Aluminum oxide production intelligent quality control method and system based on image recognition

Abstract

The invention relates to the technical field of aluminum oxide production, solves the technical problem of inaccurate analysis of an image recognition segmentation method in the prior art of aluminum oxide production, and particularly relates to an intelligent quality control method for aluminum oxide production based on image recognition, which comprises the following steps: S1, obtaining an original image of a target to be detected, s2, preprocessing the pixel image to obtain a de-noised image; and S3, converting the de-noised image into a two-dimensional histogram and establishing a two-dimensional coordinate system, the histogram is subjected to beveling through a beveling analysis method, and compared with a conventional rectangular segmentation mode, the beveling mode can better analyze an area near a diagonal line of the histogram, so that image recognition and analysis are more comprehensive and accurate, and the image recognition efficiency is improved. The conditions of inaccurate analysis and even neglecting of areas near diagonals are avoided, and the accuracy of image recognition is greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of alumina production, and in particular to an intelligent quality control method and system for alumina production based on image recognition. Background Art

[0002] Rotary kilns are often used in the sintering production of alumina. The working process of the rotary kiln is divided into five stages according to different stages: preheating zone, firing zone, drying zone, decomposition zone and cooling zone. The state of the firing zone has an important influence on whether clinker can be produced with high quality, high yield and low consumption. Its temperature can reflect the internal situation of the rotary kiln, which directly affects the sintering quality of alumina. The existing technology generally obtains the temperature situation in the kiln through the observation hole of the rotary kiln. The commonly used firing zone temperature detection method mostly uses the manual fire observation method, but this method has a large error and has high requirements on the technical level, psychological quality, mental state and sense of responsibility of the staff. Therefore, the cost is high. Another colorimetric temperature measurement method uses the principle of infrared radiation temperature measurement, but due to the influence of internal environmental factors and narrow and long space of the rotary kiln, the measurement error is also large. The existing image recognition uses a histogram to analyze the image, which cannot fully analyze the image and is even easy to ignore the area near the diagonal line in the image, resulting in unsatisfactory analysis results. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present invention provides an intelligent quality control method and system for alumina production based on image recognition, which solves the technical problem of inaccurate analysis of image recognition and segmentation methods in the prior art of alumina production, and achieves the purpose of increasing the accuracy of image recognition and analysis by segmenting images using a new method.

[0004] In order to solve the above technical problems, the present invention provides the following technical solutions: an intelligent quality control method for alumina production based on image recognition, the method comprising the following steps:

[0005] S1, obtaining an original image of the target to be detected, and converting the original image into a pixel image;

[0006] S2, preprocessing the pixel image to obtain a denoised image;

[0007] S3, converting the denoised image into a two-dimensional histogram and establishing a two-dimensional coordinate system;

[0008] S4, convert the two-dimensional coordinate system of the two-dimensional histogram into an oblique cut coordinate system according to the oblique cut method, and according to the probability point P of the oblique cut coordinate system b (i, j) gets the probability h a (i, j);

[0009] S5. According to the probability h a(i, j) calculate the threshold selection function D(i), and obtain the optimal threshold t according to the threshold selection function D(i);

[0010] S6, converting the two-dimensional histogram into a binary image according to the optimal threshold t, and segmenting the binary image to obtain a segmented image;

[0011] S7. Analyze the segmented image to obtain the quality status and send it to the quality control end.

[0012] Preferably, in step S1, the specific implementation steps are as follows:

[0013] S11, input the original image into the RGB channel and convert it into an RGB image;

[0014] S12, determine the brightness L of any pixel in the RGB image a And adjustment coefficient A, calculate the brightness L of the pixel a The low brightness value L min , the calculation formula is as follows:

[0015] L min =A×L a

[0016] Among them, L min Indicates the low brightness value of the pixel, A indicates the adjustment coefficient and 30%≤A≤70%, L a Indicates the brightness of a pixel in an RGB image;

[0017] S13, repeat step S12 until the brightness of all pixels in the RGB image is adjusted to generate a pixel image.

