Steel structure apparent corrosion dynamic analysis method and device and storage medium

Abstract

The invention relates to the technical field of steel structure corrosion, in particular to a steel structure apparent corrosion dynamic analysis method and device and a storage medium. The method comprises the steps of obtaining steel structure image data; preprocessing the steel structure image data; inputting the preprocessed steel structure image data into a corrosion time sequence prediction model to obtain corrosion prediction image data; performing image segmentation on the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data; evaluating the corrosion segmentation image data through an evaluation standard to obtain an evaluation grade; and performing early warning monitoring according to the corrosion prediction image data, the corrosion segmentation image data and the evaluation level to obtain early warning information. The future corrosion state and development trend of the steel structure can be predicted, and an intelligent early warning mechanism is constructed by quantifying the corrosion level.

Description

Technical Field

[0001] The present application relates to the technical field of steel structure corrosion, and in particular, to a method, device and storage medium for dynamically analyzing the apparent corrosion of steel structures. Background Art

[0002] In the related art, as a core material in modern engineering construction, steel structures are widely used in fields such as bridges, buildings, and marine engineering. However, under the long-term erosion of marine environments such as high humidity, high salinity, and variable climates, steel structures are extremely prone to corrosion. This corrosion not only weakens the load-bearing capacity of the structure, thus posing a threat to public safety, but also causes large-scale economic losses. Existing corrosion detection technologies mainly rely on manual or semi-automatic analysis of static images. However, static detection cannot capture the evolution law of corrosion in the time dimension, and thus cannot predict the future corrosion development trend. Traditional manual or semi-automatic analysis methods have low detection accuracy in complex environments and are difficult to meet the requirements of high-precision and high-real-time monitoring in actual engineering. Moreover, these methods often show obvious deficiencies when dealing with details such as fuzzy corrosion boundaries or crack propagation, and cannot provide a comprehensive and reliable quantitative assessment of the corrosion state.

[0003] In summary, the technical problems existing in the related art need to be improved. Summary of the Invention

[0004] The main purpose of the embodiments of the present application is to propose a method, device and storage medium for dynamically analyzing the apparent corrosion of steel structures, which can predict the future corrosion state and development trend of steel structures, give high-precision prediction images, and construct an intelligent early warning mechanism by quantifying the corrosion grade.

[0005] To achieve the above object, on the one hand, an embodiment of the present application proposes a method for dynamically analyzing the apparent corrosion of steel structures, the method comprising:

[0006] Obtaining steel structure image data;

[0007] Preprocessing the steel structure image data;

[0008] Inputting the preprocessed steel structure image data into a corrosion time series prediction model to obtain corrosion prediction image data;

[0009] Performing image segmentation on the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data;

[0010] Evaluating the corrosion segmentation image data through an evaluation criterion to obtain an evaluation grade;

[0011] Performing early warning monitoring based on the corrosion prediction image data, the corrosion segmentation image data and the evaluation grade to obtain early warning information.

[0012] In some embodiments, the corrosion time series prediction model includes an improved ConvLSTM network model, and the improved ConvLSTM network model includes a residual connection, a multi-scale feature fusion module, and a spatio-temporal feature fusion module connected by a feature information flow.

[0013] In some embodiments, inputting the preprocessed steel structure image data into the corrosion time series prediction model to obtain corrosion prediction image data includes:

[0014] Inputting the preprocessed steel structure image data into the network part including the residual connection in the improved ConvLSTM network model, and performing feature extraction on the steel structure image data through a convolution operation to obtain first steel structure feature data;

[0015] Performing feature extraction and fusion on the first steel structure feature data through the multi-scale feature fusion module in the improved ConvLSTM network model to obtain a comprehensive feature representation;

[0016] Performing feature extraction on the comprehensive feature representation through the spatio-temporal feature fusion module in the improved ConvLSTM network model, and modeling according to the extracted features to obtain second steel structure feature data;

[0017] Mapping the second steel structure feature data to a prediction result space through a fully connected layer to obtain preliminary corrosion prediction image data;

[0018] Correcting the preliminary corrosion prediction image data through a physical model to obtain the corrosion prediction image data.

[0019] In some embodiments, the corrosion segmentation model includes an improved HRNetV2 segmentation model, and the improved HRNetV2 segmentation model includes a dynamic depth adjustment mechanism module, a sparse connection and gated feature fusion module, and a depthwise separable convolution and dilated convolution module connected by a feature information flow.

[0020] In some embodiments, segmenting the corrosion prediction image data through the corrosion segmentation model to obtain corrosion segmentation image data includes:

[0021] Inputting the corrosion prediction image data into the high-resolution subnet in the improved HRNetV2 segmentation model, and performing feature extraction on the corrosion prediction image data in the high-resolution subnet through the depthwise separable convolution and dilated convolution module in the improved HRNetV2 segmentation model to obtain initial feature map data;

[0022] Parallelly compute the initial feature map data within multiple subnets through the improved HRNetV2 segmentation model to obtain depth feature map data; the network depth of the connection between the subnet and the high-resolution subnet is determined by the dynamic depth adjustment mechanism module;

[0023] Perform feature fusion and resolution restoration on the depth feature map data through the sparse connection and gated feature fusion module in the improved HRNetV2 segmentation model to obtain a segmentation prediction feature map; wherein, the number of channels of the segmentation prediction feature map is equal to the number of categories to be segmented, and each channel corresponds to a feature value;

[0024] Convert the feature value into a prediction probability for the corresponding category through an activation function;

[0025] Segment the prediction probability according to a prediction probability threshold to obtain the corrosion segmentation image data.

