CN-121982457-A - Evaluation method for coke quality stability

CN121982457ACN 121982457 ACN121982457 ACN 121982457ACN-121982457-A

Abstract

The invention relates to the technical field of image processing and discloses a coke quality stability evaluation method which comprises the following steps of S101, S102, S103, S104, S105, S106, and S106, wherein the normalized angle frame sequence is obtained, the intensity of each pixel is subjected to weighted least square fitting along with a polarization angle change curve, the primary amplitude ratio of the polarization double harmonic of the pixel is obtained, the primary amplitude ratio of the polarization double harmonic of the view field is obtained, the stability distance is obtained, and the polarization stability index is obtained. According to the invention, gain drift influence is eliminated through angle frame spatial mean normalization, a harmonic projection layer adapts non-equiangular or unfilled-angle sampling through weighted least square, and generates reliability weight to eliminate abnormal angle frame interference, and visual indexes are output by combining effective domain screening, distribution comparison and quantization batch difference and index mapping, so that anisotropy and stability characteristics of coke are accurately reflected.

Inventors

WU JIANZHONG
ZHANG QIANG
XUE HAIJUN
GUO JIPING
MA JUN
ZHANG PENG
ZHANG JIE
JIA WENTAO
Zou Nengsheng
HAO JUAN

Assignees

乌海市广纳煤焦化有限公司

Dates

Publication Date: 20260505
Application Date: 20251203

Claims (10)

1. The method for evaluating the quality stability of the coke is characterized by comprising the following steps of: step S101, acquiring an angle frame sequence under the condition of reflection polarization microscopy, calculating an angle frame spatial average value of each angle frame, and normalizing the pixel intensity of the frame according to the angle frame spatial average value to obtain a normalized angle frame sequence; Step S102, inputting the normalized angle frame sequence into a neural network comprising a harmonic projection layer, wherein the harmonic projection layer carries out weighted least square fitting on the intensity of each pixel along with a polarization angle change curve, and outputs an angle average intensity, a double angle cosine coefficient and a double angle sine coefficient; Step S103, calculating and obtaining a double-harmonic synthesized amplitude according to the double-angle cosine coefficient and the double-angle sine coefficient, taking the ratio of the double-harmonic synthesized amplitude to the average intensity of angles as a pixel-level polarization double-harmonic normalization-amplitude ratio, determining an effective domain by taking a numerical stability constant as a lower limit, and only reserving the pixel-level polarization double-harmonic normalization-amplitude ratio in the effective domain; step S104, obtaining a field-of-view polarization double-harmonic normalization ratio for the median of the pixel-level polarization double-harmonic normalization ratio in the effective domain, and collecting multiple fields of view to construct a batch sample set; step S105, determining experience distribution according to a batch sample set, comparing the experience distribution with the experience distribution of a reference sample set determined at one time to obtain a first-order distribution carrying distance, and combining the median of the reference sample set to obtain a stability distance; And S106, applying an index mapping function to the stability distance to obtain a polarization stability index as a result of coke quality stability.
2. The method for evaluating coke quality stability according to claim 1 is characterized by establishing a polarization angle set comprising at least eight polarization angles, collecting angle frames corresponding to each polarization angle respectively under the condition of reflection polarization microscopy to form an angle frame sequence, wherein each angle frame comprises pixel intensities of all pixels under the polarization angle, obtaining an angle frame spatial average value corresponding to the angle frame by calculating an average value of pixel intensities of all pixels in the angle frame, dividing the pixel intensities of all pixels in each angle frame by the angle frame spatial average value corresponding to the angle frame respectively to obtain normalized angle frames, and forming the normalized angle frame sequence by all normalized angle frames.
3. The method for evaluating coke quality stability according to claim 1, wherein the normalized angle frame sequence and the polarized angle set are input into a neural network including a harmonic projection layer, the neural network generates a reliability weight for each polarized angle at each pixel, and the reliability weight has a value ranging from 0 to 1.
4. The method for evaluating quality stability of coke according to claim 1, wherein a design matrix, an observation vector and a weight matrix are constructed for each pixel, wherein the design matrix comprises a constant term, a cosine value and a sine value of each polarization angle double angle, the observation vector comprises normalized intensity of the pixel under each polarization angle, the weight matrix is a diagonal matrix, diagonal elements are reliability weights corresponding to each polarization angle at the pixel, and the angular average intensity, the double angle cosine coefficient and the double angle sine coefficient of the pixel are obtained through weighted least square calculation.
5. The method for evaluating coke quality stability according to claim 1, wherein in the neural network training stage, two constraints are applied, namely a harmonic reconstruction constraint, so that the weighted error between the normalized intensity of each pixel under each polarization angle and the intensity obtained by fitting an angle average intensity, a double angle cosine coefficient and a double angle sine coefficient is minimum, the fitting only comprises a constant term and a double angle harmonic term, and a gain invariance constraint, wherein after integral intensity scaling is applied to the normalized angle frame sequence, the ratio of the double angle modulation amplitude obtained by the same harmonic projection layer to the angle average intensity is kept consistent with the corresponding ratio before scaling.
6. The method for evaluating coke quality stability according to claim 1, wherein square operation is performed on the double angle cosine coefficient of each pixel, square operation is performed on the double angle sine coefficient of the pixel, square root is taken after summation of two square operation results, and double harmonic synthesized amplitude of the pixel is obtained; dividing the double-harmonic synthesized amplitude of each pixel by the angle average intensity of the pixel to obtain the pixel-level polarization double-harmonic normalized amplitude ratio of the pixel; Setting a numerical stability constant, screening out pixels with the angle average intensity larger than the numerical stability constant in a view field pixel set to form an effective domain, outputting a pixel-level polarization double-harmonic normalization amplitude ratio corresponding to all pixels in the effective domain, and simultaneously keeping the effective domain and the view field pixel set, wherein the view field pixel set represents a set formed by all pixels participating in calculation in the same view field, and corresponds to a region corresponding to the view field pixel set under the condition of reflection polarization microscopy.
7. The method for evaluating coke quality stability according to claim 1 is characterized by comprising the steps of performing median calculation on all pixel-level polarization double-harmonic normalization magnitude ratios in an effective domain to obtain field-level polarization double-harmonic normalization magnitude ratios corresponding to the field of view, determining all fields of view contained in a batch to be evaluated, collecting field-level polarization double-harmonic normalization magnitude ratios corresponding to each field of view, and sorting the field-level polarization double-harmonic normalization magnitude ratios to form a batch sample set.
8. The method for evaluating the quality stability of coke according to claim 1, wherein the first-order distribution conveying distance is obtained by calculating the integral of the absolute difference of the corresponding split values of the two empirical split functions over the split horizontal section with the median of the reference sample set as a dimensionless factor.
9. The method for evaluating the quality stability of coke according to claim 1, wherein the number of discrete quantiles is set, quantile level values distributed equidistantly are taken in a quantile level interval, absolute differences of corresponding quantile values of two empirical quantile functions at each quantile level value are calculated, all the absolute differences are summed, and then the sum is divided by the product of the number of discrete quantiles and the median of a reference sample set, so that a discrete approximation result of the stability distance is obtained.
10. The method for evaluating the quality stability of coke according to claim 1, wherein the stability distance is mapped by a natural exponential function, a negative value of the stability distance is used as an index of the natural exponential function, the value of the mapped result ranges from 0 to 1, and when the stability distance is 0, the mapped result is 1, and the mapped result is used as a polarization stability index.

