Method and system for real-time detection of metal corrosion depth based on image recognition algorithm

Abstract

The present invention discloses a method and system for real-time detection of metal corrosion depth based on image recognition algorithm. The image recognition algorithm is used to detect the corrosion depth, and a database of morphological features and corrosion depth of large metal components is constructed according to existing detection results, and the corrosion depth of metal components is fed back in real time by image matching. The long-term stable tracking and prediction of metal corrosion status and service life are guaranteed, and accurate and timely corrosion data are provided for real-time detection of clients, reducing the experimental work volume and work cost of on-site detection. Compared with traditional detection methods, this method can greatly reduce the frequency and workload of detection, and can detect the degree of corrosion more intelligently and in real time. The corrosion morphology is associated with the corrosion depth, and different morphologies have different depths to detect the corrosion depth of metal components, which reduces the workload of large-scale on-site detection and provides theoretical support for long-term detection.

Description

Technical Field

[0001] The present invention relates to the technical field of metal component corrosion detection, and in particular to a method and system for real-time detection of metal corrosion depth based on an image recognition algorithm. Background Art

[0002] Metal materials have the advantages of high physical strength, high stability, and excellent durability, and are widely used in the construction field. However, metal corrosion is the core problem that limits its service life. Especially in water conservancy and civil engineering buildings with long life cycles, regular and long-term monitoring of metal corrosion is extremely important.

[0003] Traditional metal corrosion detection mostly uses resistance method, ultrasonic thickness measurement method, potentiometric method and other methods, which are still applicable to conventional metal structures. However, for metal-coating systems with complex multiphase materials, traditional methods have problems such as complex and difficult on-site operation, and destructive testing affects normal use. Especially for metal structures in port and terminal structures, they must be inspected through grinding and polishing and ultrasonic thickness measurement. The operation is extremely difficult and requires a large number of inspection personnel and supporting equipment. In addition, the feedback time is long, the data obtained is small, and the level of intelligence is low.

[0004] Therefore, a new method for measuring the depth of metal corrosion is urgently needed. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method and system for real-time detection of metal corrosion depth based on image recognition algorithm, which detects the corrosion depth by utilizing the correspondence between the topographic image of the metal component and the corrosion depth.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] The method for real-time detection of metal corrosion depth based on an image recognition algorithm provided by the present invention comprises the following steps:

[0008] Obtaining a microscopic topographic image of the detection point;

[0009] Obtain the corrosion depth of the detection point;

[0010] Establishing a mesoscopic database based on the mesoscopic topography image and the corrosion depth of the detection point; the mesoscopic database is used to store the corresponding relationship between the detection point position and the corrosion depth;

[0011] Acquire a wide-angle topographic image of the metal component at the point to be inspected; the wide-angle topographic image is a microscopic topographic image including a plurality of inspection points;

[0012] Constructing a hierarchical tree-like matching model for a microscopic topography image by indexing detection points in a wide-angle topography image;

[0013] The corrosion depth is determined based on the mesomorphological images indexed by the mesoscopic database.

[0014] Furthermore, the detection points are set at different elevation positions on the metal component, and the microscopic topography image is obtained by collecting the detection points at different elevation positions according to a preset area.

[0015] Furthermore, the preset area is formed by setting a collection frame on the mask plate; detection points are set at different elevations on the same metal component, and the detection points of different metal components constitute a detection point matrix;

[0016] Furthermore, the corrosion depth of the detection point is obtained by measuring the thickness of the corresponding position of the detection point after grinding and polishing using an ultrasonic thickness gauge.

[0017] Furthermore, the hierarchical tree matching model adopts the FLANN matching algorithm.

[0018] Furthermore, the method further comprises the following steps:

[0019] A corrosion depth prediction model is constructed, wherein the corrosion depth prediction model is used to establish a corrosion depth prediction model based on the relationship between the microscopic morphology image and the corrosion depth; and the corrosion depth of the point to be detected at the next moment is predicted through the corrosion depth prediction model diagram.

