A device and method for detecting surface defects of a submarine pipeline

Abstract

The application discloses a kind of submarine pipeline surface defect detection device and detection method, it is collected through introducing trained neural network to the submarine pipeline surface image and carries out defect detection processing, improves processing efficiency, simultaneously, in combination with the positioning information generated simultaneously when image acquisition, so that the result of detection neural network output is when pipeline has defect, the submarine pipeline surface image with defect and its associated positioning information are rapidly output with defect, the mechanism makes that maintenance personnel can quickly locate the position of pipeline defect, to improve the formation efficiency of repair measure, scheme, in addition, the present scheme is also through actual application scene to construct and the adaptability of submarine water body state situation denoising training set, then through training denoising neural network to the submarine pipeline surface image collected is carried out denoising processing, it can higher quality, higher efficiency and higher accuracy to realize the denoising of submarine image, provides data guarantee for the identification detection of detection neural network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pipeline maintenance, in particular to a submarine pipeline surface defect detection device and method. BACKGROUND

[0002] As the lifeline of offshore oil and gas fields, submarine pipelines play an important role in offshore oil development. However, under harsh working conditions, submarine pipelines often fail due to corrosion, wave scouring or other mechanical damage, etc. Therefore, the safety of submarine pipelines has always been a problem in the field. With the wide application of "robot" technology, the application of specific robots for submarine pipeline inspection has become more and more popular. There are even some cleaning robots specifically designed for cleaning the surface of submarine pipelines. However, due to the complexity of the water quality and state of the sea, the application of visual detection technology is subject to certain interference and constraints. Therefore, there is currently no widely used visual technology for submarine pipeline surface inspection and defect detection. Therefore, how to optimize the image acquisition of submarine pipeline surface and improve the accuracy and reliability of visual technology in submarine pipeline surface defect detection is a very practical research topic. SUMMARY

[0003] Therefore, the purpose of the present application is to provide a submarine pipeline surface defect detection device and method that is reliable, flexible and can enhance and modify the features of the collected submarine pipeline surface image.

[0004] In order to achieve the above technical purpose, the technical solution adopted by the present application is as follows:

[0005] A submarine pipeline surface defect detection method comprises:

[0006] S01, a database is constructed, which stores a plurality of pipeline defect images;

[0007] S02, the pipeline defect images in the database are obtained and formatted according to the preset conditions, then defect labeling is performed, and then the labeled pipeline defect images and unlabeled pipeline defect images are associated to form a first training data set;

[0008] S03, the first training data set is obtained, a preset amount of data in the first training data set is selected as training data and validation data according to the preset conditions, then the training data and validation data are imported into the neural network for training until the model converges, and a detection neural network for detecting submarine pipeline surface defects is obtained;

[0009] S04, submarine pipeline surface images are collected according to the preset conditions and positioning information associated with the submarine pipeline surface images is generated synchronously;

[0010] S05, after the acquired seabed pipeline surface image is denoised, the seabed pipeline surface image is imported into the detection neural network for detection, and a detection result is output by the detection neural network;

[0011] S06, the detection result is acquired, and when the detection result is that defects exist, the seabed pipeline surface image and the positioning information associated with the seabed pipeline surface image are output.

[0012] As a preferred selection embodiment, preferably, in the scheme S05, a denoising neural network is used to denoise the acquired seabed pipeline surface image, and a construction method of the denoising neural network comprises:

[0013] A01, a simulation scene corresponding to a seabed pipeline laying sea area is constructed, and a scene background is transparent or pure color;

[0014] A02, the seawater state under different preset flow rates, fluid states and / or light spot conditions is manufactured in the simulation scene, then image acquisition is performed and the image is segmented into multiple image frames, after the background color of the image frame is removed, the image frame transparency is set to 30%-60%, and multiple image noise data under different image transparencies are output, after the image noise data are collected, a fluid noise data set is obtained;

[0015] A03, pipeline defect images in the database are acquired, and the pipeline defect images are format-adjusted according to preset conditions, then the image noise data in the fluid noise data set and the pipeline defect images are randomly combined and superimposed to obtain noise-processed pipeline defect images, and after the noise-processed pipeline defect images and the pipeline defect images without noise processing are associated, a second training data set is formed by collection;

[0016] A04, the second training data set is acquired, a preset amount of data in the second training data set is selected as training data and verification data according to preset conditions, then the training data and the verification data are imported into the neural network for training until the model converges, and a denoising neural network for denoising the seabed pipeline surface image is obtained.

