Ship prefabricated pipe intelligent identification equipment and method based on visual inspection

Abstract

A ship prefabricated pipe intelligent identification device based on visual inspection comprises edge device software and edge computing device hardware. The edge equipment software comprises an image acquisition preprocessing module, a pipe fitting feature recognition module, a character feature recognition module, a feature combination judgment module, a character combination judgment module, an engineering information database, a human-computer interaction interface and a process library; the edge computing equipment hardware comprises a camera, a touch screen and a communication module, the camera is used for shooting and obtaining pipe fitting image information, the touch screen is used for manual interface operation, and the communication module is responsible for information transmission of all modules of the software. According to the method, the image processing technology and the deep learning algorithm are utilized, automatic identification of the ship prefabricated pipe fitting is achieved through intelligent identification, and the assembling efficiency of ship pipe units, ship unit modules, ship segments and other scenes can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent ship manufacturing, in particular to an intelligent identification device for ship prefabricated pipe fittings based on visual detection, and also to a method for intelligent identification using the intelligent identification device for ship prefabricated pipe fittings based on visual detection. Background Art

[0002] The installation of pipe fittings during shipbuilding is an important part of shipbuilding operations. At present, prefabricated pipe fittings need to be manufactured in pipe processing and manufacturing workshops and placed on pallets by unit or area classification. Prefabricated pipe fittings are non-standard pipe fittings with various types and shapes, and there are many similar parts. However, in the process of pipe assembly of traditional ship units, areas or sections, workers mainly rely on the graphics and numbers of paper drawings and other information to compare with the pipe numbers in the pallet. The search for prefabricated pipe fittings often requires the process of prefabricated pipe fittings inventory, number comparison, paper process information viewing, etc. The overall efficiency is low and prone to errors, which greatly affects the assembly quality and efficiency of ship pipe fittings. Summary of the invention

[0003] The technical problem to be solved by the present invention is to address the deficiencies in the prior art and to provide an intelligent recognition device for prefabricated ship pipe fittings based on visual inspection, which can effectively improve the assembly efficiency of ship pipe units, ship unit modules, ship sections and other scenarios through automatic recognition of prefabricated ship pipe fittings.

[0004] Another technical problem to be solved by the present invention is to provide a method for intelligent identification using the above-mentioned ship prefabricated pipe intelligent identification device based on visual detection.

[0005] The technical problem to be solved by the present invention is achieved through the following technical solutions. The present invention is a ship prefabricated pipe intelligent identification device based on visual inspection, including edge device software and edge computing device hardware, the edge device software includes an image acquisition preprocessing module, a pipe feature recognition module, a text feature recognition module, a feature combination judgment module, a text combination judgment module, an engineering information database, a human-computer interaction interface and a process library; The image acquisition preprocessing module is used to acquire and process the pipe image and transmit it to the pipe feature recognition module and the text feature recognition module; The pipe feature recognition module is used to recognize feature point information on the pipe image and send it to the feature combination judgment module; The text feature recognition module is used to recognize the pipe image coding information and send it to the text combination judgment module; The feature combination judgment module is used to perform statistical analysis and combination classification on the identified feature points, and compare the combination features with the engineering information database to obtain the pipe fitting number with the highest probability and send it to the human-machine interface interaction module; The character combination judgment module is used to search and compare the recognized character combination with the engineering information database; The engineering information database is used to store ship production design data and process data in advance through data extraction; The human-machine interface interaction includes a prefabricated pipe number confirmation interface module and a process guidance display interface module. The prefabricated pipe number confirmation interface module is used to display the pipe model obtained by comparison and judgment according to probability, and the process guidance display interface is used to view the pipe assembly process information; The edge computing device hardware includes a camera, a touch screen and a communication module. The camera is used to capture and obtain pipe image information, the touch screen is used for manual interface operation, and the communication module is responsible for information transmission among software modules.

