An image recognition-based corrosion typing method and system

Abstract

The application relates to the technical field of image recognition, and particularly discloses a corrosion typing method and system based on image recognition, which comprises the following steps: collecting a to-be-detected image of a pipeline inner wall, and dividing the to-be-detected image into multiple local areas; quantitatively evaluating optical interference backgrounds of the local areas affected by a deposition layer, and generating background description information for representing the coverage state of the deposition layer of the local areas; adjusting algorithm parameters of image preprocessing according to the background description information, and performing image preprocessing on each local area to suppress or compensate for optical interference caused by the deposition layer; extracting corrosion feature information from the preprocessed local areas; and combining the background description information and the corrosion feature information to judge the corrosion type of the local areas. Through self-adaptive preprocessing based on quantitative evaluation, the application penetrates or corrects the interference of the deposition layer, and reduces the risk of missed detection caused by features being shielded or distorted.

Description

Technical Field

[0001] This application relates to the field of image recognition technology, and in particular to an image recognition-based erosion classification method and system. Background Technology

[0002] In the daily operation and safety assurance of nuclear power plants, the health status of long-distance pipelines is of paramount importance. To promptly detect and address corrosion within these pipelines, existing technology has introduced an automated inspection system based on image recognition. This system typically consists of an inspection robot carrying a high-resolution industrial camera and illumination equipment, which acquires images along the inner wall of the pipeline and transmits them to a rear analysis station. The analysis station's recognition program learns from a large number of standard corrosion photographs, mastering the texture, color, and shape characteristics of various corrosion types (such as pitting corrosion, crevice corrosion, uniform corrosion patches, and crack corrosion), thereby determining the type and severity of corrosion.

[0003] However, in the actual operating environment of long-distance nuclear power transmission pipelines, the inner walls of the pipelines are generally covered by complex deposits of varying thickness, color, and texture. These deposits are formed by the deposition and solidification of minerals, metal ions, and chemical agents in the water under high temperature and pressure. Their distribution is uneven; a thin and dense iron oxide layer may form in straight pipe sections, while loose and unevenly thick deposits may accumulate at bends, welds, or near valves. This ever-changing deposit layer severely interferes with optical imaging, causing the camera to capture not the true surface of the metal pipe itself, but the surface covered by the deposit layer.

[0004] The presence of such deposits poses a significant challenge to existing image recognition methods. First, the deposit layer can obscure, distort, or even completely cover the true corrosion morphology of the underlying layer. For example, typical pitting corrosion may be filled in by deposits, appearing only as slightly darker spots on the image, easily misidentified as harmless stains; highly damaging crack corrosion may also disappear completely due to the deposit layer, leading to catastrophic missed detections. Second, the deposits themselves may form pseudo-features resembling corrosion, such as pits or textures, during crystallization or peeling, causing the recognition program to generate numerous false alarms. Therefore, in practical applications, existing image recognition-based automated inspection systems suffer from a sharp decline in reliability due to recognition failures caused by actual changes in the surface condition of the pipe's inner wall, facing a large number of missed detections and misjudgments. Summary of the Invention

[0005] This application proposes a corrosion classification method and system based on image recognition, aiming to solve the technical problem of missed detection and misjudgment caused by image recognition of complex deposition layers in the internal inspection of long-distance nuclear power transmission pipelines.

[0006] In a first aspect, this application provides an image recognition-based corrosion typing method for detecting the interior of long-distance nuclear power transmission pipelines. The method includes the following steps:

[0007] Acquire an image of the inner wall of the pipe to be inspected, and segment the image into multiple local regions;

[0008] The optical interference background of the local area affected by the deposition layer is quantitatively evaluated to generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes information on the degree of light occlusion, information on color performance differences, information on the complexity of surface structure, and information on image sharpness.

[0009] Based on the background description information, the algorithm parameters for image preprocessing are adjusted, and image preprocessing is performed on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters for image preprocessing includes: adjusting the parameters of the brightness and contrast enhancement algorithm based on the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information; and adjusting the parameters of the texture smoothing algorithm based on the surface structure complexity information;

[0010] Extract corrosion feature information from the preprocessed local area;

[0011] By combining the background description information and the corrosion feature information, the corrosion type of the local area is determined.

[0012] According to some embodiments of this application, the light occlusion information is obtained by calculating the relative attenuation ratio of the average pixel brightness of the local area to a preset brightness reference value; the color performance difference information is obtained by calculating the difference between the average color vector of the local area and a preset color vector in a predetermined color space; the surface structure complexity information is obtained by extracting the texture features of the local area and calculating the statistical complexity index of the texture features; and the image sharpness information is obtained by analyzing the intensity of image gradient changes in the local area.

[0013] According to some embodiments of this application, the step of acquiring an image of the inner wall of the pipe to be detected and segmenting the image into multiple local regions further includes:

[0014] Simultaneously collect temperature and pressure parameters inside the pipeline;

[0015] After the step of quantifying and evaluating the optical interference background affected by the deposition layer in the local area and generating background description information to characterize the coverage state of the deposition layer in the local area, the method further includes:

[0016] Based on the temperature and pressure parameters, physical predictions of the optical performance of the deposited layer are generated; wherein, the physical predictions include predicted light occlusion and predicted image sharpness.

[0017] If the difference between the predicted light occlusion degree and the light occlusion degree information exceeds a preset occlusion degree threshold, and / or the difference between the predicted image clarity and the image clarity information exceeds a preset clarity threshold, then based on the tunable multispectral light source and industrial camera, multiple different wavelengths of illumination light are scanned on the local area, and reflected light images at each wavelength are acquired simultaneously.

[0018] Based on the reflected light image, a characteristic spectrum of the reflection intensity of the local region as a function of wavelength is generated and used to correct the background description information.

[0019] According to some embodiments of this application, the step of generating physical predictions of the optical performance of the deposited layer based on the temperature and pressure parameters includes:

[0020] If, within a preset time period, the temperature drop and pressure fluctuation both exceed the corresponding preset threshold values, then based on the temperature drop and fluctuation, and according to preset physical association rules, physical prediction values ​​including the predicted light occlusion degree of the deposition layer and the predicted image clarity are generated.

[0021] According to some embodiments of this application, the correction process for the background description information includes:

[0022] Acquire a pre-stored first reference spectrum characterizing the deposition layer in a stable optical state and a second reference spectrum characterizing the deposition layer in a transient optical deviation state;

[0023] The similarity between the feature spectrum and the first reference spectrum and the second reference spectrum is calculated respectively to obtain the first similarity and the second similarity.

[0024] If the second similarity is higher than the first similarity and higher than the preset similarity threshold, then the second similarity is used as a transient influence parameter.

[0025] The corrected light occlusion information is obtained by using a preset first correction formula; the first correction formula is: L'=L+(1-L)×T×α; where L' is the corrected light occlusion information, L is the light occlusion information before correction, T is the transient influence parameter, and α is the preset first correction factor;

[0026] The corrected image sharpness information is obtained through a preset second correction formula; the second correction formula is: C'=CC×T×β; where C' is the corrected image sharpness information, C is the image sharpness information before correction, T is the transient influence parameter, and β is the preset second correction factor;

[0027] The corrected light occlusion information and / or image sharpness information are then used to update the background description information to obtain the corrected background description information.

[0028] According to some embodiments of this application, the steps of adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information include:

[0029] A probe wave is emitted into the local area, and an echo signal is received;

[0030] The echo signal is analyzed to determine the microstructure state of the deposition layer in the local area and the state-influencing parameters of the microstructure state.

[0031] The microstructure state and the state influence parameters are input into a preset fuzzy kernel parameter mapping model for calculation to obtain a fuzzy kernel function; wherein, the fuzzy kernel parameter mapping model includes the correspondence between fuzzy kernel type and microstructure state, the functional relationship between fuzzy kernel strength and state influence parameters, and the calculation rules for fuzzy kernel spatial distribution parameters.

