Ocean platform corrosion identification and prediction method and system based on machine vision

Abstract

The invention discloses an ocean platform corrosion identification prediction method and system based on machine vision, and the method comprises the steps: carrying out the segmentation of an image through a multi-scale segmentation algorithm according to an ocean platform corrosion image data set, obtaining a candidate corrosion region set, correcting a candidate region through morphological operation, removing a noise region with a too small area, and carrying out the recognition and prediction of the ocean platform corrosion. Obtaining a refined corrosion area; secondly, extracting a plurality of feature vectors for the corrosion area, constructing multi-view corrosion feature vectors, fusing typical feature modes of corrosion types, and optimizing features through similarity calculation; and finally, analyzing the evolution process of the corrosion area by using a time series data mining technology, constructing a Markov random field model to predict a future corrosion trend, and generating a corrosion development map by using a visualization technology. According to the method, accurate identification and dynamic prediction of corrosion of the ocean platform are realized by highlighting discriminative characteristics of different corrosion types, and the accuracy and efficiency of corrosion monitoring are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine platform corrosion identification and prediction, and in particular relates to a marine platform corrosion identification and prediction method and system based on machine vision. Background Art

[0002] There is a key technical problem in the identification of offshore platform corrosion, that is, how to accurately distinguish and identify different types of corrosion patterns. Common types of corrosion include pitting, uniform corrosion, and crack corrosion, which have obvious differences in morphology, texture, and color. Although traditional edge detection algorithms such as Canny and Sobel can extract the contour of the corrosion area, it is difficult to achieve accurate classification for complex corrosion morphologies such as irregular pitting and cracks based on contour information alone. Directly applying machine learning algorithms such as support vector machines and random forests for classification faces the problem of insufficient feature representation and is difficult to characterize the detailed features of corrosion samples. Although convolutional neural networks can automatically extract deep features, they are prone to overfitting in small sample scenarios. In addition, simple classification models lack consideration of the corrosion evolution process and cannot reveal the development trend of corrosion. Therefore, how to characterize corrosion features from multiple angles such as texture, color, and shape, and combine prior knowledge to build a robust corrosion classification and evolution prediction model, is a technical problem that needs to be solved urgently. This requires in-depth exploration in multiple fields such as machine vision, machine learning, and materials science, and the comprehensive use of multi-source data and interdisciplinary knowledge to break through bottlenecks and achieve accurate identification and prediction of offshore platform corrosion. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method and system for identifying and predicting marine platform corrosion based on machine vision to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a method for identifying and predicting marine platform corrosion based on machine vision, comprising the following steps:

[0005] According to the offshore platform corrosion image dataset, a multi-scale segmentation algorithm is used to segment the image to obtain a set of candidate corrosion regions. The candidate regions are corrected through morphological operations, and the noise regions with too small areas are eliminated to obtain the refined corrosion regions.

[0006] For the corrosion area, a number of feature vectors are extracted, and corrosion feature vectors of multiple perspectives are constructed based on the several feature vectors;

[0007] On the basis of feature vectors, typical feature patterns of corrosion types are integrated, and features are optimized through similarity calculation to highlight the discriminative features of different corrosion types and form knowledge-enhanced corrosion feature representation.

[0008] Through time series data mining technology, the evolution process of the corrosion area is analyzed, the spatiotemporal characteristics of corrosion development are extracted, and a Markov random field model is constructed to describe the dynamic changes of corrosion and predict future corrosion trends.

[0009] Visualization technology is used to generate corrosion development maps, which present the corrosion distribution and severity in different time periods through color coding and heat maps.

[0010] Optionally, segmenting the image using a multi-scale segmentation algorithm according to the marine platform corrosion image dataset comprises the following steps:

[0011] The Gaussian filtering algorithm is used to pre-process the image to remove noise and obtain the image after noise suppression;

[0012] A multi-scale watershed segmentation algorithm is used to obtain candidate corrosion area sets at different scales;

[0013] The support vector machine algorithm is used to classify the candidate corrosion areas into two categories to obtain a preliminary screened set of corrosion areas;

[0014] The noise area with too small area is removed by morphological opening operation, and the edge of the eroded area is corrected by closing operation to obtain the refined eroded area set.

[0015] Optionally, extracting texture features, color features and shape features from the corrosion area includes the following steps:

[0016] The gray-level co-occurrence matrix algorithm is used to extract the texture features of the corrosion area and obtain the texture feature vector;

[0017] The color histogram algorithm is used to extract the color distribution characteristics and obtain the color histogram feature vector;

[0018] The color moment algorithm is used to extract color statistical features and obtain the color moment feature vector;

[0019] The shape invariant features are extracted by using the invariant moment algorithm to obtain the invariant moment feature vector;

[0020] The Fourier descriptor algorithm is used to extract the shape frequency domain features and obtain the Fourier descriptor feature vector;

[0021] Each feature vector is fused to construct a multi-view erosion feature vector.

[0022] Optionally, the step of fusing typical characteristic patterns of the corrosion type based on the characteristic vector comprises the following steps:

[0023] Obtain material corrosion data, obtain the feature vector of the corrosion image, and establish a typical feature pattern library of different corrosion types based on prior knowledge in the field of material science;

[0024] Calculate the similarity between the feature vector and each corrosion type in the pattern library to obtain a similarity vector of the feature vector;

[0025] If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the typical feature pattern of the type is merged into the feature vector;

[0026] The support vector machine algorithm is used to train the corrosion type classification model with the knowledge-enhanced corrosion feature representation as input;

[0027] Use the trained model to identify the corrosion type of the new corrosion image and output the recognition result.

[0028] Optionally, if the similarity of a certain type in the similarity vector is greater than a preset threshold, then the feature vector is judged to belong to the corrosion type, and the following steps are also included:

[0029] A similarity vector of the feature vector is constructed according to the similarity values, and each element in the vector represents the degree of similarity with the corresponding type; the similarity values ​​in the similarity vector are sorted, and the corrosion type with the largest similarity is taken as the candidate type.

[0030] Optionally, after training the corrosion type classification model, also include:

[0031] For the corrosion feature vector of each sample, a meta-learning algorithm is used, with the corrosion feature vector as input and the corresponding corrosion type as output, to establish the corrosion type classification model as a feature-to-type mapping model;

[0032] For the mapping model, adversarial samples are generated through the adversarial learning algorithm, and the adversarial samples are added to the training set to retrain the model;

[0033] Obtain unlabeled new corrosion samples and extract the corrosion feature vector of each sample;

[0034] Input the feature vector of the new sample into the enhanced mapping model to determine the corrosion type of each sample;

[0035] Conduct confidence analysis on the corrosion type results output by the model, and manually review samples with confidence levels lower than the preset threshold to improve classification accuracy;

[0036] Feedback the classification results, add new samples and their predicted types to the training set, retrain the model regularly, and complete continuous optimization of the model.

[0037] Optionally, analyzing the evolution of the corrosion area by using the time series data mining technology comprises the following steps:

[0038] According to the corrosion classification results, the time series data of the corrosion area is obtained, and the evolution process of the corrosion area is analyzed by using the time series data mining technology;

[0039] By constructing a Markov random field model, the extracted spatiotemporal characteristics of corrosion development are used as observation data of the model;

[0040] The transfer probability matrix and emission probability matrix of the model are determined by using parameter estimation method, and a probability model reflecting the dynamic change law of corrosion is obtained.

[0041] The trained Markov random field model is used to predict the distribution of the corrosion area at the next moment according to the spatiotemporal characteristics of the current corrosion area.

