CN115953410B - Corrosion pit automatic detection method based on target detection supervised learning - Google Patents
Corrosion pit automatic detection method based on target detection supervised learning Download PDFInfo
- Publication number
- CN115953410B CN115953410B CN202310248341.XA CN202310248341A CN115953410B CN 115953410 B CN115953410 B CN 115953410B CN 202310248341 A CN202310248341 A CN 202310248341A CN 115953410 B CN115953410 B CN 115953410B
- Authority
- CN
- China
- Prior art keywords
- corrosion
- target detection
- pits
- pit
- hemispherical
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic corrosion pit detection method based on target detection supervised learning, which automatically identifies and analyzes corrosion pits on objects (such as containers and pipelines) by using a method based on three-dimensional point cloud and automatically judges whether the corrosion pits meet the requirements or not, and comprises the following steps: 1) Automatically generating a large amount of corrosion pit data according to the characteristics of the corrosion pits; 2) Performing mixed training on the target detection model by utilizing the automatically generated corrosion pit data and the real corrosion pit data; 3) And automatically identifying the corrosion pit by using the trained target detection model.
Description
Technical Field
The invention relates to the technical field of detection, in particular to an automatic detection method for a corrosion pit based on target detection supervised learning.
Background
According to the requirements in the safety technical specifications of special equipment, the pressure container can have corrosion surface and pit phenomena after long-term use, objects (such as containers and pipelines) need to be detected regularly, whether corrosion is allowed or not is calculated according to the specifications in the technical specifications, and if the corrosion is not allowed, the objects (such as containers and pipelines) need to be scrapped.
Generally, the corrosion pits are measured by adopting a manual measurement mode, so that great manpower and material resources are consumed, and the working efficiency is extremely low.
Disclosure of Invention
At present, a main difficulty in identifying corrosion pits on objects (such as containers and pipelines) by using a deep learning model is that three-dimensional point cloud data of the objects with the corrosion pits are lacking, and if the data volume is insufficient, the trained accuracy and robustness of automatically identifying the corrosion pits on the objects are also insufficient.
The invention is realized by the following technical scheme: an automatic corrosion pit detection method based on target detection supervised learning comprises the following steps:
1) Automatically generating a large amount of corrosion pit data according to the corrosion pit characteristics;
2) Performing mixed training on the target detection model by utilizing the automatically generated corrosion pit data and the real corrosion pit data;
3) And automatically identifying the corrosion pit by using the trained target detection model.
Further, in order to better realize the automatic corrosion pit detection method based on target detection supervised learning, the following arrangement mode is adopted: said step 1) comprises the steps of:
1.1 Generating a point cloud on the surface of an ideal object (such as a container and a pipeline), wherein the radius range of the point cloud is 50-1000 mm according to the actual situation;
1.2 In actual operation, the binocular camera is not necessarily absolutely perpendicular to the surface of an object (such as a container or a pipeline) during use, so that the authenticity of the generated data is improved by adopting the following modes: rotating the surface point cloud of an ideal object (such as a container and a pipeline) around a y axis and a z axis, wherein the random angle range of rotation is set to be +/-15 degrees;
1.3 Converting the surface point cloud of the rotated ideal object (such as a container and a pipeline) into a depth image, and changing the z-axis into the image depth of the depth image;
1.4 Generating a hemispherical corrosion pit three-dimensional point cloud according to the depth and the size of the corrosion pit in practice;
1.5 A random rotation of the hemispherical etch pits due to the random orientation of the etch pits on the walls of the object (e.g., vessel, pipe);
1.6 Converting the rotated hemispherical etch pit into a depth image, and changing the z-axis into an image depth of the depth image;
1.7 Because the size of the corrosion pit is random, the aspect ratio of the hemispherical corrosion pit is randomly changed, and the ratio of the width to the height is 5-0.5;
1.8 Since the pits are not all round or oval in shape, the shapes of the hemispherical pits are changed by using the outline templates of the various targets in the coco data set to obtain various shapes of the hemispherical pits;
1.9 Adding or subtracting a depth image of the hemispherical etch pit to or from a region of a depth image of a surface of a desired object (e.g., vessel, pipe); wherein, the addition is a convex corrosion pit, and the subtraction is a concave corrosion pit;
1.10 Cycling the steps 1.4) to 1.9), and repeatedly adding hemispherical corrosion pits with various shapes until the number of the hemispherical corrosion pits with various shapes reaches a set value, and storing the positions of all the hemispherical corrosion pits with various shapes;
1.11 To prevent overlapping between corrosion pits, overlapping data needs to be deleted: requiring generated data, wherein corrosion pits on each object (such as a container and a pipeline) cannot be overlapped with each other, so that the iou value of each corrosion pit and other corrosion pits on the object (such as the container and the pipeline) is calculated, if the iou value is greater than 0, the corrosion pits are deleted, and the corrosion pits overlapped with each other are deleted;
1.12 After the overlapping etch pits on the objects (e.g., containers, pipes) are deleted in step 1.11), the depth image of the object without the overlapping etch pits is stored as a 32-bit image (with insufficient precision if stored as an 8-bit image), and the non-overlapping etch pit position data on all the objects (e.g., containers, pipes) is stored in the corresponding json file.
