A method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace

Abstract

The invention relates to the technical field of surface temperature extraction, and provides a method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace. The method firstly uses a deep neural network combined with an attention mechanism to accurately locate a furnace tube in a swinging state in a complex infrared thermal radiation environment, and separates it from a homogeneous background. Then, based on the infrared radiation transmission theory, a lookup table (LUT) is established, and prior knowledge of three physical parameters, namely, the furnace tube surface temperature, the equivalent background temperature, and the emissivity, is fully utilized to accurately deduct the influence of the background's reflected radiation, path radiation, emissivity, etc. on the infrared radiation received by the detector, so as to accurately quantify and calculate the surface temperature distribution of the furnace tube.

Description

Technical Field

[0001] The invention relates to the technical field of surface temperature extraction, and more specifically to a method for extracting the surface temperature of a furnace tube of an automatic coking ethylene cracking furnace. Background Art

[0002] The ethylene cracking furnace is a tubular heating furnace. The thermal cracking chemical reaction in the furnace tube decomposes the material into important industrial raw materials such as ethylene. As the material in the furnace tube is deposited and scaled, carbon deposits and carburization are formed, causing the thermal resistance of the furnace tube to increase rapidly, and eventually forming a local high-temperature area, i.e., a hot spot, on the surface of the furnace tube. If the hot spot exists for a long time without decoking (burning), it may cause the thermal stress of the furnace tube to increase, and eventually cause the tube to burst and break.

[0003] At present, the charring method of cracking furnace is generally a programmed manual operation method. To realize automatic control of charring, the surface temperature of the furnace tube must be accurately measured. Since the infrared radiation characteristics of the furnace tube and the furnace wall background are basically the same during the charring of the furnace tube, the infrared image contrast, signal-to-noise ratio, and information content output by the infrared temperature sensor are extremely low, resulting in the inability to accurately measure the furnace tube temperature. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace. By combining a deep convolutional neural network with an attention mechanism, the accurate positioning and background separation of the furnace tube in a complex infrared thermal radiation environment are achieved. Based on the infrared radiation transmission theory, the lookup table and the prior knowledge of the furnace tube surface temperature, the equivalent background temperature and the emissivity are used to deduct the background interference and accurately invert the surface temperature distribution of the furnace tube, thereby solving the problem of inaccurate surface temperature measurement caused by the similar background infrared characteristics of the furnace tube and the furnace wall in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace comprises the following steps:

[0007] Step S1, using an infrared detector to collect a real-time infrared image of a cracking furnace tube to obtain initial data including the surface temperature information of the furnace tube;

[0008] Step S2, based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, performing real-time image enhancement processing on the infrared image;

[0009] Step S3, using an image segmentation algorithm based on a deep convolutional neural network combined with an attention mechanism to perform semantic segmentation on the enhanced infrared image, separate the furnace tube area from the background area, and extract the infrared image of the furnace tube area;

[0010] Step S4, based on the infrared radiation transmission theory and combined with the multi-spectral temperature inversion equation, the temperature distribution on the furnace tube surface is determined by calculating the infrared radiation received by the detector, and the temperature inversion of each pixel is processed in parallel by a lookup table method;

[0011] Step S5, generating surface temperature distribution data of all furnace tubes within the field of view, and outputting the results to the coking control system.

[0012] As a further solution of the present invention, in step S2, the real-time image enhancement step is based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, and the computational complexity is optimized by the following method:

[0013] The gray value of the image to be processed in the histogram specification is The cumulative distribution function of The grayscale of the image after histogram normalization is The cumulative distribution function of The monotonically increasing characteristic of The solution range is limited to the number of gray levels ,in, is the maximum gray level of the original image; after solving the initial value, based on the characteristics of the monotonically increasing function, when the gray value of the image to be processed is known, Value The solution to the algebraic equation is Then, the next algebraic equation The search scope is reduced to , discard interval , the computational complexity is optimized to .

