A method and system for detecting corrosion on a surface of a large device

Abstract

The application discloses a large device surface corrosion detection method and system, and aims to solve the technical problem of poor accuracy of target detection caused by the existing large device surface corrosion detection technology. The method comprises the following steps: acquiring a large device corrosion image to be detected, and inputting the large device corrosion image to be detected into a preset large device surface corrosion detection model. The preset large device surface corrosion detection model comprises a fusion attention feature extraction network, a target neck network and a target detection head. The fusion attention feature extraction network is used for performing feature strengthening extraction on the large device corrosion image to be detected, so as to generate a strengthened feature map. The target neck network is used for performing feature convolution fusion on the strengthened feature map, so as to output a fusion feature map. The fusion feature map is input into the target detection head for target detection, and a corrosion detection result is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a large device surface corrosion detection method and system. BACKGROUND

[0002] Corrosion morphology image features are the basis for corrosion morphology recognition and classification, including gray (density, color) features, texture features, and corrosion shape features, such as the distribution of corrosion points, the size of corrosion points, and corrosion area, etc. The recognition of corrosion morphology features is a difficulty in corrosion image processing and quantitative evaluation.

[0003] With the progress of machine learning and machine vision technology, a large number of studies have been conducted on structure corrosion and coating defect detection and recognition based on vision, and traditional machine learning algorithms mainly including decision tree, random forest, support vector machine (SVM) have made great progress. Huynh et al. proposed a multi-class support vector machine (SVM) model for evaluating coating condition, and its efficient rust damage recognition performance has been verified by high-definition images collected from Sydney Harbor Bridge.

[0004] Most of the existing large device surface corrosion detection technologies decompose images by means of continuous wavelet transform, and then extract energy values as feature information for further analysis of the correlation between image feature values and corrosion weight loss data. However, in the face of severe corrosion or large noise images, the detailed features of the corrosion image are easily disturbed and become blurred, which will make the key information in the image unable to be effectively recognized, resulting in poor accuracy of target detection. SUMMARY

[0005] The present application provides a large device surface corrosion detection method and system, which solves the technical problem of poor accuracy of target detection caused by the existing large device surface corrosion detection technology.

[0006] The first aspect of the present application provides a large device surface corrosion detection method, comprising:

[0007] Obtaining a large device corrosion image to be detected, and inputting the large device corrosion image to be detected into a preset large device surface corrosion detection model, wherein the preset large device surface corrosion detection model comprises a fusion attention feature extraction network, a target neck network, and a target detection head;

[0008] The fusion attention feature extraction network is used to perform feature enhancement extraction on the large device corrosion image to be detected, to generate an enhanced feature map;

[0009] The target neck network is used to perform feature convolution fusion on the enhanced feature map, to output a fused feature map;

[0010] inputting the fusion feature map into the target detection head for target detection, and outputting a corrosion detection result.

[0011] Optionally, the fusion attention feature extraction network comprises a convolution fusion attention mechanism module and a multi-scale feature fusion module; the reinforced feature map comprises a first reinforced feature map, a second reinforced feature map and a third reinforced feature map; the feature reinforcement extraction of the corrosion image of the large-scale device to be detected by using the fusion attention feature extraction network comprises:

[0012] convolution fusion of the corrosion image of the large-scale device to be detected by using the convolution fusion attention mechanism module to generate a first convolution fusion feature map;

[0013] convolution fusion of the first convolution fusion feature map by the convolution fusion attention mechanism module to output a first reinforced feature map;

[0014] inputting the first reinforced feature map into the convolution fusion attention mechanism module for convolution fusion to determine a second reinforced feature map;

[0015] convolution fusion of the second reinforced feature map by using the convolution fusion attention mechanism module to output a second convolution fusion feature map;

[0016] multi-scale feature fusion of the second convolution fusion feature map by the multi-scale feature fusion module to generate a third reinforced feature map.

[0017] Optionally, the convolution fusion attention mechanism module comprises two convolution layers connected in series, a convolution block attention mechanism layer, a splicing layer and a convolution layer; the image processing steps of the convolution fusion attention mechanism module comprise:

[0018] convolution operation of the corrosion input image input into the convolution fusion attention mechanism module by the two convolution layers connected in series to generate a first convolution feature map;

[0019] segmentation of the first convolution feature map to output a first segmented feature map and a second segmented feature map;

[0020] convolution operation of the second segmented feature map by the two convolution layers connected in series to generate a second convolution feature map;

[0021] feature weight adjustment of the second convolution feature map by the convolution block attention mechanism layer to output a first convolution block attention feature map;

[0022] The first convolutional block attention feature map and the second segmentation feature map are spliced to generate a first spliced feature map, and the first spliced feature map is input into two convolutional layers in series for convolution operation to output a third convolutional feature map;

[0023] The third convolutional feature map is subjected to feature weight adjustment by using a convolutional block attention mechanism layer to generate a second convolutional block attention feature map;

[0024] The second convolutional block attention feature map and the first spliced feature map are spliced to output a second spliced feature map;

[0025] The first segmentation feature map, the first convolutional block attention feature map, the second spliced feature map, and the second segmentation feature map are spliced to generate a third spliced feature map;

[0026] The third spliced feature map is subjected to convolution operation by a convolutional layer to output an erosion output map.

[0027] Optionally, the multi-scale feature fusion module includes three cascaded average pooling layers, three cascaded maximum pooling layers, a splicing layer, and a convolutional layer; the multi-scale feature fusion of the second convolutional fusion feature map by the multi-scale feature fusion module to generate a third enhanced feature map includes:

[0028] The second convolutional fusion feature map is subjected to convolution operation by a convolutional layer to generate an enhanced convolutional feature map;

[0029] The enhanced convolutional feature map is respectively input into three cascaded average pooling layers and three cascaded maximum pooling layers for average pooling and maximum pooling to output an average pooling feature map and a maximum pooling feature map;

[0030] The maximum pooling feature map, the average pooling feature map, and the enhanced convolutional feature map are spliced by using a splicing layer to output a spliced pooling feature map;

[0031] The spliced pooling feature map is subjected to convolution operation by a convolutional layer to generate a third enhanced feature map.

