Method, medium and system for monitoring corrosion degree of high-altitude large steel structure after construction
Through multi-modal image data analysis and neural network model, the detection range limitations and insufficient accuracy of corrosion evaluation of high-altitude large steel structures are solved, and comprehensive and accurate corrosion evaluation of high-altitude large steel structures is achieved, which reduces labor costs and increases the Monitor efficiency.
Patent Information
- Application Number
- CN202410991440.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-12
AI Technical Summary
It is difficult for the prior art to comprehensively and accurately evaluate the corrosion conditions of large steel structures at high altitudes. The detection range is limited and the accuracy is not high, so it is difficult to detect corrosion hazards in a timely manner.
The visible light camera and ultrasonic flaw detector are used to obtain multimodal image data. Through clustering analysis, citric acid solution treatment and neural network model, combined with image difference and feature extraction, the corrosion degree of the steel structure is calculated to achieve comprehensive and comprehensive corrosion conditions. Accurate assessment.
Comprehensive monitoring and accurate quantitative analysis of large steel structures at high altitudes has been achieved, labor costs have been reduced, monitoring efficiency and frequency have been improved, corrosion conditions can be evaluated automatically, and the accuracy and comprehensiveness of the assessment are ensured.
Smart Images

Figure CN118937334A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of high-altitude large-scale steel structures, and in particular, relates to a method, medium and system for monitoring the degree of corrosion of high-altitude large-scale steel structures after construction. Background Art
[0002] Large-scale high-altitude steel structures are widely used in bridges, communication towers, wind power generation equipment and other fields. Such steel structures play an important supporting role. However, due to long-term exposure to harsh natural environments, these high-altitude steel structures are susceptible to corrosion, resulting in a gradual decline in structural performance and even serious safety hazards. Therefore, how to effectively monitor and evaluate the corrosion status of high-altitude steel structures has become a key issue to ensure their safety and service life.
[0003] The existing steel structure corrosion monitoring technologies mainly include the following:
[0004] 1. Manual visual inspection method: Professional inspectors regularly visually observe the surface of the steel structure and record the visible corrosion. This method relies on manual inspection, which requires high professional skills of inspectors and can only evaluate surface corrosion, but cannot find hidden defects in the internal structure. At the same time, due to the limitations of inspection in high-altitude environments, it is difficult to fully grasp the overall corrosion status.
[0005] 2. Traditional non-destructive testing methods: ultrasonic testing, magnetic particle testing and other technologies are used to perform non-destructive testing on steel structures. These methods can more deeply evaluate the internal structural changes of steel, but they require professional equipment and personnel, and the testing cost is high. At the same time, the limitations of the test are also strong, and it is difficult to conduct a comprehensive evaluation of the entire steel structure surface.
[0006] 3. Electrochemical corrosion monitoring method: Electrochemical sensors are installed on the surface of the steel structure to monitor the changes in corrosion current and potential in real time. This method can provide continuous corrosion trend data, but it requires a large number of sensors to be fixed on the steel structure, which has high installation and maintenance costs, and it is also impossible to evaluate the overall corrosion distribution.
[0007] The above-mentioned prior art has the following major problems:
[0008] 1. The detection range is limited, and it is difficult to fully understand the overall corrosion situation. Manual inspection and local non-destructive testing cannot cover the entire surface of the steel structure, and electrochemical monitoring can only evaluate individual locations.
[0009] 2. The detection accuracy is not high, and it is difficult to detect corrosion hazards in time. Manual visual inspection is greatly affected by factors such as the inspector's experience and environment, and it is impossible to detect subtle corrosion defects in a timely and accurate manner. Traditional non-destructive testing also has certain limitations, and it is difficult to accurately and quantitatively assess the degree of corrosion.
[0010] Therefore, there is an urgent need to develop a new steel structure corrosion monitoring method that can comprehensively and accurately evaluate the corrosion status of large-scale high-altitude steel structures. Summary of the invention
[0011] In view of this, the present invention provides a method, medium and system for monitoring the corrosion degree of large-scale high-altitude steel structures after construction, which can solve the technical problem that it is difficult to comprehensively and accurately evaluate the corrosion status of large-scale high-altitude steel structures in the prior art.
[0012] The present invention is achieved in that:
[0013] A first aspect of the present invention provides a method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction, which comprises the following steps:
[0014] S01, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction, and preprocessing to obtain original visible light images and original ultrasonic images;
[0015] S02, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction within the monitoring period, and pre-processing to obtain monitoring visible light images and monitoring ultrasonic images;
[0016] S03, performing cluster analysis on the monitoring visible light image to obtain a plurality of cluster center regions of the monitoring visible light image; performing cluster analysis on the monitoring ultrasound image to obtain a plurality of cluster center regions of the monitoring ultrasound image;
[0017] S04, obtaining a union of a plurality of cluster centers of the monitoring visible light image and a plurality of cluster center regions of the monitoring ultrasound image as a typical region set, including a plurality of typical regions;
[0018] S05, after uniformly spraying the citric acid solution on each typical area in the typical area set for 30 minutes, obtaining a visible light image and an ultrasonic image of the surface of the typical area, and pre-processing to obtain a first visible light image and a first ultrasonic image;
[0019] S06. After the surface of the steel structure is cleaned and dried, a visible light image and an ultrasonic image of the surface of the typical area are obtained, and a second visible light image and a second ultrasonic image are obtained by preprocessing;
[0020] S07. For each typical region, establish a region image set, including an original region image, an original ultrasound region image, a monitoring region image, a monitoring ultrasound region image, a first visible light image, a first ultrasound image, a second visible light image, and a second ultrasound image, wherein the original region image and the original ultrasound region image are images of regions corresponding to the typical region in the original visible light image and the original ultrasound image; the monitoring region image and the monitoring ultrasound region image are images of regions corresponding to the typical region in the monitoring visible light image and the monitoring ultrasound image;
[0021] S08. For each regional image set of a typical region, using the first visible light image and the first ultrasonic image, the second visible light image and the second ultrasonic image, adopting an image difference and feature extraction method in combination with a multimodal weighted analysis method, calculate the corrosion degree of the typical region;
[0022] S09, establishing a training data set, including the original regional image, the original ultrasonic regional image, the monitoring regional image, the monitoring ultrasonic regional image and the corrosion degree of the typical region in the regional image set of each typical region, training a neural network model, and obtaining a steel structure corrosion degree model;
[0023] S10. Using the steel structure corrosion degree model, inputting the original visible light image and the original ultrasonic image as well as the monitoring visible light image and the monitoring ultrasonic image, obtaining a corrosion degree matrix of the entire steel structure surface, and outputting the matrix.
[0024] Wherein, the step S01 specifically includes: first, use a high-resolution visible light camera and an ultrasonic flaw detector to comprehensively scan the entire surface of the steel structure to obtain the original visible light image and the original ultrasonic image. During scanning, the influence of factors such as shadows and reflections on the image quality should be avoided as much as possible, and the time correspondence of the two imaging data should be ensured. Secondly, it is necessary to perform preprocessing operations such as denoising, correction and color correction on the obtained original image data to improve the image quality and obtain the original visible light image and the original ultrasonic image for subsequent analysis. The purpose of this step is to obtain high-quality original image data and lay a reliable foundation for subsequent analysis and evaluation.
[0025] Wherein, the step S02 specifically includes: firstly, repeating step S01 within a certain time interval, using the same visible light camera and ultrasonic flaw detector to comprehensively scan the surface of the steel structure, and obtaining new visible light images and new ultrasonic images reflecting the changes in the monitoring time period. Secondly, the obtained new image data is subjected to the same preprocessing operation as step S01, including denoising, correction and color correction of visible light images, and denoising, image enhancement and signal correction of ultrasonic images, etc., to obtain monitoring visible light images and monitoring ultrasonic images for subsequent analysis. The purpose of this step is to obtain monitoring image data reflecting the changes in the state of the steel structure, laying a foundation for the next step of comparative analysis and corrosion degree assessment.
