Online detection method, system and equipment for particle size distribution of pulverized coal

Abstract

The invention relates to an on-line detection method for particle size distribution of pulverized coal. The method comprises the following steps: collecting a pulverized coal image; preprocessing the pulverized coal image; constructing a Mask R-CNN network model, inputting the preprocessed pulverized coal image into the Mask R-CNN network model, and executing the training of the Mask R-CNN network model; and detecting the preprocessed to-be-detected pulverized coal image by adopting the trained Mask R-CNN network model to obtain the number of pixel points corresponding to each pulverized coal particle, and calculating the particle size of each pulverized coal particle and the average particle size of the pulverized coal particles. The method comprises the following steps: acquiring a pulverized coal image of pulverized coal falling at an outlet of a pulverized coal silo in real time through a high-speed camera, preprocessing the pulverized coal image to obtain an effective screenshot of the pulverized coal, calculating the particle size of each pulverized coal particle and the average particle size of the pulverized coal particles through a Mask R-CNN network model, and obtaining the distribution condition of the particle size of the pulverized coal. The particle size of the pulverized coal can be detected on line in real time, and the operation efficiency and stability of a CFB unit are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of coal powder particle size detection, and in particular relates to an online detection method, system and equipment for coal powder particle size distribution. Background Art

[0002] As an advanced combustion technology, circulating fluidized bed (CFB) boilers are widely used in energy production and environmental protection. Due to their ability to adapt to a variety of fuels (such as coal, waste, biomass, etc.) and low emission characteristics, CFB boilers are increasingly used in electricity, thermal energy and industrial processes. However, the performance of CFB boilers and their combustion efficiency are closely related to the particle size of the fuel. The particle size not only affects the combustion speed and heat release rate of the fuel, but is also directly related to the combustion stability of the boiler and the emission level of pollutants.

[0003] Traditional particle size detection methods usually rely on laboratory sampling and analysis, involving multiple steps such as sampling, screening, and measurement. This method has problems such as sampling errors, time delays, and the inability to achieve real-time monitoring, which makes it difficult to meet the needs of modern CFB boilers for dynamic monitoring of fuel particle size. Especially in a continuously operating industrial environment, real-time, online fuel particle size monitoring is particularly important.

[0004] In addition, changes in fuel particle size may lead to unstable boiler operating conditions, which in turn affects the thermal efficiency and safety of the boiler. Summary of the invention

[0005] The purpose of the present invention is to provide a method, system and device for online detection of coal powder particle size distribution, which can realize online and real-time detection of coal powder particle size, improve the operating efficiency and stability of CFB units, help operators to adjust combustion strategies in time, optimize fuel utilization efficiency, and reduce emissions of environmental pollutants;

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for online detection of coal powder particle size distribution comprises the following steps:

[0008] Collect coal powder images;

[0009] Preprocessing the coal powder image;

[0010] Construct a Mask R-CNN network model, input the preprocessed coal powder image into the Mask R-CNN network model, and perform training of the Mask R-CNN network model;

[0011] The trained Mask R-CNN network model is used to detect the preprocessed coal powder image to obtain the number of pixels corresponding to each coal powder particle, and the particle size of each coal powder particle and the average particle size of the coal powder particles are calculated;

[0012] And, foreign matter detection is performed on the coal powder image to be detected.

[0013] Preferably, the pulverized coal image is a pulverized coal image captured when the pulverized coal falls into a pulverized coal conveyor belt at an outlet of a pulverized coal silo of a boiler.

[0014] Preferably, the method further comprises the following steps:

[0015] If the difference between the first average particle size of coal powder obtained by the Mask R-CNN network model and the second average particle size of coal powder obtained according to the boiler furnace pressure is greater than the threshold value for m consecutive times, the sampled coal powder particles are subjected to off-line detection to obtain the third average particle size of coal powder particles;

[0016] If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is less than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

[0017] Preferably, the method further comprises the following steps:

[0018] If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is also greater than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

[0019] Preferably, the method further comprises the following steps:

[0020] If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is also greater than the threshold, then a search for the historical average particle size of the coal powder particles is performed; if the difference between the historical average particle size of the coal powder particles and the third average particle size of the coal powder particles is less than the threshold, the coal powder image collected at this time is preprocessed, and the Mask R-CNN network model corresponding to the historical average particle size of the coal powder particles is used for detection and calculation; if the difference between the average particle size of the coal powder particles calculated by detection and the third average particle size of the coal powder particles is less than the threshold, the Mask R-CNN network model corresponding to the historical average particle size of the coal powder particles is used to detect the coal powder image to be detected after subsequent preprocessing; if the difference between the average particle size of the coal powder particles calculated by detection and the third average particle size of the coal powder particles is greater than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

[0021] Preferably, the preprocessing of the coal powder image also includes the following steps: converting the coal powder image into a grayscale image using a weighted average method; calculating the global optimal threshold of the grayscale image using an Otsu's algorithm; binarizing the grayscale image according to the global optimal threshold to obtain a binary image; and taking a screenshot of M×N pixels of the coal powder area in the binary image.

[0022] Preferably, the foreign body detection of the coal powder image to be detected includes the following steps: converting the coal powder image to be detected into a grayscale image; denoising the grayscale image using Gaussian filtering; then performing Fourier transform on the denoised grayscale image to obtain a spectrum diagram, and performing foreign body detection based on the spectrum diagram.

[0023] The present invention also provides an online detection system for coal powder particle size distribution, comprising:

[0024] An image acquisition module is used to acquire images of pulverized coal. The image acquisition module is installed between the outlet of the pulverized coal silo of the boiler and the pulverized coal conveyor belt. The image acquisition module acquires images of pulverized coal at the outlet of the pulverized coal silo of the boiler when the pulverized coal falls into the pulverized coal conveyor belt.

[0025] A preprocessing module, used for preprocessing the coal powder image;

[0026] A training module is used to build a Mask R-CNN network model, input the preprocessed coal powder image into the Mask R-CNN network model, and perform training of the Mask R-CNN network model;

[0027] The particle size detection and calculation module is used to detect the preprocessed coal powder image to be detected using the trained Mask R-CNN network model, obtain the number of pixels corresponding to each coal powder particle, and calculate the particle size of each coal powder particle and the average particle size of the coal powder particles;

[0028] And, a foreign matter detection module is used to perform foreign matter detection on the coal powder image to be detected.

[0029] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned on-line detection method for coal powder particle size distribution are executed.

[0030] The present invention also provides an online detection device for coal powder particle size distribution, comprising:

[0031] Memory, for storing software applications,

[0032] The processor is used to execute the software application, and each program of the software application correspondingly executes the steps in the above-mentioned coal powder particle size distribution online detection method.

[0033] The present invention uses a high-speed camera to collect coal powder images dropped at the outlet of the coal powder silo in real time, pre-processes the coal powder images, obtains effective screenshots of the coal powder, and uses a Mask R-CNN network model to obtain the number of pixel points corresponding to each coal powder particle in the effective screenshot of the coal powder, calculates the particle size of each coal powder particle and the average particle size of the coal powder particles, obtains the distribution of the coal powder particle size, and then uses a foreign matter detection module to detect abnormal objects. Finally, the results are uploaded to a control center to assist CFB boiler combustion, thereby improving the operating efficiency and stability of the CFB unit.

[0034] By comparing the third average particle size of coal powder particles with the first average particle size of coal powder particles and the second average particle size of coal powder particles, the present invention can effectively determine whether an error occurs in the network model or the coal powder quality changes. When the coal powder quality changes and the current network model is no longer applicable, the old model is saved and a historical search is performed based on the non-online detection results. If there are historical results that are similar to the non-online detection results, the old model is used to calculate and determine whether it is appropriate. If it is appropriate, the old model is not trained and the detection is continued using the old model. If it is not appropriate, a new model is trained to replace the old model. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a schematic diagram of the boiler coal powder silo outlet structure;

[0036] Figure 2 It is a schematic diagram of the internal structure of the coal powder channel;

[0037] Figure 3 It is a flow chart of the online detection method;

[0038] Figure 4 It is the network model update flow chart. DETAILED DESCRIPTION

[0039] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0040] like Figure 3 As shown, the present invention provides an online detection method for coal powder particle size distribution, comprising the following steps:

[0041] S1, collecting coal powder images;

[0042] Specifically, this embodiment uses a high-speed camera 3 to collect coal powder images, such as Figure 1-2 The high-speed camera 3 is installed between the boiler coal powder silo outlet 1 and the coal powder conveyor belt, and an LED lamp is installed on one side of the high-speed camera 3 to supplement the lighting and ensure that there is no shadow in the shooting; a coal plowing car is installed at the boiler coal powder silo outlet 1, and the coal powder in the silo is pushed out by the coal plowing car. A coal powder channel 2 is provided on the outside of the boiler coal powder silo outlet 1, and the coal powder conveyor belt is at the bottom of the coal powder channel 2. The coal powder pushed out by the coal plowing car passes through the coal powder channel 2 and falls into the coal powder conveyor belt, and the coal powder is sent to the boiler through the coal powder conveyor belt, wherein the high-speed camera 3 is installed on the inner wall of the coal powder channel 2, and the high-speed camera 3 is parallel to the falling direction of the coal powder and vertically faces the coal powder. The high-speed camera 3 collects the coal powder image when the coal powder at the boiler coal powder silo outlet 1 falls into the coal powder conveyor belt. Since the coal powder falls more dispersedly, by installing the high-speed camera 3 in the coal powder channel 2, it is convenient to detect and calculate the coal powder particle size;

[0043] Among them, n images taken by a high-speed camera can be defined as one acquisition.

[0044] S2, preprocessing the coal powder image;

[0045] Specifically, the preprocessing of the coal powder image further includes the following steps:

[0046] S21, converting the color coal powder image into a grayscale image by using a weighted average method;

[0047] More specifically, an RGB color image refers to a color pixel in which each pixel of the image is formed by different combinations of three primary colors (Three Primary Colors) - red (R), green (G), and blue (B). The weighted average method is to perform weighted average on the three primary color components of R, G, and B with corresponding weights, and the obtained value is used as the grayscale value of the obtained grayscale image. The present invention performs weighted average according to formula (1-1) to obtain a more reasonable grayscale value from the perspective of human biology:

[0048] Y=0.3×R+0.59×G+0.11×B(1-1)

[0049] S22, grayscale image coordinate convention,

[0050] More specifically, since the position and angle of the camera are fixed, the pixel size of the captured image is fixed, and the actual volume represented by each pixel is also fixed. The image can be regarded as a two-dimensional plane, with the starting point being the lower left corner of the image (1, 1), where x is the horizontal coordinate and y is the vertical coordinate, (x, y) is the spatial (plane) coordinate and is also a pixel point, so the grayscale of each pixel point can be represented by a function f(x, y).

[0051] S23, using Otsu's algorithm to calculate the global optimal threshold of the grayscale image;

[0052] More specifically, Otsu's method is a threshold segmentation technique widely used in image processing to automatically determine the optimal threshold in an image to segment the image into foreground and background. This method maximizes the between-class variance to find the optimal grayscale threshold so that the image can be divided into two categories: foreground and background;

[0053] S24, binarizing the grayscale image according to the global optimal threshold to obtain a binarized image;

[0054] More specifically, an image in which a pixel has only two colors, white and black, is called a binary image. The binarization of a grayscale image is calculated according to the formula g(x,y)(1-2).

[0055]

[0056] The threshold here is the global optimal threshold obtained by Otsu's algorithm;

[0057] Grayscale images are constructed by adding many color depths between black and white to binary images. This type of image can display all grayscales from the darkest black to the brightest white. Each grayscale level represents a color depth, usually represented by L. In grayscale images, pixels can take values ​​between 0 and L-1. The value of L varies depending on the data type used to store the grayscale value. There may be 256 or 2k values. When k=1, the grayscale image is a binary image.

[0058] S25, taking a screenshot of M×N pixels of the coal powder area in the binary image.

[0059] More specifically, since the camera may have a large shooting range and may capture non-coal powder areas, affecting subsequent detection, the present invention randomly captures a screenshot of M×N pixels in the coal powder area.