[0006] The technical problem to be solved by the present invention can also be further achieved through the following technical solutions. For the above-mentioned ship prefabricated pipe intelligent identification equipment based on visual inspection, the prefabricated pipe detection model adopted by the equipment is based on the YOLO algorithm deep learning training pipe image features and text AI recognition model.

[0007] The technical problem to be solved by the present invention can also be further achieved through the following technical solutions. For the above-mentioned intelligent identification equipment for prefabricated ship pipe fittings based on visual inspection, the image acquisition and preprocessing module collects images of ship pipe fittings by arranging cameras or handheld devices on site, and performs multi-angle and all-round image capture and acquisition of the pipe fittings in real time, and focuses on capturing local parts of the pipe fittings to obtain high-quality pipe fitting images, and views the pipe fitting numbers as text identification symbols for auxiliary identification, and at the same time performs grayscale and contrast enhancement preprocessing operations on the collected pipe fitting images to improve the clarity and contrast of the image.

[0008] The technical problem to be solved by the present invention can also be further achieved by the following technical solutions. For the above-mentioned intelligent recognition device for prefabricated ship pipe fittings based on visual inspection, the pipe fitting feature recognition module is trained with key features by deep learning training. The training types include flange joints, branch pipe joints, elbows, tees, and crosses. Each category is divided into different subcategories by color. The pipe feature recognition module is a software module developed by the secondary development of the intelligent recognition model trained by the deep learning training environment. The recognition model in the pipe feature recognition module is usually trained by the following methods: Step 1: Dataset Preparation In the ship production environment, a large number of characteristic pictures of flange joints, branch joints, elbows, tees, and crosses are taken. Each feature needs to collect different surface colors and environmental backgrounds; Step 2: Dataset annotation: Use annotation software to mark the features of the pipes in the image and add feature attributes; Step 3: Train the AI ​​model using the YOLO algorithm First, the script converts the format and sends the produced image data set to the AI ​​model. The AI ​​model locates the bounding box through the features of the pipe image set and accurately detects the location of the image pipe features. According to the feature information reflected in the image data set, the algorithm training parameters are adjusted, and the algorithm for iteratively identifying pipe features is trained and optimized to obtain an AI recognition model with good accuracy in terms of accuracy, recall rate, average precision AP, and average precision mean AP. pt weight file. Under the premise of ensuring accuracy, the parameters can be adjusted to obtain a lightweight and fast recognition speed weight file. When identifying pipe fittings, the program calls and outputs the trained AI model .pt weight file for prefabricated ship fittings, and identifies the features of the pipe fittings through the AI ​​model. The output information includes a combination of prefabricated pipe fitting feature type, correct probability and identification location.

[0009] The technical problem to be solved by the present invention can be further achieved by the following technical solutions. For the above-mentioned intelligent recognition device for prefabricated ship pipes based on visual detection, since the pipe feature recognition module recognizes only the local features of the pipe, rather than the whole prefabricated pipe, it is necessary to further judge according to the recognition result of the pipe feature recognition module; The judgment process is as follows: The first step is to count the feature point combination data of the detected object according to the results of the pipe identification module, including the number and type of feature points, location and accuracy information; In the second step, the number and type of feature points of the detected object are globally searched with the project feature database to obtain the possible prefabricated pipe fitting numbers; The comparison method includes two methods: accurate judgment method and fuzzy judgment method; Use quasi-accurate judgment method to compare with the database to obtain fully matched prefabricated pipe fittings; Through the fuzzy judgment method, it is judged that the feature combination belongs to the subset of prefabricated pipe fittings in the database, and the prefabricated pipe fittings that exit the matching are obtained; Matching rate calculation method: Matching rate = sum of the probability of matching feature points / total number of features; The possible prefabricated pipe fitting numbers are obtained by calculating the percentage of the number of matching feature points and the total number of feature points under the number, and are sorted by the feature matching rate.