[0026] Wherein, the step S03 specifically includes: using the K-means clustering algorithm to analyze the monitoring visible light image, dividing the pixels on the image into multiple clusters, each cluster having a cluster center. In this way, multiple cluster center areas in the monitoring visible light image can be obtained, reflecting the appearance changes of the steel structure surface to varying degrees. Using the DBSCAN density clustering algorithm to analyze the monitoring ultrasonic image, the dense areas in the image can be automatically found and divided into different clusters. In this way, multiple cluster center areas in the monitoring ultrasonic image can be obtained, reflecting the changes in the internal structure of the steel structure. The purpose of this step is to extract representative key areas from the monitoring image to provide focus objects for subsequent analysis and comparison.
[0027] Wherein, the step S04 specifically includes: first, according to the result of step S03, obtain K cluster center areas of the monitoring visible light image and M cluster center areas of the monitoring ultrasonic image. Then, these visible light and ultrasonic cluster center areas are set and combined to obtain a set of N typical areas, namely, a typical area set. These typical areas are areas where there are significant changes in both visible light and ultrasonic images, reflecting the comprehensive changes in the surface and internal structure of the steel structure. The purpose of this step is to extract representative key areas from a large amount of monitoring image data to provide reliable sample data for subsequent corrosion analysis and prediction.
[0028] Wherein, the step S05 specifically includes: firstly, preparing a citric acid solution with a concentration of 10% to 15%. Citric acid has good cleaning and passivation effects and can effectively remove rust on the steel surface. Then, the citric acid solution is evenly sprayed on the surface of each typical area by spraying for 30 minutes. After the spraying is completed, the surface of the treated typical area is immediately scanned by a visible light camera and an ultrasonic flaw detector to obtain the first batch of visible light images and ultrasonic images. Finally, the first batch of visible light images and ultrasonic images obtained are pre-processed to obtain the first visible light image and the first ultrasonic image. The purpose of this step is to remove part of the rust on the steel surface by treatment with the citric acid solution, and to obtain the initial image data after treatment, so as to lay the foundation for the subsequent comparative analysis with the untreated state.
[0029] Wherein, the step S06 specifically includes: first, cleaning and drying the surface of the steel structure sprayed with the citric acid solution in step S05 to remove the residual citric acid solution. Then, the visible light camera and the ultrasonic flaw detector are used again to scan the surface of the typical area after cleaning and drying to obtain a second batch of visible light images and ultrasonic images. Finally, the obtained second batch of visible light images and ultrasonic images are preprocessed to obtain a second visible light image and a second ultrasonic image. The purpose of this step is to obtain the latest state image data of the typical area of the steel structure after removing the influence of the citric acid solution, so as to provide a basis for subsequent comparative analysis and corrosion degree calculation.
[0030] Wherein, the step S07 specifically includes: for each typical area, establishing a regional image set including the following contents: original regional image and original ultrasonic regional image, monitoring regional image and monitoring ultrasonic regional image, first visible light image and first ultrasonic image, second visible light image and second ultrasonic image. By establishing such a regional image set including multiple image data, a comprehensive information basis can be provided for subsequent comparative analysis and corrosion degree calculation. These image data cover the changes of typical areas of steel structures at different time points and processing states.
[0031] Wherein, the step S08 specifically includes: firstly, finely aligning the four images (first visible light, first ultrasound, second visible light, second ultrasound) to ensure the correspondence at the pixel level. Then, multiple indicators such as visible light image difference, color feature change, ultrasound image difference, surface roughness change, material density feature change and texture feature change are calculated. Finally, these feature values are weighted and fused to obtain the calculation formula for the rust degree of the typical area. By comprehensively utilizing the advantages of visible light and ultrasound imaging data, the corrosion changes on the surface and inside of the steel structure can be fully reflected.
[0032] Wherein, the step S09 specifically includes: firstly, extracting the original regional image, the original ultrasonic regional image, the monitoring regional image, the monitoring ultrasonic regional image and the corresponding corrosion degree from the regional image set established in step S07 to form a training data set. Then, a multi-branch convolutional neural network is selected as the model structure, including a module for extracting the original visible light feature, the original ultrasonic feature, the monitoring visible light feature and the monitoring ultrasonic feature, and a module for fusing microscopic and macroscopic features. Finally, the neural network model is trained using the training data set to obtain a model for predicting the overall corrosion degree of the steel structure. This method can make full use of the advantages of multimodal image data and improve the accuracy and robustness of corrosion degree prediction.
[0033] Wherein, the step S10 specifically includes: inputting the original visible light image and the original ultrasonic image obtained in step S01, and the monitoring visible light image and the monitoring ultrasonic image obtained in step S02 into the steel structure corrosion prediction model trained in step S09. The prediction model will automatically extract the features in these input images, and calculate the corrosion matrix of the entire steel structure surface, reflecting the corrosion degree of each part. Finally, the calculated overall corrosion matrix is output as the final result of the method of the present invention. This method greatly improves the analysis efficiency, while also ensuring the accuracy and comprehensiveness of the evaluation, providing an important reference for subsequent maintenance.
[0034] Wherein, the K-means clustering algorithm is used to perform cluster analysis on the monitored visible light images.
[0035] Optionally, the K value of the K-means clustering algorithm is set to 5 or 10 to ensure that different degrees of corrosion areas can be captured. The algorithm can divide the pixels on the monitored visible light image into multiple clusters, each cluster has a representative central area, reflecting the different degrees of appearance changes on the surface of the steel structure. Through cluster analysis, key areas with significant changes in the image can be automatically discovered, providing key focus objects for subsequent corrosion assessment.
[0036] In addition, the DBSCAN density clustering algorithm is used to perform cluster analysis on the monitoring ultrasound images.
[0037] Optionally, the DBSCAN density clustering algorithm analyzes the monitoring ultrasound image. The algorithm can automatically find dense areas in the image and divide them into different clusters, overcoming the interference of noise and outliers. Through DBSCAN clustering, multiple cluster center areas in the monitoring ultrasound image can be obtained, reflecting the changes in the internal structure of the steel structure. These areas, together with the visible light cluster center areas determined in step S04, constitute a typical area set that fully reflects the changes in the condition of the steel structure.
[0038] Wherein, the concentration of the citric acid solution is 10% to 15%.
[0039] The neural network model adopts a multi-branch convolutional neural network structure, and each layer is an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer, wherein the convolution layer includes 5 modules, which are as follows:
[0040] The original visible light module is used to perform multi-scale convolution and maximum pooling operations on the input original area image to obtain the original visible light feature map;
[0041] The original ultrasound module is used to perform spatial pyramid pooling and attention mechanism processing on the input original ultrasound region image to obtain the original ultrasound feature map;
[0042] The monitoring visible light module is used to perform deep residual network processing and hole convolution operations on the input monitoring area image to obtain the monitoring visible light feature map;
[0043] The monitoring ultrasound module is used to perform feature pyramid network processing and channel attention mechanism operation on the input monitoring ultrasound area image to obtain a monitoring ultrasound feature map;
[0044] The fusion module uses feature concatenation, weighted summation and adaptive fusion to achieve comprehensive analysis and fusion of the four feature maps.
[0045] Furthermore, the fusion module includes micro-feature fusion and macro-feature fusion. The micro-feature fusion adopts a multi-head self-attention mechanism structure to capture subtle corrosion features and local changes; the macro-feature fusion adopts a bilinear pooling structure to extract the overall corrosion distribution and trend characteristics.
[0046] Specifically:
[0047] Micro-feature fusion uses a multi-head self-attention mechanism, which can simultaneously focus on the feature interactions of different subspaces and better capture subtle local features such as rust spots and cracks.
[0048] Macro feature fusion adopts bilinear pooling. By calculating the outer product between feature maps, high-order features can be effectively extracted to better express the overall corrosion distribution and change trend.
[0049] Finally, the micro and macro features are combined through a concatenation operation and then passed through one or more fully connected layers to output the rust degree prediction value.
