An image recognition-based industrial furnace and kiln condition pattern recognition system

Abstract

This invention discloses an industrial furnace operating condition pattern recognition system based on image recognition, comprising an image acquisition module, an edge processing module, an image texture analysis module, a data association module, and an operating condition recognition module. The image acquisition module acquires and preprocesses source image data from within the furnace using a near-infrared camera and a polarizing filter, then constructs a thermal radiation distribution map reflecting differences in material composition. The edge processing module performs frequency domain decomposition of the source image data in scale and direction using a Log-Gabor wavelet filter bank algorithm to obtain pixel phase values. This invention utilizes a combination of a near-infrared camera and a polarizing filter in the image acquisition module, providing a foundation of image data for subsequent texture analysis. Furthermore, the edge processing module can extract closed, continuous edge lines formed in the early stages of crust formation, improving the sensitivity to detect subtle textures and effectively distinguishing between the periodic structure of the crust region and the random textures generated by material flow.

Description

Technical Field

[0001] This invention relates to the field of furnace and kiln operating condition pattern recognition technology, and specifically to an industrial furnace and kiln operating condition pattern recognition system based on image recognition. Background Technology

[0002] Industrial furnaces, as key thermal equipment in industrial production, are widely used in high-energy-consuming industries such as metallurgy, building materials, and chemicals, undertaking core process tasks such as raw material roasting, metal smelting, and material sintering. Their operating status directly affects material conversion rate, product density, and energy utilization efficiency. However, furnace wall crusting can significantly increase the thermal efficiency and fuel consumption of industrial furnaces. Furthermore, crusting narrows the airflow channels within the furnace, and abnormally increases the pressure gradient, potentially leading to furnace sealing failure and flue gas leakage. Therefore, an industrial furnace operating condition pattern recognition system is necessary.

[0003] Currently, most industrial furnace and kiln condition pattern recognition systems only use traditional edge detection operators (such as Canny and Sobel) to recognize low signal-to-noise ratio images. They cannot recognize the subtle texture changes in the image during the early stage of furnace and kiln crust formation. At the same time, the random textures generated by the material flow in the furnace and kiln can mask the periodic structure of the crust region, resulting in insufficient response of the system to the weak features of the early crust region and reduced recognition performance. Summary of the Invention

[0004] To address these issues, the present invention provides an image recognition-based industrial furnace operating condition pattern recognition system.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An industrial furnace and kiln condition pattern recognition system based on image recognition includes an image acquisition module, an edge processing module, an image texture analysis module, a data association module, and a condition recognition module.

[0007] The image acquisition module can acquire and preprocess source image data inside the furnace using a near-infrared camera and polarizing filter, and then construct a thermal radiation distribution map that reflects the differences in material composition.

[0008] The edge processing module can perform frequency domain decomposition of the source image data in terms of scale and orientation using the Log-Gabor wavelet filter bank algorithm to obtain pixel phase values. Pixel phase value Edge information used to identify subtle textures; pixel phase value The calculation formula is as follows:

[0009] in, For the first Wavelet response at a scale For the first The angle of wavelet response at each scale The phase is the value decomposed in the frequency domain;

[0010] The image texture analysis module can divide the thermal radiation distribution map into local regions and generate orientation histograms, and calculate the periodic intensity of each region. And based on the peak intensity and periodic intensity of the orientation histogram Generate texture feature vectors;

[0011] The data association module can align texture feature vectors and thermal radiation distribution maps in time, and calculate the crusting probability of each image region based on the aligned image feature data. Generate a comprehensive shell probability diagram;

[0012] The operating condition identification module classifies the operating conditions of the furnace based on a comprehensive crusting probability map and a preset threshold.

[0013] Furthermore, the polarizing filter is installed in front of the lens of the near-infrared camera to filter out the strong specular reflection light generated by the high-temperature walls and molten materials inside the furnace.

[0014] Furthermore, the image acquisition module also includes a preprocessing submodule, which is used to perform partition correction processing on the source image data.

[0015] Furthermore, the edge processing module can also adjust the threshold value based on a preset high threshold. and low threshold Determine pixel phase value Edge point strength; if Mark them as core edge points; If it is, then it is marked as a weak edge point.

[0016] Furthermore, the image texture analysis module includes an extraction subunit, a recognition subunit, and a generation subunit;

[0017] The extraction subunit can divide the partitioned image of the thermal radiation distribution map into multiple local regions, and statistically analyze the distribution frequency of the edge line in the range of 0-180 degrees in each region to form a direction histogram.