[0010] The technical problem to be solved by the present invention can be further achieved through the following technical solutions. For the above-mentioned intelligent recognition equipment for prefabricated ship pipe fittings based on visual detection, since the text feature recognition module of the prefabricated pipe fittings usually recognizes some numbers, letters or symbols, the coding of the prefabricated pipe fittings needs to be compared with the database for inference. The inference method judges through global search similarity. The recognized text is searched in sequence with the number information of the engineering information database and sorted by similarity.

[0011] The technical problem to be solved by the present invention can also be further achieved by the following technical solutions. For the above-mentioned intelligent recognition equipment for prefabricated ship pipe fittings based on visual detection, the prefabricated pipe fittings text feature recognition module is trained on handwritten, steel stamped, and printed letters and numbers in the deep learning environment AI model to obtain a text feature recognition AI model. The prefabricated pipe fittings text feature recognition module realizes the recognition of pipe fitting number text features by calling the text feature recognition AI model: Step 1: Text image acquisition and processing: a large number of photos are taken in the ship production environment to obtain the entire string of text area numbers on the pipes, and the acquired images are enhanced and processed; Step 2: Coded text marking, the rectangular boundary box marks the location of the letter or number features of the pipe fitting text number; Step 3: Encoded text feature extraction: Use the AI ​​model to accurately detect the bounding box of the text feature location in the image set, extract and separate the detected text numbers, adjust the algorithm training parameters based on the feature information reflected in the image data set, and optimize and iteratively train the AI ​​recognition model; Step 4: Text feature recognition. Pipe fitting text feature recognition calculates the similarity between image text features by calling the AI ​​recognition model to find the numbering information with the highest matching degree.

[0012] The technical problem to be solved by the present invention can be further achieved through the following technical solutions. For the above-mentioned ship prefabricated pipe intelligent identification equipment based on visual inspection, the engineering information database is also responsible for storing the prefabricated pipe numbers and related process information. The data needs to be preset before inspection, and the data information is extracted and stored by the pipe number, feature point information and process information.

[0013] The technical problem to be solved by the present invention can also be further achieved by the following technical solutions. For the above-mentioned intelligent identification equipment for prefabricated ship pipes based on visual inspection, the human-computer interaction module presents the pipe number information sorting by probability through the page, and the worker confirms the pipe number and clicks it according to the on-site confirmation situation. The process guidance module obtains relevant process information through the number, and the process information is retrieved from the prefabricated process information library or from the cloud site for on-site viewing: Step 1: The human-computer interaction module obtains the pipe fitting number sequence determined by the pipe fitting features and the pipe system number sequence determined by the text features; The second step is to sort the pipe fitting numbers comprehensively according to (text probability of ID + feature recognition probability of ID) / number of recognitions; The third step is for the user to confirm whether the number is correct, confirm the number is correct or correct the number; The fourth step is to display the prefabricated pipe fitting assembly process information in the process window according to the number display.

[0014] The technical problem to be solved by the present invention can be further achieved by the following technical solutions. For the above-mentioned intelligent identification device for prefabricated ship pipes based on visual detection, a method for intelligent identification of prefabricated ship pipes based on visual detection is provided, and the steps are as follows: Step 1: Prefabrication project information database; The contents of the ship pipe design CAD and process data are converted into matching geometric attribute data in XML, JSON or other formats through the data extraction module, and stored in the edge engineering information database; Step 2: At the work site; In the prefabricated pipe pallet assembly environment, take out the pipes to be assembled to ensure that there is no stacking and the pipes are clearly labeled. On-site workers use pads to scan the pipes from multiple angles and in all directions. Step 3: The software performs intelligent recognition calculations in the background; The edge device detects and captures the image of the pipe fitting. The device software calls the deep learning training AI recognition model to identify the feature points and coded text logos of the captured pipe fitting image, and classifies the identified feature points. The pipe fitting feature combination is sent to the engineering information database for search and comparison. There may be some assembly pipe fittings with similar features but different models. The pipe fitting model with a higher probability is obtained and sent to the human-machine interface. Step 4: Process confirmation and review; On-site workers check the human-machine interface and select and confirm the pipe number based on probability sorting, or search for the pipe number; based on the determined pipe model, call the process data in the engineering information database for analysis and processing, and display the information on the pipe assembly method, assembly location and process in the form of animation and video for reference by on-site workers.