[0050] Among them, the scope of micro-feature fusion is:
[0051] The first signs of rust, such as small changes in color or subtle changes in surface texture
[0052] Tiny rust spots, usually less than 1mm in diameter
[0053] Tiny cracks or holes, usually less than 0.1mm in width
[0054] Slight changes in local surface roughness
[0055] Small echo changes in ultrasound images reflect subtle changes in the internal structure of the material
[0056] The scope of macro feature fusion is:
[0057] Large areas of rust, usually covering an area greater than 1cm 2
[0058] Visible color changes, such as large reddish-brown areas
[0059] Large cracks or peeling, width or area exceeding 1mm
[0060] Changes in the overall surface morphology, such as large-scale unevenness
[0061] Large-scale features in ultrasound images, such as overall changes in material thickness
[0062] Overlap range:
[0063] Medium-sized rusted area, about 1 mm in size 2 To 1cm 2 between
[0064] The process of the rust spots gradually expanding from tiny rust spots to visible rust spots
[0065] A gradual change in surface texture, from microscopic roughness to macroscopically visible roughness
[0066] The development process of cracks, from small cracks to visible cracks
[0067] Medium-scale features in ultrasound images, such as local thickness variations or areas of gradual changes in material density
[0068] The advantages of this design are:
[0069] 1. Comprehensiveness: By focusing on both microscopic and macroscopic features, all stages and manifestations of rust can be fully captured.
[0070] 2. Continuity: Overlapping ranges ensure a continuous transition from micro to macro, avoiding “blind spots” in feature extraction.
[0071] 3. Complementarity: Microscopic features help with early rust detection and detailed assessment, while macroscopic features reflect the overall degree of rust, and the two complement each other.
[0072] 4. Multi-scale analysis: This method achieves true multi-scale analysis and can adapt to rust conditions of different sizes and degrees.
[0073] 5. Robustness: By comprehensively considering the characteristics of different scales, the model's adaptability and prediction accuracy to various corrosion conditions are improved.
[0074] Through this combination of microscopic and macroscopic features, as well as the setting of overlapping ranges, the model can more comprehensively and accurately assess the corrosion condition of steel structures, effectively capturing and analyzing everything from early minor changes to large-scale severe corrosion. This design combines a variety of advanced deep learning techniques, which can simultaneously take advantage of visible light and ultrasonic images, capturing subtle corrosion features while reflecting the overall corrosion condition, thereby improving the accuracy and robustness of corrosion prediction.
[0075] The step of calculating the corrosion degree of the typical area is specifically:
[0076] Step 801. Register the four images to ensure pixel-level correspondence;
[0077] Step 802: Calculate the pixel-level difference between the first visible light image and the second visible light image to obtain a visible light difference value D_visible;
[0078] Step 803: Extract color features, pay special attention to the color changes related to rust, which include reddish brown, orange brown and dark red, calculate the pixel proportion changes of these colors, and obtain the color change value C_change;
[0079] Step 804: Calculate the pixel-level difference between the first ultrasonic image and the second ultrasonic image to obtain an ultrasonic difference value D_ultrasound;
[0080] Step 805: Analyze the change of depth information, pay attention to the increase of surface roughness, specifically calculate the surface profile change rate R_profile and the depth variance increase degree V_depth of the ultrasound image;
[0081] Step 806: extracting material density related features, including the change of the sound wave reflection intensity ΔI_reflection and the change of the sound wave attenuation coefficient ΔA_attenuation;
[0082] Step 807: Calculate the texture feature change, including the change ΔT_GLCM of the gray level co-occurrence matrix (GLCM) feature;
[0083] Step 808: extract morphological features and calculate the area change ratio ΔS_area of the rusted area;
[0084] Step 809. Define the corrosion degree calculation formula as follows:
[0085] Rust_Degree=w1*D_visible+w2*C_change+w3*D_ultrasound+w4*R_profile+w5*V_depth
[0086] +w6*ΔI_reflection+w7*ΔA_attenuation+w8*ΔT_GLCM+w9*ΔS_area
[0087] Among them, w1 to w9 are weight coefficients, which must satisfy Σwi=1;
[0088] Step 810. Normalize the calculated Rust_Degree value to a range of 0-100%, where 0% indicates no rust and 100% indicates complete rust;
[0089] The normalization formula is:
[0090] Normalized_Rust_Degree=(Rust_Degree-Min_Rust_Degree) / (Max_Rust_Degree-Min_Rust_Degree)*100%;
[0091] Among them, Min_Rust_Degree and Max_Rust_Degree are the theoretical minimum and maximum rust values respectively;
[0092] Step 811: Output the normalized rust value as the final rust evaluation result of the typical area.
[0093] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction.
[0094] A third aspect of the present invention provides a system for monitoring the degree of corrosion of a large-scale high-altitude steel structure after construction, which includes the above-mentioned computer-readable storage medium.
[0095] Compared with the prior art, the present invention provides a method, medium and system for monitoring the corrosion degree of large-scale high-altitude steel structures after construction, which have the following beneficial effects:
[0096] 1. Comprehensively monitor the overall corrosion status of the steel structure. This method uses a visible light camera and an ultrasonic flaw detector to comprehensively scan the entire surface of the steel structure to obtain multimodal image data covering surface appearance and internal structure information. By comprehensively analyzing these data, the overall corrosion degree of the steel structure can be accurately assessed, not limited to individual parts.
[0097] 2. Accurately and quantitatively analyze the degree of corrosion. This method uses image difference, feature extraction, and multimodal fusion techniques to conduct detailed analysis and quantification of various indicators in visible light images and ultrasonic images, which can more accurately reflect the specific corrosion conditions on the surface and inside of the steel structure. Compared with manual visual inspection and simple non-destructive testing, this method can more carefully discover and evaluate corrosion hazards.
[0098] 3. High degree of automation, greatly reducing labor costs. This method uses advanced deep learning algorithms to automatically evaluate corrosion conditions, without relying on a large number of professionals for manual inspection. Compared with manual visual inspection and traditional non-destructive testing methods, the installation, maintenance and operation costs of this method are greatly reduced. At the same time, due to its automation and intelligent characteristics, it also greatly improves the efficiency and frequency of monitoring.
[0099] In summary, the high-altitude large-scale steel structure corrosion monitoring method proposed in the present invention fully integrates the advantages of multimodal image data and adopts advanced digital analysis technology to achieve a comprehensive and accurate assessment of the overall corrosion status of the steel structure. BRIEF DESCRIPTION OF THE DRAWINGS
[0100] Figure 1 A flow chart of the method provided by the present invention. DETAILED DESCRIPTION
[0101] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0102] like Figure 1 As shown, it is a flow chart of a method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction provided by the present invention. The method comprises the following steps:
[0103] S01, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction, and preprocessing to obtain original visible light images and original ultrasonic images;
[0104] S02, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction within the monitoring period, and pre-processing to obtain monitoring visible light images and monitoring ultrasonic images;
[0105] S03, performing cluster analysis on the monitoring visible light image to obtain multiple cluster center regions of the monitoring visible light image; performing cluster analysis on the monitoring ultrasound image to obtain multiple cluster center regions of the monitoring ultrasound image;
[0106] S04, obtaining a union of multiple cluster centers of the monitoring visible light image and multiple cluster center regions of the monitoring ultrasound image as a typical region set, including multiple typical regions;
[0107] S05, after uniformly spraying the citric acid solution on each typical area in the typical area set for 30 minutes, obtaining a visible light image and an ultrasonic image of the surface of the typical area, and pre-processing to obtain a first visible light image and a first ultrasonic image;
[0108] S06. After cleaning and drying the surface of the steel structure, obtain a visible light image and an ultrasonic image of the surface of a typical area, and pre-process to obtain a second visible light image and a second ultrasonic image;
[0109] S07. For each typical region, establish a region image set, including an original region image, an original ultrasound region image, a monitoring region image, a monitoring ultrasound region image, a first visible light image, a first ultrasound image, a second visible light image, and a second ultrasound image, wherein the original region image and the original ultrasound region image are images of regions corresponding to the typical region in the original visible light image and the original ultrasound image; the monitoring region image and the monitoring ultrasound region image are images of regions corresponding to the typical region in the monitoring visible light image and the monitoring ultrasound image;
[0110] S08. For each regional image set of a typical region, using the first visible light image and the first ultrasonic image, the second visible light image and the second ultrasonic image, adopting an image difference and feature extraction method in combination with a multimodal weighted analysis method, calculate the corrosion degree of the typical region;
[0111] S09, establishing a training data set, including the original regional image, the original ultrasonic regional image, the monitoring regional image, the monitoring ultrasonic regional image and the corrosion degree of the typical region in the regional image set of each typical region, training a neural network model, and obtaining a steel structure corrosion degree model;
[0112] S10. Using the steel structure corrosion degree model, the original visible light image and the original ultrasonic image as well as the monitoring visible light image and the monitoring ultrasonic image are input to obtain the corrosion degree matrix of the entire steel structure surface, and output it.