[0017] As a preferred selection embodiment, preferably, in the scheme S04, the seabed pipeline surface image is acquired and the positioning information associated with the seabed pipeline surface image is generated synchronously according to preset conditions, which comprises:

[0018] S041, the seabed pipeline surface image is collected in a spiral or reciprocating swing manner within a preset radian with the length direction of the seabed pipeline as the collection advancing direction and the radial center of the seabed pipeline as the rotation center, and the collection time is recorded at the same time;

[0019] S042. Using the starting point of the surface image acquisition of the subsea pipeline as one of the positioning points, establish the first coordinate (x1, y1, z1), and use the acquisition endpoint as the Nth coordinate (x1, y1, z1). n ,y n ,z n The system records the acquisition trajectory and coordinate changes in real time. Then, using the acquisition time and coordinate recording time as common factors, it associates the surface image of the subsea pipeline with the coordinate information to generate positioning information.

[0020] As a preferred implementation method, the image range of the submarine pipeline surface image acquired from the first coordinate to the Nth coordinate covers the non-field blind area and non-image acquisition blind area of ​​the submarine pipeline.

[0021] As a preferred implementation method, this solution further includes:

[0022] S07. Based on the collected surface images and positioning information of the subsea pipeline, perform image modeling of the subsea pipeline;

[0023] S08. Obtain the detection results. When the detection results indicate the presence of defects, locate and visualize the defect in the image modeling based on the surface image of the subsea pipeline and its associated positioning information.

[0024] As a preferred implementation method, preferably, in this solution S07, the data for image modeling of the subsea pipeline based on the collected surface images and positioning information of the subsea pipeline is also stored in a database, and the corresponding modeling time and the subsea pipeline surface image acquisition time are recorded.

[0025] As a preferred implementation method, this solution further includes:

[0026] S09. Obtain the surface image of the subsea pipeline with defects and its associated location information, and generate corresponding maintenance information. The maintenance personnel shall obtain the maintenance information and carry out maintenance on the subsea pipeline area with defects. After the maintenance is completed, upload the surface image of the subsea pipeline in the repair area.

[0027] S10. Obtain the surface image of the subsea pipeline in the repair area, update and replace it in the image model of the subsea pipeline corresponding to the maintenance information, and save and associate the replaced subsea pipeline surface image.

[0028] As a preferred implementation method, this solution obtains the detection results. When the detection result indicates the presence of a defect, it reads the database to determine whether there are historical records of submarine pipeline surface image acquisition and image modeling records. If so, it retrieves the most recently acquired submarine pipeline surface image of the submarine pipeline for which the current detection result indicates a defect and outputs it in conjunction with the data.

[0029] Based on the above, the present application also provides a submarine pipeline surface defect detection device, which comprises:

[0030] A data storage unit is configured to build a database in which a plurality of pipeline defect images are stored.

[0031] An image processing unit is configured to acquire the pipeline defect images in the database, adjust the format of the pipeline defect images according to preset conditions, label the defects, and then correlate the labeled pipeline defect images and the unlabeled pipeline defect images to form a first training data set.

[0032] A detection neural network unit is configured to detect the submarine pipeline surface defects and output the detection results.

[0033] A data scheduling unit is configured to acquire the first training data set, select a preset amount of data in the first training data set as training data and validation data according to preset conditions, and then import the training data and the validation data into the neural network to train the model to convergence, thereby obtaining a detection neural network for detecting the submarine pipeline surface defects.

[0034] An image acquisition unit is configured to acquire submarine pipeline surface images according to preset conditions.

[0035] A positioning unit is configured to synchronously generate positioning information associated with the submarine pipeline surface images.

[0036] An image preprocessing unit is configured to acquire the acquired submarine pipeline surface images, perform denoising processing on the submarine pipeline surface images, and then import the submarine pipeline surface images into the detection neural network for detection, thereby outputting the detection results by the detection neural network.

[0037] An information scheduling unit is configured to acquire the detection results, and when the detection results indicate that there are defects, output the submarine pipeline surface images and the positioning information associated therewith.