[0032] Optionally, the target neck network includes a partial convolutional fusion module, a splicing layer, an up-sampling layer, and a convolutional layer; the fusion feature map includes a first fusion feature map, a second fusion feature map, and a third fusion feature map; the feature convolution fusion of the enhanced feature map by the target neck network to output a fusion feature map includes:

[0033] The third enhanced feature map is subjected to up-sampling by using an up-sampling layer to output a first up-sampled feature map, and the first up-sampled feature map and the second enhanced feature map are spliced by using a splicing layer to generate a first intermediate spliced feature map;

[0034] inputting the first intermediate splicing feature map into a partial convolution fusion module for partial convolution fusion, outputting a first partial convolution fusion feature map, and using an up-sampling layer to up-sample the first partial convolution fusion feature map to generate a second up-sampling feature map;

[0035] splicing the second up-sampling feature map and the first reinforced feature map through a splicing layer to output a second intermediate splicing feature map, inputting the second intermediate splicing feature map into a partial convolution fusion module for partial convolution fusion, and outputting a first fusion feature map;

[0036] convolving the first fusion feature map using a convolution layer to generate a first fusion convolution feature map, and inputting the first fusion convolution feature map and the first partial convolution fusion feature map into a splicing layer for splicing to output a third intermediate splicing feature map;

[0037] generating a second fusion feature map by performing partial convolution fusion on the third intermediate splicing feature map through a partial convolution fusion module, and outputting a second fusion convolution feature map by performing convolution operation on the second fusion feature map using a convolution layer;

[0038] inputting the second fusion convolution feature map and the third reinforced feature map into a splicing layer for splicing to output a fourth intermediate splicing feature map, and generating a third fusion feature map by performing partial convolution fusion on the fourth intermediate splicing feature map through a partial convolution fusion module.

[0039] Optionally, the training process of the preset large device surface corrosion detection model comprises:

[0040] obtaining a large device corrosion image data set to be trained, and labeling the large device corrosion image data set to be trained to generate a labeled corrosion image data set;

[0041] performing image combination on the labeled corrosion image data set using a preset mosaic method to generate a plurality of combined corrosion images;

[0042] respectively cutting and adjusting each of the combined corrosion images to generate a plurality of adjusted corrosion images;

[0043] using a plurality of the adjusted corrosion images to construct a target corrosion image data set;

[0044] using the target corrosion image data set to train an initial large device surface corrosion detection model to determine the preset large device surface corrosion detection model that has been trained.

[0045] The second aspect of the present application provides a large device surface corrosion detection system, comprising:

[0046] An acquisition module is configured to acquire a large device corrosion image to be detected and input the large device corrosion image to be detected into a preset large device surface corrosion detection model, the preset large device surface corrosion detection model comprising a fusion attention feature extraction network, a target neck network and a target detection head;

[0047] A feature extraction module is configured to perform feature enhancement extraction on the large device corrosion image to be detected by using the fusion attention feature extraction network, to generate an enhanced feature map;

[0048] A convolution fusion module is configured to perform feature convolution fusion on the enhanced feature map by using the target neck network, to output a fused feature map;

[0049] A target detection module is configured to input the fused feature map into the target detection head to perform target detection, and output a corrosion detection result.

[0050] The third aspect of the present application provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the steps of the large device surface corrosion detection method according to any one of the above aspects.

[0051] The fourth aspect of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed to implement the steps of the large device surface corrosion detection method according to any one of the above aspects.

[0052] The fifth aspect of the present application provides a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the large device surface corrosion detection method according to any one of the above aspects.

[0053] From the above technical solutions, the present application has the following advantages:

[0054] The technical scheme of the present application provides a large device surface corrosion detection method, first, the corrosion image of the large device to be detected is obtained, and the corrosion image of the large device to be detected is input into the preset large device surface corrosion detection model, the preset large device surface corrosion detection model includes a fusion attention feature extraction network, a target neck network and a target detection head; then, the fusion attention feature extraction network is used to perform feature enhancement extraction on the corrosion image of the large device to be detected, and an enhanced feature map is generated; the enhanced feature map is subjected to feature convolution fusion through the target neck network, and a fused feature map is output; finally, the fused feature map is input into the target detection head for target detection, and a corrosion detection result is output; based on the above scheme, after the corrosion image of the large device to be detected is obtained and subjected to feature enhancement extraction and feature convolution fusion by the fusion attention feature extraction network and the target neck network, the fused feature map obtained is input into the target detection head for target detection, and the corrosion detection result is output, which can automatically focus on the salient region in the image, enhance the capture ability of the key information, and thus improve the accuracy of target detection. BRIEF DESCRIPTION OF DRAWINGS

[0055] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0056] Figure 1 A step flow chart of a large device surface corrosion detection method provided for the first embodiment of the present application;

[0057] Figure 2 A structure schematic diagram of a preset large device surface corrosion detection model provided for the first embodiment of the present application;

[0058] Figure 3 A structure schematic diagram of a C2f-CBAM module provided for the first embodiment of the present application;

[0059] Figure 4 A structure schematic diagram of a multi-scale feature fusion module provided for the first embodiment of the present application;

[0060] Figure 5 A model training step flow chart of a preset large device surface corrosion detection model provided for the second embodiment of the present application;

[0061] Figure 6 A model training flow chart of a preset large device surface corrosion detection model provided for the second embodiment of the present application;

[0062] Figure 7A structural block diagram of a large device surface corrosion detection system according to an embodiment of the present application is provided. DETAILED DESCRIPTION

[0063] The embodiment of the present application provides a large device surface corrosion detection method and system, and aims to solve the technical problem of poor accuracy of target detection caused by the existing large device surface corrosion detection technology.

[0064] In order to make the purpose, characteristics and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the following described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0065] Please refer to Figure 1 , Figure 1 A step flowchart of a large device surface corrosion detection method according to the first embodiment of the present application is provided.

[0066] The large device surface corrosion detection method provided by the present application comprises the following steps:

[0067] In step 101, a large device corrosion image to be detected is acquired, and the large device corrosion image to be detected is input into a preset large device surface corrosion detection model. The preset large device surface corrosion detection model comprises a fusion attention feature extraction network, a target neck network and a target detection head.