[0113] The specific implementation methods of the above steps are described in detail below:
[0114] Step S01: Obtain visible light images and ultrasonic images of all surfaces of the high-altitude large-scale steel structure after construction, and pre-process to obtain original visible light images and original ultrasonic images.
[0115] The specific implementation method of this step is as follows:
[0116] First, it is necessary to use a high-resolution visible light camera and ultrasonic flaw detector to comprehensively scan the entire steel structure surface to obtain the original visible light image and ultrasonic image. It should be noted here that factors such as shadows and reflections should be avoided as much as possible during scanning to affect the image quality. At the same time, the acquisition of visible light images and ultrasonic images should be completed in the same time period to ensure the time correspondence of the two imaging data.
[0117] Secondly, the acquired raw image data needs to be preprocessed. For visible light images, denoising, correction, and color correction are required to improve image quality. For ultrasound images, denoising, image enhancement, and signal correction are required to improve the signal-to-noise ratio and spatial resolution of the echo signal.
[0118] After the preprocessing is completed, the original visible light image and original ultrasonic image for subsequent analysis can be obtained. These two types of imaging data reflect the appearance characteristics of the steel structure surface and the internal structure information, respectively, laying the foundation for subsequent multimodal analysis.
[0119] The main purpose of this step is to obtain high-quality original image data to ensure the reliability and accuracy of subsequent analysis and evaluation. The use of professional imaging equipment and appropriate preprocessing techniques can minimize the interference of external factors on image information and improve the signal-to-noise ratio of the data.
[0120] Step S02: Obtain visible light images and ultrasonic images of all surfaces of the large-scale high-altitude steel structure after construction within the monitoring time period, and pre-process to obtain monitoring visible light images and monitoring ultrasonic images.
[0121] The specific implementation method of this step is as follows:
[0122] First, it is necessary to repeat step S01 at a certain time interval (for example, every 6 months or 1 year), use the same visible light camera and ultrasonic flaw detector to fully scan the surface of the steel structure, and obtain new visible light images and ultrasonic images. These image data reflect the changes of the steel structure during the monitoring period.
[0123] Secondly, the acquired new image data needs to be preprocessed, which is the same as the preprocessing method in step S01, including denoising, correction and color correction of visible light images, and denoising, image enhancement and signal correction of ultrasound images.
[0124] After the preprocessing is completed, the monitoring visible light image and monitoring ultrasonic image for subsequent analysis can be obtained. These two types of data respectively represent the appearance changes and internal structure changes of the steel structure surface during the monitoring period.
[0125] The main purpose of this step is to obtain monitoring image data that reflects the changes in the state of the steel structure, laying the foundation for the next step of comparative analysis and corrosion assessment. Regular monitoring can dynamically grasp the corrosion status of the steel structure and provide a basis for timely maintenance.
[0126] Step S03: performing cluster analysis on the monitoring visible light image to obtain multiple cluster center areas of the monitoring visible light image; performing cluster analysis on the monitoring ultrasound image to obtain multiple cluster center areas of the monitoring ultrasound image.
[0127] The specific implementation method of this step is as follows:
[0128] For the monitoring visible light images, the K-means clustering algorithm is used for analysis. K-means clustering is an unsupervised learning algorithm that can divide the pixels on the image into K clusters, each with a cluster center. Here, it is necessary to set a suitable K value, such as K = 5 or K = 10, to ensure that different degrees of corrosion areas can be captured. Through K-means clustering, multiple cluster center areas in the monitoring visible light images can be obtained, which reflect the different degrees of appearance changes on the surface of the steel structure.
[0129] For monitoring ultrasound images, the DBSCAN density clustering algorithm is used for analysis. DBSCAN is a density-based clustering algorithm that can automatically discover dense areas in images and divide them into different clusters. This method is highly robust to noise and outliers and is very suitable for ultrasound image analysis. Through DBSCAN clustering, multiple cluster center areas in the monitoring ultrasound images can be obtained, which reflect the changes in the internal structure of the steel structure.
[0130] The main purpose of this step is to extract representative key areas from the monitoring images to provide focus for subsequent analysis and comparison. Through cluster analysis, areas with significant changes on the surface and internal structure of the steel structure can be automatically identified, laying the foundation for accurate assessment of corrosion conditions.
[0131] Step S04: obtaining a union of multiple cluster centers of the monitoring visible light image and multiple cluster center areas of the monitoring ultrasound image as a typical area set, including multiple typical areas.
[0132] The specific implementation method of this step is as follows:
[0133] First, according to the result of step S03, K cluster center regions for monitoring visible light images and M cluster center regions for monitoring ultrasound images are obtained.
[0134] Then, these visible light and ultrasonic cluster center regions are set and unioned to obtain a set of N typical regions, namely the typical region set. These typical regions are regions with significant changes in both visible light and ultrasonic images, reflecting the comprehensive changes in the surface and internal structure of the steel structure.
[0135] In order to ensure a comprehensive analysis of the corrosion of steel structures, these typical areas need to be selected as key areas of focus for detailed monitoring and evaluation in subsequent steps.
[0136] The main purpose of this step is to extract representative key areas from a large amount of monitoring image data to provide reliable sample data for subsequent corrosion analysis and prediction. Through the union operation, it can ensure that the most critical change areas on the surface and internal structure of the steel structure are included, laying the foundation for a comprehensive and accurate corrosion assessment.
[0137] Step S05: After uniformly spraying the citric acid solution on each typical area in the typical area set for 30 minutes, obtaining a visible light image and an ultrasonic image of the surface of the typical area, and pre-processing to obtain a first visible light image and a first ultrasonic image.
[0138] The specific implementation method of this step is as follows:
[0139] First, prepare a 10% to 15% concentration of citric acid solution. Citric acid has good cleaning and passivation effects, which can effectively remove rust on the steel surface and protect the steel from further corrosion.
[0140] Then, spray the citric acid solution evenly on the surface of each typical area by spraying, and continue spraying for 30 minutes. This ensures that the solution is in full contact with the steel surface and fully exerts its cleaning and passivation effects.
[0141] 30 minutes after spraying the citric acid solution, the surface of the typical treated area was scanned using a visible light camera and an ultrasonic flaw detector to obtain the first batch of visible light images and ultrasonic images. These image data reflect the initial changes in the surface and internal structure of the steel structure after the citric acid treatment.
[0142] Finally, the first batch of visible light images and ultrasonic images are preprocessed, including denoising, correction and other operations, to improve the image quality and obtain the first visible light image and the first ultrasonic image.
[0143] The main purpose of this step is to remove some of the rust on the steel surface through citric acid solution treatment and obtain the initial image data after treatment, laying the foundation for subsequent comparative analysis with the untreated state. The concentration and action time of the citric acid solution need to be optimized to ensure that its cleaning and passivation effects are fully exerted without causing excessive damage to the steel.
[0144] Step S06: After the surface of the steel structure is cleaned and dried, a visible light image and an ultrasonic image of the surface of the typical area are obtained, and a second visible light image and a second ultrasonic image are obtained by preprocessing.
[0145] The specific implementation method of this step is as follows:
[0146] First, the surface of the steel structure sprayed with the citric acid solution in step S05 is cleaned and dried. Cold water flushing and drying fans can be used to ensure that the surface is clean and dry and to remove the residual citric acid solution.
[0147] Then, the visible light camera and ultrasonic flaw detector were used to scan the typical area surface after cleaning and drying to obtain the second batch of visible light images and ultrasonic images. These image data reflect the further changes in the surface and internal structure of the steel structure after citric acid treatment.
[0148] Finally, the acquired second batch of visible light images and ultrasonic images are preprocessed, including denoising, correction and other operations, to improve the image quality and obtain the second visible light images and the second ultrasonic images.
[0149] The main purpose of this step is to obtain the latest status image data of the typical area of the steel structure after removing the influence of the citric acid solution, so as to provide a basis for subsequent comparative analysis and corrosion degree calculation. Through cleaning and drying, the interference of the citric acid solution on the subsequent analysis can be avoided.
[0150] Step S07: For each typical region, a regional image set is established, including an original regional image, an original ultrasonic regional image, a monitoring regional image, a monitoring ultrasonic regional image, a first visible light image, a first ultrasonic image, a second visible light image, and a second ultrasonic image.
[0151] The specific implementation method of this step is as follows:
[0152] First, based on the results of the previous steps, a regional image set is established for each typical region. This image set includes the following:
[0153] 1. Original region image and original ultrasound region image: that is, the local image region corresponding to the typical region in the original visible light and original ultrasound images acquired in step S01.
[0154] 2. Monitoring area image and monitoring ultrasound area image: that is, the local image area corresponding to the typical area in the monitoring visible light and monitoring ultrasound images obtained in step S02.
[0155] 3. The first visible light image and the first ultrasonic image: namely, the visible light and ultrasonic images after spraying the citric acid solution obtained in step S05.
[0156] 4. The second visible light image and the second ultrasonic image: namely, the visible light and ultrasonic images after cleaning and drying obtained in step S06.
[0157] By establishing such a regional image set containing a variety of image data, a comprehensive information basis can be provided for subsequent comparative analysis and corrosion degree calculation. These image data cover the changes in typical areas of steel structures at different time points and treatment states, laying a solid data foundation for accurately evaluating the degree of corrosion.
[0158] Step S08: For each regional image set of a typical region, the first visible light image and the first ultrasonic image, the second visible light image and the second ultrasonic image are used to calculate the degree of corrosion of the typical region by adopting an image difference and feature extraction method in combination with a multimodal weighted analysis method.
[0159] The specific implementation method of this step is as follows:
[0160] First, the four images (first visible light, first ultrasound, second visible light, second ultrasound) need to be finely registered to ensure pixel-level correspondence. This step is critical because pixel-level comparison and analysis are required later.
[0161] 1. Visible light image difference: Calculate the pixel-level difference between the first visible light image and the second visible light image to obtain the visible light difference value D visible This value reflects the degree of variation in color and texture of the steel surface.
[0162] 2. Color feature extraction: Focus on extracting color changes related to rust, such as reddish brown, orange brown, and dark red. Calculate the pixel ratio changes of these colors to obtain the color change value C change . This value indicates the degree of surface corrosion.
[0163] 3. Ultrasonic image difference: Calculate the pixel-level difference between the first ultrasonic image and the second ultrasonic image to obtain the ultrasonic difference value D ultrasound . This value reflects the changes in the internal structure of the steel.
[0164] 4. Surface roughness analysis: Calculate the surface contour change rate R of the ultrasound image profile and the depth variance increases by V depth , to evaluate the changes in surface morphology.
[0165] 5. Material density characteristics: extract the change in sound wave reflection intensity ΔI reflection and the change in the acoustic attenuation coefficient ΔA attenuation , reflecting the change of material density.
[0166] 6. Texture feature changes: Calculate the change ΔT of the gray level co-occurrence matrix (GLCM) features GLCM , characterizing the changes in surface texture features.
[0167] 7. Morphological characteristics: Calculate the area change ratio ΔS of the rusted area area , reflecting the growth of corrosion area.
[0168] Finally, the above multiple eigenvalues are weighted and fused to obtain the corrosion degree calculation formula of the typical area as follows:
[0169] Rust_Degree=w1·D visible +w2·C change +w3·D ultrasound +w4·R profile +w5·V depth +w6·ΔI reflection +w7·ΔA attenuation +w8·ΔT GLCM +w9·ΔS area
[0170] Among them, w1 to w9 are weight coefficients that need to be adjusted according to actual conditions, but they need to satisfy
[0171] This weighted fusion method can make full use of the different types of feature information contained in the visible light and ultrasonic imaging data to comprehensively reflect the corrosion changes on the surface and inside of the steel structure. The weights between different features can be adjusted according to actual conditions to maximize the accuracy and reliability of corrosion assessment.
[0172] For example, in the initial stage of slight rust, the weights of color features and surface texture features can be appropriately increased to better capture subtle changes in appearance; while in the severe corrosion stage, the weights of material density features and morphological features can be increased to more accurately reflect deep structural changes. Through this dynamic adjustment, it can be ensured that the model can give reliable evaluation results at various corrosion levels.
[0173] Finally, the calculated Rust_Degree value is normalized to obtain a normalized rust degree in the range of 0-100%:
[0174]
[0175] Among them, Min_Rust_Degree and Max_Rust_Degree are the theoretically possible minimum and maximum rust values, respectively. In this way, the rust degrees of different typical areas can be unified into a standard percentage range, which is convenient for comparative analysis and final evaluation.
[0176] The main purpose of this step is to accurately calculate the corrosion degree of each typical area through the extraction and fusion of multiple image features, laying the foundation for subsequent model training and overall evaluation. Through comparative analysis, feature extraction and weighted fusion, the corrosion changes on the surface and inside of the steel structure can be fully reflected.
[0177] Step S09: Establish a training data set, including the original area image, the original ultrasonic area image, the monitoring area image, the monitoring ultrasonic area image and the corrosion degree of the typical area in the area image set of each typical area, train a neural network model, and obtain a steel structure corrosion degree model.
[0178] The specific implementation method of this step is as follows:
[0179] First, based on the regional image set of each typical region established in step S07, the following data is extracted:
[0180] 1. Original region image and original ultrasound region image
[0181] 2. Monitoring area image and monitoring ultrasound area image
[0182] 3. The rust degree of the typical area calculated in step S08
[0183] These data are combined into a training data set for subsequent neural network model training.
[0184] Then, a multi-branch convolutional neural network structure is selected as the model architecture. The model contains the following main modules:
[0185] 1. Original visible light module: Perform multi-scale convolution and maximum pooling on the original area image to extract the original visible light features.
[0186] 2. Original ultrasound module: Perform spatial pyramid pooling and attention mechanism processing on the original ultrasound area image to extract the original ultrasound features.
[0187] 3. Monitoring visible light module: Perform deep residual network processing and hole convolution on the monitoring area image to extract the monitoring visible light features.
[0188] 4. Monitoring ultrasound module: Perform feature pyramid network processing and channel attention mechanism on the monitoring ultrasound area image to extract monitoring ultrasound features.
[0189] 5. Fusion module: It adopts multi-head self-attention mechanism and bilinear pooling structure to realize the fusion of micro and macro features.
[0190] Finally, the fused features are passed through the fully connected layer to output the final corrosion degree prediction value.
[0191] This multi-branch neural network structure can fully utilize the advantages of both visible light and ultrasonic imaging data to capture subtle local features and extract overall corrosion distribution and trend characteristics, thereby improving the accuracy and robustness of rust degree prediction.
[0192] By training this neural network model, a model for predicting the overall corrosion degree of steel structures can be obtained. The model can take the original visible light image, the original ultrasonic image, the monitoring visible light image and the monitoring ultrasonic image as input, and output the corrosion degree matrix of the entire steel structure surface, providing a basis for subsequent visualization and analysis.
[0193] Step S10: using the steel structure corrosion degree model, inputting the original visible light image and the original ultrasonic image as well as the monitoring visible light image and the monitoring ultrasonic image, obtaining the corrosion degree matrix of the entire steel structure surface, and outputting it.
[0194] The specific implementation method of this step is as follows:
[0195] First, the original visible light image and the original ultrasonic image obtained in step S01, and the monitoring visible light image and the monitoring ultrasonic image obtained in step S02, are input into the steel structure corrosion degree prediction model trained in step S09.
[0196] The prediction model automatically extracts the features in these input images and calculates the rust matrix of the entire steel structure surface through the weight parameters obtained through previous training. Each element in this rust matrix represents the degree of corrosion of the steel at the corresponding position.
[0197] Finally, the calculated overall corrosion matrix is output as the final result of the method of the present invention. This corrosion matrix can be used for subsequent visualization, data analysis, health assessment and other applications.
[0198] By utilizing the trained neural network model, the corrosion condition of the entire steel structure surface can be quickly and automatically assessed without having to check each typical area one by one. This approach greatly improves the efficiency of the analysis while also ensuring the accuracy and comprehensiveness of the assessment.
[0199] The output rust matrix data can intuitively reflect the degree of corrosion of each part of the steel structure surface, providing an important reference for subsequent maintenance. At the same time, by comparing the rust matrix of different time periods, the development trend of steel structure corrosion can also be analyzed to provide support for predictive maintenance.
[0200] In summary, the main purpose of step S10 is to use the prediction model established in the previous steps to comprehensively evaluate the corrosion condition of the entire steel structure surface, output detailed rust matrix results, and provide a reliable basis for subsequent status analysis and maintenance decisions.
[0201] The whole method uses multimodal analysis of visible light images and ultrasonic images, combined with deep learning technology, to achieve a comprehensive and accurate assessment of the corrosion of steel structures. Through step-by-step design and specific implementation, it ensures that each link can be supported by reliable data, and the final output of the rust matrix can truly reflect the corrosion status of the surface and interior of the steel structure.
[0202] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction.
[0203] A third aspect of the present invention provides a system for monitoring the degree of corrosion of a large-scale high-altitude steel structure after construction, which includes the above-mentioned computer-readable storage medium.
[0204] Specifically, the principle of the present invention is to use two multimodal imaging data, visible light image and ultrasonic image, to achieve a comprehensive assessment of the overall corrosion status through systematic image preprocessing, cluster analysis, feature extraction and neural network modeling. The reason why this method can effectively solve the problems existing in the prior art is mainly reflected in the following aspects:
[0205] 1. Multimodal data fusion for comprehensive perception. Visible light images can reflect the appearance characteristics of the steel structure surface, such as color and texture, while ultrasonic images can provide information on the internal structure of the steel, such as surface roughness and material density. By fusing and analyzing these two types of imaging data, we can fully perceive the corrosion changes on the surface and inside of the steel structure, without missing the details of surface corrosion or ignoring the hidden defects of the internal structure. This multimodal data fusion lays a solid foundation for subsequent accurate corrosion assessment.
[0206] 2. Typical area analysis, focusing on key areas. By clustering the monitoring images, this method can automatically find key areas with significant changes on the surface and inside of the steel structure, which serve as the focus of subsequent analysis. These typical areas not only reflect the overall corrosion trend, but also include the most severely corroded areas. In-depth analysis of these key areas can more specifically evaluate the overall corrosion status and improve the accuracy of the evaluation.
[0207] 3. Multi-feature fusion for accurate quantitative analysis. This method not only extracts various corrosion-related features in visible light images and ultrasonic images, such as color change, surface roughness, material density, etc., but also uses weighted fusion to comprehensively analyze these features. This multi-dimensional feature fusion can more comprehensively reflect all aspects of steel structure corrosion, thereby obtaining more accurate quantitative evaluation results. Compared with single manual visual inspection or simple non-destructive testing, this method has stronger analytical capabilities and reliability.
[0208] 4. Neural network modeling to achieve intelligent evaluation. The present invention adopts a multi-branch convolutional neural network model, which can automatically extract features and perform fusion analysis on the input visible light and ultrasonic images, and finally output the overall corrosion assessment results. This intelligent analysis method based on deep learning gets rid of the dependence on professionals and greatly improves the efficiency and replicability of monitoring. At the same time, through continuous model training and optimization, this method can also continuously improve the accuracy and robustness of corrosion assessment.
[0209] In order to better understand and implement the present invention, a specific embodiment of the present invention is provided below: A steel group has a large-scale high-altitude communication tower, which is 200 meters high and uses a steel structure to bear the weight. With long-term exposure to complex natural environments, the surface and internal structure of the communication tower began to show varying degrees of corrosion. In order to ensure the safe use of the communication tower, the steel group decided to use the post-construction rust degree monitoring method of the high-altitude large-scale steel structure proposed in the present invention to conduct a comprehensive assessment of the corrosion status of the communication tower.
[0210] The first step is to obtain the original visible light image and ultrasonic image of the entire surface of the communication tower.
[0211] The steel group selected a high-resolution visible light camera and an advanced ultrasonic flaw detector to conduct a comprehensive scan of the entire surface of the communication tower. The image resolution of the visible light camera is 4096×3072 pixels, and the image resolution of the ultrasonic flaw detector is 2048×1536 pixels. In order to avoid the influence of factors such as shadows and reflections on the image quality, they chose to carry out the scanning operation under sunny and light wind conditions. At the same time, they also ensured that the visible light camera and the ultrasonic flaw detector worked synchronously to ensure the time correspondence of the collected data. After the scan is completed, a series of preprocessing operations are performed to obtain the final original visible light image I origin,visible and the original ultrasound image I origin,ultrasound For visible light images, denoising, distortion correction and color correction are first performed to improve image quality; for ultrasound images, denoising, echo enhancement and signal correction are performed to enhance the signal-to-noise ratio and spatial resolution of echo signals. After these preprocessing steps, the original image data is optimized and enhanced, laying a solid foundation for subsequent analysis and evaluation.
[0212] The second step is to obtain the latest visible light image and ultrasonic image of the communication tower during the monitoring period.
[0213] In order to dynamically monitor the corrosion status of the communication tower, the company decided to conduct a scan inspection every six months. Six months after the previous scan, they used the same visible light camera and ultrasonic flaw detector to conduct a comprehensive scan of the communication tower surface and obtained the latest monitoring visible light image. monitor,visible and monitoring ultrasound images I monitor,ultrasound Consistent with the preprocessing method in step 1, these newly acquired image data were also denoised, corrected, and enhanced to ensure that the image quality met the requirements of subsequent analysis.
[0214] By comparing the original image and the monitoring image, we can preliminarily observe some changes that have occurred on the surface and internal structure of the communication tower in the past six months. For example, some new dark red and orange-brown spots appeared in the visible light image, and the ultrasonic image also showed that the echo signal in some areas had attenuated. These changes may be related to the increased degree of corrosion of the communication tower. In order to further analyze and quantify these changes, subsequent clustering analysis and feature extraction are required.
[0215] The third step is to perform cluster analysis on the monitoring visible light images and monitoring ultrasound images.
[0216] In order to accurately locate the key areas where there are significant changes in the surface and internal structure of the communication tower, the research team first monitored the visible light image I monitor,visible and monitoring ultrasound images I monitor,ultrasound Cluster analysis was performed.
[0217] For monitoring visible light images I monitor,visible , they used the K-means clustering algorithm for analysis. After many attempts, they finally determined that the number of cluster centers K was 8, which can better capture the different degrees of corrosion areas on the surface of the communication tower. The K-means algorithm divides the pixels on the image into 8 clusters, each of which has a representative center area C visible These cluster center areas reflect the obvious color changes and texture differences on the surface of the communication tower, which are consistent with the dark red and orange-brown spots initially observed.
[0218] For monitoring ultrasound images I monitor,ultrasound The research team used the DBSCAN density clustering algorithm for analysis. The DBSCAN algorithm can automatically find dense areas in the image and divide them into different clusters. By adjusting the algorithm parameters ∈ and MinPts, M = 12 ultrasound cluster center areas C were finally determined. ultrasound These areas reflect some changes in the internal structure of the communication tower, such as the obvious weakening of the echo signal in some areas, which may be caused by the reduction of material density or the increase of surface roughness.
[0219] Through the above cluster analysis, the research team identified a total of N = 20 representative key areas in the monitoring visible light images and monitoring ultrasound images, namely the typical area set C typical These areas cover areas where significant changes have occurred on the tower surface and in its internal structure, providing key areas of focus for subsequent corrosion assessments.
[0220] In the fourth step, each area in the typical area set was treated with citric acid solution and imaged.
[0221] In order to better analyze the extent of corrosion in typical areas of communication towers, the research team decided to treat these areas with a citric acid solution and record the image changes before and after treatment.
[0222] First, they prepared a 12% citric acid solution c citric =0.12. Citric acid has good cleaning and passivation effects, which can effectively remove rust on the steel surface and prevent further corrosion.
[0223] Then, spray the citric acid solution evenly on each typical area for 30 minutes. spray = 30min. This ensures that the solution is in full contact with the steel surface and exerts the maximum cleaning and passivation effect.
[0224] Immediately after the spraying was completed, the research team used a visible light camera and an ultrasonic flaw detector to scan a typical treated area and obtained the first batch of visible light images.first,visible and ultrasound images I first,ultrasound These images show the initial changes to the tower's surface and internal structure after treatment with a citric acid solution.
[0225] Afterwards, the research team carefully cleaned and dried the surface of the communication tower treated with citric acid to remove the residual solution. The same typical area was scanned again using a visible light camera and an ultrasonic flaw detector to obtain a second batch of visible light images. second,visible and ultrasound images I second,ultrasound The image data reflects the latest status of the communication tower after cleaning and drying.
[0226] By comparing and analyzing these two batches of images, the research team was able to initially observe changes in typical areas of the communication tower before and after the citric acid treatment. first,visible In the I first,ultrasound The echo signal intensity in some areas also increased. This shows that the citric acid solution does play a role in cleaning and passivation, but to further quantify the changes in the degree of corrosion, it is necessary to extract various characteristic indicators for comprehensive evaluation.
[0227] The fifth step is to establish a regional image set for each typical area to provide a data basis for subsequent analysis.
[0228] In order to make full use of the various types of image data obtained above, the research team established a regional image set R for each typical area, which includes the following:
[0229] 1. Original visible light region image I origin,visible and the original ultrasound region image I origin,ultrasound : That is, the local image area corresponding to the typical area in the original image data obtained in step one.
[0230] 2. Monitoring of visible light region images I monitor,visible and monitoring ultrasound area image I monitor,ultrasound : That is, the local image area corresponding to the typical area in the monitoring image data obtained in step 2.
[0231] 3. The first batch of visible light images I first,visible and the first ultrasound region images I first,ultrasound : That is, the first batch of image data obtained after citric acid treatment in step 4 corresponds to the typical area.
[0232] 4. The second batch of visible light images I second,visible and the second batch of ultrasound region images I second,ultrasound : That is, the second batch of image data obtained after cleaning and drying in step 4 corresponds to the typical area.
[0233] By establishing such a regional image set containing a variety of image data, the research team provided a comprehensive data basis for subsequent comparative analysis and corrosion assessment. These image data cover the changes in typical areas of communication towers at different time points and treatment states, laying a solid foundation for accurately assessing the degree of corrosion.
[0234] The sixth step is to extract features from the regional image set and calculate the corrosion degree of each typical area.
[0235] With abundant regional image data, the research team began to extract various characteristic indicators related to corrosion for each typical area, and weightedly fused these indicators to calculate the final corrosion degree of the area.
[0236] Specifically, they first first,visible ,I first,ultrasound ,I second,visible and I second,ultrasound The four images were finely registered at the pixel level to ensure that subsequent feature extraction and comparative analysis were based on accurate spatial correspondence.
[0237] The research team then calculated the following characteristic metrics:
[0238] 1. Visible light difference value D visible : By calculating I second,visible and I first,visible The pixel-level difference reflects the overall degree of change in the visible light image.
[0239] 2. Color change value C change : Focus on extracting colors related to rust, such as dark red, orange-brown, etc., and calculate their second,visible The change in the pixel ratio in reflects the change in surface color.
[0240] 3. Ultrasonic difference value D ultrasound : By calculating I second,ultrasound and I first,ultrasound The pixel-level difference reflects the overall change degree in the ultrasound image.
[0241] 4. Surface profile change rate R profile :Calculation I first,ultrasound and I second,ultrasound The rate of change of the surface contour reflects the increase in surface roughness.
[0242] 5. Depth variance increase V depth :Calculation I first,ultrasound and I second,ultrasound The increase in the variance of mid-depth information also reflects the change in surface roughness.
[0243] 6. Change in sound wave reflection intensity ΔI reflection :Calculation I first,ultrasound and I second,ultrasound The change in the intensity of sound wave reflection reflects the change in material density.
[0244] 7. Sound wave attenuation coefficient change ΔA attenuation :Calculation I first,ultrasound and I second,ultrasound The change degree of the sound wave attenuation coefficient also reflects the change of material density.
[0245] 8. Texture feature change ΔT GLCM : Extract the gray level co-occurrence matrix (GLCM) texture features and calculate I first,visible ,I second,visible The change of this feature reflects the change of surface texture.
[0246] Finally, the research team weighted and integrated the above eight characteristic indicators to calculate the corrosion degree Rust_Degree of each typical area:
[0247] Rust_Degree=w1·D visible +w2·C change +w3·D ultrasound +w4·R profile +w5·V depth +w6·ΔI reflection +w7·ΔA attenuation +w8·ΔT GLCM
[0248] Among them, w1 to w8 are weight coefficients adjusted according to actual conditions, satisfying
[0249] Through this multi-feature fusion approach, the research team was able to comprehensively reflect the degree of corrosion in each typical area. For example, the difference in visible light images and color changes reflect changes in surface appearance, the difference in ultrasonic images and changes in material density reflect changes in internal structure, and surface roughness and texture changes comprehensively reflect the overall situation of corrosion. By weighting and combining these indicators, a more accurate corrosion assessment result can be obtained.
[0250] For example, in a typical area of a communication tower, the research team calculated that the corrosion degree was Rust_Degree = 78. This value indicates that the area has already experienced serious corrosion and needs to be paid close attention to.
[0251] The seventh step is to establish a training data set and train a neural network model for predicting the overall corrosion degree.
[0252] With the corrosion calculation results of each typical area, the research team began to build a neural network model for predicting the overall corrosion. The input of this model will be the original visible light image, the original ultrasonic image, the monitoring visible light image and the monitoring ultrasonic image, and the output will be the corrosion matrix of the entire communication tower surface.
[0253] First, the research team extracted the following data from the regional image set R of each typical area established above to form a training data set:
[0254] 1. Original visible light region image I origin,visible and the original ultrasound region image I origin,ultrasound
[0255] 2. Monitoring of visible light region images I monitor,visible and monitoring ultrasound area image I monitor,ultrasound
[0256] 3. The corrosion degree Rust_Degree of each typical area calculated in step 6
[0257] With such a training dataset containing image data and corrosion degree labels, the research team then built a multi-branch convolutional neural network as a prediction model. The main structure of the model includes:
[0258] 1. Original visible light feature extraction module f origin,visible :Yes origin,visible Perform multi-scale convolution and pooling to extract original visible light features.
[0259] 2. Original ultrasonic feature extraction module f origin,ultrasound :Yes origin,ultrasound Spatial pyramid pooling and attention mechanism processing are performed to extract original ultrasonic features.
[0260] 3. Monitoring visible light feature extraction module f monitor,visible :Yes monitor,visible Deep residual network processing and dilated convolution are performed to extract and monitor visible light features.
[0261] 4. Monitoring Ultrasonic Feature Extraction Module f monitor,ultrasound :Yes monitor,ultrasound Feature pyramid network processing and channel attention mechanism are used to extract monitoring ultrasound features.
[0262] 5. Feature fusion module f fusion :The multi-scale features in the original and monitored data are integrated by using micro-feature fusion (multi-head self-attention mechanism) and macro-feature fusion (bilinear pooling).
[0263] Finally, the fused features are passed through the fully connected layer to output the final corrosion degree prediction value Rust_Degree pred .
[0264] By repeatedly training and optimizing this neural network model, the research team finally obtained a model that can accurately predict the overall corrosion degree. This model can fully explore various corrosion-related features contained in visible light and ultrasonic image data, and integrate them in an appropriate way, ultimately giving a more comprehensive and accurate evaluation result.
[0265] In the eighth step, the trained neural network model is used to evaluate the corrosion condition of the entire communication tower.
