Steel structure corrosion identification and grade evaluation method based on deep learning

Abstract

The embodiment of the invention provides a steel structure corrosion identification and grade evaluation method based on deep learning, and belongs to the technical field of steel structure corrosion. The method comprises the following steps: acquiring an experimental corrosion image of an indoor accelerated corrosion experiment and a to-be-detected field corrosion image; performing image segmentation training according to the experimental corrosion image to obtain a corrosion region segmentation model; inputting the on-site corrosion image into a corrosion area segmentation model for corrosion area identification to obtain a corrosion mask image; calculating the corrosion area according to the corrosion mask image to obtain a corrosion area proportion; and performing grade evaluation on the corrosion area proportion according to a preset corrosion grade standard to obtain a corrosion grade label. According to the embodiment of the invention, manual intervention can be reduced, the corrosion detection efficiency is improved, the method is suitable for corrosion detection under different environmental conditions, and high-precision automatic identification and evaluation of the corrosion state can be realized.

Description

Technical Field

[0001] This application relates to the field of steel structure corrosion technology, and in particular to a deep learning-based method for steel structure corrosion identification and severity assessment. Background Technology

[0002] With the continuous advancement of modern infrastructure and marine engineering construction, steel structures, due to their high strength, ease of construction, and good reusability, have become a key load-bearing material for engineering systems such as bridges, port facilities, offshore wind power, offshore platforms, and high-rise buildings. However, during long-term service, steel structures are often exposed to complex environments with high humidity, high salinity, and extreme temperature differences. Especially in coastal and marine atmospheric environments, they are highly susceptible to electrochemical corrosion due to the deposition of sea salt particles, erosion by moist aerosols, and periodic wet-dry cycles. This corrosion not only gradually causes the surface protective layer to fail but may also lead to local thinning, crack initiation, and degradation of the strength of connecting components, thereby causing a decrease in load-bearing capacity, node instability, and even overall structural failure, seriously threatening engineering safety and service life.

[0003] Currently, the assessment of steel structure corrosion mainly relies on visual inspection or standardized corrosion grading systems. These standards are mostly based on indicators such as corrosion area, rust morphology, and corrosion depth, and are used for qualitative or semi-quantitative grading. However, in complex service scenarios such as marine environments, steel structure corrosion exhibits multi-stage, multi-form, and non-uniform development characteristics. Manual assessment methods have significant limitations: they are not only labor-intensive and inefficient, but also heavily influenced by the assessor's experience and subjective judgment, making it difficult to guarantee the objectivity and consistency of the assessment results.

[0004] In addition, on-site testing is often limited by marine climate conditions and construction safety requirements, resulting in discontinuous data acquisition and insufficient information accuracy, which further exacerbates the uncertainty of corrosion assessment. Summary of the Invention

[0005] The main objective of this application is to propose a deep learning-based method for steel structure corrosion identification and grade assessment, which aims to reduce manual intervention, improve the efficiency of corrosion detection, adapt to corrosion detection under different environmental conditions, and achieve high-precision automatic identification and assessment of corrosion status.

[0006] To achieve the above objectives, one aspect of this application proposes a deep learning-based method for steel structure corrosion identification and severity assessment, the method comprising the following steps: Acquire experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images of the material to be tested; Based on the experimental corrosion images, an image segmentation training was performed to obtain a corrosion region segmentation model; The on-site corrosion image is input into the corrosion region segmentation model to identify the corrosion region and obtain a corrosion mask image; The corrosion area ratio is obtained by calculating the corrosion area ratio based on the corrosion mask image. The corrosion area ratio is evaluated according to a preset corrosion level standard to obtain a corrosion level label.

[0007] In some embodiments, acquiring experimental corrosion images from indoor accelerated corrosion experiments includes the following steps: Acquire outdoor corrosion images of natural exposure and indoor corrosion images simulating marine atmospheric corrosion conditions; The outdoor corrosion image and the indoor corrosion image are mapped to stages according to the acquisition time interval to obtain image sequence pairs. The image sequence pairs include outdoor corrosion images and indoor corrosion images with consistent corrosion stages, and the corrosion stages include initial stage, development stage and stable stage. Feature extraction is performed on the image sequence pairs to obtain image evolution features, wherein the image evolution features include texture features, color features and morphological features. The texture features are used to characterize the changes in surface roughness and rust morphology, and the color features are used to characterize the degree of rust and differences in oxide types. The gray relational degree is calculated separately for each type of feature in the image evolution features to obtain the relational degree value corresponding to each type of feature. The relational degree value is used to measure the consistency of changes between indoor and outdoor in the time dimension. The overall correlation score is obtained by taking a weighted average of all the correlation scores. The indoor corrosion images whose comprehensive correlation degree meets the preset correlation degree threshold are used as experimental corrosion images.

[0008] In some embodiments, the erosion region segmentation model is an improved BlendMask model, which includes an image input layer, a feature extraction layer, a feature fusion layer, an erosion region recognition layer, and a result output layer. The feature fusion layer includes a channel attention module and a spatial attention module.

[0009] In some embodiments, inputting the on-site corrosion image into the corrosion region segmentation model for corrosion region identification to obtain a corrosion mask image includes the following steps: The on-site corrosion image is input into the image input layer for data conversion to obtain an input image in digital matrix form; The input image is subjected to multi-scale erosion feature extraction through the feature extraction layer to obtain shallow and deep features; The shallow and deep features are fused and attention is enhanced by the feature fusion layer to obtain an output feature map; The corrosion region identification layer spatially locates and divides the corrosion region in the output feature map to obtain the mask outline of the corrosion region. The output layer performs image binarization based on the mask contour to obtain a mask erosion image, wherein the eroded area in the mask erosion image is the first pixel value and the non-eroded area is the second pixel value.