[0018] The identification subunit is used to calculate the periodic intensity of the edge line in the dominant direction in each local region. ;

[0019] The generating subunit can generate a texture feature vector for each local region, including the peak intensity and periodic intensity of the histogram. .

[0020] Furthermore, the periodic intensity The calculation formula is as follows:

[0021] in, and The minimum and maximum displacement ranges of the edge line. The correlation function for the dominant direction of the histogram; periodic intensity A value close to 1 indicates strong periodicity, corresponding to a crust region; periodicity intensity A value close to 0 indicates a random texture.

[0022] Furthermore, the probability of crust formation Indicates the possibility of crust formation; probability of crust formation. The calculation formula is as follows:

[0023]

[0024] in, and All are weighting coefficients. This represents the ratio of thermal radiation.

[0025] Furthermore, the thermal radiation ratio The calculation formula is as follows:

[0026]

[0027] in, and These are the minimum and maximum values ​​in the thermal radiation distribution map.

[0028] The present invention has the following advantages: The present invention uses a combination of a near-infrared camera and a polarizing filter in the image acquisition module, and constructs a thermal radiation distribution map by using the 850nm and 950nm dual-band ratio method, which can identify silicate-rich areas from the material composition level; at the same time, the polarizing filter effectively suppresses specular reflection interference, providing an image data basis for subsequent texture analysis, and the edge processing module can extract the closed continuous edge lines formed in the early stage of crust formation, improving the detection sensitivity of fine textures.

[0029] Meanwhile, the image texture analysis module effectively distinguishes the periodic structure of the crust region from the random texture generated by material flow through directional histograms and periodicity calculations, solving the problem of random textures masking crust features and improving the accuracy and reliability of early crust identification.

[0030] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. Attached Figure Description

[0031] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0032] Figure 1 This is a flowchart illustrating the implementation of an image recognition-based industrial furnace operating condition pattern recognition system according to the present invention. Detailed Implementation

[0033] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figure 1 An industrial furnace and kiln condition pattern recognition system based on image recognition includes an image acquisition module, an edge processing module, an image texture analysis module, a data association module, and a condition recognition module.

[0035] The image acquisition module can acquire and preprocess source image data inside the furnace using a near-infrared camera and polarizing filter, and then construct a thermal radiation distribution map that reflects the differences in material composition, which facilitates the subsequent identification of crusting conditions.

[0036] Among them, the near-infrared camera utilizes the difference in reflectivity of materials in the near-infrared band (800-1000nm) between the furnace crust and the normal furnace wall to capture deep thermal state information that cannot be presented by visible light; the silicate compounds in the primary crust layer show significant absorption peaks in this band, while the normal kiln wall material has a higher reflectivity.

[0037] When constructing the thermal radiation distribution map, firstly, a narrowband filter is used to separate two characteristic bands: 850nm and 950nm. Then, the ratio of the radiation intensity of the 950nm band to the radiation intensity of the 850nm band at each pixel is calculated. , Finally, the ratio of all pixels Constructing a thermal radiation distribution map, ratio Lower pixel regions (i.e., regions where 950nm radiation is strongly absorbed) indicate enrichment of silicate crust components; while the ratio Higher and more uniform areas correspond to normal kiln wall materials.

[0038] A polarizing filter is installed in front of the lens of a near-infrared camera to filter out the strong specular reflection light generated by the high-temperature walls and molten materials inside the furnace, while retaining image data that reflects the surface texture characteristics.

[0039] The image acquisition module also includes a preprocessing submodule, which performs zonal correction on the source image data. The specific details are as follows:

[0040] 1) The probabilistic Hough transform algorithm is used to analyze the kiln wall contour in the source image. This algorithm determines the arc parameters of the kiln wall contour by randomly sampling image edge points and accumulating votes in the parameter space;

[0041] 2) Establish a proportional relationship between the known actual physical dimensions of the furnace and the detected pixel dimensions, and construct a projection model of a three-dimensional coordinate system based on the installation position of the near-infrared camera to ensure the accuracy of geometric correction.

