Optical image enhancement and recognition method and system for valves in severe weather conditions based on multispectral fusion

Abstract

The present application provides a method and system for enhanced optical image recognition of valves in severe weather conditions based on multispectral fusion, wherein the method comprises: obtaining multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions; separating the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture characteristics associated with the reflective properties of the valve material; performing dual amplitude correction on the multi-band texture characteristics in both the spatial and frequency domains; assigning spectral fusion weights to the multispectral optical image data based on the corrected multi-band texture characteristics; and generating an enhanced optical image that matches the actual mechanical state of the valve based on the spectral fusion weights and the corrected multi-band texture characteristics. The present application improves the recognition accuracy of defects such as surface cracks and deformations of valves in severe weather conditions.

Description

Technical Field

[0001] The present application relates to the field of multispectral fusion technology, and in particular to a method and system for enhancing the optical image recognition of valves in severe weather conditions based on multispectral fusion. Background Art

[0002] When industrial pipeline valves operate in harsh weather conditions such as strong winds, dust storms, rain, and fog, their surfaces are susceptible to environmental interference (such as dust adhesion and water vapor scattering). This makes it difficult for traditional optical inspection methods to accurately identify defects such as valve cracks and corrosion. Existing technologies must address the collaborative optimization of noise suppression and effective feature enhancement in multispectral imagery. Furthermore, they must integrate valve mechanical state data (such as vibration and stress) to enhance the physical relevance of defect detection.

[0003] Currently, the more advanced solution uses an image enhancement method based on the fusion of infrared and visible light dual bands. By establishing a statistical model of dust scattering noise, background difference processing is performed on the multispectral image, and the visible light texture details and infrared thermal radiation characteristics are fused with fixed weights to finally output an enhanced image.

[0004] The problems of this scheme are that it is difficult to distinguish between dust adhesion and real defects; the fixed weight fusion strategy cannot adapt to the reliability changes of each band feature under different weather conditions; and the structural deformation features in the enhanced image have a low degree of match with the actual mechanical state. Summary of the Invention

[0005] The present application provides a method and system for optical image enhancement recognition of valves in severe weather conditions based on multi-spectral fusion, which is used to solve the problem of low recognition accuracy of defects such as cracks and deformation on the valve surface in severe weather conditions in the prior art.

[0006] In a first aspect, the present application provides a valve optical image enhancement and recognition method under severe weather conditions based on multispectral fusion, comprising:

[0007] Acquire multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions;

[0008] Based on the mechanical dynamic response characteristics, the multispectral optical image data is separated to extract multi-band texture features associated with the reflective characteristics of the valve material;

[0009] Performing dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture features;

[0010] assigning a spectral fusion weight to the multispectral optical image data according to the corrected multi-band texture features;

[0011] Based on the spectral fusion weights and the modified multi-band texture features, an enhanced optical image matching the actual mechanical state of the valve is generated.

[0012] Optionally, the multispectral optical image data is separated based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the valve material reflectance characteristics, including:

[0013] Constructing a dynamic noise separation model related to dust scattering interference, wherein the parameters of the dynamic noise separation model are jointly determined by the vibration frequency time variation curve and the mechanical stress spatial gradient distribution in the mechanical dynamic response characteristics;

[0014] determining, according to the dynamic noise separation model, the time-varying interference intensity and the noise diffusion area of the dust scattering noise in the multispectral optical image data;

[0015] Couple the time-varying interference intensity with the noise diffusion area to determine a noise separation boundary condition;

[0016] performing a cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition, so as to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold, and treating a set of all target pixels as a dust scattering noise component;

[0017] Multi-band texture features are extracted from the multispectral optical image data from which the dust scattering noise component is removed.

[0018] Optionally, performing a cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold includes:

[0019] Based on the noise separation boundary conditions and the noise diffusion area, a mapping table of inverse variation relationships between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates is established;

[0020] Based on the inverse change relationship mapping table, a joint determination rule for the attenuation amplitude of visible light reflectivity and the increase in infrared radiation intensity is constructed. The boundary conditions of the joint determination rule include a first threshold and a second threshold. The first threshold is obtained by dynamically calculating the difference between the average reflectivity of the visible light band in the absence of dust interference and the current time-varying interference intensity. The second threshold is obtained by superimposing the baseline intensity of the infrared band within the normal radiation range of the valve material and the radiation increment of the noise diffusion area.

[0021] Target pixels whose reflectivity in the visible light band is lower than the first threshold and whose radiation intensity in the infrared band is higher than the second threshold are screened out.

[0022] Optionally, establishing a mapping table of inverse variation relationships between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates based on the noise separation boundary condition and the noise diffusion area includes:

[0023] Converting the time-varying interference intensity in the noise separation boundary condition into a space-time matrix according to the amplitude mutation point distribution of the vibration frequency time variation curve;

[0024] Calculating a diffusion area ratio parameter at each spatial coordinate position according to the spatial coverage shape of the noise diffusion area;

[0025] Performing a pixel-by-pixel multiplication operation on the space-time matrix and the diffusion area ratio parameter to generate a noise interference superposition weight map;

[0026] Based on the gradient distribution direction in the noise interference superposition weight map, low reflectivity pixels whose gradient distribution direction angle is greater than a set angle are identified in the visible light band reflectivity, and high radiation intensity pixels whose gradient distribution direction angle is less than a set angle are identified in the infrared band radiation intensity;

[0027] A reverse change relationship mapping table is constructed using the spatial overlap between the low reflectivity pixels and the high radiation intensity pixels as index key values.

[0028] Optionally, coupling calculation of the time-varying interference intensity and the noise diffusion area to determine a noise separation boundary condition includes:

[0029] Calculating the instantaneous impact value of the time-varying interference intensity based on the vibration frequency time variation curve;

[0030] dividing the noise diffusion area into a core influence area and an edge transition area according to the spatial gradient distribution of the mechanical stress;

[0031] In the core influence area, a convolution operation is performed on the time-varying interference intensity and the spatial stress gradient to generate the interference intensity of the core influence area, and expansion compensation is performed in the edge transition area to obtain the interference intensity of the edge transition area;

[0032] According to the interference intensity of the core influence area and the interference intensity of the edge transition area, the values of the parameters used to separate the core influence area and the edge transition area are adjusted to obtain the noise separation boundary conditions.

[0033] Optionally, generating an enhanced optical image matching an actual mechanical state of the valve based on the spectral fusion weight and the modified multi-band texture feature includes:

[0034] Dynamically adjusting the spectral fusion weight according to the mechanical dynamic response characteristics;

[0035] According to the corrected multi-band texture features, the multispectral optical image data is layered and fused according to the adjusted spectral fusion weights;

[0036] Based on the mechanical stress spatial gradient distribution, marking potential structural abnormality areas in the layered fused multispectral optical image data;

[0037] Based on the multispectral optical image data marked with the potential structural abnormality area, an enhanced optical image matching the actual mechanical state of the valve is generated.

[0038] Optionally, performing dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture feature includes:

[0039] Based on the mechanical dynamic response characteristics, the main peak value of the energy distribution of the multi-band texture feature in the frequency domain is extracted, and the matching degree between the frequency component amplitude corresponding to the main peak value of the energy distribution and the natural vibration frequency of the valve material is used as the frequency domain amplitude correction coefficient;

[0040] Calculating the anisotropic intensity ratio of the multi-band texture feature in the spatial domain according to the maximum gradient direction of the mechanical stress spatial gradient distribution;

[0041] In the spatial domain, the amplitude of the multi-band texture feature is enhanced and corrected in proportion to the anisotropic intensity ratio along the direction of the maximum gradient, and an attenuation correction is performed inversely proportional to the anisotropic intensity ratio in the direction perpendicular to the maximum gradient. In the frequency domain, amplitude suppression inversely proportional to the frequency domain amplitude correction coefficient is applied to the multi-band texture feature to obtain a modified multi-band texture feature.

