Steel structure local corrosion intelligent identification and performance evaluation method based on deep learning

Abstract

The invention discloses a steel structural member local corrosion intelligent identification and evaluation method based on deep learning, and the method comprises the steps: obtaining a steel structure surface image through an imaging device, carrying out the feature extraction and corrosion region identification of an input image through a YOLOv12 semantic segmentation model which introduces a CBAM self-attention mechanism, and generating a pixel-level corrosion mask; the actual physical area of the corrosion area is automatically calculated based on an area calculation algorithm of conversion of mask integration and ground sampling distance; the residual bearing performance of the corroded component is evaluated in combination with the corrosion area ratio and the morphological characteristic prediction model; and finally realizing visual display and report output of detection results, area data and performance indexes through a graphical interface. According to the invention, full-process automatic processing of local corrosion of the steel structural member from detection, segmentation, quantitative measurement and calculation to member performance evaluation is realized, and closed-loop intelligent evaluation from visual detection to mechanical evaluation is realized.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring and computer vision technology, and in particular relates to a method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning. Background Technology

[0002] Steel structures are widely used in modern infrastructure such as bridges, tunnels, offshore platforms, factories, and transmission towers. However, during service, steel surfaces are highly susceptible to corrosion due to factors such as air humidity, chloride ions, acidic substances, and electrochemical reactions. Corrosion leads to cross-sectional weakening, loosening of connections, stress concentration, and fatigue crack initiation, significantly reducing the overall stiffness and load-bearing capacity of the structure. Currently, corrosion detection in engineering mainly relies on manual inspection or manual interpretation of monocular images. This method is inefficient, prone to errors, and lacks quantitative indicators. In recent years, with the rapid development of deep learning technology, steel structure corrosion identification methods based on convolutional neural networks (CNNs) and segmentation networks (such as U-Net, Mask R-CNN, and DeepLabV3+) have been gradually applied to the field of structural health monitoring. However, these methods still have many shortcomings: they are less sensitive to small-area corrosion, especially in areas of light rust or fine cracking, where they are prone to missed detection; at the same time, different shooting distances and camera parameters lead to inconsistent pixel scales, and the lack of a unified scale calibration mechanism makes it difficult to achieve accurate corrosion area conversion; furthermore, existing methods mostly remain at the image segmentation level, only outputting corrosion mask images, failing to effectively connect with the structural performance evaluation process, and making it difficult to directly reflect the safety status and load-bearing capacity degradation of components. Therefore, there is an urgent need to develop an intelligent system that integrates corrosion detection, area quantification, and performance evaluation to achieve automated and closed-loop processing from image-level feature recognition to component performance quantification. Summary of the Invention

[0003] Purpose of the invention: The present invention aims to provide an intelligent method and system that can automatically complete the detection of corrosion areas on the surface of steel structures, physical conversion of corrosion area, and assessment of remaining bearing capacity, overcoming the problems of low detection accuracy, isolated data, and discontinuous assessment in existing methods, and realizing the coupling and automation of visual inspection and engineering performance assessment.

[0004] Technical solution: The present invention provides a method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning, comprising the following steps:

[0005] (1) Collect corrosion image data of steel structure surface, expand it with existing public datasets, and perform image annotation, data division and preprocessing;

[0006] (2) The YOLOv12 semantic segmentation model with CBAM self-attention mechanism was used to train and validate the corrosion data to obtain a corrosion detection model for corrosion region identification.

[0007] (3) Generate pixel-level corrosion masks based on the trained corrosion detection model;

[0008] (4) Binarize the image based on the pixel-level erosion mask, and sum the values ​​of all pixels in the binary image to obtain the total number of pixels in the erosion area; calculate the actual erosion area of ​​the erosion area automatically based on the area calculation algorithm that converts the total number of pixels in the erosion mask to the ground sampling distance.

[0009] (5) Construct a quantitative relationship model between the corrosion area ratio and the bearing capacity reduction coefficient to evaluate the remaining bearing capacity of the component after corrosion;

[0010] (6) Visualize and output reports of test results, area data and performance indicators through a graphical interface.