[0014] As a further solution of the present invention, in step S3, the image segmentation algorithm adopts a combination of DeepLabv3+ deep convolutional neural network and CBAM attention mechanism, wherein the CBAM attention mechanism is embedded in the encoder stage of DeepLabv3+ deep convolutional neural network, which is used to enhance the expression of feature maps in channel dimension and spatial dimension through channel attention module and spatial attention module, and combines the low-level features and high-level features fusion processing of the decoder stage to segment the furnace pipe area from the background area, and extract the infrared image of the furnace pipe area.

[0015] As a further solution of the present invention, in step S4, the temperature inversion is performed based on the multi-spectral temperature inversion equation, and the expression of the multi-spectral temperature inversion equation is:

[0016] ;

[0017] In the formula, is the surface temperature of the furnace tube to be inverted, is the equivalent background temperature, is the smoke transmittance, The detector is at wavelength The response value to temperature is is the detector response at two characteristic wavelengths, is the apparent temperature, is the emissivity of the furnace tube surface.

[0018] As a further solution of the present invention, in step S4, the temperature inversion includes establishing a lookup table, wherein the establishing of the lookup table is implemented by the following steps:

[0019] Step A1, establishing a two-dimensional search grid for the inverted furnace tube surface temperature and equivalent background temperature, and setting a step size, wherein the range of each dimension is the temperature value range of the furnace tube surface temperature and the equivalent background temperature;

[0020] Step A2, traverse any grid point in the two-dimensional search grid ,in Indicates The furnace tube surface temperature at each grid point is Indicates The equivalent background temperature of the grid points; Substitute the two equations of the multi-spectral temperature inversion equation to calculate the emissivity of the grid point at two characteristic wavelengths respectively;

[0021] Step A3, matching the emissivity value of each grid point, when the difference between the emissivity at two characteristic wavelengths is less than a preset threshold and both are within a specified range, the corresponding grid point is determined. is the inversion temperature.

[0022] As a further solution of the present invention, in step S4, the temperature inversion process adopts CUDA multi-core parallel computing technology based on NVIDIA GPU, and queries each pixel of the furnace tube infrared image through a lookup table to invert the surface temperature distribution of all furnace tubes within the field of view.

[0023] Compared with the prior art, the method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace of the present invention has the following beneficial effects:

[0024] The present invention combines deep convolutional neural networks and CBAM attention mechanisms to accurately locate and separate the furnace tubes from the background in a complex infrared thermal radiation environment, overcoming the difficulty of identifying the target due to the similar infrared radiation characteristics of the furnace tubes and the background. This method effectively enhances the expression ability of the feature map in the channel and spatial dimensions through the attention mechanism, further improves the segmentation accuracy of the furnace tube area, and lays a solid foundation for subsequent temperature inversion.

[0025] The present invention is based on the infrared radiation transmission theory, and by making full use of the prior knowledge of the furnace tube surface temperature, equivalent background temperature and emissivity, the interference of background reflected radiation and path radiation is accurately deducted to achieve rapid and accurate inversion of the furnace tube surface temperature. Compared with the traditional method, the present invention significantly improves the accuracy and real-time performance of temperature measurement, and provides reliable technical support for the automated decoking control of the ethylene cracking furnace. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Define the schematic diagram for the histogram.

[0027] Figure 2 This is a furnace tube image enhancement diagram of a method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace according to the present invention.

[0028] Figure 3 This is the original Deeplabv3+ network architecture diagram.

[0029] Figure 4 This is the basic principle diagram of CBAM.

[0030] Figure 5 This is the improved Deeplabv3+ network architecture diagram after adding the attention mechanism in the present invention.

[0031] Figure 6 This is the infrared image of the original furnace tube.

[0032] Figure 7 This is the result of the original Deeplabv3+ network segmentation.

[0033] Figure 8 This is the improved Deeplabv3+ network segmentation result diagram after adding the attention mechanism in the present invention.

