A boiler scale detection method and device based on multimodal images

Abstract

The present application relates to the field of image detection technology, and specifically to a boiler scale detection method and device based on multimodal images, the method comprising: performing a panoramic scan of the boiler inner wall through a dual-spectral imaging system at a detection temperature, and synchronously collecting visible light images and infrared thermal imaging images; judging the scale type based on the scale chromaticity difference in the visible light image and generating a scale classification map; based on a standard infrared thermal imaging image of a scale-free boiler obtained at the detection temperature, converting the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks; inputting the temperature anomaly distribution map into a deep neural network to output a scale thickness distribution map, aligning the scale classification map with the scale thickness distribution map through spatial registration technology, ensuring that pixels at the same spatial position contain both scale type and thickness values, generating a scale feature fusion map showing scale types and corresponding thicknesses in different color areas, and realizing accurate detection and quantification of boiler scale.

Description

Technical Field

[0001] The present application relates to the field of image detection technology, and in particular to a boiler scale detection method and device based on multimodal images. Background Art

[0002] In natural gas station process systems, boilers serve as a key heat source, continuously supplying high-temperature hot water to heat exchangers. This hot water transfers heat energy to the natural gas flow through the heat exchanger walls, compensating for the temperature loss caused by the subsequent pressure drop. This heat transfer process is crucial for maintaining the natural gas transmission temperature and preventing hydrate blockage. However, scale accumulates on the boiler's heating surfaces over long-term operation. This scale layer hinders the proper diffusion of heat into the water, reducing heat transfer efficiency.

[0003] Existing boiler scale detection technology uses a single infrared thermal imaging method to capture the temperature distribution on the boiler surface during operation and identify scale deposits. This method relies on the physical properties of heat conduction: the boiler's metal substrate, due to its high thermal conductivity, quickly transfers heat to the internal working fluid, resulting in a cool, dark appearance in thermal imaging. However, the scale layer adhering to the metal surface hinders heat transfer due to its thermal resistance effect, leading to localized heat accumulation and the formation of high-temperature areas, which appear as bright patches in the infrared thermal image. In engineering practice, the "brighter, thicker" rule of thumb is often used: the brightness of patches in the infrared thermal image is used to qualitatively determine the extent of scale deposition, with the center of the bright spot interpreted as the thickest area of ​​scale. However, this method, based on a single infrared thermal image, can only roughly indicate abnormal areas and is difficult to support precise boiler maintenance decisions. Summary of the Invention

[0004] To address the above issues, the present application provides a boiler scale detection method and device based on multimodal images, which realizes accurate detection and quantification of boiler scale through the image features of visible light images and infrared thermal imaging and heat conduction theory.

[0005] To achieve the purpose of this application, this application provides the following technical solutions:

[0006] In a first aspect, the present application provides a boiler scale detection method based on multimodal images, comprising:

[0007] At the detection temperature, a dual-spectral imaging system is used to perform a panoramic scan of the boiler inner wall, synchronously capturing visible light images and infrared thermal imaging images;

[0008] Performing polarization filtering on the visible light image to suppress the mirror reflection of the boiler metal substrate and retain the diffuse reflection area on the scale surface. Then, performing LAB color space conversion on the diffuse reflection area, based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated.

[0009] Based on a standard infrared thermal imaging image of a non-scaling boiler obtained at the detection temperature, converting the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition;

[0010] Input the temperature anomaly distribution map into a deep neural network, extract the multi-scale features of the temperature anomaly distribution image through an encoder downsampling structure, generate a thermal resistance variation law, gradually convert the thermal resistance variation law learned by the encoder into a spatially distributed thickness value through a decoder upsampling structure, and output a scale thickness distribution map, wherein each pixel in the scale thickness distribution map stores a floating-point thickness value;

[0011] The scale classification map is aligned with the scale thickness distribution map through spatial registration technology to ensure that pixels at the same spatial position contain both scale type and thickness value, thereby generating a scale feature fusion map showing scale types and corresponding thicknesses in different color areas.

[0012] In one possible implementation, polarization filtering is performed on the visible light image to suppress the mirror reflection of the boiler metal substrate and retain the diffuse reflection area on the scale surface. The diffuse reflection area is then converted to the LAB color space. Based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated, further comprising:

[0013] By analyzing the intensity changes of the visible light image at different polarization angles, the mirror reflection area generated by the boiler metal substrate and the diffuse reflection area generated by the scale surface are distinguished, and the mirror reflection information is suppressed while the diffuse reflection information is retained;

[0014] The difference between the mirror reflection area and the diffuse reflection area is enhanced by adjusting the local contrast and the global grayscale distribution, so as to obtain a visible light image retaining the diffuse reflection area;

[0015] The visible light image retaining the diffuse reflection area is converted into a color space to generate a LAB image including chromaticity channels A and B. The scale type is determined based on the scale accumulation areas in chromaticity channels A and B, and the scale classification map is generated, wherein each pixel in the scale classification map stores a scale type label value, and the scale types include rust scale, carbonate scale, and silicate scale, which are different in the accumulation areas.

