calculated as:

[0024] ;

[0025] in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i, is the total number of pixels in the first neighborhood of pixel j on edge i, is the total number of edge centers within the second neighborhood of pixel point k within the first neighborhood of pixel point j, is an absolute value.

[0026] The larger it is, the less uniform the distribution of edge i and its adjacent edges is.

[0027] Optionally, the degree of non-uniformity is calculated as:

[0028] ;

[0029] in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i, is the total number of edge centers within the second neighborhood of pixel j of edge i.

[0030] The larger it is, the less uniform the distribution of edge i and its adjacent edges is, which means that the channel corresponding to edge i is more likely to be blocked.

[0031] Optionally, the training process of the neural network model includes:

[0032] Input the training set into the neural network model to obtain a trained neural network model;

[0033] The training set includes multiple samples, each sample includes an edge and a label corresponding to the edge, and the label is used to indicate whether the channel corresponding to the edge is blocked.

[0034] Optionally, the neural network model is a ResNet neural network model.

[0035] Optionally, the loss function of the neural network model is a cross entropy loss function.

[0036] Beneficial effects of the present invention:

[0037] The greater the similarity of the edge to be detected, the less likely the channel corresponding to the edge is to be blocked. The greater the non-uniformity of the edge to be detected, the more likely the channel corresponding to the edge is to be blocked. The weighted calculation of the blockage index of edge i to be detected makes the calculated blockage index more able to reflect the possibility of blockage in the channel corresponding to edge i. The blocked channel has a certain aggregation, and the area of ​​the blocked area is large. The above method can more accurately determine the blocked area and whether the catalyst module is blocked, thereby improving the accuracy of the catalyst module blockage detection.

[0038] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0040] Figure 1 is a flow chart schematically showing a method for intelligently detecting blockage of a honeycomb catalyst module according to an embodiment of the present invention;

[0041] Figure 2 is a schematic diagram schematically illustrating shadows caused by illumination according to an embodiment of the present invention;

[0042] Figure 3 is a schematic diagram showing an embodiment of the present invention Figure 2 Schematic diagram of the corresponding edge image. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0044] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0045] Due to the influence of factors such as illumination, the edge image of the unblocked honeycomb catalyst module includes the edge corresponding to the shadow of the catalyst module edge, which leads to poor accuracy in detecting whether the honeycomb catalyst module is blocked. Figure 2 As shown, Figure 2 FIG. 2 is a schematic diagram schematically showing shadows caused by illumination according to an embodiment of the present invention. Figure 3 As shown, Figure 3 is a schematic diagram showing an embodiment of the present invention Figure 2Schematic diagram of the corresponding edge image.

[0046] exist Figure 3 In the example, multiple adjacent edges are relatively similar, and the distribution of multiple adjacent edges is relatively uniform.

[0047] In the edge image of the honeycomb catalyst module without blockage and module edge shadow caused by light, since the pipeline of the honeycomb catalyst has a certain thickness, when the edge detection algorithm is used to obtain the edge image, the lines on both sides of the pipeline correspond to two edges, that is, each channel corresponds to an edge similar to a hexagon. Multiple adjacent edges are also relatively similar, and the distribution of multiple adjacent edges is also relatively uniform.

[0048] In the edge image of the clogged honeycomb catalyst module, the edge of the clogged area is messy and disordered, the shape difference between the edge of the clogged area and its adjacent edges is large and the similarity is low, and the distribution uniformity between the edge of the clogged area and its adjacent edges is low.

[0049] Based on this, the present invention further determines whether the honeycomb catalyst is blocked according to the similarity and uniformity of the adjacent multiple edges, so as to improve the accuracy of detecting whether the honeycomb catalyst is blocked.

[0050] Figure 1 FIG. 1 is a flow chart schematically showing a method for intelligently detecting blockage of a honeycomb catalyst module according to an embodiment of the present invention. Figure 1 The specific instructions are as follows:

[0051] S1: Acquire an edge image of the honeycomb catalyst module to be detected.