[0018] Preferably, in step S2, the specific implementation steps are as follows:

[0019] S21. Determine the threshold function sgt(b) based on the wavelet transform formula, and its function expression is:

[0020]

[0021] Wherein, sgt(b) represents the threshold function, and b represents the independent variable of the threshold function sgt(b);

[0022] S22, setting the original wavelet coefficient w based on the historical data of the wavelet function y , set the threshold T based on historical data, and calculate the optimized wavelet coefficient w according to the threshold function sgt(b) a , the calculation formula is as follows:

[0023]

[0024] Among them, w aDenote the optimized wavelet coefficient as w y Denote the original wavelet coefficient, and T represents the threshold;

[0025] S23. Obtain the wavelet coefficient w of the smallest layer by using the iterative decomposition method according to the optimized wavelet coefficient w a ; min ;

[0026] S24. Calculate the standard deviation noise value σ according to the wavelet coefficient w of the smallest layer. The calculation formula is as follows: min ;

[0027]

[0028] where σ represents the standard deviation noise value, and median(w min ) represents the median of the wavelet coefficient of the smallest layer;

[0029] S25. Obtain the signal-to-noise ratio of the pixel image by using the peak signal-to-noise ratio method, and determine the decomposition scale B according to the signal-to-noise ratio;

[0030] S26. Obtain the width W and height H of the pixel image, and calculate the threshold function K according to the decomposition scale B c , and the calculation formula is as follows:

[0031]

[0032] where K c represents the threshold function, σ represents the standard deviation noise value, and B represents the decomposition scale;

[0033] S27. Set the denoising threshold H based on the historical data of denoising, and judge whether the pixel point is removed according to the threshold function K c ;

[0034] If K c ≥H, then the pixel point is a noise point and is removed;

[0035] If K c <H, then the pixel point is a non-noise point and is retained;

[0036] S28. After repeating step S27 to traverse all pixel points of the pixel image, obtain the denoised image.

[0037] Preferably, in step S3, the specific implementation steps are as follows:

[0038] S31. Convert the denoised image into an HSV image through the HSV color space;

[0039] S32. Create a canvas to be drawn, and convert the HSV image into a two-dimensional histogram by using the OpenCV algorithm;

[0040] S33. Select a point O on the two-dimensional histogram a As the coordinate origin and establish a two-dimensional coordinate system.

[0041] Preferably, in step S4, the specific implementation steps are as follows:

[0042] S41, setting the width and height of the two-dimensional histogram to L-1;

[0043] S42, converting the x-axis and y-axis in the two-dimensional coordinate system into the x'-axis and y'-axis of the oblique coordinate system, the conversion formula is as follows:

[0044]

[0045] Wherein, x' represents the transformed oblique cut horizontal axis, and y' represents the transformed oblique cut vertical axis;

[0046] S43, obtaining the total number of pixels U according to the two-dimensional histogram a , create probability point P in the oblique coordinate system b (i, j);

[0047] S44, obtain probability point P according to the two-dimensional histogram b The frequency of occurrence of (i, j) is E ij , and calculate the probability point P b The probability of (i, j) appearing h a (i, j), the calculation formula is as follows:

[0048]

[0049] Among them, h a (i, j) represents the probability point P b The probability of (i, j) occurring and E ij Represents the probability point P b The frequency of occurrence of (i, j), U a Represents the total number of pixels in a two-dimensional histogram.

[0050] Preferably, in step S5, the specific implementation steps are as follows:

[0051] S51, dividing the two-dimensional histogram into two parts i≤t and i>t in the oblique coordinate system;

[0052] S52. According to the probability h a (i, j) calculates the first coefficient f1, and the calculation formula is as follows:

[0053]

[0054] Among them, f1 represents the first coefficient, ha (i) represents the probability with abscissa i, and t represents the number of probabilities;

[0055] S53. According to the probability h a (i, j) calculates the second coefficient f2, and the calculation formula is as follows:

[0056]

[0057] in, represents the probability quantity;

[0058] S54. According to the probability h a (i, j) and probability point P b The first correlation value β1 is calculated by the abscissa value i of (i, j), and the calculation formula is as follows:

[0059]

[0060] Among them, i represents the probability point P b The horizontal coordinate value of (i, j);

[0061] S55. According to the probability h a (i, j) and probability point P b The second correlation value β2 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows:

[0062]

[0063] Wherein, β2 represents the second correlation value;

[0064] S56. According to the probability h a (i, j) and probability point P b The third correlation value β3 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows:

[0065]

[0066] Wherein, β3 represents the third correlation value;

[0067] S57, calculating the selected value D(i) according to the first coefficient f1, the second coefficient f2, the first correlation value β1, the second correlation value β2 and the third correlation value β3, the calculation formula is as follows:

[0068] D(i)=f1(β1-β3) 2 +f2(β2-β3) 2

[0069] Where D(i) represents the selected value;

[0070] S58, judging whether the selected value D(i) is the optimal threshold;

[0071] If D(i) is the maximum value, the process ends and the independent variable i is used as the optimal threshold value t, and the process proceeds to step S6;

[0072] If D(i) is not the maximum value, step S58 is repeated until the maximum value of D(i) is found.