[0026] In some embodiments, the parallelly computing the initial feature map data within multiple subnets through the improved HRNetV2 segmentation model to obtain depth feature map data includes:

[0027] Parallelly compute the initial feature map data within each subnet through the sparse connection and gated feature fusion module and the depthwise separable convolution and dilated convolution module in the improved HRNetV2 segmentation model to obtain the depth feature map data;

[0028] During the parallel computing process, perform feature extraction on the initial feature map data through the depthwise separable convolution in the depthwise separable convolution and dilated convolution module, expand the receptive field of the initial feature map data through the dilated convolution in the depthwise separable convolution and dilated convolution module, transfer the features extracted from the initial feature map data between the subnets through the sparse connection in the sparse connection and gated feature fusion module, and fuse the initial feature map data after weighting through the gated unit in the gated feature fusion in the sparse connection and gated feature fusion module.

[0029] In some embodiments, the evaluating the corrosion segmentation image data through an evaluation criterion to obtain an evaluation level includes:

[0030] Statistically calculate the corrosion area of the corrosion segmentation image data through a method based on pixel counting to obtain the area of the corrosion region;

[0031] Calculate the corrosion rate by calculating the area of the corrosion region and the area of the steel structure image data through a corrosion rate calculation formula;

[0032] Match the corrosion rate according to the comprehensive standard table for evaluating the corrosion degree of steel structures to obtain the evaluation level.

[0033] In some embodiments, the method further includes the following steps:

[0034] Respond to the display instruction of the user display interface;

[0035] Control the user display interface to display the corrosion prediction image data, the corrosion segmentation image data, the evaluation level, and the warning information according to the display instruction;

[0036] Among them, the user display interface interacts through the Streamlit framework.

[0037] To achieve the above object, on the other hand, an embodiment of the present application proposes a device for dynamic analysis of the apparent corrosion of steel structures, and the device includes:

[0038] The first module is used to obtain steel structure image data;

[0039] The second module is used to preprocess the steel structure image data;

[0040] The third module is used to input the preprocessed steel structure image data into the corrosion time series prediction model to obtain corrosion prediction image data;

[0041] The fourth module is used to perform image segmentation on the corrosion prediction image data through the corrosion segmentation model to obtain corrosion segmentation image data;

[0042] The fifth module is used to evaluate the corrosion segmentation image data through the evaluation criteria to obtain the evaluation level;

[0043] The sixth module is used to perform warning monitoring according to the corrosion prediction image data, the corrosion segmentation image data, and the evaluation level to obtain warning information.

[0044] To achieve the above object, on the other hand, an embodiment of the present application proposes a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0045] The embodiments of the present application at least include the following beneficial effects: The present application provides a method, device and storage medium for dynamic analysis of the apparent corrosion of steel structures. This solution predicts the steel structure image data through a corrosion time series prediction model, accurately captures the dynamic evolution process of corrosion development, can effectively predict the future corrosion morphology and its development trend, and real-time tracks the corrosion changes. Image segmentation is performed on the corrosion prediction image data through a corrosion segmentation model, which greatly enhances the ability to capture corrosion boundaries and details, and ensures the high-precision and robustness of the segmentation. Through the evaluation level, the severity of corrosion can be quantitatively evaluated, and its potential risk level can be accurately judged. Through early warning monitoring, not only the safety of steel structures can be improved, but also sudden accidents caused by corrosion problems can be significantly reduced, and the scientific nature and efficiency of operation and maintenance management are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flowchart of the method for dynamic analysis of the apparent corrosion of steel structures provided by the embodiments of the present application;

[0047] Figure 2 is Figure 1 a flowchart of step S300 in

[0048] Figure 3 a schematic diagram of the first real-time series steel structure image data;

[0049] Figure 4 a schematic diagram of the second real-time series steel structure image data;

[0050] Figure 5 a schematic diagram of the third real-time series steel structure image data;

[0051] Figure 6 a schematic diagram of the predicted corrosion prediction image data;

[0052] Figure 7 is Figure 1 a flowchart of step S400 in

[0053] Figure 8 a schematic diagram of the first corrosion prediction image data;

[0054] Figure 9 a schematic diagram of the first corrosion segmentation image data;

[0055] Figure 10 a schematic diagram of the second corrosion prediction image data;

[0056] Figure 11 a schematic diagram of the second corrosion segmentation image data;

[0057] Figure 12 a schematic diagram of the third corrosion prediction image data;

[0058] Figure 13 It is a schematic diagram of the third corrosion segmentation image data;

[0059] Figure 14 It is a table of the comprehensive standard for evaluating the corrosion degree of steel structures;

[0060] Figure 15 It is a schematic diagram of the user display interface;

[0061] Figure 16 It is a schematic structural diagram of a device for dynamic analysis of the apparent corrosion of steel structures. Detailed implementation manners

[0062] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the embodiments of the present application. They are only examples of devices and methods that are consistent with some aspects of the embodiments of the present application.

[0063] It can be understood that the terms "first", "second", etc. used in the present application can be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of the present application, the first information can also be referred to as the second information. Similarly, the second information can also be referred to as the first information. Depending on the context, as used herein, the words "if", "when" can be interpreted as "when...", "when...", or "in response to determining".

[0064] The terms "at least one", "a plurality", "each", "any one", etc. used in the present application, at least one includes one, two or more than two, a plurality includes two or more than two, each refers to each of the corresponding plurality, and any one refers to any one of the plurality.

[0065] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present application belongs. The terms used herein are only for the purpose of describing the embodiments of the present application and are not intended to limit the present application.

[0066] Before elaborating on the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application will be described first. The nouns and terms involved in the embodiments of the present application are applicable to the following explanations.

[0067] LSTM: Long Short-Term Memory, is a special type of Recurrent Neural Network (RNN).

[0068] ConvLSTM: Convolutional Long Short-Term Memory, which is a variant of LSTM.

[0069] RGB image: A common representation of color images, where RGB stands for red, green, and blue respectively.

[0070] HRNetV2: The second version of the High-Resolution Network, which is a deep neural network architecture for computer vision tasks such as image segmentation and pose estimation.

