Petroleum product copper sheet corrosion online detection method and system based on computer vision

Abstract

The invention relates to the technical field of copper corrosion detection, and discloses a petroleum product copper sheet corrosion online detection method and system based on computer vision, and the method comprises the steps: constructing a macroscopic data collection point of a test copper sheet, and constructing a microscopic data collection point of the test copper sheet; performing data preprocessing on the historical microscopic data and the current microscopic data, establishing a concentration change model of the microscopic data acquisition points, and analyzing the corrosion diffusion degree corresponding to the current data through the concentration change model; selecting corrosion abnormal points from the macroscopic data acquisition points; performing equal-interval processing on the historical macroscopic data to generate a gray model of the equal-length historical data, performing equal-interval processing on the current macroscopic data, and outputting corrosion severity corresponding to the equal-length current data through the gray model; and selecting an image acquisition point from the abnormal corrosion points, performing image acquisition on the image acquisition point, and performing corrosion detection on the image acquisition point by using the acquired image. According to the invention, the data volume during copper sheet corrosion detection can be reduced.

Description

Technical Field

[0001] The invention relates to an online detection method and system for copper sheet corrosion of petroleum products based on computer vision, belonging to the technical field of copper corrosion detection. Background Art

[0002] In the quality control process of petroleum products, copper strip corrosion test is an important indicator, which can reflect the corrosiveness of metallic copper in petroleum products. The traditional copper strip corrosion detection method mainly relies on manual visual inspection and standard color plate comparison and judgment, which has the problems of strong subjectivity, low efficiency and low precision. Therefore, the development of a device that can automatically detect the degree of copper strip corrosion online is of great significance to improving the detection efficiency and quality control level of petroleum products.

[0003] At present, when the copper sheet is large in size and area, the image recognition technology cannot completely capture all areas of the copper sheet for detection. The primary reason is that the area of ​​the copper sheet recognized by the image recognition technology at one time cannot completely cover the total area occupied by the copper sheet. That is to say, it is necessary to use the image recognition technology to scan the copper sheet area multiple times and at multiple positions before the image data of the copper sheet surface can be collected. For example, when the area occupied by the copper sheet is 20m*20m and the image recognition technology recognizes the copper sheet area of ​​5m*5m at a time, it is necessary to use the image collector to scan the copper sheet surface 4 times before the complete image of the copper sheet can be collected. Secondly, when collecting the copper sheet image, it is also necessary to collect copper sheet samples from the complete copper sheet. To collect copper sheet samples, it is necessary to capture the copper sheet image samples whose colors are not affected when the copper sheet is not contaminated with oil. However, such opportunities are rare. This is because when the copper sheet is in use (at this time, the copper sheet is in contact with oil), it is very difficult to obtain a copper sheet that is not contaminated by oil, especially when all areas of the copper sheet are not contaminated by oil to capture images of the copper sheet in all areas, which makes corrosion detection of the copper sheet area more difficult. For example, when the area occupied by the copper sheet is 20m*20m and the area of ​​the copper sheet that is not contaminated by oil is 4m*4m, only the 4m*4m area can be imaged, while the image data of other areas contaminated with oil need to wait until the oil disappears before image data can be collected. In addition, since it is not possible to determine which areas need to be focused on and which areas need to be ignored, it is still necessary to collect images of the entire copper sheet area, which results in a long time period, a huge amount of image data that needs to be collected and processed, and low efficiency of copper sheet corrosion detection.

[0004] Therefore, the amount of copper sheet image data during copper sheet corrosion detection is large. Summary of the invention

[0005] The present invention provides a method and system for online detection of copper sheet corrosion of petroleum products based on computer vision, the main purpose of which is to reduce the amount of data during copper sheet corrosion detection.

[0006] To achieve the above object, the present invention provides an online detection method for copper sheet corrosion of petroleum products based on computer vision, comprising:

[0007] Constructing macro data collection points of the test copper sheet, and constructing micro data collection points of the test copper sheet, and collecting historical micro data and current micro data at the micro data collection points;

[0008] Preprocessing the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establishing a concentration change model of the micro data collection point according to the pre-historical data, and analyzing the corrosion diffusion degree corresponding to the pre-current data through the concentration change model;

[0009] According to the micro data collection point and the corrosion diffusion degree, a corrosion abnormal point is selected from the macro data collection point, and historical macro data and current macro data are collected at the corrosion abnormal point;

[0010] Performing equal-interval processing on the historical macro data to obtain equal-length historical data, generating a grey model for the equal-length historical data, performing equal-interval processing on the current macro data to obtain equal-length current data, and outputting the corrosion severity corresponding to the equal-length current data through the grey model;

[0011] Based on the severity of the corrosion, an image acquisition point is selected from the corrosion abnormal point, image acquisition is performed at the image acquisition point to obtain an acquired image, a training image corresponding to the acquired image is obtained, image preprocessing is performed on the training image to obtain a preprocessed image, and based on the preprocessed image, a trained encoder and a trained decoder of the acquired image are generated, and corrosion detection is performed on the image acquisition point using the acquired image according to the trained encoder and the trained decoder to obtain a corrosion detection result of the test copper sheet.

[0012] Optionally, the macro data collection points of the test copper sheet are constructed, including:

[0013] Acquire the macro data type of the test copper sheet;

[0014] Based on the macro data type, a macro data acquisition device of the test copper sheet is set;

[0015] Querying the number of ranges of the collection range of the macro data collection device contained in the range corresponding to the test copper sheet;

[0016] Determining the number of devices of the macro data acquisition device using the number of ranges;

[0017] Generating random positions of the macro data acquisition devices within the range corresponding to the test copper sheet according to the number of the devices;

[0018] When the coverage area of ​​the macro data acquisition device at the random position is not less than the range corresponding to the test copper sheet, the random position is used as the macro data acquisition point of the test copper sheet.