[0073] Preferably, in step S6, the specific implementation steps are as follows:

[0074] S61, obtain the segmentation function G according to the optimal threshold t a ;

[0075] S62, randomly select a pixel point M in the two-dimensional histogram a (c, d), according to the partition function G a Determine pixel M a Pixel values ​​of (c, d);

[0076] like Then the pixel value is 0 and ends;

[0077] like Then the pixel value is 1 and ends;

[0078] S63, repeating step S62 to traverse all pixels in the two-dimensional histogram to obtain a binary image;

[0079] S64, segmenting the binary image along the edge of the binary image and generating a segmented image.

[0080] Preferably, the segmentation function G a The function formula is as follows:

[0081]

[0082] Among them, G a represents the segmentation function, j represents the probability point P b The ordinate of (i, j), t represents the optimal threshold.

[0083] Preferably, in step S7, the specific implementation steps are as follows:

[0084] S71. Obtaining a segmentation function G in the segmented image a Count the number of pixels Z in the zero-value image.

[0085] S72, setting a threshold interval [k, l] based on historical data of aluminum oxide production to determine whether the quality status of the zero-value image is good;

[0086] If k ≥ Z ≥ l, the quality status is good and the quality status is sent to the quality control end;

[0087] If Z > k or Z < l, the quality status is abnormal and the quality status is sent to the quality control terminal.

[0088] Preferably, the technical solution further provides a system for the intelligent quality control method of alumina production described above. The system includes: an image conversion module that acquires the original image of the target to be detected and converts the original image into a pixel image;

[0089] A preprocessing module that preprocesses the pixel image to obtain a denoised image;

[0090] A two-dimensional processing module that converts the denoised image into a two-dimensional histogram and establishes a two-dimensional coordinate system;

[0091] A probability analysis module that converts the two-dimensional coordinate system of the two-dimensional histogram into an oblique cutting coordinate system according to the oblique cutting method, and obtains the probability h b (i, j) based on the probability points P a (i, j) of the oblique cutting coordinate system;

[0092] A threshold analysis module that calculates the threshold selection function D(i) according to the probability h a (i, j), and obtains the optimal threshold t according to the threshold selection function D(i);

[0093] A segmented image module that converts the two-dimensional histogram into a binary image according to the optimal threshold t, and segments the binary image to obtain a segmented image;

[0094] A quality analysis module that analyzes the segmented image to obtain the quality status and sends it to the quality control.

[0095] By means of the above technical solution, the present invention provides an intelligent quality control method and system for alumina production based on image recognition, which at least has the following beneficial effects:

[0096] 1. The present invention performs oblique cutting on the histogram through the oblique cutting analysis method. Compared with the conventional rectangular segmentation method, the oblique cutting method can better analyze the area near the diagonal of the histogram, making the image recognition and analysis more comprehensive and accurate, avoiding inaccurate analysis or even ignoring the area near the diagonal, and greatly improving the accuracy of image recognition.

[0097] 2. In the denoising process of the present invention, optimized wavelet coefficients are used for denoising processing, and the pixel size of the image is combined in the calculation and analysis, which can improve the denoising effect, increase the work efficiency of image denoising, and make the image clearer after denoising.

[0098] 3. The present invention utilizes image recognition to detect the quality of alumina production through image acquisition at the observation port of the rotary kiln. This is more accurate than manual or infrared temperature measurement methods. The quality status of alumina can be accurately judged by the flame color and area size in the image, thereby improving the quality of the produced alumina. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0100] Figure 1 This is a flow chart of an intelligent quality control method for aluminum oxide production based on image recognition according to the present invention;

[0101] Figure 2 This is a structural block diagram of an intelligent quality control system for aluminum oxide production based on image recognition according to the present invention. DETAILED DESCRIPTION

[0102] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods, so that the implementation process of how the present application uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly.