[0010] In some embodiments, the step of fusing and enhancing the shallow and deep features through the feature fusion layer to obtain an output feature map includes the following steps: The shallow features and the deep features are fused to obtain a fused feature map; The channel attention module adaptively adjusts the weights of the fused feature map according to the channel importance to obtain the first enhanced feature map; The spatial attention module adaptively adjusts the weights of the first enhanced feature map based on its spatial distribution characteristics to obtain the output feature map.

[0011] In some embodiments, the step of calculating the area of ​​the etched region based on the etched mask image to obtain the etched area ratio includes the following steps: The total number of pixels in the etched mask image is obtained by counting the number of pixels. The total number of pixels in the eroded area is obtained by counting the number of pixels in the area. The erosion area ratio is obtained by calculating the ratio between the total number of pixels in the image and the total number of pixels in the region.

[0012] In some embodiments, the method further includes the following steps: If the difference between the corrosion area ratio and the grade boundary of the corrosion grade label is less than a preset warning threshold, the corrosion grade label will be reviewed.

[0013] To achieve the above objectives, another aspect of this application proposes a deep learning-based system for steel structure corrosion identification and severity assessment, the system comprising: The first module is used to acquire experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected. The second module is used to perform image segmentation training based on the experimental corrosion image to obtain a corrosion region segmentation model. The third module is used to input the on-site corrosion image into the corrosion region segmentation model to identify the corrosion region and obtain a corrosion mask image. The fourth module is used to calculate the area of ​​the etched region based on the etched mask image to obtain the etched area ratio; The fifth module is used to evaluate the corrosion area ratio according to a preset corrosion level standard to obtain a corrosion level label.

[0014] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0016] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0017] The embodiments of this application include at least the following beneficial effects: This application provides a deep learning-based method for steel structure corrosion identification and grade assessment. This method acquires experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected; it trains image segmentation based on the experimental corrosion images to obtain a corrosion region segmentation model; it inputs the on-site corrosion images into the corrosion region segmentation model for corrosion region identification to obtain a corrosion mask image; it calculates the corrosion region area based on the corrosion mask image to obtain the corrosion area ratio; and it assesses the corrosion area ratio according to a preset corrosion grade standard to obtain a corrosion grade label. The embodiments of this application can reduce manual intervention, improve the efficiency of corrosion detection, adapt to corrosion detection under different environmental conditions, and achieve high-precision automatic identification and assessment of corrosion status. Attached Figure Description

[0018] Figure 1 This is a flowchart of the deep learning-based steel structure corrosion identification and grade assessment method provided in the embodiments of this application; Figure 2 This is a schematic diagram illustrating the effect of image segmentation results provided in the embodiments of this application; Figure 3 This is a schematic diagram illustrating the effect of the grade evaluation results provided in the embodiments of this application; Figure 4 This is a schematic diagram of the steel structure corrosion identification and grading system based on deep learning provided in an embodiment of this application; Figure 5 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of systems and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] Before providing a detailed description of the embodiments of this application, the relevant technologies involved in the embodiments of this application will be described first.

[0022] In recent years, with the rapid development of computer vision and intelligent recognition technologies, image recognition-based corrosion detection methods have gradually emerged. Some studies have attempted to identify and segment corrosion areas on steel structure surfaces using image processing or deep learning algorithms, which has improved detection efficiency to some extent. However, these methods still face problems such as insufficient segmentation accuracy, strong interference from complex backgrounds, and difficulty in directly correlating evaluation results with standard systems. More importantly, there is currently a lack of high-quality image databases with physicochemical index labels that can reflect the laws of marine corrosion, which limits the generalization ability of model training and makes it difficult to achieve effective transfer and application in real engineering scenarios.

[0023] Therefore, there is an urgent need to build an intelligent identification and grading system for steel structure corrosion in marine environments. Based on high-quality corrosion image data, combined with physicochemical performance parameters and a standardized grading system, this system can achieve accurate identification of corrosion areas, quantitative assessment of corrosion degree, and intelligent determination of corrosion level.

[0024] In view of this, this application provides a deep learning-based method for steel structure corrosion identification and severity assessment. This method acquires experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected; it trains an image segmentation model based on the experimental corrosion images to obtain a corrosion region segmentation model; it inputs the on-site corrosion images into the corrosion region segmentation model for corrosion region identification to obtain a corrosion mask image; it calculates the corrosion area ratio based on the corrosion mask image; and it assesses the corrosion area ratio according to a preset corrosion severity standard to obtain a corrosion severity label. This application reduces manual intervention, improves corrosion detection efficiency, adapts to corrosion detection under different environmental conditions, and enables high-precision automatic identification and assessment of corrosion status.

[0025] The deep learning-based steel structure corrosion identification and severity assessment method provided in this application relates to the field of steel structure corrosion technology. This deep learning-based steel structure corrosion identification and severity assessment method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited to these. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the deep learning-based steel structure corrosion identification and severity assessment method, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] Figure 1 This is an optional flowchart of the deep learning-based steel structure corrosion identification and severity assessment method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0028] Step S101: Obtain experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected.

[0029] Step S102: Perform image segmentation training based on the experimental corrosion image to obtain the corrosion region segmentation model.

[0030] Step S103: Input the on-site corrosion image into the corrosion region segmentation model to identify the corrosion region and obtain the corrosion mask image.

[0031] Step S104: Calculate the area of ​​the etched region based on the etched mask image to obtain the etched area ratio.

[0032] Step S105: Evaluate the corrosion area ratio according to the preset corrosion level standard to obtain the corrosion level label.