Further, in order to better realize the automatic corrosion pit detection method based on target detection supervised learning, which is disclosed by the invention, the same points as the technical scheme are not repeated here: said step 2) comprises the steps of:
2.1 Initializing the weight of the target detection model;
2.2 The weight of the target detection model pre-trained in the coco data set is adopted for transfer learning, so that the model robustness can be greatly improved.
2.3 The initial learning rate is 0.0001, the automatically generated corrosion pit data is used for carrying out the coarse training of the target detection model, and noise with different scales is required to be added to the automatically generated corrosion pit data in the training process so as to simulate the state that a real object (such as a container and a pipeline) is not absolutely smooth;
2.4 After the target detection model is trained to 40 epochs, the learning rate is reduced to 0.00001, and at the moment, automatically generated corrosion pit data and real corrosion pit data are input in equal quantity for training;
2.5 After the target detection model is trained to 5epoch, the loss tends to stably stop the target detection model training.
Further, in order to better realize the automatic corrosion pit detection method based on target detection supervised learning, which is disclosed by the invention, the same points as the technical scheme are not repeated here: the data volume ratio of the automatically generated corrosion pit data to the real corrosion pit data is as follows: 1000:1.
further, in order to better realize the automatic corrosion pit detection method based on target detection supervised learning, which is disclosed by the invention, the same points as the technical scheme are not repeated here: the target detection model adopts a yolov4 network structure.
Further, in order to better realize the automatic corrosion pit detection method based on target detection supervised learning, which is disclosed by the invention, the same points as the technical scheme are not repeated here: said step 3) comprises the steps of:
3.1 Shooting the surface of the object to be detected by using a depth camera to generate a three-dimensional point cloud of the surface of the object (such as a container and a pipeline);
3.2 Intercepting a 30cm multiplied by 30cm size of a central area of a three-dimensional point cloud of the surface of an object (such as a container and a pipeline) and converting the three-dimensional point cloud into a depth image;
3.3 Preprocessing the depth image obtained in step 3.2), dividing the depth image by 250;
3.4 Inputting the target detection model obtained in the step 3.3) into the trained target detection model in the step 2), analyzing the depth image, and returning all mark frames related to the corrosion pit;
3.5 The depth of each etch pit mark frame is determined as the etch pit depth, and the length and width of the mark frame is the etch pit length and width.
Compared with the prior art, the invention has the following advantages:
the method automatically identifies and analyzes the corrosion pits on the objects (such as containers and pipelines) by using a method based on three-dimensional point cloud, and automatically judges whether the corrosion pits meet the requirements.
The invention can improve the working efficiency of workers and the detection precision.
According to the invention, when the actual three-dimensional point cloud data of the object is applied, the object defect can be detected by training out the target detection model without manually marking the data.
According to the method, three-dimensional point cloud data of the object (such as a container and a pipeline) with the corrosion pit are automatically generated through program simulation, and a deep learning model can be trained to robustly and automatically identify the corrosion pit on the object (such as the container and the pipeline) without or with a small amount of manual labeling data.
Detailed Description
The present invention will be described in further detail with reference to examples, but embodiments of the present invention are not limited thereto.
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, based on the embodiments of the invention, which are apparent to those of ordinary skill in the art without inventive faculty, are intended to be within the scope of the invention. Thus, the following detailed description of the embodiments of the invention is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.