[0071] Streamlit: An open-source Python framework designed specifically for machine learning and data science projects, used to quickly build interactive web applications.

[0072] In the related art, Figure 1 is an optional flowchart of the method for dynamic analysis of the apparent corrosion of steel structures provided by the embodiments of the present application. Figure 1 The method in may include but is not limited to steps S100 to S500:

[0073] Step S100: Obtain steel structure image data;

[0074] Step S200: Preprocess the steel structure image data;

[0075] Step S300: Input the preprocessed steel structure image data into the corrosion time series prediction model to obtain corrosion prediction image data;

[0076] Step S400: Segment the corrosion prediction image data through the corrosion segmentation model to obtain corrosion segmentation image data;

[0077] Step S500: Evaluate the corrosion segmentation image data according to the evaluation criteria to obtain an evaluation level;

[0078] Step S600: Perform early warning monitoring based on the corrosion prediction image data, the corrosion segmentation image data, and the evaluation level to obtain early warning information.

[0079] In some embodiments, in step S100, to obtain steel structure image data, professional equipment such as industrial cameras can be set at the steel structure to be detected or monitored. The apparent state image data of the steel structure to be detected or monitored is obtained through the professional equipment for obtaining image data, and the obtained image data is used to prepare for subsequent dynamic detection and monitoring.

[0080] In some embodiments, in step S200, preprocessing operations are performed on the obtained image data, including size normalization, pixel value standardization, data augmentation, etc., to improve the training effect and generalization ability of the model. The methods of data augmentation include random flipping, rotation, scaling, adding noise, etc., to increase the richness of the data.

[0081] Figure 2 Yes Figure 1 is the flowchart of step S300 in Figure 1 Step S300 of includes but is not limited to steps S310 to S350:

[0082] Step S310: Input the preprocessed steel structure image data into the network part with residual connections in the improved ConvLSTM network model, and perform feature extraction on the steel structure image data through convolution operations to obtain the first steel structure feature data;

[0083] Step S320: Perform feature extraction and fusion on the first steel structure feature data through the multi-scale feature fusion module in the improved ConvLSTM network model to obtain a comprehensive feature representation;

[0084] Step S330: Perform feature extraction on the comprehensive feature representation through the spatio-temporal feature fusion module in the improved ConvLSTM network model, and model according to the extracted features to obtain the second steel structure feature data;

[0085] Step S340: Map the second steel structure feature data to the prediction result space through a fully connected layer to obtain preliminary corrosion prediction image data;

[0086] Step S350: Correct the preliminary corrosion prediction image data through a physical model to obtain corrosion prediction image data.

[0087] In some embodiments, in step S310, the preprocessed steel structure image data is input into the trained improved ConvLSTM network model. Ensure that the preprocessing of each frame is consistent with the training data, such as size, pixel value normalization, etc., to guarantee input consistency. The input data enters the network part with residual connections. If there are residual connections between some ConvLSTM layers, after the spatio-temporal feature extraction of the first ConvLSTM layer, part of the features are directly transmitted to the subsequent residual connection nodes, and the other part is also transmitted here after being processed by the intermediate layer. The two are added together to obtain the first steel structure feature data. This enables the network to learn richer features, prevents gradient disappearance and degradation, provides representative information for prediction, and helps predict subtle structural changes.

[0088] In some embodiments, in step S320, the features after the residual connection processing enter the multi-scale feature extraction branch. Feature extraction modules of different scales start to work. The large convolution kernel module extracts the overall large-scale structure features, and the small convolution kernel module captures the detailed features. Then, they are fused in a specific fusion layer through methods such as splicing or weighted summation to obtain a comprehensive feature representation.

[0089] In some embodiments, in step S330, for the processed image sequence data, the data after the previous residual connection and multi-scale feature fusion module processing enters the spatio-temporal feature fusion structure. The spatial convolution operation extracts spatial features on each frame of the image, and the LSTM operation processes the features of consecutive frames along the time dimension, memorizing the feature changes at different moments, and capturing the object movement trajectory and dynamic changes. Through the extracted features for modeling, the second steel structure feature data is obtained. It can be understood that spatio-temporal feature modeling requires obtaining the necessary information and data basis through feature extraction. Only by first extracting effective spatio-temporal features can higher-level modeling be carried out on this basis to reveal the complex relationships and patterns in spatio-temporal data.

[0090] In some embodiments, in step S340, the second steel structure feature data obtained after the above series of processes, in the final stage of the network, through the fully connected layer or other specific output layers, maps the fused features to the final prediction result space, and outputs the preliminary corrosion prediction image data according to the task.

[0091] In some embodiments, in step S350, through the continuous prediction results, the expansion trend of the corrosion area can be analyzed, and combined with the physical model for correction to obtain the final corrosion prediction image data.

[0092] In some embodiments, in combination with steps S310 to S350, it can be understood that the input data first enters the ConvLSTM layer part including residual connections. The residual connections enable part of the features to be directly skipped and transferred to subsequent specific positions after the first ConvLSTM layer extracts spatio-temporal features from the data, and part of the features are processed by the intermediate ConvLSTM layers and then merged and added to avoid gradient problems and enhance feature learning. Next is the multi-scale feature fusion module. The network sets up different-scale feature extraction branches. The large convolutional kernel module is responsible for extracting large-scale overall structural features to grasp the general outline and layout of the image; the small convolutional kernel module focuses on capturing detailed features. These different-scale features are fused at a specific fusion layer by means of concatenation or weighted summation, etc., to generate a feature representation containing multi-scale information, improving the ability to understand objects in complex scenes. Then comes the spatio-temporal feature fusion part. For image sequence data, spatial convolution operations first extract spatial features on each frame of the image to capture information such as the shape and texture of the object. Subsequently, the LSTM operation processes the features of consecutive frames along the time dimension, memorizing the feature changes at different times and capturing the motion trajectories and dynamic changes of the object. The spatial convolution features are combined with the hidden states output by the LSTM in a weighted manner, etc., enabling the network to comprehensively consider the spatio-temporal information of the object. At the backend of the network, the feature data processed as above is mapped to the final prediction result space through a fully connected layer or a specific output layer, and the prediction results corresponding to the task are output, such as classification, object detection, or image prediction results, etc.