[0033] In this embodiment, to address a series of issues in steel structure corrosion identification and assessment technology, particularly the challenges in corrosion zone identification accuracy, corrosion level assessment automation, and engineering applicability, this embodiment integrates data-driven approaches with engineering standards. This effectively supports health monitoring, maintenance decision-making, and life prediction of marine engineering structures, promotes the transformation of steel structure operation and maintenance from experience-based to scientific decision-making, and provides reliable assurance for the long-term safety and sustainable service of marine infrastructure.

[0034] Specifically, this embodiment first systematically obtains experimental corrosion images that can characterize corrosion evolution through indoor accelerated corrosion experiments, laying a data foundation for model training.

[0035] Given the long-term and uncontrollable nature of corrosion processes in marine environments, directly relying on outdoor data for model training presents numerous limitations. Firstly, outdoor environmental factors are complex and variable, significantly influenced by climate, humidity, salinity, and lighting conditions, resulting in long data acquisition cycles, poor controllability, and difficulty in achieving consistent recording throughout the entire process. Secondly, on-site images often suffer from uneven lighting, occlusion, and surface contamination, which negatively impact the stability of model feature extraction and training. In contrast, indoor accelerated corrosion experiments can systematically reproduce different stages of corrosion under controlled conditions, achieving complete temporal tracking and high-precision recording of spatial features, thereby constructing a representative and transferable high-quality corrosion dataset.

[0036] In addition, to enhance the representativeness and engineering applicability of the data, this embodiment fully considers the characteristics of the engineering environment in the design of the indoor accelerated corrosion test. The indoor accelerated corrosion test specifically simulates corrosion environment conditions, including but not limited to typical marine corrosion environment conditions, industrial atmospheric corrosion environment conditions, and chemical media corrosion environment conditions, to ensure that the experimental data can reflect the actual service status of the steel structure in the project.

[0037] For example, marine corrosive environments are characterized by alternating cycles of high salt spray concentration, high humidity, and high temperature, which can significantly accelerate the electrochemical corrosion process on the surface of steel structures.

[0038] Industrial atmospheric corrosion environments are characterized by industrial waste gases and suspended particulate matter rich in sulfur dioxide (SO2), hydrogen sulfide (H2S), nitrogen oxides (NOx), etc. In simulation, constant or periodic acidic gases of the above-mentioned gases can be introduced into a closed environment. The morphology and distribution of corrosion products are different from those of marine environments, often showing a more uniform rust layer or specific corrosion product colors.

[0039] Corrosion in chemical media is dominated by specific chemical media and can be simulated by immersion in or exposure to aerosols containing acid, alkali, or salt solutions of specific concentrations. For example, for wastewater treatment plants, low pH and chloride-containing environments can be simulated.

[0040] It should be noted that, since there may be differences between the data from indoor accelerated corrosion experiments and actual external environmental corrosion, which may affect the generalization ability of subsequent models, this embodiment introduces grey system theory to address the issue that indoor accelerated corrosion experiment data cannot be directly transferred. By performing correlation analysis on the experimental corrosion images from indoor accelerated corrosion experiments, experimental corrosion images that can highly represent the real long-term corrosion process in terms of image feature evolution are selected for subsequent model training.

[0041] Furthermore, after obtaining experimental corrosion images with engineering representative pre-migration value, the second stage begins: constructing and optimizing a corrosion region segmentation model for image segmentation. This model aims to achieve high-precision automatic identification of corrosion regions, providing reliable data support for subsequent corrosion area calculation and grade assessment.

[0042] Specifically, experimental corrosion images are pre-labeled, and the parameters of the model are trained using a large number of labeled experimental corrosion images. This enables the model to master the distinguishable features of the corrosion area from the background area in terms of color, texture, and morphology, so as to have the ability to identify and extract the corrosion area from the background.

[0043] Subsequently, the trained corrosion region segmentation model is put into application. In the model application stage, it is necessary to obtain the on-site corrosion image to be detected as the input of the model. This image is obtained by acquiring the steel structure surface in the actual scene through a camera or sensor.

[0044] For example, the on-site corrosion images to be detected can be acquired using a high-resolution industrial camera under natural light conditions on the surface of a steel component.

[0045] The corrosion image to be detected is input into the corrosion region segmentation model for corrosion region identification. The corrosion region segmentation model automatically focuses on the key features of the corrosion region through a complete data processing chain from the original image to accurate segmentation of the corrosion region, separating the corrosion region from the non-corrosion region. Then, it distinguishes the corrosion region from the non-corrosion region by different pixel values, obtaining a corrosion mask image corresponding to the input image space, such as... Figure 2 As shown, an erosion mask image is a binary image in which each pixel has only two possible values. For example, for eroded areas, the pixel value is set to red, and for non-eroded areas, the pixel value is set to black, thus making the shape and extent of the eroded area clearly visible. The erosion mask image can preserve the geometric shape and boundary contour of the original eroded area, realize pixel-level positioning of erosion, and provide a good guarantee for the accuracy of subsequent calculation of erosion area.

[0046] It should be noted that the erosion mask image can also suppress background interference such as uneven lighting and shadows that may exist in the on-site erosion image. These are effectively filtered out and set to black in the erosion mask image, so that subsequent image analysis can focus entirely on the eroded area.

[0047] Furthermore, by counting the number of pixels with different pixel values ​​in the etched mask image, the proportion of the etched area in the etched mask image is calculated, and the image recognition result is converted into a quantifiable value to obtain the etched area proportion.

[0048] Finally, the corrosion area ratio is matched with the standardized preset corrosion level standard to automatically generate the corresponding corrosion level label (such as A0, A1, A2, etc.).

