Steel structure surface corrosion identification method and device, medium and equipment

Abstract

The invention discloses a steel structure surface corrosion identification method and device, a medium and equipment, and relates to the technical field of image identification. The method comprises the steps of obtaining an image data set of a steel structure surface; according to the corrosion parameter of each image in the image data set, performing corrosion degree marking on each image to obtain a training data set; a Faster ViT module is used for replacing a C2f layer in a neck network of the YOLOv8 network, and an improved YOLOv8 network is obtained; taking the training data set as input, taking the corrosion degree of each image as output, and training the improved YOLOv8 network to obtain a corrosion identification model; and inputting a steel structure surface image acquired in real time into the corrosion identification model to obtain a corrosion identification result of the steel structure surface. According to the scheme, the accuracy and efficiency of identifying the surface corrosion of the steel structure can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly to a method, device, medium, and equipment for identifying corrosion on the surface of steel structures. Background Art

[0002] Due to the characteristics of high strength and high stability, steel structures are widely used in buildings for support. Since outdoor steel structures are long-term exposed to the external environment, they will be eroded by natural factors such as wind, rain, snow, frost, and sunlight, resulting in corrosion on the surface of the steel structure, thereby reducing the strength of the steel structure and posing potential safety hazards. Therefore, it is necessary to identify the corrosion on the surface of the steel structure and regularly maintain the corroded areas.

[0003] The traditional method for identifying corrosion on the surface of steel structures is mainly through manual identification, but the efficiency of manual identification of corrosion on the surface of steel structures is low and the accuracy is poor. With the development of artificial intelligence technology, the steel structure corrosion identification technology with computer vision as the core has developed rapidly. First, it uses an Unmanned Aerial Vehicle (UAV) to take all-round and multi-angle photos of the steel structure to be detected, and then uses image recognition algorithms to process the taken images, which can timely detect the corroded areas on the surface of the steel structure and avoid potential safety accidents. The existing algorithms for identifying corrosion on the surface of steel structures are mainly trained based on the YOLO series of algorithms. The more advanced one is YOLOv8 proposed by Ultralytics in 2023, which has relatively high accuracy and speed for the corrosion on the surface of steel structures.

[0004] However, since the YOLOv8 network occupies a large amount of computing resources during the process of identifying corrosion on the surface of steel structures, it cannot simultaneously identify a large number of steel structure surface images, and the accuracy of the YOLOv8 network in identifying images containing small target corrosion areas is insufficient. Therefore, the existing technology cannot efficiently and accurately identify the corrosion on the surface of steel structures. Summary of the Invention

[0005] Based on this, in view of the technical problem that the existing technology cannot efficiently and accurately identify the corrosion on the surface of steel structures, it is necessary to provide a method, device, medium, and equipment for identifying corrosion on the surface of steel structures.

[0006] The present invention adopts the following technical solutions: In a first aspect, the present invention provides a method for identifying corrosion on the surface of a steel structure, the method comprising: Obtaining an image dataset of the surface of the steel structure; According to the corrosion parameters of each image in the image dataset, label the corrosion degree of each image to obtain a training dataset, where the corrosion degree includes mild corrosion, moderate corrosion, severe corrosion, and extremely severe corrosion; Replace the C2f layer in the neck network of the YOLOv8 network with the FasterViT module to obtain an improved YOLOv8 network; Use the training dataset as the input and the corrosion degree of each image as the output to train the improved YOLOv8 network to obtain a corrosion recognition model; Input the steel structure surface image collected in real time into the corrosion recognition model to obtain the corrosion recognition result of the steel structure surface.

[0007] Further, the corrosion parameters specifically include: The corrosion area, corrosion color, corrosion color depth, and surface coating peeling rate of the steel structure surface.