[0019] Optionally, establishing a concentration change model of the microscopic data collection point according to the pre-historical data includes:

[0020] Query the microscopic concentration and copper flake concentration of the microscopic data collection point from the pre-historical data;

[0021] Comparing the copper sheet concentration with the microscopic concentration to obtain a concentration comparison result;

[0022] Determining the diffusion parameters of the microscopic data collection points using the concentration comparison results;

[0023] According to the pre-historical data and the diffusion parameters, the concentration change model of the microscopic data collection point is established using the following formula:

[0024]

[0025] in, represents the concentration change model, s(z,t) represents the concentration at point z at time t, Indicates the neighborhood of point z in the pre-historical data The concentration of the point, Represents the z point and The diffusion parameter between points, Indicates the historical data The volume of a point, D z Represents the neighborhood of point z.

[0026] Optionally, analyzing the corrosion diffusion degree corresponding to the pre-current data by using the concentration change model includes:

[0027] Analyzing the current concentration value corresponding to the pre-current data using the concentration change model;

[0028] According to the current concentration value, the corrosion diffusion degree corresponding to the pre-current data is calculated using the following formula:

[0029]

[0030] Among them, q(z′,t) represents the degree of corrosion diffusion, s(z′,t) represents the current concentration value, and s 1 Indicates the standard concentration of the test copper sheet, s 2Indicates the standard concentration of petroleum.

[0031] Optionally, selecting the corrosion abnormality point from the macro data collection point according to the micro data collection point and the corrosion diffusion degree includes:

[0032] Determining whether the corrosion diffusion degree is an abnormal diffusion degree;

[0033] When the corrosion diffusion degree is an abnormal diffusion degree, querying a first target collection point corresponding to the corrosion diffusion degree in the micro data collection point;

[0034] Identifying a second target collection point among the macro data collection points that covers the first target collection point;

[0035] The second target acquisition point is taken as a corrosion abnormal point.

[0036] Optionally, the performing equal-interval processing on the historical macro data to obtain equal-length historical data includes:

[0037] Extract the macro data moments of the historical macro data; sort the macro data moments to obtain a data moment sequence; and calculate the standard time interval of the data moment sequence using the following formula:

[0038]

[0039] Wherein, Δt represents the standard time interval, m represents the length of the data time sequence, and t m represents the maximum time in the data time sequence, t 1 represents the minimum time in the data time sequence;

[0040] According to the standard time interval, the macro data sequence corresponding to the data time sequence is processed at equal intervals using the following formula to obtain equal-length historical data:

[0041]

[0042] Among them, y(t j ) represents the same length of historical data, x(t j ) indicates that the macro data sequence is at t j The historical macro data at the time, j represents the sequence number of the time in the data time series, t j represents the jth moment in the data time sequence, x(t j -1) represents x(t j ) is the predecessor in the macro data series, and Δt represents the standard time interval.

[0043] Optionally, the generating of the grey model of the equal-length historical data includes:

[0044] The same length historical data is accumulated using the following formula to obtain accumulated historical data:

[0045]

[0046] Among them, y (1) (k) represents the accumulated historical data, y (0) (i) represents the i-th data in the same length historical data, n represents the number of values ​​of the same type in the same length historical data, and k represents the sequence number less than n;

[0047] According to the accumulated historical data, the grey model of the equal-length historical data is constructed using the following formula:

[0048]

[0049] Y n =BU

[0050]

[0051] in, represents the grey model, a and u in U represent unknown parameters, U represents the matrix composed of unknown parameters, y (0) (2) y (0) (3) y (0) (n) represents y (0) (i) When taking different values ​​of i, Y n Indicated by y (0) (2) y (0) (3) y (0) (n) constitutes a matrix of equal length historical data, y (1) (1) y (1) (2) y (1) (3) y (1) (n-1), y (1) (n) represents y (1) (k) takes different values ​​when k is used, B represents the matrix composed of accumulated historical data, U represents the calculation result of U, a represents the calculation result of a, u represents the calculation result of u, Indicates the value of the accumulated historical data in the next period, y (1) (t) represents any type of data in the accumulated historical data, and t represents the time.

[0052] Optionally, outputting the corrosion severity corresponding to the equal-length current data through the grey model includes:

[0053] According to the grey model, calculating the planned current data of the equal-length historical data corresponding to the grey model;

[0054] Query the historical corrosion degree corresponding to the historical data of equal length;

[0055] Constructing a grey analysis model of the historical corrosion degree;

[0056] Determine whether the planned current data is consistent with the equal-length current data;

[0057] When the planned current data is consistent with the equal-length current data, outputting the corrosion severity corresponding to the equal-length current data using the grey analysis model;

[0058] When the planned current data is inconsistent with the equal-length current data, the severity of the copper sheet corrosion is used as the corrosion severity corresponding to the equal-length current data.

[0059] Optionally, generating a trained encoder and a trained decoder of the acquired image based on the preprocessed image includes:

[0060] Dividing the preprocessed image into multi-view sample anchors, positive samples and negative samples;

[0061] In the preset encoder, the multi-view sample anchor point, the positive sample and the negative sample are respectively mapped to the feature latent space using the following formulas to obtain the anchor point feature, the positive sample feature and the negative sample feature:

[0062] f a =Φ(s a ),f p =Φ(s p ),f n =Φ(s n )

[0063] Among them, f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s a represents the multi-view sample anchor, Φ represents the encoder, s p represents a positive sample, s n represents negative samples;

[0064] The first loss value corresponding to the anchor point feature, the positive sample feature and the negative sample feature is calculated using the following formula:

[0065]

[0066] L 2 =||Rec(f a )-s a ||

[0067] Among them, L1 Indicates the loss value between the anchor point feature and the positive sample feature and the negative sample feature in the first loss value, L 2 represents the loss value between the anchor feature and the multi-view sample anchor in the first loss value, f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s a Represents the multi-view sample anchor point, Rec represents the reconstruction model, and Rec is used to transform f a Reconstruct the original image and ensure that the latent features extracted by the encoder Φ are complete;