[0068] It should be noted that the CBAM (Convolutional Block Attention Module) attention mechanism is introduced into the C2f (Cross Stage Partial Bottleneck with 2 Convolutions) model in the YOLOv8 model in the present application, the channel and spatial attention bias of the corrosion topography feature information is increased, the corresponding weight matrix is added to the target feature map through the fusion attention mechanism, and fine features are realized. On the basis of the traditional SPPF (Spatial Pyramid Pooling Fast) module of the YOLOv8 model, an average pooling layer is added, so that the feature information can be fused and integrated at different spatial scales. Compared with single-scale feature extraction, multi-scale feature fusion can better process targets of different sizes, shapes and positions in the image, so as to construct a target detection model based on the improved YOLOv8, i.e. the preset large device surface corrosion detection model.

[0069] Please refer to Figure 2The preset large device surface corrosion detection model comprises a fusion attention feature extraction network, a target neck network and a target detection head; the fusion attention feature extraction network is composed of four convolution fusion attention mechanism modules and a multi-scale feature fusion module; the convolution fusion attention mechanism module is composed of a convolution layer and a C2f-CBAM module connected in series; and the target neck network is composed of four partial convolution fusion modules (C2f), four splicing layers (Concat), two up-sampling layers (Upsample) and two convolution layers.

[0070] In step 102, the fusion attention feature extraction network is used to perform feature enhancement extraction on the large device corrosion image to be detected, and an enhanced feature map is generated.

[0071] The fusion attention feature extraction network comprises a convolution fusion attention mechanism module and a multi-scale feature fusion module, please refer to Figure 3 The convolution fusion attention mechanism module is composed of a convolution layer and a C2f-CBAM module connected in series, and the C2f-CBAM module is composed of six convolution layers, two convolution block attention mechanism layers and three splicing layers.

[0072] Further, please refer to Figure 4 The multi-scale feature fusion module is composed of two convolution layers, three cascaded average pooling layers, three cascaded maximum pooling layers and a splicing layer.

[0073] The enhanced feature map comprises a first enhanced feature map, a second enhanced feature map and a third enhanced feature map.

[0074] Further, step 102 can comprise the following sub-steps S21-S25:

[0075] In step S21, the convolution fusion attention mechanism module is used to perform convolution fusion on the large device corrosion image to be detected, and a first convolution fusion feature map is generated.

[0076] Optionally, the convolution fusion attention mechanism module comprises two convolution layers connected in series, a convolution block attention mechanism layer, a splicing layer and a convolution layer; the image processing steps of the convolution fusion attention mechanism module comprise:

[0077] In S01, the corrosion input image input into the convolution fusion attention mechanism module is subjected to convolution operation through two convolution layers connected in series, and a first convolution feature map is generated.

[0078] In S02, the first convolution feature map is segmented, and a first segmented feature map and a second segmented feature map are output.

[0079] In S03, the second segmented feature map is subjected to convolution operation through two convolution layers connected in series, and a second convolution feature map is generated.

[0080] S04, performing feature weight adjustment on the second convolutional feature map through a convolution block attention mechanism layer, and outputting a first convolution block attention feature map;

[0081] S05, splicing the first convolution block attention feature map and the second segmentation feature map to generate a first spliced feature map, and inputting the first spliced feature map into two convolution layers in series for convolution operation, and outputting a third convolutional feature map;

[0082] S06, performing feature weight adjustment on the third convolutional feature map using a convolution block attention mechanism layer to generate a second convolution block attention feature map;

[0083] S07, splicing the second convolution block attention feature map and the first spliced feature map to output a second spliced feature map;

[0084] S08, splicing the first segmentation feature map, the first convolution block attention feature map, the second spliced feature map, and the second segmentation feature map to generate a third spliced feature map;

[0085] S09, performing convolution operation on the third spliced feature map through a convolution layer to output an erosion output map.

[0086] The erosion input map is a feature map input into the convolution fusion attention mechanism module. It can be understood that the erosion input map can correspond to any feature map input into the convolution fusion attention mechanism module for image processing in the process of model training or erosion detection.

[0087] The first convolutional feature map, the first segmentation feature map, the second segmentation feature map, the second convolutional feature map, the first convolution block attention feature map, the first spliced feature map, the third convolutional feature map, the second convolution block attention feature map, the second spliced feature map, and the third spliced feature map are all intermediate maps generated in the convolution fusion attention mechanism module.

[0088] The erosion output map is a feature map output by the convolution fusion attention mechanism module. It can be understood that it can correspond to any output feature map after the convolution fusion attention mechanism module processes the image in the process of model training or detection.

[0089] It should be noted that the C2f module is a component in the YOLOv8 model used for feature fusion, which functions to fuse feature maps between different stages to obtain higher quality feature representations. This cross-stage fusion helps improve the performance of object detection. CBAM (Convolutional Block Attention Mechanism) is an attention mechanism used to enhance the performance of convolutional neural networks. The convolutional block attention mechanism layer consists of two parts: channel attention and spatial attention. The channel attention module is used to adaptively adjust the feature map weights of different channels to highlight important features. The spatial attention module is used to adaptively adjust the feature map weights of different spatial positions to improve the clarity of the feature map.

[0090] Based on the above, the present application combines C2f and CBAM attention mechanisms. The C2f-CBAM module takes the C2f fused feature map as input and further enhances the representation ability of the feature map through the CBAM attention mechanism. Therefore, replacing the C2f module of the Backbone backbone network in YOLOv8 with the C2f-CBAM module helps improve the performance of the object detection model, making it more effective in capturing important features in corrosion images and improving the accuracy of object detection.

[0091] Step S22, convolve the first convolutional fusion feature map through the convolutional fusion attention mechanism module to output a first enhanced feature map;

[0092] Step S23, input the first enhanced feature map into the convolutional fusion attention mechanism module for convolutional fusion to determine a second enhanced feature map;

[0093] Step S24, convolve the second enhanced feature map using the convolutional fusion attention mechanism module to output a second convolutional fusion feature map;

[0094] Step S25, perform multi-scale feature fusion on the second convolutional fusion feature map through a multi-scale feature fusion module to generate a third enhanced feature map.

[0095] The multi-scale feature fusion module includes three cascaded average pooling layers, three cascaded maximum pooling layers, a concatenation layer, and a convolutional layer.

