CN-121982354-A - Shipyard safety production hidden trouble investigation method and system based on AI image identification

CN121982354ACN 121982354 ACN121982354 ACN 121982354ACN-121982354-A

Abstract

The invention discloses a shipyard safety production hidden trouble checking method and system based on AI image recognition, and relates to the technical field of safety production. The method comprises the steps of collecting shipyard scene data, training a target detection model according to hidden danger data, detecting hidden danger, generating a visual map through hidden danger detection, and conducting safety production hidden danger investigation, wherein the shipyard scene data comprise personnel individual images, personnel action images, equipment operation images and shipyard environment images. The invention is based on deep learning models such as YOLO, CNN and the like, can accurately identify illegal behaviors such as unworn safety helmet, smoking in a smoking forbidden area and the like, has high identification accuracy, can complete detection in a short time, and can discover potential safety hazards in time. For example, the identification of the action of not wearing the safety helmet can be completed in a short time, and the real-time performance is ensured.

Inventors

JIN HAO
WANG JIWU
KANG JIAN

Assignees

船舶信息研究中心(中国船舶集团有限公司第七一四研究所)
北京石油化工学院

Dates

Publication Date: 20260505
Application Date: 20251205

Claims (10)

1. The potential safety hazard investigation method for shipyard safety production based on AI image identification is characterized by comprising the following steps: S1, collecting shipyard scene data, wherein the shipyard scene data comprise personnel individual images, personnel action images, equipment running images and shipyard environment images; S2, training a target detection model according to hidden danger data; s3, detecting hidden danger; s4, generating a visual map through hidden danger detection, and conducting safety production hidden danger investigation.
2. The method for checking hidden danger in shipyard safety production based on AI image recognition according to claim 1, wherein in step S1, the method for collecting and integrating hidden danger data comprises: S11, marking hidden danger information in a scene image of a shipyard; s12, storing the marked hidden danger information into a database, and performing normalization processing; s13, setting preset time intervals, and cleaning data in a database every preset time interval to remove repeated and error information; and S14, classifying and summarizing the cleaned data according to the characteristics of hidden danger position coordinates, types and the like, and generating an intermediate data file.
3. The method for checking hidden danger in shipyard safety production based on AI image recognition according to claim 1, wherein in step S2, the method for training the target detection model comprises: s21, training a YOLO model through an unworn safety helmet image and a safety belt image marked in an individual image of a person and a smoke image marked in a shipyard environment image; s22, introducing an OpenCV, and training the OpenCV through smoking action images marked in the personnel action images; s23, adopting a data enhancement technology to improve the generalization capability of the model.
4. The method for checking hidden danger in shipyard safety production based on AI image recognition according to claim 1, wherein in step S3, the method for detecting hidden danger comprises: after the personal images are collected, a model is called to carry out target detection, and the behavior of not wearing the safety helmet is detected; According to the real-time motion image acquired by the smoking-forbidden zone, combining motion capture and smoke feature analysis, extracting a hand motion profile and a motion track through OpenCV, and judging whether suspicious smoking motion exists or not; And analyzing the smoke drifting track according to the smoke shape image and the optical flow method, and judging that the smoking is illegal when the action and the smoke characteristic conditions are met.
5. The method for checking hidden danger in shipyard safety production based on AI image recognition of claim 4, wherein in step S3, the method for detecting hidden danger further comprises: welding spark splash detection specifically includes: Collecting an image of a welding area by using a camera, and highlighting a spark area through image graying and binarization processing; based on a connected domain analysis algorithm, counting the number and distribution of sparks, and judging whether the sparks are splashed out of a safe range by combining welding process parameters, and if so, sending a deceleration or pause signal to welding equipment; the equipment high temperature anomaly identification detection specifically comprises: acquiring the temperature distribution image of the surface of the equipment in real time by an infrared thermal imaging camera, Building a thermal imaging analysis model in the YOLO model, positioning a temperature abnormal region, and performing early warning by the system when the equipment temperature exceeds a preset threshold value; Fire control passageway jam detects, specifically includes: the passage area is distinguished from the stacked objects by adopting a semantic segmentation technology, when the proportion of the occupied passage area of the objects exceeds a preset blockage threshold value, the objects are judged to be blocked, an alarm is triggered, Generating a report containing the location and severity of the blockage; the identifying and detecting of the hidden defect of the ship structure specifically comprises the following steps: acquiring an image acquired by the X-ray detection device and data acquired by the ultrasonic detection device, Analyzing the X-ray image through a CNN model, identifying the trend and the length of the crack, And combining ultrasonic signal characteristic extraction to diagnose hidden danger of the ship structure.
6. The method for checking hidden danger in shipyard safety production based on AI image recognition of claim 1, wherein in step S4, the method for generating and displaying a visual map comprises: Importing hidden danger data into plant floor plan to QGIS software, wherein the floor plan comprises buildings, roads and equipment layout; Creating a layer in QGIS, associating hidden danger data with the layer, and automatically generating a point position mark on a plan according to hidden danger position coordinates; Adopting a color coding system, and simultaneously superposing thermodynamic diagrams to present a risk aggregation area; And embedding the QGIS generated visual map into a Grafana open source visual platform, and setting a timing refreshing mechanism.
7. The method for checking potential safety hazards in shipyard production based on AI image recognition according to any one of claims 1-6, wherein after step S4, further comprises: s5, rectifying and modifying hidden danger information, specifically comprising the following steps: Establishing a hidden trouble rectifying and modifying feedback system; the staff enters a hidden danger detail page through hidden danger point location information; After the correction is finished, a worker uploads a correction picture in a hidden danger correction feedback system, the hidden danger correction feedback system identifies hidden danger positions and correction conditions in the picture, and judges whether the correction meets the standards or not through comparison of an image comparison algorithm and the picture when the hidden danger is reported; If yes, marking the hidden danger state as checked and accepted, and generating a closed loop record comprising photos before and after correction, correction time and correction personnel; if not, returning to carry out the renovation, wherein the renovation progress is dynamically displayed on the signboard in real time; hidden danger state change triggers database update and automatically generates a time axis component.
8. The shipyard safety production hidden trouble investigation system based on AI image recognition is characterized by being applied to the shipyard safety production hidden trouble investigation method based on AI image recognition as claimed in claims 1-7, comprising: The camera comprises an action camera and a thermal imaging camera, and is arranged at a shipyard key point and used for acquiring shipyard scene data; The data processing center is provided with a server cluster and is used for running a deep learning model and storing massive images and hidden danger data; the wireless network is used for connecting the camera and the data processing center and transmitting shipyard scene data acquired by the camera to the data processing center.
9. The AI-image-recognition-based shipyard safety production hidden trouble shooting system of claim 8, wherein an edge computing device is further installed near the camera and is used for reducing data transmission delay and performing preliminary processing on images.
10. The AI image recognition-based shipyard safety production hidden trouble shooting system of claim 8, wherein the data processing center is provided with a CentOS Linux operating system and an OpenCV image processing library, and PyTorch is selected as a deep learning framework.

Description

Shipyard safety production hidden trouble investigation method and system based on AI image identification Technical Field The invention relates to the technical field of safety production, in particular to a shipyard safety production hidden trouble checking method and system based on AI image recognition. Background In the marine manufacturing industry, safety production management is always facing extremely complex challenges. The shipyard operation environment has the characteristics of high risk, dynamic performance and space complexity, and relates to multiple risk superposition scenes such as high-altitude operation, three-dimensional cross construction, closed cabin operation, large-scale hoisting equipment operation, inflammable and explosive storage and the like, and potential hidden hazards comprise major accident risks such as structural collapse, mechanical injury, gas poisoning, fire explosion and the like. The traditional safety management highly depends on a manual inspection and paper recording system, has the obvious defects that the manual inspection is insufficient in hidden danger identification rate such as chemical mixing of paint warehouse, mounting position error of a gas detector (such as that a combustible gas probe is arranged at a height of 1.5 m from 0.3 m), missing of anti-explosion equipment and the like, especially in a large gantry crane track area, a ballast tank and other visual blind areas, the paper flow causes long average time from hidden danger discovery to maintenance approval, the illegal live inspection cannot be timely stopped, dynamic risks such as high-altitude walking of a safety belt are not tied, data analysis is difficult to realize by a dispersed hidden danger recording book, and the risks such as high incidence rule of illegal operation in a specific period (such as night overtime), relevance of a deep-layer continuous accident caused by a specific equipment (such as a cutting machine) fault and the like cannot be predicted. The prior trial transformation technology only realizes local optimization, namely a factory video monitoring system is limited to post evidence obtaining, lacks the capability of actively identifying illegal behaviors such as unworn safety helmet smoking in a smoking-forbidden area, focuses on production progress management, and lacks an intelligent diagnosis engine to diagnose hidden equipment status hazards such as rail gnawing of a crane, grounding abnormality of a welding machine and the like in real time. Industry needs a set of novel system of depth fusion real-time image intelligent diagnosis and global hidden trouble space visualization technology to solve the problems of multiple manual inspection blind areas, serious response delay and blank core pain points of risk prediction, and truly realize the fundamental transition from passive treatment to active defense. Disclosure of Invention The invention provides a shipyard safety production hidden trouble investigation method and system based on AI image recognition, and aims to solve the problems of more artificial inspection blind areas, serious response delay and blank risk prediction in the prior art. In order to solve the technical problems, the invention adopts the following technical scheme: the invention provides a shipyard safety production hidden trouble checking method based on AI image identification, which comprises the following steps: S1, collecting shipyard scene data, wherein the shipyard scene data comprise personnel individual images, personnel action images, equipment running images and shipyard environment images; S2, training a target detection model according to hidden danger data; s3, detecting hidden danger; s4, generating a visual map through hidden danger detection, and conducting safety production hidden danger investigation. On the basis, the invention can be further improved in that in the step S1, the method for collecting and integrating hidden danger data comprises the following steps: S11, marking hidden danger information in a scene image of a shipyard; s12, storing the marked hidden danger information into a database, and performing normalization processing; s13, setting preset time intervals, and cleaning data in a database every preset time interval to remove repeated and error information; and S14, classifying and summarizing the cleaned data according to the characteristics of hidden danger position coordinates, types and the like, and generating an intermediate data file. On the basis, the invention can be further improved in the step S2, the method for training the target detection model comprises the following steps: s21, training a YOLO model through an unworn safety helmet image and a safety belt image marked in an individual image of a person and a smoke image marked in a shipyard environment image; s22, introducing an OpenCV, and training the OpenCV through smoking action images marked in the personnel action images; s23, adopting