[0068] Obtain a feature set corresponding to the anchor feature, the positive sample feature, and the negative sample feature; in a preset decoder, use the following formula to perform feature decoding on the feature set to obtain a decoding category:

[0069] o i =Ψ(f i )

[0070] Among them, i Indicates the decoding category corresponding to the i-th feature in the feature set, f i represents the i-th feature in the feature set, Ψ represents the decoder, the decoder Ψ consists of 2 layers of fully connected layers, and the decoder Ψ is used to directly map the latent features to the category space;

[0071] The second loss value corresponding to the decoding category is calculated using the following formula:

[0072] L 3 =Softmax(o i ,g i )

[0073] Among them, L 3 represents the second loss value, o i Indicates the decoding category corresponding to the i-th feature in the feature set, g i Indicates o i The corresponding label tag;

[0074] Based on the first loss value and the second loss value, the encoder and the decoder are trained to obtain a trained encoder and a trained decoder.

[0075] In order to solve the above problems, the present invention also provides an online detection system for copper sheet corrosion of petroleum products based on computer vision, the system comprising:

[0076] A microscopic acquisition module is used to construct a macroscopic data acquisition point of the test copper sheet, and to construct a microscopic data acquisition point of the test copper sheet, and to collect historical microscopic data and current microscopic data at the microscopic data acquisition point;

[0077] A diffusion analysis module, for performing data preprocessing on the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establishing a concentration change model of the micro data collection point according to the pre-historical data, and analyzing the corrosion diffusion degree corresponding to the pre-current data through the concentration change model;

[0078] A macroscopic acquisition module, for selecting a corrosion abnormality point from the macroscopic data acquisition points according to the microscopic data acquisition points and the corrosion diffusion degree, and collecting historical macroscopic data and current macroscopic data at the corrosion abnormality point;

[0079] A degree output module is used to perform equal-interval processing on the historical macro data to obtain equal-length historical data, generate a grey model of the equal-length historical data, perform equal-interval processing on the current macro data to obtain equal-length current data, and output the corrosion severity corresponding to the equal-length current data through the grey model;

[0080] A corrosion detection module is used to select an image acquisition point from the corrosion abnormal point based on the corrosion severity, perform image acquisition at the image acquisition point to obtain an acquired image, obtain a training image corresponding to the acquired image, perform image preprocessing on the training image to obtain a preprocessed image, generate a trained encoder and a trained decoder of the acquired image based on the preprocessed image, perform corrosion detection on the image acquisition point using the acquired image according to the trained encoder and the trained decoder, and obtain a corrosion detection result of the test copper sheet.

[0081] Compared with the problem described in the background technology, the embodiment of the present invention establishes a concentration change model of the micro data collection points according to the pre-historical data, so as to use the change of the petroleum concentration on the test copper sheet to identify whether corrosion has occurred on the surface of the copper sheet. If the calculated copper sheet concentration is reduced compared with the copper sheet concentration of the standard version, it means that the copper sheet has been corroded. The embodiment of the present invention selects corrosion abnormal points from the macro data collection points according to the micro data collection points and the corrosion diffusion degree, so as to select part or a small number of collection points from a large number of micro data collection points, thereby reducing the amount of corrosion detection data. Furthermore, the embodiment of the present invention performs corrosion detection on the image collection points by using the collected image to obtain the corrosion detection result of the test copper sheet, so as to reduce the amount of corrosion detection data of the copper sheet. Therefore, the online detection method for copper sheet corrosion of petroleum products based on computer vision proposed in the present invention can reduce the amount of data during copper sheet corrosion detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A schematic diagram of a process flow of an online detection method for copper corrosion of petroleum products based on computer vision provided by an embodiment of the present invention;

[0083] Figure 2 A schematic diagram of an encoder for an online detection method for copper sheet corrosion of petroleum products based on computer vision provided by an embodiment of the present invention;

[0084] Figure 3 A schematic diagram of a decoder for an online detection method for copper corrosion of petroleum products based on computer vision provided by an embodiment of the present invention;

[0085] Figure 4 A schematic diagram of a module for implementing the computer vision-based online detection method for copper sheet corrosion in petroleum products provided in one embodiment of the present invention.

[0086] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0087] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0088] The embodiment of the present application provides an online detection method for copper corrosion of petroleum products based on computer vision. The execution subject of the online detection method for copper corrosion of petroleum products based on computer vision includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the online detection method for copper corrosion of petroleum products based on computer vision can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0089] Embodiment 1:

[0090] Reference Figure 1 FIG. 1 is a flow chart of an online detection method for copper sheet corrosion of petroleum products based on computer vision according to an embodiment of the present invention. In this embodiment, the online detection method for copper sheet corrosion of petroleum products based on computer vision includes:

[0091] S1. Constructing macro data collection points of the test copper sheet, and constructing micro data collection points of the test copper sheet, and collecting historical micro data and current micro data at the micro data collection points.

[0092] In the embodiment of the present invention, the test copper sheet refers to the copper sheet in contact with petroleum, the macro data collection point refers to the location of the data collection device for collecting macro data on the copper sheet, and the macro data refers to temperature data and time data.

[0093] In one embodiment of the present invention, the construction of the macro data collection point of the test copper sheet includes: obtaining the macro data type of the test copper sheet; setting the macro data collection device of the test copper sheet based on the macro data type; querying the number of ranges of the collection range of the macro data collection device contained in the range corresponding to the test copper sheet; determining the number of devices of the macro data collection device using the number of ranges; generating a random position of the macro data collection device within the range corresponding to the test copper sheet according to the number of devices; and when the coverage area of ​​the macro data collection device at the random position is not less than the range corresponding to the test copper sheet, using the random position as the macro data collection point of the test copper sheet.

[0094] Among them, the macro data type includes temperature type and time type, the macro data acquisition device includes a temperature acquisition device and a time recording device, the range corresponding to the test copper sheet refers to the area and region of the test copper sheet, the acquisition range refers to the maximum area and region that can collect data for each of the macro data acquisition devices, the number of devices is consistent with the number of ranges, and the coverage area refers to the area occupied by the areas and regions of multiple macro data acquisition devices when collecting data on the test copper sheet.