[0042] If the predicted future corrosion trend exceeds the preset threshold, the current corrosion area is judged to have a development risk and the early warning mechanism is triggered;

[0043] The environmental parameters and material property information of the corrosion area are obtained and used as auxiliary inputs of the Markov random field model.

[0044] Optionally, the use of visualization technology to generate a corrosion development map includes the following steps:

[0045] Obtain historical corrosion data and current corrosion status data of offshore platforms, pre-process the data, unify the data format and unit, and process missing values ​​and outliers;

[0046] Based on the preprocessed data, a time series prediction algorithm is used to establish a corrosion development prediction model to predict the corrosion distribution and severity in different time periods in the future;

[0047] The prediction results are visualized and color-coded to represent corrosion of different severity with different colors to generate a corrosion distribution map.

[0048] On the basis of the corrosion distribution map, a heat map is superimposed, and different colors and brightness are used to highlight the high-risk areas according to the severity of corrosion;

[0049] Compare and analyze the corrosion distribution maps and thermograms of different time periods to determine the corrosion development trend and rate, and identify areas where corrosion is accelerated or slowed down; comprehensively consider the corrosion distribution, severity and development trend factors, evaluate the corrosion risk level of various parts of the offshore platform, and generate a risk assessment report.

[0050] The present invention also provides a machine vision-based marine platform corrosion identification and prediction system, which is used to implement a machine vision-based marine platform corrosion identification and prediction method, and the system comprises:

[0051] Image acquisition module, data processing module, feature extraction module, model training module and visualization module;

[0052] The image acquisition module is used to obtain corrosion image data of the offshore platform;

[0053] The data processing module is used to pre-process the image data, use a Gaussian filter algorithm to remove noise, and then use a multi-scale segmentation algorithm to segment the image to obtain a set of candidate corrosion regions;

[0054] The feature extraction module is used to extract texture features, color features and shape features of the corrosion area to form a multi-view corrosion feature vector;

[0055] The model training module is used to establish a corrosion type classification model and identify the corrosion type through knowledge-enhanced corrosion feature representation;

[0056] The visualization module is used to generate a corrosion development map, which presents the corrosion distribution and severity in different time periods in the form of color coding and heat map.

[0057] Optionally, the data processing module also includes a morphological operation unit for performing opening and closing operations on the candidate corrosion areas, removing noise areas and correcting edges to obtain a refined set of corrosion areas. In the feature extraction module, the texture feature extraction is implemented through a grayscale co-occurrence matrix algorithm, the color feature is extracted through a color histogram and a color moment algorithm, and the shape feature is extracted through an invariant moment and a Fourier descriptor algorithm to form a complete feature vector.

[0058] Optionally, the model training module adopts a small sample learning strategy, combined with a meta-learning algorithm and an adversarial learning algorithm, to train feature vectors to obtain a mapping model from features to corrosion types. After the output results of the mapping model are subjected to confidence analysis, low-confidence samples are marked as requiring review.

[0059] Optionally, the visualization module processes historical corrosion data and current status data through a time series prediction algorithm to generate a corrosion development prediction model and generate visualization results including corrosion distribution maps and thermal maps. The results are used to evaluate the corrosion risk level of each part and output a risk assessment report.

[0060] Optionally, the system further includes a database module for storing image data, corrosion feature vectors, model parameters and risk assessment reports. The database module is connected to each functional module to achieve data sharing and real-time updating.

[0061] Optionally, in the model training module, based on the feature vector, typical feature patterns of the corrosion type are integrated, including:

[0062] Obtain material corrosion data, obtain the feature vector of the corrosion image, and establish a typical feature pattern library of different corrosion types based on prior knowledge in the field of material science;

[0063] Calculate the similarity between the feature vector and each corrosion type in the pattern library to obtain a similarity vector of the feature vector;

[0064] If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the typical feature pattern of the type is merged into the feature vector;

[0065] The support vector machine algorithm is used to train the corrosion type classification model with the knowledge-enhanced corrosion feature representation as input;

[0066] Use the trained model to identify the corrosion type of the new corrosion image and output the recognition result.

[0067] Optionally, in the model training module,

[0068] If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the following steps are also included:

[0069] A similarity vector of the feature vector is constructed according to the similarity values, and each element in the vector represents the degree of similarity with the corresponding type; the similarity values ​​in the similarity vector are sorted, and the corrosion type with the largest similarity is taken as the candidate type.

[0070] Optionally, in the model training module, after training the corrosion type classification model, the following further comprises:

[0071] For the corrosion feature vector of each sample, a meta-learning algorithm is used, with the corrosion feature vector as input and the corresponding corrosion type as output, to establish the corrosion type classification model as a feature-to-type mapping model;

[0072] For the mapping model, adversarial samples are generated through the adversarial learning algorithm, and the adversarial samples are added to the training set to retrain the model;

[0073] Obtain unlabeled new corrosion samples and extract the corrosion feature vector of each sample;

[0074] Input the feature vector of the new sample into the enhanced mapping model to determine the corrosion type of each sample;

[0075] Conduct confidence analysis on the corrosion type results output by the model, and manually review samples with confidence levels lower than the preset threshold to improve classification accuracy;

[0076] Feedback the classification results, add new samples and their predicted types to the training set, retrain the model regularly, and complete continuous optimization of the model.

[0077] The present invention also provides a computer terminal device, comprising:

[0078] one or more processors;

[0079] A memory, coupled to the processor, for storing one or more programs;

[0080] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of a method for identifying and predicting marine platform corrosion based on machine vision.

[0081] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for identifying and predicting marine platform corrosion based on machine vision are implemented.

[0082] Compared with the prior art, the present invention has the following advantages and technical effects:

[0083] The present invention provides a method for identifying and predicting marine platform corrosion based on machine vision, comprising: firstly, according to a marine platform corrosion image data set, a multi-scale segmentation algorithm is used to segment the image to obtain a set of candidate corrosion areas, the candidate areas are corrected by morphological operations, and noise areas with too small areas are eliminated to obtain refined corrosion areas; secondly, for the corrosion areas, a number of feature vectors are extracted, and multi-perspective corrosion feature vectors are constructed based on the number of feature vectors; on the basis of the feature vectors, typical feature patterns of corrosion types are integrated, and the features are optimized by similarity calculation to highlight the discriminative features of different corrosion types, so as to form a knowledge-enhanced corrosion feature representation; finally, the evolution process of the corrosion area is analyzed by time series data mining technology, the spatiotemporal features of corrosion development are extracted, a Markov random field model is constructed to characterize the dynamic changes of corrosion, and future corrosion trends are predicted; and a corrosion development map is generated by using visualization technology, and the corrosion distribution and severity in different time periods are presented by color coding and heat maps. The machine vision-based marine platform corrosion identification method of the present invention realizes efficient identification, dynamic monitoring and future trend prediction of corrosion areas through the comprehensive application of multi-scale segmentation, feature extraction and fusion, time series data analysis and visualization technology, and significantly improves the accuracy and efficiency of marine platform corrosion management. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0085] Figure 1 is a general method flow chart of an embodiment of the present invention;

[0086] Figure 2 This is a schematic diagram of model training in step S103 of an embodiment of the present invention;

[0087] Figure 3 The system structure and flow chart of the embodiment of the present invention are shown in FIG. DETAILED DESCRIPTION

[0088] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0089] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0090] Embodiment 1

[0091] like Figure 1 As shown, this embodiment provides a method for identifying and predicting marine platform corrosion based on machine vision, including:

[0092] According to the offshore platform corrosion image dataset, a multi-scale segmentation algorithm is used to segment the image to obtain a set of candidate corrosion regions. The candidate regions are corrected through morphological operations, and the noise regions with too small areas are eliminated to obtain the refined corrosion regions.

[0093] For the corrosion area, a number of feature vectors are extracted, and corrosion feature vectors of multiple perspectives are constructed based on the several feature vectors;

[0094] On the basis of feature vectors, typical feature patterns of corrosion types are integrated, and features are optimized through similarity calculation to highlight the discriminative features of different corrosion types and form knowledge-enhanced corrosion feature representation.

[0095] Through time series data mining technology, the evolution process of the corrosion area is analyzed, the spatiotemporal characteristics of corrosion development are extracted, and a Markov random field model is constructed to describe the dynamic changes of corrosion and predict future corrosion trends.

[0096] Visualization technology is used to generate corrosion development maps, which present the corrosion distribution and severity in different time periods through color coding and heat maps.

[0097] S101, based on the marine platform corrosion image dataset, a multi-scale segmentation algorithm is used to segment the image, including the following steps:

[0098] The Gaussian filtering algorithm is used to pre-process the image to remove noise and obtain the image after noise suppression;

[0099] A multi-scale watershed segmentation algorithm is used to obtain candidate corrosion area sets at different scales;

[0100] The support vector machine algorithm is used to classify the candidate corrosion areas into two categories to obtain a preliminary screened set of corrosion areas;

[0101] The noise area with too small area is removed by morphological opening operation, and the edge of the eroded area is corrected by closing operation to obtain the refined eroded area set.

[0102] As an optional implementation of this embodiment, based on the marine platform corrosion image dataset, a Gaussian filtering algorithm is used to perform denoising preprocessing on the image to obtain a noise-suppressed image. For the preprocessed image, a multi-scale watershed segmentation algorithm is used to obtain a set of candidate corrosion regions at different scales by setting structural elements of different scales. According to the grayscale, texture, shape and other features of the candidate corrosion region, a regional feature vector is constructed, and the candidate region is binary classified using a support vector machine algorithm to obtain a preliminarily screened set of corrosion regions. For the preliminarily screened set of corrosion regions, a morphological opening operation is used to remove noise regions with too small an area, and then a morphological closing operation is used to correct the edge of the corrosion region to obtain a refined set of corrosion regions.

[0103] Specifically, Gaussian filtering is a commonly used image denoising method. It uses the weights of normal distribution to perform weighted averaging on each pixel in the image, thereby smoothing the image and suppressing noise. For example, for an image of marine platform corrosion, there may be some random noise points in the image, which will interfere with the subsequent identification of corrosion areas. A Gaussian filter with a size of 5x5 and a standard deviation of 1.5 can be used to process the image, replacing the value of each pixel with the weighted average of its surrounding pixels. In this way, the influence of the noise point will be smoothed by the surrounding pixel values, thereby achieving the purpose of denoising, making the image clearer and facilitating subsequent segmentation and feature extraction. Image segmentation is the process of dividing an image into several regions with similar features. The multi-scale watershed segmentation algorithm is a segmentation method based on image morphology, which can effectively segment different regions in an image. The algorithm regards the image as a topographic map, and the grayscale value represents the height of the terrain. By simulating the spreading process of water flow, the image is finally segmented into different regions. The preprocessed image can be segmented multiple times using structural elements of different sizes (e.g., circular or square structural elements of 3x 3, 5x 5, 7x 7). The size of the structural element determines the scale of the segmentation. Smaller structural elements can segment smaller areas, while larger structural elements can segment larger areas. In this way, a set of candidate erosion regions at multiple scales can be obtained, thereby improving the accuracy of erosion region identification. For example, for an image containing multiple erosion regions of different sizes, a smaller structural element can be used to segment small erosion regions, while a larger structural element can be used to segment large erosion regions, thereby identifying all erosion regions. After obtaining the candidate erosion regions, their features need to be extracted for subsequent classification. Grayscale features (e.g., average grayscale, grayscale variance), texture features (e.g., statistics of grayscale co-occurrence matrix), and shape features (e.g., area, perimeter, circularity) of the candidate regions can be extracted. For example, the grayscale value of the erosion region is usually lower than that of the surrounding area. The average grayscale value of the candidate region can be calculated and used as a component of the feature vector. The texture of the corroded area may also be different from that of the surrounding area. The energy, entropy and other features of the gray-level co-occurrence matrix can be calculated and used as components of the feature vector. The shape of the corroded area may also have certain characteristics. The area, perimeter, circularity and other aspects of the candidate area can be calculated and used as components of the feature vector. Combining these features into a feature vector can be used to describe the characteristics of the candidate area. Support vector machine is a commonly used binary classification algorithm. It separates samples of different categories by finding an optimal hyperplane. The extracted regional feature vector can be used as input to classify the candidate area using the support vector machine to determine whether it is a corroded area. For example, the feature vectors of samples that are known to be corroded areas and not corroded areas can be input into the support vector machine for training to obtain a classification model.Then, the feature vector of the candidate region to be classified is input into the trained model to determine whether it is a corrosion region. In this way, the corrosion region can be preliminarily screened out. Morphological opening and closing operations are two commonly used image processing methods. Opening operations can remove small noise regions in the image, while closing operations can connect disconnected regions and smooth edges. After the corrosion regions are preliminarily screened out, there may be some noise regions with too small areas, which are not real corrosion regions. Morphological opening operations can be used to remove these noise regions. For example, using a 3x 3 square structure element for opening operations can effectively remove noise regions smaller than the structure element, while retaining corrosion regions with larger areas. After the opening operation, the edge of the corrosion region may not be smooth enough or even broken. Morphological closing operations can be used to correct the edge of the corrosion region. For example, using a 5x 5 square structure element for closing operations can connect disconnected corrosion regions, smooth the edge of the corrosion region, and make the outline of the corrosion region more complete. After the processing of morphological opening and closing operations, a more accurate set of corrosion regions can be obtained.

[0104] S102, extracting texture features, color features and shape features for the corrosion area, including the following steps:

[0105] The gray-level co-occurrence matrix algorithm is used to extract the texture features of the corrosion area and obtain the texture feature vector;

[0106] The color histogram algorithm is used to extract the color distribution characteristics and obtain the color histogram feature vector;

[0107] The color moment algorithm is used to extract color statistical features and obtain the color moment feature vector;

[0108] The shape invariant features are extracted by using the invariant moment algorithm to obtain the invariant moment feature vector;

[0109] The Fourier descriptor algorithm is used to extract the shape frequency domain features and obtain the Fourier descriptor feature vector;

[0110] Each feature vector is fused to construct a multi-view erosion feature vector.

[0111] As an optional implementation method of this embodiment, based on the corrosion area image, a gray level co-occurrence matrix algorithm is used to extract the texture features of the corrosion area to obtain a texture feature vector; based on the corrosion area image, a wavelet transform algorithm is used to extract the multi-scale texture features of the corrosion area to obtain a wavelet transform coefficient feature vector; based on the corrosion area image, a color histogram algorithm is used to extract the color distribution features of the corrosion area to obtain a color histogram feature vector; based on the corrosion area image, a color moment algorithm is used to extract the color statistical features of the corrosion area to obtain a color moment feature vector; based on the corrosion area contour, an invariant moment algorithm is used to extract the shape invariant features of the corrosion area to obtain an invariant moment feature vector; based on the corrosion area contour, a Fourier descriptor algorithm is used to extract the shape frequency domain features of the corrosion area to obtain a Fourier descriptor feature vector; the texture feature vector, wavelet transform coefficient feature vector, color histogram feature vector, color moment feature vector, invariant moment feature vector and Fourier descriptor feature vector are fused to construct a multi-view corrosion feature vector for subsequent corrosion detection and analysis.

