Image-based flame motion feature extraction method

Abstract

The invention discloses an image-based flame motion feature extraction method, and belongs to the technical field of flame feature extraction. Comprising the following steps: step 1, carrying out preprocessing operations of gray level conversion and filtering and denoising on a flame image collected by a camera in real time; 2, inputting two continuous frames of original images, carrying out downsampling to obtain an image pyramid, calculating a top velocity field, and then carrying out deformation operation on the images; 3, calculating a bottom layer image of the image pyramid by using an iterative calculation formula to obtain a flame optical flow velocity field; and step 4, calculating a velocity mean value and a velocity direction mean value of each frame of pixel point in the flame image, and completing flame motion feature extraction. The method can accurately and reliably extract the effective feature information of the combustor flame combustion process in space and time, qualitatively and quantitatively reflect the motion features of the combustion process, and improves the economy, safety and environmental protection of a power plant.

Description

Technical Field

[0001] The invention relates to the technical field of flame feature extraction, and in particular to an image-based flame motion feature extraction method. Background Art

[0002] In order to absorb more new energy sources, thermal power units need to have the ability to flexibly adjust the peak load. In the future, thermal power units will even be under low-load conditions for a long time, which is likely to cause a series of combustion problems. In addition, due to the price of coal, most power plants choose to burn multiple fuels or burn low-quality coal, which has a greater impact on combustion. Unstable combustion not only reduces the thermal efficiency of the boiler, but also produces a large amount of pollutants and loud noise, and may even cause safety accidents such as furnace fire extinguishing and deflagration. The boiler combustion process is very complicated. The static characteristics of the flame can only characterize part of the information of the combustion process and cannot reflect the dynamic process of combustion over time, which is not conducive to long-term monitoring of the combustion state in the furnace. Due to the harsh furnace environment, complex fuel composition, and turbulent characteristics of flame combustion in industrial combustion systems, many advanced flame feature extraction technologies are difficult to apply. The existing flame feature extraction methods extract flame geometric parameters, luminous parameters, and thermodynamic parameters to reflect the flame state, ignoring the continuous dynamic process of the flame.

[0003] Therefore, a HS optical flow method based on iterative improvement of image pyramid is needed to extract the dynamic characteristics of furnace flames in order to describe the motion characteristics of flame combustion. Summary of the invention

[0004] The purpose of the present invention is to provide a method for extracting flame motion features based on an image, comprising the following steps:

[0005] Step 1: Perform grayscale conversion, filtering and denoising preprocessing operations on the flame image collected by the camera in real time;

[0006] Step 2: Input two consecutive frames of original images and downsample them to obtain an image pyramid. After calculating the top-level velocity field, perform a deformation operation on the image.

[0007] Step 3: Calculate the bottom image of the image pyramid using an iterative calculation formula to obtain the flame optical flow velocity field;

[0008] The iterative calculation formula is:

[0009]

[0010] where u k+1 、v k+1 represents the speed of a pixel in the x-axis and y-axis directions at the k+1th iteration; u k 、v k Respectively represent the speed of the pixel point on the x and y axes at the kth iteration; Ix ,I y ,I t Represents the gradient of the gray value of the pixel in the x, y and time directions, I1 is the gradient value of the pixel before iteration; m and c are the coefficients of the image gray value, represents the local mean, is the velocity field of the previous image in the pyramid, α c , α s , α m represents the weight coefficient of each item, and β is the reduction coefficient;

[0011] Step 4: Calculate the mean value of the velocity magnitude and the mean value of the velocity direction of each pixel in the flame image to complete the flame motion feature extraction.

[0012] The closer to the upper layer of the image pyramid in step 2, the smaller the size of the image.

[0013] The beneficial effects of the present invention are:

[0014] The method of the present invention can accurately and reliably extract effective characteristic information of the burner flame combustion process in space and time, qualitatively and quantitatively reflect the movement characteristics of the combustion process, and improve the economy, safety and environmental protection of the power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of the flame motion feature extraction method based on image of the present invention;

[0016] Figure 2 The curves of coal supply and air supply as a function of frame number;

[0017] Figure 3 This is the curve of flame light flow speed changing with frame number;

[0018] Figure 4 This is the curve of the flame light flow velocity direction changing with the frame number. DETAILED DESCRIPTION

[0019] The present invention proposes a method for extracting flame motion features based on an image, which will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0020] Figure 1 The flowchart of the flame motion feature extraction method based on image of the present invention is as follows:

[0021] 1. Perform grayscale conversion, filtering and denoising preprocessing operations on the flame image collected by the camera in real time;

[0022] Second, input two consecutive frames of original images and downsample them to obtain an image pyramid. After calculating the top-level velocity field, the image is deformed.

[0023] Flame optical flow speed magnitude and direction: Calculate the average speed magnitude and speed direction of each pixel in the flame image. Starting from the basic formula of the HS optical flow method, combined with the pyramid iteration method, according to the brightness conservation assumption of the optical flow method, the following optical flow constraint equation is listed for each pixel of the image:

[0024] I(x+dx,y+dy,t+dt)=I(x,y,t)

[0025] Among them, I(x, y, t) represents the grayscale value of the image position at time t.

[0026] The basic idea of ​​the HS optical flow method is to solve the optical flow field (u, v) by minimizing the following energy function:

[0027]

[0028] in: is a regularization term. The larger the value, the less smooth the velocity field is. α is a smoothing parameter.