[0038] Based on the above, the present application also provides a computer-readable storage medium, which stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, at least one program, code set or instruction set is loaded and executed by a processor to realize the submarine pipeline surface defect detection method described above.

[0039] Compared with the prior art, the method has the beneficial effects that: the method ingeniously introduces a trained neural network to perform defect detection and processing on the collected seabed pipeline surface image, greatly improves the processing efficiency, and combines the positioning information generated synchronously during image collection, so that when the pipeline has a defect, the result output by the detection neural network is the seabed pipeline surface image with the defect and the positioning information associated with the seabed pipeline surface image, which enables the maintenance personnel to quickly locate the position of the pipeline defect, improves the formation efficiency of the maintenance measures and schemes, and in addition, the method further constructs a denoising training set adaptive to the state of the seabed water body according to the actual application scenario, and then trains a denoising neural network to perform denoising processing on the collected seabed pipeline surface image, which can realize the denoising of the seabed image with high quality, high efficiency and high accuracy, provides data guarantee for the recognition and detection of the detection neural network, and makes the method more reliable and more flexible in application. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0041] Figure 1 is one of the schematic diagrams of the brief implementation process of the present application;

[0042] Figure 2 is the schematic diagram of the brief implementation process of constructing a denoising neural network of the present application;

[0043] Figure 3 is the schematic diagram of the detailed process of step S4 of the present application;

[0044] Figure 4 is one of the schematic diagrams of the local process of the present application;

[0045] Figure 5 is the second schematic diagram of the local process of the present application;

[0046] Figure 6 is the schematic diagram of the unit module connection of the present application. DETAILED DESCRIPTION

[0047] The application will be described in further detail below with reference to the drawings and embodiments. It is particularly pointed out that the following embodiments are only for illustration of the application, but do not limit the scope of the application. Similarly, the following embodiments are only part of the embodiments of the application, not all embodiments, and all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of the application.

[0048] As shown in Figure 1 , the embodiment of the application provides a method for detecting defects on the surface of a submarine pipeline, which comprises:

[0049] S01, constructing a database in which a plurality of pipeline defect images are stored;

[0050] S02, obtaining the pipeline defect images in the database and adjusting their formats according to preset conditions, then performing defect labeling, and then associating the labeled pipeline defect images with the unlabeled pipeline defect images to form a first training data set;

[0051] S03, obtaining the first training data set, selecting a preset amount of data in the first training data set as training data and validation data according to preset conditions, then importing the training data and validation data into a neural network for training until the model converges, and obtaining a detection neural network for detecting defects on the surface of a submarine pipeline;

[0052] S04, collecting submarine pipeline surface images according to preset conditions and synchronously generating positioning information associated with the submarine pipeline surface images;

[0053] S05, obtaining the collected submarine pipeline surface images and performing denoising processing thereon, then importing them into the detection neural network for detection, and outputting the detection results from the detection neural network;

[0054] S06, obtaining the detection results, and when the detection results indicate the presence of defects, outputting the submarine pipeline surface images and the positioning information associated therewith.

[0055] As shown in Figure 2 , the seawater in the submarine may exhibit different fluid states due to various external factors, and it is inevitable that when taking images of the submarine pipeline, the image of the seawater will also be taken. At this time, the image obtained will show that the pipeline image is superimposed with a certain transparency of water body, and manual judgment will result in the need to occupy a large amount of manpower. In order to improve the processing efficiency and save manpower, the embodiment introduces a neural network for auxiliary processing. Specifically, as a preferred selection implementation, preferably, a denoising neural network is used in S05 of the present application to denoise the collected submarine pipeline surface images. The construction method of the denoising neural network comprises:

[0056] A01. Construct a simulation scene corresponding to the sea area where the submarine pipeline will be laid, and make the background of the scene transparent or solid color.

[0057] A02. Create seawater conditions with different preset flow rates, fluid states and / or light spots in the simulation scenario, then acquire images of them and divide them into multiple image frames. After removing the background color from the image frames, set the transparency of the image frames between 30% and 60% and output multiple image noise data under different image transparency. After collecting the image noise data, obtain the fluid noise dataset.

[0058] A03. Obtain pipeline defect images from the database and adjust their format according to preset conditions. Then, randomly combine and superimpose the image noise data in the fluid noise dataset with the pipeline defect images to obtain noise-processed pipeline defect images. After associating the noise-processed pipeline defect images with the unprocessed pipeline defect images, they are aggregated to form the second training dataset.