In the present invention, unless explicitly specified and limited otherwise, the terms "mounted," "connected," "disposed," "deployed," "secured," and the like are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally formed, particularly by means other than by screwing, interference fit, riveting, screw-assisted connection, and the like, in any of a variety of conventional mechanical connection means. The specific meaning of the above terms in the present invention can be understood by those of ordinary skill in the art according to the specific circumstances.
In the present invention, unless expressly stated or limited otherwise, a first feature "above" or "below" a second feature may include both the first and second features being in direct contact, as well as the first and second features not being in direct contact but being in contact with each other through additional features therebetween. Moreover, a first feature being "above," "over" and "on" a second feature includes the first feature being directly above and obliquely above the second feature, or simply indicating that the first feature is higher in level than the second feature. The first feature being "under", "below" and "beneath" the second feature includes the first feature being directly under and obliquely below the second feature, or simply means that the first feature is less level than the second feature.
Noun interpretation
coco dataset: the data includes an image, and a dataset of positional information for various objects in the image.
json file: a file storage format.
iou value: a measure of the distance between the location of the object identified by the deep learning model and the location of the real object.
loss: loss value.
yolov4 network: a deep learning model for detecting the position of a target in an image.
epoch: and twice.
Example 1
The automatic corrosion pit detection method based on target detection supervised learning solves the problem of insufficient data, trains a deep learning model by utilizing a mode of mixing and training automatically generated data and real data, achieves the effect of automatically detecting the corrosion pit by utilizing the deep learning model, and comprises the following steps of:
1) Automatically generating a large amount of corrosion pit data according to the corrosion pit characteristics;
2) Performing mixed training on the target detection model by utilizing the automatically generated corrosion pit data and the real corrosion pit data;
3) And automatically identifying the corrosion pit by using the trained target detection model.
Example 2
The embodiment is further optimized on the basis of the embodiment, and the same points as the technical scheme are not repeated here, so that the automatic corrosion pit detection method based on target detection supervision learning is further realized better, and particularly the following setting mode is adopted: said step 1) comprises the steps of:
1.1 Generating a point cloud on the surface of an ideal object (such as a container and a pipeline), wherein the radius range of the point cloud is 50-1000 mm according to the actual situation;
1.2 In actual operation, the binocular camera is not necessarily absolutely perpendicular to the surface of an object (such as a container or a pipeline) during use, so that the authenticity of the generated data is improved by adopting the following modes: rotating the surface point cloud of an ideal object (such as a container and a pipeline) around a y axis and a z axis, wherein the random angle range of rotation is set to be +/-15 degrees;
1.3 Converting the surface point cloud of the rotated ideal object (such as a container and a pipeline) into a depth image, and changing the z-axis into the image depth of the depth image;
1.4 Generating a hemispherical corrosion pit three-dimensional point cloud according to the depth and the size of the corrosion pit in practice;
1.5 A random rotation of the hemispherical etch pits due to the random orientation of the etch pits on the walls of the object (e.g., vessel, pipe);
1.6 Converting the rotated hemispherical etch pit into a depth image, and changing the z-axis into an image depth of the depth image;
1.7 Because the size of the corrosion pit is random, the aspect ratio of the hemispherical corrosion pit is randomly changed, and the ratio of the width to the height is 5-0.5;
1.8 Since the pits are not all round or oval in shape, the shapes of the hemispherical pits are changed by using the outline templates of the various targets in the coco data set to obtain various shapes of the hemispherical pits;
1.9 Adding or subtracting a depth image of the hemispherical etch pit to or from a region of a depth image of a surface of a desired object (e.g., vessel, pipe); wherein, the addition is a convex corrosion pit, and the subtraction is a concave corrosion pit;
1.10 Cycling the steps 1.4) to 1.9), and repeatedly adding hemispherical corrosion pits with various shapes until the number of the hemispherical corrosion pits with various shapes reaches a set value, and storing the positions of all the hemispherical corrosion pits with various shapes;
1.11 To prevent overlapping between corrosion pits, overlapping data needs to be deleted: we require the data generated that the etch pits on each object do not overlap each other, thus calculating the iou value of each etch pit on the object (e.g., container, pipe) with other etch pits, and if the iou value is greater than 0, deleting the etch pit, thus deleting the etch pits that overlap each other;
1.12 After the superimposed etch pits on the object are deleted in step 1.11), the depth image of the object without superimposed etch pits is stored as a 32-bit image (with insufficient precision if stored as an 8-bit image), and the non-superimposed etch pit position data on all objects (e.g., containers, pipes) are stored in the corresponding json file.