[0096] Further, step S25 can include the following sub-steps S251-S254:

[0097] Step S251, perform convolutional operation on the second convolutional fusion feature map through the convolutional layer to generate an enhanced convolutional feature map;

[0098] Step S252, input the enhanced convolution feature map into three cascaded average pooling layers and three cascaded maximum pooling layers respectively for average pooling and maximum pooling, and output average pooling feature map and maximum pooling feature map;

[0099] Step S253, using a concatenation layer to concatenate the maximum pooling feature map, the average pooling feature map and the enhanced convolution feature map, and output a concatenated pooling feature map;

[0100] Step S254, performing convolution operation on the concatenated pooling feature map through a convolution layer to generate a third enhanced feature map.

[0101] It should be noted that SPPF is different from the traditional SPP which uses multiple different size pooling kernels. SPPF usually uses multiple small pooling layers (such as three 3x3 maximum pooling) to replace the pooling operation of a single large kernel. This strategy reduces the amount of calculation while maintaining the ability of multi-scale feature fusion. After the fast pooling operation, the SPPF layer concatenates the outputs of different pooling layers to form a fixed-length feature vector. This feature vector can then be used as the input of the subsequent fully connected layer or other network layers.

[0102] Further, in order to further improve the feature extraction capability of the SPPF module, the present application proposes a multi-scale feature fusion scheme. The core idea of this scheme is to capture the feature information of the image at different scales to enhance the network's perception ability of multi-scale targets. Specifically, based on the traditional SPPF, an average pooling layer is added, so that the feature information can be fused and integrated at different spatial scales. Compared with single-scale feature extraction, multi-scale feature fusion can better handle targets of different sizes, shapes and positions in the image, and improve the network's expression ability and adaptability to complex scenes.

[0103] Further, the maximum pooling extracts the maximum value from the feature map of the local receptive field. It retains the most significant features, which are usually strong responses to edges, textures or prominent target regions. The average pooling generates a smoother feature map by calculating the average of all values within the local receptive field. It retains the overall information within the local region, rather than focusing only on the maximum value. By introducing an average pooling layer, not only can key information be effectively retained, but also information loss caused by pooling operations can be reduced, thereby achieving more efficient and accurate feature representation. The combination of average pooling layer and maximum pooling layer can integrate different statistical information, further improve the model's ability to capture detailed features, and provide more rich and recognizable feature maps for downstream target detection tasks. At the same time, this scheme improves the generalization performance of the model while maintaining the computational efficiency, and is suitable for various application scenarios with limited computing resources.

[0104] Step 103, performing feature convolution fusion on the reinforced feature map through the target neck network, and outputting a fused feature map.

[0105] The target neck network comprises a partial convolution fusion module, a splicing layer, an up-sampling layer, and a convolution layer; and the fused feature map comprises a first fused feature map, a second fused feature map, and a third fused feature map.

[0106] The fused feature map comprises a first fused feature map, a second fused feature map, and a third fused feature map.

[0107] Further, step 103 can comprise the following sub-steps S31-S36:

[0108] Step S31, performing up-sampling on the third reinforced feature map through the up-sampling layer to output a first up-sampled feature map, and splicing the first up-sampled feature map and the second reinforced feature map through the splicing layer to generate a first intermediate splicing feature map;

[0109] Step S32, inputting the first intermediate splicing feature map into the partial convolution fusion module to perform partial convolution fusion, outputting a first partial convolution fused feature map, and performing up-sampling on the first partial convolution fused feature map through the up-sampling layer to generate a second up-sampled feature map;

[0110] Step S33, splicing the second up-sampled feature map and the first reinforced feature map through the splicing layer to output a second intermediate splicing feature map, inputting the second intermediate splicing feature map into the partial convolution fusion module to perform partial convolution fusion, and outputting a first fused feature map;

[0111] Step S34, performing convolution operation on the first fused feature map through the convolution layer to generate a first fused convolution feature map, and splicing the first fused convolution feature map and the first partial convolution fused feature map through the splicing layer to output a third intermediate splicing feature map;

[0112] Step S35, performing partial convolution fusion on the third intermediate splicing feature map through the partial convolution fusion module to generate a second fused feature map, and performing convolution operation on the second fused feature map through the convolution layer to output a second fused convolution feature map;

[0113] Step S36, splicing the second fused convolution feature map and the third reinforced feature map through the splicing layer to output a fourth intermediate splicing feature map, and performing partial convolution fusion on the fourth intermediate splicing feature map through the partial convolution fusion module to generate a third fused feature map.

[0114] Step 104, inputting the fused feature map into a target detection head to perform target detection, and outputting an erosion detection result.

[0115] The corrosion detection result includes a first corrosion detection result, a second corrosion detection result, and a third corrosion detection result.

[0116] It should be noted that the first fusion feature map, the second fusion feature map, and the third fusion feature map are respectively input to the corresponding target detection head for target detection, and the first corrosion detection result corresponding to the first fusion feature map, the second corrosion detection result corresponding to the second fusion feature map, and the third corrosion detection result corresponding to the third fusion feature map are output.

[0117] In the embodiment of the present application, the present application provides a large device surface corrosion detection method. First, the corrosion image of the large device to be detected is obtained, and the corrosion image of the large device to be detected is input to the preset large device surface corrosion detection model. The preset large device surface corrosion detection model includes a fusion attention feature extraction network, a target neck network, and a target detection head. Then, the fusion attention feature extraction network is used to extract the features of the corrosion image of the large device to be detected, and a strengthened feature map is generated. The strengthened feature map is convolved and fused by the target neck network, and a fused feature map is output. Finally, the fused feature map is input to the target detection head for target detection, and a corrosion detection result is output. Based on the above scheme, after the fusion attention feature extraction network and the target neck network are used to extract the features of the corrosion image of the large device to be detected and convolve and fuse the features, the fused feature map obtained is input to the target detection head for target detection, and a corrosion detection result is output. The process can automatically focus on the salient region in the image, enhance the capture ability of the key information, and improve the accuracy of target detection.

[0118] For better illustration, refer to Figure 5 , a step flow chart of model training of the preset large device surface corrosion detection model provided by the embodiment two of the present application is shown. The process can include the following steps:

[0119] Step 501, obtaining a large device corrosion image data set to be trained, and labeling the large device corrosion image data set to be trained to generate a labeled corrosion image data set.