[0112] Specifically, the gray-level co-occurrence matrix algorithm can effectively extract the texture features of the image. Its basic principle is to count the probability of different gray-value pixel pairs in the image appearing in different directions and at different distances. For example, for the marine platform corrosion image, the gray-level co-occurrence matrix with a distance of one pixel in the horizontal direction can be calculated. By calculating the statistics of the gray-level co-occurrence matrix such as energy, entropy, contrast, and correlation, a texture feature vector can be constructed to describe the texture features of the corrosion area. The texture of the corrosion area is usually rough, so the entropy value of its gray-level co-occurrence matrix may be high. The wavelet transform algorithm can decompose the image into sub-bands of different scales and directions, thereby extracting the multi-scale texture features of the image. For example, the Haar wavelet can be used to decompose the marine platform corrosion image to obtain detail sub-bands in the horizontal, vertical, and diagonal directions. These sub-bands reflect the texture information of the image in different directions. By calculating the statistics of the wavelet coefficients of each sub-band such as energy, mean, and variance, a wavelet transform coefficient feature vector can be constructed to describe the multi-scale texture features of the corrosion area. The corrosion area may show different texture characteristics at different scales, and the wavelet transform can effectively capture these characteristics. The color histogram algorithm can count the frequency of different colors in the image, thereby extracting the color distribution characteristics of the image. For example, for the marine platform corrosion image, the distribution of pixel values ​​of each color channel (such as RGB) in the image can be counted. The corrosion area usually presents a specific color, such as reddish brown. By calculating the color histogram, a color histogram feature vector can be obtained to describe the color distribution characteristics of the corrosion area. The color histogram can reflect the main color composition of the corrosion area. The color moment algorithm can extract the color statistical characteristics of the image, such as mean, variance, and skewness. These statistics can describe the color distribution of the image. For example, for the marine platform corrosion image, the mean, variance, and skewness of the three RGB color channels of the corrosion area can be calculated. Combining these statistics into a color moment feature vector can be used to describe the color characteristics of the corrosion area. The color of the corrosion area is usually different from that of the surrounding area, and the color moment can capture these differences. The invariant moment algorithm can extract the shape invariant features of the image, such as area, perimeter, circularity, etc. These features are not affected by transformations such as translation, rotation, and scaling of the image. For example, seven invariant moments of the corrosion area can be calculated and combined into an invariant moment feature vector. The shape of the eroded area may show certain regularity, such as being approximately circular or elliptical. Invariant moments can effectively describe these shape features. The Fourier descriptor algorithm can represent the contour of the image as coefficients in the frequency domain, thereby extracting the shape frequency domain features of the image. For example, the contour coordinates of the eroded area can be Fourier transformed to obtain a series of Fourier coefficients. A part of the low-frequency coefficients is selected to construct the Fourier descriptor feature vector. These low-frequency coefficients reflect the main shape characteristics of the contour of the eroded area.The contour shape of the corrosion area may be complex, and Fourier descriptors can effectively describe these complex shapes. By fusing the texture feature vector, wavelet transform coefficient feature vector, color histogram feature vector, color moment feature vector, invariant moment feature vector and Fourier descriptor feature vector extracted above, a multi-view corrosion feature vector can be constructed. Doing so can more comprehensively describe the characteristics of the corrosion area and improve the accuracy and robustness of corrosion detection and analysis. For example, these feature vectors can be connected into a high-dimensional vector, or a more complex fusion method can be used, such as weight-based fusion. Multi-view feature fusion can comprehensively utilize the advantages of different features to better identify and analyze corrosion areas.

[0113] S103, based on the feature vector, integrating typical feature patterns of the corrosion type, including the following steps:

[0114] Obtain material corrosion data, obtain the feature vector of the corrosion image, and establish a typical feature pattern library of different corrosion types based on prior knowledge in the field of material science;

[0115] Calculate the similarity between the feature vector and each corrosion type in the pattern library to obtain a similarity vector of the feature vector;

[0116] If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the typical feature pattern of the type is merged into the feature vector;

[0117] The support vector machine algorithm is used to train the corrosion type classification model with the knowledge-enhanced corrosion feature representation as input;

[0118] Use the trained model to identify the corrosion type of the new corrosion image and output the recognition result.

[0119] As an optional implementation method of this embodiment, material corrosion data is obtained to obtain a feature vector of the corrosion image. Based on prior knowledge in the field of materials science, a typical feature pattern library of different corrosion types is established. The similarity between the feature vector and each corrosion type in the typical feature pattern library is calculated to obtain a similarity vector of the feature vector. If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the typical feature pattern of the type is fused into the feature vector. By fusing typical feature patterns, optimizing the original feature vector, highlighting the discriminative features of different corrosion types, and obtaining a knowledge-enhanced corrosion feature representation. Using a support vector machine algorithm, the corrosion type classification model is trained with the knowledge-enhanced corrosion feature representation as input. Using the trained classification model, the corrosion type of the new corrosion image is identified, and the identification result is output.

[0120] Specifically, material corrosion is a ubiquitous phenomenon that can seriously affect the performance and service life of materials. In order to effectively identify and control corrosion, it is necessary to analyze the corrosion image and determine the type of corrosion. First, it is necessary to obtain material corrosion data and extract corrosion images from them. For example, corrosion images can be collected from actual engineering structures such as offshore platforms, bridges, and pipelines. After obtaining the image, it is necessary to extract the feature vector of the image to describe the characteristics of the corrosion area. Here, a variety of feature extraction methods can be used, such as the texture features (gray level co-occurrence matrix, wavelet transform coefficients), color features (color histogram, color moment), and shape features (invariant moment, Fourier descriptor) mentioned above. Suppose there is an offshore platform corrosion image, its texture, color, and shape features are extracted, and they are combined into a feature vector, such as [2,5,8,...,1]. This vector contains various feature information of the corrosion area. Next, it is necessary to establish a typical feature pattern library of different corrosion types based on prior knowledge in the field of materials science. Corrosion types can be classified according to the mechanism and morphology of corrosion, such as pitting, crevice corrosion, uniform corrosion, etc. For each corrosion type, a large number of sample images can be collected and their feature vectors can be extracted. Then, the average value of these feature vectors is calculated as the typical feature pattern of the corrosion type. For example, the typical feature pattern of pitting corrosion may be [8,2,1,... ,9], and the typical feature pattern of crevice corrosion may be [1,7,3,... ,2]. Then, it is necessary to calculate the similarity between the feature vector of the corrosion image and each corrosion type in the typical feature pattern library. The similarity can be calculated using methods such as cosine similarity. For example, assuming that the calculated similarity vector is [9,2,1], this means that the corrosion image has the highest similarity with pitting corrosion. If the similarity of a certain type in the similarity vector is greater than a preset threshold, such as 8, then the feature vector is judged to belong to the corrosion type. The advantage of this is that prior knowledge can be used to assist in the identification of corrosion types and improve the accuracy of identification. At the same time, the typical feature pattern of the type is fused into the feature vector. For example, if the corrosion image is judged to belong to pitting corrosion, the typical feature pattern of pitting corrosion [8,2,1,... ,9] are fused into the original feature vector [2,5,8,... ,1]. The fusion method can be a simple weighted average, such as multiplying the two vectors by a certain weight and then adding them. Fusion of typical feature patterns can optimize the original feature vector, highlight the discriminative features of different corrosion types, and obtain a knowledge-enhanced corrosion feature representation. For example, the fused feature vector may be [7,3,15,... ,8]. Finally, the support vector machine algorithm is used to train the corrosion type classification model with the knowledge-enhanced corrosion feature representation as input. Support vector machine is a commonly used classification algorithm with good generalization ability. After training, the trained classification model can be used to identify the corrosion type of new corrosion images.For example, if the feature vector of a new corrosion image is input into the trained model, the model will output the corrosion type to which the image belongs, such as pitting. This can realize automatic corrosion type identification and improve the efficiency of corrosion detection and analysis. In this way, different types of corrosion can be more effectively identified and evaluated, thus providing guidance for subsequent maintenance and repair work.