[0020] Furthermore, the corrosion depth prediction model is established according to the following steps:

[0021] The processed mesomorphology images are obtained, and the RGB image of a single mesomorphology image is binarized by gradient thresholding. The color difference gradient is used as the step size and it is iterated repeatedly. Based on the continuously iterated mesomorphology images of steel substrate corrosion and the corrosion depth test results of field tests, the correspondence between the corrosion depth gradient and the mesomorphology of metal corrosion can be quantified. Based on the established correspondence, an in-depth analysis of the mesomorphology to be tested is performed.

[0022] Furthermore, the mesomorphology image is processed using the VGG16 algorithm, which includes 13 convolutions and 3 full connections. Through 13 convolution calculations, 5 maximum pooling, and 3 full connection calculations, the floating rust and residual coating elements in the mesomorphology image are finally captured.

[0023] The present invention provides a system for real-time detection of metal corrosion depth based on an image recognition algorithm, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor implements the above-mentioned method when executing the program.

[0024] The beneficial effects of the present invention are:

[0025] The present invention provides a method and system for real-time detection of metal corrosion depth based on an image recognition algorithm. This method detects the corrosion depth by using an image recognition algorithm, constructs a database of the morphological features and corrosion depth of large metal components based on existing detection results, and provides real-time feedback on the corrosion depth of metal components through image matching. This ensures long-term stable tracking and prediction of metal corrosion status and service life, provides accurate and timely corrosion data for real-time detection of clients, and reduces the volume of experimental work and work costs of on-site detection. Compared with traditional detection methods, this method can greatly reduce the frequency and workload of detection, and can detect the degree of corrosion more intelligently and in real time. By using the correlation between corrosion morphology and corrosion depth, different morphologies have different depths to detect the corrosion depth of metal components, reducing the workload of large-scale on-site detection, while providing theoretical support for long-term detection.

[0026] The method provided by the present invention obtains macroscopic, microscopic, and depth images through on-site image sampling, and then establishes corresponding models for the macroscopic, microscopic, and depth images. Image matching can be performed using drone cruise photos to find the macroscopic wide-angle image contained in the cruise photos, and then the microscopic morphology image can be found through the macroscopic wide-angle image. The corresponding average corrosion depth can be determined. Furthermore, the VGG16 algorithm can be used to remove factors unrelated to rust in the image, obtaining a large amount of image data. Through gradient binarization, the precise correspondence between microscopic morphology and corrosion depth can be determined. Based on this correspondence, the local corrosion depth of the cruise photos can be calculated, ultimately achieving a unified result of rust measurement and prediction.

[0027] Other advantages, objects, and features of the present invention will be described in part in the following description and, in part, will be apparent to those skilled in the art upon examination of the following or may be learned from practice of the present invention. The objects and other advantages of the present invention may be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the purpose, technical solutions and beneficial effects of the present invention more clear, the present invention provides the following drawings for illustration:

[0029] Figure 1 This is a flow chart of a method for real-time detection of metal corrosion depth based on image recognition algorithm.

[0030] Figure 2 Schematic diagram of sampling the rust appearance of metal structures for masking panels.

[0031] Figure 3 This is a microscopic morphology diagram of the test point.

[0032] Figure 4 Schematic diagram of the wide-angle morphology photograph of the metal component.

[0033] Figure 5 It is a hierarchical tree matching algorithm to optimize the logical diagram.

[0034] Figure 6 This is the convolutional neural network corrosion depth prediction model diagram.

[0035] Figure 7 It is a schematic diagram of the morphological image.

[0036] Figure 8 It is the VGG convolutional neural network morphology processing flow chart and effect diagram.

[0037] Figure 9 It is the gradient binarization effect diagram.

[0038] Figure 10 Schematic diagram of the microscopic picture to be tested.

[0039] Figure 11 It is the correction graph of the neural network prediction model.

[0040] Figure 12 This is a three-dimensional solid model diagram of the hydraulic structure.

[0041] Figure 13 Rendering of the entity simulation model.