[0052] In one embodiment, an industrial camera is used to collect an image of the honeycomb catalyst module to be inspected. To facilitate subsequent processing, the collected image is grayed to obtain a grayscale image.

[0053] In one embodiment, in order to reduce the degree of interference of noise on the grayscale image, Gaussian filtering is used to denoise the grayscale image to obtain a denoised grayscale image.

[0054] In one embodiment, the Canny edge detection algorithm is used to extract the edge of the grayscale image to obtain an edge image.

[0055] S2: Train the neural network model.

[0056] The training process includes: inputting the training set into the neural network model to obtain a trained neural network model. The training set includes multiple samples, each sample includes an edge and a label corresponding to the edge, and the label is used to indicate whether the channel corresponding to the edge is blocked. The above labels are manually annotated, and the label of the edge corresponding to no blockage is 0, and the label of the edge corresponding to blockage is 1. It should be noted that the amount of data with different labels should be kept balanced.

[0057] In one embodiment, the acquisition process of the edge in the training set is as follows: in multiple batches of honeycomb catalysts, multiple edges of honeycomb catalyst models without blockage are selected, and multiple edges of blocked areas of honeycomb catalyst models with blockage are selected. The total number of edges in the training set can be 500.

[0058] It should be noted that in order to improve the accuracy of blockage recognition, the edges in the training set should include the edges of the catalyst modules in as many cases as possible.

[0059] In one embodiment, the neural network model is a ResNet neural network model, and the loss function of the neural network model is a cross entropy loss function.

[0060] In one embodiment, in response to the number of training times reaching a preset number of training times, the training is stopped. The number of training times may be 500. In another embodiment, when the loss is less than a preset loss threshold, the training is stopped.

[0061] Other technical details in the training process are common sense and will not be elaborated on.

[0062] At this point, a trained neural network model is obtained.

[0063] S3: Input each edge in the edge image into a trained neural network model, and output the confidence level of whether the channel corresponding to each edge is blocked.

[0064] The confidence value ranges from 0 to 1. The confidence is positively correlated with the possibility that the channel corresponding to the edge is blocked.

[0065] S4: Calculate the similarity between any edge and other edges in the edge image.

[0066] In one embodiment, the calculation formula for the similarity is:

[0067] ;

[0068] in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i (the calculation method of circularity is the prior art), is the total number of pixels contained in the convex hull of edge i (indicating the area of ​​the convex hull), is the edge in the edge image except the edge i (No. The total number of pixels of the edge, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull of is the central abscissa of edge i, is the center ordinate of edge i, For the edge The central abscissa of For the edge The center vertical coordinate of .

[0069] It should be noted that edge i is the i-th edge in the edge image.

[0070] is a positive correlation normalization function. In this article, the positive correlation normalization function can be a maximum and minimum normalization function of the positive correlation.

[0071] For example, The interpretation is: the edge i Denoted as distance A, the distance A of each edge is calculated, the maximum value of the maximum and minimum normalization function is the maximum value of all distances A, and the minimum value is the minimum value of all distances A. Other normalization processes in this article will not be repeated unless otherwise specified.

[0072] When the catalyst module without blockage is affected by light, the similarity of the adjacent multiple edges is high. For the catalyst module without blockage and not affected by light, the similarity of the adjacent multiple edges is high. For the blocked area of ​​the catalyst module, the adjacent multiple edges are disordered and the similarity of the adjacent multiple edges is low.

[0073] The smaller the edge i is, the closer it is to the edge The closer the distance; The larger the edge i is, the closer it is to the edge The greater the similarity.

[0074] The larger the value is, the more likely it is that edge i is similar to its adjacent edges, which means that the channel corresponding to edge i is less likely to be blocked.

[0075] In another embodiment, the calculation formula of the similarity is:

[0076] ;

[0077] in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i, is the total number of pixels contained in the convex hull of edge i, is the edge in the edge image except the edge i The total number of pixels, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull.

[0078] At this point, the similarity degree corresponding to each edge is obtained.