[0008] Further, according to the corrosion parameters of each image in the image dataset, labeling the corrosion degree of each image specifically includes: For any one of the images, when the corrosion area on the steel structure surface in the any one image is the first corrosion area, the corrosion color depth is the first color depth, and the surface coating peeling rate is the first peeling rate, label the any one image as mild corrosion; When the corrosion area on the steel structure surface in the any one image is the first corrosion area, the corrosion color depth is the second color depth, and the corrosion color is light brown, label the any one image as moderate corrosion; When the corrosion area on the steel structure surface in the any one image is the second corrosion area, the corrosion color depth is the second color depth, the corrosion color is brown, and the surface coating peeling rate is the second peeling rate, label the any one image as severe corrosion; When the corrosion area on the steel structure surface in the any one image is the third corrosion area, the corrosion color depth is the third color depth, the corrosion color is black or dark red, and the surface coating peeling rate is the third peeling rate, label the any one image as extremely severe corrosion; Among them, the area sizes of the first corrosion area, the second corrosion area, and the third corrosion area increase in sequence; the color depths of the first color depth, the second color depth, and the third color depth increase in sequence; the values of the first peeling rate, the second peeling rate, and the third peeling rate increase in sequence.

[0009] Further, labeling the corrosion degree of each image specifically includes: An acquisition module for acquiring an image dataset of the steel structure surface; A labeling module, configured to label the corrosion degree of each image in the image dataset according to the corrosion parameters of each image, so as to obtain a training dataset, where the corrosion degree includes mild corrosion, moderate corrosion, severe corrosion and extremely severe corrosion; An optimization module, configured to replace the C2f layer in the neck network of the YOLOv8 network with a FasterViT module to obtain an improved YOLOv8 network; A training module, configured to use the training dataset as input and the corrosion degree of each image as output to train the improved YOLOv8 network to obtain a corrosion recognition model; A recognition module, configured to input a steel structure surface image collected in real time into the corrosion recognition model to obtain a corrosion recognition result of the steel structure surface.

[0010] The present invention provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the method for recognizing corrosion on the surface of a steel structure is implemented.

[0011] The present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method for recognizing corrosion on the surface of a steel structure is implemented.

[0012] At least one technical solution adopted by the present invention can achieve the following beneficial effects: The present invention first labels the corrosion degree of each image according to the corrosion parameters of each image in the image dataset of the steel structure surface, and replaces the C2f layer in the neck network of the YOLOv8 network with a FasterViT module to obtain an improved YOLOv8 network, and then trains a corrosion recognition model based on the improved YOLOv8 network. The FasterViT module can normalize the spatial feature application layer of the steel structure surface image, so as to more reasonably balance the accuracy of corrosion recognition and the throughput of the steel structure surface image, ensuring the accuracy of corrosion recognition on the steel structure surface and the efficiency of corrosion recognition. In the case of significant changes in the characteristics such as the size and shape of corrosion, the attention mechanism of the FasterViT module can intelligently adjust the weights of the importance of different corrosion samples, ensuring that the corrosion recognition model focuses on diverse features during training, greatly enhancing the adaptability and robustness of the corrosion recognition task from the perspective of an unmanned aerial vehicle, enabling the corrosion recognition model to more effectively capture and learn the corrosion features contained in the steel structure surface image in a complex scene, thereby improving the accuracy of the corrosion recognition model for recognizing corrosion on the steel structure surface. Therefore, this method can improve the accuracy and efficiency of recognizing corrosion on the steel structure surface. Description of the Drawings

[0013] The accompanying drawings described herein are used to provide a further understanding of the present invention and form a part of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0014] Figure 1 It is a flowchart of a method for identifying surface corrosion of a steel structure provided by the present invention; Figure 2 It is an image of the surface of a steel structure taken by a drone provided by the present invention; Figure 3 It is a schematic diagram of marking the corrosion degree of the surface image of a steel structure provided by the present invention; Figure 4 It is a structural diagram of the YOLOv8 network provided by the present invention; Figure 5 It is a structural diagram of each module in the YOLOv8 network provided by the present invention; Figure 6 It is a structural diagram of the FasterViT module provided by the present invention; Figure 7 It is a structural diagram of the improved YOLOv8 network provided by the present invention; Figure 8 It is a schematic diagram of a device for identifying surface corrosion of a steel structure provided by the present invention; Figure 9 It is a schematic diagram of a computer device for implementing a method for identifying surface corrosion of a steel structure provided by the present invention. Detailed Embodiments