[0093] In some embodiments, in step S300, for the corrosion time-series sequencing model, an improved ConvLSTM network model is adopted. The improvements to the improved ConvLSTM network model include adding residual connections, a multi-scale feature fusion module, and spatio-temporal feature fusion.

[0094] Specifically, for the corrosion time-series sequencing model, the input data first enters the network part including residual connections. In the improved ConvLSTM network model, residual connections are constructed between certain ConvLSTM layers. When the data passes through the first ConvLSTM layer for spatio-temporal feature extraction, part of the features are directly transferred to the residual connection node after the next ConvLSTM layer. At the same time, part of the features also reach this node after being processed by the intermediate ConvLSTM layers. These two parts of the features are added together, enabling the network to learn richer features, avoiding the problems of gradient disappearance and degradation that occur as the network depth increases, and thus providing more representative feature information for subsequent predictions. For example, when predicting subtle structural changes in an image, the residual connections can ensure that the key detailed features learned at the bottom layer are not lost and are smoothly transferred to subsequent layers for calculation.

[0095] In the multi-scale feature fusion module for the corrosion time-series sequencing model, multiple-scale feature extraction branches are designed in the network. After some basic convolutional and ConvLSTM layer operations, the feature extraction modules of different scales start to work. For example, a module with a larger convolutional kernel is responsible for extracting the overall structural features at a large scale, while a module with a smaller convolutional kernel focuses on capturing the detailed features of the image. These features of different scales are fused at a specific fusion layer. A common fusion method is concatenation, where the feature maps of different scales are concatenated along the channel dimension to form a feature representation containing multi-scale information. Another method is weighted summation, where the features of different scales are weighted and added according to the weights automatically learned during training to highlight the scale information that is more important for the prediction task. For example, when predicting an object in a complex scene, the large-scale features can help locate the approximate position of the object, while the small-scale features are helpful for identifying the specific category and detailed features of the object. The information after the multi-scale feature fusion module enables the model to make more accurate predictions.

[0096] In the corrosion time-series sequencing model, for image sequence data, the spatio-temporal feature fusion module plays a key role. After being processed by the previous residual connection and multi-scale feature fusion module, the data enters the spatio-temporal feature fusion structure. In this structure, the spatial convolutional operation and the LSTM operation on the time series are closely combined. The spatial convolutional operation extracts spatial features from each frame of the image, capturing information such as the shape and texture of the objects in the image. The LSTM operation processes the features of consecutive frames along the time dimension, memorizing the feature changes at different times and capturing the motion trajectories and dynamic changes of the objects. For example, when predicting the future motion state of an object in a video, the spatial features of each frame of the object extracted by the spatial convolution are transmitted and updated in the time dimension through the memory units of the LSTM, enabling the model to comprehensively consider the spatial positions and motion trends of the object at different times, and thus predict the position and state of the object at the next moment.

[0097] In some embodiments, a residual connection is added based on the improved ConvLSTM network model, that is, information loss is avoided through skip connections to ensure the stable improvement of performance when the number of network layers increases. The expression is as follows:

[0098] y = f(x) + x;

[0099] where y represents the output, f(x) represents the feature map calculated through the convolutional network, and x represents the input.

[0100] Based on the improved ConvLSTM network model, a multi-scale feature extraction mechanism is adopted. Different-scale feature maps are obtained through multi-resolution convolutional operations, and then feature fusion is performed. This strategy can enhance the model's ability to capture corrosion details and large-scale trends, and reduce information loss caused by the blurred corrosion boundary.

[0101] Based on the improved ConvLSTM network model, through the spatio-temporal feature fusion module, combining the advantages of spatial convolution and temporal LSTM units to model the historical image sequence, enhancing the network's prediction ability for the dynamic changes of corrosion. The spatio-temporal features are modeled through the following expression:

[0102] h t = LSTM(x t , h t-1 );

[0103] Among them, h t represents the hidden state at the current moment, x t represents the current input, h t-1 represents the hidden state at the previous moment, and LSTM represents the long short-term memory network.

[0104] In some embodiments, for the training of the corrosion time series prediction model, based on a large amount of indoor salt spray test data, real corrosion sequence images are used for model training and optimization. To improve the training effect, the following strategies are adopted:

[0105] Data augmentation: Generate diverse corrosion scenarios through methods such as rotation, flipping, and scale change to simulate the impact of different climate and environmental conditions on corrosion evolution.

[0106] Loss function design: Adopt the weighted cross-entropy loss function and L2 regularization to balance the training of the model, avoid overfitting, and ensure the accuracy of the time series prediction images. The form of the loss function is as follows:

[0107]

[0108] Among them, L cross represents the cross-entropy loss, L reg represents the L2 regularization term, α represents the weight coefficient of the cross-entropy loss, β represents the weight coefficient of the L2 regularization term, and L represents the total loss function.

[0109] By adjusting the weight coefficients of the weighted cross-entropy loss function and the L2 regularization term, the focus on the classification accuracy and prevention of overfitting of the model can be flexibly balanced according to the specific tasks and data characteristics. For example, in the case of a small amount of data and easy overfitting, the weight of the L2 regularization can be appropriately increased; if the problem of data sample imbalance is more prominent, the weight of the weighted cross-entropy loss function can be increased. The loss function generated by combining the two can not only handle the problem of sample imbalance but also effectively prevent the model from overfitting. In actual model training, especially when facing complex data sets and large-scale models, these two problems often occur simultaneously, so this combined loss function can well handle these problems.