[0120] It should be noted that the obtained image is labeled, that is, the steel corrosion part in the image is labeled, so that the new data generated subsequently has the same label information as the original data.

[0121] Step 502, using a preset mosaic method to combine the labeled corrosion image data set to generate a plurality of combined corrosion images.

[0122] It should be noted that after a series of steel corrosion pictures are collected as original data and are labeled, four pictures are randomly selected from them, and based on a preset Mosaic method, the four pictures are combined into a large Mosaic image (combined corrosion image) in a random manner. The four pictures can be placed in four different areas of the Mosaic image respectively, and it is ensured that they have certain overlapping parts, so that the model can learn the continuity between different parts.

[0123] In the embodiment, the Mosaic method is used to expand the steel corrosion picture data set, increase the diversity and complexity of the data, and improve the generalization ability and performance of the model.

[0124] Step 503, respectively, cutting and adjusting each combined corrosion image to generate a plurality of adjusted corrosion images.

[0125] It should be noted that random cutting and adjustment are performed on the generated Mosaic image, the cutting position and cutting size are randomly selected, and size adjustment, brightness and contrast adjustment and other operations are performed on the cut image to increase the diversity of the data.

[0126] Step 504, using a plurality of adjusted corrosion images to construct a target corrosion image data set.

[0127] It should be noted that the Mosaic image (adjusted corrosion image) after cutting and adjustment and the corresponding label information in the image are saved as new training data, that is, the target corrosion image data set.

[0128] Step 505, using the target corrosion image data set to train the initial large device surface corrosion detection model to determine the trained preset large device surface corrosion detection model.

[0129] It should be noted that in the target detection task, the model presets a large number of prior frames for each image. In the training process, the network updates the weight parameters according to the overlapping degree of the predicted area and the actual area, that is, the IOU of the two areas. In order to evaluate the prediction result and compare with other methods, the present application uses IOU to quantitatively evaluate the performance of the detection model (trained preset large device surface corrosion detection model). The IOU is defined as follows:

[0130] ;

[0131] Where, G T is a real boundary box (Ground Truth), that is, a labeled target box; IOU(G T , P M) To evaluate the index, different sizes can be set. When the IOU between the predicted area and the true area is greater than 0.5, the predicted frame is considered to be a true prediction, called a positive sample, otherwise a negative sample; P M is a representative predicted bounding box, that is, the detection box output by the model.

[0132] Further, in the evaluation stage, the accuracy is defined as the ratio of the number of positive samples to the number of identified samples. The recall rate is defined as the ratio of the correctly identified objects in the test data set to the total number of such objects. The precision (Precision) and recall (Recall) are defined as follows:

[0133] ;

[0134] where TP is the positive sample detection (True Positive); FN is the negative sample (False Negative) that is incorrectly detected; FP is the positive sample that is incorrectly detected; Recall is the recall rate; Precision is the precision.

[0135] Further, accuracy and recall are often contradictory and are not sufficient to measure model performance alone, so the present application introduces AP (average precision) to intuitively represent the performance of the predictor. It indicates the performance of the model on a particular class of objects. The value of AP is the area enclosed by the PR (Precision-Recall) curve and the coordinate axes. However, AP represents the accuracy of a certain class and does not measure the overall performance of the entire model. In order to comprehensively evaluate the performance of the model and compare it with mainstream models, the present application introduces mAP (mean Average Precision) as an evaluation index. mAP represents the average accuracy of each class of the model in the data set, and can well reflect the comprehensive performance of the model. AP and mAP are defined as follows:

[0136] ;

[0137] where AveP(q) is the average accuracy of the qth class; Q is the total number of classes.

[0138] Exemplarily, refer to Figure 6, based on the image acquisition device to obtain the original image set, the original image set using Mosaic method, randomly select 4 image mosaic to generate a new image, until all the pictures complete splicing, build data diversity and higher model robustness of the data set; image enhancement data set is divided into training set and validation set according to 8:2 proportion; the part of the data set is used to train the improved corrosion detection model, and the improved model is obtained, and the trained model is evaluated using the validation data set. The benchmark YOLOv8 model and the improved model are compared, and the final steel corrosion detection model is obtained.

[0139] As a comparison of technical effects, reference can be made in combination with the prior art. In the field of corrosion science, corrosion digital image processing technology was first applied to the identification and evaluation of stainless steel pitting corrosion. In 1981, D. Zhak et al. studied the corrosion image of AISI304 stainless steel in 10% FeC13 by image method. The image is binarized, and the pitting corrosion rate value of 9.73% is obtained by statistics.

[0140] Kurose et al. analyzed the distribution and development process of stress corrosion crack of pure copper by image analysis technology, and proved that fractal dimension is an effective method to evaluate the occurrence stage of copper corrosion. Fractal dimension is applied to feature extraction of corrosion morphology image. Wang Shouyan et al. took the fractal dimension, generalized dimension, area factor and void characteristics of corrosion specimen image as input, and established a corrosion morphology image diagnosis system based on analysis characteristics by fuzzy pattern recognition theory. The surface corrosion morphology of metal is judged by the fractal characteristic value of the image.

[0141] Song Shizhe, Wang Shouyan et al. decomposed the image by continuous wavelet transform and extracted the energy value as feature information, analyzed the correlation between image feature value and corrosion weight loss data, and proved that this method can effectively judge and predict the early behavior of atmospheric corrosion of pure zinc. Mohammad et al. used wavelet analysis technology to perform two-dimensional wavelet transform on the corrosion morphology image with pitting and crack, decomposed it into sub-images with various special scales and orientation information, extracted energy features, and applied neural network method to realize good corrosion morphology classification. Tao Lei et al. used wavelet image analysis technology to decompose the LC4CS and LY12CZ aluminum alloy corrosion morphology image after filtering and extract the image feature value δ of the sub-image, which is used to analyze the influence of each test factor on the test acceleration. Yan Yunhui et al. took the L1 norm, energy and entropy of each frequency band output after wavelet transform as the features of metal fracture identification and classification, and used linear minimum distance classifier to accurately identify and classify six typical fracture surfaces such as corrosion fatigue and ductile fatigue.