[0015] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0016] Currently, the server mentioned in the present invention can be a server set up on a business platform or a device such as a desktop computer or a laptop computer that can execute the solution of the present invention. For the convenience of description, only the server is used as the execution entity for description below. The technical solutions provided by each embodiment of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Refer to Figure 1 , a method for identifying surface corrosion of a steel structure in the present invention specifically includes the following steps: S10: Obtain an image dataset of the surface of the steel structure.

[0018] In this embodiment, images of the steel structure surface are captured by a drone to obtain an image dataset covering different angles and various environmental interference factors.

[0019] Preferably, after obtaining the image dataset of the steel structure surface, all the images in the image dataset are preprocessed. The preprocessing method is as follows: all the images in the image dataset are uniformly processed into 360 pixels × 360 pixels, and images with blurred pixels or irrelevant to corrosion are screened out manually.

[0020] Schematically, referring to Figure 2 , Figure 2 in (a) of Figure 2 in (b) of Figure 2 in (c) of Figure 2 and (d) of

[0021] are images of the steel structure surface at different angles captured by the drone.

[0022] S20: According to the corrosion parameters of each image in the image dataset, each image is labeled with the corrosion degree to obtain a training dataset. The corrosion degrees include mild corrosion, moderate corrosion, severe corrosion, and extremely severe corrosion.

[0023] In this embodiment, the method of labeling the corrosion degree of each image is as follows: The corrosion degree of each image is labeled by the labelimg tool.

[0024] Schematically, referring to Figure 3 , Figure 3 in (a) of Figure 3 and (b) of Figure 3 are the results of labeling the steel structure surface images by the lableimg tool. Among them, the area enclosed by the white rectangular frame is the corrosion area. Figure 3Among them, (b) contains four areas of moderate corrosion and one area of severe corrosion. After annotation, the corresponding annotation coordinates are obtained and are in one-to-one correspondence with the corrosion images. The label images obtained by this data annotation method have high annotation accuracy, and the annotation process avoids complex edge corrosion areas and dense pitting areas. After annotation, the corrosion images and annotation coordinates are combined into a complete corrosion image dataset

[0025] S30: Replace the C2f layer in the neck network of the YOLOv8 network with the FasterViT module to obtain an improved YOLOv8 network

[0026] In this embodiment, the C2f layer in the neck network of the YOLOv8 network is replaced with the FasterViT module to obtain an improved YOLOv8 network, namely the FasterViT-YOLOv8 network. Refer to Figure 4 , the YOLOv8 network includes a backbone network, a neck network, and a head network. Among them, the backbone network consists of Conv layers, C2f layers, and SPPF layers, the neck network consists of C2f layers, a concatenation layer Concat, an upsampling layer Upsample, and a convolutional layer Conv, and the head network is three detection heads (Detect). As Figure 4 shown, input the image containing steel structure corrosion into the YOLOv8 network, and the detection result is that there are three areas of moderate corrosion and one area of severe corrosion in the steel structure

[0027] YOLOv8 effectively extracts multi-level feature information in images through densely stacked convolutional layers. At the same time, in order to alleviate the problem of gradient disappearance or explosion during deep network training and further improve the network's representation ability and performance, a residual connection mechanism is incorporated. This mechanism helps to reduce the effective depth of the model, thereby improving the training efficiency and final performance of the network without significantly increasing the computational complexity. YOLOv8 adopts a new backbone network and detection heads, achieving a high detection speed while maintaining high accuracy, and is suitable for real-time applications. It uses techniques such as depthwise separable convolutions, Feature Pyramid Networks (FPN), and Path Aggregation Network (PAN) to reduce the computational amount and improve the computational efficiency, and shows high robustness and stability in various complex scenarios. Compared with the previous versions in this series, YOLOv8 has higher object detection capabilities