[0032] According to some embodiments of this application, the step of analyzing the echo signal to determine the microstructure state of the deposition layer in the local region and the state influence parameters of the microstructure state includes:

[0033] Calculate the amplitude attenuation rate and / or spectral characteristic changes of the echo signal;

[0034] If the amplitude attenuation rate exceeds a first preset threshold, and / or the spectral characteristic change indicates that the high-frequency component has attenuation or resonance in a predetermined mode, then the microstructure state of the deposition layer in the local area is determined to be the presence of micron-sized bubbles or density inhomogeneity, and the state influence parameters are calculated based on the amplitude attenuation rate and / or spectral characteristic change.

[0035] According to some embodiments of this application, the step of extracting corrosion feature information from the preprocessed local region includes:

[0036] Based on the background description information, the feature extraction strategy and sensitivity are adaptively adjusted to extract the suspected corrosion area and corrosion feature information of the suspected corrosion area from the preprocessed local area. The corrosion feature information includes: local feature points and descriptors obtained by multi-scale feature extraction algorithm; color features of corrosion products obtained by color space analysis; and geometric features obtained by shape analysis.

[0037] According to some embodiments of this application, the step of determining the corrosion type of the local area by combining the background description information and the corrosion characteristic information includes:

[0038] The local feature points, descriptors, color features, and geometric features extracted from the suspected corrosion area are integrated with the background description information to form a feature evidence vector;

[0039] The feature evidence vector is input into a preset classification inference model for calculation to obtain the confidence level of the suspected corrosion area belonging to various corrosion types;

[0040] The corrosion type with the highest confidence level is identified as the candidate type for the suspected corrosion region;

[0041] Based on the light occlusion information and the image clarity information, the confidence threshold used to confirm the corrosion type is dynamically adjusted;

[0042] For suspected corrosion areas where the confidence level of the candidate type is higher than the adjusted confidence level threshold: if the component values ​​of the color feature in the preset color space all fall within the color threshold range after the dynamic adjustment of the color performance difference information, and the difference between the texture feature complexity index extracted from the suspected corrosion area and the surface structure complexity information is less than the preset texture consistency threshold, then it is determined to be a pseudo-deposit area and is excluded; otherwise, the candidate type is taken as the final corrosion type of the suspected corrosion area.

[0043] Secondly, this application also provides an image recognition-based corrosion typing system for detecting the interior of long-distance nuclear power transmission pipelines. The system includes:

[0044] The image acquisition and region segmentation module is used to acquire the image to be detected on the inner wall of the pipe and segment the image to be detected into multiple local regions;

[0045] The information generation module is used to quantitatively evaluate the optical interference background of the local area affected by the deposition layer, and generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes light occlusion information, color performance difference information, surface structure complexity information, and image sharpness information.

[0046] An image processing module is used to adjust the algorithm parameters of image preprocessing according to the background description information, and to perform image preprocessing on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters of image preprocessing includes: adjusting the parameters of the brightness and contrast enhancement algorithm according to the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of the deblurring algorithm according to the image sharpness information; and adjusting the parameters of the texture smoothing algorithm according to the surface structure complexity information;

[0047] The feature extraction module is used to extract corrosion feature information from the preprocessed local area;

[0048] The corrosion determination module is used to determine the corrosion type of the local area by combining the background description information and the corrosion feature information.

[0049] The technical solution according to the embodiments of this application has at least the following beneficial effects:

[0050] This application, through adaptive preprocessing based on quantitative assessment, effectively penetrates or corrects the interference of sediment layers on light, color, texture, and clarity, improving the input conditions for corrosion feature extraction and reducing the risk of missed detections due to feature occlusion or distortion. By incorporating background description information into the final judgment, it can more intelligently weigh evidence, effectively distinguishing between genuine corrosion and false features formed by the sediment itself, thereby reducing misjudgments. Overall, this application upgrades the detection process from a passive, one-size-fits-all approach to the environment to an intelligent diagnosis that proactively senses and adapts to the environment, enhancing the robustness and accuracy of corrosion identification in the extremely complex and variable sediment layer environment of nuclear power pipelines, providing a more reliable technical guarantee for pipeline safety.

[0051] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0052] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0053] Figure 1 This is a flowchart illustrating an image recognition-based erosion classification method provided in an embodiment of this application.

[0054] Figure 2 This is a flowchart illustrating the process of adjusting the parameters and intensity of the deblurring algorithm in an embodiment of this application.

[0055] Figure 3 This is a flowchart illustrating the process of determining the corrosion type of the local area in an embodiment of this application.

[0056] Figure 4 This is a schematic diagram of the architecture of an image recognition-based erosion classification system provided in an embodiment of this application. Detailed Implementation

[0057] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0058] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0059] In nuclear power plant operation, the health of long-distance pipelines is crucial. Existing technologies typically employ automated inspection systems based on image recognition. These systems use inspection robots to collect images of the pipeline's inner wall, and a background algorithm identifies the type of corrosion based on pre-learned corrosion features. However, in real-world operating environments, the inner walls of pipelines are generally covered by a complex deposition layer with varying thickness, color, and texture. This deposition layer severely interferes with optical imaging: on the one hand, it can obscure, distort, or even completely cover the true corrosion morphology of the underlying layer, leading to missed detections; on the other hand, the deposits themselves may form pseudo-features resembling corrosion, causing misjudgments. Therefore, existing recognition methods that rely on training with fixed standard models show a significant decrease in reliability when faced with such complex and variable deposition layer interference, making it difficult to meet the requirements for accurate detection.

[0060] In this regard, such as Figure 1 As shown, this application discloses an image recognition-based corrosion typing method, applied to the internal inspection of long-distance nuclear power transmission pipelines. The method includes the following steps:

[0061] S110, Acquire the image to be detected on the inner wall of the pipe, and segment the image to be detected into multiple local regions;

[0062] S120, quantify and evaluate the optical interference background of the local area affected by the deposition layer, and generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes light occlusion information, color performance difference information, surface structure complexity information, and image sharpness information.

[0063] S130, based on the background description information, adjust the algorithm parameters of the image preprocessing and perform image preprocessing on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters of the image preprocessing includes: adjusting the parameters of the brightness and contrast enhancement algorithm based on the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information; and adjusting the parameters of the texture smoothing algorithm based on the surface structure complexity information;

[0064] S140, extract corrosion feature information from the preprocessed local area;

[0065] S150, combining the background description information and the corrosion feature information, the corrosion type of the local area is determined.

[0066] In this application, "image to be inspected" refers to raw image data acquired from the inner wall of a pipe using equipment such as an industrial camera. "Local region" refers to dividing the entire image to be inspected into smaller, independently analyzable image blocks. "Deposition layer" refers to the accumulation of various substances on the inner wall of the pipe; its complex and variable optical properties are a major factor contributing to difficulties in image recognition. "Background description information" is comprehensive data used to quantitatively assess the optical interference of the deposition layer and characterize its coverage state. "Corrosion feature information" refers to visual features extracted from the image used to identify the type and degree of corrosion. "Corrosion classification" refers to classifying corrosion phenomena into different types based on the extracted feature information.

[0067] The acquisition of images to be inspected can be accomplished by an inspection robot deployed inside the pipeline, carrying a high-resolution industrial camera. For example, the robot can move along the pipeline axis at a preset speed and periodically trigger the camera to capture images. The acquired raw images are usually high-resolution, and to facilitate subsequent processing, these images are segmented into multiple smaller local regions. One segmentation method is to use a fixed-size grid to uniformly divide the image into several rectangular regions.

[0068] Information on the degree of light shading can be obtained by analyzing the distribution pattern of the brightness histogram in a local area. Deposition layer coverage usually causes the brightness distribution to shift to the left (dark areas) or become distorted. By quantifying the degree of deviation of this distribution pattern from the standard histogram of a clean surface, the severity of light shading can be assessed.

[0069] Information on color representation differences can be obtained through dominant color analysis or color clustering techniques. For example, color clustering (such as K-means clustering) can be performed on a local area image to extract its dominant hue (such as the center values ​​of 1-3 main color clusters), and then compared with a pre-stored reference color set representing the dominant hue of a clean metal surface. The difference in color representation can be quantified by calculating the minimum matching distance or overall color difference of the dominant hue in a color space (such as Lab color space).