[0059] A04. Obtain the second training dataset. Select a preset amount of data from the second training dataset as training data and validation data according to preset conditions. Then import the training data and validation data into the neural network and train until the model converges to obtain a denoising neural network for denoising the surface image of the subsea pipeline.

[0060] In A02, the transparency of the image frame can be set to 30%, 35%, 40%, 45%, 50%, or 60%.

[0061] Combination Figure 3 As shown, in order to facilitate the association between the acquired surface images of the subsea pipeline and their acquired location coordinates, the following steps are taken in scheme S04: The subsea pipeline surface images are acquired according to preset conditions, and the positioning information associated with the subsea pipeline surface images is generated simultaneously:

[0062] S041. Using the length of the subsea pipeline as the forward direction of data acquisition, and the radial center of the subsea pipeline as the rotation center, the surface image of the subsea pipeline is acquired by reciprocating in a spiral or within a preset arc, while recording the acquisition time.

[0063] S042. Using the starting point of the surface image acquisition of the subsea pipeline as one of the positioning points, establish the first coordinate (x1, y1, z1), and use the acquisition endpoint as the Nth coordinate (x1, y1, z1). n ,y n ,z n The system records the acquisition trajectory and coordinate changes in real time. Then, using the acquisition time and coordinate recording time as common factors, it associates the surface image of the subsea pipeline with the coordinate information to generate positioning information.

[0064] In order to obtain the image of the submarine pipeline as much as possible, avoid unnecessary local loss and cause inspection omission, as a preferred selection implementation, preferably, the image range of the submarine pipeline surface image collected by the first coordinate to the Nth coordinate covers the non-visual field blind area and the non-image collection blind area of the submarine pipeline.

[0065] In combination Figure 4 As shown in the scheme, for remote monitoring, if only the collected submarine pipeline surface image is quickly reviewed to understand the overall situation, it is likely to cause visual fatigue and miss more content details, as a preferred selection implementation, preferably, the scheme further comprises:

[0066] S07, according to the collected submarine pipeline surface image and positioning information, image modeling of the submarine pipeline is carried out, wherein the submarine pipeline surface image used is the denoised image;

[0067] S08, obtaining the detection result, when the detection result is that there is a defect, positioning and visual output in image modeling are carried out according to the submarine pipeline surface image and the positioning information associated therewith.

[0068] In order to further intuitively compare the submarine pipeline surface images at different detection times, as a preferred selection implementation, preferably, the data of the image modeling of the submarine pipeline according to the collected submarine pipeline surface image and positioning information in the scheme S07 is also stored in the database and the corresponding modeling time and submarine pipeline surface image collection time are recorded.

[0069] In combination Figure 5 As shown in the scheme, in order to further improve the humanization and intuitiveness of the scheme and follow up the repair situation on the mechanism shown, Figure 4 As shown in the scheme, in order to further improve the humanization and intuitiveness of the scheme and follow up the repair situation on the mechanism shown,

[0070] S09, obtaining the submarine pipeline surface image with defects and the positioning information associated therewith, and generating repair information accordingly, obtaining the repair information by the repair and maintenance personnel, repairing the submarine pipeline area with defects, and uploading the submarine pipeline surface image of the repaired area after the repair is completed;

[0071] S10, obtaining the submarine pipeline surface image of the repaired area, and updating and replacing it into the image modeling of the submarine pipeline corresponding to the repair information, and saving and associating the replaced submarine pipeline surface image.

[0072] As a preferred selection implementation, preferably, the scheme obtains a detection result, when the detection result is that there is a defect, reads the database to determine whether there is a history record of seabed pipeline surface image collection and image modeling record of the seabed pipeline, when there is, calls the seabed pipeline surface image collected at the latest time when the detection result is that there is a defect of the seabed pipeline and outputs in association.

[0073] By outputting the seabed pipeline surface image with defects and the last detected image in association, the scheme can effectively assist the background maintenance personnel in analysis and comparison, and at the same time, the mechanism also helps to call historical data in time, saves labor, and greatly improves the material collection efficiency of the background maintenance personnel.

[0074] Since the scheme can be adapted to seabed pipeline detection in different regions, in order to enable the seabed pipeline surface image in use to benefit the neural network model, so that the neural network model can further improve the processing accuracy and detection reliability following the use of the scheme, the scheme can further include: further marking the seabed pipeline image with defects in the detection result by artificial means, when the marking meets the preset requirements, associating and storing the detection result and the seabed pipeline image in the first training set, at the same time, the training data in the first training set are all recorded with numbers, then according to a preset time period, the data in the first training set numbered in the preset region are imported into the detection neural network for training and verification, when the training is preset times, the verification accuracy reaches the preset value, the detection neural network is updated, when the training is preset times, the verification accuracy does not reach the preset value, then the training data are continuously imported into the detection neural network for retraining until the model converges.

[0075] As shown in Figure 6 Based on the above, the embodiment of the scheme also provides a seabed pipeline surface defect detection device, which comprises:

[0076] A data storage unit is configured to construct a database, wherein the database stores a plurality of pipeline defect images and is also configured to store a first training data set and seabed pipeline surface images obtained and associated positioning information;

[0077] An image processing unit is configured to obtain pipeline defect images in the database, adjust the format of the pipeline defect images according to preset conditions, perform defect labeling, associate the labeled pipeline defect images and unlabeled pipeline defect images, and then collect the first training data set;

[0078] A detection neural network unit is configured to detect seabed pipeline surface defects and output a detection result;

[0079] The data scheduling unit is configured to obtain a first training data set, select a preset amount of data in the first training data set as training data and verification data according to a preset condition, and then import the training data and the verification data into the neural network for training until the model converges, so as to obtain a detection neural network for detecting surface defects of a submarine pipeline.

[0080] The image acquisition unit is configured to acquire submarine pipeline surface images according to a preset condition.

[0081] The positioning unit is configured to synchronously generate positioning information associated with the submarine pipeline surface images.

[0082] The image preprocessing unit is configured to acquire the acquired submarine pipeline surface images, perform denoising processing on the images, and then import the images into the detection neural network for detection, so as to output a detection result by the detection neural network.

[0083] The information scheduling unit is configured to acquire the detection result, and output the submarine pipeline surface images and the positioning information associated therewith when the detection result indicates that there is a defect.

[0084] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0085] If the integrated unit is realized in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0086] The above only describes some embodiments of the present application, and does not limit the protection scope of the present application, and any equivalent device or equivalent process transformation made by using the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method of detecting surface defects of a subsea pipeline, characterized in that, The application comprises the following steps: constructing a database in which a plurality of pipeline defect images are stored; acquiring pipeline defect images in the database and adjusting the format of the images according to preset conditions, then performing defect labeling, associating the labeled pipeline defect images with the unlabeled pipeline defect images, and then collecting to form a first training data set; acquiring the first training data set, selecting a preset amount of data in the first training data set as training data and validation data according to preset conditions, then importing the training data and validation data into a neural network for training until the model converges, and obtaining a detection neural network for detecting defects on the surface of a submarine pipeline; acquiring submarine pipeline surface images according to preset conditions, collecting the pipeline surface images in a spiral or reciprocating manner within a preset arc with the length direction of the submarine pipeline as the forward direction of collection and the radial center of the submarine pipeline as the rotation center, synchronously recording the collection time and generating coordinate information associated with the collected images, so that the image range from the starting point to the ending point of the image collection covers the non-visual field blind area and the non-image collection blind area of the submarine pipeline, and the collection time and the coordinates are used as common associated factors to generate positioning information; acquiring the collected submarine pipeline surface images and performing denoising processing, then importing the images into the detection neural network for detection, and outputting the detection results from the detection neural network; acquiring the detection results, and when the detection results indicate the presence of defects, outputting the submarine pipeline surface images and the positioning information associated therewith; wherein the collected submarine pipeline surface images are denoised using a denoising neural network, and the construction method of the denoising neural network comprises: constructing a simulation scene corresponding to the submarine pipeline deployment sea area and setting the scene background to be transparent or pure color; manufacturing seawater states under different preset flow rates, fluid states and / or light spot conditions in the simulation scene, then collecting images and dividing them into multiple image frames, removing the background color of the image frames, setting the image frame transparency to between 30% and 60%, and outputting multiple image noise data under different image transparencies, and after collecting the image noise data, obtaining a fluid noise data set; acquiring pipeline defect images in the database and adjusting the format of the images according to preset conditions, then randomly combining and superimposing the image noise data in the fluid noise data set with the pipeline defect images to obtain noise-processed pipeline defect images, then associating the noise-processed pipeline defect images with the pipeline defect images that have not been noise-processed, and then collecting to form a second training data set; acquiring the second training data set, selecting a preset amount of data in the second training data set as training data and validation data according to preset conditions, then importing the training data and validation data into a neural network for training until the model converges, and obtaining a denoising neural network for denoising submarine pipeline surface images.