[0103] Due to the technical problem of inaccurate analysis of image recognition and segmentation methods in the existing technology of alumina production, please refer to Figure 1 This embodiment provides an intelligent quality control method and system for aluminum oxide production based on image recognition, which can segment images using a new method to increase the accuracy of image recognition and analysis. The method includes the following steps:

[0104] S1, obtaining the original image of the target to be detected, and converting the original image into a pixel image. The present invention uses an image acquisition device to acquire an image of the internal flame of the alumina rotary kiln from the observation port of the rotary kiln. In step S1, the specific implementation steps are as follows:

[0105] S11, input the original image into the RGB channel and convert it into an RGB image;

[0106] S12, determine the brightness L of any pixel in the RGB image a And the adjustment coefficient A, calculate the brightness L of the pixel a The low brightness value L min , the calculation formula is as follows:

[0107] L min =A×L a

[0108] Among them, L min Indicates the low brightness value of the pixel, A indicates the adjustment coefficient and 30%≤A≤70%, L a Indicates the brightness of the pixel in the RGB image. During the image acquisition process, the flame color in alumina production is very bright and light, and the direct observation accuracy is low. The continuous high brightness of the screen is harmful to the observer's eyes and the luminous screen. During the analysis process, the image brightness needs to be adjusted to a low brightness to help protect the human eye and the screen.

[0109] S13, repeat step S12 until the brightness of all pixels in the RGB image is adjusted to generate a pixel image.

[0110] S2, pre-process the pixel image to obtain a denoised image. Due to the influence of the ambient temperature and particles in the kiln, the image inside the rotary kiln contains a lot of noise and needs to be denoised. In step S2, the specific implementation steps are as follows:

[0111] S21. Determine the threshold function sgt(b) based on the wavelet transform formula, and its function expression is:

[0112]

[0113] Wherein, sgt(b) represents a threshold function, b represents an independent variable of the threshold function sgt(b), some symbol functions are required in wavelet transform, and the present invention adopts the threshold function sgt(b) as a function for calculating wavelet coefficients.

[0114] S22, setting the original wavelet coefficient w based on the historical data of the wavelet function y , set the threshold T based on historical data, and calculate the optimized wavelet coefficient w according to the threshold function sgt(b) a , the calculation formula is as follows:

[0115]

[0116] Among them, w a represents the optimized wavelet coefficients, w y Represents the original wavelet coefficient, T represents the threshold value, and the historical data of the wavelet function is the relevant data stored in the database of the alumina rotary kiln. In practice, each production will record the wavelet coefficients, flame color, flame temperature, alumina quality and other data after denoising, which is helpful for quality analysis of alumina production through images.

[0117] S23, according to the optimized wavelet coefficient w a The iterative decomposition method is used to obtain the wavelet coefficient w of the minimum layer min, The iterative decomposition method is a commonly used method for obtaining the wavelet coefficients of the minimum layer. The analysis result can be obtained only by using the iterative decomposition method, which will not be elaborated here.

[0118] S24. According to the wavelet coefficients w of the minimum layer min Calculate the standard deviation noise value σ, and the calculation formula is as follows:

[0119]

[0120] where σ represents the standard deviation noise value, and median(w min ) represents the median of the wavelet coefficients of the minimum layer;

[0121] S25. Use the peak signal-to-noise ratio method to obtain the signal-to-noise ratio of the pixel image, and determine the decomposition scale B according to the signal-to-noise ratio. The peak signal-to-noise ratio method is a common method for obtaining the signal-to-noise ratio. In the present invention, the desired result can be obtained only by using the peak signal-to-noise ratio method, which will not be elaborated here.

[0122] S26. Obtain the width W and height H of the pixel image, and calculate the threshold function K according to the decomposition scale B c , and the calculation formula is as follows:

[0123]

[0124] where K c represents the threshold function, σ represents the standard deviation noise value, and B represents the decomposition scale;

[0125] S27. Set the denoising threshold H based on the historical data of denoising, and judge whether the pixel points are eliminated according to the threshold function K c ;

[0126] If K c ≥H, then the pixel point is a noise point and is eliminated;

[0127] If K c <H, then the pixel point is a non-noise point and is retained. Removing the noise in the image through denoising helps the subsequent steps.

[0128] S28. Repeat step S27 to traverse all pixel points of the pixel image to obtain the denoised image.