[0110] Optimization algorithm: The Adam optimizer is used for parameter update, and the learning rate is dynamically adjusted to improve the convergence speed and stability.

[0111] In some embodiments, as Figures 3 to 6 shown, Figure 3 is a schematic diagram of the true time-series steel structure image data of the first steel structure appearance, Figure 4 is a schematic diagram of the second true time-series steel structure image data of the steel structure appearance, Figure 5 is a schematic diagram of the third true time-series steel structure image data of the steel structure appearance, Figure 6 is a schematic diagram of the predicted corrosion prediction image data. Among them, Figure 3 is taken earlier in time series than Figure 4 , Figure 4 is taken earlier in time series than Figure 5 , and the steel structure time-series prediction images of Figure 3 , Figure 4 and Figure 5 processed by the corrosion time-series prediction model are obtained Figure 6 schematic diagram of the predicted corrosion prediction image data. Specifically, the model outputs the corrosion prediction image data for several future time steps. The corrosion prediction image data not only provides input for subsequent corrosion area segmentation but also provides data support for corrosion level assessment and intelligent warning systems. The model generates the corrosion development images for each time step, capturing the spatial distribution and dynamic evolution of the corrosion area.

[0112] In some embodiments, the corrosion segmentation model includes an improved HRNetV2 segmentation model. The improved HRNetV2 segmentation model includes a dynamic depth adjustment mechanism module, a sparse connection and gated feature fusion module, and a depthwise separable convolution and dilated convolution module. In the initial high-resolution subnet part, depthwise separable convolution is used for feature extraction to reduce the computational amount. As the network extends, according to the dynamic depth adjustment mechanism module, the network depth is flexibly determined based on the complexity of the input image. When connecting different-resolution subnets, redundant connections are reduced using sparse connections to lower the model complexity. In each subnet, dilated convolution is used for the convolutional layers that need to expand the receptive field to obtain a wider context information without increasing the number of parameters. And in the feature fusion stage of different-resolution subnets, a gated feature fusion mechanism is introduced. The gating unit assigns weights to different-resolution feature maps according to the feature importance and adaptively adjusts the fusion ratio. Different-resolution subnets continuously perform parallel calculations, repeating the operations of the depthwise separable convolution, dilated convolution, sparse connection, and gated feature fusion module until reaching the network output layer. Through such a construction structure, the improved HRNetV2 segmentation model reduces the computational cost while enhancing the ability to capture different-scale features, improving the model performance, and being able to more effectively complete tasks such as image segmentation.

[0113] Figure 7 Yes Figure 1 is the flowchart of step S400 in Figure 1 Step S400 of includes but is not limited to steps S410 to S460:

[0114] Step S410: Input the corrosion prediction image data into the high-resolution subnet in the improved HRNetV2 segmentation model, and perform feature extraction on the corrosion prediction image data in the high-resolution subnet through the depthwise separable convolution and dilated convolution module in the improved HRNetV2 segmentation model to obtain the initial feature map data;

[0115] Step S420: Perform parallel computing on the initial feature map data in multiple subnets through the improved HRNetV2 segmentation model to obtain the depth feature map data; the network depth of the connection between the subnet and the high-resolution subnet is determined by the dynamic depth adjustment mechanism module;

[0116] Step S430: Perform feature fusion and resolution restoration on the depth feature map data through the sparse connection and gated feature fusion module in the improved HRNetV2 segmentation model to obtain the segmentation prediction feature map; among them, the number of channels of the segmentation prediction feature map is equal to the number of categories to be segmented, and each channel corresponds to a feature value;

[0117] Step S440: Convert the feature value into the prediction probability of the corresponding category through the activation function;

[0118] Step S450: Segment the prediction probability according to the prediction probability threshold to obtain the corrosion segmentation image data.

[0119] In some embodiments, in step S410, the corrosion prediction image data is input into the high-resolution subnet of the improved HRNetV2 segmentation model. This subnet first uses depthwise separable convolution to perform preliminary feature extraction on the image. Depthwise separable convolution decomposes the standard convolution into depth convolution and pointwise convolution. Depth convolution performs convolution operations independently on each input channel, and pointwise convolution is used to adjust the number of channels. For example, for an input RGB image with 3 channels, depth convolution will perform convolution on the red, green, and blue channels respectively to extract the features of each channel, and then pointwise convolution will integrate and transform these features to obtain the initial feature map data.

[0120] In some embodiments, in step S420, according to the network depth determined by the dynamic depth adjustment mechanism module, the high-resolution subnet gradually and parallelly connects multiple subnets with different resolutions for parallel computing. During the computing process, depthwise separable convolution is used for feature extraction, dilated convolution is used to expand the receptive field, sparse connection is used to transfer the extracted features between subnets, and the gating unit in gated feature fusion assigns weights to the initial feature map data and then performs fusion. In each subnet, depthwise separable convolution is continued to be used for feature extraction to obtain depth feature map data. Specifically, for some convolutional layers that need to expand the receptive field, dilated convolution is adopted. Dilated convolution inserts holes in the standard convolutional kernel, enabling the convolutional operation to obtain more extensive context information without increasing the number of parameters. During the computing process of subnets with different resolutions, depth feature map data of different scales will be generated, and these depth feature map data contain various information of the image from details to the whole.

[0121] In some embodiments, in step S430, when performing feature transfer between subnets with different resolutions, a sparse connection mechanism is applied. Sparse connection does not fully connect all neurons, but only retains some important connections. This means that during the feature transfer process, only those connections that make important contributions to subsequent feature extraction and fusion will be retained, reducing unnecessary computations and the number of parameters. Feature fusion operations are performed on the depth feature map data through a series of convolutional layers to further integrate the features of different subnets. At the same time, upsampling techniques are used to gradually restore the resolution of the depth feature map data, where the upsampling techniques include bilinear interpolation upsampling or deconvolution operations, etc., to make it the same size as the input image. During this process, the convolutional layers continue to refine and adjust the features to meet the requirements of the segmentation task. After feature fusion and resolution restoration, a segmentation prediction feature map with the same size as the input image is obtained. The number of channels of the segmentation prediction feature map is equal to the number of categories to be segmented. For example, for the steel structure corrosion segmentation task, it may include categories such as background, mild corrosion, moderate corrosion, severe corrosion, etc. Each channel corresponds to the prediction information of a category.

[0122] When depth feature map data of different resolutions are passed to the gated feature fusion module, the gating mechanism starts to take effect. The gating unit assigns different weights to each depth feature map data according to the importance of the input features. These weights are learned and can adaptively adjust the fusion ratio of features with different resolutions. By means of weighted summation and other methods, the depth feature map data of different resolutions are fused together to obtain a more representative feature representation. The fused depth feature map data continue to be calculated in the subsequent layers of the network, repeating operations such as depthwise separable convolution, dilated convolution, sparse connection, and gated feature fusion until the final output layer of the network is reached. In this process, the network continuously extracts, transforms, and fuses features, gradually extracting higher-level and more abstract features. After a series of previous calculations, it finally reaches the output layer of the network. The output layer is usually a convolutional layer, and the number of channels of the output segmentation prediction feature map is the same as the number of categories to be segmented, and each channel corresponds to a feature value.

[0123] In some embodiments, in steps S440 to S450, for each pixel point of the output layer, the feature value is converted into the corresponding class prediction probability through an activation function, so that each pixel point has a probability distribution belonging to different classes. According to the specific task requirements, a suitable threshold is selected to segment the prediction probability. If the probability of a certain pixel point in a certain class is greater than the threshold, the pixel point is classified into that class; otherwise, it is classified into other classes. In this way, the output probability map is converted into a specific class label map to obtain the final corrosion segmentation image data.

[0124] In some embodiments, the dynamic depth adjustment mechanism module dynamically selects whether to skip some convolutional layers by analyzing the complexity of the input features. In scenarios with low feature complexity, it automatically and selectively skips redundant calculations, significantly improving the inference efficiency and reducing the computational burden. This mechanism uses the statistical information of the input features (such as variance, mean, etc.) to determine whether to perform the forward calculation of certain layers. The expression is:

[0125] skip_layer = f(μ(X), σ(X));

[0126] where f represents the gating function, μ(X) represents the mean of the input features Figure X and σ(X) represents the variance of the input features Figure X , and skip_layer represents the network depth. These statistics are used to determine whether to skip some convolutional layers.

[0127] For the sparse connection and gated feature fusion module, a sparse connectivity mechanism is introduced. During the feature fusion process, only some important branches are selected for fusion, thereby reducing the transmission of redundant features and improving computational efficiency. At the same time, a gated mechanism (Gated Feature Fusion) is adopted to dynamically activate specific branches according to the characteristics of the input features, thus strengthening the selective fusion of features with different resolutions and avoiding interference from irrelevant information. The expression is:

[0128] Foutput = G(F1,F2,…,F n );

[0129] Among them, F i represents the features of each branch, G represents the gating function, and Foutput represents the fusion output.

[0130] For the depthwise separable convolution and dilated convolution module, depthwise separable convolution (Depthwise Separable Convolution) is introduced to reduce computational complexity while retaining segmentation accuracy, and dilated convolution (Dilated Convolution) is combined to enhance the receptive field, thereby expanding the context modeling ability, especially at the boundaries and details of the corroded area, and improving the recognition ability of large-scale corroded areas. The expression of depthwise separable convolution is:

[0131] Y = W d *X;

[0132] Among them, W d is the depth convolution kernel, * represents the depthwise separable convolution operation, X is the input feature, and Y is the output feature. By performing the depthwise separable convolution operation independently in each channel, the computational amount is reduced.

[0133] The expression of dilated convolution is:

[0134] Y = W d * D X;

[0135] Among them, * represents the dilated convolution operation, D is the dilation rate, W d is the depth convolution kernel, X is the input feature, and Y is the output feature.

[0136] In some embodiments, for the dynamic depth adjustment mechanism module, when the improved HRNetV2 segmentation model processes images of different complexities, the dynamic depth adjustment mechanism module can automatically adjust the depth of the network according to the characteristics of the input image. This enables the improved HRNetV2 segmentation model to more flexibly adapt to different image data, enhancing the overall performance and efficiency. By dynamically adjusting the network depth, the waste of resources caused by using a fixed-depth network in all cases is avoided, allowing the improved HRNetV2 segmentation model to operate efficiently even in resource-constrained environments. For the sparse connection and gated feature fusion module, sparse connections can reduce unnecessary connections in the network, lowering the complexity and computational load of the model. In the improved HRNetV2 segmentation model, some connections that contribute less to feature extraction can be removed through sparse connections, making the network more lightweight. The gated feature fusion mechanism can screen and enhance input features according to their importance. During the multi-resolution feature fusion process of HRNetV2, the gating mechanism can adaptively adjust the weights of features at different resolutions, highlighting important features and suppressing irrelevant or noisy features, thereby improving the effect of feature fusion. In the depthwise separable convolution and dilated convolution module, depthwise separable convolution decomposes the standard convolution into depth convolution and pointwise convolution, significantly reducing the computational load and the number of parameters of the convolution operation. Using depthwise separable convolution in the improved HRNetV2 segmentation model can reduce the complexity of the model and improve the training and inference speeds. Dilated convolution can increase the receptive field of the convolutional layer without increasing the size of the convolution kernel and the number of parameters by inserting holes in the convolution kernel. In the improved HRNetV2 segmentation model, dilated convolution helps the model obtain more extensive context information and enhances the segmentation ability for objects at different scales.

[0137] In some embodiments, such as Figures 8 to 13 shown, Figure 8 is a schematic diagram of the first corrosion prediction image data, Figure 9 is a schematic diagram of the first corrosion segmentation image data, Figure 10 is a schematic diagram of the second corrosion prediction image data, Figure 11 is a schematic diagram of the second corrosion segmentation image data, Figure 12 is a schematic diagram of the third corrosion prediction image data, Figure 13 is a schematic diagram of the third corrosion segmentation image data. Specifically, Figure 8 the yellow part of Figure 9 is the first corrosion prediction image data obtained by predicting the preprocessed steel structure image data through the corrosion timing prediction model, Figure 10 and Figure 11 the red part of Figure 12 is the first corrosion segmentation image data obtained by segmenting the corrosion prediction image data using the corrosion segmentation model. Similarly, Figure 13They correspond to each other, which are respectively the second and third corrosion prediction image data corresponding to the yellow part and the second and third corrosion segmentation image data of the red part. By comparing the diagrams of the yellow and red parts, it is obvious that the segmentation result clearly shows the outline of the corrosion area, and the ability to capture the corrosion boundary and details is greatly enhanced, ensuring high-precision segmentation.

[0138] In some embodiments, in step S500, through an improved ConvLSTM model for time series prediction, the evolution image of steel structure corrosion in the future for a period of time is predicted. On this basis, an improved HRNetV2 segmentation algorithm is used to accurately segment the corrosion area in the predicted image, and combined with the corrosion level evaluation standard, the progress and development trend of corrosion are quantified. Specifically, the quantification of the corrosion level not only considers the proportion of the current corrosion area, but also combines historical data for time series analysis to predict the possible future corrosion range and risk level. The area of the corrosion area is statistically obtained by counting the pixels of the corrosion segmentation image data. The corrosion rate is calculated by calculating the area of the corrosion area and the area of the steel structure image data. The corrosion rate formula is:

[0139]

[0140] Among them, A corrosion represents the area of the corrosion area, A total represents the total area of the image, and CorrosionArea represents the corrosion rate.

[0141] Combined with Figure 14 the comprehensive standard table for evaluating the corrosion degree of the steel structure shown in the figure to match the corrosion rate, the evaluation level is obtained. In the table, the corresponding evaluation level and the corresponding anti-corrosion measures are obtained according to the corrosion rate.

[0142] In some embodiments, in step S600, based on the above prediction and segmentation results, when the corrosion level in the prediction result exceeds the set danger threshold, the system will automatically trigger the warning mechanism. Specifically, threshold judgment is performed by comparing the corrosion level with the set danger threshold to automatically identify the severity of the current corrosion. For example, when the corrosion area exceeds a certain percentage, the system considers that it has entered a dangerous state. Once it is detected that the corrosion level reaches the warning standard, the system will generate an alarm signal, automatically send a warning message to the relevant maintenance personnel or decision-makers, and attach a detailed corrosion analysis report. Through the warning message, not only the safety of the steel structure can be improved, but also the sudden accidents caused by corrosion problems can be significantly reduced, improving the scientificity and efficiency of operation and maintenance management. Through accurate corrosion prediction and real-time warning, the system provides strong support for the long-term operation and maintenance of the steel structure, effectively reducing the operation cost and extending the service life of the structure.

[0143] In some embodiments, such as Figure 15 As shown in the schematic diagram of the user display interface, the method further includes responding to a display instruction of the user display interface; controlling the user display interface to display corrosion prediction image data, corrosion segmentation image data, an evaluation level, and warning information according to the display instruction; wherein, the user display interface interacts through the Streamlit framework. This interface can not only display the real-time corrosion status, but also visually present the prediction results, corrosion evolution trend, and corresponding warning information. Specifically, it shows the current corrosion status of the steel structure, including the predicted future corrosion images and segmentation results. When the system triggers a warning, the user interface will update the warning status in real time and automatically pop up relevant safety tips and maintenance suggestions to ensure that users can respond within the shortest time. Through this intelligent warning mechanism, the efficiency of steel structure corrosion monitoring and management can be significantly improved, and strong guarantee for structural safety can be provided in practical applications.

[0144] In some embodiments, the present application can handle various complex corrosion scenarios, especially performing excellently in the corrosion detection and monitoring of steel structures in marine environments. Its core technology has good scalability in aspects such as steel structure health detection and monitoring, corrosion assessment, and risk prediction. It is not only applicable to fields such as bridges, buildings, and ocean engineering, but also can be customized according to different environments and corrosion characteristics. The high adaptability of the system enables it to adapt to changing external environments and different corrosion forms, meeting the high requirements for the accuracy, efficiency, and scalability of detection and monitoring in practical applications.

[0145] As Figure 16 shown, a device for dynamic analysis of apparent corrosion of steel structures, the device includes:

[0146] A first module for acquiring steel structure image data;

[0147] A second module for preprocessing the steel structure image data;

[0148] A third module for inputting the preprocessed steel structure image data into a corrosion time series prediction model to obtain corrosion prediction image data;

[0149] A fourth module for performing image segmentation on the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data;

[0150] A fifth module for evaluating the corrosion segmentation image data according to an evaluation criterion to obtain an evaluation level;

[0151] A sixth module for performing warning monitoring according to the corrosion prediction image data, corrosion segmentation image data, and evaluation level to obtain warning information.

[0152] It can be understood that the content in the above method embodiments is applicable to the device embodiments of the present application. The functions specifically implemented by the device embodiments of the present application are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0153] The present application can be used in numerous general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. The present 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. The present application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0154] The embodiment of the present application also provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned method for automatically analyzing the state of an electrochemical gas sensor. The computer device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0155] It can be understood that the content in the above method embodiments is applicable to the device embodiments of the present application. The functions specifically implemented by the device embodiments of the present application are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0156] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include high-speed random access memory, and can also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory optionally includes a memory remotely located relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0157] The embodiments described in the embodiments of the present application are for more clearly explaining the technical solutions of the embodiments of the present application, and do not constitute a limitation to the technical solutions provided by the embodiments of the present application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0158] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation to the embodiments of the present application, and may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0159] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, and their appropriate combinations.

[0161] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0162] The units described as separate components above may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0163] In addition, in each embodiment of the present application, the functional units may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0164] The preferred embodiments of the embodiments of the present application have been described above with reference to the accompanying drawings. However, this does not limit the scope of the rights of the embodiments of the present application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of the present application shall fall within the scope of the rights of the embodiments of the present application.

Claims

1. A method for dynamic analysis of the apparent corrosion of a steel structure, characterized in that, The method includes: Obtain steel structure image data; Preprocess the steel structure image data; Input the preprocessed steel structure image data into a corrosion time series prediction model to obtain corrosion prediction image data; Perform image segmentation on the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data; Evaluate the corrosion segmentation image data according to an evaluation criterion to obtain an evaluation level; Carry out early warning monitoring based on the corrosion prediction image data, the corrosion segmentation image data, and the evaluation level to obtain early warning information.

2. The method according to claim 1, characterized in that The corrosion time series prediction model includes an improved ConvLSTM network model, and the improved ConvLSTM network model includes a residual connection, a multi-scale feature fusion module, and a spatio-temporal feature fusion module connected by a feature information flow.

3. The method according to claim 2, wherein The step of inputting the preprocessed steel structure image data into a corrosion time series prediction model to obtain corrosion prediction image data includes: Input the preprocessed steel structure image data into the network part including the residual connection in the improved ConvLSTM network model, and perform feature extraction on the steel structure image data through a convolution operation to obtain first steel structure feature data; Perform feature extraction and fusion on the first steel structure feature data through the multi-scale feature fusion module in the improved ConvLSTM network model to obtain a comprehensive feature representation; Perform feature extraction on the comprehensive feature representation through the spatio-temporal feature fusion module in the improved ConvLSTM network model, and perform modeling according to the extracted features to obtain second steel structure feature data; Map the second steel structure feature data to a prediction result space through a fully connected layer to obtain preliminary corrosion prediction image data; Correct the preliminary corrosion prediction image data through a physical model to obtain the corrosion prediction image data.

4. The method according to claim 1, wherein The corrosion segmentation model includes an improved HRNetV2 segmentation model, and the improved HRNetV2 segmentation model includes a dynamic depth adjustment mechanism module, a sparse connection and gated feature fusion module, and a depthwise separable convolution and dilated convolution module connected by a feature information flow.

5. The method according to claim 4, wherein The step of performing image segmentation on the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data includes: Input the corrosion prediction image data into the high-resolution subnet in the improved HRNetV2 segmentation model, and perform feature extraction on the corrosion prediction image data in the high-resolution subnet through the depthwise separable convolution and dilated convolution module in the improved HRNetV2 segmentation model to obtain initial feature map data; Perform parallel computing on the initial feature map data in multiple subnets through the improved HRNetV2 segmentation model to obtain depth feature map data; the network depth of the connection between the subnet and the high-resolution subnet is determined by the dynamic depth adjustment mechanism module; Feature fusion and resolution recovery are performed on the depth feature map data through the sparse connection and gated feature fusion module in the improved HRNetV2 segmentation model to obtain a segmentation prediction feature map; wherein, the number of channels of the segmentation prediction feature map is equal to the number of categories to be segmented, and each channel corresponds to a feature value; The feature value is converted into a prediction probability for the corresponding category through an activation function; The prediction probability is segmented according to a prediction probability threshold to obtain the corrosion segmentation image data.

6. The method according to claim 5, wherein The parallel calculation of the initial feature map data in multiple subnets through the improved HRNetV2 segmentation model to obtain depth feature map data includes: Parallel calculation is performed on the initial feature map data in each subnet through the sparse connection and gated feature fusion module and the depthwise separable convolution and dilated convolution module in the improved HRNetV2 segmentation model to obtain the depth feature map data; During the parallel calculation process, feature extraction is performed on the initial feature map data through the depthwise separable convolution in the depthwise separable convolution and dilated convolution module, the receptive field of the initial feature map data is enlarged through the dilated convolution in the depthwise separable convolution and dilated convolution module, the features extracted from the initial feature map data are transferred between the subnets through the sparse connection in the sparse connection and gated feature fusion module, and the initial feature map data is fused after being weighted by the gated unit in the gated feature fusion in the sparse connection and gated feature fusion module.

7. The method according to claim 1, characterized in that The evaluation of the corrosion segmentation image data through an evaluation criterion to obtain an evaluation level includes: The area of the corrosion region is obtained by statistically counting the corrosion area of the corrosion segmentation image data through a method based on pixel counting; The corrosion rate is obtained by calculating the area of the corrosion region and the area of the steel structure image data through a corrosion rate calculation formula; The evaluation level is obtained by matching the corrosion rate according to the comprehensive standard table for evaluating the corrosion degree of the steel structure.

8. The method according to claim 1, characterized in that The method further includes the following steps: Responding to a display instruction of the user display interface; Controlling the user display interface to display the corrosion prediction image data, the corrosion segmentation image data, the evaluation level, and the warning information according to the display instruction; Wherein, the user display interface interacts through the Streamlit framework.

9. A device for dynamically analyzing the apparent corrosion of a steel structure, characterized in that, The device includes: A first module for acquiring steel structure image data; A second module for preprocessing the steel structure image data; A third module for inputting the preprocessed steel structure image data into a corrosion time series prediction model to obtain corrosion prediction image data; A fourth module for segmenting the corrosion prediction image data through a corrosion segmentation model to obtain corrosion segmentation image data; A fifth module for evaluating the corrosion segmentation image data through an evaluation criterion to obtain an evaluation level; A sixth module for performing warning monitoring based on the corrosion prediction image data, the corrosion segmentation image data, and the evaluation level to obtain warning information.

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 according to any one of claims 1 to 8.

Images