A method for analyzing the impurity content of desulfurization slurry gypsum particles

Abstract

The application discloses a kind of desulfurization slurry gypsum particle impurity rate analysis methods, comprising: using holographic reconstruction method to the holographic image of desulfurization slurry is locally reconstructed;The cross section of local reconstruction is axially positioned, adopts depth of field expansion method to focus image fusion in the same cross section in particle, obtain depth of field expansion chart;With neural network model identification particle image in depth of field expansion chart, output binary image and particle contour coordinates of depth of field expansion chart;With the neural network model with self-attention mechanism module, according to particle contour coordinates, single particle gray scale chart is intercepted from depth of field expansion chart, then according to gray distribution feature is divided into gypsum particle and impurity particle;The geometric parameter information of gypsum particle and impurity particle is extracted, combined with particle category, calculates the impurity rate of desulfurization slurry gypsum particle.The analysis method provided by the application improves the accuracy of particle identification and classification, and improves the accuracy of the calculation of the impurity rate of desulfurization slurry gypsum particle.

Description

A method for analyzing the impurity content of gypsum particles in desulfurization slurry Technical Field

[0001] This invention relates to the field of multiphase flow measurement, and specifically to a method for analyzing the impurity content of gypsum particles in desulfurization slurry. Background Technology

[0002] Flue gas desulfurization (FGD) is a crucial technology for treating SO2 emissions. Among these technologies, limestone-gypsum wet FGD is the most widely used, holding over 90% of the market share. Its main principle involves counter-current contact between flue gas and circulating limestone slurry within the desulfurization tower, resulting in a chemical reaction that removes sulfur dioxide from the flue gas and generates reusable gypsum, thus improving both environmental protection and economic efficiency.

[0003] The quality of gypsum will greatly affect the economic benefits of flue gas desulfurization gypsum, and the quality of gypsum is determined by factors such as the moisture content, particle size distribution, and impurity content of gypsum.

[0004] Currently, offline analysis remains the primary method for measuring the impurity content of desulfurized gypsum slurry. For example, the chemical titration method specified in the national standard GB / T 7698-2014 requires a series of pretreatment steps such as sampling, filtration, and drying before titrating to determine the specific components and their content. This method is cumbersome and can take several hours or even days. The X-ray diffraction method used in Pan et al. CIESC Journal 66.11 (2015): 4618 requires sending the desulfurized gypsum sample to an X-ray diffractometer for testing. This equipment is bulky and cannot reflect parameter changes in real time. Other methods, such as the infrared spectroscopy method reported in Guan et al. Journal of Central South University 26.12 (2019): 3213-3224 and the drying difference method specified in the national standard GB / T 5484-2012, also suffer from complex operation and poor real-time measurement.

[0005] Therefore, real-time measurement of the gypsum impurity content in wet desulfurization reaction slurry will effectively improve gypsum production efficiency and quality, enhance economic benefits, and increase flue gas desulfurization efficiency, which is of great significance for operators to control reaction conditions. Summary of the Invention

[0006] The purpose of this invention is to provide a method for analyzing the impurity content of gypsum particles in desulfurization slurry, which improves the accuracy of particle identification and classification, as well as the accuracy of calculating the impurity content of gypsum particles in desulfurization slurry.

[0007] This invention provides the following technical solution:

[0008] A method for analyzing the impurity content of gypsum particles in desulfurization slurry, the method comprising:

[0009] (1) Local reconstruction of the holographic image of the desulfurization slurry was performed using the holographic reconstruction method;

[0010] (2) Axially locate the locally reconstructed cross-sectional image, and use the depth-of-field extension method to fuse the particle focusing image into the same cross-sectional image to obtain the depth-of-field extension image;

[0011] (3) Use a neural network model to identify the particle images in the depth-of-field extended map and output the binary image of the depth-of-field extended map and the particle contour coordinates;

[0012] (4) Using a neural network model with a self-attention mechanism module, extract the grayscale image of a single particle from the depth-of-field extension map based on the particle outline coordinates, and then divide the grayscale image of a single particle into gypsum particles and impurity particles according to the grayscale distribution characteristics.

[0013] (5) Extract the geometric parameter information of gypsum particles and impurity particles, and calculate the impurity content of gypsum particles in desulfurization slurry in combination with particle type.

[0014] The analytical method provided by this invention achieves online measurement of the impurity content of gypsum particles in desulfurization slurry by holographic reconstruction and depth extension of digital holograms of gypsum and impurity particles captured in the desulfurization slurry, and then using a neural network model to identify and classify the particles.