[0034] Fig. 9 Schematic diagram of temperature inversion based on infrared radiation transmission.

[0035] Fig.10 The present invention is a schematic flow diagram of a method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace.

[0036] In the figure, Encoder: encoder; Decoder: decoder; Atrous Conv: hole convolution; DCNN: deep convolutional neural network; Spatial Attention: spatial attention; Channel Attention: channel attention; 1x1 Conv: 1×1 convolution; 3x3 Conv rate 6: 3×3 convolution, expansion rate is 6; 3x3 Conv rate 12: 3×3 convolution, expansion rate is 12; 3x3 Conv rate 18: 3×3 convolution, expansion rate is 18; Image Pooling: image pooling; Low-LevelFeatures: low-level features; Concat: feature concatenation; Upsample by 4: the upsampling factor is 4. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] Example 1

[0039] A method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace comprises the following steps:

[0040] Step S1, using an infrared detector to collect a real-time infrared image of a cracking furnace tube to obtain initial data including the surface temperature information of the furnace tube;

[0041] Step S2, based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, real-time image enhancement processing is performed on the infrared image to improve the contrast and signal-to-noise ratio of the image, laying a foundation for subsequent segmentation and temperature calculation;

[0042] Step S3, using an image segmentation algorithm based on a combination of DeepLabv3+ deep convolutional neural network and CBAM attention mechanism to perform semantic segmentation on the enhanced infrared image, separate the furnace tube area from the background area, and extract the infrared image of the furnace tube area;

[0043] Step S4, based on the infrared radiation transmission theory and combined with the multi-spectral temperature inversion equation, the temperature distribution on the furnace tube surface is determined by calculating the infrared radiation received by the detector, and the temperature inversion of each pixel is processed in parallel by a lookup table method;

[0044] Step S5, generating surface temperature distribution data of all furnace tubes within the field of view, and outputting the results to the coking control system to achieve accurate monitoring of furnace tube hot spots and real-time control of automated coking clearing.

[0045] In step S2 of the embodiment of the present invention, the real-time image enhancement step is based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, and the calculation complexity is optimized by the following method:

[0046] The gray value of the image to be processed in the histogram specification is The cumulative distribution function of The grayscale of the image after histogram normalization is The cumulative distribution function of The monotonically increasing characteristic of The solution range is limited to the number of gray levels ,in, is the maximum gray level of the original image; through exhaustive method, The algebraic equation is used to solve the initial gray value, and the total number of searches is ; After solving the initial value, based on the characteristics of the monotonically increasing function, when the gray value of the image to be processed is known Value The solution to the algebraic equation is Then, the next algebraic equation The search scope is reduced to , discard interval , thus reducing the computational complexity from Optimized for .

[0047] In step S3 of the embodiment of the present invention, the image segmentation algorithm adopts a combination of DeepLabv3+ deep convolutional neural network and CBAM attention mechanism, wherein the channel attention module and spatial attention module of the CBAM attention mechanism are embedded in the encoder of the DeepLabv3+ deep convolutional neural network, and the channel attention module generates a pooling vector describing the global distribution of the feature channel by performing global maximum pooling and global average pooling on the feature map, and inputs the pooling vector to the shared fully connected layer to generate a channel weight matrix; the spatial attention module generates a two-dimensional spatial representation of the feature map by performing maximum pooling and average pooling on the feature map along the channel direction, and generates a spatial weight matrix through convolution operation.

[0048] The CBAM attention mechanism in the embodiment of the present invention is embedded in the encoder stage of the DeepLabv3+ deep convolutional neural network to enhance the expression of the feature map in the channel dimension and the spatial dimension, and the fused features are processed through the decoder stage to achieve the segmentation of the furnace pipe area and the background area, and extract the infrared image of the furnace pipe area.