Example 3
The present embodiment is further optimized based on any one of the foregoing embodiments, and the same parts as the foregoing technical solutions are not repeated herein, and further for better implementing the method for automatically detecting a corrosion pit based on target detection supervised learning described in the present invention, the same parts as the foregoing technical solutions are not repeated herein: said step 2) comprises the steps of:
2.1 Initializing the weight of the target detection model;
2.2 The weight of the target detection model pre-trained in the coco data set is adopted for transfer learning, so that the model robustness can be greatly improved.
2.3 The initial learning rate is 0.0001, the automatically generated corrosion pit data is used for carrying out the coarse training of the target detection model, and noise with different scales is required to be added to the automatically generated corrosion pit data in the training process so as to simulate the state that a real object (such as a container and a pipeline) is not absolutely smooth;
2.4 After the target detection model is trained to 40 epochs, the learning rate is reduced to 0.00001, and at the moment, automatically generated corrosion pit data and real corrosion pit data are input in equal quantity for training;
2.5 After the target detection model is trained to 5epoch, the loss tends to stably stop the target detection model training.
Example 4
The present embodiment is further optimized based on any one of the foregoing embodiments, and the same parts as the foregoing technical solutions are not repeated herein, and further for better implementing the method for automatically detecting a corrosion pit based on target detection supervised learning described in the present invention, the same parts as the foregoing technical solutions are not repeated herein: the data volume ratio of the automatically generated corrosion pit data to the real corrosion pit data is as follows: 1000:1.
example 5
The present embodiment is further optimized based on any one of the foregoing embodiments, and the same parts as the foregoing technical solutions are not repeated herein, and further for better implementing the method for automatically detecting a corrosion pit based on target detection supervised learning described in the present invention, the same parts as the foregoing technical solutions are not repeated herein: the target detection model adopts a yolov4 network structure.
Example 6
The present embodiment is further optimized based on any one of the foregoing embodiments, and the same parts as the foregoing technical solutions are not repeated herein, and further for better implementing the method for automatically detecting a corrosion pit based on target detection supervised learning described in the present invention, the same parts as the foregoing technical solutions are not repeated herein: said step 3) comprises the steps of:
3.1 Shooting the surface of an object to be detected (such as a container and a pipeline) by using a depth camera to generate an object surface three-dimensional point cloud;
3.2 Intercepting a 30cm multiplied by 30cm size of a central area of a three-dimensional point cloud of the surface of an object (such as a container and a pipeline) and converting the three-dimensional point cloud into a depth image;
3.3 Preprocessing the depth image obtained in step 3.2), dividing the depth image by 250;
3.4 Inputting the target detection model obtained in the step 3.3) into the trained target detection model in the step 2), analyzing the depth image, and returning all mark frames related to the corrosion pit;
3.5 The depth of each etch pit mark frame is determined as the etch pit depth, and the length and width of the mark frame is the etch pit length and width.
Example 7
The present embodiment is further optimized based on any one of the above embodiments, and the same features as the foregoing technical solutions are not described herein in detail, and the method for automatically detecting a corrosion pit based on target detection supervised learning includes the following steps:
(1) Generating an ideal pipeline surface point cloud, wherein the radius range of the ideal pipeline surface point cloud is 50-1000 mm according to the actual situation;
(2) In actual operation, the binocular camera is not necessarily absolutely perpendicular to the surface of the pipeline in the use process, so that the authenticity of the generated data is improved by adopting the following mode: rotating the ideal pipeline surface point cloud around a y axis and a z axis, wherein the random angle range of rotation is set to be +/-15 degrees;
(3) Converting the rotated ideal pipeline surface point cloud into an ideal pipeline surface depth image, and changing the z-axis into the image depth of the ideal pipeline surface depth image;