[0049] For example, the main task of the corrosion level assessment mechanism is to automatically classify the corrosion status of steel structures into different levels based on the proportion of corroded areas in the image and in conjunction with standardized corrosion level classification rules. The assessment system is constructed based on the corrosion area extraction results from image segmentation, combined with standardized corrosion area and actual steel structure corrosion status assessment standards.

[0050] This embodiment constructs an automated, preset corrosion level evaluation system based on international standards. By evaluating the corrosion area, the corrosion state is divided into several levels to accurately determine the degree of corrosion. The preset corrosion level standards are shown in the table below: Table 1 Classification of Corrosion Grades

[0051] like Figure 3 As shown, by matching the corrosion area ratio calculated from the corrosion mask image, the corrosion level label is A3, indicating that there is obvious corrosion in a local area on the steel structure in the on-site corrosion image corresponding to the corrosion mask image, and the surface coating may be damaged.

[0052] The aforementioned classification system makes the assessment of corrosion levels more objective and standardized, avoids subjective errors in manual assessment, and enables efficient and accurate quantitative evaluation of the corrosion status of steel structures, thus achieving intelligent assessment of the corrosion status of steel structures.

[0053] In some embodiments, obtaining experimental corrosion images of indoor accelerated corrosion experiments in step S101 may include, but is not limited to, steps S201 to S206.

[0054] Step S201: Obtain outdoor corrosion images of natural outdoor exposure and indoor corrosion images of simulated marine atmospheric corrosion conditions.

[0055] Step S202: Perform stage mapping on the outdoor corrosion images and indoor corrosion images according to the acquisition time interval to obtain image sequence pairs.

[0056] Step S203: Extract features from the image sequence pairs to obtain image evolution features.

[0057] Step S204: Calculate the gray relational degree for each type of feature in the image evolution features separately to obtain the relational degree value corresponding to each type of feature.

[0058] Step S205: Calculate the weighted average of all correlation values ​​to obtain the comprehensive correlation score.

[0059] Step S206: Use the indoor corrosion images whose comprehensive correlation degree meets the preset correlation degree threshold as experimental corrosion images.

[0060] In this embodiment, taking the typical marine atmospheric environment of Sanya as an example, considering the characteristics of the marine environment, such as high salt spray concentration, high humidity and high temperature alternating cycles, the accelerated test is designed to simulate marine atmospheric corrosion conditions by using salt spray accelerated testing, selecting Q235 steel as the standard specimen, and placing it in a neutral salt spray environment.

[0061] During the experiment, high-frequency image acquisition was used to fully cover the entire corrosion process. For example, indoor corrosion images were recorded at intervals of 0h, 2h, 4h, 6h, 8h, and 10h to ensure that a continuous and detailed corrosion evolution sequence was formed in a short period of time.

[0062] The outdoor corrosion images were obtained from steel component exposure experiments under typical marine atmospheric corrosion conditions along the Sanya coast. The specimen material and surface treatment were consistent with the indoor samples, and images of the corroded surface were acquired using a high-resolution industrial camera under natural light conditions.

[0063] Optionally, the acquired indoor and outdoor corrosion images are cropped, normalized, brightness corrected, and enhanced (rotated, flipped, noise added, etc.) to improve data quality and robustness, providing reliable high-precision input data for subsequent model training.

[0064] Understandably, since the timescale of indoor accelerated corrosion experiments is much shorter than that of outdoor natural exposure (indoor data is collected in hours / days, while outdoor data is collected in months), the two cannot be directly synchronized in time. Therefore, this embodiment adopts a stage mapping method to divide the indoor accelerated corrosion process into three stages and align them with the corresponding corrosion stages in the outdoor exposure experiment, resulting in image sequence pairs with consistent corrosion stages, thus achieving equivalent process comparison. The corrosion stages include the initial stage, the development stage, and the stable stage.

[0065] Specifically, the process of stage mapping is not a simple time alignment, but a process alignment based on the morphological similarity of corrosion development. For example, the initial stage is characterized by the appearance of rust spots, and the development stage is characterized by the rapid development of rust. By using the characteristics of different corrosion stages, morphologically similar outdoor corrosion images and indoor corrosion images are used as a set of image sequence pairs, thereby establishing a correlation between indoor corrosion images with outdoor corrosion images at different time scales.

[0066] Next, grey relational analysis was performed on the image sequences. To verify the engineering representativeness of indoor accelerated corrosion data in terms of image feature evolution, this embodiment performed feature-by-feature comparison between indoor corrosion image sequences and outdoor corrosion image sequences from long-term exposure.

[0067] Specifically, to achieve feature layer alignment of indoor and outdoor images, this embodiment uses image texture, color, and morphological features as core indicators to extract image evolution features. Image evolution features include texture features: gray-level co-occurrence matrix indicators (energy, entropy, contrast, correlation), used to reflect changes in surface roughness and rust morphology; color features: RGB and HSV color space parameters, characterizing the degree of rust and differences in oxide types, such as the reddish-brown tone changes brought about by rust formation.

[0068] For each type of feature in the image evolution features, a sequence is formed in chronological order. For example, suppose the indoor accelerated test sequence is X1=[ x 1,1 , x 1,2 ,..., x 1,n The actual environmental corrosion data sequence is X2=[ x 2,1 , x 2,2 ,..., x 2,n ], for each type of feature extracted from the aforementioned image evolution features i Calculate grey relational degree independently γ i This yields a correlation degree value used to measure the consistency of changes between indoor and outdoor environments over time. The formula for calculating the grey correlation degree is as follows: ; In the formula, ρ is the resolution coefficient, which is usually taken as 0.5. i These are feature numbers, representing different image evolution features (such as texture, color, etc.). γ i It is the first i The correlation value of a feature is such that the closer the value is to 1, the more consistent the indoor and outdoor features are in terms of time evolution.