[0060] S3, constructing a Mask R-CNN network model, inputting the preprocessed coal powder image into the Mask R-CNN network model, and performing training of the Mask R-CNN network model;

[0061] Specifically, the Mask R-CNN neural network obtains the segmentation mask of each particle in the image, and each pixel value in each mask is 1 or 0, indicating whether the pixel belongs to the particle;

[0062] More specifically, the training of the Mask R-CNN network model includes the following steps:

[0063] A set of preprocessed coal powder sample images captured by the high-speed camera 3 are used in advance to generate image annotations using existing tools (labelme or COCO-annotator), including the category, bounding box and segmentation mask of each object, and a training set and a test set are established according to a ratio of 0.8:0.2;

[0064] Construct the Mask R-CNN network model, use the backbone network (ResNet+FPN) to extract image features, use RPN to generate candidate regions, use ROI Align to align candidate regions, establish classification and bounding box regression branches to determine the target category and location, use the Mask branch to generate instance masks for each target, and finally use the loss function to optimize the network and output detection and segmentation results;

[0065] Use the training set to input images and corresponding labeled data, and perform forward propagation, backpropagation, and optimization; use the test set to evaluate the model.

[0066] S4. Using the trained Mask R-CNN network model to detect the preprocessed coal powder image to be detected, obtain the number of pixels corresponding to each coal powder particle, and calculate the particle size of each coal powder particle and the average particle size of the coal powder particles;

[0067] Specifically, during the detection, the high-speed camera 3 collects the coal powder image to be detected. After the coal powder image to be detected is preprocessed, the captured coal powder image is input into the trained Mask R-CNN neural network model to obtain the number of each coal powder particle in the image and the segmentation mask of each particle; each mask is summed to count the number of pixels of each coal powder particle;

[0068] Count the number of pixels occupied by each particle and convert it into actual area. Use a circle of equal area and use the formula S = π × R 2, calculate the particle size of each particle and the average particle size of the whole, and make statistics on all the results, and upload the statistical results to the control center. The actual area is converted based on the camera's viewing angle and the straight-line distance to the coal powder, and the camera's field of view is calculated to obtain the actual size of the captured image;

[0069] The above processing can be performed on n images collected at one time, and n average particle sizes can be calculated and compared. If there are obvious differences, the screenshot calculation results with differences are removed, and the remaining calculation results are comprehensively counted to obtain the particle size-ratio result and average particle size collected at one time.

[0070] And, S5, performing foreign matter detection on the coal powder image to be detected;

[0071] Specifically, after the coal powder image to be detected is converted into a grayscale image, two copies are made, one for foreign body detection and the other for preprocessing. Gaussian filtering is used to denoise the image for foreign body detection to reduce the interference of subtle noise on frequency domain analysis. Then, Fourier transform is used to convert the coal powder image to be detected from the spatial domain to the frequency domain to obtain a spectrum diagram (showing the composition of different frequencies in the image).

[0072] The Fourier transform is the conversion of the image from the spatial domain to the frequency domain, and the conversion is continued using formula 1-3:

[0073]

[0074] Among them, f(x,y) is the image in the spatial domain, F(u,v) is the image in the frequency domain, W and H are the width and height of the image respectively;

[0075] Since the frequency of coal powder images is usually smooth or regular, and foreign matter often introduces different frequency characteristics, high-pass filtering is used to retain high-frequency components (foreign matter details) and suppress low-frequency components (smooth areas of coal powder); the location of the high-frequency components in the spectrum graph is marked and its coordinates are recorded and uploaded to the control center, and the corresponding coordinates of the original image are marked for early warning.

[0076] like Figure 4 As shown, the following steps are also included:

[0077] S6. If the difference between the first average particle size of the pulverized coal particles obtained by the Mask R-CNN network model and the second average particle size of the pulverized coal particles obtained according to the boiler furnace pressure is greater than the threshold value for m consecutive times, the pulverized coal particles are sampled for off-line detection to obtain a third average particle size of the pulverized coal particles;

[0078] Specifically, when the pulverized coal enters the furnace of the boiler for combustion, the differential pressure inside the furnace is generally detected by the pressure detection module in the furnace of the boiler. The particle size distribution of the fuel currently entering the furnace for combustion can be inferred by the differential pressure inside the furnace. Since the detection is performed after the pulverized coal is burned, it has obvious hysteresis. Therefore, it can be used in the present invention to analyze the accuracy of the particle size distribution detected online.