(4) Generating a hemispherical corrosion pit three-dimensional point cloud according to the depth and the size of the corrosion pit in practice;
(5) The direction of the corrosion pits on the pipeline wall is random, so that the hemispherical corrosion pits are randomly rotated;
(6) Converting the rotated hemispherical etch pit into a hemispherical etch pit depth image, and changing the z-axis to the image depth of the hemispherical etch pit depth image;
(7) The size of the corrosion pit is random, so that the aspect ratio of the hemispherical corrosion pit is randomly changed, and the ratio of the width to the height is 5-0.5;
(8) Since the pits are not all round or oval in shape, the shapes of the hemispherical pits are changed by using the outline templates of the various targets in the coco data set to obtain various shapes of the hemispherical pits;
(9) Adding or subtracting a certain area of the hemispherical corrosion pit depth image and the ideal pipeline surface depth image; wherein, the addition is a convex corrosion pit, and the subtraction is a concave corrosion pit;
(10) Cycling the steps (4) to (9), repeatedly adding hemispherical corrosion pits with various shapes until the number of the hemispherical corrosion pits with various shapes reaches a set value, and storing the positions of all the hemispherical corrosion pits with various shapes;
(11) To prevent overlapping of corrosion pits with each other, it is necessary to delete overlapping data: requiring generated data, wherein corrosion pits on each object cannot overlap each other, so that the iou value of each corrosion pit and other corrosion pits on the pipeline is calculated, if the iou value is greater than 0, the corrosion pits are deleted, and the corrosion pits which overlap each other are deleted;
(12) After deleting the overlapped data, storing the depth image with the corrosion pits according to a 32-bit image (if the depth image is stored as an 8-bit image, the accuracy is insufficient), and storing non-overlapped corrosion pit position data on all pipelines into corresponding json files;
(13) Selecting a yolov4 network structure as a target detection model, and performing target detection model mixed training by using the corrosion pit data automatically generated in the steps (1) - (12) and a small amount of real corrosion pit data (manual marking data); wherein, the data volume ratio of the automatically generated corrosion pit data and a small amount of real corrosion pit data is 1000:1, in practice, the data size of the real corrosion pit data is 20, and when the weight of the target detection model is initialized, the weight of the target detection model pre-trained in the coco data set is adopted to perform migration learning, so that the model robustness can be greatly improved.
The specific mixed training step comprises the following steps:
(14) After initialization, the initial learning rate of 0.0001 is adopted, and the automatically generated corrosion pit data is used for carrying out coarse training on the target detection model, wherein noise with different scales is required to be added to the automatically generated corrosion pit data in the training process so as to simulate a state that a real pipeline is not absolutely smooth;
(15) After the target detection model is trained to 40 epochs, the learning rate is reduced to 0.00001, and at the moment, automatically generated corrosion pit data and real corrosion pit data are input in equal quantity for training;
(16) After the target detection model is trained to 5epoch, loss tends to stably stop the target detection model training.
And then automatically identifying corrosion pits by using a trained target detection model, and specifically comprising the following steps:
(17) Shooting the surface of the pipeline to be detected by using a depth camera to generate a three-dimensional point cloud of the surface of the object;
(18) Intercepting the 30cm multiplied by 30cm of the three-dimensional point cloud center area of the pipeline surface, and converting the three-dimensional point cloud center area into a depth image;
(19) Preprocessing the depth image obtained in step (18), dividing the depth image by 250;
(20) Inputting the target detection model obtained in the step (19) after training, analyzing the depth image, and returning all mark frames related to the corrosion pit;
(21) The depth of each corrosion pit marking frame is obtained, namely the depth of the corrosion pit, and the length and the width of the marking frame are the length and the width of the corrosion pit.
The foregoing description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and any simple modification, equivalent variation, etc. of the above embodiment according to the technical matter of the present invention are within the scope of the present invention.