[0069] The comprehensive correlation degree Γ is obtained by taking a weighted average of the correlation values ​​calculated from all features using grey relational analysis. ; In the formula, Γ represents the overall correlation degree, indicating the temporal consistency between indoor and outdoor corrosion images based on all features. For the first i The weight of each feature reflects its contribution to the overall consistency. It can be set according to the stability or sensitivity of the feature, for example, the total weight can be 1.

[0070] For example, the weighting method can be set according to the stability and sensitivity of the feature, such as a weight of 0.6 for texture features and 0.4 for color features.

[0071] Assuming a preset correlation threshold of 0.85, when the overall correlation Γ between indoor and outdoor corrosion images is ≥ 0.85, it can be considered that the indoor corrosion images, after verification by grey relational analysis, demonstrate significant consistency in the development trend and manifestation of rust on the surface of Q235 steel under the selected accelerated corrosion environment with the real engineering environment. Therefore, it can be concluded that the image data collected in the indoor accelerated testing is effective and representative for training deep learning recognition models. Conversely, if the overall correlation does not meet the preset correlation threshold, it indicates that the current accelerated experimental conditions are poorly set and the data is not representative enough. Experimental parameters, such as salt spray concentration and spray cycle, need to be readjusted, and the above analysis and verification need to be repeated until the preset correlation threshold is met before using the indoor corrosion images as experimental corrosion images for subsequent model training.

[0072] This embodiment introduces grey system theory to establish the correlation between indoor and outdoor corrosion data, which solves the problem of insufficient representativeness of experimental images in model training in the past. This ensures the generalization performance of the subsequent corrosion recognition model in real engineering scenarios and provides basic support for the reliability of the entire system.

[0073] In some embodiments, the erosion region segmentation model is an improved BlendMask model. The improved BlendMask model includes an image input layer, a feature extraction layer, a feature fusion layer, an erosion region recognition layer, and a result output layer. The feature fusion layer includes a channel attention module and a spatial attention module. Step S103 may include, but is not limited to, steps S301 to S305.

[0074] Step S301: The on-site corrosion image is input to the image input layer for data conversion to obtain an input image in the form of a digital matrix.

[0075] Step S302: Multi-scale erosion feature extraction is performed on the input image through the feature extraction layer to obtain shallow and deep features.

[0076] Step S303: The shallow and deep features are fused and attention is enhanced through the feature fusion layer to obtain the output feature map.

[0077] Step S304: Spatial localization and region segmentation of the erosion region in the output feature map are performed by the erosion region recognition layer to obtain the mask contour of the erosion region.

[0078] Step S305: The image is binarized according to the mask contour through the result output layer to obtain the mask erosion image.

[0079] In this embodiment, an efficient image segmentation model for steel structure corrosion identification is established based on the BlendMask network structure. This image segmentation model learns from experimental corrosion images and can form a complete data processing link from the original image to accurate segmentation of the corrosion area. To enhance the model's recognition stability and feature representation ability in complex environments, this embodiment introduces the CBAM (Convolutional Block Attention Module) attention mechanism on top of BlendMask, enabling the model to automatically focus on key features of the corrosion area, improving boundary recognition accuracy and detail fidelity.

[0080] Specifically, the improved BlendMask model consists of an image input layer, a feature extraction layer, a feature fusion layer, an erosion region recognition layer, and a result output layer. During model operation, data sequentially passes through five stages: input, extraction, fusion, recognition, and output, with each stage undertaking a specific function.

[0081] The input phase unifies the data format and standardizes features by reading standardized corrosion images. This phase is accomplished through the image input layer, which is primarily responsible for inputting preprocessed corrosion images into the model. These images include experimental corrosion images from salt spray accelerated experiments (training phase) or field-collected corrosion image data (application phase). This phase ensures the consistency of the images in terms of brightness, size, and resolution, and transforms the input images into a digital matrix form that the model can recognize, laying the data foundation for subsequent feature extraction.

[0082] The extraction stage extracts multi-scale erosion features from the image, including but not limited to texture, morphology, and structural features. This stage is accomplished through a feature extraction layer, the core of the model, implemented based on the BlendMask backbone structure. This layer extracts multi-scale features of the eroded regions from the input image, including color variations and texture differences. During this process, the model gradually separates the differential features between the eroded regions and the background regions through multiple convolutions and feature mappings, providing a foundation for subsequent refined recognition.

[0083] The fusion stage integrates multi-scale feature information and enhances the response to key erosion features through an attention mechanism. This stage is accomplished through a feature fusion layer, which integrates erosion feature information from different levels, fusing detailed information from shallow features with overall structural information from deep features. This stage introduces the CBAM (Convolutional Block Attention Module) attention mechanism, which adaptively allocates feature weights across both channel and spatial dimensions, enabling the model to focus on erosion-related region features and suppress interference from complex backgrounds. This maintains stable recognition performance under diverse environmental conditions, resulting in an output feature map.

[0084] The identification stage is responsible for generating the spatial distribution of eroded regions and defining their boundaries. This stage is accomplished by an erosion region identification layer. This layer receives the fused output feature map and, based on the trained model parameters, spatially locates and segments the eroded regions in the image. The model then generates a mask contour for the eroded regions based on the feature distribution, achieving automatic separation between eroded and non-eroded regions. The output of this stage is an intermediate layer result, which can be used for performance optimization and visualization verification during model training.

[0085] The output stage outputs a mask image of the corroded area, providing input for subsequent quantitative analysis and corrosion level assessment. This stage is completed through the result output layer, which converts the recognition results output by the corrosion area recognition layer into a standardized corrosion mask image. This mask image has the same size as the input image and uses binarization, setting pixels in the corroded areas to the first pixel value (e.g., 1) and pixels in the non-corroded areas to the second pixel value (e.g., 0), thus obtaining... Figure 2 The image displayed matches the size of the input field corrosion image and shows a masked corrosion image of the corroded area. Comparison shows that the corroded areas in the masked corrosion image are spatially perfectly aligned with those in the field corrosion image. This masked corrosion image can be directly input into the subsequent corrosion level assessment mechanism to automatically calculate the corrosion area ratio and determine the corrosion level.

[0086] This embodiment, through the aforementioned hierarchical structure and orderly data flow, achieves the entire process from raw image to quantitative identification of corroded areas. This model not only maintains high recognition accuracy under complex lighting and background conditions but also operates stably in highly corrosive environments such as marine salt spray and humidity, providing a reliable technical foundation for monitoring the service status and life assessment of steel structures.

[0087] In some embodiments, step S303 may include, but is not limited to, steps S401 to S403.

[0088] Step S401: The shallow features and deep features are fused to obtain a fused feature map.

[0089] Step S402: The channel attention module adaptively adjusts the weights of the fused feature map according to the channel importance to obtain the first enhanced feature map.

[0090] Step S403: The spatial attention module adaptively adjusts the weights of the first enhanced feature map according to the spatial distribution characteristics to obtain the output feature map.

[0091] In this embodiment, deep features are feature maps rich in semantic information, while shallow features are feature maps with high resolution and rich details. The two are combined through specific interactions (such as multiplication, addition, or splicing operations) to obtain a fused feature map that simultaneously contains detailed information from the shallow layer (such as erosion boundaries) and semantic information from the deep layer (such as "erosion regions").

[0092] Next, the CBAM attention mechanism is used to adaptively assign feature weights to the fused feature map in both channel and spatial dimensions, where the channel attention mechanism and the spatial attention mechanism are sequentially connected.

[0093] The CBAM attention mechanism first analyzes the importance of different channels in the fused feature map through the channel attention module to determine which types of features are more discriminative for erosion recognition. Then, it highlights key channels and suppresses redundant features by adjusting the weights to obtain the first enhanced feature map.

[0094] For example, the channel attention mechanism can assign higher weights to channels that appear reddish-brown and have a rough texture, while assigning lower weights to channels that have a smooth surface and a metallic sheen.

[0095] Subsequently, the spatial attention module receives the first enhanced feature map after channel weighting, further analyzes its spatial distribution characteristics on the image plane, determines the specific location and boundary range of the eroded area in the image, thereby strengthening the model's focusing ability at the spatial level, and adjusts the weights according to the spatial distribution characteristics to strengthen the areas with dense erosion or large feature changes in the first enhanced feature map, thus obtaining the output feature map.

[0096] For example, the spatial attention mechanism can assign higher weights to large areas that are reddish-brown and have a rough texture to enhance areas of dense corrosion, while assigning higher weights to boundary lines where there are drastic changes in features at the junction of corroded and non-corroded areas to define the boundary range.

[0097] Through this "channel-first, space-later" feature optimization path, the corrosion region segmentation model can adaptively focus on corrosion features at different scales and locations, enabling the output feature map to have both strong semantic expression and spatial positioning capabilities, providing high-precision feature support for subsequent corrosion region identification and corrosion level assessment.

[0098] In some embodiments, step S104 may include, but is not limited to, steps S501 to S503.

[0099] Step S501: Count the number of pixels in the etched mask image to obtain the total number of pixels in the image.

[0100] Step S502: Count the number of pixels in the eroded area to obtain the total number of pixels in the area.

[0101] Step S503: Calculate the erosion area ratio based on the ratio of the total number of pixels in the image to the total number of pixels in the region.

[0102] In this embodiment, after image segmentation is completed, the area ratio of the etched region is calculated on the mask etched image to achieve the conversion from visual recognition to quantitative analysis.

[0103] Specifically, the erosion mask image is a two-dimensional digital matrix composed of pixels. Its size has been standardized at output, for example, the height is set to 512 pixels and the width is set to 512 pixels. The total number of pixels in the erosion mask image can be obtained by multiplying the height by the width.

[0104] Since the eroded and non-eroded areas are modeled with different pixel values ​​when generating the erosion mask image, it is only necessary to count the pixels in the erosion mask image whose pixel value is the first pixel value, and traverse each pixel to obtain the total number of pixels in the eroded area.

[0105] The ratio of the total number of pixels in the region to the total number of pixels in the image is calculated to obtain a relative area of ​​the corroded area in the currently acquired on-site corrosion image, thus converting the pixel recognition result into a corrosion area ratio that can be used for quantitative comparison.

[0106] In some embodiments, the deep learning-based steel structure corrosion identification and rating method may also include, but is not limited to, step S601.

[0107] Step S601: When the difference between the corrosion area ratio and the grade boundary of the corrosion grade label is less than the preset warning threshold, the corrosion grade label is reviewed.

[0108] In this embodiment, to further improve the reliability of the corrosion level assessment, the deep learning-based steel structure corrosion identification and level assessment method also introduces an intelligent boundary early warning and verification mechanism to verify the assessment results when the corrosion area ratio is at the critical edge of the corrosion level.

[0109] For example, taking A2 and A3 in Table 1 as examples, the two corrosion level labels share a level boundary value of 0.5. Assuming that the corrosion area ratio obtained after corrosion area recognition and area ratio calculation of a certain input image is 0.49, then during the level evaluation, based on this corrosion area ratio, the corrosion level label of A2 will be assigned first. Then, the corrosion area ratio will be subtracted from the level boundary, and the absolute value of the difference will be 0.01. If the preset warning threshold is 3% of the range of the same corrosion level label, then the preset warning threshold of A2 is (0.5-0.05)*3%=0.0135. By comparison, it can be seen that the difference of 0.01 is less than 0.0135, which means that the condition for reviewing the corrosion level label is met.

[0110] Optionally, the review can be conducted manually or by machine. If machine review is used, another backup model trained with high recall as the optimization objective can be started to reprocess the original image. By comparing the outputs of the two models, if the corrosion level labels are consistent, the original corrosion level label is confirmed; if there are differences, further confirmation can be made through manual review.

[0111] The corrosion level labels that trigger manual review conditions are marked, such as by highlighting them or by displaying pop-up prompts. At the same time, the on-site corrosion image, corrosion mask image, and the assessed corrosion level label are displayed so that users can confirm or correct the corrosion level label based on their professional experience.

[0112] The following is a detailed description and explanation of the solutions in the embodiments of the present invention, using specific application examples: The method provided in this application relates to the interdisciplinary field of civil engineering, marine engineering, and artificial intelligence, and falls under the category of structural health monitoring and intelligent detection technology. This method addresses typical corrosion scenarios such as marine atmospheres, tidal zones, and salt spray environments. It constructs an image database and combines an improved BlendMask instance segmentation model with an attention mechanism to achieve accurate identification and automated level assessment of corrosion areas in steel structures. This effectively solves the problems of complex corrosion morphologies and difficult detection in marine environments, and can be widely applied to engineering facilities such as offshore bridges, port terminals, and offshore platforms, providing efficient and reliable technical support for the safe service and intelligent operation and maintenance of marine infrastructure.

[0113] Specifically, in order to achieve efficient and automated corrosion level assessment, this embodiment designs a complete corrosion assessment system, which includes a data processing and report generation module.

[0114] The data processing module is responsible for preprocessing the input image, including normalization, cropping, and enhancement, to ensure the quality of the image data. This module also converts the image segmentation results into a binary erosion region mask, such as... Figure 2As shown, the area ratio of the corroded region is calculated.

[0115] The report generation module, based on image segmentation results, calculates the area of ​​the corroded region and matches it with a standardized corrosion grading system to automatically generate corresponding corrosion grade labels, such as... Figure 3 As shown in the diagram. This module also includes intelligent evaluation logic; when the corrosion area approaches the grade boundary, the system will automatically prompt that manual verification may be necessary to ensure the accuracy of the evaluation.

[0116] The workflow of the corrosion level assessment system is roughly as follows: First, image data of the steel structure surface is acquired through a camera or sensor; then, the input image is processed using a corrosion area segmentation model trained with indoor accelerated experimental data to extract the corrosion area; the area ratio of the corrosion area is calculated based on the segmentation results; the corrosion level is automatically assessed based on the corrosion area and standardized level classification rules; the system provides feedback on the results, automatically detects possible errors, and prompts for manual review.

[0117] The corrosion assessment system provided in this embodiment not only incorporates the BlendMask model from the CBAM module, enabling accurate identification of corrosion areas in steel structures even in complex backgrounds and with blurred corrosion edges, but also maintains a stable IoU (Intersection over Union) of over 85% in various testing environments, effectively identifying corrosion areas under different conditions and ensuring high-precision corrosion detection. Furthermore, compared to manual visual inspection methods, the automated image segmentation system achieves a processing time of less than one second per image, significantly improving corrosion detection efficiency. The system can automatically complete the entire process from image acquisition to corrosion level assessment, reducing manual intervention and improving operational efficiency, demonstrating strong automation and high efficiency. More importantly, this technology is widely applicable to corrosion detection of various steel structures, especially showing promising application prospects in industrial structures such as bridges, factories, and pipelines. The system supports automatic corrosion level assessment according to international standards (such as ASTM and ISO) and can flexibly adapt to corrosion detection under different environmental conditions, exhibiting strong adaptability and scalability. In addition, through clear binary mask images and corrosion area distribution maps, the severity and location of corrosion are intuitively presented. Users can quickly understand the test results through a simple interface. The operation is simple and suitable for rapid application in engineering sites, with good operability and visualization.

[0118] In summary, the embodiments of this application specifically solve the following technical problems: 1. Manual assessment is inefficient and highly subjective: Current steel structure corrosion assessment relies on manual visual inspection, which is inefficient and easily influenced by the assessor's experience, leading to inconsistent and inaccurate results. This application solves the problems of low efficiency and high subjectivity in manual assessment by using automated image segmentation technology, achieving accurate automatic identification and assessment of corrosion status.

[0119] 2. The issue of correlation between indoor and outdoor corrosion data: There may be discrepancies between indoor corrosion experimental data and actual external environmental corrosion, affecting the generalization ability of the model. This application introduces grey system theory to establish a correlation model between indoor and outdoor corrosion data, solving the problem that indoor accelerated corrosion experimental data cannot be directly transferred to the actual environment, thereby improving the applicability of the model in practical engineering.

[0120] 3. Insufficient Image Processing Accuracy: In related technologies, image processing methods cannot accurately identify eroded areas in complex backgrounds, especially when the eroded boundaries are blurred or the details are small, resulting in low segmentation accuracy. This application improves the BlendMask model and incorporates an attention mechanism to enhance the model's accuracy in identifying eroded area boundaries and details, thus solving the problem of insufficient accuracy in traditional methods.

[0121] 4. Lack of standardized corrosion level assessment methods: The assessment of corrosion levels in mother corrosive organisms lacks a unified standardized method, relying on manual judgment or imprecise scoring standards, making it difficult to promote and apply in practical engineering. This application solves the problem of inconsistent assessment standards by aligning image segmentation results with international standards and constructing an automated corrosion area ratio-based level assessment system.

[0122] Reference Figure 4 This application also provides a deep learning-based steel structure corrosion identification and severity assessment system, which can implement the above-mentioned method. The system includes: The first module is used to acquire experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected.

[0123] The second module is used to train image segmentation based on experimental corrosion images to obtain a corrosion region segmentation model.

[0124] The third module is used to input the on-site corrosion image into the corrosion region segmentation model for corrosion region identification, and obtain the corrosion mask image.

[0125] The fourth module is used to calculate the area of ​​the etched region based on the etched mask image, and obtain the etched area ratio.

[0126] The fifth module is used to evaluate the corrosion area ratio according to the preset corrosion level standard and obtain the corrosion level label.

[0127] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0128] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0129] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0130] Reference Figure 5 , Figure 5 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0131] The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the methods described in the embodiments of this application.

[0132] The input / output interface 903 is used to implement information input and output.

[0133] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0134] Bus 905 transmits information between various components of the device, such as processor 901, memory 902, input / output interface 903, and communication interface 904.

[0135] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0136] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0137] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0138] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0139] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0140] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0141] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0142] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0143] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0145] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0146] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0147] In the embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0148] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A deep learning-based method for steel structure corrosion identification and severity assessment, characterized in that, The method includes the following steps: Acquire experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images of the material to be tested; Based on the experimental corrosion images, an image segmentation training was performed to obtain a corrosion region segmentation model; The on-site corrosion image is input into the corrosion region segmentation model to identify the corrosion region and obtain a corrosion mask image; The corrosion area area is calculated based on the corrosion mask image to obtain the corrosion area ratio. The corrosion area ratio is evaluated according to a preset corrosion level standard to obtain a corrosion level label.

2. The method according to claim 1, characterized in that, The process of obtaining experimental corrosion images from indoor accelerated corrosion experiments includes the following steps: Acquire outdoor corrosion images of natural exposure and indoor corrosion images simulating marine atmospheric corrosion conditions; The outdoor corrosion image and the indoor corrosion image are mapped to stages according to the acquisition time interval to obtain image sequence pairs. The image sequence pairs include outdoor corrosion images and indoor corrosion images with consistent corrosion stages, and the corrosion stages include initial stage, development stage and stable stage. Feature extraction is performed on the image sequence pairs to obtain image evolution features, wherein the image evolution features include texture features, color features and morphological features. The texture features are used to characterize the changes in surface roughness and rust morphology, and the color features are used to characterize the degree of rust and differences in oxide types. The gray relational degree is calculated separately for each type of feature in the image evolution features to obtain the relational degree value corresponding to each type of feature. The relational degree value is used to measure the consistency of changes between indoor and outdoor in the time dimension. The overall correlation score is obtained by taking a weighted average of all the correlation scores. The indoor corrosion images whose comprehensive correlation degree meets the preset correlation degree threshold are used as experimental corrosion images.

3. The method according to claim 1, characterized in that, The erosion region segmentation model is an improved BlendMask model, which includes an image input layer, a feature extraction layer, a feature fusion layer, an erosion region recognition layer, and a result output layer. The feature fusion layer includes a channel attention module and a spatial attention module.

4. The method according to claim 3, characterized in that, The process of inputting the on-site corrosion image into the corrosion region segmentation model for corrosion region identification to obtain a corrosion mask image includes the following steps: The on-site corrosion image is input into the image input layer for data conversion to obtain an input image in digital matrix form; The input image is subjected to multi-scale erosion feature extraction through the feature extraction layer to obtain shallow and deep features; The shallow and deep features are fused and attention is enhanced by the feature fusion layer to obtain an output feature map; The corrosion region identification layer spatially locates and divides the corrosion region in the output feature map to obtain the mask outline of the corrosion region. The output layer performs image binarization based on the mask contour to obtain a mask erosion image, wherein the eroded area in the mask erosion image is the first pixel value and the non-eroded area is the second pixel value.

5. The method according to claim 4, characterized in that, The step of fusing and enhancing the shallow and deep features through the feature fusion layer to obtain an output feature map includes the following steps: The shallow features and the deep features are fused to obtain a fused feature map; The channel attention module adaptively adjusts the weights of the fused feature map according to the channel importance to obtain the first enhanced feature map; The spatial attention module adaptively adjusts the weights of the first enhanced feature map based on its spatial distribution characteristics to obtain the output feature map.

6. The method according to claim 1, characterized in that, The step of calculating the area of ​​the etched region based on the etched mask image to obtain the etched area ratio includes the following steps: The total number of pixels in the etched mask image is obtained by counting the number of pixels. The total number of pixels in the eroded area is obtained by counting the number of pixels in the area. The erosion area ratio is obtained by calculating the ratio between the total number of pixels in the image and the total number of pixels in the region.

7. The method according to claim 1, characterized in that, The method further includes the following steps: When the difference between the corrosion area ratio and the grade boundary of the corrosion grade label is less than a preset warning threshold, the corrosion grade label is reviewed.

8. A deep learning-based system for identifying and assessing corrosion levels in steel structures, characterized in that, The system includes: The first module is used to acquire experimental corrosion images from indoor accelerated corrosion experiments and on-site corrosion images to be detected. The second module is used to perform image segmentation training based on the experimental corrosion image to obtain a corrosion region segmentation model. The third module is used to input the on-site corrosion image into the corrosion region segmentation model to identify the corrosion region and obtain a corrosion mask image. The fourth module is used to calculate the area of ​​the etched region based on the etched mask image to obtain the etched area ratio; The fifth module is used to evaluate the corrosion area ratio according to a preset corrosion level standard to obtain a corrosion level label.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images