[0015] Compared with the prior art, the present invention provides an intelligent recognition method for prefabricated ship pipe fittings based on visual inspection. The pipe feature recognition AI model trained by image processing technology and deep learning algorithm is used to identify the feature points on the pipe fittings. The recognized feature points are combined and classified, and compared with the geometric data information of the pipe fittings in the database to obtain the pipe fitting type number with a higher probability and feed it back to the human-computer interaction interface. The on-site workers confirm the specific model of the pipe fitting through the human-computer interaction operation interface of the equipment. If there is no such number, they manually enter the graphic number. The human-computer interaction retrieves the cloud process guidance file according to the specific model information of the pipe fitting, and feeds back the guidance information such as the pipe fitting assembly method and assembly position for the on-site workers to refer to. Through intelligent recognition, automatic recognition and assembly of prefabricated ship pipe fittings can be realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A diagram of an equipment system of the present invention; Figure 2 The principle flow chart of the pipe feature recognition of the present invention; Figure 3 The principle flow chart of the pipe fitting text feature recognition of the present invention; Figure 4 Edge equipment pipe fitting detection and assembly flow chart of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] Reference Figure 1-4 The present invention provides an intelligent recognition device and method for ship prefabricated pipe fittings based on visual detection, which uses image processing technology and deep learning algorithms to realize automatic recognition of ship prefabricated pipe fittings through intelligent recognition, and can effectively improve the assembly efficiency of ship pipe units, ship unit modules, ship sections and other scenes. The specific technical scheme is as follows: During the ship assembly and welding process, on-site workers use edge devices to visually inspect the prefabricated installation and welding pipes of the ship sections. The edge device display interface can view the specific models of the pipes identified by the detection and related information of the pipe process guidance. The edge device system Figure 1 shown.

[0019] The edge device software mainly consists of image acquisition and preprocessing module, pipe feature recognition module, text feature recognition module, feature combination judgment module, text combination judgment module, engineering information database, human-computer interaction interface, and process library; The image acquisition preprocessing module acquires and processes the pipe fitting image and transmits it to the pipe fitting feature recognition module and the text feature recognition module; the pipe fitting feature recognition module recognizes the feature point information on the pipe fitting image and sends it to the feature combination judgment module, which performs statistical analysis and combination classification on the recognized feature points; compares the combination features with the engineering information database, obtains the pipe fitting number with the highest probability and sends it to the human-computer interface interaction module; the text feature recognition module recognizes the pipe fitting image coding information and sends it to the text combination judgment module, which searches and compares the recognized text combination with the engineering information database; The human-computer interface interaction includes the prefabricated pipe fitting number confirmation interface module and the process guidance display interface module; the prefabricated pipe fitting number confirmation interface module displays the comparison and judgment to obtain the pipe fitting model according to probability, and the process guidance display interface views the pipe fitting assembly process information; the engineering information database stores the ship production design data and process data in advance through data extraction; the prefabricated pipe fitting detection model is based on the YOLO algorithm deep learning training of pipe fitting image features and text AI recognition model.

[0020] The hardware part of the edge computing device consists of a camera, a touch screen and a communication module. The camera captures the image information of the pipe, the touch screen is used for manual interface operation, and the communication module is responsible for information transmission among the software modules.