[0142] In recent years, with the rapid development of computer technology and the advent of the big data era, deep learning technology has begun to develop rapidly. The convolutional neural network trained by deep learning classification algorithm performs efficiently and accurately in image recognition and feature extraction. The reason is that the deep convolutional neural network adds convolutional and pooling layers compared to traditional neural networks, which can train and learn the features we need from high-dimensional output features, not only reducing the computational workload but also significantly improving the classification accuracy, making it have a unique advantage in image classification processing. The LeNet, AleXNet, GoogLeNet, VGGNet, and other convolutional neural network models that have won the ILSVRC competition have high accuracy in image classification and recognition. Wang Minghao uses deep learning methods to identify and retrieve the feature information of NiCrAlY coating / Ni-based high-temperature alloy micro-morphology images. The test set recognition accuracy of the neural network with RMS Prop optimizer and multiple convolutional layers is 98% and 90.67%, respectively. The recognition accuracy of the convolutional neural network trained with Adam optimizer is further improved to 99.17%. Wang uses an unsupervised image segmentation method based on texture to integrate a classifier based on red, green, and blue (RGB) features to build a pixel-based corrosion classifier. Users only need to use a small group of manually labeled images to build an automatic labeling tool for similar corrosion features in the image.

[0143] Therefore, as a method that has been popular in recent years and has performed well, deep learning is used as the core method for researching corrosion image detection and recognition. A corrosion detection algorithm model with good performance and application prospects is proposed. According to the current research difficulties of lack of corrosion picture data and types, the influence of insufficient data and difficult data labeling on the performance of the method model in deep learning is overcome.

[0144] YOLOv8 is a real-time object detection model that has been optimized in terms of accuracy and speed, using a more efficient network architecture that can better handle objects in complex scenarios. YOLOv8 can accurately identify and locate various types of corrosion features such as rust, peeling, and cracks, which are very important for maintenance and repair. In many industrial scenarios, corrosion may occur in complex backgrounds, and YOLOv8 performs well in handling such cases, effectively distinguishing corrosion features from background interference. However, high-quality and large amounts of labeled data are necessary for model training to ensure the accuracy of the model, especially in specific fields such as corrosion detection. And when dealing with very small corrosion features, the model may still have the risk of missing detection.

[0145] Based on the above, the existing technology in the field of corrosion image processing faces many challenges, the most pressing of which is the severe shortage of corrosion image dataset. Most of the public datasets are small in size and limited in the types and scenarios of corrosion, making it difficult to fully reflect the complexity of the actual industrial environment. Therefore, the existing datasets are not highly representative and cannot cover the diversity of different materials, corrosion types and environmental conditions, resulting in insufficient generalization ability of the model.

[0146] In addition, there is a lack of standardized datasets in the field, making it difficult for researchers to evaluate the performance of models uniformly. The lack of a unified benchmark makes it difficult to compare different methods horizontally and accurately measure their strengths and weaknesses. At the same time, due to the lack of data, the training of deep learning models is often difficult to fully carry out, resulting in lower accuracy in practical applications. Deep learning models usually rely on large amounts of data to extract features and patterns, and the existing scarce data cannot effectively support the learning process of the model, limiting its ability to detect corrosion.

[0147] In summary, the corrosion image processing technology needs to address the problems of small dataset size, insufficient representation and lack of standard benchmark. Only with significant improvements in data quantity and quality can the further application of deep learning models in this field be promoted, thereby improving the accuracy and reliability of corrosion detection.

[0148] To address the above problems, the present application provides a large-scale device surface corrosion detection method, which makes full use of limited data sets and adopts feature fusion and enhanced attention mechanism to provide accurate and efficient detection of corrosion images. Specifically, by fusing attention mechanism and multi-scale feature fusion, the accuracy and feature extraction ability are realized, so that the model can automatically focus on the salient regions in the image, thereby improving the ability to capture key information. Compared with existing feature extraction methods, feature extraction methods often only process single-scale features, which limits the model's ability to effectively recognize targets of different sizes, shapes and distributions. The present application uses a multi-scale feature fusion method to enable the model to extract and integrate information at different spatial scales, thereby enhancing the perception of multiple targets, especially when facing targets of different sizes and shapes.

[0149] Please refer to Figure 7 , Figure 7 The structure block diagram of a large-scale device surface corrosion detection system according to an embodiment of the present application is provided.

[0150] The large-scale device surface corrosion detection system provided by the present application comprises:

[0151] The acquisition module 701 is configured to acquire a large device corrosion image to be detected, and input the large device corrosion image to be detected to a preset large device surface corrosion detection model, the preset large device surface corrosion detection model comprising a fusion attention feature extraction network, a target neck network and a target detection head.

[0152] The feature extraction module 702 is configured to perform feature enhancement extraction on the large device corrosion image to be detected by using the fusion attention feature extraction network, and generate an enhanced feature map.

[0153] The convolution fusion module 703 is configured to perform feature convolution fusion on the enhanced feature map by using the target neck network, and output a fused feature map.

[0154] The target detection module 704 is configured to input the fused feature map to the target detection head for target detection, and output a corrosion detection result.

[0155] Further, the fusion attention feature extraction network comprises a convolution fusion attention mechanism module and a multi-scale feature fusion module; the enhanced feature map comprises a first enhanced feature map, a second enhanced feature map and a third enhanced feature map; and the feature extraction module 702 comprises:

[0156] The first sub-module is configured to perform convolution fusion on the large device corrosion image to be detected by using the convolution fusion attention mechanism module, and generate a first convolution fusion feature map.

[0157] The second sub-module is configured to perform convolution fusion on the first convolution fusion feature map by using the convolution fusion attention mechanism module, and output the first enhanced feature map.

[0158] The third sub-module is configured to input the first enhanced feature map to the convolution fusion attention mechanism module for convolution fusion, and determine a second enhanced feature map.

[0159] The fourth sub-module is configured to perform convolution fusion on the second enhanced feature map by using the convolution fusion attention mechanism module, and output a second convolution fusion feature map.

[0160] The fifth sub-module is configured to perform multi-scale feature fusion on the second convolution fusion feature map by using the multi-scale feature fusion module, and generate a third enhanced feature map.

[0161] Optionally, the convolution fusion attention mechanism module comprises two convolution layers connected in series, a convolution block attention mechanism layer, a concatenation layer and a convolution layer; and the image processing steps of the convolution fusion attention mechanism module comprise:

[0162] The corrosion input image input to the convolution fusion attention mechanism module is subjected to convolution operation by the two convolution layers connected in series, and a first convolution feature map is generated.