[0011] Furthermore, the CBAM self-attention mechanism achieves feature weighting through joint modeling of channel attention and spatial attention, with channel attention weights... The calculation method is as follows:

[0012] ;

[0013] in, For the input feature map, For the Sigmoid function, It is a two-layer fully connected network;

[0014] Spatial attention weights The calculation is as follows:

[0015] ;

[0016] in, This represents a 7×7 convolution operation. [This refers to channel concatenation; the final output is:]

[0017] ;

[0018] in, This represents an element-wise multiplication operation; this mechanism enables adaptive weighting and spatial enhancement of features, improving the response intensity of eroded edges and small regions.

[0019] Furthermore, the actual corrosion area Determined by the following formula:

[0020] ;

[0021] in, This represents the total number of pixels in the etch mask. The ground sampling distance is determined according to the following formula:

[0022] ;

[0023] in, For the image sensor width, For shooting distance, For the lens focal length, The width is the image width in pixels.

[0024] Furthermore, the implementation process of step (5) is as follows:

[0025] Corrosion area ratio The calculation formula is:

[0026] ;

[0027] in, This represents the actual corrosion area of ​​the corroded region. This refers to the effective inspected area of ​​the component;

[0028] The bearing capacity reduction factor is calculated using a quantitative relationship model established based on experimental data between the corrosion area ratio and the bearing capacity reduction factor. Remaining bearing capacity Determined by the following formula:

[0029] ;

[0030] in, This represents the theoretical load-bearing capacity of the component when it is not corroded.

[0031] Furthermore, the bearing capacity reduction factor Ratio of corrosion area The relationship is determined by the user based on the component type, corrosion morphology, or fitting accuracy, including:

[0032] Linear model: It is suitable for situations where corrosion is relatively mild and load-bearing capacity degrades approximately linearly;

[0033] Power function model: It is applicable to the nonlinear decay law under moderate to severe corrosion.

[0034] Exponential model: It is suitable for situations where the load-bearing capacity decreases rapidly with the corrosion area;

[0035] Wherein, k and m are fitting coefficients, obtained by fitting the results of corrosion component load-bearing tests or finite element analysis using the least squares method. The parameters corresponding to different component types and corrosion morphologies can be calibrated and updated based on experimental data. The value of k is 0.015 to 0.025; the value of m is 1.0 to 1.3.

[0036] Furthermore, the aforementioned The value range is divided into four levels to characterize the degree of component performance degradation and safety status, specifically:

[0037] Level I: Safe condition. ≥0.9 indicates that the component is in good condition or slightly corroded, and has no significant impact on the overall load-bearing capacity;

[0038] Grade II: Slight corrosion, 0.8≤ <0.9 indicates that the component has a small amount of surface corrosion and a slight decrease in local load-bearing capacity, but still meets the design requirements;

[0039] Level III: Moderate corrosion, 0.6≤ If the value is less than 0.8, and obvious corrosion areas appear in localized parts of the component, the load-bearing capacity is significantly reduced, and it should be included in the key monitoring scope.

[0040] Level IV: Severe corrosion. If the corrosion area of ​​the component is large or the cross-section is significantly weakened, the remaining load-bearing capacity will decrease significantly, and reinforcement or replacement measures should be taken.

[0041] This invention also provides a deep learning-based intelligent identification and performance evaluation system for localized corrosion of steel structures, implemented using a PyQt5 architecture. The system includes: a model loading module, an image input module, a corrosion detection module, a scale calibration module, an area calculation module, a performance evaluation module, and a result export module. The model loading module loads model weight files. The image input module receives and preprocesses images of the steel structure to be detected. The corrosion detection module uses a YOLOv12 network with an embedded CBAM attention mechanism to improve the recognition accuracy of small-scale corrosion areas and achieve automatic segmentation of corrosion areas. The scale calibration module supports both automatic and manual calibration methods and is used to calculate the ground sampling distance based on image capture parameters. The area calculation module performs physical conversion of the corrosion area based on mask integral and scale calibration. The performance evaluation module calculates the remaining load-bearing capacity of the component based on a corrosion area ratio and load-bearing capacity reduction coefficient model. The result export module generates a comprehensive report containing a corrosion distribution map, an area statistics table, and the calculated remaining load-bearing capacity, and exports it in CSV or PDF format.