[0021] Specific: The image acquisition and preprocessing module collects images of ship pipes by arranging cameras or handheld devices on site, captures images of pipes from multiple angles and in all directions in real time, and focuses on capturing local parts of pipes to ensure high-quality images of pipes. Pipe numbers are viewed as text symbols for auxiliary identification. At the same time, the collected pipe images are preprocessed by grayscale, contrast enhancement and other operations to improve image clarity and contrast.

[0022] The pipe feature recognition module uses deep learning training to train key features. The training types include flange joints, branch joints, elbows, tees, crosses and other major categories. Each category is divided into different subcategories by color. This is because prefabricated ship fittings will undergo different pretreatments before installation depending on the medium or installation process, and the final assembly form will be exported in different colors.

[0023] The pipe feature recognition module is a software module formed by secondary development of the intelligent recognition model trained in the deep learning training environment.

[0024] The recognition model in the pipe feature recognition module is usually trained by the following methods, such as Figure 2 As shown; Step 1: Dataset preparation: Take a large number of feature images of flange joints, branch joints, elbows, tees, crosses, etc. in the ship production environment. Each feature needs to be collected with different surface colors and environmental backgrounds. Step 2: Dataset annotation: Use annotation software to frame and annotate the features of the pipe fittings in the image and add feature attributes, such as XX pipe flange joint; Step 3: Use the YOLO algorithm to train the AI ​​model; First, the script converts the format and sends the produced image data set to the AI ​​model. The AI ​​model locates the bounding box through the features of the pipe image set and accurately detects the location of the image pipe features. According to the feature information reflected in the image data set, the algorithm training parameters are adjusted, and the algorithm for iteratively identifying pipe features is trained and optimized to obtain an AI recognition model with good accuracy such as precision, recall, average precision AP, and average precision mean mAP. pt weight files can be adjusted while ensuring accuracy to obtain lightweight and fast recognition speed weight files.

[0025] When identifying pipe fittings, the program side calls and outputs the trained AI model .pt weight file for prefabricated ship pipe fittings, and uses the AI ​​model to identify the features of the pipe fittings. The output information includes the prefabricated pipe fitting feature type, correct probability, and identification position, etc. For example, when an on-site worker identifies a pipe fitting, the visual inspection equipment takes a picture of the pipe fitting through the program side. When the program side identifies the flange and tee features, the annotation box displays the flange name and probability.

[0026] Since the pipe feature recognition module only recognizes the local features of the pipe, rather than the entire prefabricated pipe, further judgment is required based on the recognition results of the pipe feature recognition module; The judgment process is as follows: The first step is to count the combined data of feature points of the detected object according to the results of the pipe identification module, including the number and type of feature points, location, accuracy and other information; In the second step, the number and type of feature points of the detected object are searched globally with the project feature database to obtain possible prefabricated pipe fitting numbers; the comparison method usually consists of two methods: accurate judgment method and fuzzy judgment method; using accurate judgment, by comparing with the database, a completely matching prefabricated pipe fitting is obtained; using fuzzy judgment method, it is determined that the feature combination belongs to a subset of the prefabricated pipe fittings in the database, and the prefabricated pipe fittings that exit the match are obtained; Match rate calculation method: Matching rate = sum of the probability of matching feature points / total number of features; It is obtained by calculating the percentage of the number of feature points that match the number and the total number of feature points under the number; The possible prefabricated pipe fitting numbers obtained are sorted by feature matching rate.

[0027] In the deep learning environment, the AI ​​model is trained on handwritten, steel-stamped, and printed letters and numbers to obtain a text feature recognition AI model. The prefabricated pipe fitting text feature recognition module recognizes the pipe fitting number text features by calling the text feature recognition AI model. Figure 3 As shown; Step 1: Text image acquisition and processing: a large number of photos are taken in the ship production environment to obtain the entire string of text area numbers on the pipes, and the acquired images are enhanced and processed; Step 2: Coded text marking, the rectangular boundary box marks the location of the letter or number features of the pipe fitting text number; Step 3: Encoded text feature extraction: Use the AI ​​model to accurately detect the bounding box of the text feature location in the image set, extract and separate the detected text numbers, adjust the algorithm training parameters based on the feature information reflected in the image data set, and optimize and iteratively train the AI ​​recognition model; Step 4: Text feature recognition. Pipe fitting text feature recognition calculates the similarity between image text features by calling the AI ​​recognition model to find the numbering information with the highest matching degree.