[0028] Specifically, refer toFigure 5 The C2f layer is composed of a sequentially connected Conv layer, Split layer, Bottleneck layer, Bottleneck layer, Concat layer, and Conv layer. The Split layer, Bottleneck layer, Bottleneck layer, and Concat layer are respectively connected to the Concat layer. The Bottleneck layer is composed of two sequentially connected Conv layers. The Conv layer is composed of a sequentially connected Conv2d layer, BatchNorm2d layer, and SILU layer. The SPPF layer is composed of a sequentially connected Conv layer, Maxpool2d layer, Maxpool2d layer, Maxpool2d layer, Concat layer, and Conv layer. The outputs of the Conv layer and Maxpool2d layer are respectively connected to the Concat layer.

[0029] Reference Figure 6 The FasterViT (Fast Vision Transformers with Hierarchical Attention) module is composed of a sequentially connected input layer, Conv layer, Conv layer, Downsample layer, ConvBlock layer, Downsample layer, ConvBlock layer, Downsample layer, Hieraechical Attention layer, Downsample layer, Hieraechical Attention layer, and Head layer. After the outputs of the latter two Downsample layers are processed by the CTinit layer, they are input into the Hieraechical Attention layer again.

[0030] FasterViT combines the local feature learning characteristics of CNN and the global modeling characteristics of ViT, and introduces the Hierarchical Attention (HAT) method to increase the interaction between windows while reducing the computational cost. In various CV tasks including classification, object detection, and segmentation, FasterViT is faster in terms of accuracy and image throughput, and HAT can be used as a plug-and-play enhancement module. FasterViT consists of four different stages. Residual convolutional blocks are used in the high-resolution stage, and Transformer blocks are used in subsequent stages. The input image resolution is reduced and the number of channels is doubled through stride convolutional layers between stages, and hierarchical attention blocks are used to extract long- and short-distance spatial relationships for effective cross-window interaction.

[0031] Specifically, the schematic diagram of using the FasterViT module to replace the C2f layer in the neck network of the YOLOv8 network is as Figure 7 shown: The backbone of YOLOv8 generates feature maps of various resolutions through a series of convolutional operations. These feature maps have their own characteristics and cover visual information from fine details to the overall view. Subsequently, these feature maps are fed into the neck network for in-depth feature fusion and fine processing.

[0032] Input image Through two consecutive 3×3 convolutional layers, with a stride of 2 for each convolutional layer, and the GELU activation function is used after each convolution. Then FasterViT applies layer normalization to the spatial features and uses a convolutional layer with a kernel of 3×3 and a stride of 2 to reduce the spatial resolution by 2 times. The specific expression is: = GELU ( BN ( ( x ))) x = BN ( ( ))+ x ) Where, x represents the input value, represents the convolutional layer with a kernel of 3×3 and a stride of 2, BN represents batch normalization, GELU is the activation function, represents the output feature value.

[0033] A novel window attention module in FasterViT. The core is to introduce carrier tokens (CTs) on the local window of Swin Transformer to summarize the information of the local window. Subsequently, information interaction between local windows based on CTs has higher-order carrier tokens. When given an output feature map , where H , W and d represent the height, width, and number of the feature map. When setting H = W , in order to n = divide the input feature map into n × n local windows. The expression is:

[0034] = ( x ) Where, represents the output feature value, A function that divides the input feature map in a specific dimension, k refers to the window size.

[0035] By specifying carrier tokens (CTs), the cost is reduced to obtain an attention footprint much larger than the local window. First, the CTs are initialized by pooling the l2c tokens of each window. The initialization expression is as follows: = ( x ) = ( ) where is the positional encoding for valid positions, and represent the carrier token and the feature pooling operation respectively.

[0036] Inside each HAT module, the CTs go through the attention process. The expression for the attention process is: = + ( LN ( )) = + ( LN ( )) where LN represents layer normalization, MHSA represents multi-head self-attention, γ is a learnable channel-specific scaling factor, is a two-layer MLP structure with the GELU activation function.