[0042] 3) Unfold the arc parameter image of the kiln wall contour into a planar rectangle using a projection model. This process uses the detected arc center as the pole, sets the sampling radius range radially (from the inner diameter to the outer diameter of the kiln wall), and sets the sampling angle circumferentially from 0 to 360 degrees. For each target pixel in the unfolded image, calculate its corresponding polar coordinates (…). , ), through polar coordinates ( , Find its corresponding rectangular coordinates in the source image data. This allows for the determination of the position of each point in the unfolded image within the original image, thus preserving the actual physical size proportions of the kiln wall surface. This provides a precise geometric basis for the subsequent construction of thermal radiation distribution maps and quantitative analysis of the crust region. Polar coordinates ( , The calculation formula for ) is as follows:

[0043]

[0044]

[0045] in, and These are the starting and ending angles of the arc of the kiln wall in the unfolded diagram. and Let these be the inner and outer radii of the kiln wall. The circumferential angle of the kiln wall. This represents the radial depth of the kiln wall. The total width of the unfolded diagram. This represents the total height of the unfolded diagram.

[0046] Rectangular coordinates The calculation formula is as follows:

[0047]

[0048]

[0049] in, These are the coordinates of the center of the arc.

[0050] The edge processing module can perform scale and orientation frequency domain decomposition on the source image data using the Log-Gabor wavelet filter bank algorithm to obtain pixel phase values. This allows each scale to correspond to a different spatial frequency response, used to identify the edge lines of fine texture pixels; each direction covers a specific angular range, used to detect anisotropic features; it facilitates and effectively extracts and enhances the fine texture features of the crust on the inner wall of the furnace, providing accurate and reliable edge information for subsequent working condition identification.

[0051] Pixel phase value Edge information used to identify subtle textures; higher values ​​indicate stronger edge saliency. Pixel phase value. The calculation formula is as follows:

[0052] in, For the first Wavelet response at a scale For the first The angle of wavelet response at each scale The phase is the value obtained from the frequency domain decomposition.

[0053] The edge processing module can also adjust the threshold based on a preset high threshold. and low threshold Determine pixel phase value Edge point strength; if Mark them as core edge points; If a weak edge point is identified, it is then checked whether a core edge point exists in its neighborhood. If it does, the weak edge point is considered to be a continuation of the real edge and is connected to form a closed and continuous fine texture pixel edge line used to characterize the shell contour.

[0054] The aforementioned Log-Gabor wavelet filter bank algorithm refers to a class of multi-scale, multi-directional frequency domain analysis methods based on logarithmic polar coordinate transformation. Its core feature lies in employing an asymmetric Log-Gabor kernel function for image decomposition. By constructing a bandpass filter bank with logarithmic distribution characteristics in the frequency domain, this algorithm can more effectively capture the local features of different frequency components in an image, improving the detection accuracy of subtle textures and edge information.

[0055] The image texture analysis module can divide the thermal radiation distribution map into local regions and generate orientation histograms, calculating the periodic intensity of each region. To identify the periodic structure of the crust, and based on the peak intensity and periodic intensity of the orientation histogram. Generate texture feature vectors.

[0056] The image texture analysis module includes extraction sub-units, recognition sub-units, and generation sub-units. The extraction sub-unit divides the partitioned image based on the thermal radiation distribution map into multiple local regions, and statistically analyzes the distribution frequency of edge lines within the range of 0-180 degrees in each region to form an orientation histogram to capture the dominant direction of the texture;

[0057] Identify sub-units to calculate the periodic intensity of the edge line in the dominant direction in each local region. By measuring the regularity of the spacing between edge lines, the periodic intensity... A value close to 1 indicates strong periodicity, corresponding to a crust region; periodicity intensity A value close to 0 indicates a random texture.

[0058] Periodic intensity The calculation formula is as follows:

[0059] in, and The minimum and maximum displacement ranges of the edge line. This is the correlation function for the dominant direction of the histogram.

[0060] Related functions The calculation formula is as follows:

[0061] in, For edge images, As the dominant direction, This represents the displacement distance.

[0062] The generating subunit can generate a texture feature vector for each local region, including the peak intensity and periodic intensity of the histogram. Peak intensity and periodic intensity Regions exceeding the threshold are considered as crust-like feature regions.

[0063] The data association module can temporally align texture feature vectors and thermal radiation distribution maps, and calculate the crusting probability of each image region based on the aligned image feature data. Generate a comprehensive shelling probability map, where each pixel region has a probability value representing the likelihood of shelling; shelling probability The calculation formula is as follows:

[0064]

[0065] in, and All are weighting coefficients. + =1, The thermal radiation ratio is calculated using the following formula:

[0066]

[0067] in, and These are the minimum and maximum values ​​in the thermal radiation distribution map.

[0068] Time alignment refers to establishing a unified timestamp benchmark. By using the timestamp benchmark, it is ensured that the texture feature vector obtained from the image texture analysis module and the thermal radiation distribution map obtained from the image acquisition module correspond to the kiln wall state data at the same acquisition time, thereby eliminating the feature mismatch problem caused by asynchronous data acquisition time.