[0129] S3. Convert the denoised image into a two-dimensional histogram and establish a two-dimensional coordinate system. Establishing a two-dimensional coordinate system helps to segment and analyze the image and can lay a foundation for the subsequent steps. In step S3, the specific implementation steps are as follows:

[0130] S31. Convert the denoised image into an HSV image through the HSV color space. The HSV color space is also called the HSV color model. The color parameters in this model are hue, saturation and brightness. Its advantages mainly include intuitiveness, independence, uniformity and easy conversion.

[0131] S32. Create a canvas to be drawn, and use the OpenCV algorithm to convert the HSV image into a two-dimensional histogram. The OpenCV algorithm is an open source computer vision and machine learning software library, which contains a variety of computer vision algorithms, such as image processing, feature detection, face recognition, object recognition, etc. The algorithm can convert the HSV image into a two-dimensional histogram, which will not be repeated here.

[0132] S33. Select a point O on the two-dimensional histogram a As the coordinate origin and establish a two-dimensional coordinate system.

[0133] S4, convert the two-dimensional coordinate system of the two-dimensional histogram into an oblique cut coordinate system according to the oblique cut method, and according to the probability point P of the oblique cut coordinate system b (i, j) gets the probability h a (i, j), the oblique cutting method is to transform the conventional horizontal and vertical two-dimensional coordinate system of the two-dimensional histogram into a coordinate system inclined at 45 degrees. The inclined coordinate system is helpful for analyzing the image and avoids ignoring the pixels near the diagonal of the conventional histogram. In step S4, the specific implementation steps are as follows:

[0134] S41, setting the width and height of the two-dimensional histogram to L-1;

[0135] S42, converting the x-axis and y-axis in the two-dimensional coordinate system into the x'-axis and y'-axis of the oblique coordinate system, the conversion formula is as follows:

[0136]

[0137]

[0138] Among them, x' represents the transformed oblique horizontal axis, and y' represents the transformed oblique vertical axis. After the coordinate system is converted into the oblique coordinate system, since the inclination angle is 45 degrees, the coordinate system needs to be divided by

[0139] S43, obtaining the total number of pixels U according to the two-dimensional histogram a , create probability point P on the oblique coordinate system b (i, j);

[0140] S44, obtain probability point P according to the two-dimensional histogram b The frequency of occurrence of (i, j) is E ij , and calculate the probability point Pb The probability of (i, j) appearing h a (i, j), the calculation formula is as follows:

[0141]

[0142] Among them, h a (i, j) represents the probability point P b The probability of (i, j) occurring and E ij Represents the probability point P b The frequency of occurrence of (i, j), U a Represents the total number of pixels in the two-dimensional histogram. It is the value after conversion under the oblique coordinate system.

[0143] S5. According to the probability h a (i, j) calculates the threshold selection function D(i), and obtains the optimal threshold t according to the threshold selection function D(i). In the process of selecting the optimal threshold, the coefficients and related values ​​need to be calculated, which is related to the probability h a (i, j) are closely related, and the maximum value among the thresholds is found by calculation as the optimal threshold. In step S5, the specific implementation steps are as follows:

[0144] S51, dividing the two-dimensional histogram into two parts i≤t and i>t in the oblique coordinate system;

[0145] S52. According to the probability h a (i, j) calculates the first coefficient f1, and the calculation formula is as follows:

[0146]

[0147] Among them, f1 represents the first coefficient, h a (i) represents the probability with abscissa i, and t represents the number of probabilities;

[0148] S53. According to the probability h a (i, j) calculates the second coefficient f2, and the calculation formula is as follows:

[0149]

[0150] in, It represents the probability quantity, which is the value converted from the two-dimensional histogram with the width and height of L-1 under the oblique coordinate system;

[0151] S54. According to the probability h a (i, j) and probability point P b The first correlation value β1 is calculated by the abscissa value i of (i, j), and the calculation formula is as follows:

[0152]

[0153] Among them, i represents the probability point P b The horizontal coordinate value of (i, j);

[0154] S55. According to the probability h a (i, j) and probability point P b The second correlation value β2 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows:

[0155]

[0156] Wherein, β2 represents the second correlation value;

[0157] S56. According to the probability h a (i, j) and probability point P b The third correlation value β3 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows:

[0158]

[0159] Wherein, β3 represents the third correlation value;

[0160] S57, calculating the selected value D(i) according to the first coefficient f1, the second coefficient f2, the first correlation value β1, the second correlation value β2 and the third correlation value β3, the calculation formula is as follows:

[0161] D(i)=f1(β1-β3) 2 +f2(β2-β3) 2

[0162] Where D(i) represents the selected value;

[0163] S58, judging whether the selected value D(i) is the optimal threshold;

[0164] If D(i) is the maximum value, the process ends and the independent variable i is used as the optimal threshold value t, and the process proceeds to step S6;

[0165] If D(i) is not the maximum value, repeat step S58 until the maximum value of D(i) is found, and determine the maximum value among the selected values ​​D(i) through a sorting algorithm. This method can quickly find the maximum value and determine it as the optimal threshold t.

[0166] S6. According to the optimal threshold value t, the two-dimensional histogram is converted into a binary image, and the binary image is segmented to obtain a segmented image. The image is segmented by the segmentation function using the optimal threshold value t. In conventional methods, the most critical step for image recognition and quality inspection of alumina production is image segmentation. Segmentation by oblique cutting coordinate system can not only make the recognition result more accurate, but also optimize the denoising effect to a certain extent. In step S6, the specific implementation steps are as follows:

[0167] S61, obtain the segmentation function G according to the optimal threshold t a , the partition function G a The function formula is as follows:

[0168]

[0169] Among them, G a represents the segmentation function, j represents the probability point P b The ordinate of (i, j), t represents the optimal threshold, originally used Calculation, but based on actual operating experience, it is found that using It can better optimize the tiny noise in the image while segmenting the image.

[0170] S62, randomly select a pixel point M in the two-dimensional histogram a (c, d), according to the partition function G a Determine pixel M a Pixel values ​​of (c, d);

[0171] like Then the pixel value is 0 and ends;

[0172] like Then the pixel value is 1 and ends;

[0173] S63, repeating step S62 to traverse all pixels in the two-dimensional histogram to obtain a binary image;

[0174] S64, segmenting the binary image along the edge of the binary image and generating a segmented image. Obvious boundary areas can be seen in the binary image. Segmenting the image does not mean splitting the image, but dividing the image area to facilitate subsequent steps.

[0175] S7, analyzing the segmented image to obtain the quality status and sending it to the quality control end. The segmented image is further analyzed to know the flame conditions in the alumina production process, and then judge the quality status of the alumina production. In step S7, the specific implementation steps are as follows:

[0176] S71. Obtaining a segmentation function G in the segmented image aPixels with a value of 0 are used to generate a zero-value image, and the number Z of pixels in the zero-value image is counted. According to the number of pixels in the zero-value image, it is equivalent to the area size of the zero-value image.

[0177] S72. Set a threshold interval [k, l] based on the historical data of alumina production, and determine whether the quality status of the zero-value image is good.

[0178] If k ≥ Z ≥ l, the quality status is good and the quality status is sent to the quality control end.

[0179] If Z > k or Z < l, the quality status is abnormal and the quality status is sent to the quality control end. The quality status can enable the quality control end to operate more accurately. The staff at the quality control end can judge the next step of alumina production based on the quality status to ensure the production quality of alumina.

[0180] Please refer to Figure 2 , which shows the structural block diagram of the intelligent quality control system for alumina production provided by this embodiment. The quality control system includes an image conversion module, a preprocessing module, a two-dimensional processing module, a probability analysis module, a threshold analysis module, a segmented image module, and a quality analysis module.

[0181] The image conversion module acquires the original image of the target to be detected and converts the original image into a pixel image; the preprocessing module preprocesses the pixel image to obtain a denoised image; the two-dimensional processing module converts the denoised image into a two-dimensional histogram and establishes a two-dimensional coordinate system; the probability analysis module converts the two-dimensional coordinate system of the two-dimensional histogram into an oblique cutting coordinate system according to the oblique cutting method, and obtains the probability h b (i, j) based on the probability points P a (i, j) of the oblique cutting coordinate system; the threshold analysis module calculates the threshold selection function D(i) according to the probability h a (i), j), and obtains the optimal threshold t according to the threshold selection function D(i); the segmented image module converts the two-dimensional histogram into a binary image according to the optimal threshold t and segments the binary image to obtain a segmented image; the quality analysis module analyzes the segmented image to obtain the quality status and sends it to the quality control end.