[0121] S1031, if the similarity of a certain type in the similarity vector is greater than a preset threshold, it is determined that the feature vector belongs to the corrosion type, and the following steps are also included:

[0122] A similarity vector of the feature vector is constructed according to the similarity values, and each element in the vector represents the degree of similarity with the corresponding type; the similarity values ​​in the similarity vector are sorted, and the corrosion type with the largest similarity is taken as the candidate type.

[0123] As an optional implementation of this embodiment, the similarity between the feature vector and each corrosion type in the typical feature pattern library is calculated to obtain a similarity vector of the feature vector.

[0124] Obtain the feature vector to be analyzed, which represents the corrosion feature information of the material surface. Obtain a pre-established typical corrosion feature pattern library, which contains feature patterns of multiple typical corrosion types. Calculate the similarity between the feature vector and the feature pattern of each corrosion type in the pattern library, and use algorithms such as cosine similarity to obtain the similarity value between the feature vector and each type. Based on the calculated similarity value, construct a similarity vector of the feature vector, and each element in the vector represents the degree of similarity with the corresponding type. Sort the similarity values ​​in the similarity vector, and use the corrosion type with the greatest similarity as the candidate type.

[0125] Specifically, obtain the feature vector to be analyzed, which represents the characteristic information of material corrosion. Imagine taking a corrosion photo, extracting some key features such as color, texture, shape, etc. through image processing technology, and then converting these features into numbers to form a vector, which is the so-called feature vector. For example, a feature vector may be [2,5,8,1], where each number represents the intensity of a feature. Next, a typical corrosion feature pattern library is needed. This library is like a dictionary of corrosion types, which stores the characteristic patterns of various known corrosion types. Each corrosion type has a corresponding feature vector, which represents the most typical feature of that type of corrosion. For example, the feature vector of pitting corrosion may be [8,2,1,9], and the feature vector of crevice corrosion may be [1,7,3,2]. Building this library requires a large amount of corrosion sample data and professional material science knowledge. Then, calculate the similarity between the feature vector to be analyzed and the feature vector of each corrosion type in the pattern library. The cosine similarity algorithm can be used here. Cosine similarity can be understood as the cosine value of the angle between two vectors. The smaller the angle, the larger the cosine value, and the higher the similarity. For example, the cosine similarity between the feature vector [2,5,8,1] to be analyzed and the feature vector [8,2,1,9] of pitting corrosion is calculated to be 0.2, and the cosine similarity between the feature vector [2,5,8,1] to be analyzed and the feature vector [8,2,1,9] of crevice corrosion is calculated to be 0.4. After the similarity value is calculated, a similarity vector can be constructed. Each element in this vector represents the similarity between the feature vector to be analyzed and the corresponding corrosion type. For example, if there are three types of corrosion in the library: pitting corrosion, crevice corrosion, and uniform corrosion, then the similarity vector may be [2,4,6], indicating that the similarity between the feature vector to be analyzed and pitting corrosion is 0.2, the similarity between crevice corrosion and uniform corrosion is 0.4, and the similarity between uniform corrosion and uniform corrosion is 0.6. Finally, the values ​​in the similarity vector are sorted. The corrosion type with the greatest similarity will be considered as a candidate type. In the above example, the similarity value of uniform corrosion is 0.6, which is the largest, so it is considered that the corrosion image to be analyzed is most likely uniform corrosion. Doing so can help quickly screen out the most likely corrosion type and improve the efficiency of corrosion recognition. The advantage of this is that existing knowledge can be used to guide the identification of corrosion types, rather than relying entirely on complex models. Especially when sample data is insufficient, this method can effectively improve the accuracy of identification.

[0126] S1032. After training the corrosion type classification model, it also includes:

[0127] like Figure 2 As shown, for the corrosion feature vector of each sample, a meta-learning algorithm is used, with the corrosion feature vector as input and the corresponding corrosion type as output, to establish the corrosion type classification model as a feature-to-type mapping model;

[0128] For the mapping model, adversarial samples are generated through the adversarial learning algorithm, and the adversarial samples are added to the training set to retrain the model;

[0129] Obtain unlabeled new corrosion samples and extract the corrosion feature vector of each sample;

[0130] Input the feature vector of the new sample into the enhanced mapping model to determine the corrosion type of each sample;

[0131] Conduct confidence analysis on the corrosion type results output by the model, and manually review samples with confidence levels lower than the preset threshold to improve classification accuracy;

[0132] Feedback the classification results, add new samples and their predicted types to the training set, retrain the model regularly, and complete continuous optimization of the model.

[0133] As an optional implementation method of this embodiment, a small sample learning strategy is adopted. On limited labeled samples, a mapping model from corrosion features to types is established through a meta-learning algorithm. The model is used to migrate to new corrosion samples to achieve accurate classification of corrosion types.

[0134] Corrosion feature vector for each sample. A meta-learning algorithm is used, with the corrosion feature vector as input and the corresponding corrosion type as output, to train and establish a mapping model from feature to type. For the trained mapping model, adversarial samples are generated through the adversarial learning algorithm, and the adversarial samples are added to the training set to retrain the model to improve the robustness of the model. A batch of unlabeled new corrosion samples are obtained from the business scenario, and the corrosion feature vector of each sample is extracted. The feature vector of the new sample is input into the enhanced mapping model, and the corrosion type of each sample is determined by transfer learning. The confidence analysis of the corrosion type results output by the model is performed, and samples with confidence lower than the preset threshold are manually reviewed to improve the classification accuracy. The classification results are returned to the business system, and the new samples and their predicted types are added to the training set. The model is retrained regularly to achieve continuous optimization and updating of the model.

[0135] Specifically, the corrosion feature vector can be understood as a key indicator that describes the corrosion morphology in numbers. For example, the average corrosion depth, corrosion area percentage, number of corrosion holes and other indicators can be combined into a vector, such as [5,2,10], which is a corrosion feature vector that represents the corrosion situation of a certain sample. The typical corrosion feature pattern library is like a "picture book" of corrosion types, which stores the feature vectors corresponding to various known corrosion types. For example, the feature vector of pitting may be [1,05,50], indicating that the pitting depth is shallow, the area percentage is small, but the number of holes is large. The goal of the meta-learning algorithm is to learn a mapping relationship from features to types, just like training a translator to translate the "language" of corrosion features into the "language" of corrosion types. Machine learning algorithms such as support vector machines (SVM) can be used to implement this mapping model. For example, if the pitting feature vector [1,05,50] is input, the model can output the type of "pitting". The adversarial learning algorithm is like a "sparring partner", which will generate some special samples to challenge the mapping model. These samples are called adversarial samples. They are slightly different from real corrosion samples, but they are enough to confuse the model. For example, it may generate a feature vector [15, 06, 45] that is very close to the feature vector of pitting corrosion, but it actually belongs to another type. By adding these adversarial samples to training, the robustness of the model can be improved, making it less likely to be "fooled". For new samples, their corrosion feature vectors can be extracted and then input into the trained mapping model to let the model predict their corrosion type. Transfer learning can help use existing knowledge to process new samples. For example, if the previously trained model is mainly for carbon steel corrosion, and now it is necessary to process stainless steel corrosion samples, transfer learning can help the model quickly adapt to new materials. The output results of the model are not 100% accurate. Confidence analysis is to evaluate the confidence of the model in each prediction result. For example, if the model predicts that a sample is pitting corrosion with a confidence of 9, it means that the model is 90% sure that it is pitting corrosion. If the confidence is lower than the preset threshold, such as 7, manual review is required, and experts are required to judge its type, which can improve the accuracy of classification. Finally, the classification results are returned to the business system, such as the corrosion management system, so that appropriate measures can be taken. At the same time, new samples and their predicted types are added to the training set, and the model is retrained regularly so that the model can continuously learn new knowledge and improve performance. This is like an "expert" who is constantly learning. As experience accumulates, the judgment becomes more and more accurate. For example, suppose there is a batch of corrosion samples from an oil pipeline, their corrosion feature vectors are extracted, and a corrosion type recognition model is trained using a meta-learning algorithm. Then, a new corrosion sample is obtained from the field, and its feature vector is [3,1,20]. This vector is input into the model, and the model predicts that it is uniform corrosion with a confidence of 85. Since the confidence is higher than the preset threshold of 7, the model's prediction result is accepted and the result is returned to the pipeline management system.At the same time, the sample and its predicted type (uniform corrosion) are added to the training set for subsequent model updates. After a period of time, a new batch of samples are obtained. The feature vector of one sample is [1,01,80]. The model predicts that it is pitting corrosion, but the confidence level is only 6. Since the confidence level is lower than the threshold, the sample is submitted to the corrosion expert for manual review. The expert confirms that the sample is indeed pitting corrosion and feeds the result back to the system. The results confirmed by the expert are added to the training set, and the model is retrained so that the model can better identify similar samples.

[0136] S104, analyzing the evolution process of the corrosion area by using time series data mining technology, including the following steps:

[0137] According to the corrosion classification results, the time series data of the corrosion area is obtained, and the evolution process of the corrosion area is analyzed by using the time series data mining technology;

[0138] By constructing a Markov random field model, the extracted spatiotemporal characteristics of corrosion development are used as observation data of the model;

[0139] The transfer probability matrix and emission probability matrix of the model are determined by using parameter estimation method, and a probability model reflecting the dynamic change law of corrosion is obtained.

[0140] The trained Markov random field model is used to predict the distribution of the corrosion area at the next moment according to the spatiotemporal characteristics of the current corrosion area.

[0141] If the predicted future corrosion trend exceeds the preset threshold, it is judged that the current corrosion area has a greater development risk and the early warning mechanism is triggered;

[0142] The environmental parameters and material property information of the corrosion area are obtained and used as auxiliary inputs of the Markov random field model.

[0143] As an optional implementation method of this embodiment, the time series data of the corrosion area is obtained according to the corrosion classification result, and the evolution process of the corrosion area is analyzed by using the time series data mining technology to extract the spatiotemporal characteristics that can characterize the development of corrosion. By constructing a Markov random field model, the extracted spatiotemporal characteristics of corrosion development are used as the observation data of the model, and the model is used to characterize the dynamic change process of corrosion. For the constructed Markov random field model, the parameter estimation method is used to determine the transfer probability matrix and the emission probability matrix of the model, and a probability model that can reflect the dynamic change law of corrosion is obtained. Using the trained Markov random field model, according to the spatiotemporal characteristics of the current corrosion area, the distribution of the corrosion area at the next moment is predicted to achieve the prediction of future corrosion trends. If the predicted future corrosion trend exceeds the preset threshold, it is judged that the current corrosion area has a greater development risk, the early warning mechanism is triggered, and corresponding protective measures are taken. The environmental parameters and material properties of the corrosion area are obtained and used as auxiliary inputs of the Markov random field model to improve the accuracy of corrosion trend prediction. According to the predicted future corrosion development trend, the corrosion area is risk assessed, the corrosion risk level of different areas is determined, and a visual corrosion risk distribution map is generated to provide decision support for subsequent anti-corrosion measures.

[0144] Specifically, the time series data of the corrosion area can be understood as a series of corrosion image data taken at different time points. The purpose of obtaining this data is to analyze the changes in the corrosion area over time. For example, a certain part of the offshore platform can be photographed at regular intervals to collect a sequence of corrosion images over a period of time. These images record the changes in the corrosion area, depth and other characteristics of the part. Using these time series data, time series data mining technology can be used to analyze the evolution of corrosion. For example, the rate of change of the corrosion area at different time points can be calculated to analyze the speed of corrosion development. Time series data mining technology can be used to extract spatiotemporal features that can characterize the development of corrosion. For example, the rate of change of the corrosion area over time, the shape change of the corrosion area, the spatial expansion speed of the corrosion area, etc. Assuming that the image data of a pipeline corrosion area within one year is obtained, the area of ​​the corrosion area each month can be extracted through image processing technology, and these area data can be combined into a time series. By analyzing this time series, it can be found that the corrosion area shows a linear growth trend, with a growth rate of 1 square centimeter per month. This growth rate is a spatiotemporal feature, which contains both time information (per month) and spatial information (area). The Markov random field model can be used to characterize the dynamic change process of corrosion. It assumes that the state of the corrosion area at the next moment is only related to its current state, but not to the previous state. The extracted spatiotemporal characteristics of corrosion development can be used as observation data for the model. For example, the corrosion area growth rate mentioned above is input into the Markov random field model as observation data. The parameter estimation method can be used to determine the transition probability matrix and emission probability matrix of the Markov random field model. The transition probability matrix describes the probability of the corrosion area transitioning from one state to another. The emission probability matrix describes the probability of observing a specific spatiotemporal feature in a given state. For example, assuming that the corrosion state is divided into three types: mild, moderate, and severe. The transition probability matrix can describe the probability of developing from mild corrosion to moderate corrosion, and from moderate corrosion to severe corrosion. The emission probability matrix describes the probability of observing different corrosion area growth rates under mild, moderate, and severe corrosion states. The trained Markov random field model can be used to predict the distribution of the corrosion area at the next moment. For example, based on the current corrosion area and growth rate, the area and corrosion state of the corrosion area in the next month can be predicted. Assuming that the current corrosion area is 10 square centimeters and the growth rate is 1 square centimeter per month, the model predicts that the corrosion area will be 11 square centimeters next month and the corrosion state will be mild. If the predicted future corrosion trend exceeds the preset threshold, for example, the corrosion area exceeds the safety threshold, it is judged that there is a greater development risk in the current corrosion area, triggering the early warning mechanism and taking corresponding protective measures. For example, if the corrosion area is predicted to reach 15 square centimeters next month and the safety threshold is 12 square centimeters, the system will issue an early warning to remind the staff to take protective measures, such as applying anti-corrosion paint or replacing damaged parts.Environmental parameters (such as temperature, humidity, salinity) and material properties (such as material composition, corrosion resistance) of the corrosion area can be used as auxiliary inputs of the Markov random field model to improve the accuracy of corrosion trend prediction. For example, in a marine environment, high salinity accelerates corrosion, so salinity information can be input into the model to more accurately predict corrosion trends. According to the predicted future corrosion development trend, the corrosion area can be risk assessed to determine the corrosion risk level of different areas. For example, the corrosion area is divided into three levels: low risk, medium risk and high risk. The evaluation results are used to generate a visual corrosion risk distribution map, with different colors representing areas of different risk levels, so as to intuitively understand the distribution of corrosion risks. This will provide decision support for subsequent anti-corrosion measures. For example, protective measures can be taken in high-risk areas as a priority.