[0163] The first convolution feature map is segmented to output a first segmentation feature map and a second segmentation feature map;

[0164] The second segmentation feature map is subjected to convolution operation by two convolution layers in series to generate a second convolution feature map;

[0165] The second convolution feature map is subjected to feature weight adjustment by a convolution block attention mechanism layer to output a first convolution block attention feature map;

[0166] The first convolution block attention feature map and the second segmentation feature map are spliced to generate a first spliced feature map, and the first spliced feature map is input to two convolution layers in series for convolution operation to output a third convolution feature map;

[0167] The third convolution feature map is subjected to feature weight adjustment by a convolution block attention mechanism layer to generate a second convolution block attention feature map;

[0168] The second convolution block attention feature map and the first spliced feature map are spliced to output a second spliced feature map;

[0169] The first segmentation feature map, the first convolution block attention feature map, the second spliced feature map, and the second segmentation feature map are spliced to generate a third spliced feature map;

[0170] The third spliced feature map is subjected to convolution operation by a convolution layer to output an erosion output map.

[0171] Further, the multi-scale feature fusion module includes three cascaded average pooling layers, three cascaded maximum pooling layers, a splicing layer, and a convolution layer; the fifth sub-module is specifically configured to:

[0172] The second convolution fusion feature map is subjected to convolution operation by a convolution layer to generate a reinforced convolution feature map;

[0173] The reinforced convolution feature map is input to the three cascaded average pooling layers and the three cascaded maximum pooling layers respectively for average pooling and maximum pooling to output an average pooling feature map and a maximum pooling feature map;

[0174] The maximum pooling feature map, the average pooling feature map, and the reinforced convolution feature map are spliced by the splicing layer to output a spliced pooling feature map;

[0175] The spliced pooling feature map is subjected to convolution operation by a convolution layer to generate a third reinforced feature map.

[0176] Further, the target neck network includes a partial convolution fusion module, a splicing layer, an up-sampling layer, and a convolution layer; the fusion feature map includes a first fusion feature map, a second fusion feature map, and a third fusion feature map; the convolution fusion module 703 is specifically configured to:

[0177] The third reinforced feature map is up-sampled by using an up-sampling layer to output a first up-sampled feature map, and the first up-sampled feature map and the second reinforced feature map are spliced by using a splicing layer to generate a first intermediate splicing feature map;

[0178] The first intermediate splicing feature map is input into a partial convolution fusion module for partial convolution fusion to output a first partial convolution fusion feature map, and the first partial convolution fusion feature map is up-sampled by using an up-sampling layer to generate a second up-sampled feature map;

[0179] The second up-sampled feature map and the first reinforced feature map are spliced by using a splicing layer to output a second intermediate splicing feature map, and the second intermediate splicing feature map is input into a partial convolution fusion module for partial convolution fusion to output a first fusion feature map;

[0180] The first fusion feature map is subjected to convolution operation by using a convolution layer to generate a first fusion convolution feature map, and the first fusion convolution feature map and the first partial convolution fusion feature map are spliced by using a splicing layer to output a third intermediate splicing feature map;

[0181] The third intermediate splicing feature map is subjected to partial convolution fusion by using a partial convolution fusion module to generate a second fusion feature map, and the second fusion feature map is subjected to convolution operation by using a convolution layer to output a second fusion convolution feature map;

[0182] The second fusion convolution feature map and the third reinforced feature map are spliced by using a splicing layer to output a fourth intermediate splicing feature map, and the fourth intermediate splicing feature map is subjected to partial convolution fusion by using a partial convolution fusion module to generate a third fusion feature map.

[0183] In an optional embodiment, the system further comprises:

[0184] The first module is configured to acquire a large device corrosion image dataset to be trained, and label the large device corrosion image dataset to be trained to generate a labeled corrosion image dataset;

[0185] The second module is configured to combine the labeled corrosion image dataset by using a preset mosaic method to generate a plurality of combined corrosion images;

[0186] The third module is configured to respectively crop and adjust each combined corrosion image to generate a plurality of adjusted corrosion images;

[0187] The fourth module is configured to use the plurality of adjusted corrosion images to construct a target corrosion image dataset;

[0188] The fifth module is configured to use the target corrosion image dataset to train an initial large device surface corrosion detection model to determine a trained preset large device surface corrosion detection model.

[0189] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, modules and sub-modules can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0190] The embodiment of the present application also provides a computer device, comprising a memory and a processor, the memory stores a computer program; the computer program is executed by the processor, so that the processor executes the steps of the large device surface corrosion detection method according to any one of the foregoing embodiments.

[0191] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program / instruction, and the computer program / instruction is executed by a processor to realize the steps of the large device surface corrosion detection method according to any one of the foregoing embodiments.

[0192] The embodiment of the present application also provides a computer program product, comprising a computer program / instruction, and the computer program / instruction is executed by a processor to realize the steps of the large device surface corrosion detection method according to any one of the foregoing embodiments.

[0193] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the above-described device embodiments are only schematic, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0194] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0195] The above-described and above-described embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the foregoing embodiments of the present application have been described in detail, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for detecting surface corrosion of large-scale equipment, characterized in that, include: A corrosion image of a large device to be detected is acquired, and the corrosion image of the large device to be detected is input into a preset large device surface corrosion detection model. The preset large device surface corrosion detection model includes a fusion attention feature extraction network, a target neck network, and a target detection head. The fusion attention feature extraction network is used to perform feature enhancement extraction on the corrosion image of the large device to be detected, and an enhanced feature map is generated. The enhanced feature map is fused by feature convolution through the target neck network, and a fused feature map is output. The fused feature map is input into the target detection head for target detection, and the corrosion detection result is output. The fusion attention feature extraction network includes a convolutional fusion attention mechanism module and a multi-scale feature fusion module; the enhanced feature map includes a first enhanced feature map, a second enhanced feature map, and a third enhanced feature map; the step of using the fusion attention feature extraction network to perform feature enhancement extraction on the corrosion image of the large device to be detected, generating an enhanced feature map, includes: The convolutional fusion attention mechanism module is used to perform convolutional fusion on the corrosion image of the large device to be detected, and a first convolutional fusion feature map is generated. The first convolutional fusion feature map is convolved and fused using the convolutional fusion attention mechanism module to output the first enhanced feature map. The first enhanced feature map is input into the convolutional fusion attention mechanism module for convolutional fusion to determine the second enhanced feature map; The second enhanced feature map is convolutionally fused using a convolutional fusion attention mechanism module to output the second convolutionally fused feature map. The second convolutional fusion feature map is fused using a multi-scale feature fusion module to generate a third enhanced feature map.