[0028] Since the text feature recognition module of prefabricated pipe fittings usually recognizes some numbers, letters or symbols, the coding of the prefabricated pipe fittings needs to be compared with the database for inference. The inference method judges through global search similarity. The recognized text is searched in sequence with the number information of the engineering information database and sorted by similarity.

[0029] The engineering information database is mainly responsible for storing the prefabricated pipe fitting numbers and related process information. Its data needs to be preset before detection. Its data information is extracted and stored by pipe fitting numbers, feature point information, and process information.

[0030] The human-computer interaction module mainly presents the order of pipe fitting number information by probability sorting through the page. Workers confirm the pipe fitting number and click it according to the on-site confirmation situation. The process guidance module obtains relevant process information through the number. The process information is obtained through the prefabricated process information database, or the relevant process information is retrieved from the cloud site for on-site viewing; Step 1: The human-computer interaction module obtains the pipe fitting number sequence determined by the pipe fitting features and the pipe system number sequence determined by the text features; The second step is to sort the fitting numbers comprehensively according to (text probability of ID + feature recognition probability of ID) / number of recognitions; The third step is for the user to confirm whether the number is correct. Confirm the number is correct or correct the number; The fourth step is to display the prefabricated pipe fitting assembly process information in the process window according to the number display.

[0031] The use process of the present invention: Step 1: Prefabrication project information database; The ship pipe design CAD and process data are converted into matching XML, JSON or other format geometric attribute data through the data extraction module, and stored in the edge engineering information database; Step 2: At the work site; In the prefabricated pipe pallet assembly environment, take out the pipes to be assembled to ensure that there is no stacking and the pipes are clearly labeled. On-site workers use pads to scan the pipes from multiple angles and in all directions. Step 3: The software performs intelligent recognition calculations in the background; The edge device detects and captures the image of the pipe fitting. The device software calls the deep learning training AI recognition model to identify the feature points and coded text logos of the captured pipe fitting image, and classifies the identified feature points. The pipe fitting feature combination is sent to the engineering information database for search and comparison. There may be some assembly pipe fittings with similar features but different models. The pipe fitting model with a higher probability is obtained and sent to the human-machine interface. Step 4: Process confirmation and review; On-site workers check the human-machine interface and select and confirm the pipe number based on probability sorting, or search for the pipe number; based on the determined pipe model, call the process data in the engineering information database for analysis and processing, and display the pipe assembly method, assembly location and process information in the form of animation and video for on-site workers to refer to for assembly.

[0032] The present invention can assist the shipbuilding and related structural parts assembly and welding manufacturing industries to reduce costs and increase efficiency, and is of great significance for promoting the digital transformation of the manufacturing industry, implementing the country's independent and controllable strategy and the development plan of new generation information technology.

Claims

1. An intelligent identification device for prefabricated ship pipes based on visual inspection, characterized in that: It includes edge device software and edge computing device hardware. The edge device software includes image acquisition preprocessing module, pipe feature recognition module, text feature recognition module, feature combination judgment module, text combination judgment module, engineering information database, human-computer interaction interface and process library; The image acquisition preprocessing module is used to acquire and process the pipe image and transmit it to the pipe feature recognition module and the text feature recognition module; The pipe feature recognition module is used to recognize feature point information on the pipe image and send it to the feature combination judgment module; The text feature recognition module is used to recognize the pipe image coding information and send it to the text combination judgment module; The feature combination judgment module is used to perform statistical analysis and combination classification on the identified feature points, and compare the combination features with the engineering information database to obtain the pipe fitting number with the highest probability and send it to the human-machine interface interaction module; The character combination judgment module is used to search and compare the recognized character combination with the engineering information database; The engineering information database is used to store ship production design data and process data in advance through data extraction; The human-machine interface interaction includes a prefabricated pipe number confirmation interface module and a process guidance display interface module. The prefabricated pipe number confirmation interface module is used to display the pipe model obtained by comparison and judgment according to probability, and the process guidance display interface is used to view the pipe assembly process information; The edge computing device hardware includes a camera, a touch screen and a communication module. The camera is used to capture and obtain pipe image information, the touch screen is used for manual interface operation, and the communication module is responsible for information transmission among software modules.