[0042] Figure 14 This is a preview of the stage implementation of the method for real-time detection of metal corrosion depth based on image recognition algorithm. DETAILED DESCRIPTION

[0043] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0044] Example 1

[0045] like Figure 1 As shown, Figure 1 This is a flow chart of a method for real-time detection of metal corrosion depth based on an image recognition algorithm. The method provided in this embodiment for real-time detection of metal corrosion depth based on an image recognition algorithm includes the following steps:

[0046] Obtaining a mesoscopic topographic image of the detection point; in this embodiment, the detection points are set at different elevations on the metal component, and the mesoscopic topographic image is obtained by collecting data from the detection points at different elevations according to a preset area; the preset area is formed by setting a collection frame on the mask plate; detection points are set at different elevations on the same metal component, and the detection points of different metal components constitute a detection point matrix;

[0047] Obtain the corrosion depth of the test point; the corrosion depth of the test point in this implementation rate is obtained by measuring the thickness of the corresponding position of the test point after grinding and polishing with an ultrasonic thickness gauge;

[0048] Establishing a mesoscopic database based on the mesoscopic topography image and the corrosion depth of the detection point; the mesoscopic database is used to store the corresponding relationship between the detection point position and the corrosion depth;

[0049] Acquire a wide-angle topographic image of the metal component at the point to be inspected; the wide-angle topographic image is a microscopic topographic image including a plurality of inspection points;

[0050] Constructing a hierarchical tree-like matching model, wherein the hierarchical tree-like matching model is used to index the mesomorphological image of the detection point in the wide-angle topography image; determining the corrosion depth based on the mesomorphological image indexed in the mesoscopic database;

[0051] This embodiment obtains the corrosion depth by collecting image data on site and polishing each inspection point of the metal component; establishes a macro database and a wide-angle feature database by collecting wide-angle morphology images; establishes a micro morphology database and a micro feature database by collecting micro morphology images; and finally constructs the information platform backend;

[0052] When it is necessary to obtain the corrosion depth of the detection target, the corrosion depth of the detection target can be indexed from the information platform backend by building the information platform frontend. The specific process is as follows:

[0053] Wide-angle topographic images of the inspection target are collected by drone patrol or workers at fixed points. The hierarchical tree matching algorithm is used to quickly index the micro-topographic images from the macro-database to determine the location of the metal structure inspection point. The correspondence between the inspection point position and the corrosion depth is then indexed from the micro-database. Finally, the corrosion depth of the inspection target is obtained and displayed on the front desk of the information platform.

[0054] Example 2

[0055] like Figures 2 to 4 As shown, the method for real-time detection of metal corrosion depth based on image recognition algorithm provided in this embodiment includes the following steps:

[0056] Step 1: On-site work:

[0057] 1. Select a specific sampling mask panel to take photos of the rusted appearance of the metal structure and select the points to be tested; Figure 2 As shown, Figure 2 Schematic diagram of sampling the rust appearance of metal structures for masking panels;

[0058] 2. Use an angle grinder to grind and polish the peeled coating, rust and other impurities at the polishing points;

[0059] 3. Use an ultrasonic thickness gauge to test the remaining thickness of the test point and obtain the corrosion depth based on relevant calculations;

[0060] Step 2: Detailed database establishment:

[0061] 1. Take photos of the measured points by field sampling and adjust them to the original size of 10*10, such as Figure 3 As shown, Figure 3 It is the microscopic morphology of the point to be measured;

[0062] 2. Use the sift algorithm to extract the feature point information of the microscopic morphology photo, and store the feature point information in a digital form as a TXT file for subsequent call processing; the Sift algorithm execution steps in this embodiment are as follows:

[0063] The Sift algorithm uses Gaussian filters of different scales and templates to process the image, and then performs a difference process to construct a Gaussian difference pyramid. This ensures its scale invariance and constructs the scale space of the image.

[0064] The Sift algorithm selects extreme points in the Gaussian difference pyramid as key points representing image features, and performs contrast and edge tests on the pre-selected feature points to eliminate those with low contrast and low noise robustness. This ensures the accuracy of their positions and precisely locates the feature points.

[0065] The Sift algorithm collects the gradient and direction distribution characteristics of the pixels in the neighborhood window of the Gaussian pyramid image where the feature point is located, and uses the histogram to count the gradient and direction of the pixels in the neighborhood as the main direction of the feature point. This ensures its rotation invariance and determines the direction of the feature point.

[0066] Combining the above three standard elements, the SIFT algorithm describes the feature points through descriptors, and can store them as data in a txt file to construct a feature database;

[0067] 3. By arranging the test calculation results of the ultrasonic thickness gauge, the corrosion depth is associated with its microscopic morphology photos, and the corresponding relationship between the photos and thickness is achieved by controlling the naming;

[0068] Step 3: Macro database establishment:

[0069] 1. Use drone cruising and camera photography to obtain wide-angle topographical photos of each metal component;

[0070] like Figure 4 As shown, Figure 4 Schematic diagram of wide-angle topographic photograph of metal components;

[0071] 2. Use the sift algorithm to extract the feature point information of the macro wide-angle morphology photo and store the feature point information in a digital form as a TXT file;

[0072] 3. Using the FLANN matching algorithm and combining it with on-site field records, the macro-wide-angle morphology photos and the micro-morphology photos are associated to achieve the corresponding relationship between the wide-angle morphology photos, the micro-morphology photos, and the corrosion depth;

[0073] The process of the Flann algorithm in this embodiment is as follows: Based on the feature point information extracted by the sift algorithm, the FLANN algorithm constructs a KD tree, inserts the feature point information into it, and uses the nearest neighbor search method to find the feature point information closest to the matching target point and calculates its matching similarity;

[0074] like Figure 5 As shown, Figure 5 This is a schematic diagram of the optimization logic of the hierarchical tree matching algorithm. This embodiment provides a hierarchical tree matching algorithm to achieve a fast matching process and optimize the matching logic of the exhaustive matching method. The existing exhaustive matching method matches the target image with each image in the microscopic image database until the matching process is completed;

[0075] This hierarchical tree-based matching algorithm uses wide-angle topography images as intermediate indexes. The specific matching process is as follows: using the similarity threshold of the FLANN algorithm, the feature point information of the macro-wide-angle topography image and the micro-topography image in the database are compared, and the database images that meet the similarity are selected to establish the correspondence between the macro-wide-angle topography image and the micro-topography image;

[0076] Based on the correspondence between the measuring points on the microscopic morphology image and the rust results of the grinding test established in step 2, the correspondence between the three databases will be established.

[0077] FLANN matching is used to compare the FLANN matching algorithm with the field operation records.

[0078] It can greatly accelerate the matching process of the target image. By matching with the wide-angle morphological image information library, the position of the target image and related morphological information can be quickly located. And through the established correspondence between the wide-angle morphological image and the microscopic morphological image, the corresponding microscopic morphological image can be quickly found, which greatly reduces the matching time of the target image.

[0079] Step 4: Based on the established macro-micro morphology image database, select some structural morphologies including floating rust, residual coating, impurities, and steel substrate, such as Figure 7 As shown, Figure 7 is a schematic diagram of the morphology image, Figure 7In the figure, the yellow-brown elements are rust products, mainly Fe2O3, the gray part is the residual coating with a significant scaly texture, and the black part is the steel substrate part with severe rust.

[0080] Using color and texture as the selection features, the established database of micromorphological images is used as the sample first-stage training set. The VGG16 algorithm is used to filter out the coating and floating rust factors in the micromorphological images.

[0081] Step 4: Based on the established microscopic image database and corrosion depth database, a convolutional neural network corrosion depth prediction model is constructed. The VGG16 neural network in this embodiment does not need to be modified too much. The VGG16 algorithm includes the first layer, the second layer, the third layer, the fourth layer, and the fifth layer. The specific steps of the VGG16 algorithm are as follows:

[0082] The input image is resized to 224×224×3, and convolution is performed twice using a 3×3 convolution kernel with 64 channels and a stride of 1. The output size is 224×224×64, completing the convolution operation of the first layer.

[0083] The output of the first layer is max-pooled with a filter of 2×2 and a stride of 2, halving the image size to 112×112×64. After two convolutions with 128 3×3 convolution kernels, the output size becomes 112×112×128, completing the convolution operation of the second layer.

[0084] The output of the second layer is max-pooled, and the size after pooling becomes 56×56×128. After 256 3×3 convolution kernels and three convolutions, the output size becomes 56×56×256, completing the convolution operation of the third layer;

[0085] The output of the third layer is pooled to a size of 28×28×256. After three convolutions with 512 3×3 convolution kernels, the output size is reduced to 28×28×512, completing the convolution operation of the fourth layer.

[0086] The output of the fourth layer is pooled using max pooling, and the size after pooling becomes 14×14×512. After 512 3×3 convolution kernels and three convolutions, the output size becomes 14×14×512, completing the convolution operation of the fifth layer.