2. The method of subsea pipeline surface defect detection of claim 1, wherein, acquiring submarine pipeline surface images according to preset conditions and synchronously generating positioning information associated with the submarine pipeline surface images comprises: The starting point of the image collection of the submarine pipeline surface is taken as one of the positioning points, a first coordinate (x1, y1, z1) is established, the ending point is taken as an Nth coordinate (x n ,y n ,z n ), the collection track and the coordinate change are recorded in real time, then the image of the submarine pipeline surface is associated with the coordinate information by taking the collection time and the coordinate recording time as common factors, and the positioning information is generated.

3. The method of subsea pipeline surface defect detection of claim 2, wherein, the image range of the submarine pipeline surface images collected from the first coordinate to the Nth coordinate covers the non-visual field blind area and the non-image collection blind area of the submarine pipeline.

4. The method of subsea pipeline surface defect detection of claim 3, wherein, The application further comprises: According to the collected seabed pipeline surface image and positioning information, the seabed pipeline is image modeled; Obtain the detection result, when the detection result is that there is a defect, position and visualize output in the image modeling according to the seabed pipeline surface image and the positioning information associated therewith.

5. The method of subsea pipeline surface defect detection of claim 4, wherein, The data of the seabed pipeline image modeling according to the collected seabed pipeline surface image and positioning information is also stored in the database and records the corresponding modeling time and seabed pipeline surface image collection time.

6. The method of subsea pipeline surface defect detection of claim 5, wherein, It also includes: Obtain the seabed pipeline surface image with the detection result of existing defects and the positioning information associated therewith, and generate the maintenance information accordingly, obtain the maintenance information by the maintenance personnel and maintain the seabed pipeline area with defects, and upload the seabed pipeline surface image of the repaired area after the maintenance is completed; Obtain the seabed pipeline surface image of the repaired area, and update and replace it into the image modeling of the seabed pipeline corresponding to the maintenance information, and save and associate the replaced seabed pipeline surface image.

7. The method of subsea pipeline surface defect detection of claim 6, wherein, Obtain the detection result, when the detection result is that there is a defect, read the database to determine whether there is a historical record of seabed pipeline surface image collection and image modeling record of the seabed pipeline, when there is, call the seabed pipeline with the detection result of existing defects in the latest collected seabed pipeline surface image and associate and output.

8. An apparatus for detecting a surface defect of a submarine pipeline, which applies the method for detecting a surface defect of a submarine pipeline according to any one of claims 1 to 7, characterized by, It includes: A data storage unit is configured to build a database, and the database stores a plurality of pipeline defect images; An image processing unit is configured to obtain the pipeline defect images in the database, adjust the formats of the pipeline defect images according to preset conditions, label the defects, associate the labeled pipeline defect images with the unlabeled pipeline defect images, and then collect the first training data set; A detection neural network unit is configured to detect the defects on the seabed pipeline surface and output a detection result; A data scheduling unit is configured to obtain the first training data set, select a preset amount of data in the first training data set as training data and verification data according to preset conditions, and then import the training data and the verification data into the neural network to train the model to converge, so as to obtain the detection neural network for detecting the defects on the seabed pipeline surface; An image acquisition unit is configured to acquire seabed pipeline surface images according to preset conditions; A positioning unit is configured to generate positioning information associated with the seabed pipeline surface images synchronously; An image preprocessing unit is configured to acquire the collected seabed pipeline surface images, perform denoising processing on the seabed pipeline surface images, and then import the seabed pipeline surface images into the detection neural network to perform detection, so as to output a detection result by the detection neural network; An information scheduling unit is configured to obtain the detection result, and when the detection result is that there is a defect, output the seabed pipeline surface image and the positioning information associated therewith.

9. A computer readable storage medium characterized by: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize the seabed pipeline surface defect detection method according to any one of claims 1 to 7.

Images