[0042] In a second aspect, the present application provides a valve optical image enhancement recognition system in severe weather conditions based on multispectral fusion, comprising:

[0043] An acquisition module is used to obtain multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions;

[0044] a separation module, configured to separate the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the reflective characteristics of the valve material;

[0045] A correction module, configured to perform dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture features;

[0046] an allocating module, configured to allocate spectral fusion weights to the multispectral optical image data according to the corrected multi-band texture features;

[0047] A generating module is used to generate an enhanced optical image that matches the actual mechanical state of the valve based on the spectral fusion weight and the modified multi-band texture feature.

[0048] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described in the first aspect for enhancing the optical image recognition of valves in severe weather conditions based on multispectral fusion.

[0049] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, which, when executed by a processor, implements a valve optical image enhancement and recognition method in severe weather conditions based on multispectral fusion as described in any one of the first aspects.

[0050] In the present application, a method for enhanced optical image recognition of valves in severe weather conditions based on multispectral fusion is provided. The method comprises: obtaining multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions; separating the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the reflective characteristics of the valve material; performing dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture features; assigning spectral fusion weights to the multispectral optical image data based on the corrected multi-band texture features; and generating an enhanced optical image that matches the actual mechanical state of the valve based on the spectral fusion weights and the corrected multi-band texture features.

[0051] The technical solution provided by this application has the following beneficial effects:

[0052] This application achieves synchronous acquisition of multimodal data on valve status, providing a time-space aligned input source for subsequent fusion processing. A physical noise model is established using vibration and stress data to improve the separation accuracy of dust scattering noise from actual valve features. Directional enhancement (in the spatial domain) and resonant frequency band screening (in the frequency domain) are used to preserve texture features strongly related to valve material and deformation. Band weights are adaptively adjusted based on the reliability of the corrected features to optimize feature complementarity under different weather conditions. The output image retains both high-resolution texture details and features associated with mechanical status, improving defect detection accuracy.

[0053] Furthermore, this application also constructs a dynamic noise separation model jointly driven by vibration frequency and mechanical stress to determine the coupling boundary conditions of the time-varying interference intensity and the noise diffusion area, and then screens noise pixels based on the inverse change characteristics of visible light reflectivity and infrared radiation intensity, and finally extracts effective multi-band texture features from the denoised data.

[0054] In addition, this solution has made a breakthrough in converting valve mechanical status data into physical constraints for image noise separation, thereby improving the recognition accuracy of dust scattering noise while avoiding the mistaken elimination of real defect features.

[0055] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0057] Figure 1 A flowchart of a valve optical image enhancement and recognition method under severe weather conditions based on multispectral fusion provided in an embodiment of the present application;

[0058] Figure 2 A schematic diagram of the structure of a valve optical image enhancement and recognition system in severe weather conditions based on multispectral fusion provided in an embodiment of the present application;

[0059] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to enable people skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0061] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0062] Researchers have found that traditional optical inspection methods for industrial pipeline valves in severe weather conditions have problems such as severe interference from dust scattering noise and inaccurate defect feature extraction, and existing technologies lack dynamic correlation with the mechanical state of the valves. Based on this, an embodiment of the present application provides a method for optical image enhancement and recognition of valves in severe weather conditions based on multispectral fusion. This method constructs a noise separation model driven by physical properties by synchronously collecting multispectral images and mechanical dynamic response data, and adopts spatial-frequency domain collaborative correction and dynamic weight fusion technology to improve the recognition accuracy of valve surface defects in dust / rain and fog environments. The technical solution of the present application can be applied to online detection and health status assessment of pipeline valves in industrial fields such as petroleum and chemical industries under severe working conditions such as strong winds, dust, and high humidity.

[0063] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0064] Figure 1 The flowchart of a valve optical image enhancement recognition method under severe weather conditions based on multi-spectral fusion provided in an embodiment of the present application is as follows: Figure 1 As shown, the method includes:

[0065] Step 101: Acquire multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves under severe weather conditions.

[0066] In this step, severe weather conditions refer to strong winds and dust conditions with a dust concentration ≥ 200 μg / m³ and visibility ≤ 500 meters, or rain and fog conditions with rainfall intensity ≥ 10 mm / h and relative humidity ≥ 90%. The dust concentration is measured in real time using a laser scattering particulate matter monitor, visibility is measured using a transmission visibility meter, rainfall intensity is collected using a tipping bucket rain gauge, and relative humidity is detected using a capacitive humidity sensor. Multispectral optical image data includes optical image data in the visible light band and at least two infrared bands. Mechanical dynamic response characteristics include vibration frequency signals (in the range of 10-2000 Hz, collected using an accelerometer), surface mechanical stress distribution (measured using a strain gauge array with a resolution of ≥ 100 points / m²), and acoustic emission signals during valve opening and closing (in the frequency range of 20-100 kHz, collected using a broadband acoustic emission sensor). The sampling rate of the vibration frequency signal and acoustic emission signal is ≥ 10 kHz, and the time synchronization error between the mechanical stress distribution data and the multispectral optical image data is ≤ 1 ms.

[0067] In this embodiment, an image acquisition device installed near the valve acquires multispectral optical image data, while a vibration sensor and a stress sensor simultaneously capture mechanical dynamic response characteristics. The image acquisition device utilizes a multispectral camera, ensuring stable operation even in harsh environments. The vibration sensor utilizes a high-precision accelerometer for real-time monitoring of valve vibration. The stress sensor utilizes a distributed strain gauge array to accurately measure the stress distribution on the valve surface. All data acquisition devices utilize a time synchronization module to ensure temporal consistency.

[0068] For example, during severe sandstorms, a chemical plant monitors valves in its gas pipelines. A multispectral camera captures visible and infrared images of the valves. Simultaneously, an accelerometer and strain gauge array mounted on the valve flange capture the valve's vibration frequency signal and surface stress distribution data in the dusty environment. This data is transmitted to a central processor via an Industrial IoT gateway for further processing.

[0069] Step 102: Separate the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the reflective characteristics of the valve material.

[0070] In this step, the valve material reflection characteristics refer to the reflection patterns of different valve material surfaces to multispectral band light. Multiband texture features refer to the physically meaningful feature combinations extracted from multispectral images.

[0071] In an embodiment of the present application, based on the collected vibration frequency signal, its time variation curve is calculated to analyze the correlation between the vibration characteristics and noise; at the same time, the spatial gradient distribution is calculated based on the stress distribution data to determine the stress concentration area. These two features are input into the dynamic noise separation model. First, the time-varying interference intensity is calculated, then the spatial range of the noise diffusion area is determined, and finally the noise separation boundary conditions are obtained through coupled calculation. According to the boundary conditions, the visible light reflectivity and infrared radiation intensity of each pixel in the image are analyzed across bands, and the pixel set that meets the noise characteristics is screened out and removed, and finally an effective multi-band texture feature is extracted.

[0072] For example, using a valve in a chemical plant as an example, the processor calculated the time-varying curve of the dominant vibration frequency based on the collected vibration signal. It detected a sudden increase in vibration frequency during a certain period. Simultaneously, the stress distribution revealed stress concentration at the bottom of the valve. The system identified this period and area as areas of high noise interference, established a corresponding noise separation boundary, and successfully separated the dust interference adhering to the valve surface from the image, preserving the authentic valve texture.

[0073] Step 103: performing dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture features.

[0074] In this step, spatial domain amplitude correction refers to the adjustment of feature intensity in the image spatial dimension, and frequency domain amplitude correction refers to the adjustment of feature intensity in the frequency dimension.

[0075] In this embodiment, the energy distribution of texture features in the frequency domain is first analyzed to identify the dominant peak frequency. The degree of matching between the peak frequency and the natural frequency of the valve material is calculated as a frequency domain correction factor. Simultaneously, in the spatial domain, the anisotropic intensity ratio is calculated based on the direction of the maximum stress gradient distribution. In the spatial domain, the feature amplitude is enhanced along the direction of the maximum gradient, while the amplitude is weakened perpendicularly. In the frequency domain, irrelevant frequency components are suppressed based on the correction factor. Ultimately, a dual correction result is achieved that preserves the key features.

[0076] For example, the system detected a significant peak in a certain frequency band for the texture characteristics of a chemical plant valve, coinciding with the natural frequency of the stainless steel valve. Spatial analysis revealed significant anisotropy along the valve's principal stress direction. Based on this, the system enhanced this characteristic frequency band in the frequency domain and the features along the principal stress direction in the spatial domain, effectively highlighting possible cracks in the valve.

[0077] Step 104: allocating spectral fusion weights to the multispectral optical image data according to the corrected multi-band texture features.

[0078] In this step, the spectral fusion weight represents the parameter that determines the weight of different band images in the final fusion result. The modified multi-band texture feature refers to the optimized feature set after the double amplitude correction process.

[0079] In this embodiment, the reliability of each band's image is dynamically determined based on the corrected multi-band texture feature quality assessment results. In severe dust conditions, the weight of the visible light band is reduced, while the weight of the infrared band is increased. In high humidity conditions, the weight ratios of the different infrared bands are appropriately adjusted. This weight adjustment takes into account factors such as texture feature clarity and compatibility with the machine's state, achieving adaptive distribution through a pre-set weight calculation model.

[0080] For example, in the case of a chemical plant, the system automatically reduces the weight of the visible light band and increases the weight of the medium-wave infrared band in sandstorm weather; when stress abnormality is detected in a certain area of the valve, the weight ratio of the area in the long-wave infrared image is increased to ensure that the characteristics of the stress abnormality area are fully displayed.

[0081] Step 105: Based on the spectral fusion weight and the modified multi-band texture feature, an enhanced optical image matching the actual mechanical state of the valve is generated.

[0082] In this step, mechanical state matching indicates the consistency between the image features and the actual physical state of the valve. Enhanced optical image indicates the high-quality valve image after processing.

[0083] In this embodiment, image bands are fused at the pixel level according to assigned weights, and feature enhancement is performed on marked potential abnormal areas. The fusion process maintains a balance between texture detail and thermal radiation characteristics, ensuring that the resulting image clearly displays both surface topography and internal stress distribution. The resulting image shows a high degree of consistency between valve defect characteristics and mechanical condition data.

[0084] For example, the final enhanced image of the chemical plant valve clearly shows a tiny crack at the bottom of the valve. The location completely corresponds to the stress concentration area, and the direction of the crack is consistent with the direction of the stress gradient, verifying the reliability of the detection results.

[0085] This method, through the collaborative processing of multispectral images and mechanical dynamic responses, effectively overcomes noise interference in harsh weather conditions and enables accurate identification of valve surface defects. The resulting enhanced image not only clearly displays surface topography but also reflects internal mechanical state changes, providing a reliable basis for valve health assessment. The entire solution demonstrates excellent adaptability and accuracy in complex industrial environments.

[0086] To address the problem of insufficient accuracy in valve surface defect detection under severe weather conditions, in some embodiments, step 102 : separating the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the valve material reflectance characteristics, includes:

[0087] Step 201: constructing a dynamic noise separation model related to dust scattering interference, wherein the parameters of the dynamic noise separation model are jointly determined by the vibration frequency time variation curve and the mechanical stress spatial gradient distribution in the mechanical dynamic response characteristics.

[0088] In step 201, the dynamic noise separation model is used to distinguish between the physical driving model of the actual valve characteristics and the ambient noise. The vibration frequency time variation curve represents a characteristic curve reflecting the temporal variation of the valve's main vibration frequency. This vibration frequency time variation curve is converted into noise model parameters through the following steps: a short-time Fourier transform is performed on the raw vibration signal collected by the accelerometer to extract a time-varying curve of the amplitude of the main frequency component in the 1-100 Hz frequency band; the standard deviation of this curve within a sliding time window (0.1 seconds) is calculated as a time-varying interference intensity baseline value; the baseline value is normalized based on the valve structure's resonant frequency (pre-stored in the material database) to generate the time-varying interference intensity parameters used in the dynamic noise separation model. The spatial gradient distribution of mechanical stress represents the spatial distribution of the stress change rate on the valve surface. The spatial gradient distribution of mechanical stress is acquired by using an embedded fiber Bragg grating sensor array (spacing ≤ 2 cm) to measure the strain distribution on the valve surface in real time. The strain data are converted into stress values based on the stress-strain constitutive relationship of the valve material (pre-stored in the material database). Finally, the spatial gradient distribution is calculated through three-dimensional spline interpolation, with a resolution that matches the spatial sampling rate of the multispectral optical image at a 1:1 ratio.

[0089] In this embodiment, a vibration sensor acquires raw vibration signals, extracts the dominant frequency component through time-frequency analysis, and generates a time-varying curve. Simultaneously, the stress gradient distribution is calculated based on strain gauge array data. Both components are input into a physically constrained neural network model, resulting in a dynamic noise separation model that reflects the vibration-stress-noise relationship.

[0090] Step 202: Determine the time-varying interference intensity and noise diffusion area of the dust scattering noise in the multispectral optical image data according to the dynamic noise separation model.

[0091] In step 202, the time-varying interference intensity represents an intensity parameter that quantifies the temporal fluctuation of dust scattering noise. The noise diffusion area represents the spatial range affected by dust in the image.

[0092] In an embodiment of the present application, a trained model is used to analyze the vibration spectrum characteristics and stress gradient distribution at the current moment, and the time-varying interference intensity value is calculated; at the same time, the main noise diffusion path is determined according to the stress gradient direction, and the current noise area range is predicted in combination with historical diffusion data.

[0093] Step 203: performing coupling calculation on the time-varying interference intensity and the noise diffusion area to determine a noise separation boundary condition.

[0094] In step 203, the coupled calculation represents the fusion operation of the time domain and the spatial domain features. The noise separation boundary condition represents the decision boundary for defining the noise and the effective features.

[0095] In this embodiment, the time-varying interference intensity is mapped onto a spatial grid within the noise diffusion region, and a two-dimensional boundary distribution map is generated through spatial convolution. The core impact region is linearly superimposed, while the edge region is weighted fused based on the stress gradient direction, ultimately generating a physically meaningful noise separation boundary.

[0096] Step 204: Based on the noise separation boundary condition, a cross-band correlation analysis is performed on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold, and a set of all target pixels is used as the dust scattering noise component.

[0097] In step 204, visible light reflectance is a quantitative indicator of the valve surface's reflectivity for visible light with a wavelength of 400-700 nm. It is acquired through the visible light channel of a multispectral camera. The calculation uses the reflectance of a standard whiteboard as a benchmark, and the ratio of the valve surface's reflected light intensity to the incident light intensity is normalized to a reflectance value within the range of 0-1. This value is influenced by the valve material, surface condition, and environmental interference. In dusty environments, dust adhesion can cause an abnormal decrease in local reflectance. Infrared radiation intensity refers to the thermal radiation energy intensity of the valve surface in a specific infrared band (such as 3-5 μm or 8-12 μm). It is measured using an infrared thermal imager or infrared multispectral camera. This value reflects the combined characteristics of the valve surface temperature and emissivity. The calculation is based on the blackbody radiation law and incorporates sensor calibration parameters to convert the raw radiation signal into a standardized radiation intensity value. Dust coverage can alter local emissivity, leading to an abnormal increase in radiation intensity. Cross-band correlation analysis represents the collaborative processing of multispectral data. Target pixels represent a set of pixels that meet noise characteristics. The dust scattering noise component refers to the collection of optical interference features caused by dust particles in multispectral images. This component manifests as pixel regions with abnormally low reflectivity in the visible light band (dust obscuration) and abnormally high infrared radiation intensity (dust thermal radiation). This component is separated from the original image through cross-band correlation analysis. Its spatial distribution matches the noise diffusion region determined by mechanical vibration and stress data, making it the primary noise source that needs to be suppressed in subsequent image enhancement processing.

[0098] In this embodiment, for each spatial coordinate point, the visible light reflectance and infrared radiation intensity are simultaneously compared to determine their compliance with boundary conditions. A point is considered a noise pixel if both the visible light reflectance is below a dynamic threshold (affected by time-varying interference intensity) and the infrared radiation intensity is above a reference value (affected by the characteristics of the noise region).

[0099] Step 205: extracting multi-band texture features from the multispectral optical image data from which the dust scattering noise component has been removed.

[0100] In step 205, the multi-band texture feature represents a valid feature set after denoising.

[0101] In an embodiment of the present application, the texture features of each band (such as gray-level co-occurrence matrix, etc.) are extracted from the multispectral data after noise pixels are removed, and the correlation features between the bands (such as reflectance ratio, gradient correlation, etc.) are calculated to form a complete feature expression.

[0102] Here's a specific example:

[0103] In a real-world monitoring case involving gas pipeline valves at a chemical plant, during severe dust storms with visibility less than 300 meters, the system detected a drop in the valve's average visible light reflectance from a normal value of 0.65 to 0.32 (calculated as follows: reflectance = measured reflected light intensity / reflected light intensity from a standard whiteboard × an environmental attenuation coefficient of 0.9). Simultaneously, the long-wave infrared radiation intensity increased from the standard value of 0.45 to 0.83 (due to the thermal radiation of dust particles). Vibration signal analysis revealed abnormal vibration at the valve flange connection at 125 Hz (normal value: 80-100 Hz), with an amplitude of 0.8 mm / s² (calculated by integrating the accelerometer). Strain gauge data revealed a stress concentration zone at the bottom of the flange with a maximum value of 25 MPa (calculated using the strain-stress conversion formula σ = Eε, where E is the elastic modulus of the valve material). After inputting these parameters into the dynamic noise model, the system determined a time-varying interference intensity of 0.72 (calculated as: 0.5 × vibration anomaly coefficient + 0.5 × stress gradient coefficient). It also delineated a noise diffusion zone with a radius of 15 cm centered on the flange (determined by a safety factor of 1.5 times the stress gradient extension length). By setting a reflectivity threshold of 0.4 and a radiation intensity threshold of 0.7 (calculated based on the 3σ principle of historical data statistics), the system successfully isolated approximately 8% of the pixels (a total of 2,356 pixels) at the flange connection as sand and dust noise components. Ultimately, two radial cracks, measuring 3.2 mm and 4.5 mm in length, were clearly revealed in the denoised image.

[0104] In the embodiment of the present application, this method effectively solves the problem of inaccurate valve surface feature extraction in harsh environments through noise separation driven by mechanical state data. The generated texture features can not only clearly reflect the surface morphology, but also maintain a high degree of consistency with the actual mechanical state of the valve, providing a reliable basis for defect diagnosis.

[0105] In order to solve the problem of severe noise interference in valve surface defect identification in a dusty environment, in some embodiments, step 203: performing a cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold, includes:

[0106] Step 301: Based on the noise separation boundary conditions and the noise diffusion area, a mapping table of the inverse variation relationship between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates is established.

[0107] In step 301 , the reverse change relationship mapping table refers to a lookup table that records the corresponding relationship between the decrease in visible light reflectance and the increase in infrared radiation.

[0108] In this embodiment, the image space is first gridded according to noise separation boundary conditions. Within each grid cell, the correspondence between the decrease in visible light reflectance and the increase in infrared radiation is calculated. A continuous surface of changes is generated using a spatial interpolation algorithm, and a mapping table is established, with spatial coordinates as indexes and reflectance-radiation intensity change rates as values. This table is updated in real time, dynamically reflecting the noise characteristics of different areas.

[0109] Step 302: Based on the inverse change relationship mapping table, a joint judgment rule for the attenuation amplitude of visible light reflectivity and the increase in infrared radiation intensity is constructed. The boundary conditions of the joint judgment rule include a first threshold and a second threshold. The first threshold is obtained by dynamically calculating the difference between the average reflectivity of the visible light band in the absence of sand and dust interference and the current time-varying interference intensity. The second threshold is obtained by superimposing the baseline intensity of the infrared band within the normal radiation range of the valve material and the radiation increment of the noise diffusion area.

[0110] In step 302, the joint decision rule represents the noise identification logic that integrates visible light and infrared features. The first threshold represents the dynamic segmentation threshold for visible light reflectance. The second threshold represents the dynamic segmentation threshold for infrared radiation intensity. The visible light band refers to the spectral acquisition channel within the wavelength range of 400-700 nm. The visible light band reflectance represents the measured value of the valve surface's reflectivity of incident light within this band. The visible light reflectance attenuation amplitude refers to the decrease in the current reflectance compared to the non-interference state. The current time-varying interference intensity specifically refers to the instantaneous interference intensity value calculated based on the real-time vibration frequency mutation point at a specific moment or time period. It is the specific instantiation of "time-varying interference intensity" on the time axis. The time-varying interference intensity refers to the dynamic variation in the intensity of dust scattering noise in the time dimension, as determined by the time-varying vibration frequency curve. The relationship between the two is as follows: "current time-varying interference intensity" is the real-time value of "time-varying interference intensity" during the dynamic noise separation process. The former inherits the definition of the latter and adds a time point constraint. The infrared band refers to the spectral acquisition channel within the wavelength range of 700nm-1mm (including at least two sub-bands described above). The infrared band radiation intensity represents the measured value of the thermal radiation energy of the valve surface within that band. The infrared radiation intensity increase refers to the increase in the current radiation intensity compared to the baseline state.

[0111] In this embodiment, a two-variable joint probability model is established based on a mapping table analysis of the statistical distribution characteristics of reflectivity and radiation intensity. The first threshold is calculated by subtracting 0.8 times the time-varying interference intensity (output from step 202) from the baseline reflectivity (obtained from the valve material parameter library). The second threshold is calculated by adding 1.2 times the noise region radiation increment (estimated using a thermal diffusion model) to the baseline radiation intensity (calculated based on the valve operating temperature). Both thresholds are automatically calibrated every hour.

[0112] Step 303: Filter out target pixels whose reflectivity in the visible light band is lower than the first threshold and whose radiation intensity in the infrared band is higher than the second threshold.

[0113] In this embodiment, a parallel processing architecture is used to perform a triple check on each pixel: first, checking whether it is within the noise diffusion area, then verifying whether the visible light reflectance is below a first threshold, and finally confirming whether the infrared radiation is above a second threshold. Pixels that meet all three conditions are marked as target pixels, and all target pixels form a spatial distribution map of the noise component.

[0114] Here's a specific example:

[0115] In an actual monitoring of gas pipeline valves at a chemical plant, the system, based on established noise separation boundary conditions (a time-varying interference intensity of 0.72 and a noise diffusion radius of 15 cm), first constructed a mapping table for the inverse relationship between visible light reflectance and infrared radiation intensity. In the flange connection area, the measured visible light reflectance decreased from a baseline value of 0.65 to 0.28 (calculated as 0.65 × (1 - 0.72 time-varying interference coefficient) × 0.95 environmental correction factor), while the infrared radiation increased from a baseline value of 0.45 to 0.81 (calculated as 0.45 + 0.3 radiation increment × 1.2 stress concentration factor). Based on this mapping, the system dynamically set the first threshold to 0.38 (calculated as 0.65 baseline reflectance - 0.72 interference intensity × 0.38 adjustment factor) and the second threshold to 0.75 (calculated as 0.45 baseline radiation + 0.25 minimum radiation increment × 1.2 safety factor). By scanning a 1280×1024 pixel image of the flange area, 2150 target pixels (visible light reflectance <0.38 and infrared radiation >0.75) were selected. After morphological closing, they were confirmed to correspond to a 4.2 cm² dust-covered area. After noise removal, the image clearly revealed a 3.8 mm long crack on the flange sealing surface. Its location corresponded exactly to the 25 MPa stress concentration zone. The crack orientation was at an 18° angle to the principal stress direction, and the crack width was 0.15 mm (calculated by converting pixel size to actual size), providing accurate defect location for emergency valve maintenance.

[0116] In the embodiment of the present application, the method achieves accurate identification and separation of sand and dust noise through dual verification of the inverse change characteristics of multi-spectral features and mechanical status data, so that the display clarity of real defects on the valve surface meets the requirements of industrial inspection standards, and effectively supports the valve health status assessment work in harsh environments.

[0117] To address the problem of insufficient noise separation accuracy in multispectral images under dusty environments, in some embodiments, step 301: establishing a mapping table of inverse variations between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates based on the noise separation boundary conditions and the noise diffusion region, includes:

[0118] Step 401: converting the time-varying interference intensity in the noise separation boundary condition into a time-space matrix according to the amplitude mutation point distribution of the vibration frequency time variation curve.

[0119] In step 401, an amplitude mutation point represents a point in the vibration frequency curve where the amplitude exceeds 30% of the average amplitude. The space-time matrix represents a two-dimensional matrix reflecting the spatial distribution of time-varying interference intensity. The row vectors of the matrix represent the change in interference intensity along the time dimension, and the column vectors represent the distribution of interference intensity along the spatial dimension.

[0120] In this embodiment, we first identify amplitude mutation points in the vibration frequency curve. We then construct an initial matrix with the duration of the mutation points as rows and the spatial location as columns. We then spatially interpolate the time-varying interference intensity based on the impact range of the mutation points, ultimately generating a spatiotemporal matrix with the same resolution as the image. Each element in this matrix represents the quantized value of the time-varying interference at the corresponding pixel.

[0121] Step 402: Calculate a diffusion area ratio parameter at each spatial coordinate position according to the spatial coverage shape of the noise diffusion area.

[0122] In step 402, the spatial coverage shape represents the noise diffusion form determined by the direction of the mechanical stress gradient. The diffusion area ratio parameter represents the degree of noise influence in the local area.

[0123] In this embodiment, a circular region with a radius of 1.2 times the stress gradient extension length is taken as the center of each pixel, and the proportion of pixels in the region belonging to the noise diffusion area is calculated as the diffusion area ratio parameter. A bilinear interpolation algorithm is used to ensure the spatial continuity of the parameter.

[0124] Step 403: performing pixel-by-pixel multiplication of the spatiotemporal matrix and the diffusion area ratio parameter to generate a noise interference superposition weight map.

[0125] In step 403, the noise interference superposition weight map represents a weight distribution map of the integrated time-varying and space-varying interferences.

[0126] In the embodiment of the present application, a dot product operation is performed on the spatiotemporal matrix and the diffusion area ratio parameter, and the result is normalized to generate a weight map in the range of 0 to 1. The area with a weight value exceeding 0.7 is determined to be a high interference area, and the area with a weight value below 0.3 is determined to be a low interference area.

[0127] Step 404: Based on the gradient distribution direction in the noise interference superposition weight map, low-reflectivity pixels whose gradient distribution direction angle is greater than a set angle are identified in the visible light band reflectivity, and high-radiation intensity pixels whose gradient distribution direction angle is less than a set angle are identified in the infrared band radiation intensity.

[0128] In step 404, the set angle represents the allowable deviation angle determined based on the stress gradient direction. Low-reflectivity pixels are image pixels with lower-than-normal reflectivity in the visible light band. Their reflectivity values are typically below a dynamic threshold determined based on the valve material properties. The gradient distribution direction represents the primary spatial diffusion direction of noise interference. High-radiance pixels are image pixels with higher-than-normal radiance in the infrared band. Their radiance values typically exceed a dynamic threshold calculated based on the valve operating temperature and material emissivity.

[0129] In this embodiment, the gradient field of the weight map is first calculated to determine the main gradient direction. In visible light images, pixels with reflectivity below the mean and an angle between the gradient direction and the main direction greater than 25 degrees are selected; in infrared images, pixels with radiation intensity above the mean and an angle less than 15 degrees are selected.

[0130] Step 405: constructing an inverse change relationship mapping table using the spatial overlap between the low reflectivity pixels and the high radiation intensity pixels as index key values.

[0131] In step 405, the spatial overlap indicates the degree of spatial consistency between two features. The specific process for generating spatial overlap is as follows: a low-reflectivity pixel set (pixels filtered from the visible light band reflectivity whose reflectivity is below a dynamically calculated threshold) is used. A high-radiance pixel set (pixels filtered from the infrared band radiation intensity whose radiation intensity is above a dynamically calculated threshold) is used. A binary decision is performed for each spatial coordinate point: if the current coordinate exists in both the low-reflectivity pixel set and the high-radiance pixel set, it is marked as an overlapping pixel; otherwise, it is marked as a non-overlapping pixel. The spatial overlap at the current coordinate is calculated as the ratio of the number of overlapping pixels to the total number of pixels within the local area (3×3 pixel range) of the noise interference overlay weight map. When the spatial gradient direction of the mechanical stress coincides with the maximum gradient direction of the noise diffusion region, a weighting coefficient proportional to the gradient strength is applied to the overlap calculation result. When the time-varying rate of change of the vibration frequency exceeds a set fluctuation threshold, an attenuation compensation based on the time-varying interference intensity is applied to the overlap calculation result. The spatial overlap of all coordinates is linearly mapped to the interval [0, 1], and the normalized value is used as the index key value of the inverse change relationship mapping table. The index key value represents the query basis of the mapping table.

[0132] In an embodiment of the present application, the coexistence of low-reflectivity pixels and high-radiation-intensity pixels at each spatial coordinate point is counted, and the ratio of coexisting pixels within a 3×3 neighborhood is calculated as the degree of overlap, and a mapping relationship table of reflectivity-radiation-intensity change rate is constructed using this key value.

[0133] Here's a specific example:

[0134] In a case study monitoring gas pipeline valves at a chemical plant, the system first converted the time-varying interference intensity of 0.72 into a 1280×1024 resolution space-time matrix based on the vibration frequency mutation point (125Hz, lasting 8 minutes). The value of the flange region in this matrix was 0.65 (calculated as 0.72 × 0.9 position attenuation coefficient). Based on the noise diffusion area (a sector with a radius of 15 cm) determined by the stress distribution, the diffusion area ratio parameter was calculated for each pixel. The parameter near the flange bolt hole reached 0.85 (calculated as: actual diffusion area 12.8 cm² / theoretical maximum area 15 cm² × π). The space-time matrix was point-by-point multiplied by the diffusion parameter to generate a noise interference superposition weight map. The weight of the area above the flange reached 0.74 (0.65 × 0.85 × 1.34 stress gradient coefficient). Based on weighted graph gradient analysis, a set of pixels with a reflectivity of 0.26 (benchmark 0.65×0.4) and a gradient angle of 32° were identified in the visible light image (accounting for 7.2% of the total pixels); a set of pixels with a radiance of 0.83 (benchmark 0.45+0.38) and a gradient angle of 12° were identified in the infrared image (accounting for 8.5% of the total pixels). The spatial overlap between the two feature sets within a 3×3 neighborhood was calculated to be 0.71 (number of overlapping pixels / total number of pixels). A reverse change mapping table constructed based on this analysis showed that when the reflectivity dropped to 0.3, the corresponding radiance increased to 0.8±0.05.

[0135] In the embodiment of the present application, the method achieves accurate quantification and positioning of dust interference through vibration-stress coordinated noise modeling and spatial correlation analysis of multi-spectral features, so that the detection rate of real valve defects meets the requirements of industrial inspection standards, providing reliable technical support for equipment status assessment in harsh environments.

[0136] To address the problem of insufficient noise boundary recognition accuracy in a dusty environment, in some embodiments, step 203: coupling the time-varying interference intensity with the noise diffusion area to determine the noise separation boundary condition includes:

[0137] Step 501: Calculate the instantaneous impact value of the time-varying interference intensity based on the vibration frequency time variation curve.

[0138] In step 501, the instantaneous impact value represents the intensity of the time-varying interference intensity on the spatial point at a specific moment.

[0139] In this embodiment, a first-order difference is first applied to the vibration frequency time curve, extracting the time intervals where the amplitude change rate exceeds a set threshold as valid intervals. Within each valid interval, the product of the square of the vibration amplitude and the duration is used as the basic impact value. This is then multiplied by the valve structure transfer coefficient (obtained from pre-stored data of the finite element model) to obtain the instantaneous impact value matrix for each spatial point.

[0140] Step 502: Divide the noise diffusion area into a core influence area and an edge transition area according to the spatial gradient distribution of the mechanical stress.

[0141] In step 502, the core impact zone represents the area with the highest mechanical stress gradient and the most severe noise interference. The extent of the core impact zone is determined by the product of the extension length and the elastic modulus of the valve surface material. The edge transition zone represents the peripheral area where the noise impact gradually decreases.

[0142] In this embodiment, based on the spatial gradient distribution of mechanical stress, the region with a gradient exceeding 1.5 times the average gradient is defined as the core impact zone (approximately 30% of the area), with the remainder being the edge transition zone. The core impact zone boundary is morphologically expanded based on the gradient direction to ensure that the stress concentration area is fully contained.

[0143] Step 503: In the core influence area, convolution operation is performed on the time-varying interference intensity and the spatial stress gradient to generate the interference intensity of the core influence area, and expansion compensation is performed in the edge transition area to obtain the interference intensity of the edge transition area.

[0144] In step 503, the interference intensity within the core influence zone represents a quantitative indicator of the overall noise level in the core area. Extended compensation utilizes a nonlinear interpolation method based on the amplitude of the mechanical stress gradient. Within the edge transition zone, the compensation intensity is attenuated inversely proportional to the square of the distance from the boundary, using the stress gradient at the core influence zone boundary as a reference. Simultaneously, the first-order derivative of the vibration frequency time curve is superimposed as a dynamic adjustment factor. When the derivative exceeds a threshold, a morphological dilation operation (3×3 circular structuring element) is initiated to ensure spatial continuity within the noise diffusion region. The interference intensity within the edge transition zone is a quantitative representation of the gradual attenuation of the noise influence from the core area toward the periphery.

[0145] In this embodiment, a 3×3 convolution kernel is used within the core region to perform spatial convolution of the time-varying interference intensity and the stress gradient. The convolution kernel weight coefficient is dynamically adjusted based on the stress gradient direction. A distance-based exponential decay compensation algorithm is used within the edge transition region. The compensation intensity is inversely proportional to the distance from the core region boundary, with a maximum compensation amount of 0.6 times the core region interference intensity.

[0146] Step 504: According to the interference intensity of the core influence area and the interference intensity of the edge transition area, adjust the values of the parameters used to separate the core influence area and the edge transition area to obtain the noise separation boundary conditions.

[0147] In step 504, the noise separation boundary condition is implemented by a two-parameter adaptive threshold function: assuming the interference intensity in the core area is Ic and the interference intensity in the edge area is Ie, the boundary condition is α·Ic+(1-α)·Ie>T, where the weight coefficient α is determined by the consistency of the mechanical stress gradient direction (0.6-0.9), and the dynamic threshold T=κ·(1+e^(-βΔt)), κ is a material-related constant, β is the time-varying interference intensity attenuation rate, and Δt is the time difference between the current moment and the vibration mutation point.

[0148] In this embodiment, the core region interference intensity and edge region interference intensity are input into a dual-threshold adaptive model. The core region adopts a rigid boundary (threshold coefficient 1.0) and the edge region adopts a flexible boundary (threshold coefficient 0.7). The boundary parameters are adjusted through an iterative optimization algorithm to ensure that the final boundary is spatially continuous and geometrically consistent with the mechanical stress distribution.

[0149] Here's a specific example:

[0150] During actual monitoring of a gas pipeline valve at a chemical plant, the system calculated an instantaneous impact value of 0.78 based on the vibration frequency time curve (abnormal vibration at 125 Hz lasting 12 minutes, with an amplitude of 0.85 mm / s²). (Calculation formula: 0.5 × vibration amplitude coefficient + 0.5 × duration coefficient, where amplitude coefficient = measured amplitude / baseline amplitude = 0.85 / 0.6 = 1.42, and duration coefficient = actual duration / standard sampling duration = 12 / 10 = 1.2.) Based on the spatial gradient distribution of mechanical stress (maximum gradient 28 MPa / m below the flange, extending 10 cm), a circular area with a radius of 12 cm was defined as the core impact zone (area 113 cm², calculated as π × 12²), with an outer 3 cm annular zone as the edge transition zone. A 3×3 convolution kernel (weight coefficients [0.2, 0.3, 0.2; 0.3, 0.5, 0.3; 0.2, 0.3, 0.2]) was used to convolve the time-varying interference intensity with the stress gradient in the core region, resulting in an interference intensity of 0.82 in the core region (calculated as 0.78×0.7+28 / 35×0.3). Distance attenuation compensation was applied to the edge transition region (compensation amount = core intensity × e^(-0.15×distance)). The interference intensity measured 1 cm from the core region boundary was 0.58 (calculated as 0.82×e^(-0.15×1)). The resulting noise separation boundary conditions showed a core region threshold of 0.75 (calculated as a 0.82×0.9 safety factor) and an edge region threshold of 0.52 (calculated as 0.58×0.9).

[0151] In the embodiment of the present application, the method achieves accurate division of noise boundaries through spatiotemporal coupling calculation of vibration and stress characteristics, so that the accuracy of identifying real valve defects meets the requirements of industrial inspection standards, and effectively solves the technical difficulties of equipment status assessment in harsh environments.

[0152] To address the problem of insufficient fusion of the multispectral image and the valve mechanical state, in some embodiments, step 105: generating an enhanced optical image matching the actual mechanical state of the valve based on the spectral fusion weights and the modified multi-band texture features, includes:

[0153] Step 601: Dynamically adjust the spectrum fusion weight according to the mechanical dynamic response characteristics.

[0154] In this embodiment, the vibration signal's dominant frequency change rate and the gradient change of the stress distribution are first analyzed to calculate the reliability coefficients of each band image. The visible light band weight is calculated as 0.6 × vibration stability coefficient + 0.4 × stress uniformity coefficient; the infrared band weight is calculated as 0.7 × thermal radiation correlation coefficient + 0.3 × vibration-thermal coupling coefficient. The weights are dynamically updated every hour to ensure adaptability to changing operating conditions.

[0155] Step 602: Based on the corrected multi-band texture features, the multispectral optical image data is layered and fused according to the adjusted spectral fusion weights.

[0156] In step 602 , hierarchical fusion represents the integration of multispectral data hierarchically by feature importance.

[0157] In this embodiment, multispectral data is processed in three layers: a base layer (80% weight) that fuses visible light textures with infrared radiation distribution; a detail layer (15% weight) that enhances anisotropic features; and an anomaly layer (5% weight) that highlights stress concentration areas. Each layer utilizes a wavelet-based fusion algorithm, with fusion parameters dynamically adjusted based on texture clarity and compatibility with the mechanical state.

[0158] Step 603: Based on the mechanical stress spatial gradient distribution, mark potential structural abnormality areas in the layered fused multispectral optical image data.

[0159] In step 603 , potential structural anomaly regions represent high-risk areas where defects may exist.

[0160] In this embodiment, texture anomalies (such as fractures and mutations) are first identified in the fused image. Spatial matching analysis is then performed with the stress gradient distribution. When the spatial overlap between the texture anomaly and the high stress gradient region (gradient value > 25 MPa / m) exceeds 70%, it is marked as a potential structural anomaly. The marking intensity is divided into three levels, corresponding to different repair priorities.

[0161] Step 604: Generate an enhanced optical image that matches the actual mechanical state of the valve based on the multispectral optical image data marked with the potential structural abnormality region.

[0162] In step 604, the specific generation process of the enhanced optical image is as follows: the multispectral optical image data of the potential structural abnormality area is processed as follows: first, anisotropic diffusion filtering based on texture sharpening is applied to the visible light band (the diffusion coefficient is inversely proportional to the mechanical stress gradient amplitude); second, the radiation intensity of the infrared band is normalized (based on the standard value in the valve material radiation characteristic database); finally, the processed multi-band data is weightedly superimposed at the pixel level according to the spectral fusion weight, wherein the fusion weight of the abnormal area pixels is additionally increased by a dynamic compensation coefficient proportional to the vibration frequency amplitude, and the final enhanced optical image is output.

[0163] In this embodiment, the marked area is subjected to targeted enhancement processing: edge sharpness is enhanced in the visible light channel (by a factor of 1.2-1.5), and radiation contrast is enhanced in the infrared channel (by a factor of 1.3-1.8). Simultaneously, the natural transition of the non-abnormal area is maintained, ultimately generating an enhanced image that maintains optical authenticity while highlighting the associated features of the mechanical state.

[0164] Here's a specific example:

[0165] In monitoring a gas pipeline valve at a chemical plant, the system dynamically adjusted spectral fusion weights based on real-time vibration data (the main frequency suddenly increased from a normal 110Hz to 148Hz) and stress data (stress concentration at the bottom of the flange reached 28MPa). The visible light weight was set to 0.4 (calculated as: 0.6 × vibration stability coefficient 0.5 + 0.4 × stress uniformity coefficient 0.25), and the mid-wave infrared weight was set to 0.6. During the layered fusion process, a weight of 0.7 was used to fuse the visible light texture (reflectance 0.35) with the mid-wave infrared features (radiance 0.78). A weight of 0.25 was used in the detail layer to enhance anisotropy along the principal stress direction (azimuth angle 135°) (anisotropy ratio 2.8). A weight of 0.05 was used in the anomaly layer to highlight stress concentration areas (gradient value 26MPa / m). After fusion, three potential anomaly areas were identified on the flange sealing surface. One of these areas had an 85% overlap with the 32 MPa stress point (calculated as: number of overlapping pixels / total number of pixels in the stress area), marking it as a Level 1 anomaly. This area underwent targeted enhancement processing: a visible light edge sharpening factor of 1.3 (baseline value 1.0 + 0.3 × stress gradient factor), and a 1.4-fold increase in long-wave infrared contrast. The resulting enhanced image clearly showed two cracks on the flange sealing surface (4.2 mm and 3.5 mm long, 0.2 mm wide), with a stress of 35 MPa at the crack tip.

[0166] In the embodiment of the present application, the method realizes the precise correlation expression between the surface morphology characteristics of the valve and the internal stress state through multi-spectral fusion and directional enhancement guided by mechanical state. The generated enhanced image meets the requirements of visual interpretation and quantitative analysis at the same time, thereby improving the reliability of valve defect detection in harsh environments.

[0167] In order to solve the problem of insufficient matching between the multi-band texture features and the valve mechanical state, in some embodiments, step 103: performing dual amplitude correction of the multi-band texture features in the spatial domain and the frequency domain includes:

[0168] Step 701: Based on the mechanical dynamic response characteristics, extract the main peak value of the energy distribution of the multi-band texture characteristics in the frequency domain, and use the matching degree between the frequency component amplitude corresponding to the main peak value of the energy distribution and the natural vibration frequency of the valve material as the frequency domain amplitude correction coefficient.

[0169] In step 701, the main peak of the energy distribution represents the frequency component with the largest amplitude in the energy spectrum of the multi-band texture feature in the frequency domain. The frequency domain amplitude correction coefficient is a parameter that represents the degree of matching between the characteristic frequency and the natural frequency of the material.

[0170] In this embodiment, a fast Fourier transform (FFT) is first performed on multi-band texture features to obtain a frequency-domain energy spectrum. A peak detection algorithm is then used to identify the frequency component corresponding to the dominant peak in the energy distribution. This frequency component is then matched against a pre-stored database of valve material natural vibration frequencies, and the degree of match is calculated (formula: match = 1 - |measured frequency - natural frequency| / natural frequency). This is then used as a frequency-domain amplitude correction factor. This factor is then used for subsequent frequency-domain amplitude suppression.

[0171] Step 702: Calculate the anisotropic intensity ratio of the multi-band texture feature in the spatial domain according to the maximum gradient direction of the mechanical stress spatial gradient distribution.

[0172] In step 702, the maximum gradient direction indicates the direction with the largest mechanical stress spatial gradient amplitude. The anisotropic intensity ratio is the ratio of the average intensity of a multi-band texture feature parallel to the direction of the maximum mechanical stress gradient to the average intensity perpendicular to that direction. It is used to quantify the directional differences in texture features.

[0173] In this embodiment, the gray-level co-occurrence matrix of the texture features in the spatial domain is first calculated, and contrast features are extracted in the four directions of 0°, 45°, 90°, and 135°. Simultaneously, the maximum gradient direction (e.g., 120°) is extracted from the spatial gradient distribution of mechanical stress. The ratio of the average intensity of all directions with an angle less than 15° to the maximum gradient direction to the average intensity in the perpendicular direction is calculated as the anisotropic intensity ratio.

[0174] Step 703: In the spatial domain, the amplitude of the multi-band texture feature is enhanced along the maximum gradient direction in proportion to the anisotropic intensity ratio, and an attenuation correction is performed inversely proportional to the anisotropic intensity ratio in the direction perpendicular to the maximum gradient. In the frequency domain, amplitude suppression inversely proportional to the frequency domain amplitude correction coefficient is applied to the multi-band texture feature to obtain a modified multi-band texture feature.

[0175] In step 703, the enhancement correction represents the texture enhancement along the main stress direction, and the amplitude suppression represents the attenuation of non-material characteristic frequencies.

[0176] In this embodiment, a directional filter bank is used in the spatial domain. Bandpass filtering (passband width = anisotropy intensity ratio × 15°) is performed along the direction of maximum gradient to enhance texture features in that direction. Low-pass filtering is performed perpendicularly to the direction to suppress irrelevant details. In the frequency domain, the energy spectrum is weighted with a weighting factor = (1 - frequency domain amplitude correction factor) × 0.8 + 0.2 to retain the characteristic frequency components of the material and suppress other frequency components. Finally, an inverse transform is performed to obtain the corrected multi-band texture features.

[0177] Here's a specific example:

[0178] In a case study monitoring gas pipeline valves at a chemical plant, the system performed frequency domain analysis on de-noised multi-band texture features and found that the main energy peak was located at 126 Hz (the natural frequency of stainless steel valves is 128 Hz), with a matching degree of 0.98 (calculated as 1-|126-128| / 128). The stress distribution revealed a stress concentration area (gradient value of 24 MPa / m) around the valve flange bolt hole, with a maximum gradient at 50°. Spatial domain analysis measured the average texture intensity along the 50° direction to be 82, and 35 along the perpendicular direction, resulting in an anisotropic intensity ratio of 2.34 (calculated as 82 / 35). During the correction process, the 120-132 Hz frequency band was retained in the frequency domain (weighting factor 0.81, calculated as (1-0.98) × 0.8 + 0.2 = 0.81), while other frequency bands were attenuated to 30% of their original amplitude. In the spatial domain, directional enhancement was performed along a 50° passband of 50° ± 17.55° (2.34 × 7.5°), and Gaussian smoothing with a value of σ = 2.34 was applied in the vertical direction. The processed image clearly revealed two defects on the flange sealing surface: a 3.5 mm long radial crack (deviation 3° from the main stress direction) and a 2.1 mm diameter erosion pit (centered at the 26 MPa stress point).

[0179] In an embodiment of the present application, the method uses space-frequency domain collaborative correction guided by mechanical state, so that the texture features not only retain the inherent frequency domain characteristics of the material, but also strengthen the spatial characteristics related to stress distribution, thereby improving the accuracy and reliability of valve defect detection.

[0180] Figure 2 A schematic diagram of the structure of a valve optical image enhancement recognition system under severe weather conditions based on multi-spectral fusion provided in an embodiment of the present application is shown as follows: Figure 2 As shown, the system includes:

[0181] The acquisition module 21 is used to acquire multi-spectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves under severe weather conditions.

[0182] The separation module 22 is configured to separate the multispectral optical image data based on the mechanical dynamic response characteristics to extract multi-band texture features associated with the reflective characteristics of the valve material.

[0183] The correction module 23 is used to perform dual amplitude correction in the spatial domain and the frequency domain on the multi-band texture features.

[0184] The allocating module 24 is configured to allocate spectral fusion weights to the multispectral optical image data according to the corrected multi-band texture features.

[0185] The generating module 25 is configured to generate an enhanced optical image that matches the actual mechanical state of the valve based on the spectral fusion weight and the modified multi-band texture feature.

[0186] Figure 2 The valve optical image enhancement recognition system based on multi-spectral fusion in severe weather conditions can be performed Figure 1 The implementation principles and technical effects of the multispectral fusion-based method for enhancing optical image recognition of valves in severe weather conditions, as described in the illustrated embodiment, will not be elaborated upon. The specific manner in which the various modules and units in the multispectral fusion-based system for enhancing optical image recognition of valves in severe weather conditions operate has been described in detail in the related embodiments and will not be further elaborated upon here.

[0187] In one possible design, Figure 2 The valve optical image enhancement recognition system based on multi-spectral fusion in severe weather conditions in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0188] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0189] The processing component 32 is as follows Figure 1 The embodiment provides a valve optical image enhancement and recognition method under severe weather conditions based on multi-spectral fusion.

[0190] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0191] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0192] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0193] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0194] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0195] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0196] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a valve optical image enhancement and recognition method under severe weather conditions based on multi-spectral fusion.

[0197] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0198] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0199] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A valve optical image enhancement recognition method based on multi-spectral fusion in severe weather conditions, characterized by: include: Acquire multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions; Constructing a dynamic noise separation model related to dust scattering interference, wherein the parameters of the dynamic noise separation model are jointly determined by the vibration frequency time variation curve and the mechanical stress spatial gradient distribution in the mechanical dynamic response characteristics; determining, according to the dynamic noise separation model, the time-varying interference intensity and the noise diffusion area of the dust scattering noise in the multispectral optical image data; Couple the time-varying interference intensity with the noise diffusion area to determine a noise separation boundary condition; performing a cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition, so as to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold, and treating a set of all target pixels as a dust scattering noise component; extracting multi-band texture features from the multispectral optical image data from which the dust scattering noise component is removed; Based on the mechanical dynamic response characteristics, the main peak value of the energy distribution of the multi-band texture feature in the frequency domain is extracted, and the matching degree between the frequency component amplitude corresponding to the main peak value of the energy distribution and the natural vibration frequency of the valve material is used as the frequency domain amplitude correction coefficient; Calculating the anisotropic intensity ratio of the multi-band texture feature in the spatial domain according to the maximum gradient direction of the mechanical stress spatial gradient distribution; In the spatial domain, an enhancement correction is performed on the amplitude of the multi-band texture feature along the maximum gradient direction in proportion to the anisotropic intensity ratio, and an attenuation correction is performed inversely proportional to the anisotropic intensity ratio in a direction perpendicular to the maximum gradient direction. In the frequency domain, an amplitude suppression is applied inversely proportional to the frequency domain amplitude correction coefficient to the multi-band texture feature to obtain a corrected multi-band texture feature. assigning a spectral fusion weight to the multispectral optical image data according to the corrected multi-band texture features; Based on the spectral fusion weights and the modified multi-band texture features, an enhanced optical image matching the actual mechanical state of the valve is generated.

2. The method according to claim 1, characterized in that The step of performing cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold includes: Based on the noise separation boundary conditions and the noise diffusion area, a mapping table of inverse variation relationships between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates is established; Based on the inverse change relationship mapping table, a joint determination rule for the attenuation amplitude of visible light reflectivity and the increase in infrared radiation intensity is constructed. The boundary conditions of the joint determination rule include a first threshold and a second threshold. The first threshold is obtained by dynamically calculating the difference between the average reflectivity of the visible light band in the absence of dust interference and the current time-varying interference intensity. The second threshold is obtained by superimposing the baseline intensity of the infrared band within the normal radiation range of the valve material and the radiation increment of the noise diffusion area. Target pixels whose reflectivity in the visible light band is lower than the first threshold and whose radiation intensity in the infrared band is higher than the second threshold are screened out.

3. The method according to claim 2, characterized in that The establishing of a mapping table of inverse variation relationships between the visible light band reflectivity and the infrared band radiation intensity at the same spatial coordinates based on the noise separation boundary condition and the noise diffusion area includes: Converting the time-varying interference intensity in the noise separation boundary condition into a space-time matrix according to the amplitude mutation point distribution of the vibration frequency time variation curve; Calculating a diffusion area ratio parameter at each spatial coordinate position according to the spatial coverage shape of the noise diffusion area; Performing a pixel-by-pixel multiplication operation on the space-time matrix and the diffusion area ratio parameter to generate a noise interference superposition weight map; Based on the gradient distribution direction in the noise interference superposition weight map, low reflectivity pixels whose gradient distribution direction angle is greater than a set angle are identified in the visible light band reflectivity, and high radiation intensity pixels whose gradient distribution direction angle is less than a set angle are identified in the infrared band radiation intensity; A reverse change relationship mapping table is constructed using the spatial overlap between the low reflectivity pixels and the high radiation intensity pixels as index key values.

4. The method according to claim 1, wherein The coupling calculation of the time-varying interference intensity and the noise diffusion area to determine the noise separation boundary condition includes: Calculating the instantaneous impact value of the time-varying interference intensity based on the vibration frequency time variation curve; dividing the noise diffusion area into a core influence area and an edge transition area according to the spatial gradient distribution of the mechanical stress; In the core influence area, a convolution operation is performed on the time-varying interference intensity and the spatial stress gradient to generate the interference intensity of the core influence area, and expansion compensation is performed in the edge transition area to obtain the interference intensity of the edge transition area; According to the interference intensity of the core influence area and the interference intensity of the edge transition area, the values of the parameters used to separate the core influence area and the edge transition area are adjusted to obtain the noise separation boundary conditions.

5. The method according to claim 1, wherein The step of generating an enhanced optical image matching the actual mechanical state of the valve based on the spectral fusion weight and the modified multi-band texture feature includes: Dynamically adjusting the spectral fusion weight according to the mechanical dynamic response characteristics; According to the corrected multi-band texture features, the multispectral optical image data is layered and fused according to the adjusted spectral fusion weights; Based on the mechanical stress spatial gradient distribution, marking potential structural abnormality areas in the layered fused multispectral optical image data; Based on the multispectral optical image data marked with the potential structural abnormality area, an enhanced optical image matching the actual mechanical state of the valve is generated.

6. A valve optical image enhancement recognition system based on multi-spectral fusion in severe weather conditions, characterized by: include: An acquisition module is used to obtain multispectral optical image data and mechanical dynamic response characteristics of industrial pipeline valves in severe weather conditions; a separation module, configured to construct a dynamic noise separation model related to dust scattering interference, wherein parameters of the dynamic noise separation model are jointly determined by a vibration frequency time variation curve and a mechanical stress spatial gradient distribution in the mechanical dynamic response characteristics; determining, according to the dynamic noise separation model, the time-varying interference intensity and the noise diffusion area of the dust scattering noise in the multispectral optical image data; Couple the time-varying interference intensity with the noise diffusion area to determine a noise separation boundary condition; performing a cross-band correlation analysis on the visible light band reflectance and the infrared band radiation intensity at the same spatial coordinates in the multispectral optical image data according to the noise separation boundary condition, so as to screen out target pixels whose visible light band reflectance is lower than a first threshold and whose infrared band radiation intensity is higher than a second threshold, and treating a set of all target pixels as a dust scattering noise component; extracting multi-band texture features from the multispectral optical image data from which the dust scattering noise component is removed; a correction module for extracting a main peak value of the energy distribution of the multi-band texture feature in the frequency domain based on the mechanical dynamic response characteristics, and using the matching degree between the frequency component amplitude corresponding to the main peak value of the energy distribution and the natural vibration frequency of the valve material as a frequency domain amplitude correction coefficient; Calculating the anisotropic intensity ratio of the multi-band texture feature in the spatial domain according to the maximum gradient direction of the mechanical stress spatial gradient distribution; In the spatial domain, an enhancement correction is performed on the amplitude of the multi-band texture feature along the maximum gradient direction in proportion to the anisotropic intensity ratio, and an attenuation correction is performed inversely proportional to the anisotropic intensity ratio in a direction perpendicular to the maximum gradient direction. In the frequency domain, an amplitude suppression is applied inversely proportional to the frequency domain amplitude correction coefficient to the multi-band texture feature to obtain a corrected multi-band texture feature. an allocating module, configured to allocate spectral fusion weights to the multispectral optical image data according to the corrected multi-band texture features; A generating module is used to generate an enhanced optical image that matches the actual mechanical state of the valve based on the spectral fusion weight and the modified multi-band texture feature.

7. A computing device, characterized in that The invention comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the optical image enhancement and recognition method of valves in severe weather based on multispectral fusion as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for enhancing the optical image recognition of valves in severe weather based on multispectral fusion according to any one of claims 1 to 5 is implemented.

Images