[0182] Those of ordinary skill in the art can understand that all or part of the steps in the method of the above embodiments can be completed by instructing relevant hardware through a program. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0183] Each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the above embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0184] The above implementation methods have been described in detail. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. An intelligent quality control method for aluminum oxide production based on image recognition, characterized in that: The method comprises the following steps: S1, obtaining an original image of the target to be detected, and converting the original image into a pixel image; S2, preprocessing the pixel image to obtain a denoised image; S3, converting the denoised image into a two-dimensional histogram and establishing a two-dimensional coordinate system; S4, convert the two-dimensional coordinate system of the two-dimensional histogram into an oblique cut coordinate system according to the oblique cut method, and according to the probability point P of the oblique cut coordinate system b (i, j) gets the probability h a (i, j); S5. According to the probability h a (i, j) calculate the threshold selection function D(i), and obtain the optimal threshold t according to the threshold selection function D(i); S6, converting the two-dimensional histogram into a binary image according to the optimal threshold t, and segmenting the binary image to obtain a segmented image; S7. Analyze the segmented image to obtain the quality status and send it to the quality control end.

2. The quality control method according to claim 1, characterized in that: In step S1, the specific implementation steps are as follows: S11, input the original image into the RGB channel and convert it into an RGB image; S12, determine the brightness L of any pixel in the RGB image a And adjustment coefficient A, calculate the brightness L of the pixel a The low brightness value L min , the calculation formula is as follows: L min =A×L a Among them, L min Indicates the low brightness value of the pixel, A indicates the adjustment coefficient and 30%≤A≤70%, L a Indicates the brightness of a pixel in an RGB image; S13, repeat step S12 until the brightness of all pixels in the RGB image is adjusted to generate a pixel image.

3. The quality control method according to claim 1, characterized in that: In step S2, the specific implementation steps are as follows: S21. Determine the threshold function sgt(b) based on the wavelet transform formula, and its function expression is: Wherein, sgt(b) represents the threshold function, and b represents the independent variable of the threshold function sgt(b); S22, setting the original wavelet coefficient w based on the historical data of the wavelet function y , set the threshold T based on historical data, and calculate the optimized wavelet coefficient w according to the threshold function sgt(b) a , the calculation formula is as follows: Among them, w a represents the optimized wavelet coefficients, w y represents the original wavelet coefficient, T represents the threshold; S23, according to the optimized wavelet coefficient w a The iterative decomposition method is used to obtain the wavelet coefficient w of the minimum layer min ; S24, according to the wavelet coefficient w of the minimum layer min Calculate the standard deviation noise value σ, the calculation formula is as follows: Where σ represents the standard deviation noise value, median(w min ) represents the median value of the wavelet coefficients of the smallest layer; S25, using a peak signal-to-noise ratio method to obtain a signal-to-noise ratio of the pixel image, and determining a decomposition scale B according to the signal-to-noise ratio; S26, obtain the width W and height H of the pixel image, and calculate the threshold function K according to the decomposition scale B c , the calculation formula is as follows: Among them, K c represents the threshold function, σ represents the standard deviation noise value, and B represents the decomposition scale; S27, set the denoising threshold H based on the denoising historical data, and c Determine whether the pixel should be removed; If K c ≥H, the pixel is a noise point and is removed; If K c <H, then the pixel is a non-noise point and is retained; S28, repeat step S27 to traverse all pixel points of the pixel image to obtain a denoised image.

4. The quality control method according to claim 1, characterized in that: In step S3, the specific implementation steps are as follows: S31, converting the denoised image into an HSV image through the HSV color space; S32, creating a canvas to be drawn, and using the OpenCV algorithm to convert the HSV image into a two-dimensional histogram; S33. Select a point O on the two-dimensional histogram a As the coordinate origin and establish a two-dimensional coordinate system.

5. The quality control method according to claim 1, characterized in that: In step S4, the specific implementation steps are as follows: S41, setting the width and height of the two-dimensional histogram to L-1; S42, converting the x-axis and y-axis in the two-dimensional coordinate system into the x'-axis and y'-axis of the oblique coordinate system, the conversion formula is as follows: Wherein, x' represents the transformed oblique cut horizontal axis, and y' represents the transformed oblique cut vertical axis; S43, obtaining the total number of pixels U according to the two-dimensional histogram a , create probability point P in the oblique coordinate system b (i, j); S44, obtain probability point P according to the two-dimensional histogram b The frequency of occurrence of (i, j) is E ij , and calculate the probability point P b The probability of (i, j) appearing h a (i, j), the calculation formula is as follows: Among them, h a (i, j) represents the probability point P b The probability of (i, j) occurring and i, E ij Represents the probability point P b The frequency of occurrence of (i, j), U a Represents the total number of pixels in a two-dimensional histogram.