[0095] Furthermore, in an embodiment of the present invention, the micro data collection point refers to the location of a data collection device for collecting micro data of petroleum contaminated on a copper sheet and the location of a device for collecting the concentration of the copper sheet, the micro data refers to the concentration data and volume data of material points in the petroleum contaminated on the copper sheet and the concentration data of the copper sheet, wherein the material points refer to metal atoms, the historical micro data refers to the micro data within a historical period, and the current micro data refers to the micro data at the current moment.

[0096] Optionally, the process of constructing the microscopic data collection points of the test copper sheet is similar to the process of constructing the macroscopic data collection points of the test copper sheet mentioned above, and will not be further elaborated here.

[0097] Optionally, the process of collecting historical micro data and current micro data at the micro data collection point refers to the process of measuring the concentration, volume and metal atom concentration of the entire copper sheet of metal atoms, which can be achieved by professional instruments, such as atomic absorption spectrometers.

[0098] S2. Preprocess the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establish a concentration change model of the micro data collection points based on the pre-historical data, and analyze the corrosion diffusion degree corresponding to the pre-current data through the concentration change model.

[0099] Optionally, the process of performing data preprocessing on the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data refers to the process of reducing or enlarging the unit of the micro data. For example, if the unit is too small, the numerical value will be too large. In this case, the unit is enlarged, and when enlarging the unit of this numerical value, it is also necessary to match the units of other data. Therefore, the units of other data must also be enlarged or reduced accordingly.

[0100] Furthermore, the embodiment of the present invention establishes a concentration change model of the micro data collection points based on the pre-historical data, so as to utilize the change in the petroleum concentration on the test copper sheet to identify whether corrosion has occurred on the surface of the copper sheet; if the calculated concentration of the copper sheet is lower than that of the standard copper sheet, it indicates that corrosion has occurred on the copper sheet.

[0101] The concentration variation model refers to the influence of the change of the concentration of the petroleum material point in the neighborhood of the material point on the copper sheet over time on the concentration of the material point on the copper sheet.

[0102] In one embodiment of the present invention, the concentration change model of the micro data collection point is established according to the pre-historical data, including: querying the micro concentration and copper sheet concentration of the micro data collection point from the pre-historical data; comparing the copper sheet concentration with the micro concentration to obtain a concentration comparison result; determining the diffusion parameter of the micro data collection point using the concentration comparison result; and establishing the concentration change model of the micro data collection point according to the pre-historical data and the diffusion parameter using the following formula:

[0103]

[0104] in, represents the concentration change model, s(z,t) represents the concentration at point z at time t, Indicates the neighborhood of point z in the pre-historical data The concentration of the point, Represents the z point and The diffusion parameter between points, Indicates the historical data The volume of a point, D z Represents the neighborhood of point z.

[0105] The microscopic concentration refers to the concentration of the substance in the oil. The copper sheet concentration refers to the concentration data in the spatial region much larger than point z, not the concentration of the copper sheet material point in this spatial region. Point z refers to the material point in the copper sheet, and the neighborhood of point z is the spherical area formed by the spherical radius of the material point in the copper sheet. Find Find the point on the copper sheet nearest to the point and take this point as the z point.

[0106] Optionally, the process of determining the diffusion parameter of the microscopic data collection point using the concentration comparison result refers to: when the copper flake concentration is greater than the microscopic concentration, the dissolution rate of the copper flake in the petroleum is used as the diffusion parameter; when the copper flake concentration is less than the microscopic concentration, the diffusion coefficient of the copper flake example in the petroleum is used as the diffusion parameter.

[0107] In one embodiment of the present invention, the step of analyzing the corrosion diffusion degree corresponding to the pre-current data by the concentration change model includes: analyzing the current concentration value corresponding to the pre-current data by using the concentration change model; and calculating the corrosion diffusion degree corresponding to the pre-current data by using the following formula according to the current concentration value:

[0108]

[0109] Among them, q(z′,t) represents the degree of corrosion diffusion, s(z′,t) represents the current concentration value, and s 1 Indicates the standard concentration of the test copper sheet, s 2 Indicates the standard concentration of petroleum.

[0110] The current concentration value refers to the value of s(z, t) based on the current data.

[0111] S3. According to the micro data collection point and the corrosion diffusion degree, a corrosion abnormal point is selected from the macro data collection point, and historical macro data and current macro data are collected at the corrosion abnormal point.

[0112] The embodiment of the present invention selects corrosion abnormal points from the macro data collection points according to the micro data collection points and the corrosion diffusion degree, so as to select a part or a small number of collection points from a large number of micro data collection points, thereby reducing the amount of corrosion detection data.

[0113] In one embodiment of the present invention, the selecting of corrosion abnormality points from the macro data collection points based on the micro data collection points and the corrosion diffusion degree includes: determining whether the corrosion diffusion degree is an abnormal diffusion degree; when the corrosion diffusion degree is an abnormal diffusion degree, querying a first target collection point corresponding to the corrosion diffusion degree in the micro data collection points; identifying a second target collection point in the macro data collection points that covers the first target collection point; and using the second target collection point as a corrosion abnormality point.

[0114] Optionally, the process of judging whether the corrosion diffusion degree is an abnormal diffusion degree is as follows: when q(z′, t) is non-zero, it indicates an abnormal diffusion degree.

[0115] Furthermore, in an embodiment of the present invention, the historical macro data refers to the macro data within a historical period, and the current macro data refers to the macro data at the current moment.

[0116] Optionally, the process of collecting historical macro data and current macro data at the corrosion abnormal point can be implemented by clock and temperature sensors.

[0117] S4. Perform equal-interval processing on the historical macro data to obtain equal-length historical data, generate a grey model for the equal-length historical data, perform equal-interval processing on the current macro data to obtain equal-length current data, and output the corrosion severity corresponding to the equal-length current data through the grey model.