[0015] In step (1), local reconstruction is: illuminating the hologram with reconstructed light of the same frequency and wavelength as the reference light to form a light wave similar to the object light wave, and reconstructing the reconstructed cross-sectional image of each axial distance.

[0016] In step (1), the hologram reconstruction method is a scalar diffraction calculation method, which refers to the spatial holographic reconstruction of the object light wave recorded by the holographic image through diffraction theory, and obtains the light intensity of the scattered light of the object at different positions in space, including angular spectrum reconstruction, fractional Fourier transform reconstruction, wavelet transform reconstruction and other methods.

[0017] Preferably, the hologram reconstruction method employs wavelet transform reconstruction. Unlike Fourier transform, wavelet functions are restricted in both the spatial and frequency domains, allowing them to effectively utilize particle interference image information while effectively shielding against noise interference. Therefore, the holographic reconstructed image obtained based on wavelet transform reconstruction has a more uniform background and a better signal-to-noise ratio.

[0018] In step (2), the depth-of-field extension method refers to determining the focused images of objects at different cross sections and fusing these focused images to form a new image; the parameters used as criteria in the depth-of-field extension method include the grayscale of the grain image and the grayscale gradient.

[0019] Furthermore, the depth-of-field extension method can be a local gradient variance method, a local brightness variance method, or a correlation coefficient method, etc.

[0020] More preferably, the depth-of-field extension method uses the local gradient variance method. For the focal plane, the gray-level gradient of the edge portion of the particle will reach an extreme value, resulting in a large gradient variance with the surrounding background area or the internal region of the particle. The gradient variance of the current local region is calculated, and then the maximum value is extracted. This value is then compared with the same region in other images, and the local region image with the largest gradient variance value is selected. These local regions are then combined to obtain the entire focused image.

[0021] In step (3), the neural network model is trained by the following method: labeling the particle images on the depth-of-field extension map as a labeled dataset, training the neural network model, and obtaining the trained neural network model.

[0022] In step (3), the neural network models include FCN, UNet, SegNet, R-CNN, etc. Unlike simple target recognition, they can not only identify the approximate location of particles, but also determine whether all pixels in the image belong to particles or background, thus completing pixel-level recognition of particles.

[0023] Preferably, in step (3), the neural network model is used to perform threshold segmentation of the image, thereby extracting the image region where the particles are located. The UNet neural network model can be used. Its characteristic is that the network structure presents a U-shape, which can simultaneously utilize low-level, high-resolution contour features and high-level, low-resolution semantic features to achieve pixel-level particle segmentation.

[0024] In step (4), the neural network model with the self-attention mechanism module is trained by the following method: providing single-particle grayscale images of gypsum particles and impurity particles as a labeled dataset, training the neural network model with the self-attention mechanism module, and obtaining the trained neural network model with the self-attention mechanism module.

[0025] Furthermore, during the training process, the self-attention mechanism module introduces pre-weights and iteratively updates them to obtain attention weights, which are then applied to the trained neural network model with the self-attention mechanism module.

[0026] In step (4), the neural network model with a self-attention mechanism module can extract image features and perform operations on the image features to determine the image type, thereby enabling particle type identification. In this invention, neural network models with self-attention mechanism modules such as VGG, ResNet, and ResNext can be used. The neural network model with a self-attention mechanism module outputs the particle categories in the depth-of-field extension map, which include two categories: gypsum particles and impurity particles.

[0027] Because gypsum particles have high transparency, while impurity particles, mainly composed of limestone, have very low transparency, their transparency characteristics differ significantly. Therefore, this invention designs a self-attention mechanism module to enable the neural network model to focus on the transparency characteristics of the particles, specifically the grayscale distribution characteristics of the particle's central region. This allows the model to strengthen or weaken the influence of different pixels based on the salience of the transparency / grayscale distribution characteristics, thus more effectively distinguishing different types of particles.

[0028] Preferably, in step (4), a ResNet neural network model with a self-attention mechanism module is used. It has a unique residual network structure, which effectively solves the problem of gradient vanishing during gradient propagation, thereby greatly increasing the number of convolutional layers in the model and enabling it to extract deeper features from the image, thus achieving higher classification accuracy.

[0029] In step (5), the method for calculating the impurity content of gypsum particles in the desulfurization slurry includes:

[0030] Extract the geometric parameters of the particles, including: the major and minor axes of the particles, cross-sectional area, and diameter of the equivalent area circle;

[0031] Based on the shape of the gypsum particles, the gypsum particles are approximated as cylinders or spheres, and their volume is solved by combining the geometric parameters of the gypsum particles.

[0032] Based on the shape of the impurity particles, the impurity particles are approximated as cylinders or spheres, and their volume is solved by combining the geometric parameters of the impurity particles.

[0033] Multiplying the volume and density gives the mass of gypsum particles and impurity particles, respectively. The impurity content is expressed as the ratio of the mass of impurity particles to the total mass of particles.

[0034] The densities of gypsum particles and impurity particles were pre-determined using the XRD method: the desulfurization slurry was analyzed using XRD to determine the density of gypsum particles and the densities of various impurity particle components. The weighted average of the densities of each impurity particle component was then used as the density of the impurity particles. The main component of the impurity particles was limestone.

[0035] The analytical method provided by this invention is based on digital holography technology. It performs holographic reconstruction and depth-of-field expansion on the digital hologram / holographic image of desulfurization slurry, realizing online extraction of particle geometric parameters. In addition, a neural network model is used for particle identification and classification, which solves a series of problems such as complex holographic backgrounds, easy misjudgment of backgrounds as particles, and difficulty in extracting image features by traditional machine learning methods. This improves the accuracy of particle identification and classification, thereby improving the accuracy of calculating the impurity rate of gypsum particles in desulfurization slurry. It overcomes a series of problems that are common in general measurement methods, such as low measurement efficiency, reliance on manual processing, cumbersome measurement operations, and poor real-time performance, and realizes high-precision real-time online measurement of the impurity rate of gypsum particles in desulfurization slurry. Attached Figure Description

[0036] Figure 1 is a flowchart of the holographic image processing of desulfurization slurry.

[0037] Figure 2 is a flowchart of the method for analyzing the impurity content of gypsum particles.

[0038] In Figure 2, the various numbered legends represent: 1. Holographic image of desulfurization slurry; 2. Depth-expanded image after holographic reconstruction and depth-expanding; 3. Binary image after recognition by neural network model; 4. Single-particle grayscale image. Detailed Implementation

[0039] The technical solutions of the embodiments given herein will now be described in detail with reference to the accompanying drawings.

[0040] The flowchart of the method for analyzing the impurity content of gypsum particles in desulfurization slurry provided by the present invention is shown in Figures 1 and 2:

[0041] (1) The desulfurization slurry was photographed using a digital holographic device to obtain a holographic image of the desulfurization slurry 1.

[0042] (2) The holographic reconstruction method is used to locally reconstruct the hologram 1 of the desulfurization slurry to obtain particle reconstruction maps (or locally reconstructed cross-sectional maps) of different sections.

[0043] This embodiment uses the wavelet transform reconstruction method, specifically:

[0044] First, use wavelet basis functions. To construct the wavelet function:

[0045] (1)

[0046] In the formula, x and y are coordinates, and α is a scale parameter. The expression is:

[0047] (2)

[0048] In the formula, λ is the laser wavelength and z is the reconstructed depth.

[0049] Since the wavelet function only exhibits significant fluctuations away from the horizontal axis near its center point and has a mean of 0, a Gaussian window function and variables are needed. :

[0050] (3)

[0051] In the formula, σ is the bandwidth factor, which is related to the sampling of the holographic system.

[0052] Wavelet transform can be used to characterize optical diffraction and holographic recording processes. The light intensity of the reconstructed image is represented as:

[0053] (4)

[0054] (3) Depth extension of particle reconstruction images of different sections: Axial positioning is performed to find the focal position of different particles, and the depth extension method is used to fuse them into a depth extension image 2 after holographic reconstruction and depth extension.

[0055] This embodiment uses the local gradient variance method for depth-of-field extension processing, specifically:

[0056] (3-1) Calculate the pixel gradient for a local region. The Sobel operator is typically used to calculate the gradients in the x and y directions. , :

[0057] (5)

[0058] (3-2) Find the gradient matrix of the local region, where m and n are pixel indices:

[0059] (6)

[0060] (3-3) Find the mean gradient of the local region :

[0061] (7)

[0062] In the formula, M and N are the number of pixels in the x and y directions of the image, respectively.

[0063] (3-4) Calculate the local gradient variance:

[0064] (8)

[0065] (3-5) Calculate for each image, then compare the corresponding local regions, and use the maximum gradient variance as the criterion to merge the particle focus maps into the same image.

[0066] (4) Use the trained neural network model to identify particles from the depth extension map 2 and obtain the binary image of the particles, i.e., the binary image 3 after identification by the neural network model; output the binary image of the depth extension map and the particle outline coordinates.

[0067] This embodiment uses the UNet neural network model for particle recognition. Manually labeled binarized particle images are used as the dataset to train the model. The input depth-of-field extended image (2) is processed by the model, and the output is a binarized particle image used to extract geometric parameters.