[0266] With the above-trained corrosion prediction model, the research team can use the original visible light image I obtained in step 1 and step 2 origin,visible , original ultrasound image I origin,ultrasound , monitor visible light image I monitor,visible and monitoring ultrasound images I monitor,ultrasound Input into the model to obtain the corrosion degree matrix Rust_Degree of the entire communication tower surface matrix .
[0267] Each element in this corrosion matrix represents the degree of corrosion at a specific location on the communication tower. Based on this matrix, the research team can intuitively understand the corrosion distribution of the entire communication tower surface.
[0268] By comparing the corrosion matrix of different time periods, the research team can also analyze the development trend of the communication tower corrosion. For example, they observed that in half a year, the corrosion degree of the area at a height of about 30 meters on the southwest side of the communication tower increased from the original 60% to 85%, which means that the corrosion in this area is intensifying. At the same time, some new corrosion points appeared at a height of about 100 meters on the northeast side of the communication tower, with a corrosion degree of about 10%. This dynamic monitoring can provide an important basis for subsequent maintenance work.
[0269] In general, using the trained neural network model, the research team was able to quickly and automatically conduct a comprehensive assessment of the corrosion status of the entire communication tower and output a detailed corrosion matrix result. This method not only improves the efficiency of monitoring, but also greatly improves the accuracy and reliability of the assessment, providing strong support for ensuring the safe use of communication towers.
Claims
1. A method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction, characterized in that: The following steps are involved: S01, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction, and preprocessing to obtain original visible light images and original ultrasonic images; S02, obtaining visible light images and ultrasonic images of all surfaces of the high-altitude large steel structure after construction within the monitoring period, and pre-processing to obtain monitoring visible light images and monitoring ultrasonic images; S03, performing cluster analysis on the monitoring visible light image to obtain a plurality of cluster center regions of the monitoring visible light image; performing cluster analysis on the monitoring ultrasound image to obtain a plurality of cluster center regions of the monitoring ultrasound image; S04, obtaining a union of a plurality of cluster centers of the monitoring visible light image and a plurality of cluster center regions of the monitoring ultrasound image as a typical region set, including a plurality of typical regions; S05, after uniformly spraying the citric acid solution on each typical area in the typical area set for 30 minutes, obtaining a visible light image and an ultrasonic image of the surface of the typical area, and pre-processing to obtain a first visible light image and a first ultrasonic image; S06. After the surface of the steel structure is cleaned and dried, a visible light image and an ultrasonic image of the surface of the typical area are obtained, and a second visible light image and a second ultrasonic image are obtained by preprocessing; S07. For each typical region, establish a region image set, including an original region image, an original ultrasound region image, a monitoring region image, a monitoring ultrasound region image, a first visible light image, a first ultrasound image, a second visible light image, and a second ultrasound image, wherein the original region image and the original ultrasound region image are images of regions corresponding to the typical region in the original visible light image and the original ultrasound image; the monitoring region image and the monitoring ultrasound region image are images of regions corresponding to the typical region in the monitoring visible light image and the monitoring ultrasound image; S08. For each regional image set of a typical region, using the first visible light image and the first ultrasonic image, the second visible light image and the second ultrasonic image, adopting an image difference and feature extraction method in combination with a multimodal weighted analysis method, calculate the corrosion degree of the typical region; S09, establishing a training data set, including the original regional image, the original ultrasonic regional image, the monitoring regional image, the monitoring ultrasonic regional image and the corrosion degree of the typical region in the regional image set of each typical region, training a neural network model, and obtaining a steel structure corrosion degree model; S10. Using the steel structure corrosion degree model, inputting the original visible light image and the original ultrasonic image as well as the monitoring visible light image and the monitoring ultrasonic image, obtaining a corrosion degree matrix of the entire steel structure surface, and outputting the matrix.
2. A method for monitoring the corrosion degree of a high-altitude large steel structure after construction according to claim 1, characterized in that: The K-means clustering algorithm is used to perform cluster analysis on the monitored visible light images.
3. The method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction according to claim 1 is characterized in that: The DBSCAN density clustering algorithm is used to perform cluster analysis on the monitoring ultrasound images.
4. The method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction according to claim 1 is characterized in that: The concentration of the citric acid solution is 10% to 15%.
5. The method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction according to claim 1 is characterized in that: The neural network model adopts a multi-branch convolutional neural network structure, and each layer is an input layer, a convolution layer, a pooling layer, a fully connected layer and an output layer, wherein the convolution layer includes 5 modules, which are as follows: The original visible light module is used to perform multi-scale convolution and maximum pooling operations on the input original area image to obtain the original visible light feature map; The original ultrasound module is used to perform spatial pyramid pooling and attention mechanism processing on the input original ultrasound region image to obtain the original ultrasound feature map; The monitoring visible light module is used to perform deep residual network processing and hole convolution operations on the input monitoring area image to obtain the monitoring visible light feature map; The monitoring ultrasound module is used to perform feature pyramid network processing and channel attention mechanism operation on the input monitoring ultrasound area image to obtain a monitoring ultrasound feature map; The fusion module uses feature concatenation, weighted summation and adaptive fusion to achieve comprehensive analysis and fusion of the four feature maps.
6. A method for monitoring the corrosion degree of a high-altitude large steel structure after construction according to claim 5, characterized in that: The fusion module includes micro-feature fusion and macro-feature fusion. The micro-feature fusion adopts a multi-head self-attention mechanism structure to capture subtle corrosion features and local changes; the macro-feature fusion adopts a bilinear pooling structure to extract the overall corrosion distribution and trend characteristics.
7. The method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction according to claim 1 is characterized in that: The steps for calculating the rust degree of the typical area are as follows: Step 801. Register the four images to ensure pixel-level correspondence; Step 802: Calculate the pixel-level difference between the first visible light image and the second visible light image to obtain a visible light difference value D_visible; Step 803: Extract color features, pay special attention to the color changes related to rust, which include reddish brown, orange brown and dark red, calculate the pixel proportion changes of these colors, and obtain the color change value C_change; Step 804: Calculate the pixel-level difference between the first ultrasonic image and the second ultrasonic image to obtain an ultrasonic difference value D_ultrasound; Step 805: Analyze the change of depth information, pay attention to the increase of surface roughness, specifically calculate the surface profile change rate R_profile and the depth variance increase degree V_depth of the ultrasound image; Step 806: extracting material density related features, including the change of the sound wave reflection intensity ΔI_reflection and the change of the sound wave attenuation coefficient ΔA_attenuation; Step 807: Calculate the texture feature change, including the change ΔT_GLCM of the gray level co-occurrence matrix (GLCM) feature; Step 808: extract morphological features and calculate the area change ratio ΔS_area of the rusted area; Step 809. Define the corrosion degree calculation formula as follows: Rust_Degree=w1*D_visible+w2*C_change+w3*D_ultrasound+w4*R_profile+w5*V_depth+w6*ΔI_reflection+w7*ΔA_attenuation+w8*ΔT_GLCM+w9*ΔS_area Among them, w1 to w9 are weight coefficients, which must satisfy Σwi=1; Step 810. Normalize the calculated Rust_Degree value to a range of 0-100%, where 0% indicates no rust and 100% indicates complete rust; The normalization formula is: Normalized_Rust_Degree=(Rust_Degree-Min_Rust_Degree) / (Max_Rust_Degree-Min_Rust_Degree)*100%; Among them, Min_Rust_Degree and Max_Rust_Degree are the theoretical minimum and maximum rust values respectively; Step 811: Output the normalized rust value as the final rust evaluation result of the typical area.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the method for monitoring the corrosion degree of a large-scale high-altitude steel structure after construction as described in any one of claims 1-7.
9. A system for monitoring the corrosion degree of large-scale high-altitude steel structures after construction, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.
Citation Information
Cited By
Remote sensing change detection method and system based on multi-modal space-time difference feature enhancement
CN119580119A
Tire internal defect detection method and system based on tactile perception and ultrasonic fusion
CN119936199A
Continuous online monitoring system for high-altitude paint spraying operation and unmanned aerial vehicle device
CN120539388A
Bridge steel bar corrosion dynamic monitoring system based on fiber bragg grating sensing
CN121521209A
Bridge steel bar corrosion dynamic monitoring system based on fiber grating sensor
CN121521209B