[0118] In one embodiment of the present invention, the equal-interval processing of the historical macro data to obtain equal-length historical data includes: extracting the macro data moments of the historical macro data; sorting the macro data moments to obtain a data moment sequence; and calculating the standard time interval of the data moment sequence using the following formula:

[0119]

[0120] Wherein, Δt represents the standard time interval, m represents the length of the data time sequence, and t m represents the maximum time in the data time sequence, t 1 represents the minimum time in the data time sequence;

[0121] According to the standard time interval, the macro data sequence corresponding to the data time sequence is processed at equal intervals using the following formula to obtain equal-length historical data:

[0122]

[0123] Among them, y(t j ) represents the same length of historical data, x(t j ) indicates that the macro data sequence is at t j The historical macro data at the time, j represents the sequence number of the time in the data time series, t j represents the jth moment in the data time sequence, x(t j -1) represents x(t j ) is the predecessor in the macro data series, and Δt represents the standard time interval.

[0124] It should be noted that when the macro data is sorted at different moments, only the data of the same type is sorted, for example, the temperature data at different moments are sorted, and the time lengths at different moments are sorted, rather than a mixed sorting of temperature and time.

[0125] In one embodiment of the present invention, the grey model for generating the equal-length historical data includes: accumulating the equal-length historical data using the following formula to obtain accumulated historical data:

[0126]

[0127] Among them, y (1) (k) represents the accumulated historical data, y (0) (i) represents the i-th data in the same length historical data, n represents the number of values ​​of the same type in the same length historical data, and k represents the sequence number less than n;

[0128] According to the accumulated historical data, the grey model of the equal-length historical data is constructed using the following formula:

[0129]

[0130] Y n =BU

[0131]

[0132] in, represents the grey model, a and u in U represent unknown parameters, U represents the matrix composed of unknown parameters, y (0) (2) y (0) (3) y (0) (n) represents y (0) (i) When taking different values ​​of i, Y n Indicated by y (0) (2) y (0) (3) y (0) (n) constitutes a matrix of equal length historical data, y (1) (1) y (1) (2) y (1) (3) y (1) (n-1), y (1) (n) represents y (1) (k) takes different values ​​when k is used, B represents the matrix composed of accumulated historical data, U represents the calculation result of U, a represents the calculation result of a, u represents the calculation result of u, y (1) (t+1) represents the value of the accumulated historical data in the next period, y (1) (t) represents any type of data in the accumulated historical data, and t represents the time.

[0133] It should be noted that the process of accumulating the historical data of equal length also only accumulates the data of the same type.

[0134] In one embodiment of the present invention, the outputting the corrosion severity corresponding to the equal-length current data through the grey model includes: calculating the planned current data of the equal-length historical data corresponding to the grey model using the following formula according to the grey model:

[0135]

[0136] in, Indicates the current planning data. Indicates the value of the accumulated historical data in the next period, y (1) (t) represents any type of data in the accumulated historical data, and t represents the time;

[0137] Query the historical corrosion degree corresponding to the equal-length historical data; construct a gray analysis model of the historical corrosion degree; determine whether the planned current data is consistent with the equal-length current data; when the planned current data is consistent with the equal-length current data, use the gray analysis model to output the corrosion severity corresponding to the equal-length current data; when the planned current data is inconsistent with the equal-length current data, use the severity of copper sheet corrosion as the corrosion severity corresponding to the equal-length current data.

[0138] Among them, the planned current data refers to the macro data at a moment consistent with the moment of the equal-length current data, and the gray analysis model refers to the gray model of the historical corrosion degree. It should be noted that when the planned current data is inconsistent with the equal-length current data, it means that an abnormality occurs when using the gray model to analyze the data in the future time period, and it is necessary to use image recognition technology in the subsequent process to further identify whether copper sheet corrosion has occurred in the equal-length current data. Therefore, the severity of copper sheet corrosion is taken as the severity of corrosion corresponding to the equal-length current data. The historical corrosion degree and the corrosion severity refer to the severity level of copper sheet corrosion, such as no corrosion, slight corrosion, severe corrosion, etc.

[0139] S5. Based on the severity of the corrosion, select an image acquisition point from the corrosion abnormal point, perform image acquisition at the image acquisition point to obtain an acquired image, obtain a training image corresponding to the acquired image, perform image preprocessing on the training image to obtain a preprocessed image, generate a trained encoder and a trained decoder of the acquired image based on the preprocessed image, perform corrosion detection on the image acquisition point using the acquired image according to the trained encoder and the trained decoder, and obtain a corrosion detection result of the test copper sheet.

[0140] Optionally, the process of selecting image acquisition points from the corrosion anomaly points based on the corrosion severity is similar to the process of selecting corrosion anomaly points from the macro data acquisition points based on the micro data acquisition points and the corrosion diffusion degree, and is not further elaborated here.

[0141] The training images refer to the image data of all copper sheet surfaces.

[0142] In one embodiment of the present invention, the image preprocessing of the training image to obtain the preprocessed image includes: performing color space conversion on the training image to obtain a color conversion image; performing histogram equalization on the training image to obtain a color enhanced image; performing edge detection on the training image to obtain an edge detection image; performing image filtering on the training image to obtain a filtered image; enhancing the texture features of the training image to obtain a feature enhanced image; fusing the color conversion image, the color enhanced image, the edge detection image, the filtered image and the feature enhanced image to obtain a multi-channel image; removing noise from the multi-channel image to obtain a denoised image; performing image normalization on the denoised image to obtain a normalized image;

[0143] Optionally, the process of performing image preprocessing on the training image to obtain a preprocessed image includes: converting the RGB image into a Lab color space to highlight the color difference in the image; using the brightness channel (L) in the Lab space to perform CLAHE (contrast limited adaptive histogram equalization) to enhance the brightness contrast of the image and make the color more obvious; using Canny edge detection to extract edge information of the image and enhance texture features; using a multi-scale, multi-directional Gabor filter to extract texture features of a specific scale and direction; using a local binary pattern (LBP) to extract local texture patterns in the image and enhance subtle surface texture features; fusing the enhanced color image, edge image, Gabor filtering results, and LBP features into a multi-channel image; using bilateral filtering to remove noise and retain edge information; and normalizing the image so that its value is between 0 and 1 for easy neural network processing.