[0042] Furthermore, the workflow of the system is as follows:

[0043] The user first loads the trained model weight file and selects the image or folder to be detected. The system then automatically performs model inference, generates a segmentation mask for the corroded area, and calculates the corresponding corrosion area and area ratio. Subsequently, the system calls the performance evaluation model to calculate the remaining load-bearing capacity of the component based on the degree of corrosion and displays the detection results in real time on the interface. Finally, the system automatically generates a detection report, which includes a corrosion distribution map, an area statistics table, the calculation results of the remaining load-bearing capacity, and a comprehensive safety level recommendation.

[0044] Beneficial Effects: Compared with existing technologies, the beneficial effects of this invention are as follows: This invention achieves integrated steel structure corrosion detection, area calculation, and performance evaluation, establishing a complete closed-loop process from image input to safety level output, overcoming the problem of separation between detection results and mechanical performance evaluation in traditional methods; secondly, by introducing the CBAM attention mechanism, the model's ability to identify fine corrosion areas and complex backgrounds is significantly improved, effectively enhancing detection accuracy and stability; this invention proposes a quantitative relationship model based on corrosion area and bearing capacity reduction coefficient, replacing traditional thickness measurement parameters with area indicators, realizing quantitative evaluation of bearing performance without ultrasonic testing, avoiding the complexity and error propagation problems of manual thickness measurement; furthermore, the performance evaluation module can automatically match parameters according to component type, material properties, and service environment, achieving multi-scenario adaptation and self-correction capabilities, and possessing good engineering scalability; finally, the system of this invention adopts the PyQt5 architecture, with an intuitive interface and simple operation, possessing result visualization and automatic report export functions, enabling rapid deployment and intelligent evaluation in various engineering scenarios such as bridges, tunnels, factories, and marine steel structures, providing an efficient, reliable, and scalable solution for the health monitoring and safety management of steel structures. Attached Figure Description

[0045] Figure 1 The flowchart shows a deep learning-based intelligent identification method for localized corrosion in steel structure components.

[0046] Figure 2 Improve the visualization of corrosion sample detection results for the YOLOv12 model;

[0047] Figure 3 Schematic diagram of the performance evaluation model of this invention;

[0048] Figure 4 This is a schematic diagram showing the output of corrosion segmentation and performance evaluation results;

[0049] Figure 5 This is a functional structure diagram of the GUI interface of the corrosion detection and area calculation system. Detailed Implementation

[0050] The present invention will now be described in further detail with reference to the accompanying drawings:

[0051] like Figure 1 As shown, this invention proposes a deep learning-based intelligent identification and performance evaluation method for localized corrosion in steel structures, specifically including the following steps:

[0052] Step 1: Collect corrosion image data of steel structure surface, expand it with existing public datasets, and perform image annotation, data segmentation and preprocessing.

[0053] Depending on the size of the object being inspected and the environmental conditions, select appropriate imaging equipment (such as industrial cameras, drone cameras, or handheld terminals) to acquire images of the component surface under natural or supplemental lighting conditions. During shooting, ensure the image resolution can reflect the details of corrosion marks, and maintain an orthogonal shooting angle as much as possible to reduce distortion. Before inputting the acquired images into the model, perform preprocessing operations such as size normalization, brightness balancing, and noise suppression to improve the stability of the inspection.

[0054] Step 2: Use the YOLOv12 semantic segmentation model with CBAM self-attention mechanism to train and validate the corrosion data to obtain a corrosion detection model for corrosion region identification.

[0055] This invention introduces an improved model with a CBAM attention mechanism into the backbone of YOLOv12 for training. The introduced CBAM attention mechanism comprises two parts: channel attention and spatial attention. In channel attention, the input feature X is subjected to average pooling and max pooling respectively, and then input into two MLP layers to calculate the channel weights.

[0056]

[0057] in, For the input feature map, For the Sigmoid function, It is a two-layer fully connected network, and then feature recalibration is performed.

[0058] Then spatial attention is calculated:

[0059]

[0060] in, This represents a 7×7 convolution operation. [This refers to channel splicing; the final fused feature is obtained as follows:]

[0061]

[0062] in, This represents an element-wise multiplication operation. This mechanism enables adaptive weighting and spatial augmentation of features, improving the response intensity of eroded edges and small regions.

[0063] During the data preparation and model training phase, the image data used included on-site measured photos, drone inspection images, and images of artificially accelerated corrosion specimens. To improve the model's generalization ability, the samples were enhanced with brightness, saturation, and rotation, and then divided into training, validation, and test sets in a 7:2:1 ratio.

[0064] After the user selects the appropriate model weights on the interface, the system automatically loads the network and performs inference, segmenting the eroded regions in the input image at the pixel level. For example... Figure 2 As shown, the detection results are displayed overlaid in the form of a blue mask, allowing users to visually observe the spatial distribution of corrosion.

[0065] Step 3: Generate a pixel-level corrosion mask based on the trained corrosion detection model and calculate the corrosion area.

[0066] The erosion mask is binarized, and the sum of all pixel values ​​in the binary image is used to obtain the total number of pixels in the eroded area.

[0067] The pixel area is converted to physical area. The spatial resolution of the image is calculated by inputting parameters such as shooting distance, focal length, and sensor size, or automatically calibrated using a reference object of known size in the image. A pixel-to-physical-domain conversion method based on Ground Sampling Distance (GSD) is used for area calculation. The formula for calculating the corrosion area is:

[0068]

[0069] in, The GSD (Ground Sampling Distance) is calculated using camera parameters to represent the total number of pixels in the erosion mask.

[0070]

[0071] in, For the image sensor width, For shooting distance, For the lens focal length, The image width is in pixels. When there are multiple areas of corrosion on a component, the area of ​​each area is calculated individually and then summed to obtain the total corrosion area. The proportion of the corrosion area to the component's surface area is then determined. This proportion is used to characterize the degree of corrosion.

[0072] Step 4: Based on the experimental results, construct a quantitative relationship model between the corrosion area ratio and the bearing capacity reduction coefficient to evaluate the remaining bearing capacity of the component after corrosion.

[0073] like Figure 3As shown, using the corrosion area ratio as input, a pre-established quantitative relationship model between the corrosion area ratio and the bearing capacity reduction coefficient is called to calculate the bearing capacity reduction coefficient and thereby deduce the remaining bearing capacity.

[0074] The corrosion area ratio is defined as:

[0075]

[0076] in, This represents the actual physical area of ​​the corroded region. This represents the effective inspected area of ​​the component, used to quantify the corrosion percentage.

[0077] The bearing capacity reduction factor is calculated using a quantitative relationship model established based on experimental data between the corrosion area ratio and the bearing capacity reduction factor. Remaining bearing capacity Determined by the following formula:

[0078]

[0079] in, This represents the theoretical load-bearing capacity of the component when it is not corroded.

[0080] Quantitative relationship models can be linear, power function, or exponential in form, and are obtained by fitting experimental data or numerical simulation results based on component type and corrosion characteristics. The bearing capacity reduction factor is then considered. Ratio of corrosion area The relationship forms can be summarized into the following three types. Users can select the appropriate model according to the component type, corrosion morphology, or fitting accuracy, namely: ① Linear model: ① Suitable for cases with mild corrosion and approximately linear degradation of load-bearing capacity; ② Power function model: ③ Exponential model: Applicable to nonlinear decay laws under moderate to severe corrosion; This model is suitable for situations where the load-bearing capacity rapidly decreases with the corrosion area. Here, k and m are fitting coefficients, which can be obtained by fitting the results of load-bearing tests or finite element analysis of corroded components using the least squares method. Parameters corresponding to different component types and corrosion morphologies can be calibrated and updated based on experimental data. The model is written into the system database for online evaluation. When measured data is unavailable, the system can perform a preliminary evaluation using preset parameter ranges (k=0.015~0.025, m=1.0~1.3).

[0081] The system automatically retrieves the corresponding model from the database and allows users to update or calibrate the model parameters based on the detection data. When the user inputs the corrosion area ratio, the system will automatically output the reduction factor, remaining bearing capacity, and corresponding safety level recommendations.

[0082] Step 5: Visualize and output reports of test results, area data, and performance indicators through a graphical interface.