Claims (5)
1. An automatic corrosion pit detection method based on target detection supervised learning is characterized by comprising the following steps of: comprising the following steps:
1) Automatically generating a large amount of corrosion pit data according to the characteristics of the corrosion pits; comprising the following steps:
1.1 Generating ideal object surface point cloud with the radius range of 50-1000 mm;
1.2 Rotating the ideal object surface point cloud around a y axis and a z axis, wherein the random angle range of rotation is set to be +/-15 degrees;
1.3 Converting the rotated ideal object surface point cloud into a depth image, and changing the z-axis into the image depth of the depth image;
1.4 Generating a hemispherical corrosion pit three-dimensional point cloud according to the depth and the size of the corrosion pit in practice;
1.5 Randomly rotating the hemispherical etch pits;
1.6 Converting the rotated hemispherical etch pit into a depth image, and changing the z-axis into an image depth of the depth image;
1.7 Randomly changing the aspect ratio of the hemispherical etching pits, wherein the ratio of the width to the height is 5-0.5;
1.8 Changing the shape of the hemispherical etch pit by using the outline templates of various targets in the coco data set to obtain hemispherical etch pits of various shapes;
1.9 Adding or subtracting the depth image of the hemispherical etch pit of various shapes to or from a region of the depth image of the ideal object surface; wherein, the addition is hemispherical corrosion pits with various convex shapes, and the subtraction is hemispherical corrosion pits with various concave shapes;
1.10 Cycling the steps 1.4) to 1.9), and repeatedly adding hemispherical corrosion pits with various shapes until the number of the hemispherical corrosion pits with various shapes reaches a set value, and storing the positions of all the hemispherical corrosion pits with various shapes;
1.11 Deleting overlapping data: calculating the iou value of each corrosion pit and other corrosion pits on the object, if the iou value is more than 0, deleting the corrosion pits, and deleting the mutually overlapped corrosion pits;
1.12 After the overlapping etch pits on the object are deleted in step 1.11), storing the depth image of the object without the overlapping etch pits according to the 32-bit image, and storing the position data of the non-overlapping etch pits on all the objects into the corresponding json file;
2) Performing mixed training on the target detection model by utilizing the automatically generated corrosion pit data and the real corrosion pit data;
3) And automatically identifying the corrosion pit by using the trained target detection model.
2. The method for automatically detecting corrosion pits based on target detection supervised learning according to claim 1, wherein: said step 2) comprises the steps of:
2.1 Initializing the weight of the target detection model;
2.2 Performing migration learning by adopting a target detection model weight pre-trained in a coco data set;
2.3 Using the initial learning rate of 0.0001 to perform coarse training of the target detection model by using the automatically generated corrosion pit data, wherein noise with different scales is required to be added to the automatically generated corrosion pit data in the training process;
2.4 After the target detection model is trained to 40 epochs, the learning rate is reduced to 0.00001, and at the moment, automatically generated corrosion pit data and real corrosion pit data are input in equal quantity for training;
2.5 After the target detection model is trained to 5epoch, the loss tends to stably stop the target detection model training.
3. The method for automatically detecting corrosion pits based on target detection supervised learning according to claim 1 or 2, wherein: the data volume ratio of the automatically generated corrosion pit data to the real corrosion pit data is as follows: 1000:1.
4. the method for automatically detecting corrosion pits based on target detection supervised learning according to claim 1 or 2, wherein: the target detection model adopts a yolov4 network structure.
5. The method for automatically detecting corrosion pits based on target detection supervised learning according to claim 1 or 2, wherein: said step 3) comprises the steps of:
3.1 Shooting the surface of the object to be detected by using a depth camera to generate an object surface three-dimensional point cloud;
3.2 Intercepting the size of 30cm multiplied by 30cm of the center area of the three-dimensional point cloud on the surface of the object, and converting the size into a depth image;
3.3 Preprocessing the depth image obtained in step 3.2), dividing the depth image by 250;
3.4 Inputting the target detection model obtained in the step 3.3) into the trained target detection model in the step 2), analyzing the depth image, and returning all mark frames related to the corrosion pit;
3.5 The depth of each etch pit mark frame is determined as the etch pit depth, and the length and width of the mark frame is the etch pit length and width.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248341.XA CN115953410B (en) | 2023-03-15 | 2023-03-15 | Corrosion pit automatic detection method based on target detection supervised learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310248341.XA CN115953410B (en) | 2023-03-15 | 2023-03-15 | Corrosion pit automatic detection method based on target detection supervised learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115953410A CN115953410A (en) | 2023-04-11 |
| CN115953410B true CN115953410B (en) | 2023-05-12 |
Family
ID=85886507