[0070] Information on the complexity of surface structures can be obtained through frequency domain texture analysis. For example, a two-dimensional Fourier transform can be performed on a local area image to analyze the distribution of its spectral energy. When the deposited layer surface is rough and has complex texture, its spectral energy is often more dispersed, with a greater distribution in the high-frequency region. The proportion of spectral energy within a preset frequency band (such as mid-to-high frequencies) or the entropy value of the radial average energy distribution curve of the spectrum can be calculated as an indicator of surface structure complexity.

[0071] Image sharpness information can be obtained through frequency domain analysis or methods based on image autocorrelation / mutual information. For example, the total energy of high-frequency components (such as frequencies above a preset frequency threshold) in a local region of the image after Fourier transform can be calculated. The more blurred the image, the more severe the loss of high-frequency details, and the lower the corresponding total high-frequency energy. Therefore, this high-frequency energy value can be directly used as a negative correlation indicator of image sharpness (or can be converted into a positively correlated sharpness indicator through a mapping function).

[0072] If the light occlusion information indicates low brightness in a local area, the gain coefficient of the brightness enhancement algorithm can be increased. If the color difference information indicates significant color deviation in a local area, the parameters of the color correction algorithm can be adjusted. For image sharpness information, if a local area is blurred, the parameters of the deblurring algorithm can be adjusted, such as choosing an appropriate blur kernel size and deblurring intensity. When the image sharpness information indicates moderate blurring, a Gaussian deblurring algorithm can be selected, with a medium-intensity blur kernel. When the image sharpness information indicates severe blurring, a more complex blind deconvolution algorithm can be selected, with higher iteration counts and regularization parameters. For surface structure complexity information, if the texture in a local area is overly complex, the parameters of the texture smoothing algorithm can be adjusted, such as increasing the window size or iteration count of the smoothing filter, to reduce pseudo-textures introduced by the deposition layer.

[0073] Corrosion feature information can encompass various types to comprehensively describe the corrosion phenomenon. For example, local feature points and descriptors can be obtained through multi-scale feature extraction algorithms (such as SIFT, SURF, ORB, etc.), which can capture discriminative local structures in the image. Furthermore, color space analysis can be used to obtain the color features of corrosion products, such as extracting the hue, saturation, and brightness distribution of the corroded region in the HSV color space. Geometric features can also be obtained through shape analysis, such as the area, perimeter, aspect ratio, and roundness of the corroded region; these features help distinguish different types of corrosion with varying shapes.

[0074] Finally, sedimentary layer information is used to assist corrosion identification and improve the accuracy of the judgment. For example, a discriminator can be configured for each major preset corrosion type (such as pitting corrosion and crack corrosion). Each discriminator focuses on matching a set of core features for the corresponding corrosion type. For example, the pitting corrosion discriminator mainly scores the matching based on the roundness and indentation of the area, while the crack corrosion discriminator mainly scores the matching based on the aspect ratio and linearity of the area. Each discriminator analyzes the current suspected corrosion area independently and outputs an initial matching confidence score. Subsequently, the confidence scores output by each discriminator are dynamically adjusted based on the background description information of the suspected area. Finally, the weighted confidence scores output by all discriminators are weighted and fused, and the corrosion type with the highest fused confidence score is determined as the final judgment result for that area.

[0075] The technical solution of this application first segments the image of the inner wall of the pipe into local regions and quantitatively evaluates the optical interference background affected by the sediment layer, generating comprehensive background description information to assess the complex coverage state of the sediment layer. Then, based on the background description information, the parameters of subsequent image preprocessing algorithms are dynamically and specifically adjusted to achieve adaptive image enhancement, suppressing or compensating for specific optical distortions caused by the sediment layer and restoring obscured details for feature extraction. Finally, in the judgment stage, the background description information characterizing environmental interference is combined with corrosion feature information extracted from the preprocessed image and input into the judgment model to intelligently distinguish between real corrosion and false sediment features.

[0076] In summary, this application, through adaptive preprocessing based on quantitative assessment, effectively penetrates or corrects the interference of sediment layers on light, color, texture, and clarity, improving the input conditions for corrosion feature extraction and reducing the risk of missed detections due to feature occlusion or distortion. By incorporating background description information into the final judgment, evidence can be weighed more intelligently, effectively distinguishing between genuine corrosion and false features formed by the sediment itself, thereby reducing misjudgments. Overall, this application upgrades the detection process from a passive, one-size-fits-all approach to the environment to an intelligent diagnosis that proactively senses and adapts to the environment, enhancing the robustness and accuracy of corrosion identification in the extremely complex and variable sediment layer environment of nuclear power pipelines, providing a more reliable technical guarantee for pipeline safety.

[0077] Preferably, in the method of this application:

[0078] The light occlusion information is obtained by calculating the relative attenuation ratio between the average pixel brightness of the local area and a preset brightness reference value.

[0079] The color performance difference information is obtained by calculating the difference between the average color vector of the local area and a preset color vector in a predetermined color space.

[0080] The surface structure complexity information is obtained by extracting the texture features of the local region and calculating the statistical complexity index of the texture features.

[0081] The image sharpness information is obtained by analyzing the intensity of image gradient changes in the local region.

[0082] The light occlusion information refers to the degree to which the deposition layer obstructs the propagation of light. The preset brightness reference value can be the image brightness under conditions without a deposition layer or standard lighting. By comparing the difference between the actual brightness and the reference value, the degree to which light is occluded by the deposition layer can be accurately reflected.

[0083] The color performance difference information is used to characterize the influence of the deposition layer on the color features of the image. The preset color vector can be derived from the average color of the area without the deposition layer, or set according to the inherent color characteristics of the pipe material, so as to effectively capture the color shift or saturation change caused by the deposition layer.

[0084] The surface structure complexity information aims to describe the physical morphological characteristics of the deposited layer surface, such as its roughness, grain distribution, or texture details. Texture features can be extracted using algorithms such as Local Binary Pattern (LBP) and Gabor features, and their complexity can be quantified by calculating statistical indicators such as entropy, contrast, and energy.

[0085] The image sharpness information directly reflects the degree of blur in the image, which is usually caused by the presence of the deposition layer or its optical properties. The greater the intensity of the image gradient change, the sharper the image edges and the richer the details; conversely, blurriness may exist, thus providing a quantitative basis for subsequent deblurring processing.

[0086] This application enables a refined, multi-dimensional quantitative assessment of the optical interference background caused by deposits on the inner wall of pipes. This detailed quantification method allows subsequent image preprocessing to more effectively suppress or compensate for the interference caused by the deposits.

[0087] In a further embodiment of this application, the step of acquiring an image of the inner wall of the pipe to be detected and segmenting the image into multiple local regions preferably includes:

[0088] Simultaneously collect temperature and pressure parameters inside the pipeline;

[0089] After the step of quantifying and evaluating the optical interference background affected by the deposition layer in the local area and generating background description information to characterize the coverage state of the deposition layer in the local area, the method further includes:

[0090] Based on the temperature and pressure parameters, physical predictions of the optical performance of the deposited layer are generated; wherein, the physical predictions include predicted light occlusion and predicted image sharpness.

[0091] If the difference between the predicted light occlusion degree and the light occlusion degree information exceeds a preset occlusion degree threshold, and / or the difference between the predicted image clarity and the image clarity information exceeds a preset clarity threshold, then based on the tunable multispectral light source and industrial camera, multiple different wavelengths of illumination light are scanned on the local area, and reflected light images at each wavelength are acquired simultaneously.

[0092] Based on the reflected light image, a characteristic spectrum of the reflection intensity of the local region as a function of wavelength is generated and used to correct the background description information.

[0093] While acquiring images, the system uses built-in or external sensors to obtain real-time temperature and pressure data inside the pipe. The temperature parameter can be understood as the real-time temperature of the fluid or pipe wall, and the pressure parameter as the real-time pressure of the fluid inside the pipe. The purpose is to provide the necessary environmental input for subsequent physical predictions.

[0094] Predicted light occlusion refers to the expected degree to which the deposition layer blocks or absorbs light under current temperature and pressure conditions; predicted image sharpness refers to the expected degree to which the deposition layer affects image detail under current temperature and pressure conditions. The purpose is to provide a reference benchmark independent of image optical evaluation for verifying the accuracy of image evaluation.

[0095] Tunable multispectral light sources can emit light of different wavelengths, such as from the visible to the near-infrared band, to detect the response of the deposited layer under different spectra. Industrial cameras are then used to capture these reflected light images with high precision. The aim is to obtain richer and more fundamental optical information about the deposited layer, overcoming the limitations that may exist under single-wavelength or broadband illumination.

[0096] Images of reflected light acquired at different wavelengths are processed to extract the reflection intensity value of each pixel or region, and plotted as a curve or dataset varying with wavelength. The characteristic spectrum reflects the absorption, scattering, and reflection characteristics of the deposited layer material, aiming to provide an objective and comprehensive description of the optical properties of the deposited layer and offer a reliable basis for correcting background description information.

[0097] This application enables effective verification of the initial quantitative assessment results of optical interference background by simultaneously acquiring temperature and pressure parameters and generating physical prediction values ​​based on these parameters. When a significant difference is found between the predicted and actual measured values, the system can intelligently initiate multispectral scanning to acquire the optical characteristic spectrum of deeper layers of the deposition layer, thereby correcting the background description information.

[0098] In a specific embodiment of this application, the step of generating physical prediction values ​​regarding the optical performance of the deposited layer based on the temperature and pressure parameters preferably includes:

[0099] If, within a preset time period, the temperature drop and pressure fluctuation both exceed the corresponding preset threshold values, then based on the temperature drop and fluctuation, and according to preset physical association rules, physical prediction values ​​including the predicted light occlusion degree of the deposition layer and the predicted image clarity are generated.

[0100] The preset duration refers to the time window for continuously monitoring changes in temperature and pressure parameters. It can be set according to the typical rate of change of the pipeline environment and the response characteristics of the deposit layer, for example, it can be several minutes to several hours. Temperature drop amplitude refers to the maximum temperature drop within the preset duration, while fluctuation amplitude refers to the maximum range of pressure variation (e.g., peak-to-valley difference) within the preset duration. When both amplitudes exceed their respective preset "amplitude thresholds," it indicates that a significant dynamic change has occurred in the pipeline's internal environment, such as rapid coolant flow or a sudden adjustment in system pressure. These changes may cause rapid alterations in the physicochemical properties of the deposit layer, thereby affecting its optical performance.

[0101] Physical correlation rules are pre-established mathematical models or empirical rules that describe the relationship between dynamic changes in a specific environment and the optical performance of a sedimentary layer. For example, a rule might be defined as follows: when the temperature drops rapidly and the pressure fluctuates significantly, the sedimentary layer may condense or agglomerate, leading to increased light shading and decreased image sharpness. These rules can be derived from laboratory simulations, field data analysis, or physical and chemical theoretical derivations, aiming to more accurately reflect the optical response of the sedimentary layer under transient conditions.

[0102] This application's solution, by introducing monitoring of the temperature drop and pressure fluctuation amplitudes and setting corresponding preset amplitude thresholds, can effectively identify rapid dynamic changes in the pipeline's internal environment. When these dynamic changes reach a level sufficient to affect the optical performance of the deposited layer, it no longer relies solely on general temperature and pressure parameters, but generates more accurate physical prediction values ​​based on preset physical correlation rules and these specific dynamic change amplitudes. This mechanism enables the system to make more targeted predictions of the optical performance of the deposited layer under transient conditions, thereby compensating for the inaccuracy of accurately capturing dynamic optical changes in the deposited layer based solely on conventional temperature and pressure parameters.

[0103] The following is a specific example to illustrate this.

[0104] Suppose that during the inspection of a nuclear power plant pipeline, the system detects that the internal temperature parameter drops from 80°C to 70°C within a continuous 15-minute period, a temperature drop of 10°C; simultaneously, the pressure parameter fluctuates from 15 MPa to 14.5 MPa and then rises back to 15.2 MPa within the same time period, a fluctuation of 0.7 MPa. The preset temperature drop threshold is 5°C, and the preset pressure fluctuation threshold is 0.5 MPa. Since both the detected temperature drop (10°C) and fluctuation (0.7 MPa) exceed their respective preset thresholds, the system determines that the current environment is in a dynamic state. At this time, the system will invoke preset physical correlation rules based on the 10°C temperature drop and the 0.7 MPa pressure fluctuation. For example, this rule might indicate that under such temperature drop and pressure fluctuations, the microstructure of the deposition layer will change, resulting in a 20% increase in predicted light shading and a 15% decrease in predicted image sharpness. Based on this, the system generates updated predicted light occlusion and predicted image sharpness. These predictions are then used to correct background description information, ensuring a more accurate assessment of optical interference in the deposition layer under dynamic conditions.

[0105] It should be noted that the correction process for the background description information specifically includes:

[0106] Acquire a pre-stored first reference spectrum characterizing the deposition layer in a stable optical state and a second reference spectrum characterizing the deposition layer in a transient optical deviation state;

[0107] The similarity between the feature spectrum and the first reference spectrum and the second reference spectrum is calculated respectively to obtain the first similarity and the second similarity.

[0108] If the second similarity is higher than the first similarity and higher than the preset similarity threshold, then the second similarity is used as a transient influence parameter.

[0109] The corrected light occlusion information is obtained by using a preset first correction formula; the first correction formula is: L'=L+(1-L)×T×α; where L' is the corrected light occlusion information, L is the light occlusion information before correction, T is the transient influence parameter, and α is the preset first correction factor;

[0110] The corrected image sharpness information is obtained through a preset second correction formula; the second correction formula is: C'=CC×T×β; where C' is the corrected image sharpness information, C is the image sharpness information before correction, T is the transient influence parameter, and β is the preset second correction factor;

[0111] The corrected light occlusion information and / or image sharpness information are then used to update the background description information to obtain the corrected background description information.

[0112] The first and second reference spectra are obtained in advance through experiments or simulations. For example, the first reference spectrum may be the reflection spectrum after the deposition layer has formed and stabilized under constant temperature, pressure, and flow rate; while the second reference spectrum may be the reflection spectrum exhibited when the surface structure or composition of the deposition layer undergoes instantaneous changes due to drastic temperature or pressure changes. The characteristic spectrum is obtained based on the above multispectral scanning and characterizes the variation of reflection intensity of the current local deposition layer at different wavelengths. The similarity between the characteristic spectrum and the first and second reference spectra can be calculated using various spectral similarity algorithms, such as spectral angle mapping, spectral information divergence, or Pearson correlation coefficient.

[0113] In practical applications, when the second similarity is not only higher than the first similarity but also exceeds a preset similarity threshold, it indicates that the current deposition layer is likely in a transient optical deviation state. In this case, the second similarity is used as the transient influence parameter T. The higher the value of parameter T, the more significant the influence of the transient deviation.

[0114] For information on light shading, the first correction formula aims to increase the original light shading level L based on the transient influence parameter T. Here, (1-L) represents the unshaded portion, and the product of T and α represents the amount of correction the transient influence makes to the light shading level. This means that the greater the transient influence, the more the light shading level will be corrected upwards, reflecting the potentially stronger light attenuation that the deposition layer may experience under transient deviations. The correction factor α is used to adjust the intensity of the correction to the light shading level by the transient influence.

[0115] For image sharpness information, the second correction formula aims to reduce the original image sharpness C based on the transient effect parameter T. Here, C × T × β represents the amount of image sharpness reduction caused by the transient effect. This means that the greater the transient effect, the more the image sharpness will be corrected downwards, reflecting the more severe image blurring or distortion that the deposition layer may cause under transient deviations. The correction factor β is used to adjust the strength of the correction to image sharpness caused by the transient effect.

[0116] Therefore, the corrected light occlusion information L' and image sharpness information C' will be updated in the background description information, thereby obtaining a more accurate background description information that reflects the current actual optical state.

[0117] The proposed scheme introduces both stable and transient reference spectra for comparison, and dynamically adjusts the degree of light occlusion and image clarity based on the comparison results. This allows the background description information to more accurately reflect the real optical interference of the deposition layer under complex environments, improving the accuracy and robustness of the background description information. In particular, when the optical properties of the deposition layer change instantaneously, it can avoid the deviation in corrosion feature extraction caused by misjudging the degree of optical interference, thus providing more reliable basic data for subsequent corrosion type determination.

[0118] The following is a specific example to illustrate this.

[0119] Suppose that in a local area of ​​a long-distance nuclear power transmission pipeline, a sudden fluctuation in the temperature or pressure of the fluid inside the pipeline causes a change in the microstructure of the deposited layer surface, thereby causing its optical properties to change from a stable state to a transient optical deviation state.

[0120] First, the system acquires a pre-stored first reference spectrum characterizing the type of deposited layer under stable optical conditions, and a second reference spectrum under specific transient optical deviation conditions. For example, the first reference spectrum may show a reflection intensity of 0.6 at a wavelength of 500 nm, while the second reference spectrum under transient deviation conditions may show a reflection intensity of 0.4 at the same wavelength, accompanied by a change in spectral shape.

[0121] Next, the characteristic spectrum of the current local area is obtained by scanning with a multispectral light source and acquiring data with an industrial camera. For example, the reflection intensity of this characteristic spectrum at a wavelength of 500 nm is 0.45, and its overall shape is closer to that of the second reference spectrum.

[0122] The system calculates the similarity between the feature spectrum and the first and second reference spectra. Assume the calculation shows a first similarity of 0.7 with the first reference spectrum and a second similarity of 0.9 with the second reference spectrum. Meanwhile, the preset similarity threshold is 0.8. Since the second similarity of 0.9 is higher than the first similarity of 0.7 and also higher than the preset similarity threshold of 0.8, the system determines that the current deposition layer is in a transient optical deviation state and uses the second similarity of 0.9 as the transient influence parameter T.

[0123] Assume the light occlusion information L before correction is 0.3, and the image sharpness information C is 0.7. The preset first correction factor α is 0.5, and the second correction factor β is 0.4.

[0124] According to the first correction formula L'=L+(1-L)×T×α, the corrected light occlusion information L' will be calculated as: L' = 0.3 + (1 - 0.3) × 0.9 × 0.5 = 0.3 + 0.7 × 0.9 × 0.5 = 0.3 + 0.315= 0.615.

[0125] According to the second correction formula C'=CC×T×β, the corrected image sharpness information C' will be calculated as: C' =0.7 - 0.7 × 0.9 × 0.4 = 0.7 - 0.7 × 0.36 = 0.7 - 0.252 = 0.448.

[0126] In some embodiments of this application, such as Figure 2 As shown, the preferred steps for adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information include:

[0127] A1, transmits a probe wave to the local area and receives the echo signal;

[0128] A2, Analyze the echo signal to determine the microstructure state of the deposition layer in the local area and the state influence parameters of the microstructure state;

[0129] A3. Input the microstructure state and the state influence parameters into a preset fuzzy kernel parameter mapping model for calculation to obtain a fuzzy kernel function; wherein, the fuzzy kernel parameter mapping model includes the correspondence between fuzzy kernel type and microstructure state, the functional relationship between fuzzy kernel strength and state influence parameters, and the calculation rules for fuzzy kernel spatial distribution parameters.

[0130] Probe waves, such as ultrasonic waves, millimeter waves, or terahertz waves, are emitted into a localized area to physically probe the deposits on the inner wall of a pipe. These probe waves can penetrate the deposits and generate reflected or scattered echo signals based on the physical properties of the deposits (such as density, acoustic impedance, and dielectric constant). Receiving these echo signals involves capturing them using appropriate sensors (such as ultrasonic transducers or millimeter-wave receivers).

[0131] By analyzing the amplitude attenuation, propagation time, spectral characteristics, or phase changes of the echo signal, we can infer the microstructural state of the deposition layer, such as its thickness, density uniformity, porosity, and the presence of micron-sized bubbles or particle distribution. State-related parameters quantify the degree to which these microstructural states affect optical blurring; examples include the degree of density non-uniformity of the deposition layer, the average size and density of bubbles, or surface roughness.

[0132] The fuzzy kernel parameter mapping model is a pre-established knowledge base or algorithm model. Its core lies in transforming the microstructural state of the sedimentary layer and its state-influencing parameters obtained from probe wave analysis into the fuzzy kernel function required for image deblurring. The correspondence between fuzzy kernel type and microstructural state refers to selecting the most suitable fuzzy kernel type (e.g., Gaussian fuzzy kernel, motion fuzzy kernel, defocus fuzzy kernel, etc.) based on the specific microstructure of the sedimentary layer (e.g., uniform scattering, non-uniform scattering, motion blur in a specific direction). The functional relationship between fuzzy kernel intensity and state-influencing parameters refers to calculating the specific intensity parameters of the selected fuzzy kernel (e.g., standard deviation of the Gaussian kernel, length and angle of the motion fuzzy kernel, etc.) based on the quantized values ​​of the state-influencing parameters. The calculation rules for the spatial distribution parameters of the fuzzy kernel are used to determine the range and distribution characteristics of the fuzzy kernel in the image space, such as whether it is isotropic blur or whether there is differential blur in local regions.

[0133] The proposed solution introduces a probe wave to physically probe the microstructure of the sediment layer, enabling more accurate identification of the physical causes and types of fuzziness, thereby generating a more realistic fuzzy kernel function. Consequently, the parameter adjustment of the deblurring algorithm becomes more precise, significantly improving the quality and realism of the preprocessed image and providing a more reliable foundation for subsequent extraction of corrosion feature information.

[0134] The following is a specific example to illustrate this.

[0135] Suppose that in a localized area of ​​a long-distance nuclear power transmission pipeline, a sedimentary layer with varying porosity and density has formed due to the deposition of suspended particles in the fluid. Traditional methods might only rely on image sharpness information to determine if this area is blurred and then apply a generic deblurring algorithm.

[0136] However, according to the scheme of this application, a high-frequency ultrasonic probe wave is first emitted into the local area. After the ultrasonic wave penetrates the sediment layer, its echo signal is scattered and attenuated by the porosity and density inhomogeneity of the sediment layer. Spectral analysis of the received echo signal reveals significant attenuation of its high-frequency components and the presence of multiple scattering peaks. This indicates the presence of micron-sized bubbles and irregular particle distribution within the sediment layer, i.e., the microstructural state is characterized by "the presence of micron-sized bubbles and density inhomogeneity." By calculating the amplitude attenuation rate and spectral characteristic changes of the echo signal, the state influence parameters can be quantified; for example, the average diameter of the bubbles is 5 micrometers, and the density inhomogeneity is 15%.

[0137] Subsequently, the microstructural state of "presence of micron-sized bubbles and density inhomogeneity" and the state-influencing parameters of "average bubble diameter of 5 micrometers and density inhomogeneity of 15%" are input into a preset blur kernel parameter mapping model. Based on the correspondence between the microstructural state of "presence of micron-sized bubbles and density inhomogeneity," this model determines that a combination of a "defocus blur kernel" and a "Gaussian blur kernel" should be used as the blur kernel type. Next, based on the state-influencing parameters of "average bubble diameter of 5 micrometers" and "density inhomogeneity of 15%," the radius parameter of the defocus blur kernel and the standard deviation parameter of the Gaussian blur kernel are calculated through a preset functional relationship. Simultaneously, based on the local characteristics of the deposition layer, the distribution parameters of the blur kernel in the image space are calculated. Finally, a blur kernel function that accurately reflects the optical blurring characteristics of the deposition layer is obtained.

[0138] By utilizing this precise blur kernel function, the deblurring algorithm can more effectively deconvolve images, remove blur caused by the microstructure of the sediment layer, and thus obtain clearer and more realistic local area images.

[0139] In a further embodiment of this application, the step of analyzing the echo signal to determine the microstructure state of the deposition layer in the local region and the state influence parameters of the microstructure state preferably includes:

[0140] Calculate the amplitude attenuation rate and / or spectral characteristic changes of the echo signal;

[0141] If the amplitude attenuation rate exceeds a first preset threshold, and / or the spectral characteristic change indicates that the high-frequency component has attenuation or resonance in a predetermined mode, then the microstructure state of the deposition layer in the local area is determined to be the presence of micron-sized bubbles or density inhomogeneity, and the state influence parameters are calculated based on the amplitude attenuation rate and / or spectral characteristic change.

[0142] The amplitude attenuation rate of the echo signal refers to the degree of attenuation of the signal strength of the probe wave after it penetrates the sediment layer and is reflected back, relative to the initial amplitude at the time of transmission. It can be calculated by comparing the initial amplitude of the transmitted signal with the amplitude of the received echo signal. Spectral characteristic changes refer to the changes in the energy distribution, peak frequency, bandwidth, and other characteristics of the echo signal in the frequency domain. These can be obtained through spectral analysis methods such as Fourier transform of the echo signal.

[0143] When the calculated amplitude attenuation rate exceeds a preset first threshold, it indicates that the probe wave encounters significant energy loss in the sediment layer, which is usually related to the presence of scattering or absorbing media within the sediment layer. When changes in spectral characteristics indicate that high-frequency components are attenuating or resonating in a predetermined mode—for example, a significant decrease in the energy of high-frequency components or the appearance of resonance peaks at specific frequency points—this further confirms the inhomogeneity of the microstructure within the sediment layer. When any of the above conditions or both are met simultaneously, it can be determined that the microstructure of the sediment layer in a local area contains micron-sized bubbles or density inhomogeneities.

[0144] State influence parameters are used to quantify the impact of micrometer-sized bubbles or density inhomogeneities on optical imaging. For example, a higher amplitude attenuation rate or a more significant change in spectral characteristics results in a larger state influence parameter value, indicating more severe optical interference. State influence parameters can be calculated based on pre-defined empirical formulas or obtained through training a machine learning model.

[0145] This application's method, through precise analysis of the amplitude attenuation rate and / or spectral characteristic changes of the echo signal, enables in-depth probing of the microstructure within the deposition layer. The amplitude attenuation rate directly reflects the energy loss of the probe wave in the medium, while spectral characteristic changes reveal the scattering, absorption, or resonance effects of the medium on different frequency components. It is precisely the presence of micron-sized bubbles or density-inhomogeneous microstructures that causes specific amplitude attenuation and spectral change patterns in the probe wave during propagation. By identifying these patterns, the microstructural state of the deposition layer can be accurately determined, and its impact on image sharpness can be quantified, thus providing a precise physical basis for subsequent parameter adjustments in deblurring algorithms.

[0146] In a specific embodiment of this application, the step of extracting corrosion feature information from the preprocessed local region preferably includes:

[0147] Based on the background description information, the feature extraction strategy and sensitivity are adaptively adjusted to extract the suspected corrosion area and corrosion feature information of the suspected corrosion area from the preprocessed local area. The corrosion feature information includes: local feature points and descriptors obtained by multi-scale feature extraction algorithm; color features of corrosion products obtained by color space analysis; and geometric features obtained by shape analysis.

[0148] Based on the background description information, an appropriate feature extraction algorithm is dynamically selected, and the algorithm parameters (e.g., threshold, filter size, scale range, etc.) are adjusted to ensure effective identification of eroded areas under different background conditions. For example, when the light occlusion information indicates that the light is dim, the sensitivity of the feature extraction algorithm to low-contrast areas can be enhanced; when the surface structure complexity information is high, the parameters of the texture analysis algorithm can be adjusted to distinguish the eroded texture from the background texture.

[0149] Suspected corrosion areas refer to image regions identified as potentially corroded during initial analysis. These areas require further feature extraction and assessment. Corrosion feature information is key data used to characterize these suspected corrosion areas, and its extraction methods include the following:

[0150] Local feature points and descriptors obtained through multi-scale feature extraction algorithms, such as SIFT, SURF, or ORB, can detect key points in images at different scales and generate scale- and rotation-invariant descriptors that can capture local texture and structural information of eroded regions.

[0151] Color features of corrosion products obtained through color space analysis refer to the analysis of pixel color distribution in suspected corrosion areas in color spaces such as HSV and Lab, and the extraction of features such as average hue, saturation, brightness, and color variance. These features help to distinguish the color of corrosion products (such as rust) from the color of normal pipe inner walls or deposits.

[0152] The geometric features obtained through shape analysis refer to the quantification of geometric attributes such as the outline, area, perimeter, aspect ratio, roundness, and convexity of suspected corrosion areas. These features are of great significance for identifying different forms of corrosion (such as pitting corrosion, line corrosion, and sheet corrosion).

[0153] This approach dynamically adjusts the feature extraction strategy and sensitivity based on the actual coverage of the deposit layer on the inner wall of the pipeline and the degree of optical interference, thereby effectively suppressing the impact of residual optical interference on feature extraction. This enables the system to more accurately identify suspected corrosion areas and extract more representative local feature points, descriptors, color features, and geometric features, providing high-quality input data for subsequent corrosion type determination, and thus improving the overall reliability and detection accuracy of the corrosion classification method.

[0154] Based on the above implementation methods, such as Figure 3 As shown, the step of determining the corrosion type of the local area by combining the background description information and the corrosion feature information preferably includes:

[0155] S151, integrate the local feature points, descriptors, color features and geometric features extracted from the suspected corrosion area with the background description information to form a feature evidence vector;

[0156] S152, The feature evidence vector is input into a preset classification inference model for calculation to obtain the confidence level of the suspected corrosion area belonging to various corrosion types;

[0157] S153, the corrosion type with the highest confidence level is determined as the candidate type of the suspected corrosion area;

[0158] S154, dynamically adjust the confidence threshold for confirming the corrosion type based on the light occlusion information and the image clarity information;

[0159] S155, for suspected corrosion areas where the confidence level of the candidate type is higher than the adjusted confidence level threshold: if the component values ​​of the color feature in the preset color space all fall within the color threshold range after the dynamic adjustment of the color performance difference information, and the difference between the texture feature complexity index extracted from the suspected corrosion area and the surface structure complexity information is less than the preset texture consistency threshold, then it is determined to be a pseudo-deposit area and is excluded; otherwise, the candidate type is taken as the final corrosion type of the suspected corrosion area.

[0160] Background description information and corrosion feature information are combined into a comprehensive feature evidence vector, which can fully characterize the visual characteristics of the suspected corrosion area and its environmental background.

[0161] The classification inference model can be a machine learning model, such as a support vector machine, neural network, or deep learning model. Trained on a large number of labeled corrosion images and background images, it can output the confidence level of whether the suspected corrosion region belongs to various predefined corrosion types (e.g., pitting corrosion, uniform corrosion, crevice corrosion, etc.) based on the input feature evidence vector. The corrosion type with the highest confidence level is initially identified as the candidate type of the suspected corrosion region.

[0162] Furthermore, to improve the accuracy and robustness of the judgment, this application dynamically adjusts the confidence threshold used to confirm the corrosion type based on the light occlusion information and the image clarity information. For example, when the light occlusion is high or the image clarity is low, the classification model may be more prone to misjudgment due to the decreased image quality. In this case, the confidence threshold can be appropriately increased to reduce the false positive rate. Conversely, when the image quality is good, the threshold can be appropriately decreased to improve the detection sensitivity.