[0068] (5) Using a neural network model with a self-attention mechanism module, extract the grayscale image of a single particle from the depth-of-field extension map based on the particle outline coordinates, and then divide the grayscale image of a single particle into gypsum particles and impurity particles according to the grayscale distribution characteristics.

[0069] This embodiment uses the ResNet neural network model for particle classification. The model is trained using single-particle grayscale images of gypsum and impurities as a labeled dataset. Single-particle grayscale images are extracted from the depth-of-field extension map (4), achieving image thresholding and extraction of the image region containing the particles. The model's self-attention mechanism module weights each pixel of the particle image, focusing on the grayscale distribution characteristics of the central region of the particle image, making the differences in transparency between different types of particles more significant, and using transparency differences to distinguish between gypsum particles and impurity particles.

[0070] (6) Extract the geometric parameters and category information of the particles and calculate the volume and mass of the particles: Based on the morphological characteristics of the gypsum particle image, the gypsum particles are regarded as cylinders or spheres, and their density is taken as the density of gypsum; Based on the morphological characteristics of the impurity particle image, the impurity particles are regarded as cylinders or spheres, and their main component is limestone. Their density is obtained in advance by XRD analysis; The particle volume and density are multiplied to obtain the mass of gypsum and impurity particles respectively. The impurity content is expressed as the ratio of the mass of impurity particles to the total mass of particles.

[0071] The method proposed in this invention uses digital holography to reconstruct and extend the depth of field of the digital hologram of desulfurization slurry, realizing online particle measurement; at the same time, it uses a neural network model to identify and classify particles, improving the accuracy of particle identification and classification in the complex background of the hologram, and further realizing real-time and accurate measurement of the impurity content of gypsum particles in desulfurization slurry.

Claims

1. A method for analyzing the impurity content of gypsum particles in desulfurization slurry, characterized in that, The method includes: (1) performing local reconstruction of the holographic image of the desulfurization slurry using a holographic reconstruction method; (2) performing axial positioning of the locally reconstructed cross-sectional image, and using a depth-of-field extension method to fuse the particle focusing images into the same cross-sectional image to obtain a depth-of-field extension image; (3) using a neural network model to identify the particle images in the depth-of-field extension image, and outputting the binary image of the depth-of-field extension image and the particle contour coordinates; (4) using a neural network model with a self-attention mechanism module to extract the single particle grayscale image from the depth-of-field extension image based on the particle contour coordinates, and then dividing the single particle grayscale image into gypsum according to the grayscale distribution characteristics. Particles and impurity particles; In step (4), the neural network model with self-attention mechanism module is trained by the following method: providing single-particle grayscale images of gypsum particles and impurity particles as labeled datasets, training the neural network model with self-attention mechanism module, and obtaining the trained neural network model with self-attention mechanism module; In step (4), a ResNet neural network model with self-attention mechanism module is used; (5) Extracting the geometric parameter information of gypsum particles and impurity particles, and calculating the impurity rate of gypsum particles in desulfurization slurry in combination with particle category.

2. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, In step (1), the holographic reconstruction method is a scalar diffraction calculation method, selected from angular spectrum reconstruction, fractional Fourier transform reconstruction or wavelet transform reconstruction.

3. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, In step (2), the parameters used as criteria in the depth-of-field extension method include the gray level of the particle image and the gray level gradient.

4. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, In step (3), the neural network model is trained by the following method: labeling particle images on the depth-of-field extension map as a labeled dataset, training the neural network model, and obtaining the trained neural network model.

5. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, The UNet neural network model is used in step (3).

6. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, During training, the self-attention mechanism module introduces pre-weights and iteratively updates them to obtain attention weights, which are then applied to the trained neural network model with the self-attention mechanism module.

7. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 1, characterized in that, In step (5), the method for calculating the impurity content of gypsum particles in desulfurization slurry includes: extracting the geometric parameter information of the particles, including: the major and minor axes, cross-sectional area, and diameter of the equivalent area circle of the particles; approximating the gypsum particles as prisms or spheres based on their shape, and solving for the volume of the gypsum particles by combining the geometric parameters of the gypsum particles; approximating the impurity particles as prisms or spheres based on their shape, and solving for the volume of the impurity particles by combining the geometric parameters of the impurity particles; multiplying the volume and density to obtain the mass of the gypsum particles and the mass of the impurity particles, respectively, and the impurity content is expressed as the ratio of the mass of the impurity particles to the mass of the total particles.

8. The method for analyzing the impurity content of gypsum particles in desulfurization slurry according to claim 7, characterized in that, The density of gypsum particles and impurity particles in the desulfurization slurry was analyzed by XRD method; the weighted average of the densities of various impurity particle components was taken as the density of the impurity particles.

Images