6. The quality control method according to claim 1, characterized in that: In step S5, the specific implementation steps are as follows: S51, dividing the two-dimensional histogram into two parts i≤t and i>t in the oblique coordinate system; S52. According to the probability h a (i, j) calculates the first coefficient f1, and the calculation formula is as follows: Among them, f1 represents the first coefficient, h a (i) represents the probability with abscissa i, and t represents the number of probabilities; S53, according to the probability h a (i, j) calculates the second coefficient f2, and the calculation formula is as follows: in, represents the probability quantity; S54. According to the probability h a (i, j) and probability point P b The first correlation value β1 is calculated by the abscissa value i of (i, j), and the calculation formula is as follows: Among them, i represents the probability point P b The horizontal coordinate value of (i, j); S55. According to the probability h a (i, j) and probability point P b The second correlation value β2 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows: Wherein, β2 represents the second correlation value; S56. According to the probability h a (i, j) and probability point P b The third correlation value β3 is calculated from the abscissa value i of (i, j), and the calculation formula is as follows: Wherein, β3 represents the third correlation value; S57, calculating the selected value D(i) according to the first coefficient f1, the second coefficient f2, the first correlation value β1, the second correlation value β2 and the third correlation value β3, the calculation formula is as follows: D(i)=f1(β1-β3) 2 +f2(β2-β3) 2 Where D(i) represents the selected value; S58, judging whether the selected value D(i) is the optimal threshold; If D(i) is the maximum value, the process ends and the independent variable i is used as the optimal threshold value t, and the process proceeds to step S6; If D(i) is not the maximum value, step S58 is repeated until the maximum value of D(i) is found.

7. The quality control method according to claim 1, characterized in that: In step S6, the specific implementation steps are as follows: S61, obtain the segmentation function G according to the optimal threshold t a ; S62, randomly select a pixel point M in the two-dimensional histogram a (c, d), according to the partition function G a Determine pixel M a Pixel values ​​of (c, d); like Then the pixel value is 0 and ends; like Then the pixel value is 1 and ends; S63, repeating step S62 to traverse all pixels in the two-dimensional histogram to obtain a binary image; S64, segmenting the binary image along the edge of the binary image and generating a segmented image.

8. The quality control method according to claim 1, characterized in that: The segmentation function G a The function formula is as follows: Among them, G a represents the segmentation function, j represents the probability point P b The ordinate of (i, j), t represents the optimal threshold.

9. The quality control method according to claim 1, characterized in that: In step S7, the specific implementation steps are as follows: S71. Obtaining a segmentation function G in the segmented image a Count the number of pixels Z in the zero-value image. S72, setting a threshold interval [k, l] based on historical data of aluminum oxide production to determine whether the quality status of the zero-value image is good; If k ≥ Z ≥ l, the quality status is good and the quality status is sent to the quality control end; If Z > k or Z < l, the quality status is abnormal and the quality status is sent to the quality control terminal.

10. A system comprising the intelligent quality control method for aluminum oxide production according to any one of claims 1 to 9, characterized in that: The system includes: An image conversion module, which acquires the original image of the target to be detected and converts the original image into a pixel image; A preprocessing module, which preprocesses the pixel image to obtain a denoised image; A two-dimensional processing module, which converts the denoised image into a two-dimensional histogram and establishes a two-dimensional coordinate system; The probability analysis module converts the two-dimensional coordinate system of the two-dimensional histogram into the oblique cut coordinate system according to the oblique cut method, and b (i, j) gets the probability h a (i, j); Threshold analysis module, according to the probability h a (i, j) calculate the threshold selection function D(i), and obtain the optimal threshold t according to the threshold selection function D(i); An image segmentation module, which converts the two-dimensional histogram into a binary image according to the optimal threshold t, and segments the binary image to obtain a segmented image; A quality analysis module, which analyzes the segmented image to obtain the quality status and sends it to the quality control terminal.

Images