[0049] In step S4 of the embodiment of the present invention, the temperature inversion is performed based on the multi-spectral temperature inversion equation, and the expression of the multi-spectral temperature inversion equation is:

[0050] ;

[0051] In the formula, is the surface temperature of the furnace tube to be inverted, is the equivalent background temperature, is the smoke transmittance, which can be estimated from experiments. The detector is at wavelength The response value to temperature is is the detector response at two characteristic wavelengths, is the apparent temperature, is the emissivity of the furnace tube surface.

[0052] In step S4 of the embodiment of the present invention, the temperature inversion includes establishing a lookup table, wherein the establishing of the lookup table is implemented by the following steps:

[0053] Step A1, establishing a two-dimensional search grid for the inverted furnace tube surface temperature and equivalent background temperature, and setting a step size, wherein the range of each dimension is the temperature value range of the furnace tube surface temperature and the equivalent background temperature;

[0054] Step A2, traverse any grid point in the two-dimensional search grid ,in Indicates The furnace tube surface temperature at each grid point is Indicates The equivalent background temperature of the grid points; Substitute the two formulas of the multi-spectral temperature inversion equation to calculate the emissivity corresponding to the grid point at two characteristic wavelengths;

[0055] Step A3, matching the emissivity value of each grid point, when the difference between the emissivity at two characteristic wavelengths is less than a preset threshold and both are within a specified range, the corresponding grid point is determined. is the inversion temperature.

[0056] In step S4 of the embodiment of the present invention, the temperature inversion process adopts CUDA multi-core parallel computing technology based on NVIDIA GPU, and queries each pixel of the infrared image of the furnace tube through a lookup table to invert the surface temperature distribution of all furnace tubes within the field of view.

[0057] Example 2

[0058] In a complex thermal radiation environment, especially when the furnace tube is in a charred state, the infrared image output by the sensor exhibits the characteristics of high noise and low contrast.

[0059] like Figure 1 The following is a schematic diagram of the histogram specification, where: The gray value of the image to be processed is The cumulative distribution function of The grayscale of the image after histogram normalization is The cumulative distribution function of . Histogram normalization is to know , , In the case of , that is, solving the following nonlinear algebraic equation:

[0060] ;

[0061] in is the maximum grayscale level of the original image, usually 256.

[0062] Considering and The discrete characteristics and relatively narrow value range of The simplest and most effective way to solve the above equation is the exhaustive method. , the number of times the above algebraic equation is searched is times, so the total number of solution searches is .

[0063] In the embodiment of the present invention, the cumulative distribution function is considered and The properties of a monotonically increasing function, when it is known that The solution to the algebraic equation is Then, for the algebraic equation The solution, The search space can be reduced to This allows you to discard spaces that don’t need to be searched. , thus the number of searches can be greatly reduced, and the computational complexity is reduced by Reduce to , ensuring the real-time nature of data processing.

[0064] like Figure 2 The figure shows the image enhancement of the furnace tube of the automatic coking ethylene cracking furnace tube surface temperature extraction method of the present invention. Based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, the complexity of finding the root of the nonlinear equation is reduced from O(L 2 ) is reduced to O(Llog(L)), where L is the grayscale of the image. This satisfies the ultimate requirement of real-time image enhancement and lays a good initial condition for the spatial positioning of the furnace tube and background separation (i.e., image segmentation).

[0065] Example 3

[0066] The scorch control system requires accurate temperature distribution characteristics of the furnace tube surface, but the temperature inverted from the infrared image is the temperature distribution of all scenes within the full field of view, including the furnace tube, furnace wall, burner, etc. Therefore, it is necessary to locate the furnace tube and completely separate the furnace tube from the background, so as to extract the temperature distribution characteristics that belong entirely to the furnace tube area. During the scorch period of the furnace tube, various molecular dynamic chemical reactions continue to occur in the tube, and the temperature distribution of the furnace tube in the radial and axial directions changes dramatically. At the same time, the combustion state of the burner in the furnace is also unstable, which ultimately leads to a dramatic change in the infrared radiation characteristics of the furnace tube as the detection target and the furnace wall, burner, etc. as the background. Therefore, the infrared image output by the sensor has extremely low contrast, signal-to-noise ratio, and information content, making it difficult to automatically identify the furnace tube area.