[0144] In one embodiment of the present invention, generating a trained encoder and a trained decoder of the captured image based on the preprocessed image includes: dividing the preprocessed image into a multi-view sample anchor point, a positive sample, and a negative sample; in a preset encoder, mapping the multi-view sample anchor point, the positive sample, and the negative sample to a feature latent space using the following formulas, respectively, to obtain anchor point features, positive sample features, and negative sample features:

[0145] f a =Φ(s a ),f p =Φ(s p ),f n =Φ(s n )

[0146] Among them, f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s arepresents the multi-view sample anchor, Φ represents the encoder, s p represents a positive sample, s n represents negative samples;

[0147] The first loss value corresponding to the anchor point feature, the positive sample feature and the negative sample feature is calculated using the following formula:

[0148]

[0149] L 2 =||Rec(f a )-s a ||

[0150] Among them, L 1 Indicates the loss value between the anchor point feature and the positive sample feature and the negative sample feature in the first loss value, L 2 represents the loss value between the anchor feature and the multi-view sample anchor in the first loss value, f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s a Represents the multi-view sample anchor point, Rec represents the reconstruction model, and Rec is used to transform f a Reconstruct the original image and ensure that the latent features extracted by the encoder Φ are complete;

[0151] Obtain a feature set corresponding to the anchor feature, the positive sample feature, and the negative sample feature; in a preset decoder, use the following formula to perform feature decoding on the feature set to obtain a decoding category:

[0152] o i =Ψ(f i )

[0153] Among them, i Indicates the decoding category corresponding to the i-th feature in the feature set, f i represents the i-th feature in the feature set, Ψ represents the decoder, the decoder Ψ consists of 2 layers of fully connected layers, and the decoder Ψ is used to directly map the latent features to the category space;

[0154] The second loss value corresponding to the decoding category is calculated using the following formula:

[0155] L 3 =Softmax(o i ,g i )

[0156] Among them, L 3 represents the second loss value, o i Indicates the decoding category corresponding to the i-th feature in the feature set, gi Indicates o i The corresponding label tag;

[0157] Based on the first loss value and the second loss value, the encoder and the decoder are trained to obtain a trained encoder and a trained decoder.

[0158] Among them, the network structure of the encoder Φ can explore the best structure in practical applications, which can be the mainstream ResNet, or the latest ViT, or a simple 4-layer convolutional neural network. The patent of this invention aims to propose a general algorithm, and the network structure can be arbitrarily designed. The network structure of Rec is the same as that of the encoder Φ.

[0159] It should be noted that It is used to shorten the distance between the anchor point and the positive sample features and to lengthen the distance between the anchor point and the negative sample. 2 =||Rec(f a )-s a || is used to ensure that the feature encoder has a complete understanding of the sample, L 3

[0160] =Softmax(o i ,g i ) is used for end-to-end learning. The first loss value is used to optimize the encoder parameters to obtain more discriminative features. These features can then be mapped to the final degree of corrosion through the decoder. In this process, only the decoder needs to be learned. This is an end-to-end learning process. The output of the decoder is compared with the manually annotated labels, and the loss value of the comparison result is back-propagated to learn the decoder.

[0161] See also Figure 2 , which is a schematic diagram of an encoder for an online detection method for copper sheet corrosion of petroleum products based on computer vision provided by an embodiment of the present invention.

[0162] See also Figure 3 , which is a schematic diagram of a decoder of an online detection method for copper sheet corrosion of petroleum products based on computer vision provided by an embodiment of the present invention.

[0163] Furthermore, the embodiment of the present invention uses the acquired image to perform corrosion detection on the image acquisition point according to the trained encoder and the trained decoder, so as to reduce the amount of corrosion detection data of the copper sheet.

[0164] Optionally, the process of using the acquired image to perform corrosion detection on the image acquisition point to obtain the corrosion detection result of the test copper sheet refers to the process of performing image recognition using an encoder and a decoder with optimized parameters, thereby obtaining the category of corrosion, which is similar to the aforementioned corrosion severity, including uncorroded, slightly corroded, severely corroded, etc.

[0165] Compared with the problem described in the background technology, the embodiment of the present invention establishes a concentration change model of the micro data collection points according to the pre-historical data, so as to use the change of the petroleum concentration on the test copper sheet to identify whether corrosion has occurred on the surface of the copper sheet. If the calculated copper sheet concentration is reduced compared with the copper sheet concentration of the standard version, it means that the copper sheet has been corroded. The embodiment of the present invention selects corrosion abnormal points from the macro data collection points according to the micro data collection points and the corrosion diffusion degree, so as to select part or a small number of collection points from a large number of micro data collection points, thereby reducing the amount of corrosion detection data. Furthermore, the embodiment of the present invention performs corrosion detection on the image collection points by using the collected image to obtain the corrosion detection result of the test copper sheet, so as to reduce the amount of corrosion detection data of the copper sheet. Therefore, the online detection method for copper sheet corrosion of petroleum products based on computer vision proposed in the present invention can reduce the amount of data during copper sheet corrosion detection.

[0166] Embodiment 2:

[0167] like Figure 4 The figure shows a functional module diagram of an online detection system for copper sheet corrosion of petroleum products based on computer vision according to the present invention.