[0083] To enable rapid assessment of the condition of corroded components, the system incorporates safety level classification logic within its performance evaluation module. The system classifies component safety levels based on the range of load-bearing capacity reduction coefficients. This invention uses the load-bearing capacity reduction coefficient... The value range is divided into four levels to characterize the degree of component performance degradation and safety status:

[0084] Level I: Safe Status (≥ 0.9), the component is in good condition or slightly corroded, and has no significant impact on the overall load-bearing capacity.

[0085] Level II: Slight corrosion (0.8 ≤ (< 0.9) The component showed a small amount of surface corrosion, and the local load-bearing capacity decreased slightly, but it still met the design requirements.

[0086] Level III: Moderate corrosion (0.6 ≤ If the corrosion rate is less than 0.8%, and there are obvious corrosion areas in some parts of the component, the load-bearing capacity is significantly reduced, and it should be included in the key monitoring scope.

[0087] Level IV: Severe corrosion ( If the corrosion area of ​​the component is less than 0.6, or if there is significant cross-sectional weakening and a significant decrease in the remaining load-bearing capacity, reinforcement or replacement measures should be taken.

[0088] Finally, the system outputs the following on the interface: Figure 4 The system displays corrosion distribution maps, corrosion area statistics, remaining bearing capacity results, and safety level descriptions, and can generate report files for engineers to archive and manage.

[0089] This invention also proposes a deep learning-based intelligent identification and performance evaluation system for localized corrosion of steel structures, comprising: a model loading module, an image input module, a corrosion detection module, a scale calibration module, an area calculation module, a performance evaluation module, and a result export module. The model loading module loads the model weight file; the image input module receives and preprocesses the image of the steel structure to be detected; the corrosion detection module uses a YOLOv12 network with an embedded CBAM attention mechanism to improve the recognition accuracy of small-scale corrosion areas and achieve automatic segmentation of corrosion areas; the scale calibration module supports both automatic and manual calibration methods and is used to calculate the ground sampling distance based on image capture parameters; the area calculation module performs physical conversion of the corrosion area based on mask integral and scale calibration; the performance evaluation module calculates the remaining load-bearing capacity of the component based on a corrosion area ratio and load-bearing capacity reduction coefficient model; and the result export module generates a comprehensive report containing a corrosion distribution map, an area statistics table, and the calculated remaining load-bearing capacity, and exports it in CSV or PDF format.

[0090] The system's workflow is as follows: First, the user loads the trained model weight file and selects the image or folder to be detected. The system then automatically performs model inference, generates a segmentation mask for the corroded area, and calculates the corresponding corrosion area and area ratio. Subsequently, the system calls the performance evaluation model to calculate the remaining load-bearing capacity of the component based on the degree of corrosion and displays the detection results in real time on the interface. Finally, the system automatically generates a detection report, which includes a corrosion distribution map, an area statistics table, the calculation results of the remaining load-bearing capacity, and a comprehensive safety level recommendation.

[0091] like Figure 5 As shown, this invention preferably uses the PyQt5 framework to implement the graphical user interface and deploys a trained corrosion detection and area calculation model using a deep learning inference framework. Subsequently, a quantitative relationship model between corrosion area ratio and bearing capacity reduction coefficient is integrated into the system. The various functional modules are sequentially connected via data interfaces to automate image input, detection inference, area conversion, performance evaluation, and result display. The system supports single-image analysis and batch file processing, adapting to the needs of multiple scenarios in laboratory research and engineering sites.

[0092] Taking the exposed steel plate component of a steel bridge beam as an example, during on-site inspection, the exposed steel plate of the bridge beam was selected as the inspection object. A high-resolution camera with a resolution of approximately 5000×3500 pixels was used to take pictures at a distance of about two meters. To ensure image quality, an orthophoto angle was used during shooting, and strong light or shadow interference was avoided. After shooting, the user selects the "Import Image" function in the system interface to input the acquired corrosion image into the system database. The system automatically completes format recognition and path registration, and displays the preview information of the image to be inspected.