[0163] Furthermore, for suspected corrosion areas whose confidence levels for candidate types are higher than the adjusted confidence threshold, further discrimination is performed. Specifically, if the component values ​​of the color features in the preset color space all fall within the color threshold range dynamically adjusted by the color performance difference information, and the difference between the texture feature complexity index extracted from the suspected corrosion area and the surface structure complexity information is less than the preset texture consistency threshold, then the suspected corrosion area is determined to be a pseudo-deposit area and excluded. The color threshold can be dynamically adjusted based on the color performance difference information to adapt to the color characteristics of different deposition layers; the texture consistency threshold is used to measure the similarity between the texture of the suspected corrosion area and the texture of the background deposition layer. If the above conditions are not met, i.e., the area does not meet the characteristics of a pseudo-deposit, then the candidate type is taken as the final corrosion type of the suspected corrosion area.

[0164] This application's solution deeply integrates corrosion feature information with background description information to form a comprehensive feature evidence vector, thus providing richer and more discriminative input for the classification inference model. This integration allows the model to fully consider the impact of environmental factors on image performance during initial classification. Furthermore, by dynamically adjusting the confidence threshold based on light occlusion and image sharpness information, it can adapt to image quality variations under different optical interference conditions, avoiding misjudgments caused by image quality degradation in harsh environments, while maintaining detection sensitivity under favorable conditions. Moreover, by introducing a pseudo-deposit exclusion mechanism based on comparing color and texture features with background description information, it can effectively distinguish between genuine corrosion and deposits with similar visual characteristics, thereby reducing the false positive rate and improving the accuracy and reliability of corrosion type determination.

[0165] The following is a specific example to illustrate this.

[0166] Suppose that during the inspection of a local area of ​​a nuclear power plant pipeline, the image processing module has preprocessed the image, and the feature extraction module has extracted multiple suspected corrosion areas and their corrosion characteristics. Now, the corrosion judgment module needs to determine the type of corrosion in these suspected corrosion areas.

[0167] Specifically, for a region A identified as potentially corroded, its extracted local feature points, descriptors, color features (e.g., reddish-brown), and geometric features (e.g., irregular shape) are collected. Simultaneously, background descriptive information of this local region, including moderate light occlusion, slight color variations, moderate surface complexity, and good image sharpness, is also integrated. This information collectively constitutes a feature evidence vector.

[0168] The feature evidence vector was then fed into a pre-trained deep neural network classification and inference model. The model output showed that the confidence level for suspected corrosion region A belonging to the "pitting" type was 0.85, the confidence level for belonging to the "uniform corrosion" type was 0.10, and the confidence level for other types was low. Therefore, "pitting" was identified as a candidate type for suspected corrosion region A.

[0169] At this point, the system dynamically adjusts the confidence threshold used to confirm the corrosion type based on the current light occlusion level (medium) and image clarity information (good). For example, the adjusted confidence threshold is set to 0.75. Since the confidence level of the candidate type "pitting corrosion" (0.85) is higher than the adjusted threshold of 0.75, the system will further perform a false deposition exclusion judgment.

[0170] During the pseudo-deposit exclusion phase, the system examines the color characteristics of suspected corrosion area A. Assuming that its color characteristics, including hue, saturation, and lightness components in the preset HSV color space, all fall within the color threshold range dynamically adjusted by color performance difference information, this indicates that its color has a certain similarity to the background deposit color. Simultaneously, the system calculates the texture complexity index of suspected corrosion area A and compares it with the surface structure complexity information of that local area. If the difference between the two is less than a preset texture consistency threshold—for example, if the texture of this area is highly similar to the texture of the surrounding deposits—the system determines that suspected corrosion area A is actually a pseudo-deposit area and excludes it from the corrosion list.

[0171] Conversely, if the color or texture characteristics of the suspected corrosion area A differ significantly from those of the background sediments and do not meet the criteria for identifying pseudo-sediments, then "pitting corrosion" will be ultimately determined as the corrosion type of the suspected corrosion area A. Through this series of refined judgment procedures, this application can effectively avoid misclassifying sediments as corrosion, thereby improving the accuracy of the detection results.

[0172] like Figure 4 As shown, this application also discloses an image recognition-based corrosion typing system for detecting the interior of long-distance nuclear power transmission pipelines. The system includes:

[0173] The image acquisition and region segmentation module 210 is used to acquire the image to be detected on the inner wall of the pipe and segment the image to be detected into multiple local regions;

[0174] The information generation module 220 is used to quantitatively evaluate the optical interference background of the local area affected by the deposition layer, and generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes light occlusion information, color performance difference information, surface structure complexity information, and image clarity information.

[0175] Image processing module 230 is used to adjust the algorithm parameters of image preprocessing according to the background description information, and to perform image preprocessing on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters of image preprocessing includes: adjusting the parameters of brightness and contrast enhancement algorithms according to the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of deblurring algorithms according to the image sharpness information; and adjusting the parameters of texture smoothing algorithms according to the surface structure complexity information;

[0176] The feature extraction module 240 is used to extract corrosion feature information from the preprocessed local area;

[0177] The corrosion judgment module 250 is used to judge the corrosion type of the local area by combining the background description information and the corrosion feature information.

[0178] The image acquisition and region segmentation module 210 can consist of one or more hardware components and corresponding software drivers. For example, it may include one or more high-resolution industrial cameras, an image acquisition card, and an image preprocessing unit. The industrial camera is responsible for acquiring images of the pipe's inner wall to be inspected; the image acquisition card is responsible for converting analog signals into digital signals and transmitting them to the processing unit; and the image preprocessing unit is responsible for segmenting the image to be inspected into multiple local regions. Image segmentation can be achieved using a fixed grid-based segmentation method or through adaptive segmentation using image processing algorithms (such as edge detection and region growing).

[0179] The information generation module 220 may consist of one or more processors (such as a CPU, GPU, or dedicated DSP) and memory, wherein the memory stores algorithm programs for performing quantization evaluation. For example, the information generation module 220 may calculate the average pixel brightness of a local area to obtain information on the degree of light occlusion, calculate the average color vector in a predetermined color space to obtain information on color representation differences, obtain information on the complexity of surface structure through texture feature extraction and statistical analysis, and obtain information on image sharpness by analyzing the intensity of image gradient changes. These calculation results are integrated into background description information.

[0180] The image processing module 230 can consist of a processor and a memory. The memory stores various image preprocessing algorithms (such as brightness and contrast enhancement algorithms, deblurring algorithms, and texture smoothing algorithms) and their parameter adjustment logic. For example, when receiving information on the degree of light occlusion and / or color performance differences, it can dynamically adjust the gain and offset of the brightness and contrast enhancement algorithm. When receiving image sharpness information, it can select and adjust the parameters and intensity of the deblurring algorithm, such as selecting Gaussian blur, motion blur, or blind deconvolution algorithm according to the degree of blur, and setting the corresponding blur kernel size and number of iterations. When receiving information on the complexity of the surface structure, it can adjust the parameters of the texture smoothing algorithm, such as adjusting the window size or smoothing intensity of the smoothing filter, to suppress or compensate for optical interference caused by the deposition layer.

[0181] The feature extraction module 240 may consist of one or more processors and a memory, in which various feature extraction algorithms are stored. For example, the feature extraction module 240 may implement multi-scale feature extraction algorithms (such as SIFT, SURF, ORB) to obtain local feature points and descriptors, perform color space analysis to obtain the color features of corrosion products, and perform shape analysis to obtain the geometric features of the corrosion region.

[0182] The corrosion judgment module 250 may consist of one or more processors and a memory. The memory stores a preset classification inference model (such as a support vector machine, neural network, or deep learning model) and corresponding judgment logic. The corrosion judgment module 250 inputs the integrated feature evidence vector into the classification inference model, calculates the confidence score of each corrosion type, and determines candidate corrosion types based on the confidence score. Furthermore, the corrosion judgment module 250 can dynamically adjust the confidence threshold based on background description information (such as light occlusion information and image sharpness information), and combine color features and texture feature complexity indicators to exclude false sediment areas, thereby improving the accuracy of corrosion judgment.

[0183] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0184] The preferred embodiments of this application have been described in detail above, but this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application.