[0067] like Figure 3 The figure shows the original Deeplabv3+ network architecture. The deep convolutional neural network DeepLabv3+ adopts an encoder-decoder architecture, uses the improved Xception model as the backbone network, and combines structures such as dilated spatial pyramid pooling, dilated separable convolution, and residual blocks, which greatly improves the feature extraction capability and the generalization ability of reasoning, thereby improving the accuracy of target semantic segmentation.

[0068] like Figure 4 The following is a basic principle diagram of CBAM. The attention mechanism draws on the characteristics of the human visual system. The importance of features of things is different, which is reflected in the convolutional network, that is, the importance of each feature map is different. The core idea of ​​the attention mechanism is to obtain the difference in the importance of each feature map through certain means, invest more computing resources of the neural network in more important tasks, and use the task results to reversely guide the weight update of the feature map, so as to complete the corresponding task efficiently and quickly.

[0069] CBAM combines two sub-modules, channel attention and spatial attention, aiming to capture important information of feature maps in both channel and spatial dimensions at the same time. Therefore, it has powerful feature representation capabilities, strong flexibility and embeddability, and lightweight design, which can enhance the convolutional neural network's ability to focus on targets and capture details.

[0070] In the embodiment of the present invention, a deep convolutional neural network deeplabv3+ is combined with a CBAM attention mechanism to realize semantic segmentation of the furnace tube infrared image, thereby achieving accurate separation of the furnace tube area and the background area, that is, extracting the image of the furnace tube area.

[0071] like Figure 5 The figure shows the improved Deeplabv3+ network architecture diagram after adding the attention mechanism of the present invention. The network adopts an encoder-decoder structure, expands the receptive field by introducing a dilated convolution, and embeds the channel attention module and spatial attention module of the CBAM attention mechanism in the encoder stage to enhance the expressiveness of the feature map in the channel and spatial dimensions. The channel attention module generates the channel weights of the feature map through global maximum pooling and global average pooling, thereby adjusting the importance of the feature channel; the spatial attention module generates two-dimensional spatial weights through pooling operations along the channel direction to strengthen the spatial feature expression. In the decoder stage, the network restores the resolution of the feature map by fusing low-level features and high-level features and gradually upsampling, and finally achieves high-precision semantic segmentation.

[0072] like Figure 6 The infrared image of the original furnace tube is shown below. Figure 7 The result of the original Deeplabv3+ network segmentation is shown in Figure 1. Figure 8 The figure shows the improved Deeplabv3+ network segmentation result after adding the attention mechanism in the present invention. The selected superimposed area in the figure is the segmentation result. Figure 7 and Figure 8 From the segmentation results of the dotted box area, it can be clearly seen that the improved image segmentation algorithm can better learn and capture the subtle features of the furnace tube image, such as darker targets, smaller targets, and relatively complex and chaotic backgrounds, thereby achieving better image segmentation effects.

[0073] Example 4

[0074] like Fig. 9 The figure shows the schematic diagram of temperature inversion based on infrared radiation transmission. The infrared radiation received by the detector consists of three parts: spontaneous radiation of the furnace tube, reflection of background radiation by the furnace tube, and spontaneous radiation of flue gas (i.e., process radiation). Among them, spontaneous radiation of the furnace tube is infrared radiation directly generated by the surface of the furnace tube, background radiation reflected by the furnace tube is due to the reflection of background infrared radiation by the surface of the furnace tube, and spontaneous radiation of flue gas is infrared radiation generated inside the flue gas. These three types of radiation work together to form the total radiation signal received by the detector.