2. The method for detecting surface corrosion of large equipment according to claim 1, characterized in that, The convolutional fusion attention mechanism module includes two cascaded convolutional layers, a convolutional block attention mechanism layer, a concatenation layer, and another convolutional layer; the image processing steps of the convolutional fusion attention mechanism module include: The eroded input map of the input convolutional fusion attention mechanism module is convolved through two cascaded convolutional layers to generate the first convolutional feature map; The first convolutional feature map is segmented to output a first segmented feature map and a second segmented feature map. Two cascaded convolutional layers are used to perform convolution operations on the second segmentation feature map to generate a second convolutional feature map; The second convolutional feature map is adjusted by the convolutional block attention mechanism layer, and the first convolutional block attention feature map is output. The attention feature map of the first convolutional block and the segmentation feature map are concatenated to generate a first concatenated feature map. The first concatenated feature map is then input into two concatenated convolutional layers for convolution operation to output a third convolutional feature map. A convolutional block attention mechanism layer is used to adjust the feature weights of the third convolutional feature map to generate a second convolutional block attention feature map; The attention feature map of the second convolutional block and the first concatenated feature map are concatenated to output the second concatenated feature map; The first segmentation feature map, the first convolutional block attention feature map, the second concatenation feature map, and the second segmentation feature map are concatenated to generate a third concatenation feature map; The third stitched feature map is convolutionally processed by a convolutional layer to output an erosion output map.

3. The method for detecting surface corrosion of large equipment according to claim 1, characterized in that, The multi-scale feature fusion module includes three cascaded average pooling layers, three cascaded max pooling layers, a concatenation layer, and a convolutional layer; the step of performing multi-scale feature fusion on the second convolutional fusion feature map through the multi-scale feature fusion module to generate a third enhanced feature map includes: The second convolutional fusion feature map is convolved by a convolutional layer to generate an enhanced convolutional feature map. The enhanced convolutional feature map is input into three cascaded average pooling layers and three cascaded max pooling layers respectively for average pooling and max pooling, and the average pooling feature map and max pooling feature map are output. A concatenation layer is used to concatenate the max pooling feature map, the average pooling feature map, and the enhanced convolutional feature map to output a concatenated pooling feature map; A third enhanced feature map is generated by performing convolution operations on the spliced ​​pooling feature map through a convolutional layer.

4. The method for detecting surface corrosion of large equipment according to claim 1, characterized in that, The target neck network includes a partial convolutional fusion module, a splicing layer, an upsampling layer, and a convolutional layer; the fused feature map includes a first fused feature map, a second fused feature map, and a third fused feature map; the step of performing feature convolutional fusion on the enhanced feature map through the target neck network to output a fused feature map includes: An upsampling layer is used to upsample the third enhanced feature map to output a first upsampled feature map. The first upsampled feature map and the second enhanced feature map are then spliced ​​together by a splicing layer to generate a first intermediate spliced ​​feature map. The first intermediate stitched feature map is input into the partial convolutional fusion module for partial convolutional fusion, and the first partial convolutional fusion feature map is output. The first partial convolutional fusion feature map is then upsampled using an upsampling layer to generate a second upsampled feature map. The second upsampled feature map and the first enhanced feature map are concatenated by a concatenation layer to output a second intermediate concatenated feature map. The second intermediate concatenated feature map is then input into a partial convolutional fusion module for partial convolutional fusion to output a first fused feature map. A convolutional layer is used to perform convolution operations on the first fused feature map to generate a first fused convolutional feature map. The first fused convolutional feature map and the first part of the convolutional fused feature map are then input into a stitching layer for stitching, and a third intermediate stitched feature map is output. The third intermediate stitched feature map is partially convolutionally fused by a partial convolutional fusion module to generate a second fused feature map. The second fused feature map is then convolutionally fused by a convolutional layer to output the second fused convolutional feature map. The second fused convolutional feature map and the third enhanced feature map are input into the splicing layer for splicing, and a fourth intermediate spliced ​​feature map is output. The fourth intermediate spliced ​​feature map is then partially spliced ​​by a partial convolutional fusion module to generate a third fused feature map.

5. The method for detecting surface corrosion of large equipment according to claim 1, characterized in that, The training process of the pre-set large device surface corrosion detection model includes: Obtain a dataset of erosion images of a large device to be trained, and annotate the dataset to generate an annotated erosion image dataset. The labeled erosion image dataset is combined using a pre-set mosaic method to generate multiple combined erosion images. Each of the combined erosion images is cropped and adjusted to generate multiple adjusted erosion images; A target corrosion image dataset is constructed using multiple adjusted corrosion images. The target corrosion image dataset is used to train the initial large-scale device surface corrosion detection model, and the trained pre-set large-scale device surface corrosion detection model is determined.

6. A surface corrosion detection system for large-scale equipment, applied to the surface corrosion detection method for large-scale equipment as described in claim 1, characterized in that, include: The acquisition module is used to acquire corrosion images of the large device to be detected and input the corrosion images of the large device to be detected into a preset large device surface corrosion detection model. The preset large device surface corrosion detection model includes a fusion attention feature extraction network, a target neck network and a target detection head. The feature extraction module is used to perform feature enhancement extraction on the corrosion image of the large device to be detected using the fusion attention feature extraction network, and generate an enhanced feature map; The convolutional fusion module is used to perform feature convolutional fusion on the enhanced feature map through the target neck network and output a fused feature map. The target detection module is used to input the fused feature map into the target detection head for target detection and output corrosion detection results.

7. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the large device surface corrosion detection method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed, it implements the method for detecting surface corrosion of large devices as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, wherein when the program instructions are executed by a computer, the computer performs the large device surface corrosion detection method as described in any one of claims 1-5.

Images