2. According to claim 1, the intelligent identification device for prefabricated ship pipes based on visual detection is characterized in that: The prefabricated pipe detection model used by this equipment is based on the YOLO algorithm deep learning training of pipe image features and text AI recognition model.

3. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: The image acquisition and preprocessing module collects images of ship pipes by arranging cameras or handheld devices on site, captures images of the pipes from multiple angles and in all directions in real time, and focuses on capturing local parts of the pipes to obtain high-quality images of the pipes. The pipe numbers are viewed as text symbols for auxiliary identification, and the collected pipe images are preprocessed by grayscale and contrast enhancement to improve image clarity and contrast.

4. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: The pipe feature recognition module is trained with key features by deep learning training. The training types include flange joints, branch joints, elbows, tees, and crosses. Each category is divided into different subcategories by color. The pipe feature recognition module is a software module developed by the secondary development of the intelligent recognition model trained by the deep learning training environment. The recognition model in the pipe feature recognition module is usually trained by the following methods: Step 1: Dataset Preparation In the ship production environment, a large number of characteristic pictures of flange joints, branch joints, elbows, tees, and crosses are taken. Each feature needs to collect different surface colors and environmental backgrounds; Step 2: Dataset annotation: Use annotation software to mark the features of the pipes in the image and add feature attributes; Step 3: Train the AI ​​model using the YOLO algorithm First, the script converts the format and sends the produced image data set to the AI ​​model. The AI ​​model locates the bounding box through the features of the pipe image set and accurately detects the location of the image pipe features. According to the feature information reflected in the image data set, the algorithm training parameters are adjusted, and the algorithm for iteratively identifying pipe features is trained and optimized to obtain an AI recognition model with good accuracy in terms of accuracy, recall rate, average precision AP, and average precision mean AP. pt weight file. Under the premise of ensuring accuracy, the parameters can be adjusted to obtain a lightweight and fast recognition speed weight file. When identifying pipe fittings, the program calls and outputs the trained AI model .pt weight file for prefabricated ship fittings, and identifies the features of the pipe fittings through the AI ​​model. The output information includes a combination of prefabricated pipe fitting feature type, correct probability and identification location.

5. The intelligent identification device for prefabricated ship pipes based on visual detection according to claim 1 or 4, characterized in that: Since the pipe feature recognition module only recognizes the local features of the pipe, rather than the entire prefabricated pipe, further judgment is required based on the recognition results of the pipe feature recognition module; The judgment process is as follows: The first step is to count the feature point combination data of the detected object according to the results of the pipe identification module, including the number and type of feature points, location and accuracy information; In the second step, the number and type of feature points of the detected object are globally searched with the project feature database to obtain the possible prefabricated pipe fitting numbers; The comparison method includes two methods: accurate judgment method and fuzzy judgment method; Use quasi-accurate judgment method to compare with the database to obtain fully matched prefabricated pipe fittings; Through the fuzzy judgment method, it is judged that the feature combination belongs to the subset of prefabricated pipe fittings in the database, and the prefabricated pipe fittings that exit the matching are obtained; Matching rate calculation method: Matching rate = sum of the probability of matching feature points / total number of features; The possible prefabricated pipe number is obtained by calculating the percentage of the number of characteristic points that meet the number and the total number of characteristic points under the number. Sort by feature matching rate.

6. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: Since the text feature recognition module of prefabricated pipe fittings usually recognizes some numbers, letters or symbols, the coding of the prefabricated pipe fittings needs to be compared with the database for inference. The inference method judges through global search similarity. The recognized text is searched in sequence with the number information of the engineering information database and sorted by similarity.

7. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: The prefabricated pipe fittings text feature recognition module trains handwritten, steel stamped, and printed letters and numbers in the deep learning environment AI model to obtain a text feature recognition AI model. The prefabricated pipe fittings text feature recognition module recognizes the text features of the pipe fitting numbers by calling the text feature recognition AI model: Step 1: Text image acquisition and processing: a large number of photos are taken in the ship production environment to obtain the entire string of text area numbers on the pipes, and the acquired images are enhanced and processed; Step 2: Coded text marking, the rectangular boundary box marks the location of the letter or number features of the pipe fitting text number; Step 3: Encoded text feature extraction: Use the AI ​​model to accurately detect the bounding box of the text feature location in the image set, extract and separate the detected text numbers, adjust the algorithm training parameters based on the feature information reflected in the image data set, and optimize and iteratively train the AI ​​recognition model; Step 4: Text feature recognition. Pipe fitting text feature recognition calculates the similarity between image text features by calling the AI ​​recognition model to find the numbering information with the highest matching degree.

8. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: The engineering information database is also responsible for storing the prefabricated pipe fitting numbers and related process information, and its data needs to be preset before detection. Its data information is extracted and stored by pipe fitting numbers, feature point information and process information.

9. The intelligent identification device for prefabricated ship pipes based on visual inspection according to claim 1 is characterized in that: The human-computer interaction module presents the order of pipe fitting number information by sorting the page according to probability. The worker confirms the pipe fitting number and clicks it according to the on-site confirmation situation. The process guidance module obtains relevant process information through the number. The process information is obtained through the prefabricated process information library, or the relevant process information is retrieved from the cloud site for on-site viewing: Step 1: The human-computer interaction module obtains the pipe fitting number sequence determined by the pipe fitting features and the pipe system number sequence determined by the text features; The second step is to sort the pipe fitting numbers comprehensively according to (text probability of ID + feature recognition probability of ID) / number of recognitions; The third step is for the user to confirm whether the number is correct, confirm the number is correct or correct the number; The fourth step is to display the prefabricated pipe fitting assembly process information in the process window according to the number display.

10. A method for intelligent identification of prefabricated ship pipes based on visual inspection, characterized in that: The method uses the intelligent identification device for prefabricated ship pipes based on visual detection according to any one of claims 1 to 9, and the steps are as follows: Step 1: Prefabrication project information database; The contents of the ship pipe design CAD and process data are converted into matching geometric attribute data in XML, JSON or other formats through the data extraction module, and stored in the edge engineering information database; Step 2: At the work site; In the prefabricated pipe pallet assembly environment, take out the pipes to be assembled to ensure that there is no stacking and the pipes are clearly labeled. On-site workers use pads to scan the pipes from multiple angles and in all directions. Step 3: The software performs intelligent recognition calculations in the background; The edge device detects and captures the image of the pipe fitting. The device software calls the deep learning training AI recognition model to identify the feature points and coded text logos of the captured pipe fitting image, and classifies the identified feature points. The pipe fitting feature combination is sent to the engineering information database for search and comparison. There may be some assembly pipe fittings with similar features but different models. The pipe fitting model with a higher probability is obtained and sent to the human-machine interface. Step 4: Process confirmation and review; On-site workers check the human-machine interface and select and confirm the pipe number based on probability sorting, or search for the pipe number; based on the determined pipe model, call the process data in the engineering information database for analysis and processing, and display the information on the pipe assembly method, assembly location and process in the form of animation and video for reference by on-site workers.

Images