Claims

1. A corrosion typing method based on image recognition, applied to the internal inspection of long-distance nuclear power transmission pipelines, characterized in that, The method includes the following steps: Acquire an image of the inner wall of the pipe to be inspected, and segment the image into multiple local regions; The optical interference background of the local area affected by the deposition layer is quantitatively evaluated to generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes information on the degree of light occlusion, information on color performance differences, information on the complexity of surface structure, and information on image sharpness. Based on the background description information, the algorithm parameters for image preprocessing are adjusted, and image preprocessing is performed on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters for image preprocessing includes: adjusting the parameters of the brightness and contrast enhancement algorithm based on the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information; and adjusting the parameters of the texture smoothing algorithm based on the surface structure complexity information; Extract corrosion feature information from the preprocessed local area; By combining the background description information and the corrosion feature information, the corrosion type of the local area is determined.

2. The erosion classification method based on image recognition according to claim 1, characterized in that, The light occlusion information is obtained by calculating the relative attenuation ratio between the average pixel brightness of the local area and a preset brightness reference value; The color performance difference information is obtained by calculating the difference between the average color vector of the local area and a preset color vector in a predetermined color space. The surface structure complexity information is obtained by extracting the texture features of the local region and calculating the statistical complexity index of the texture features; The image sharpness information is obtained by analyzing the intensity of image gradient changes in the local region.

3. The erosion classification method based on image recognition according to claim 1, characterized in that, The step of acquiring the image to be detected from the inner wall of the pipe and segmenting the image to be detected into multiple local regions further includes: Simultaneously collect temperature and pressure parameters inside the pipeline; After the step of quantifying and evaluating the optical interference background affected by the deposition layer in the local area and generating background description information to characterize the coverage state of the deposition layer in the local area, the method further includes: Based on the temperature and pressure parameters, physical predictions of the optical performance of the deposited layer are generated; wherein, the physical predictions include predicted light occlusion and predicted image sharpness. If the difference between the predicted light occlusion degree and the light occlusion degree information exceeds a preset occlusion degree threshold, and / or the difference between the predicted image clarity and the image clarity information exceeds a preset clarity threshold, then based on the tunable multispectral light source and industrial camera, multiple different wavelengths of illumination light are scanned on the local area, and reflected light images at each wavelength are acquired simultaneously. Based on the reflected light image, a characteristic spectrum of the reflection intensity of the local region as a function of wavelength is generated and used to correct the background description information.

4. The erosion typing method based on image recognition according to claim 3, characterized in that, The step of generating physical predictions of the optical performance of the deposited layer based on the temperature and pressure parameters includes: If, within a preset time period, the temperature drop and pressure fluctuation both exceed the corresponding preset threshold values, then based on the temperature drop and fluctuation, and according to preset physical association rules, physical prediction values ​​including the predicted light occlusion degree of the deposition layer and the predicted image clarity are generated.

5. The erosion typing method based on image recognition according to claim 3, characterized in that, The correction process for the background description information includes: Acquire a pre-stored first reference spectrum characterizing the deposition layer in a stable optical state and a second reference spectrum characterizing the deposition layer in a transient optical deviation state; The similarity between the feature spectrum and the first reference spectrum and the second reference spectrum is calculated respectively to obtain the first similarity and the second similarity. If the second similarity is higher than the first similarity and higher than the preset similarity threshold, then the second similarity is used as a transient influence parameter. The corrected light occlusion information is obtained by using a preset first correction formula; the first correction formula is: L'=L+(1-L)×T×α; where L' is the corrected light occlusion information, L is the light occlusion information before correction, T is the transient influence parameter, and α is the preset first correction factor; The corrected image sharpness information is obtained through a preset second correction formula; the second correction formula is: C'=CC×T×β; where C' is the corrected image sharpness information, C is the image sharpness information before correction, T is the transient influence parameter, and β is the preset second correction factor; The corrected light occlusion information and / or image sharpness information are then used to update the background description information to obtain the corrected background description information.

6. The erosion classification method based on image recognition according to claim 1, characterized in that, The steps for adjusting the parameters and intensity of the deblurring algorithm based on the image sharpness information include: A probe wave is emitted into the local area, and an echo signal is received; The echo signal is analyzed to determine the microstructure state of the deposition layer in the local area and the state-influencing parameters of the microstructure state. The microstructure state and the state influence parameters are input into a preset fuzzy kernel parameter mapping model for calculation to obtain a fuzzy kernel function; wherein, the fuzzy kernel parameter mapping model includes the correspondence between fuzzy kernel type and microstructure state, the functional relationship between fuzzy kernel strength and state influence parameters, and the calculation rules for fuzzy kernel spatial distribution parameters.

7. The erosion typing method based on image recognition according to claim 6, characterized in that, The step of analyzing the echo signal to determine the microstructure state of the deposition layer in the local region and the state-influencing parameters of the microstructure state includes: Calculate the amplitude attenuation rate and / or spectral characteristic changes of the echo signal; If the amplitude attenuation rate exceeds a first preset threshold, and / or the spectral characteristic change indicates that the high-frequency component has attenuation or resonance in a predetermined mode, then the microstructure state of the deposition layer in the local area is determined to be the presence of micron-sized bubbles or density inhomogeneity, and the state influence parameters are calculated based on the amplitude attenuation rate and / or spectral characteristic change.

8. The erosion classification method based on image recognition according to claim 1, characterized in that, The step of extracting corrosion feature information from the preprocessed local region includes: Based on the background description information, the feature extraction strategy and sensitivity are adaptively adjusted to extract the suspected corrosion area and corrosion feature information of the suspected corrosion area from the preprocessed local area. The corrosion feature information includes: local feature points and descriptors obtained by multi-scale feature extraction algorithm; color features of corrosion products obtained by color space analysis; and geometric features obtained by shape analysis.

9. The erosion classification method based on image recognition according to claim 8, characterized in that, The step of determining the corrosion type of the local area by combining the background description information and the corrosion feature information includes: The local feature points, descriptors, color features, and geometric features extracted from the suspected corrosion area are integrated with the background description information to form a feature evidence vector; The feature evidence vector is input into a preset classification inference model for calculation to obtain the confidence level of the suspected corrosion area belonging to various corrosion types; The corrosion type with the highest confidence level is identified as the candidate type for the suspected corrosion region; Based on the light occlusion information and the image clarity information, the confidence threshold used to confirm the corrosion type is dynamically adjusted; For suspected corrosion areas where the confidence level of the candidate type is higher than the adjusted confidence level threshold: if the component values ​​of the color feature in the preset color space all fall within the color threshold range after the dynamic adjustment of the color performance difference information, and the difference between the texture feature complexity index extracted from the suspected corrosion area and the surface structure complexity information is less than the preset texture consistency threshold, then it is determined to be a pseudo-deposit area and is excluded; otherwise, the candidate type is taken as the final corrosion type of the suspected corrosion area.

10. A corrosion typing system based on image recognition, applied to the internal inspection of long-distance nuclear power transmission pipelines, characterized in that, The system includes: The image acquisition and region segmentation module is used to acquire the image to be detected on the inner wall of the pipe and segment the image to be detected into multiple local regions; The information generation module is used to quantitatively evaluate the optical interference background of the local area affected by the deposition layer, and generate background description information to characterize the coverage state of the deposition layer in the local area; wherein, the background description information includes light occlusion information, color performance difference information, surface structure complexity information, and image sharpness information. An image processing module is used to adjust the algorithm parameters of image preprocessing according to the background description information, and to perform image preprocessing on each local region to suppress or compensate for optical interference caused by the deposition layer; wherein, adjusting the algorithm parameters of image preprocessing includes: adjusting the parameters of the brightness and contrast enhancement algorithm according to the light occlusion information and / or color performance difference information; adjusting the parameters and intensity of the deblurring algorithm according to the image sharpness information; and adjusting the parameters of the texture smoothing algorithm according to the surface structure complexity information; The feature extraction module is used to extract corrosion feature information from the preprocessed local area; The corrosion determination module is used to determine the corrosion type of the local area by combining the background description information and the corrosion feature information.

Images