[0093] Users select the pre-trained deep learning corrosion detection model (YOLOv12-CBAM) on the interface, and the system automatically loads the model and enters detection mode. During detection, the model performs feature extraction and pixel-level segmentation on the input image, identifying multiple corrosion regions in the image. The detection results are displayed as a pseudo-color mask overlaid on the original image, and the system interface simultaneously displays the number of detected corrosion regions, confidence information, and visualization results of the segmentation mask. Users can adjust the display transparency or view individual regions as needed.

[0094] The system automatically performs pixel statistics on detected corrosion masks and calculates the number of corroded pixels. To convert pixels to physical scale, users can input parameters such as shooting distance, focal length, and sensor width into the interface. The system then performs spatial calibration and calculates the spatial resolution of the image. After calibration, the system automatically calculates the actual area of ​​each corrosion region. If multiple corrosion regions exist in the image, the system will calculate the physical area of ​​each region sequentially and sum them to obtain the total corrosion area. Simultaneously, the system calculates the corrosion area percentage based on the inspected surface area of ​​the component and automatically generates a corrosion area statistics table, including information such as region number, area value, and total corrosion area ratio.

[0095] The system automatically transmits the obtained corrosion area ratio to the performance evaluation module. This module calls upon the established quantitative relationship model between "corrosion area ratio and bearing capacity reduction factor" in the database to calculate the bearing capacity reduction factor of the component and further estimate the remaining bearing capacity. Model parameters can be selected or adjusted in the interface according to different component types and corrosion characteristics. After the evaluation is completed, the system generates the bearing capacity calculation results and displays them on the interface in numerical and graphical form.

[0096] The system assesses the safety level of components based on the calculated bearing capacity reduction factor. The classification references national and industry standards, categorizing components into four levels: safe, slightly corroded, moderately corroded, and severely corroded. The system uses different colors to indicate the level range on the interface and outputs corresponding text descriptions. Users can simultaneously view the original image, corrosion segmentation results, area statistics table, remaining bearing capacity results, and safety level prompts through the interface, achieving full visualization of the process from visual inspection to performance evaluation.

[0097] After the assessment is completed, users can select the "Export Results" function. The system automatically generates an inspection report file, which includes: the original image, corrosion mask image, corrosion area statistics, performance evaluation calculation table, and safety level conclusions. The report supports export in PDF or CSV format, facilitating archiving, comparison, or periodic tracking by bridge maintenance units. The system also stores the inspection results in a database for subsequent retrieval and model optimization.

[0098] It should be understood that the embodiments and descriptions above are only the principles, main features and advantages of the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope of the invention, and all such changes and modifications fall within the protection scope of the present invention.

Claims

1. A method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning, characterized in that, Includes the following steps: (1) Collect corrosion image data of steel structure surface, expand it with existing public datasets, and perform image annotation, data division and preprocessing; (2) The YOLOv12 semantic segmentation model with CBAM self-attention mechanism was used to train and validate the corrosion data to obtain a corrosion detection model for corrosion region identification. (3) Generate pixel-level corrosion masks based on the trained corrosion detection model; (4) Binarize the image based on the pixel-level erosion mask, and sum the values ​​of all pixels in the binary image to obtain the total number of pixels in the erosion area; calculate the actual erosion area of ​​the erosion area automatically based on the area calculation algorithm that converts the total number of pixels in the erosion mask to the ground sampling distance. (5) Construct a quantitative relationship model between the corrosion area ratio and the bearing capacity reduction coefficient to evaluate the remaining bearing capacity of the component after corrosion; (6) Visualize and output reports of test results, area data and performance indicators through a graphical interface.

2. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 1, characterized in that, The CBAM self-attention mechanism achieves feature weighting through joint modeling of channel attention and spatial attention, with channel attention weights... The calculation method is as follows: ; in, For the input feature map, For the Sigmoid function, It is a two-layer fully connected network; Spatial attention weights The calculation is as follows: ； in, This represents a 7×7 convolution operation. [] represents channel concatenation; the final output is: ； in, This represents an element-wise multiplication operation; this mechanism enables adaptive weighting and spatial enhancement of features, improving the response intensity of eroded edges and small regions.

3. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 1, characterized in that, The actual corrosion area Determined by the following formula: ； in, This represents the total number of pixels in the etch mask. The ground sampling distance is determined according to the following formula: ； in, For the image sensor width, For shooting distance, For the lens focal length, The width is the image width in pixels.

4. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 1, characterized in that, The implementation process of step (5) is as follows: Corrosion area ratio The calculation formula is: ； in, This represents the actual corrosion area of ​​the corroded region. This refers to the effective inspected area of ​​the component; The bearing capacity reduction factor is calculated using a quantitative relationship model established based on experimental data between the corrosion area ratio and the bearing capacity reduction factor. Remaining bearing capacity Determined by the following formula: ； in, This represents the theoretical load-bearing capacity of the component when it is not corroded.

5. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 4, characterized in that, The bearing capacity reduction factor Ratio of corrosion area The relationship is determined by the user based on the component type, corrosion morphology, or fitting accuracy, including: Linear model: It is suitable for situations where corrosion is relatively mild and load-bearing capacity degrades approximately linearly; Power function model: It is applicable to the nonlinear decay law under moderate to severe corrosion. Exponential model: It is suitable for situations where the load-bearing capacity decreases rapidly with the corrosion area; Among them, k and m are fitting coefficients, which are obtained by fitting the results of corrosion component load test or finite element analysis using the least squares method. The parameters corresponding to different component types and corrosion morphologies can be calibrated and updated based on experimental data.

6. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 5, characterized in that, The value of k is 0.015 to 0.025; the value of m is 1.0 to 1.

3.

7. The method for intelligent identification and performance evaluation of localized corrosion in steel structures based on deep learning according to claim 4, characterized in that, The The value range is divided into four levels to characterize component performance. The degree of degradation and safety status are as follows: Level I: Safe condition. ≥0.9 indicates that the component is in good condition or slightly corroded, and has no significant impact on the overall load-bearing capacity; Grade II: Slight corrosion, 0.8≤ <0.9 indicates that the component has a small amount of surface corrosion and a slight decrease in local load-bearing capacity, but still meets the design requirements; Level III: Moderate corrosion, 0.6 ≤ If the value is less than 0.8, and obvious corrosion areas appear in localized parts of the component, the load-bearing capacity is significantly reduced, and it should be included in the key monitoring scope. Level IV: Severe corrosion. If the corrosion area of ​​the component is large or the cross-section is significantly weakened, the remaining load-bearing capacity will decrease significantly, and reinforcement or replacement measures should be taken.

8. A deep learning-based intelligent identification and performance evaluation system for localized corrosion of steel structures, employing the methods described in claims 1 to 7, characterized in that, include: The system comprises a model loading module, an image input module, a corrosion detection module, a scale calibration module, an area calculation module, a performance evaluation module, and a result export module. The model loading module loads the model weight file. The image input module receives and preprocesses the image of the steel structure to be detected. The corrosion detection module uses a YOLOv12 network with an embedded CBAM attention mechanism to improve the recognition accuracy of small-scale corrosion areas and achieve automatic segmentation of corrosion areas. The scale calibration module supports both automatic and manual calibration methods and is used to calculate the ground sampling distance based on image capture parameters. The area calculation module performs physical conversion of the corrosion area based on mask integral and scale calibration. The performance evaluation module calculates the remaining load-bearing capacity of the component based on a corrosion area ratio and load-bearing capacity reduction coefficient model. The result export module generates a comprehensive report containing a corrosion distribution map, an area statistics table, and the calculated remaining load-bearing capacity, and exports it in CSV or PDF format.

9. The intelligent identification and performance evaluation system for localized corrosion of steel structures based on deep learning according to claim 8, characterized in that, The system's workflow is as follows: The user first loads the trained model weight file and selects the image or folder to be detected. The system then automatically performs model inference, generates a segmentation mask for the eroded area, and calculates the corresponding eroded area and area ratio. Subsequently, the system calls the performance evaluation model to calculate the remaining load-bearing capacity of the component based on the degree of corrosion, and displays the test results in real time on the interface; finally, the system automatically generates a test report, which includes a corrosion distribution map, an area statistics table, the calculation results of the remaining load-bearing capacity, and a comprehensive safety level recommendation.

10. The intelligent identification and performance evaluation system for localized corrosion of steel structures based on deep learning according to claim 8, characterized in that, The system is implemented using the PyQt5 architecture.

Images