[0069] The operating condition identification module classifies the operating conditions of the furnace based on the comprehensive crusting probability map and preset thresholds, distinguishing between normal operating conditions, early stage of crusting, and severe crusting, and outputs operating condition label.

[0070] For example: under normal operating conditions: the probability of crust formation P is lower than the threshold 1; in the early stage of crust formation: the probability of crust formation P is between the threshold 1 and the threshold 2; in severe crust formation: the probability of crust formation P is higher than the threshold 2.

[0071] This invention employs a near-infrared camera and polarizing filter combination in the image acquisition module. By constructing a thermal radiation distribution map using the 850nm and 950nm dual-band ratio method, it can identify silicate-rich areas at the material composition level. At the same time, the polarizing filter effectively suppresses specular reflection interference, providing an image data foundation for subsequent texture analysis. Furthermore, the edge processing module can extract closed continuous edge lines formed in the early stage of crust formation, improving the detection sensitivity of fine textures.

[0072] Meanwhile, the image texture analysis module effectively distinguishes the periodic structure of the crust region from the random texture generated by material flow through directional histograms and periodicity calculations, solving the problem of random textures masking crust features and improving the accuracy and reliability of early crust identification.

[0073] 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, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An industrial furnace / kiln operating condition pattern recognition system based on image recognition, characterized in that, It includes an image acquisition module, an edge processing module, an image texture analysis module, a data association module, and a working condition recognition module; The image acquisition module can acquire and preprocess source image data inside the furnace through a near-infrared camera and a polarizing filter, and then construct a thermal radiation distribution map that reflects the differences in material composition. The edge processing module can perform frequency domain decomposition of the source image data in terms of scale and orientation using the Log-Gabor wavelet filter bank algorithm to obtain pixel phase values. Pixel phase value Edge information used to identify subtle textures; pixel phase value The calculation formula is as follows: , in, For the first Wavelet response at a scale For the first The angle of wavelet response at each scale The phase is the value decomposed in the frequency domain; The image texture analysis module can divide the thermal radiation distribution map into local regions and generate orientation histograms, and calculate the periodic intensity of each region. And based on the peak intensity and periodic intensity of the orientation histogram Generate texture feature vectors; the periodic intensity The calculation formula is as follows: , in, and The minimum and maximum displacement ranges of the edge line. The correlation function for the dominant direction of the histogram; periodic intensity A value close to 1 indicates strong periodicity, corresponding to a crust region; periodicity intensity A value close to 0 indicates a random texture; The data association module can align texture feature vectors and thermal radiation distribution maps in time, and calculate the crusting probability of each image region based on the aligned image feature data. Generate a comprehensive crust probability map; the crust probability... Indicates the possibility of crust formation; probability of crust formation. The calculation formula is as follows: ， in, and All are weighting coefficients. The ratio of thermal radiation; The thermal radiation ratio The calculation formula is as follows: ， in, and These represent the minimum and maximum values ​​in the thermal radiation distribution map; The operating condition identification module classifies the operating conditions of the furnace based on a comprehensive crusting probability map and a preset threshold.

2. The industrial furnace and kiln condition pattern recognition system based on image recognition according to claim 1, characterized in that, The polarizing filter is installed in front of the lens of the near-infrared camera to filter out the strong specular reflection light generated by the high-temperature walls and molten materials inside the furnace.

3. The industrial furnace and kiln operating condition pattern recognition system based on image recognition according to claim 1, characterized in that, The image acquisition module also includes a preprocessing submodule, which is used to perform partition correction processing on the source image data.

4. The industrial furnace and kiln operating condition pattern recognition system based on image recognition according to claim 1, characterized in that, The edge processing module can also adjust the threshold based on a preset high threshold. and low threshold Determine pixel phase value Edge point strength; if Mark them as core edge points; If it is, then it is marked as a weak edge point.

5. The industrial furnace and kiln operating condition pattern recognition system based on image recognition according to claim 1, characterized in that, The image texture analysis module includes an extraction subunit, a recognition subunit, and a generation subunit; The extraction subunit can divide the partitioned image of the thermal radiation distribution map into multiple local regions, and statistically analyze the distribution frequency of the edge line in the range of 0-180 degrees in each region to form a direction histogram. The identification subunit is used to calculate the periodic intensity of the edge line in the dominant direction in each local region. ; The generating subunit can generate a texture feature vector for each local region, including the peak intensity and periodic intensity of the histogram. .

Images