Description

Evaluation method for coke quality stability Technical Field The invention relates to the technical field of image processing, in particular to a method for evaluating coke quality stability. Background The method comprises the steps of adopting multi-source coal to produce coke in proportion by iron and steel enterprises, acquiring image sequences of the same view field under a plurality of polarization angles in a reflection polarization microscopic mode, establishing quantitative indexes related to coke stability, supporting production quality control, carrying out actual production and detection links across teams and laboratories, ensuring that polarization angle stepping is difficult to keep strictly equal intervals, ensuring that partial angle frames are lost due to operation errors or equipment faults, and simultaneously, generating fine fluctuation of light source output intensity and imaging device response along with angle change or time to form integral gain drift of the angle frames, wherein the objective conditions become the preconditions that cross-batch evaluation must face. In the traditional evaluation method, if angle sampling is uneven or missing, deviation of a sampling strategy is brought into a quantization process based on a simple comparison or conventional Fourier estimation calculation mode, when integral gain drift exists in an angle frame, the intensity amplitude and the average level can be amplified or reduced to different degrees to destroy the stable proportional relation of the intensity amplitude and the average level, and when the two types of influences are simultaneously generated, even if coke in the same coal blending route is detected, systematic difference can occur in quantization parameters which are obtained in different batches and are related to anisotropy, and the real characteristics of the material are deviated. The deviation transmission path from the process to the result is clear, namely, the continuous process is formed by on-site sampling, polishing sample preparation, polarization angle setting, angle frame acquisition and storage to subsequent quantization calculation, wherein uneven angles, unfilled angles or gain drift of any link can be accumulated and solidified into deviation in quantization parameter calculation, the deviation is not derived from coal blending difference or coke quality change, but is mistakenly regarded as a signal of stability difference during cross-batch comparison, stability judgment distortion is directly caused, the evaluation results of different coal blending batches can not be directly compared, and the accuracy of coke production quality control is restricted. Disclosure of Invention The invention provides a method for evaluating coke quality stability, which solves the technical problems in the prior art. The invention provides a coke quality stability evaluation method, which comprises the following steps: step S101, acquiring an angle frame sequence under the condition of reflection polarization microscopy, calculating an angle frame spatial average value of each angle frame, and normalizing the pixel intensity of the frame according to the angle frame spatial average value to obtain a normalized angle frame sequence; Step S102, inputting the normalized angle frame sequence into a neural network comprising a harmonic projection layer, wherein the harmonic projection layer carries out weighted least square fitting on the intensity of each pixel along with a polarization angle change curve, and outputs an angle average intensity, a double angle cosine coefficient and a double angle sine coefficient; Step S103, calculating and obtaining a double-harmonic synthesized amplitude according to the double-angle cosine coefficient and the double-angle sine coefficient, taking the ratio of the double-harmonic synthesized amplitude to the average intensity of angles as a pixel-level polarization double-harmonic normalization-amplitude ratio, determining an effective domain by taking a numerical stability constant as a lower limit, and only reserving the pixel-level polarization double-harmonic normalization-amplitude ratio in the effective domain; step S104, obtaining a field-of-view polarization double-harmonic normalization ratio for the median of the pixel-level polarization double-harmonic normalization ratio in the effective domain, and collecting multiple fields of view to construct a batch sample set; step S105, determining experience distribution according to a batch sample set, comparing the experience distribution with the experience distribution of a reference sample set determined at one time to obtain a first-order distribution carrying distance, and combining the median of the reference sample set to obtain a stability distance; And S106, applying an index mapping function to the stability distance to obtain a polarization stability index as a result of coke quality stability. Further, a polarization angle set c