[0079] S5: Calculate the degree of non-uniformity between any edge and multiple adjacent edges.

[0080] In one embodiment, the calculation formula of the non-uniformity is:

[0081] ;

[0082] in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i; for pixel j (jth pixel) on edge i, the first neighborhood range is determined with pixel j as the center and radius R1, is the total number of pixels in the first neighborhood range corresponding to pixel j; for pixel k (kth pixel) in the first neighborhood range corresponding to pixel j, a second neighborhood range is determined with pixel k as the center and a radius of R2, is the total number of edge centers within the second neighborhood corresponding to the pixel point k ( represents the density of edges within the second neighborhood); is an absolute value.

[0083] Among them, those skilled in the art can set R1 and R2 by themselves. For example, R1 and R2 can both be 90. The unit length 1 represents the Euclidean distance between the coordinates of two adjacent pixel points.

[0084] In the above formula, j is used to represent the count of pixels on the edge, and k is used to represent the count of pixels within the first neighborhood.

[0085] When an unobstructed catalyst module is affected by light, the distribution of multiple adjacent edges is more uniform and the arrangement is more orderly; for a catalyst module that is not obstructed and not affected by light, the distribution of multiple adjacent edges is more uniform and the arrangement is more orderly; for the blocked area of ​​the catalyst module, the multiple adjacent edges are disordered, the distribution of multiple adjacent edges is less uniform and the arrangement is less orderly.

[0086] The larger it is, the greater the difference between the densities corresponding to the pixels in the first neighborhood of the pixel j is, and the greater the discreteness of the density is (the density corresponding to the pixel represents the total number of edge centers in the second neighborhood of the pixel j).

[0087] The larger it is, the less uniform the distribution of edge i and its adjacent edges and the less neat their arrangement, which means that the channel corresponding to edge i is more likely to be blocked.

[0088] In another embodiment, the calculation formula of the non-uniformity degree is:

[0089] ;

[0090] in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i, is the total number of edge centers within the second neighborhood of pixel point j on edge i.

[0091] So far, the non-uniformity degree corresponding to each edge is obtained.

[0092] S6: Calculate the blocking index of each edge.

[0093] In one embodiment, the calculation formula of the blocking index of each edge is:

[0094] ,in, is the clogging index of edge i in the edge image, is the similarity corresponding to the edge i, is the non-uniformity degree corresponding to the edge i, is the confidence corresponding to the edge i, is the positive correlation normalization function.

[0095] right Explanation: Select the similarity corresponding to each edge in the edge image, select the similarity corresponding to the edges of multiple blocked areas, select the similarity corresponding to the edges of multiple normal areas, and the maximum and minimum values ​​in the maximum and minimum normalization functions are the maximum and minimum values ​​of the similarity selected above. The purpose of selecting the similarity corresponding to the edges of multiple blocked areas is to make The accuracy of indicating that the channel corresponding to edge i is blocked is higher. The explanation is the same and I will not repeat it here.

[0096] The larger it is, the greater the possibility that the channel corresponding to edge i is blocked. The larger it is, the greater the possibility that the channel corresponding to edge i is blocked. The larger it is, the greater the possibility that the channel corresponding to edge i is blocked. It characterizes the possibility of blockage in the channel corresponding to edge i.

[0097] by Calculate the weight , which improves the accuracy of catalyst module blockage detection.

[0098] At this point, the blocking index of each edge is obtained.

[0099] S7: Determine whether the honeycomb catalyst module to be detected is blocked.

[0100] The edge with a blockage index greater than the blockage threshold in the edge image is taken as a suspected blockage edge, the suspected blockage edges less than the distance threshold are connected, and the convex hull detection is performed on the connected suspected blockage edges to obtain the suspected blockage area (i.e., convex hull area) of the honeycomb catalyst module to be detected. If the total number of pixels in any suspected blockage area is greater than the number threshold, then the any suspected blockage area is judged to be a blockage area, and it is judged that the honeycomb catalyst module to be detected is blocked. The remaining suspected blockage areas are traversed to determine whether the remaining suspected blockage areas are blockage areas.