[0079] More specifically, if the quality of the pulverized coal at the silo outlet changes, the Mask R-CNN network model used in the current particle size calculation may not be suitable for image processing at this time.

[0080] If the difference between the first average particle size of coal powder obtained by the Mask R-CNN network model and the second average particle size of coal powder obtained according to the boiler furnace pressure is greater than the threshold value for m consecutive times, the detection will be suspended but the image collection will continue, and the sampled coal powder particles will be tested for non-online detection to obtain the third average particle size of coal powder particles, where the non-online detection is the screening detection of the sampled transported coal powder, and the threshold is the particle size difference of the coal powder particle size that may affect the boiler;

[0081] If the third average particle size of the coal powder particles is significantly different from the first average particle size of the coal powder particles, that is, the difference is greater than the threshold, and the third average particle size of the coal powder particles is not significantly different from the second average particle size of the coal powder particles, that is, the difference is less than the threshold, then it means that the current coal powder quality has not changed but the model is wrong. At this time, the newly collected coal powder images are used to establish training sets and test sets, and the Mask R-CNN network model continues to be trained, and the new model is used to replace the old model, and the old model is not saved.

[0082] If the third average particle size of the coal powder particles is not significantly different from the first average particle size of the coal powder particles, and the third average particle size of the coal powder particles is significantly different from the second average particle size of the coal powder particles, it means that the coal powder quality has indeed changed but the current model is still applicable, and normal detection is sufficient;

[0083] If the third coal powder particle average particle size is significantly different from the first coal powder particle average particle size, and the third coal powder particle average particle size is also significantly different from the second coal powder particle average particle size, it means that the coal powder quality has changed and the current model is no longer applicable. At this time, save the old model and perform a historical search based on the offline results. If there are historical results that are similar to the offline results, use the old model to calculate and determine whether it is appropriate. If it is appropriate, do not train the new model and use the old model to continue testing. If it is not appropriate, train a new model to replace the old model. By saving the old model, the next time the model needs to be updated, you can consider using the old model, which saves training time and allows the online detection system to recover quickly.

[0084] The present invention also provides an online detection system for coal powder particle size distribution, comprising:

[0085] An image acquisition module is used to acquire images of pulverized coal. The image acquisition module is installed between the boiler pulverized coal silo outlet 1 and the pulverized coal conveyor belt, and the image acquisition module acquires images of pulverized coal at the boiler pulverized coal silo outlet 1 when the pulverized coal falls into the pulverized coal conveyor belt;

[0086] A preprocessing module, used for preprocessing the coal powder image;

[0087] A training module is used to build a Mask R-CNN network model, input the preprocessed coal powder image into the Mask R-CNN network model, and perform training of the Mask R-CNN network model;

[0088] The particle size detection and calculation module is used to detect the preprocessed coal powder image to be detected using the trained Mask R-CNN network model, obtain the number of pixels corresponding to each coal powder particle, and calculate the particle size of each coal powder particle and the average particle size of the coal powder particles;

[0089] And, a foreign matter detection module is used to perform foreign matter detection on the coal powder image to be detected.

[0090] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned on-line detection method for coal powder particle size distribution are executed.

[0091] The present invention also provides an online detection device for coal powder particle size distribution, comprising:

[0092] Memory, for storing software applications,

[0093] The processor is used to execute the software application, and each program of the software application correspondingly executes the steps in the above-mentioned coal powder particle size distribution online detection method.

Claims

1. A method for online detection of coal powder particle size distribution, characterized in that: The following steps are involved: Collect coal powder images; Preprocessing the coal powder image; Construct a Mask R-CNN network model, input the preprocessed coal powder image into the Mask R-CNN network model, and perform training of the Mask R-CNN network model; The trained Mask R-CNN network model is used to detect the preprocessed coal powder image to obtain the number of pixels corresponding to each coal powder particle, and the particle size of each coal powder particle and the average particle size of the coal powder particles are calculated; And, foreign matter detection is performed on the coal powder image to be detected.