[0016] In one possible embodiment, based on a standard infrared thermal imaging image of a scale-free boiler obtained at the detection temperature, converting the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition, further comprising:

[0017] At the same detection temperature, obtaining a standard infrared thermal imaging image of a non-scaling boiler of the same model and specification, and aligning the standard infrared thermal imaging image with the infrared thermal imaging image;

[0018] Using a sliding window to traverse the aligned infrared thermal imaging image and the standard infrared thermal imaging image, a thermal radiation intensity mean difference and a thermal radiation fluctuation difference within the window are calculated, wherein the thermal radiation intensity mean difference is the difference between the window mean in the infrared thermal imaging image and the window mean in the standard infrared thermal imaging image, and the thermal radiation fluctuation difference is the absolute value of the difference between the window standard deviation in the infrared thermal imaging image and the window standard deviation in the standard infrared thermal imaging image;

[0019] Δμ w >2℃ and corresponding Δσ w Pixels with a temperature < 0.5°C are selected as candidate seed points. Continuous adjacent candidate seed points are clustered, and the geometric center of each cluster is taken as the starting point for growth.

[0020] If the neighborhood pixels of the growth starting point satisfy Δμ n >1.5℃ and |Δσ n -Δσ o |<0.3℃, the neighborhood pixel is included in the scale area, and a temperature anomaly distribution map showing a temperature anomaly block is obtained, where the neighborhood pixel includes multiple pixels around the growth starting point, Δμ represents the mean difference in thermal radiation intensity, Δσ represents the thermal radiation fluctuation difference, and Δσ represents the thermal radiation fluctuation difference. o It represents the temperature standard deviation of the area where the growth starting point is located, w represents the pixel point in the window, and n represents the pixel point in the neighborhood.

[0021] In one possible implementation, the temperature anomaly distribution map is input into a deep neural network, multi-scale features of the temperature anomaly distribution map are extracted through an encoder downsampling structure to generate a thermal resistance variation law, the thermal resistance variation law learned by the encoder is gradually converted into a spatially distributed thickness value through a decoder upsampling structure, and a scale thickness distribution map is output, further comprising:

[0022] The encoder's four-layer downsampling structure gradually extracts multi-scale thermal conductivity features from temperature anomaly images. Each layer uses convolution operations to capture the thermal resistance variations in different regions. Combined with a maximum pooling operation, this reduces spatial resolution while increasing the number of feature channels. The third layer introduces a deformable convolution module to adaptively adjust the sampling position to fit the heat flow distortion characteristics caused by sudden changes in thermal conductivity.

[0023] The spatial resolution is gradually restored through the decoder's symmetrical four-layer upsampling structure, and the feature map output by each layer is jump-connected with the encoder's features at the same scale to fuse the local morphology of the scale with the global thermal resistance semantic features to generate a fused feature.

[0024] The fused features are compressed into a single channel through a 1×1 convolution of the network output layer, normalized using a Sigmoid function, and linearly mapped to a thickness value of 0.1-5.0 mm to output a pixel-level scale thickness distribution map.

[0025] In one possible implementation, the scale classification map is aligned with the scale thickness distribution map by using a spatial registration technology to ensure that pixels at the same spatial position contain both scale types and thickness values, thereby generating a scale feature fusion map showing scale types and corresponding thicknesses in different color areas, further comprising:

[0026] The scale classification map obtained based on the visible light image analysis and the thickness distribution map generated based on the infrared thermal imaging image are aligned at the sub-pixel level to form a composite data matrix, wherein each spatial position in the matrix corresponds to a two-tuple including a type label and a thickness value, and statistics are collected on the coverage area percentage and average thickness of each type of scale.

[0027] In some embodiments, the above scale detection method is also applicable to equipment that relies on heat transfer processes and in which scale causes observable heat conduction anomalies, such as distillation tanks, distiller for seawater desalination, and condensers in refrigeration equipment of power plants.