[0075] Based on the infrared radiation transmission theory, the total infrared radiation transmission equation can be used to describe the total radiation received by the detector. The expression of the infrared total radiation transmission equation is:

[0076] ;

[0077] The infrared total radiation transfer equation can be simplified as:

[0078] ;

[0079] In the formula, is the surface temperature of the furnace tube to be inverted; and are the equivalent background temperature and the temperature of the flue gas, respectively. In a packaged environment such as a cracking furnace, the furnace tube is surrounded by flue gas, so these two are approximately the same; is the smoke transmittance, which can be estimated by experiment; is the response value of the detector; is the apparent temperature; is the response curve of the detector, obtained by calibration with a blackbody furnace; is the emissivity of the furnace tube, is the reflectivity. Since the furnace tube is opaque, .

[0080] From the infrared total radiation transmission equation, we can know that , , Under known conditions, if we want to solve You must first obtain and ,However, It is closely related to the surface morphology, detection direction, and temperature of the object to be measured. It is generally difficult to accurately estimate it through experiments. is the equivalent background temperature. Due to the complex physical and chemical processes such as heat transfer, mass transfer, and combustion taking place in the furnace all the time, It has been fluctuating constantly.

[0081] In order to improve the temperature measurement accuracy in the embodiment of the present invention, the parameters and The over-reliance on multi-spectral temperature inversion equation is proposed. Taking dual spectrum as an example, the expression of the equation is:

[0082] ;

[0083] In the formula, is the surface temperature of the furnace tube to be inverted, is the equivalent background temperature, is the smoke transmittance, The detector is at wavelength The response value to temperature is is the detector response at two characteristic wavelengths, is the apparent temperature, is the emissivity of the furnace tube surface.

[0084] In the embodiment of the present invention, the ethylene cracking furnace is selected according to the operating characteristics. and A reasonable range is given, and a temperature inversion process is proposed, including establishing a lookup table and temperature inversion based on the lookup table.

[0085] The establishment of the lookup table in the embodiment of the present invention is achieved by the following steps:

[0086] Step 1: Inverted furnace tube surface temperature and the equivalent background temperature Establish a two-dimensional search grid and set a reasonable step size. The range of each dimension is and Reasonable temperature range for each;

[0087] Step 2: Traverse any grid point , substitute it into the two equations in the multi-spectral temperature inversion equation, and calculate the emissivity respectively and ;

[0088] Step three, when and When the values ​​are closest and both are within the reasonable range, then is the inversion temperature.

[0089] The temperature inversion based on the lookup table in the embodiment of the present invention includes querying each pixel of the infrared image of the furnace tube through the lookup table to invert the surface temperature distribution of all the furnace tubes within the field of view.

[0090] Since the temperature inversion of each pixel position is independent of each other and has a natural parallel characteristic, the embodiment of the present invention uses CUDA based on the multi-core parallel computing characteristics of NVIDIA GPU to achieve fast temperature inversion.

[0091] Example 5

[0092] like Fig.10The figure shows a flow chart of a method for automatically extracting the surface temperature of a furnace tube of an ethylene cracking furnace of the present invention. First, an infrared image is acquired by an infrared detector, and the infrared image is enhanced in real time. At the same time, a lookup table is established, and the enhanced image is input into a deep neural network with an attention mechanism added to realize accurate identification of the furnace tube and separation of the background area. The temperature distribution of the entire scene is generated by subtracting reflected radiation and path radiation and avoiding emissivity. Finally, based on the above data, accurate extraction of the furnace tube temperature is realized to provide data support for subsequent decoking control.