[0145] S105, using visualization technology to generate a corrosion development map, including the following steps:

[0146] Obtain historical corrosion data and current corrosion status data of offshore platforms, pre-process the data, unify the data format and unit, and process missing values ​​and outliers;

[0147] Based on the preprocessed data, a time series prediction algorithm is used to establish a corrosion development prediction model to predict the corrosion distribution and severity in different time periods in the future;

[0148] The prediction results are visualized and color-coded to represent corrosion of different severity with different colors to generate a corrosion distribution map.

[0149] On the basis of the corrosion distribution map, a heat map is superimposed, and different colors and brightness are used to highlight the high-risk areas according to the severity of corrosion;

[0150] Compare and analyze the corrosion distribution maps and thermograms of different time periods to determine the corrosion development trend and rate, and identify areas where corrosion is accelerated or slowed down; comprehensively consider the corrosion distribution, severity and development trend factors, evaluate the corrosion risk level of various parts of the offshore platform, and generate a risk assessment report.

[0151] As an optional implementation method of this embodiment, the historical corrosion data and current corrosion status data of the marine platform are obtained, the data is preprocessed, the data format and unit are unified, and the missing values ​​and outliers are processed. According to the preprocessed data, a time series prediction algorithm, such as an ARIMA model or an LSTM neural network, is used to establish a corrosion development prediction model to predict the corrosion distribution and severity in different time periods in the future. The prediction results are visualized, and the corrosion of different severity is represented by different colors using a color coding method to generate an intuitive corrosion distribution map. On the basis of the corrosion distribution map, a heat map is superimposed, and different color depths and brightness are used to represent the high-risk areas according to the severity of the corrosion. The corrosion distribution map and the heat map of different time periods are compared and analyzed to determine the corrosion development trend and rate, and identify the areas where the corrosion is accelerated or slowed down. Comprehensively considering factors such as corrosion distribution, severity, and development trend, the corrosion risk level of each part of the marine platform is evaluated, and a risk assessment report is generated. According to the risk assessment results, targeted maintenance decision suggestions are automatically generated, such as recommended inspection frequency, maintenance priority, anti-corrosion measures, etc., to provide an intuitive and operational basis for the operation and maintenance of the marine platform.

[0152] Specifically, the offshore platform is in the seawater environment for a long time, and the corrosion problem seriously affects its safety and service life. In order to effectively predict and control corrosion, it is necessary to analyze and process the corrosion data of the platform. First, it is necessary to obtain the historical corrosion data of the offshore platform, for example, to obtain the corrosion thickness, corrosion area and other data at different locations and time points through sensors, regular inspection reports, corrosion monitoring systems and other methods. At the same time, it is also necessary to obtain the current corrosion status data, for example, to obtain the latest corrosion images, videos and other data through drone inspections, diver inspections and other methods. These data are usually in different formats and units, and need to be preprocessed, for example, converting data in different formats into a unified format, and converting data in different units into a unified unit (such as millimeters). At the same time, it is also necessary to process the missing values ​​and outliers in the data, for example, using interpolation methods, average methods and other methods to fill missing values, and using box plots and other methods to identify and remove outliers. The preprocessed data can be used to establish a corrosion development prediction model. Time series prediction algorithms, such as ARIMA models or LSTM neural networks, can be used to predict the distribution and severity of corrosion in different time periods in the future. For example, corrosion data from the past decade can be used to train an LSTM neural network model, which can then be used to predict corrosion conditions in each quarter of the next year. Assuming that the corrosion thickness data for a certain area over the past decade shows obvious periodic changes and a certain upward trend, the LSTM model can learn these patterns and predict that the corrosion thickness in the area will continue to increase in the next year and reach a peak in a specific quarter. The prediction results need to be visualized to intuitively understand the distribution and severity of corrosion. For example, color coding can be used to represent corrosion of different severity levels with different colors to generate an intuitive corrosion distribution map. For example, mild corrosion can be represented by green, moderate corrosion by yellow, and severe corrosion by red. On the basis of the corrosion distribution map, a heat map can also be superimposed, and different color depths and brightness can be used to highlight high-risk areas according to the severity of corrosion. For example, the most severely corroded area can be highlighted in dark red, the second most severely corroded area can be highlighted in orange, and so on. By comparing and analyzing the corrosion distribution map and heat map of different time periods, the trend and rate of corrosion development can be determined, and areas where corrosion is accelerated or slowed can be identified. For example, if the color of an area gradually changes from green to yellow and then to red in the past year, it means that the corrosion in that area is accelerating. Conversely, if the color of an area gradually changes from red to yellow and then to green, it means that the corrosion in that area is slowing down. Taking into account factors such as corrosion distribution, severity, and development trend, the corrosion risk level of each part of the offshore platform can be evaluated. For example, the corrosion risk level can be divided into three levels: low, medium, and high.Areas with severe and rapid corrosion can be assessed as high-risk areas, and areas with mild and slow corrosion can be assessed as low-risk areas. Finally, a risk assessment report is generated, which describes in detail the corrosion risk level of each area and the corresponding maintenance recommendations. According to the risk assessment results, targeted maintenance decision recommendations can be automatically generated. For example, for high-risk areas, it can be recommended to increase the inspection frequency, improve the maintenance priority, and take more effective anti-corrosion measures, such as replacing anti-corrosion coatings and installing cathodic protection systems. For low-risk areas, the inspection frequency can be appropriately reduced and the maintenance priority can be reduced. These recommendations can provide intuitive and operational basis for the operation and maintenance of offshore platforms, thereby improving maintenance efficiency, reducing maintenance costs, and ensuring the safe operation of the platform. For example, for a certain supporting structure of the platform, if the prediction model shows that the corrosion risk of the structure will reach a high risk level in the next year, the system can automatically generate recommendations to increase the inspection frequency of the structure, such as changing from once every six months to once every quarter, and recommend the purchase of anti-corrosion materials required for replacement in advance.

[0153] Based on this, an embodiment of the present invention provides a method for identifying and predicting marine platform corrosion based on machine vision. To address the problem that the corrosion state of marine platforms is difficult to accurately identify and predict, a multi-scale segmentation algorithm is used to obtain the corrosion area, extract multi-dimensional features such as texture, color and shape, and integrate prior knowledge in the field of materials science to construct a knowledge-enhanced feature representation. A corrosion classification model is established using small sample learning and adversarial training strategies, and the corrosion evolution law is analyzed in combination with time series data mining and Markov random field models to achieve accurate identification of corrosion types and prediction of development trends. Finally, a corrosion development map is generated through visualization technology to provide a scientific basis for marine platform maintenance decisions. The present invention significantly improves the accuracy and predictive ability of corrosion diagnosis, can effectively guide the preventive maintenance of marine platforms, and reduce safety hazards and maintenance costs.

[0154] Embodiment 2

[0155] Based on the same general inventive concept, the present invention also provides a marine platform corrosion identification and prediction system based on machine vision. Figure 3 As shown, the marine platform corrosion identification and prediction system based on machine vision provided by the present invention is described below. The marine platform corrosion identification and prediction system based on machine vision described below and the marine platform corrosion identification and prediction method based on machine vision described above can be referred to each other. The system includes:

[0156] Image acquisition module, data processing module, feature extraction module, model training module and visualization module;

[0157] The image acquisition module is used to obtain corrosion image data of the offshore platform;

[0158] The data processing module is used to pre-process the image data, use the Gaussian filter algorithm to remove noise, and then use the multi-scale segmentation algorithm to segment the image to obtain a set of candidate corrosion areas;

[0159] The feature extraction module is used to extract the texture features, color features and shape features of the corrosion area to form a multi-view corrosion feature vector;

[0160] The model training module is used to establish a corrosion type classification model and identify the corrosion type through knowledge-enhanced corrosion feature representation;

[0161] The visualization module is used to generate corrosion development maps, presenting the corrosion distribution and severity in different time periods in the form of color coding and heat maps.