[0028] In one possible implementation, the method further includes:

[0029] Classifying the scale in the scale thickness distribution map into thin scale, medium-thick scale and heavy scale according to the thickness value, and establishing a piecewise linear mapping function between scale thickness and nozzle pressure;

[0030] Calculating the nozzle pressure corresponding to the thickest thickness of the scale area in the scale thickness distribution diagram according to the piecewise linear mapping function;

[0031] When the area of ​​the scale region is smaller than a threshold area, the nozzle is set to be oriented toward the thickest point of the scale region for spraying;

[0032] When the area of ​​the scale area is greater than the threshold area, the center of mass of the scale area is used as the reference coordinate, and the offset vector is calculated in combination with the coordinates of the thickest point of the scale area. The nozzle is set to perform spiral motion with the thickest point as the center and the offset vector as the diameter to spray the scale area.

[0033] In a second aspect, the present application provides a boiler scale detection device based on multimodal images, comprising:

[0034] An image acquisition module is used to perform a panoramic scan of the boiler inner wall at the detection temperature using a dual-spectral imaging system, synchronously acquiring visible light images and infrared thermal imaging images;

[0035] a first image processing module, configured to perform polarization filtering on the visible light image to suppress the mirror reflection of the boiler metal substrate, retain the diffuse reflection area on the scale surface, and then perform LAB color space conversion on the diffuse reflection area. Based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated;

[0036] a second image processing module for converting, based on a standard infrared thermal imaging image of a scale-free boiler acquired at the detection temperature, the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition;

[0037] An analysis module is configured to input the temperature anomaly distribution map into a deep neural network, extract multi-scale features of the temperature anomaly distribution image through an encoder downsampling structure, generate a thermal resistance variation law, gradually convert the thermal resistance variation law learned by the encoder into spatially distributed thickness values ​​through a decoder upsampling structure, and output a scale thickness distribution map, wherein each pixel in the scale thickness distribution map stores a floating-point thickness value;

[0038] A generation module is used to align the scale classification map with the scale thickness distribution map through spatial registration technology, ensure that pixels at the same spatial position contain both scale type and thickness value, and generate a scale feature fusion map showing scale types and corresponding thicknesses in different color areas.

[0039] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, executes the steps of the boiler scale detection method based on multimodal images.

[0040] The embodiments of the present application provide a boiler scale detection method and device based on multimodal images. Through multimodal image fusion technology, image characteristic analysis and heat conduction modeling are combined to achieve accurate detection and quantification of boiler scale. Specifically, visible light polarization imaging is used to effectively remove metal substrate interference, and scale type identification is achieved through the chromaticity characteristics of diffuse reflection of scale. At the same time, through infrared thermal imaging combined with a deep learning network, temperature anomalies are quantified as scale thickness. The spatial registration and fusion of the two modal image data can not only distinguish the chemical composition of scale, but also quantify the deposition thickness, thereby improving the reliability of mixed scale identification and providing a comprehensive basis for formulating precise cleaning strategies. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The accompanying drawings are used to provide a further understanding of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application.

[0042] Figure 1 Flowchart of a boiler scale detection method based on multimodal images provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions of this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of this application.

[0044] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the quantity of the technical features indicated. Thus, features specified as "first" or "second" may explicitly or implicitly include one or more of such features; and in the description of this application, unless otherwise specified, "plurality" means two or more.

[0045] In order to solve the above technical problems, this application proposes the following technical solutions and corresponding embodiments.

[0046] Figure 1 The flowchart of the boiler scale detection method based on multimodal images provided in the embodiment of the present application is as follows: Figure 1 As shown, the embodiment of the present application provides a boiler scale detection method based on multimodal images, comprising the following steps:

[0047] At the detection temperature, the S100 uses a dual-spectral imaging system to perform a panoramic scan of the boiler inner wall, simultaneously capturing visible light images and infrared thermal imaging images.

[0048] The boiler inner wall includes the inner wall of the boiler barrel and the inner walls of the upper and lower heads. The dual-spectral imaging system includes a coaxially mounted visible light imaging module and an infrared thermal imaging module. The present embodiment achieves scale detection by synchronously capturing multimodal images of visible light and infrared thermal imaging. The visible light image captures the morphological and color characteristics of the scale, while the temperature distribution characteristics in the infrared thermal imaging image accurately identify the scale location and assess its thickness distribution.

[0049] Visible light images have high spatial resolution and can clearly display the microscopic morphology and color characteristics of scale. However, they are greatly affected by surface reflections, such as the reflection interference of the metal base of the boiler. Therefore, it is necessary to use a polarization filter to collect visible light images of the boiler inner wall at different rotation angles.

[0050] The infrared thermal imaging module utilizes a thermal imager with a temperature resolution of at least 0.1°C. Its detection principle is based on the significant difference in thermal conductivity between scale and the metal substrate. Scale's thermal conductivity is typically only 1 / 50-1 / 30 of that of the boiler's metal substrate. For every 1mm of scale thickness in industrial boilers, thermal efficiency decreases by approximately 5%. This thermal conductivity difference results in detectable temperature differences between the metal substrate and scale, as well as between areas of varying scale thickness. To effectively detect these temperature differences, the boiler must be heated to generate sufficient thermal conductivity differences. Specifically, the boiler is first heated to operating temperature, then quickly drained of hot water. Image acquisition is performed when the boiler's internal temperature is between 80°C and 120°C (the detection temperature). This is when the temperature difference between the metal substrate and scale is most pronounced. Furthermore, draining the boiler prevents liquid water from interfering with the image. This method is suitable for industrial boilers operating at lower pressures, ensuring detection accuracy while also balancing engineering feasibility. The temperature distribution characteristics in the thermal image can accurately identify scale locations and assess their thickness.

[0051] It should also be noted that when collecting multimodal images, a pan-tilt-controlled dual-spectral imaging system is used to scan along the axial and circumferential directions of the boiler. During scanning, the overlap rate of adjacent image frames is ensured to be no less than 30%. The scanning path can be pre-planned based on the three-dimensional model of the boiler, and the shooting distance is corrected in real time using a laser rangefinder during the scanning process.

[0052] The S200 performs polarization filtering on the visible light image to suppress the mirror reflection of the boiler metal substrate and retain the diffuse reflection area on the scale surface. It then performs LAB color space conversion on the diffuse reflection area. Based on the chromaticity differences of different types of scale in the AB chromaticity channels, it determines the scale type and generates a scale classification map. This further includes:

[0053] By analyzing the intensity changes of visible light images at different polarization angles, the mirror reflection area generated by the boiler metal substrate and the diffuse reflection area generated by the scale surface are distinguished, and the mirror reflection information is suppressed while the diffuse reflection information is retained.

[0054] The difference between the mirror reflection area and the diffuse reflection area is enhanced by adjusting the local contrast and the global grayscale distribution, and a visible light image with the diffuse reflection area is obtained.

[0055] The visible light image retaining the diffuse reflection area is converted to a color space to generate a LAB image including the A and B chromaticity channels. The scale type is determined based on the scale accumulation areas in the A and B chromaticity channels, and a scale classification map is generated. In the scale classification map, each pixel stores a type label value. For example, a label value of 0 indicates rust scale, a label value of 1 indicates carbonate scale, and a label value of 2 indicates silicate scale.

[0056] It should be noted that since the RGB channels of visible light images are affected by light intensity, the RGB values ​​of the same scale under strong light and weak light are significantly different. Therefore, in order to avoid light interference and accurately obtain the chromaticity information of the scale to avoid misjudgment of the scale type, we convert the visible light image into independent LAB images with luminance channel and chrominance channel.

[0057] By analyzing and identifying a large number of scale images, it was found that carbonate scale is yellow-white and granular in cross-section; silicate scale is off-white; rust scale is black-brown or reddish-brown, and its density is more than twice that of general scale, which is further reflected in the color channel. The scale types include rust scale with an aggregation area of ​​A > 25 and B < 12, carbonate scale with an aggregation area of ​​A∈[-12, 25] and B > 12, and silicate scale with an aggregation area of ​​A∈[-20,0] and B∈[-12, 12].

[0058] S300 converts a standard infrared thermal imaging image of a scale-free boiler obtained at a detection temperature into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition.

[0059] In some embodiments, because scale has a much lower thermal conductivity than the metal substrate, areas covered by scale can produce distinct localized high-temperature features. Therefore, a temperature anomaly distribution map is generated by comparing infrared thermal images of a scale-free boiler with the currently acquired infrared thermal image. Specifically, a standard infrared thermal image of a scale-free boiler of the same model and specifications is acquired at the same test temperature and aligned with the infrared thermal image.

[0060] Then, a sliding window is used to traverse the aligned infrared thermal imaging image and the standard infrared thermal imaging image to calculate the mean difference in thermal radiation intensity and the thermal radiation fluctuation difference within the window. Among them, the mean difference in thermal radiation intensity directly reflects the thermal resistance effect caused by scale. The mean difference in thermal radiation intensity (Δμ w ) = Infrared thermal imaging image window mean - standard infrared thermal imaging image window temperature mean, thermal radiation fluctuation difference reflects the uneven distribution of scale in the window, thermal radiation fluctuation difference (Δσ w) = |standard deviation of the window in the infrared thermal imaging image and standard deviation of the window temperature in the standard infrared thermal imaging image|, where w represents the pixel point in the window.

[0061] Finally, Δμ w >2℃ and corresponding Δσ w Pixels with a temperature <0.5℃ are selected as candidate seed points. Continuous adjacent candidate seed points are clustered and the geometric center of each cluster is taken as the starting point of growth. If the neighboring pixels of the starting point of growth satisfy Δμ n >1.5℃ and |Δσ n -Δσ o |<0.3℃, the neighborhood pixel is included in the temperature anomaly block, and a temperature anomaly distribution map showing the temperature anomaly block is obtained, where the neighborhood pixel includes multiple pixels around the growth starting point, Δσ o represents the temperature standard deviation of the area where the growth starting point is located, and n represents the number of pixels in the neighborhood. When the neighborhood pixel is at the image boundary, the growth stops.

[0062] This embodiment makes collaborative judgments based on the thermodynamic characteristics Δμ and Δσ. Taking advantage of the fact that real scale has the characteristics of high Δμ and low Δσ, while noise / artifacts have the characteristics of high Δμ and high Δσ, it effectively distinguishes real scale from noise / artifacts and accurately identifies the geometric shape and area of ​​the temperature anomaly block.

[0063] The above temperature anomaly distribution map identifies areas with abnormal heat conduction, but it cannot directly quantify the relationship between the degree of anomaly and scale thickness. To address this issue, according to Fourier's law of heat conduction, scale deposits form an additional thermal resistance layer, causing abnormally high local temperatures. This temperature change exhibits a nonlinear positive correlation with scale thickness: the greater the thickness, the more significant the temperature difference ΔT. This embodiment uses an encoder-decoder network architecture to accurately translate temperature anomalies into scale thickness.

[0064] S400 inputs the temperature anomaly distribution map into a deep neural network, extracts the multi-scale features of the temperature anomaly distribution map through the encoder downsampling structure, generates the thermal resistance variation law, and gradually converts the thermal resistance variation law learned by the encoder into spatially distributed thickness values ​​through the decoder upsampling structure, outputting the scale thickness distribution map, where each pixel in the temperature anomaly distribution map stores a floating-point thickness value. Further comprising:

[0065] The encoder's four-layer downsampling structure progressively extracts multi-scale thermal conductivity features from the temperature anomaly image. Each layer uses convolution to capture the thermal resistance variation patterns of different regions. Combined with a maximum pooling operation, this reduces spatial resolution while increasing the number of feature channels. The third layer introduces a deformable convolution module to adaptively adjust the sampling position to fit the heat flow distortion characteristics caused by thermal conductivity abrupt changes. The heat flow distortion characteristics characterize the local heat conduction path variations caused by scale deposition. When heat flows through the interface between scale and the boiler metal substrate, the difference in thermal conductivity between the scale and the substrate causes heat flow lines to deflect and change in density. This feature is captured through offset learning in the deformable convolution to locate thickness abrupt changes at the scale edge. The thermal resistance variation pattern represents the mapping between the spatial distribution of scale and thermal resistance, including the cumulative effect of thermal resistance along the thickness of the boiler wall and the thermal conductivity differences between scale compositions (such as rust and carbonate).

[0066] The spatial resolution is gradually restored through the decoder's symmetrical four-layer upsampling structure, and the feature map output by each layer is jump-connected to the encoder's features of the same scale to fuse the local morphology of the scale with the global thermal resistance semantic features to generate a fusion feature. The thermal resistance variation law includes the nonlinear mapping relationship between temperature and thickness learned by the deep neural network, which conforms to Fourier's law of heat conduction. The global thermal resistance semantic feature is a high-order representation of the temperature-thickness relationship output by the deep decoder of the network. In essence, it is the learned equivalent thermal resistance model R fouling :R fouling =d / λ scale =ΔT / q,λ scale Representing thermal conductivity, the global thermal resistance semantic feature establishes physical constraints between temperature anomaly (ΔT) and thickness (d) through heat flux density (q), ensuring that the prediction complies with thermodynamic laws.

[0067] The fused features are compressed into a single channel through the 1×1 convolution of the network output layer, normalized by the Sigmoid function, and linearly mapped to the actual thickness range of 0.1-5.0 mm to output a pixel-level scale thickness distribution map.

[0068] This embodiment forms a technical closed loop of "temperature anomaly input - thermal resistance feature extraction - thickness distribution output", in which the multi-scale thermal resistance features output by the encoder serve as the input basis of the decoder, and the thickness prediction generated by the decoder provides a spatial registration basis for subsequent scale composition analysis.

[0069] S500 uses spatial registration technology to align the scale classification map with the thickness distribution map, ensuring that pixels at the same spatial location contain both scale type and thickness values, generating a scale feature fusion map showing scale types and their corresponding thicknesses in different color areas. This further includes:

[0070] The scale classification map derived from visible light image analysis and the thickness distribution map generated from infrared thermal imaging are aligned at the subpixel level to form a composite data matrix. In this matrix, each spatial location (x, y) corresponds to a two-tuple (type label, thickness value). For example, a pixel storing <0, 1.2> indicates a 1.2mm thick rust deposit at that location. This data structure is intuitively presented using pseudo-color rendering: rust areas are represented in reddish-brown with overlaid contour lines showing a 1.0-3.0mm thickness gradient. Carbonate areas are represented in off-white with a 0.3-1.5mm color scale gradient. Silicate areas are labeled in light gray with a 0.1-0.8mm thickness gradient. This pixel-level type-thickness binding enables a complete characterization of scale characteristics. At the same time, rust scale is displayed as reddish-brown with high-thickness contour lines, carbonate scale is displayed as yellow-white with medium-thickness color gradations, and silicate scale is displayed as grayish-white with medium-thickness color gradations. Mixed deposits are distinguished by cross-hatching patterns. At the same time, the coverage percentage and average thickness statistics of each type of scale are output. This achieves the improvement of detection reliability by establishing physical association rules between color features and thickness parameters and mutual verification based on multimodal data. In some embodiments, the method of the embodiment of the present application also includes determining the nozzle pressure and injection method in the high-pressure nozzle cleaning device based on the scale thickness in the scale thickness distribution map, and providing a scale cleaning solution. It further includes:

[0071] According to the thickness value, the scale in the scale thickness distribution map is divided into thin scale, medium-thick scale and heavy scale, and a piecewise linear mapping function of scale thickness and nozzle pressure is established; according to the piecewise linear mapping function, the nozzle pressure corresponding to the thickest thickness of the scale area in the scale thickness distribution map is calculated; when the area of ​​the scale area is smaller than the threshold area, the nozzle is set to be oriented toward the thickest point of the scale area for spraying; when the area of ​​the scale area is larger than the threshold area, the center of mass of the scale area is used as the reference coordinate, and the offset vector is calculated in combination with the coordinates of the thickest point, and the nozzle is set to perform a spiral motion with the thickest point as the center and the offset vector as the diameter to spray the scale area.

[0072] In some embodiments, the coverage percentage and average thickness of each type of scale are used to infer the softening of hard water by the water softening device at the front end of the boiler. The distribution of scale directly reflects the excessive concentration of specific metal ions in the boiler water. For example, when the iron oxide content in the scale is significantly increased, it indicates that the water has Fe 3+ The concentration has exceeded the safety threshold, which indirectly proves that the adsorption capacity of the ion exchange resin of the front-end water softening device for iron ions has decreased or the oxidation inhibition has failed; similarly, the increase in carbonate scale and silicate scale indicates that the hardness filtration efficiency of the water softening device has decreased. Based on this, the water softening device at the front end of the boiler can be adjusted accordingly.

[0073] In some embodiments, the present application provides a boiler scale detection device based on multimodal images, comprising:

[0074] The image acquisition module is used to perform a panoramic scan of the boiler inner wall through a dual-spectrum imaging system at the detection temperature, and synchronously acquire visible light images and infrared thermal imaging images.

[0075] The first image processing module is used to perform polarization filtering on the visible light image, suppress the mirror reflection of the boiler metal base, retain the diffuse reflection area on the scale surface, and then perform LAB color space conversion on the diffuse reflection area. Based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated.

[0076] The second image processing module is used to convert the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks based on a standard infrared thermal imaging image of a scale-free boiler obtained at a detection temperature, wherein the temperature anomaly blocks represent the location and range of scale deposition.

[0077] The analysis module is used to input the temperature anomaly distribution map into the deep neural network, extract the multi-scale features of the temperature anomaly distribution image through the encoder downsampling structure, generate the thermal resistance variation law, and gradually convert the thermal resistance variation law learned by the encoder into spatially distributed thickness values ​​through the decoder upsampling structure, and output the scale thickness distribution map, where each pixel in the scale thickness distribution map stores a floating-point thickness value.

[0078] The generation module is used to align the scale classification map with the scale thickness distribution map through spatial registration technology, ensuring that pixels at the same spatial position contain both scale type and thickness values, and generate a scale feature fusion map showing scale types and their corresponding thicknesses in different color areas.

[0079] In some embodiments, the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the boiler scale detection method based on multimodal images are executed.

[0080] This application uses multimodal image fusion technology to combine image characteristic analysis with heat conduction modeling to achieve accurate detection and quantitative evaluation of boiler scale. Specifically, visible light polarization imaging is used to effectively remove interference from the metal substrate, and the chromaticity characteristics of diffuse reflection of scale are used to identify the type of scale. At the same time, infrared thermal imaging is combined with a deep learning network to quantify temperature anomalies as scale thickness. The spatial registration and fusion of the two modal image data can not only distinguish the chemical composition of scale, but also quantify the deposition thickness, thereby improving the reliability of mixed scale identification and providing a comprehensive basis for formulating precise cleaning strategies.

[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of the boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0082] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0083] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0084] In the several embodiments provided in this application, it should be understood that the disclosed systems, modules and methods can be implemented in other ways. For example, the module embodiments described above are merely illustrative. For example, the division of units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of modules or units, which can be electrical, mechanical or other forms.

[0085] The above embodiments are intended only to illustrate the technical solutions of the present application and are not intended to limit them. The present application is not limited to the precise structures described above and illustrated in the accompanying drawings, and it cannot be assumed that the specific implementation of the present application is limited to these descriptions. For those skilled in the art of the present application, any changes and modifications made without departing from the concept of the present application should be deemed to fall within the scope of protection of the present application.

Claims

1. A boiler scale detection method based on multimodal images, characterized in that: include: At the detection temperature, a dual-spectral imaging system is used to perform a panoramic scan of the boiler inner wall, synchronously capturing visible light images and infrared thermal imaging images; Performing polarization filtering on the visible light image to suppress the mirror reflection of the boiler metal substrate and retain the diffuse reflection area on the scale surface. Then, performing LAB color space conversion on the diffuse reflection area, based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated. Based on a standard infrared thermal imaging image of a non-scaling boiler obtained at the detection temperature, converting the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition; Input the temperature anomaly distribution map into a deep neural network, extract the multi-scale features of the temperature anomaly distribution image through an encoder downsampling structure, generate a thermal resistance variation law, gradually convert the thermal resistance variation law learned by the encoder into a spatially distributed thickness value through a decoder upsampling structure, and output a scale thickness distribution map, wherein each pixel in the scale thickness distribution map stores a floating-point thickness value; The scale classification map is aligned with the scale thickness distribution map through spatial registration technology to ensure that pixels at the same spatial position contain both scale type and thickness value, thereby generating a scale feature fusion map showing scale types and corresponding thicknesses in different color areas.

2. The boiler scale detection method based on multimodal images according to claim 1, characterized in that: The visible light image is subjected to polarization filtering processing to suppress the mirror reflection of the boiler metal substrate and retain the diffuse reflection area of ​​the scale surface. The diffuse reflection area is then subjected to LAB color space conversion. Based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated, further comprising: By analyzing the intensity changes of the visible light image at different polarization angles, the mirror reflection area generated by the boiler metal substrate and the diffuse reflection area generated by the scale surface are distinguished, and the mirror reflection information is suppressed while the diffuse reflection information is retained; The difference between the mirror reflection area and the diffuse reflection area is enhanced by adjusting the local contrast and the global grayscale distribution, so as to obtain a visible light image retaining the diffuse reflection area; The visible light image retaining the diffuse reflection area is converted into a color space to generate a LAB image including chromaticity channels A and B. The scale type is determined based on the scale accumulation areas in chromaticity channels A and B, and the scale classification map is generated, wherein each pixel in the scale classification map stores a scale type label value, and the scale types include rust scale, carbonate scale, and silicate scale, which are different in the accumulation areas.

3. The boiler scale detection method based on multimodal images according to claim 1, characterized in that: Based on a standard infrared thermal imaging image of a non-scaling boiler obtained at the detection temperature, converting the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition, further comprising: At the same detection temperature, obtaining a standard infrared thermal imaging image of a non-scaling boiler of the same model and specification, and aligning the standard infrared thermal imaging image with the infrared thermal imaging image; Using a sliding window to traverse the aligned infrared thermal imaging image and the standard infrared thermal imaging image, a thermal radiation intensity mean difference and a thermal radiation fluctuation difference within the window are calculated, wherein the thermal radiation intensity mean difference is the difference between the window mean in the infrared thermal imaging image and the window mean in the standard infrared thermal imaging image, and the thermal radiation fluctuation difference is the absolute value of the difference between the window standard deviation in the infrared thermal imaging image and the window standard deviation in the standard infrared thermal imaging image; Δμ w >2℃ and corresponding Δσ w Pixels with a temperature < 0.5°C are selected as candidate seed points. Continuous adjacent candidate seed points are clustered, and the geometric center of each cluster is taken as the starting point for growth. If the neighborhood pixels of the growth starting point satisfy Δμ n >1.5℃ and |Δσ n -Δσ o |<0.3℃, the neighborhood pixel is included in the scale area, and a temperature anomaly distribution map showing a temperature anomaly block is obtained, where the neighborhood pixel includes multiple pixels around the growth starting point, Δμ represents the mean difference in thermal radiation intensity, Δσ represents the thermal radiation fluctuation difference, and Δσ represents the thermal radiation fluctuation difference. o It represents the temperature standard deviation of the area where the growth starting point is located, w represents the pixel point in the window, and n represents the pixel point in the neighborhood.

4. The boiler scale detection method based on multimodal images according to claim 3 is characterized in that: The temperature anomaly distribution map is input into a deep neural network, multi-scale features of the temperature anomaly distribution map are extracted through an encoder downsampling structure to generate a thermal resistance variation law, the thermal resistance variation law learned by the encoder is gradually converted into a spatially distributed thickness value through a decoder upsampling structure, and a scale thickness distribution map is output, further comprising: The encoder's four-layer downsampling structure gradually extracts multi-scale thermal conductivity features from temperature anomaly images. Each layer uses convolution operations to capture the thermal resistance variations in different regions. Combined with a maximum pooling operation, this reduces spatial resolution while increasing the number of feature channels. The third layer introduces a deformable convolution module to adaptively adjust the sampling position to fit the heat flow distortion characteristics caused by sudden changes in thermal conductivity. The spatial resolution is gradually restored through the decoder's symmetrical four-layer upsampling structure, and the feature map output by each layer is jump-connected with the encoder's features at the same scale to fuse the local morphology of the scale with the global thermal resistance semantic features to generate a fused feature. The fused features are compressed into a single channel through a 1×1 convolution of the network output layer, normalized using a Sigmoid function, and linearly mapped to a thickness value of 0.1-5.0 mm to output a pixel-level scale thickness distribution map.

5. The boiler scale detection method based on multimodal images according to claim 4 is characterized in that: The scale classification map is aligned with the scale thickness distribution map by using a spatial registration technology to ensure that pixels at the same spatial position contain both scale types and thickness values, thereby generating a scale feature fusion map showing scale types and corresponding thicknesses in different color areas, further comprising: The scale classification map obtained based on the visible light image analysis and the thickness distribution map generated based on the infrared thermal imaging image are aligned at the sub-pixel level to form a composite data matrix, wherein each spatial position in the matrix corresponds to a two-tuple including a type label and a thickness value, and statistics are collected on the coverage area percentage and average thickness of each type of scale.

6. The boiler scale detection method based on multimodal images according to claim 4, characterized in that: The method further comprises: Classifying the scale in the scale thickness distribution map into thin scale, medium-thick scale and heavy scale according to the thickness value, and establishing a piecewise linear mapping function between scale thickness and nozzle pressure; Calculating the nozzle pressure corresponding to the thickest thickness of the scale area in the scale thickness distribution diagram according to the piecewise linear mapping function; When the area of ​​the scale region is smaller than a threshold area, the nozzle is set to be oriented toward the thickest point of the scale region for spraying; When the area of ​​the scale area is greater than the threshold area, the center of mass of the scale area is used as the reference coordinate, and the offset vector is calculated in combination with the coordinates of the thickest point of the scale area. The nozzle is set to perform spiral motion with the thickest point as the center and the offset vector as the diameter to spray the scale area.

7. A boiler scale detection device based on multimodal images, characterized in that: include: An image acquisition module is used to perform a panoramic scan of the boiler inner wall at the detection temperature using a dual-spectral imaging system, synchronously acquiring visible light images and infrared thermal imaging images; a first image processing module, configured to perform polarization filtering on the visible light image to suppress the mirror reflection of the boiler metal substrate, retain the diffuse reflection area on the scale surface, and then perform LAB color space conversion on the diffuse reflection area. Based on the chromaticity difference of different types of scale in the AB chromaticity channel, the scale type is determined and a scale classification map is generated; a second image processing module for converting, based on a standard infrared thermal imaging image of a scale-free boiler acquired at the detection temperature, the infrared thermal imaging image into a temperature anomaly distribution map showing temperature anomaly blocks, wherein the temperature anomaly blocks represent the location and range of scale deposition; An analysis module is configured to input the temperature anomaly distribution map into a deep neural network, extract multi-scale features of the temperature anomaly distribution image through an encoder downsampling structure, generate a thermal resistance variation law, gradually convert the thermal resistance variation law learned by the encoder into spatially distributed thickness values ​​through a decoder upsampling structure, and output a scale thickness distribution map, wherein each pixel in the scale thickness distribution map stores a floating-point thickness value; A generation module is used to align the scale classification map with the scale thickness distribution map through spatial registration technology, ensure that pixels at the same spatial position contain both scale type and thickness value, and generate a scale feature fusion map showing scale types and corresponding thicknesses in different color areas.

8. 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 boiler scale detection method based on multimodal images according to any one of claims 1 to 6 are executed.

Images