[0087] The output of the fifth layer is pooled using max pooling, and the size after pooling becomes 7×7×512. The pooling result is converted into a one-dimensional vector using the Flatten dimensionality reduction function;

[0088] After two layers of 1×1×4096 and one layer of 1×1×3 fully connected layers (three layers in total), the softmax function is used to classify the images into three categories: floating rust, corroded steel substrate, and residual coating.

[0089] This convolutional neural network model uses VGG16, including 13 convolutions and 3 full connections. Through 13 convolution calculations, 5 maximum pooling, and 3 full connection calculations, it finally captures the floating rust and residual coating elements in the microscopic morphology image. The example calculation and analysis results are shown in the figure. The floating rust and residual coating elements are extracted through the trained calculation results, and the floating rust and residual coating elements that affect the calculation in the next stage are removed from the original image. Figure 8 As shown in (a) in the figure, a microscopic morphology image is obtained that only retains the elements of the corroded steel substrate, such as Figure 8 As shown in (b), Figure 8 (b) is a schematic diagram after morphological processing. Based on the trained VGG algorithm, influencing factors such as rust and residual coating are filtered out to obtain the microscopic morphological elements of the exposed steel substrate.

[0090] Since the floating rust and residual coating are colored, they can be removed according to their own color to obtain the rust layer of the steel substrate. Rust layers of different depths have different grayscale values. After processing the wide-angle topography image, areas with different grayscale values ​​are obtained. Generally speaking, rust layer areas of various depths can be obtained from the wide-angle topography image, and the rust depth can be obtained based on the rust layer in each grayscale value area.

[0091] Step 5: Based on the microscopic morphology image with the influencing factors filtered out in step 4, the single RGB image is binarized according to different color gradients. The iterative process is performed with the color gradient as the step size to gradually filter out areas of different grayscales. After removing the floating rust, impurities and light effects, only the solid rust is retained. The rust depth is calculated based on the actual thickness values ​​corresponding to different color gradients. The calculation formula is as follows:

[0092] D=∑d i =dn

[0093] Where D is the total corrosion depth, d i is the rust depth corresponding to each color gradient, d is the average rust depth, and n is the number of color gradient layers. The following table shows the preliminary corresponding relationship for this example.

[0094] Chromatic gradient layer Color difference gradient grayscale threshold Chromatic difference gradient depth threshold 1 220～240 0.02mm 2 200～220 0.04mm 3 180～200 0.06mm 4 160～180 0.08mm 5 140～160 0.1mm 6 120～140 0.12mm

[0095] Generally speaking, after eliminating the influence of irrelevant variables, the darker the color of the rust layer, the deeper the rust; Figure 9 As shown, Figure 9This is the gradient threshold binarization processing effect diagram. Based on the continuously iterative steel substrate rust micromorphology images and the rust depth test results of the field test, the corresponding relationship between the rust depth gradient and the micromorphology of the metal rust can be quantified. Finally, a model relationship is obtained, in which the color difference gradient of different colors corresponds to different rust depths.

[0096] Based on the established correspondence, we conduct in-depth analysis of the microscopic images to be tested. The results are as follows: Figure 10 As shown, the dark purple is the deepest rusted area, with a rust depth of about 0.07 mm, and the yellow is the shallower rusted area, with a rust depth of 0.03 mm.

[0097] Step 6: Based on the corrosion depth estimation algorithm established in step 5, the field data from annual or regular large-scale inspections are used to continuously revise and calibrate the model, improve the relevant parameters of the model, form a complete corrosion depth prediction development theoretical model, and build a three-in-one intelligent platform integrating image information matching, image corrosion model, and safety assessment prediction. Further optimize the corrosion depth prediction results to ensure the accuracy of the final results, such as Figure 11 As shown, Figure 11 It is a rust development prediction map.

[0098] Step 7: Build the testing platform

[0099] 1. Establish a three-tier database correspondence and encapsulate the corresponding algorithm. The input object is any photo that meets basic accuracy requirements, and the output object is the corrosion depth of all measurements on the photo and the overall corrosion depth assessment results of the metal structure in the photo.