[0029] According to Lagrange equation:

[0030]

[0031] Combining the above two equations, we can get

[0032]

[0033] Where Δ is the Laplace operator,

[0034] Taking pixel (i, j) as an example, Δu can be expressed as

[0035] Δu i,j =(u i+1,j -u i,j )-(u i,j -u i-1,j )+(u i,j+1 -u i,j )-(u i,j -u i,j-1 )

[0036] =u i+1,j +u i-1,j +u i,j+1 +u i,j-1 -4u i,j

[0037] Discretize the equation and simplify it to get

[0038]

[0039] In order to improve the robustness of extracting flame motion features, the HS optical flow method based on the dynamic illumination equation is proposed. The basic assumption of this model is that the intensity of the relaxed light is constant, and its equation is:

[0040] I(x+dx,y+dy,t+dt)=M(x,y,t)I(x,y,t)+C(x,y,t)

[0041] Where M and C are also functions that vary in time and space. The above formula indicates that the brightness of the second image is inconsistent with the brightness of the first image.

[0042] After Taylor expansion of the above formula, we can simplify it to get:

[0043] I x u+I y v+I t =(M-1)I+C

[0044] Among them I x ,I y ,I t It represents the gradient of the gray value of a pixel in the x, y and time directions, and u and v represent the speed of the pixel in the x and y axis directions respectively.

[0045] Let m = lim dt→0 (M-1) / dt, c=lim dt→0 (C) / dt

[0046] Substitute into the above equation and eliminate dt. We get a new optical flow equation:

[0047]

[0048] In order to make the optimization problem solvable, two new variables need to be constrained. The optimization problem can be defined as follows:

[0049]

[0050] α c , α s , α m , represents the weight coefficient of each item.

[0051] After simplification, the recursive format is:

[0052]

[0053] where u k+1 、v k+1 represents the speed of a pixel in the x-axis and y-axis directions at the k+1th iteration; u k 、v kRespectively represent the speed of the pixel point on the x and y axes at the kth iteration; I x ,I y ,I t Represents the gradient of the gray value of the pixel in the x, y and time directions, I1 is the gradient value of the pixel before iteration; m and c are the coefficients of the image gray value, represents the local mean, α c , α s , α m It represents the weight coefficient of each item, and β is the reduction coefficient.

[0054] 3. Use the iterative calculation formula to calculate the bottom image of the image pyramid and obtain the flame optical flow velocity field

[0055] Although the basic form of the HS optical flow method based on the dynamic illumination equation can cope with the noise situation of grayscale value changes, the pixels of the flame image actually have a large displacement. Therefore, the image pyramid iteration algorithm is introduced to solve the situation with large displacement. Starting from the top layer of the pyramid, the optical flow velocity vector field is calculated layer by layer. Each layer of the image is sampled from the next layer with a larger size. The image size is larger as you go down. The velocity field of the bottom layer of the pyramid is finally obtained through iterative calculation.

[0056] Assumptions is the velocity field of the previous layer, then the optimization variable iteration format of the next layer is:

[0057]

[0058] For each pixel (i, j), the velocity vector n i =(u i ,v i )'s speed d i With the movement angle θ i They are:

[0059]

[0060] Fourth, calculate the mean value of the speed and direction of each pixel in the flame image to complete the flame motion feature extraction. The calculation formula is as follows:

[0061] Average speed:

[0062]

[0063] Direction size mean:

[0064]

[0065] Figure 2 The curves of coal supply and air supply as a function of frame number; Figure 3 , Figure 4 They are respectively the curves of the flame optical flow velocity magnitude changing with the frame number and the curves of the flame optical flow velocity direction changing with the frame number. By analyzing the flame image of the 10,000-frame shutdown process, the change trends of the flame optical flow velocity magnitude and the flame optical flow velocity direction conform to the dynamic change law of the flame during the unit shutdown process. As the coal supply decreases, the flame optical flow velocity magnitude shows an upward trend, while the flame optical flow velocity direction shows a slow downward trend. The upward trend after the decrease is because the coal supply decreases rapidly, while the air supply decreases slowly, causing the flame to shake violently under a large air volume and the flame turbulence is high. It can be seen that the method of the present invention can effectively extract the dynamic characteristics of the flame and provide a more effective feature for flame state detection.

Claims

1. A flame motion feature extraction method based on an image, characterized in that: The following steps are involved: Step 1: Perform grayscale conversion, filtering and denoising preprocessing operations on the flame image collected by the camera in real time; Step 2: Input two consecutive frames of original images and downsample them to obtain an image pyramid. After calculating the top-level velocity field, perform a deformation operation on the image. Step 3: Calculate the bottom image of the image pyramid using an iterative calculation formula to obtain the flame optical flow velocity field; The iterative calculation formula is: where u k+1 、v k+1 represents the speed of a pixel in the x-axis and y-axis directions at the k+1th iteration; u k 、v k Respectively represent the speed of the pixel point on the x and y axes at the kth iteration; I x ,I y ,I t Represents the gradient of the gray value of the pixel in the x, y and time directions, I1 is the gradient value of the pixel before iteration; m and c are the coefficients of the image gray value, represents the local mean, is the velocity field of the previous image in the pyramid, α c , α s , α m represents the weight coefficient of each item, and β is the reduction coefficient; Step 4: Calculate the mean value of the velocity magnitude and the mean value of the velocity direction of each pixel in the flame image to complete the flame motion feature extraction.

2. The method for extracting flame motion features based on an image according to claim 1, characterized in that: The closer the image pyramid in step 2 is to the upper layer, the smaller the size of the image is.

Images