[0162] It should be understood that the marine platform corrosion identification and prediction system based on machine vision provided in the embodiment of the present invention has all the advantages of the marine platform corrosion identification and prediction method based on machine vision provided in the above-mentioned embodiment.

[0163] The present invention also provides a computer terminal device, comprising:

[0164] one or more processors;

[0165] A memory, coupled to the processor, for storing one or more programs;

[0166] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of a method for identifying and predicting marine platform corrosion based on machine vision.

[0167] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for identifying and predicting marine platform corrosion based on machine vision are implemented.

[0168] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying and predicting marine platform corrosion based on machine vision, characterized in that: The following steps are involved: According to the offshore platform corrosion image dataset, a multi-scale segmentation algorithm is used to segment the image to obtain a set of candidate corrosion regions. The candidate regions are corrected through morphological operations, and the noise regions with too small areas are eliminated to obtain the refined corrosion regions. For the corrosion area, a number of feature vectors are extracted, and corrosion feature vectors of multiple perspectives are constructed based on the several feature vectors; On the basis of feature vectors, typical feature patterns of corrosion types are integrated, and features are optimized through similarity calculation to highlight the discriminative features of different corrosion types and form knowledge-enhanced corrosion feature representation. Through time series data mining technology, the evolution process of the corrosion area is analyzed, the spatiotemporal characteristics of corrosion development are extracted, and a Markov random field model is constructed to describe the dynamic changes of corrosion and predict future corrosion trends. Visualization technology is used to generate corrosion development maps, which present the corrosion distribution and severity in different time periods through color coding and heat maps.

2. The method according to claim 1, characterized in that The method of segmenting the image using a multi-scale segmentation algorithm based on the marine platform corrosion image dataset includes the following steps: The Gaussian filtering algorithm is used to pre-process the image to remove noise and obtain the image after noise suppression; A multi-scale watershed segmentation algorithm is used to obtain candidate corrosion area sets at different scales; The support vector machine algorithm is used to classify the candidate corrosion areas into two categories to obtain a preliminary screened set of corrosion areas; The noise area with too small area is removed by morphological opening operation, and the edge of the eroded area is corrected by closing operation to obtain the refined eroded area set.

3. The method according to claim 1, characterized in that The method of extracting texture features, color features and shape features from the corrosion area includes the following steps: The gray-level co-occurrence matrix algorithm is used to extract the texture features of the corrosion area and obtain the texture feature vector; The color histogram algorithm is used to extract the color distribution characteristics and obtain the color histogram feature vector; The color moment algorithm is used to extract color statistical features and obtain the color moment feature vector; The shape invariant features are extracted by using the invariant moment algorithm to obtain the invariant moment feature vector; The Fourier descriptor algorithm is used to extract the shape frequency domain features and obtain the Fourier descriptor feature vector; Each feature vector is fused to construct a multi-view erosion feature vector.

4. The method according to claim 1, characterized in that: The method of fusing typical characteristic patterns of corrosion types based on characteristic vectors includes the following steps: Obtain material corrosion data, obtain the feature vector of the corrosion image, and establish a typical feature pattern library of different corrosion types based on prior knowledge in the field of material science; Calculate the similarity between the feature vector and each corrosion type in the pattern library to obtain a similarity vector of the feature vector; If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the typical feature pattern of the type is merged into the feature vector; The support vector machine algorithm is used to train the corrosion type classification model with the knowledge-enhanced corrosion feature representation as input; Use the trained model to identify the corrosion type of the new corrosion image and output the recognition result.

5. The method according to claim 4, characterized in that If the similarity of a certain type in the similarity vector is greater than a preset threshold, the feature vector is judged to belong to the corrosion type, and the following steps are also included: A similarity vector of the feature vector is constructed according to the similarity values, and each element in the vector represents the degree of similarity with the corresponding type; the similarity values ​​in the similarity vector are sorted, and the corrosion type with the largest similarity is taken as the candidate type.

6. The method according to claim 4, characterized in that After training the corrosion type classification model, also include: For the corrosion feature vector of each sample, a meta-learning algorithm is used, with the corrosion feature vector as input and the corresponding corrosion type as output, to establish the corrosion type classification model as a feature-to-type mapping model; For the mapping model, adversarial samples are generated through the adversarial learning algorithm, and the adversarial samples are added to the training set to retrain the model; Obtain unlabeled new corrosion samples and extract the corrosion feature vector of each sample; Input the feature vector of the new sample into the enhanced mapping model to determine the corrosion type of each sample; Conduct confidence analysis on the corrosion type results output by the model, and manually review samples with confidence levels lower than the preset threshold to improve classification accuracy; Feedback the classification results, add new samples and their predicted types to the training set, retrain the model regularly, and complete continuous optimization of the model.

7. The method according to claim 1, characterized in that The method of analyzing the evolution of the corrosion area by using the time series data mining technology includes the following steps: According to the corrosion classification results, the time series data of the corrosion area is obtained, and the evolution process of the corrosion area is analyzed by using the time series data mining technology; By constructing a Markov random field model, the extracted spatiotemporal characteristics of corrosion development are used as observation data of the model; The transfer probability matrix and emission probability matrix of the model are determined by using parameter estimation method, and a probability model reflecting the dynamic change law of corrosion is obtained. The trained Markov random field model is used to predict the distribution of the corrosion area at the next moment according to the spatiotemporal characteristics of the current corrosion area. If the predicted future corrosion trend exceeds the preset threshold, the current corrosion area is judged to have a development risk and the early warning mechanism is triggered; The environmental parameters and material property information of the corrosion area are obtained and used as auxiliary inputs of the Markov random field model.

8. The method according to claim 1, characterized in that The method of generating a corrosion development map using visualization technology includes the following steps: Obtain historical corrosion data and current corrosion status data of offshore platforms, pre-process the data, unify the data format and unit, and process missing values ​​and outliers; Based on the preprocessed data, a time series prediction algorithm is used to establish a corrosion development prediction model to predict the corrosion distribution and severity in different time periods in the future; The prediction results are visualized and color-coded to represent corrosion of different severity with different colors to generate a corrosion distribution map. On the basis of the corrosion distribution map, a heat map is superimposed, and different colors and brightness are used to highlight the high-risk areas according to the severity of corrosion; Compare and analyze the corrosion distribution maps and thermograms of different time periods to determine the corrosion development trend and rate, and identify areas where corrosion is accelerated or slowed down; comprehensively consider the corrosion distribution, severity and development trend factors, evaluate the corrosion risk level of various parts of the offshore platform, and generate a risk assessment report.

9. A machine vision-based offshore platform corrosion identification and prediction system, characterized in that: The system is used to implement the method for identifying and predicting marine platform corrosion based on machine vision according to any one of claims 1 to 8, and comprises: an image acquisition module, a data processing module, a feature extraction module, a model training module and a visualization module; The image acquisition module is used to obtain corrosion image data of the offshore platform; The data processing module is used to pre-process the image data, use a Gaussian filter algorithm to remove noise, and then use a multi-scale segmentation algorithm to segment the image to obtain a set of candidate corrosion regions; The feature extraction module is used to extract texture features, color features and shape features of the corrosion area to form a multi-view corrosion feature vector; The model training module is used to establish a corrosion type classification model and identify the corrosion type through knowledge-enhanced corrosion feature representation; The visualization module is used to generate a corrosion development map, which presents the corrosion distribution and severity in different time periods in the form of color coding and heat map.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the method for identifying and predicting marine platform corrosion based on machine vision as described in any one of claims 1 to 8 is implemented.

Images