[0100] 2. Establish a three-dimensional solid model of the entire hydraulic structure, and use the collected macroscopic wide-angle morphological photos as textures to construct a rendering solid simulation model; Figure 12 As shown, Figure 12 It is a three-dimensional solid model diagram of the hydraulic structure; Figure 12 and Figure 13 To preview the platform display status, the present invention will continuously capture the surface topography images of each component and, based on the BIM platform, attach the images as textures to the 3D model of a large port terminal building, thereby forming a 3D preview overview presented on the display terminal of the present invention's technical platform. Figure 12 and Figure 13 The following is a schematic diagram of the attachment of some maps. When the subsequent test data is in place, a simulation model with 1:1 real-time feedback between the site and the model will be formed, such as Figure 13 As shown, Figure 13 Rendering of the entity simulation model.

[0101] 3. Combining the two main functions, a user interaction interface is constructed to achieve real-time detection and feedback of corrosion depth by providing photos or selecting simulation model structures.

[0102] like Figure 14 As shown, Figure 14 This is a preview of the stage implementation of the method for real-time detection of metal corrosion depth based on image recognition algorithm. Figure 14 This is the feedback result diagram of each background step (such as step 2 and step 3) after the present invention is actually put into use, which can prove the feasibility and authenticity of the present invention. In actual application, the present invention will rely on drone cruising to capture the overall morphology and the morphology of each structural segment of the port hydraulic structure, compare the feature points of the captured target image with the image in the macro database, and then filter out the microscopic morphology image contained therein and the rust depth corresponding to the measuring point based on the corresponding relationship.

[0103] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A method for real-time detection of metal corrosion depth based on image recognition algorithm, characterized in that: The following steps are involved: Obtain a microscopic morphological image of the detection point; Obtain rust depth samples at the inspection points; A mesoscopic database is established based on the mesoscopic morphology images and corrosion depth of the detection points; The microscopic database is used to store the corresponding relationship between the detection point position and the corrosion depth; Acquire wide-angle topographic images of the points to be inspected on the metal component; The wide-angle topographic image is a microscopic topographic image including a plurality of detection points; Constructing a hierarchical tree matching model, the hierarchical tree matching model is used to index the microscopic topography image of the detection point in the wide-angle topography image; The corrosion depth is determined by indexing the mesomorphological images in the mesoscopic database; The following steps are also included: Construct a corrosion depth prediction model based on the relationship between the microscopic morphology image and the corrosion depth; use the corrosion depth prediction model to predict the corrosion depth of the point to be detected at the next moment; The corrosion depth prediction model is established according to the following steps: Obtain processed mesomorphological images, perform gradient threshold binarization on the RGB image of a single mesomorphological image, and iterate using the color difference gradient as the step size. Based on the continuously iterated mesomorphological images of steel substrate corrosion and the corrosion depth test results of the field test, obtain the corresponding relationship between the corrosion depth gradient and the mesomorphology of metal corrosion. Based on the established corresponding relationship, calculate and analyze the mesomorphological images to be tested to obtain the corrosion depth value. The corrosion depth is obtained by the color difference gradient and calculated according to the following formula: in, is the total corrosion depth, is the rust depth corresponding to each color difference gradient, is the average corrosion depth, is the number of color gradient layers.

2. The method for real-time detection of metal corrosion depth based on image recognition algorithm according to claim 1, characterized in that: The detection points are arranged at different elevation positions on the metal component, and the microscopic topography image is obtained by collecting the detection points at different elevation positions according to a preset area.

3. The method for real-time detection of metal corrosion depth based on image recognition algorithm as claimed in claim 2, characterized in that: The preset area is formed by setting a collection frame on the mask plate; detection points are set at different elevations on the same metal component, and the detection points of different metal components constitute a detection point matrix.

4. The method for real-time detection of metal corrosion depth based on image recognition algorithm according to claim 1, characterized in that: The corrosion depth of the detection point is obtained by measuring the thickness of the corresponding position of the detection point after grinding and polishing with an ultrasonic thickness gauge.

5. The method for real-time detection of metal corrosion depth based on image recognition algorithm according to claim 1, characterized in that: The hierarchical tree matching model adopts the FLANN matching algorithm.

6. A system for real-time detection of metal corrosion depth based on an image recognition algorithm, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 5 is implemented.

Images