[0101] Those skilled in the art can set the distance threshold, the congestion threshold and the quantity threshold by themselves. For example, the distance threshold can be 30, the congestion threshold can be 0.75, and the quantity threshold can be 20.

[0102] If the honeycomb-type catalyst module is detected to be clogged, relevant personnel can be reminded to clean or replace the clogged catalyst module.

[0103] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0104] Additionally, the articles "a" and "an" as used in this disclosure and the appended claims are generally construed to mean "one or more" unless specified otherwise or clear from context to be directed to a singular form.

[0105] Where the terms "includes," "having," "having," "having," or variations thereof are used in the detailed description or claims, such terms are intended to be inclusive in a manner similar to the term "comprising."

[0106] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations appearing in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as S1, S2, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that if descriptions such as "first", "second", "A1", "B1" appear in this article, their role is to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, do not have any technical meaning, and do not limit "A1", "B1", "first" and "second" to different types. For example: the first feature of an image is the feature of the image. The role of adding "first" includes facilitating subsequent references to different objects, that is, facilitating indexing.

[0107] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for intelligently detecting blockage of a honeycomb catalyst module, characterized in that: include: Acquire an edge image of a honeycomb-type catalyst module to be inspected; Input any edge in the edge image into the trained neural network model, and output the confidence that the channel corresponding to the edge is blocked; Calculating the similarity between any one edge and other edges in the edge image; calculating the non-uniformity between any one edge and multiple adjacent edges; The formula for calculating the similarity is: ;in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i, is the total number of pixels contained in the convex hull of edge i, is the edge in the edge image except the edge i The total number of pixels, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull; The calculation formula for the degree of non-uniformity is: ;in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i, is the total number of edge centers within the second neighborhood of pixel point j of edge i; The calculation formula of the blocking index of each edge is: ,in, is the clogging index of edge i in the edge image, is the similarity corresponding to the edge i, is the non-uniformity degree corresponding to the edge i, is the confidence corresponding to the edge i, is the positive correlation normalization function; Connect the suspected blocked edges that are less than the distance threshold and perform convex hull detection to obtain the suspected blocked area of ​​the honeycomb catalyst module to be detected, wherein the blocking index of the suspected blocked edge is greater than the blocking threshold; if the total number of pixels in any suspected blocked area is greater than the number threshold, then the suspected blocked area is judged to be a blocked area; Among them, the neural network model uses the ResNet neural network model.

2. The method for intelligently detecting blockage of a honeycomb catalyst module according to claim 1, characterized in that: Or the similarity calculation formula is: ； in, is the similarity corresponding to edge i, M is the total number of edges in the edge image, represents the product, is the total number of pixels on edge i, is the circularity of the convex hull contour of edge i, is the total number of pixels contained in the convex hull of edge i, is the edge in the edge image except the edge i The total number of pixels, For the edge The circularity of the convex hull contour, For the edge The total number of pixels contained in the convex hull of is the central abscissa of edge i, is the center ordinate of edge i, For the edge The central abscissa of For the edge The center vertical coordinate of .

3. The method for intelligently detecting blockage of a honeycomb catalyst module according to claim 1, characterized in that: Or the calculation formula of non-uniformity is: ； in, is the degree of non-uniformity corresponding to edge i, is the total number of pixels on edge i, is the total number of pixels in the first neighborhood of pixel j on edge i, is the total number of edge centers within the second neighborhood of pixel point k within the first neighborhood of pixel point j, is an absolute value.

4. The method for intelligently detecting blockage of a honeycomb catalyst module according to claim 1, characterized in that: The training process of the neural network model includes: Input the training set into the neural network model to obtain a trained neural network model; The training set includes multiple samples, each sample includes an edge and a label corresponding to the edge, and the label is used to indicate whether the channel corresponding to the edge is blocked.

5. The method for intelligently detecting blockage of a honeycomb catalyst module according to claim 1, characterized in that: The loss function of the neural network model is a cross entropy loss function.

Images