[0037] To model long- and short-range spatio-temporal information, information interaction is needed between the local window and the CTs. Each local window can only access the corresponding CTs. The expression for the information interaction is as follows: = ( ) After another set of attention processes: = + ( LN ( )) = + ( LN ( )) Finally, the obtained tokens above are further split back into local features and CTs for subsequent use in the HAT module: = Split ( ) Through this embodiment, applying layer normalization to spatial features using FasterViT can more accurately measure the performance of object detection and achieve a better trade-off between accuracy and throughput. In the face of significant changes in features such as object size and shape, the attention mechanism can intelligently adjust the importance weights of different samples, ensuring that the corrosion recognition model focuses on diverse features during training. This feature greatly enhances the adaptability and robustness of the object detection task from the UAV perspective, enabling the corrosion recognition model to more effectively capture and learn object features in complex scenes, thereby improving the overall detection performance.

[0038] S40: Use the training dataset as input and the corrosion degree of each image as output to train the improved YOLOv8 network to obtain a corrosion recognition model.

[0039] In this embodiment, after obtaining the corrosion recognition model, use the validation set to verify the corrosion degree of the steel structure surface output by the corrosion recognition model, and use the test set to test the accuracy of the corrosion recognition model in identifying the corrosion degree of the steel structure surface, and adjust the parameters of the corrosion recognition model according to the accuracy. This embodiment uses the FasterViT model to optimize the YOLOv8 network to obtain an improved YOLOv8 network. The corrosion recognition model trained using the improved YOLOv8 network has improved the accuracy of identifying the corrosion of the steel structure surface by 9.0%, 8.4%, and 7.7% compared to YOLOv5, YOLOv6, and YOLOv7 respectively, and has improved by 10.9%, 10.7%, and 5.0% respectively in the mPA@0.5 metric, significantly improving the accuracy of identifying the corrosion of the steel structure surface. Applying the improved YOLOv8 network to the corrosion detection practice of the external grid structure of the training hall, the accurate recognition rate of steel pipe corrosion has been successfully achieved at 92.6%.

[0040] S50: Input the real-time collected images of the steel structure surface into the corrosion recognition model to obtain the corrosion recognition result of the steel structure surface.

[0041] Based on Figure 1A method for identifying corrosion on the surface of a steel structure is shown. First, according to the corrosion parameters of each image in the image dataset of the steel structure surface, each image is labeled with the corrosion degree. And the C2f layer in the neck network of the YOLOv8 network is replaced with the FasterViT module to obtain an improved YOLOv8 network. Then, a corrosion recognition model is trained based on the improved YOLOv8 network. It can normalize the spatial feature application layer of the steel structure surface image by using the FasterViT module, so as to more reasonably balance the accuracy of corrosion recognition and the throughput of the steel structure surface image, ensuring the accuracy of corrosion recognition on the steel structure surface and the efficiency of corrosion recognition. In the case of significant changes in the characteristics such as the size and shape of corrosion, the attention mechanism of the FasterViT module can intelligently adjust the weights of the importance of different corrosion samples, ensuring that the corrosion recognition model focuses on diverse features during training, greatly enhancing the adaptability and robustness of the corrosion recognition task from the perspective of the drone, enabling the corrosion recognition model to more effectively capture and learn the corrosion features contained in the steel structure surface image in complex scenarios, thereby improving the accuracy of the corrosion recognition model for identifying corrosion on the steel structure surface. Therefore, this method can improve the accuracy and efficiency of identifying corrosion on the steel structure surface.

[0042] When applying a method for identifying corrosion on the surface of a steel structure provided by the present invention, it is not necessary to Figure 1 execute according to the order of the steps shown. The specific execution order of each step can be determined according to needs, and the present invention does not limit this.

[0043] In addition, in one or more embodiments of the present invention, the corrosion parameters specifically include: The corrosion area, corrosion color, corrosion color depth, and surface coating peeling rate of the steel structure surface.

[0044] In this embodiment, different corrosion degrees on the steel structure surface correspond to different corrosion areas, different corrosion colors, different corrosion color depths, and different surface coating peeling rates. Therefore, each image can be labeled with the corrosion degree according to the corrosion area, corrosion color, corrosion color depth, and surface coating peeling rate of the steel structure surface in each image.

[0045] Schematically, Table 1 shows the corrosion degree classification standard of the steel structure surface provided in this embodiment: Table 1 Corrosion Degree Classification Standard Specifically, in one or more embodiments of the present invention, labeling each image with the corrosion degree according to the corrosion parameters of each image in the image dataset specifically includes: For any one of the images, when the corrosion area on the steel structure surface in any one of the images is the first corrosion area, the corrosion color depth is the first color depth, and the surface coating peeling rate is the first peeling rate, mark any one of the images as slightly corroded.

[0046] When the corrosion area on the steel structure surface in any one of the images is the first corrosion area, the corrosion color depth is the second color depth, and the corrosion color is light brown, mark any one of the images as moderately corroded.

[0047] When the corrosion area on the steel structure surface in any one of the images is the second corrosion area, the corrosion color depth is the second color depth, the corrosion color is brown, and the surface coating peeling rate is the second peeling rate, mark any one of the images as severely corroded.

[0048] When the corrosion area on the steel structure surface in any one of the images is the third corrosion area, the corrosion color depth is the third color depth, the corrosion color is black or dark red, and the surface coating peeling rate is the third peeling rate, mark any one of the images as extremely severely corroded.

[0049] Among them, the area sizes of the first corrosion area, the second corrosion area, and the third corrosion area increase in sequence; the color depths of the first color depth, the second color depth, and the third color depth increase in sequence; the values of the first peeling rate, the second peeling rate, and the third peeling rate increase in sequence.

[0050] In this embodiment, the first corrosion area refers to a relatively small corrosion area, such as less than 25 square centimeters; the second corrosion area refers to a medium-sized corrosion area, such as 25 square centimeters to 100 square centimeters; the third corrosion area refers to a relatively large corrosion area, such as greater than 100 square centimeters. The first color depth refers to a relatively light color depth, such as a color depth with a gray value less than 50; the second color depth refers to a medium color depth, such as a color depth with a gray value between 50 and 100; the third color depth refers to a relatively deep color depth, such as a color depth with a gray value greater than 100. The first peeling rate refers to a relatively low surface coating peeling ratio, such as less than or equal to 5%; the second peeling rate refers to a medium surface coating peeling ratio, such as greater than 5% and less than 20%; the third peeling rate refers to a relatively high surface coating peeling ratio, such as greater than or equal to 20%.

[0051] The solution shown in this embodiment marks the corrosion degree of each image according to the corrosion area, corrosion color, corrosion color depth, and surface coating peeling rate of the steel structure surface in each image, so that the corrosion recognition model can not only identify the corroded steel structure surface, but also accurately identify the severity of the corrosion on the steel structure surface, improving the fineness of the corrosion recognition model for the corrosion recognition of the steel structure surface.

[0052] The above is a method for identifying corrosion on the surface of a steel structure provided by one or more embodiments of the present invention. Based on the same idea, the present invention also provides a corresponding device for identifying corrosion on the surface of a steel structure, as Figure 8 shown, including: An acquisition module, configured to acquire an image data set of the surface of the steel structure.

[0053] A labeling module, configured to label the corrosion degree of each image in the image data set according to the corrosion parameters of each image in the image data set, so as to obtain a training data set, and the corrosion degree includes mild corrosion, moderate corrosion, severe corrosion, and extremely severe corrosion.

[0054] An optimization module, configured to replace the C2f layer in the neck network of the YOLOv8 network with the FasterViT module to obtain an improved YOLOv8 network.

[0055] A training module, configured to use the training data set as input and the corrosion degree of each image as output to train the improved YOLOv8 network to obtain a corrosion recognition model.

[0056] An identification module, configured to input the image of the surface of the steel structure collected in real time into the corrosion recognition model to obtain the corrosion recognition result of the surface of the steel structure.

[0057] For the specific limitations of a device for identifying corrosion on the surface of a steel structure, reference can be made to the limitations of a method for identifying corrosion on the surface of a steel structure in the above text, which will not be elaborated here. Each module in the device for identifying corrosion on the surface of a steel structure can be implemented in whole or in part by software, hardware, and their combination. Each module can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0058] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program can be used to execute the Figure 1 method for identifying corrosion on the surface of a steel structure provided above.

[0059] The present invention also provides Figure 9 a schematic structural diagram of the computer device shown in Figure 9 shown. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include other hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the Figure 1 method for identifying corrosion on the surface of a steel structure provided above.

[0060] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the described embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the various methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided by the present invention can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0061] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded by the present invention.

Claims

1. A method for identifying surface corrosion of a steel structure, characterized in that: include: Acquire an image dataset of the steel structure surface; According to the corrosion parameters of each image in the image data set, the corrosion degree of each image is labeled to obtain a training data set, wherein the corrosion degree includes light corrosion, moderate corrosion, heavy corrosion and severe corrosion; Use the FasterViT module to replace the C2f layer in the neck network of the YOLOv8 network to obtain an improved YOLOv8 network; Taking the training data set as input and the corrosion degree of each image as output, the improved YOLOv8 network is trained to obtain a corrosion recognition model; The steel structure surface image collected in real time is input into the corrosion recognition model to obtain the corrosion recognition result of the steel structure surface.

2. A method for identifying surface corrosion of a steel structure according to claim 1, characterized in that: The corrosion parameters specifically include: The corrosion area, corrosion color, corrosion color depth and surface coating peeling rate of the steel structure surface.

3. A method for identifying surface corrosion of a steel structure as claimed in claim 2, characterized in that: According to the corrosion parameters of each image in the image data set, the corrosion degree of each image is marked, specifically including: For any image among the images, when the corrosion area of ​​the steel structure surface in the image is a first corrosion area, the corrosion color depth is a first color depth, and the surface coating peeling rate is a first peeling rate, the image is marked as slightly corroded; When the corrosion area of ​​the steel structure surface in any one of the images is the first corrosion area, the corrosion color depth is the second color depth, and the corrosion color is light brown, marking any one of the images as moderately corroded; When the corrosion area of ​​the steel structure surface in any one of the images is the second corrosion area, the corrosion color depth is the second color depth, the corrosion color is brown, and the surface coating peeling rate is the second peeling rate, marking any one of the images as severely corroded; When the corrosion area of ​​the steel structure surface in any one of the images is the third corrosion area, the corrosion color depth is the third color depth, the corrosion color is black or dark red, and the surface coating peeling rate is the third peeling rate, marking any one of the images as severely corroded; Among them, the sizes of the first corrosion area, the second corrosion area and the third corrosion area increase successively; the color depths of the first color depth, the second color depth and the third color depth increase successively; and the values ​​of the first peeling rate, the second peeling rate and the third peeling rate increase successively.

4. A method for identifying surface corrosion of a steel structure according to any one of claims 1 to 3, characterized in that: The corrosion degree of each image is marked, specifically including: The corrosion degree of each image is labeled using the labelimg tool.

5. A device for identifying surface corrosion of a steel structure, characterized in that: include: An acquisition module, used for acquiring an image data set of the steel structure surface; A labeling module, used for labeling the corrosion degree of each image in the image data set according to the corrosion parameters of each image, so as to obtain a training data set, wherein the corrosion degree includes light corrosion, moderate corrosion, heavy corrosion and severe corrosion; An optimization module is used to replace the C2f layer in the neck network of the YOLOv8 network with the FasterViT module to obtain an improved YOLOv8 network; A training module, used for taking the training data set as input and the corrosion degree of each image as output, training the improved YOLOv8 network to obtain a corrosion recognition model; The recognition module is used to input the real-time collected steel structure surface image into the corrosion recognition model to obtain the corrosion recognition result of the steel structure surface.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, a method for identifying surface corrosion of a steel structure as described in any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, a method for identifying surface corrosion of a steel structure as claimed in any one of claims 1 to 4 is implemented.

Images