[0093] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0094] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting the surface temperature of a furnace tube of an automatic charring ethylene cracking furnace, characterized in that: The following steps are involved: Step S1, using an infrared detector to collect a real-time infrared image of a cracking furnace tube to obtain initial data including the surface temperature information of the furnace tube; Step S2, based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, performing real-time image enhancement processing on the infrared image; Step S3, using an image segmentation algorithm based on a deep convolutional neural network combined with an attention mechanism to perform semantic segmentation on the enhanced infrared image, separate the furnace tube area from the background area, and extract the infrared image of the furnace tube area; Step S4, based on the infrared radiation transmission theory and in combination with the multi-spectral temperature inversion equation, the temperature distribution on the furnace tube surface is determined by calculating the infrared radiation received by the detector, and the temperature inversion of each pixel is processed in parallel by a lookup table method; the temperature inversion is performed based on the multi-spectral temperature inversion equation, and the expression of the multi-spectral temperature inversion equation is: Where T0 is the surface temperature of the furnace tube to be inverted, T B is the equivalent background temperature, τ a is the smoke transmittance, I k () is the detector’s response to temperature at wavelength k, k=1, 2 is the detector’s response at two characteristic wavelengths, T r is the apparent temperature, ε(T0) is the emissivity of the furnace tube surface; Step S5, generating surface temperature distribution data of all furnace tubes within the field of view, and outputting the results to the coking control system.

2. The method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace according to claim 1, characterized in that: In step S2, the real-time image enhancement step is based on the monotonically increasing characteristic of the cumulative distribution function of the grayscale histogram, and the computational complexity is optimized by the following method: By using the monotonically increasing characteristics of the cumulative distribution function T(r) of the gray value r of the image to be processed in the histogram specification and the cumulative distribution function G(z) of the gray value z of the image after the histogram specification, the solution range of the original nonlinear algebraic equation G(z)=T(r) is limited to the gray level number z∈[0,L-1], where L is the maximum gray level of the original image; after solving the initial value, based on the characteristics of the monotonically increasing function, when the i-th value r of the gray value in the image to be processed is known i The solution to the algebraic equation is z j Then, the next step is to use the algebraic equation G(z)=T(r i+1 ) is reduced to [z j ,z L-1 ], discard the interval [z0,z i-1 ], optimizing the computational complexity to O(Llog(L)).

3. The method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace according to claim 1, characterized in that: In step S3, the image segmentation algorithm adopts a combination of DeepLabv3+ deep convolutional neural network and CBAM attention mechanism.

4. The method for extracting the surface temperature of the furnace tube of an automatic charring ethylene cracking furnace according to claim 3, characterized in that: The CBAM attention mechanism is embedded in the encoder stage of the DeepLabv3+ deep convolutional neural network, which is used to enhance the expression of the feature map in the channel dimension and spatial dimension through the channel attention module and the spatial attention module, and combines the low-level features and high-level features in the decoder stage to segment the furnace pipe area from the background area and extract the infrared image of the furnace pipe area.

5. The method for extracting the surface temperature of the tube of an automatic charring ethylene cracking furnace according to claim 1, characterized in that: In step S4, the temperature inversion includes establishing a lookup table.

6. The method for extracting the surface temperature of the tube of an automatic charring ethylene cracking furnace according to claim 5, characterized in that: The establishment of the lookup table is achieved by the following steps: Step A1, establishing a two-dimensional search grid for the inverted furnace tube surface temperature and equivalent background temperature, and setting a step size, wherein the range of each dimension is the temperature value range of the furnace tube surface temperature and the equivalent background temperature; Step A2, traverse any grid point (T 0,i ,T B,i ), where T 0,i represents the surface temperature of the furnace tube at the i-th grid point, T B,i represents the equivalent background temperature of the ith grid point; the grid point (T 0,i ,T B,i ) is substituted into the two equations of the multi-spectral temperature inversion equation to calculate the corresponding emissivity of the grid point at two characteristic wavelengths respectively; Step A3, matching the emissivity value of each grid point, when the difference between the emissivity at two characteristic wavelengths is less than a preset threshold and both are within a specified range, the corresponding grid point T is determined. 0,i is the inversion temperature.

7. The method for extracting the surface temperature of the tube of an automatic charring ethylene cracking furnace according to claim 1, characterized in that: In step S4, the temperature inversion process uses CUDA multi-core parallel computing technology based on NVIDIA GPU to query each pixel of the furnace tube infrared image through a lookup table to invert the surface temperature distribution of all furnace tubes within the field of view.

Images