2. The method for online detection of coal powder particle size distribution according to claim 1, characterized in that: The pulverized coal image is a pulverized coal image collected when the pulverized coal falls into the pulverized coal conveyor belt at the outlet of the pulverized coal silo of the boiler.

3. The method for online detection of coal powder particle size distribution according to claim 1, characterized in that: The following steps are also included: If the difference between the first average particle size of coal powder obtained by the Mask R-CNN network model and the second average particle size of coal powder obtained according to the boiler furnace pressure is greater than the threshold value for m consecutive times, the sampled coal powder particles are subjected to off-line detection to obtain the third average particle size of coal powder particles; If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is less than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

4. The method for online detection of coal powder particle size distribution according to claim 3, characterized in that: The following steps are also included: If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is also greater than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

5. The method for online detection of coal powder particle size distribution according to claim 3, characterized in that: The following steps are also included: If the difference between the third average particle size of the coal powder particles and the first average particle size of the coal powder particles is greater than the threshold, and the difference between the third average particle size of the coal powder particles and the second average particle size of the coal powder particles is also greater than the threshold, then a search for the historical average particle size of the coal powder particles is performed; if the difference between the historical average particle size of the coal powder particles and the third average particle size of the coal powder particles is less than the threshold, the coal powder image collected at this time is preprocessed, and the Mask R-CNN network model corresponding to the historical average particle size of the coal powder particles is used for detection and calculation; if the difference between the average particle size of the coal powder particles calculated by detection and the third average particle size of the coal powder particles is less than the threshold, the Mask R-CNN network model corresponding to the historical average particle size of the coal powder particles is used to detect the coal powder image to be detected after subsequent preprocessing; if the difference between the average particle size of the coal powder particles calculated by detection and the third average particle size of the coal powder particles is greater than the threshold, the newly collected coal powder image is preprocessed, and the Mask R-CNN network model is trained again, and then the newly trained Mask R-CNN network model is used to detect the preprocessed coal powder image to be detected.

6. The method for online detection of coal powder particle size distribution according to any one of claims 1, 3 to 5, characterized in that: The preprocessing of the coal powder image also includes the following steps: converting the coal powder image into a grayscale image; using the Otsu's algorithm to calculate the global optimal threshold of the grayscale image; binarizing the grayscale image according to the global optimal threshold to obtain a binary image; and taking a screenshot of M×N pixels of the coal powder area in the binary image.

7. The method for online detection of coal powder particle size distribution according to claim 6, characterized in that: The foreign body detection of the coal powder image to be detected comprises the following steps: converting the coal powder image to be detected into a grayscale image; denoising the grayscale image by Gaussian filtering; then performing Fourier transformation on the denoised grayscale image to obtain a spectrum diagram, and performing foreign body detection according to the spectrum diagram.

8. An online detection system for coal powder particle size distribution, characterized in that: include: An image acquisition module is used to acquire images of pulverized coal. The image acquisition module is installed between the outlet of the pulverized coal silo of the boiler and the pulverized coal conveyor belt. The image acquisition module acquires images of pulverized coal at the outlet of the pulverized coal silo of the boiler when the pulverized coal falls into the pulverized coal conveyor belt. A preprocessing module, used for preprocessing the coal powder image; A training module is used to build a Mask R-CNN network model, input the preprocessed coal powder image into the Mask R-CNN network model, and perform training of the Mask R-CNN network model; The particle size detection and calculation module is used to detect the preprocessed coal powder image to be detected using the trained Mask R-CNN network model, obtain the number of pixels corresponding to each coal powder particle, and calculate the particle size of each coal powder particle and the average particle size of the coal powder particles; And, a foreign matter detection module is used to perform foreign matter detection on the coal powder image to be detected.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the on-line detection method for coal powder particle size distribution described in any one of claims 1 to 7 are performed.

10. An online detection device for coal powder particle size distribution, characterized in that: include: Memory, for storing software applications, A processor is used to execute the software application, and each program of the software application correspondingly executes the steps in the online detection method for coal powder particle size distribution described in any one of claims 1 to 7.

Images