[0168] The computer vision-based online detection system 400 for copper corrosion of petroleum products of the present invention can be installed in an electronic device. According to the functions to be implemented, the computer vision-based online detection system for copper corrosion of petroleum products can include a microscopic acquisition module 401, a diffusion analysis module 402, a macroscopic acquisition module 403, a degree output module 404 and a corrosion detection module 405. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0169] In the embodiment of the present invention, the functions of each module / unit are as follows:

[0170] The microscopic acquisition module 401 is used to construct a macroscopic data acquisition point of the test copper sheet, and to construct a microscopic data acquisition point of the test copper sheet, and to collect historical microscopic data and current microscopic data at the microscopic data acquisition point;

[0171] The diffusion analysis module 402 is used to perform data preprocessing on the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establish a concentration change model of the micro data collection point according to the pre-historical data, and analyze the corrosion diffusion degree corresponding to the pre-current data through the concentration change model;

[0172] The macro data collection module 403 is used to select a corrosion abnormal point from the macro data collection points according to the micro data collection points and the corrosion diffusion degree, and collect historical macro data and current macro data at the corrosion abnormal point;

[0173] The degree output module 404 is used to perform equal-interval processing on the historical macro data to obtain equal-length historical data, generate a grey model of the equal-length historical data, perform equal-interval processing on the current macro data to obtain equal-length current data, and output the corrosion severity corresponding to the equal-length current data through the grey model;

[0174] The corrosion detection module 405 is used to select an image acquisition point from the corrosion abnormal point based on the corrosion severity, perform image acquisition at the image acquisition point to obtain an acquired image, obtain a training image corresponding to the acquired image, perform image preprocessing on the training image to obtain a preprocessed image, generate a trained encoder and a trained decoder of the acquired image based on the preprocessed image, and perform corrosion detection on the image acquisition point using the acquired image according to the trained encoder and the trained decoder to obtain a corrosion detection result of the test copper sheet.

[0175] In detail, the modules in the computer vision-based online detection system 300 for copper corrosion of petroleum products in the embodiment of the present invention are used in the same manner as described above. Figure 1 The same technical means are used as the online detection method for copper sheet corrosion of petroleum products based on computer vision described in, and can produce the same technical effects, so they will not be repeated here.

[0176] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A computer vision-based online detection method for copper corrosion in petroleum products, characterized in that: The method comprises: Constructing macro data collection points of the test copper sheet, and constructing micro data collection points of the test copper sheet, and collecting historical micro data and current micro data at the micro data collection points; Preprocessing the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establishing a concentration change model of the micro data collection point according to the pre-historical data, and analyzing the corrosion diffusion degree corresponding to the pre-current data through the concentration change model; According to the micro data collection point and the corrosion diffusion degree, a corrosion abnormal point is selected from the macro data collection point, and historical macro data and current macro data are collected at the corrosion abnormal point; Performing equal-interval processing on the historical macro data to obtain equal-length historical data, generating a grey model for the equal-length historical data, performing equal-interval processing on the current macro data to obtain equal-length current data, and outputting the corrosion severity corresponding to the equal-length current data through the grey model; Based on the severity of the corrosion, an image acquisition point is selected from the corrosion abnormal point, image acquisition is performed at the image acquisition point to obtain an acquired image, a training image corresponding to the acquired image is obtained, image preprocessing is performed on the training image to obtain a preprocessed image, and based on the preprocessed image, a trained encoder and a trained decoder of the acquired image are generated, and corrosion detection is performed on the image acquisition point using the acquired image according to the trained encoder and the trained decoder to obtain a corrosion detection result of the test copper sheet.

2. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The macro data collection points for constructing the test copper sheet include: Acquire the macro data type of the test copper sheet; Based on the macro data type, a macro data acquisition device of the test copper sheet is set; Querying the number of ranges of the collection range of the macro data collection device contained in the range corresponding to the test copper sheet; Determining the number of devices of the macro data acquisition device using the number of ranges; Generating random positions of the macro data acquisition devices within the range corresponding to the test copper sheet according to the number of the devices; When the coverage area of ​​the macro data acquisition device at the random position is not less than the range corresponding to the test copper sheet, the random position is used as the macro data acquisition point of the test copper sheet.

3. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The step of establishing a concentration variation model of the microscopic data collection point according to the pre-historical data includes: Query the microscopic concentration and copper flake concentration of the microscopic data collection point from the pre-historical data; Comparing the copper sheet concentration with the microscopic concentration to obtain a concentration comparison result; Determining the diffusion parameters of the microscopic data collection points using the concentration comparison results; According to the pre-historical data and the diffusion parameters, the concentration change model of the microscopic data collection point is established using the following formula: in, represents the concentration change model, s(z,t) represents the concentration at point z at time t, Indicates the neighborhood range of point z in the pre-historical data The concentration of the point, Represents the z point and The diffusion parameter between points, Indicates the historical data The volume of a point, D z Represents the neighborhood of point z.

4. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The analyzing the corrosion diffusion degree corresponding to the pre-current data by the concentration change model includes: Analyzing the current concentration value corresponding to the pre-current data using the concentration change model; According to the current concentration value, the corrosion diffusion degree corresponding to the pre-current data is calculated using the following formula: in, Indicates the degree of corrosion diffusion, Indicates the current concentration value, s1 indicates the standard concentration of the test copper sheet, and s2 indicates the standard concentration of petroleum.

5. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The selecting of corrosion abnormal points from the macro data collection points according to the micro data collection points and the corrosion diffusion degree comprises: Determining whether the corrosion diffusion degree is an abnormal diffusion degree; When the corrosion diffusion degree is an abnormal diffusion degree, querying a first target collection point corresponding to the corrosion diffusion degree in the micro data collection point; Identifying a second target collection point among the macro data collection points that covers the first target collection point; The second target acquisition point is taken as a corrosion abnormal point.

6. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The processing of the historical macro data at equal intervals to obtain historical data of equal length includes: Extract the macro data moments of the historical macro data; sort the macro data moments to obtain a data moment sequence; and calculate the standard time interval of the data moment sequence using the following formula: Wherein, Δt represents the standard time interval, m represents the length of the data time sequence, and t m represents the maximum time in the data time sequence, and t1 represents the minimum time in the data time sequence; According to the standard time interval, the macro data sequence corresponding to the data time sequence is processed at equal intervals using the following formula to obtain equal-length historical data: Among them, y(t j ) represents the same length of historical data, x(t j ) indicates that the macro data sequence is at t j The historical macro data at the time, j represents the sequence number of the time in the data time sequence, t j represents the jth moment in the data time sequence, x(t j -1) means x(t j ) is the predecessor in the macro data series, and Δt represents the standard time interval.

7. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The grey model for generating the equal-length historical data comprises: The same length historical data is accumulated using the following formula to obtain accumulated historical data: Among them, y (1) (k) represents the accumulated historical data, y (0) (i) represents the i-th data in the same length historical data, n represents the number of values ​​of the same type in the same length historical data, and k represents the sequence number less than n; According to the accumulated historical data, the grey model of the equal-length historical data is constructed using the following formula: in, represents the grey model, a and u in U represent unknown parameters, U represents the matrix composed of unknown parameters, y (0) (2) y (0) (3) y (0) (n) represents y (0) (i) When taking different values ​​of i, Y n Indicated by y (0) (2) y (0) (3) y (0) (n) constitutes a matrix of equal length historical data, y (1) (1) y (1) (2) y (1) (3) y (1) (n-1), y (1) (n) represents y (1) (k) takes different values ​​when k is used, B represents the matrix composed of accumulated historical data, U represents the calculation result of U, a represents the calculation result of a, u represents the calculation result of u, y (1) (t+1) represents the value of the accumulated historical data in the next period, y (1) (t) represents any type of data in the accumulated historical data, and t represents the time.

8. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: Outputting the corrosion severity corresponding to the equal-length current data through the grey model includes: According to the grey model, calculating the planned current data of the equal-length historical data corresponding to the grey model; Query the historical corrosion degree corresponding to the historical data of equal length; Constructing a grey analysis model of the historical corrosion degree; Determine whether the planned current data is consistent with the equal-length current data; When the planned current data is consistent with the equal-length current data, outputting the corrosion severity corresponding to the equal-length current data using the grey analysis model; When the planned current data is inconsistent with the equal-length current data, the severity of the copper sheet corrosion is used as the corrosion severity corresponding to the equal-length current data.

9. The computer vision-based online detection method for copper corrosion of petroleum products according to claim 1, characterized in that: The step of generating a trained encoder and a trained decoder of the acquired image based on the preprocessed image comprises: Dividing the preprocessed image into multi-view sample anchors, positive samples and negative samples; In the preset encoder, the multi-view sample anchor point, the positive sample and the negative sample are respectively mapped to the feature latent space using the following formulas to obtain the anchor point feature, the positive sample feature and the negative sample feature: f a =Φ(s a ),f p =Φ(s p ),f n =Φ(s n ) Among them, f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s a represents the multi-view sample anchor, Φ represents the encoder, s p represents a positive sample, s n represents negative samples; The first loss value corresponding to the anchor point feature, the positive sample feature and the negative sample feature is calculated using the following formula: L2=||Rec(f a )-s a || Among them, L1 represents the loss value between the anchor feature in the first loss value and the positive sample feature and the negative sample feature respectively, L2 represents the loss value between the anchor feature in the first loss value and the multi-view sample anchor point, and f a represents the anchor feature, f p represents the positive sample feature, f n represents the negative sample feature, s a Represents the multi-view sample anchor point, Rec represents the reconstruction model, and Rec is used to transform f a Reconstruct the original image and ensure that the latent features extracted by the encoder Φ are complete; Obtain a feature set corresponding to the anchor feature, the positive sample feature, and the negative sample feature; in a preset decoder, use the following formula to perform feature decoding on the feature set to obtain a decoding category: the i =Ψ(f i ) Among them, i Indicates the decoding category corresponding to the i-th feature in the feature set, f i represents the i-th feature in the feature set, Ψ represents the decoder, the decoder Ψ consists of 2 layers of fully connected layers, and the decoder Ψ is used to directly map the latent features to the category space; The second loss value corresponding to the decoding category is calculated using the following formula: <h2 style=";text-align:left;direction:ltr">L3=Softmax(o<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ,g<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> ) Among them, L3 represents the second loss value, o i Indicates the decoding category corresponding to the i-th feature in the feature set, g i Indicates o i The corresponding label tag; Based on the first loss value and the second loss value, the encoder and the decoder are trained to obtain a trained encoder and a trained decoder.

10. An online detection system for copper corrosion of petroleum products based on computer vision, characterized in that: The system comprises: A microscopic acquisition module is used to construct a macroscopic data acquisition point of the test copper sheet, and to construct a microscopic data acquisition point of the test copper sheet, and to collect historical microscopic data and current microscopic data at the microscopic data acquisition point; A diffusion analysis module, for performing data preprocessing on the historical micro data and the current micro data respectively to obtain pre-historical data and pre-current data, establishing a concentration change model of the micro data collection point according to the pre-historical data, and analyzing the corrosion diffusion degree corresponding to the pre-current data through the concentration change model; A macroscopic acquisition module, for selecting a corrosion abnormality point from the macroscopic data acquisition points according to the microscopic data acquisition points and the corrosion diffusion degree, and collecting historical macroscopic data and current macroscopic data at the corrosion abnormality point; A degree output module is used to perform equal-interval processing on the historical macro data to obtain equal-length historical data, generate a grey model of the equal-length historical data, perform equal-interval processing on the current macro data to obtain equal-length current data, and output the corrosion severity corresponding to the equal-length current data through the grey model; A corrosion detection module is used to select an image acquisition point from the corrosion abnormal point based on the severity of the corrosion, perform image acquisition at the image acquisition point to obtain an acquired image, obtain a training image corresponding to the acquired image, perform image preprocessing on the training image to obtain a preprocessed image, generate a trained encoder and a trained decoder of the acquired image based on the preprocessed image, perform corrosion detection on the image acquisition point using the acquired image according to the trained encoder and the trained decoder, and obtain a corrosion detection result of the test copper sheet.

Images