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310248341.XA Active CN115953410B (en) | 2023-03-15 | 2023-03-15 | Corrosion pit automatic detection method based on target detection supervised learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115953410B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117408999B (en) * | 2023-12-13 | 2024-02-20 | 安格利(成都)仪器设备有限公司 | Method for automatically detecting corrosion pits of containers and pipelines by utilizing point cloud complement |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113065446A (en) * | 2021-03-29 | 2021-07-02 | 青岛东坤蔚华数智能源科技有限公司 | Depth inspection method for automatically identifying ship corrosion area |
| CN113920020A (en) * | 2021-09-26 | 2022-01-11 | 中国舰船研究设计中心 | Human point cloud real-time repairing method based on depth generation model |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10861176B2 (en) * | 2018-11-27 | 2020-12-08 | GM Global Technology Operations LLC | Systems and methods for enhanced distance estimation by a mono-camera using radar and motion data |
| WO2021067665A2 (en) * | 2019-10-03 | 2021-04-08 | Photon-X, Inc. | Enhancing artificial intelligence routines using 3d data |
| WO2022088676A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Three-dimensional point cloud semantic segmentation method and apparatus, and device and medium |
| CN112560935B (en) * | 2020-12-11 | 2022-04-01 | 上海集成电路装备材料产业创新中心有限公司 | Method for improving defect detection performance |
| CN113129370B (en) * | 2021-03-04 | 2022-08-19 | 同济大学 | Semi-supervised object pose estimation method combining generated data and label-free data |
| CN113724240B (en) * | 2021-09-09 | 2023-10-17 | 中国海洋大学 | Refrigerator caster detection method, system and device based on visual identification |
| CN114463628A (en) * | 2021-12-31 | 2022-05-10 | 北方信息控制研究院集团有限公司 | Deep learning remote sensing image ship target identification method based on threshold value constraint |
| CN115164776B (en) * | 2022-07-04 | 2023-04-21 | 清华大学 | Three-dimensional measurement method and device for fusion of structured light decoding and deep learning |
| CN115618226A (en) * | 2022-10-12 | 2023-01-17 | 四川警察学院 | Encrypted traffic data synthesis method based on generation countermeasure network model |
| CN115731398A (en) * | 2022-11-18 | 2023-03-03 | 上海应用技术大学 | Point cloud and image fusion three-dimensional target detection method based on pseudo image |
-
2023
- 2023-03-15 CN CN202310248341.XA patent/CN115953410B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113065446A (en) * | 2021-03-29 | 2021-07-02 | 青岛东坤蔚华数智能源科技有限公司 | Depth inspection method for automatically identifying ship corrosion area |
| CN113920020A (en) * | 2021-09-26 | 2022-01-11 | 中国舰船研究设计中心 | Human point cloud real-time repairing method based on depth generation model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115953410A (en) | 2023-04-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115953410B (en) | Corrosion pit automatic detection method based on target detection supervised learning | |
| Bhandari et al. | A digital image-based deformation measurement system for triaxial tests | |
| Taylor et al. | Characterization of the effect of surface roughness and texture on fluid flow—past, present, and future | |
| CN108564615B (en) | Method, device and system for simulating laser radar detection and storage medium | |
| CN102183458B (en) | Method and system for measuring residual stresses | |
| CN103196370A (en) | Measuring method and measuring device of conduit connector space pose parameters | |
| CN107945178B (en) | A kind of residual hole automatic identification of explosion and feature extracting method | |
| CN109631847A (en) | Threaded target pose calculation method based on point cloud data | |
| EP3905124A3 (en) | Three-dimensional object detecting method, apparatus, device, and storage medium | |
| Agboola et al. | Comparative analysis of manual strapping method (MSM) and electro-optical distance ranging (EODR) method of tank calibration | |
| CN110705058A (en) | Near-field electromagnetic scattering simulation method for ultra-electric large-scale target | |
| JPS607318A (en) | Photogrammetry method of substance | |
| CN114359573A (en) | Pointer meter reading identification method and device and storage medium | |
| Gao et al. | Monocular vision measurement technology applied in dynamic compaction ramming settlement monitoring | |
| CN115937217B (en) | Method for automatically detecting corrosion pits by using 3D convolution and correlation operator | |
| CN112597574A (en) | Construction method and device of building information model | |
| CN108153939A (en) | Pipe lining rubber aging tendency analysis method, device, storage medium and computer equipment | |
| CN101667136A (en) | Star map simulation method based on forward ray tracking technology | |
| CN109559294B (en) | Method and device for detecting quality of circular hole of drop | |
| CN111951242A (en) | Gas leakage detection method based on multi-beam depth sounding sonar | |
| Fu et al. | Research on Image Liquid Level Measurement Technology Based on Hough Transform | |
| CN104534922A (en) | Artillery pitching radius measuring method, device and system based on double theodolites | |
| CN113702072B (en) | Gate valve wear amount measuring method, gate valve wear amount measuring equipment and readable storage medium | |
| Orlov | Hydraulic characteristics of internal protective coatings used for rehabilitation of pipelines of water supply and sanitation systems | |
| Sergeev et al. | Modeling of surface dust concentration in snow cover at industrial area using neural networks and kriging |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |