Intelligent monitoring system for ship loading and unloading hose curvature

Abstract

The application discloses a ship loading and unloading hose curvature intelligent monitoring system and relates to the field of intelligent monitoring.The system comprises a panoramic view acquisition module, a panoramic image fusion module, a simulation driving analysis module, a multi-state parallel architecture module and a hose early warning analysis module.The panoramic view acquisition module is used for acquiring visual acquisition information.The panoramic image fusion module is used for carrying out multi-dimensional non-overlapping splicing fusion processing on the visual acquisition information to acquire three-dimensional hose visual information.The simulation driving analysis module carries out simulation driving processing on the ship loading and unloading process according to the three-dimensional hose visual information to acquire corresponding fault parameter data sets.The multi-state parallel architecture module is used for setting corresponding target nodes on the three-dimensional hose visual information, carrying out space-time fusion processing on state data in the target nodes and acquiring corresponding state information.The hose early warning analysis module is used for carrying out checking and early warning analysis on the state information corresponding to each target node, thereby improving the accuracy of monitoring in the hose use process.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, and in particular to an intelligent monitoring system for the curvature of ship loading and unloading hoses. Background Technology

[0002] With the continuous improvement of my country's market competition mechanism, the application scope of hazardous chemicals such as petroleum and LNG is constantly expanding. With the expansion of the volume and scale of hazardous chemicals transportation in my country's ports, oil hoses, as an indispensable equipment component in the loading and unloading of hazardous chemicals in ports, bear important safety responsibilities. However, at present, there is a lack of effective and continuous real-time monitoring mode for oil hoses in actual port operations, and the material properties and structural characteristics of oil hoses themselves limit the use of monitoring methods. In traditional ship-port oil transfer processes, visual inspection or video system-based manual monitoring of oil hose conditions remains the most common method. However, relying entirely on human experience and sensor installation and monitoring results in large deviations and cannot achieve real-time monitoring and recording or random adjustments based on hose stress and internal / external loads. This also leads to high labor costs and low accuracy. Therefore, this invention provides an intelligent monitoring system for the curvature of ship loading and unloading hoses. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies, such as large deviations and low efficiency in the monitoring process of oil transfer hoses, and to propose an intelligent monitoring system for the curvature of ship loading and unloading hoses.

[0004] To achieve the above objectives, the present invention adopts the following technical solution: The intelligent monitoring system for the curvature of ship loading and unloading hoses includes: The panoramic view acquisition module is used to set up corresponding optical imaging devices according to the spatial layout of the storage tank and the dock, set up view acquisition nodes according to the optical imaging devices, and acquire visual acquisition information at the corresponding locations through the view acquisition nodes. The panoramic image fusion module is used to perform initial fusion processing on the visual acquisition information collected by each view acquisition node, obtain fusion coordinate points, and perform multi-dimensional non-overlapping stitching and fusion processing on the corresponding visual acquisition information based on the corresponding fusion coordinate points to obtain three-dimensional hose visual information. The simulation-driven analysis module is used to perform simulation-driven processing on the corresponding positions based on the obtained 3D hose visual information, obtain the driving parameter information of the corresponding oil hose, identify fault parameters from the obtained driving parameter information, and generate a fault parameter dataset. The multi-state parallel architecture module is used to perform target detection and target segmentation processing based on the obtained 3D hose visual information, obtain multiple target nodes according to the target processing results, perform running status analysis based on each target node, and mark them at the corresponding positions in the 3D hose visual image. The hose early warning analysis module is used to set up a fault analysis model for the corresponding fault parameter dataset, input the analysis results of the operating status of the corresponding target nodes in the 3D hose visual image into the fault analysis model for analysis and processing, and obtain the fault early warning information of the corresponding target nodes.

[0005] The above technical solution further includes: the panoramic view acquisition module includes: The visual layout unit is used to acquire information on the corresponding ship loading and unloading areas within the storage tank and the dock, acquire oil hose operation information based on the ship loading and unloading area information, arrange the position of the optical imaging device based on the oil hose operation information, and set the view acquisition nodes based on the position layout results. The visual acquisition unit is used to acquire and process data based on the optical imaging device corresponding to the view acquisition node. The view acquisition node is used to dynamically adjust the acquisition parameters of the corresponding optical imaging device according to the corresponding oil hose operation information and acquire the corresponding visual acquisition information.

[0006] Furthermore, the panoramic image fusion module includes: The feature recognition unit is used to acquire visual acquisition information obtained by each view acquisition node, perform target feature recognition on the visual acquisition information obtained by each view acquisition node, and perform labeling processing based on the target feature recognition results; The initial fusion unit is used to acquire the target feature recognition results of each marker processing, perform matching processing on the target feature recognition results, perform initial fusion processing based on the matching processing results, and generate an initial fused image. The multidimensional fusion unit is used to set fusion coordinate points for the initial fusion image corresponding to the matching processing result, and to perform multidimensional non-overlapping stitching and fusion processing on each initial fusion image according to the fusion coordinate points.

[0007] Furthermore, the multidimensional fusion unit includes: The coordinate determination unit acquires the corresponding initial fused image, sets the corresponding feature fusion layer according to the target feature recognition results included in the initial fused image, acquires the fusion coordinate points and key coordinate points corresponding to the feature fusion layer in the corresponding initial fused image, marks each fusion coordinate point and key coordinate point, and sets them as the fusion coordinate points of the initial fused image to which the corresponding feature fusion layer belongs. The stitching processing unit acquires the corresponding fusion coordinate points within the corresponding feature fusion layer, sequentially selects two or more fusion coordinate points within the corresponding feature fusion layer, performs image matching based on the selected fusion coordinate points, sets a three-dimensional spatial coordinate system, maps the position information of the view acquisition nodes corresponding to the corresponding feature fusion layer and the corresponding fusion coordinate points to the three-dimensional spatial coordinate system, acquires the corresponding fusion pixels, acquires the corresponding fusion pixels of each initial fusion image based on the mapping results of the corresponding fusion coordinate points and key coordinate points within each feature fusion layer, and performs fusion processing on the fusion pixels in the three-dimensional spatial coordinate system based on the image fusion algorithm to acquire a three-dimensional hose stitched image; The deduplication processing unit is used to monitor the overlapping areas of the 3D hose spliced ​​image obtained by the splicing processing unit, obtain the overlapping areas of pixels in the 3D hose spliced ​​image, and perform deduplication processing on the overlapping areas of pixels by adaptively adjusting the fusion parameters based on the regional features, thereby obtaining the 3D hose visual information.

[0008] Furthermore, the simulation-driven analysis module includes: The simulation driving unit is used to perform ship loading and unloading simulation driving processing on the obtained three-dimensional hose visual information, obtain hose curvature change data of the corresponding oil hose during loading and unloading, and synchronously map the hose curvature change data with the corresponding three-dimensional hose visual information to obtain the corresponding driving parameter information, and set the corresponding three-dimensional hose driving visual model according to the driving parameter information. The fault integration unit is used to perform simulation-driven processing on different types of hazardous chemicals and loading and unloading data according to the three-dimensional hose-driven visual model, obtain hose curvature change data corresponding to the oil hose, identify fault parameters from the hose curvature change data, and obtain the corresponding fault parameter dataset.

[0009] Furthermore, the multi-state parallel architecture module includes: The target processing unit acquires loading and unloading hose operation information, obtains corresponding 3D hose visual information based on the loading and unloading hose operation information, performs target detection on the 3D hose visual information, obtains corresponding hose identification area information, performs bend recognition on the hose identification area information, obtains corresponding hose bend recognition area information, performs target segmentation based on the corresponding hose identification area information and hose bend recognition area information, and sets target nodes for the corresponding hose bend recognition area information and other areas respectively. The operation processing unit is used to analyze and process the operating status of multiple target nodes based on the hose bending identification area information.

[0010] Furthermore, the operation processing unit includes: The statistical processing unit is used to analyze the running status of the corresponding identification area by the corresponding target node, obtain the view acquisition node corresponding to the corresponding identification area, set the time series axis according to the corresponding view acquisition node, and the target node maps the hose curvature change data of the corresponding identification area per unit time to the time series axis according to the corresponding view acquisition node. The unit performs change statistical analysis on the corresponding hose curvature change data according to the time series axis to obtain the sequence change data. The analysis and processing unit is used to extract features from the sequence change data corresponding to each time series axis within the corresponding target node, and obtain the state features corresponding to the corresponding time series axis respectively. The integration processing unit acquires the state features corresponding to the time series axes within the target node, integrates and statistically analyzes the frequency of occurrence of the corresponding state features, determines the type of the corresponding state features, sets the state confidence level according to the type of the corresponding state features, and integrates the state features of each time series axis within the corresponding target node according to the corresponding state confidence level to generate target state data.

[0011] Furthermore, the hose early warning analysis module includes: The fault analysis unit is used to acquire the fault parameter dataset corresponding to the 3D hose visual image, analyze and train the fault parameter dataset based on machine learning algorithm, construct a fault analysis model, input the target state data obtained from the corresponding target node into the fault analysis model for analysis and processing, and obtain fault analysis data. The fault verification unit is used to acquire fault analysis data corresponding to each target node, sort the corresponding marked target nodes according to the 3D hose visual image, acquire the correlation data of the fault analysis data corresponding to each target node, perform verification processing according to the corresponding correlation data, and output the fault verification data corresponding to the target node in the 3D hose visual image. The fault warning unit is used to distinguish the corresponding fault verification data in the three-dimensional hose visual image, generate fault warning data based on the distinction results of the fault verification data, and perform warning processing on the identification area corresponding to the corresponding target node based on the generated fault warning data.

[0012] The present invention has the following beneficial effects: 0. In this invention, the visual acquisition information collected by various optical imaging devices is initially fused, and the obtained initial fused image is set with a corresponding fusion layer according to the corresponding feature recognition results. The feature recognition results that may be related between the fused images are analyzed by coordinate analysis to obtain the corresponding fusion coordinate points. The corresponding key coordinate points are set to fuse the data information within and between the fusion layers to obtain the corresponding three-dimensional hose visual information. This improves the accuracy of the fusion process and the utilization rate of visual acquisition information, and obtains the feature information of the corresponding oil hose to the greatest extent.

[0013] 0. In this invention, by dividing the visual information of the three-dimensional hose into recognition areas and setting corresponding target nodes, and by setting corresponding time series axes according to the position information of the corresponding view acquisition nodes in the recognition area, the hose loading and unloading process in the corresponding recognition area is subjected to time and space dual coupling analysis, thereby improving the accuracy of state analysis during the hose loading and unloading process.

[0014] 0. In this invention, by verifying and analyzing the fault analysis data obtained from the target nodes corresponding to each identification area within the three-dimensional hose visual information, and by verifying the correlation data corresponding to adjacent target nodes, the accuracy and efficiency of the analysis process of abnormal loading and unloading status of the corresponding identification area during the loading and unloading of ship oil hoses are improved to a certain extent. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the intelligent monitoring system for the curvature of ship loading and unloading hoses proposed in this invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 like Figure 1 As shown, the intelligent monitoring system for the curvature of ship loading and unloading hoses proposed in this invention constructs a hose simulation monitoring and management platform. The hose simulation monitoring and management platform is equipped with a panoramic view acquisition module, a panoramic image fusion module, a simulation-driven analysis module, a multi-state parallel architecture module, and a hose early warning analysis module.

[0018] In this embodiment, the curvature calculation of the corresponding oil hoses in the simulated test site of the corresponding tank terminal and the monitoring and processing of the hazardous chemical transportation process are obtained through the hose simulation monitoring and management platform, thereby improving the safety of the corresponding hazardous chemicals during the loading and unloading of ships and reducing losses during the transfer process to a certain extent. The specific implementation process includes: The panoramic view acquisition module is used to set up corresponding optical imaging devices according to the spatial layout of the storage tank and the dock, set up view acquisition nodes according to the optical imaging devices, and acquire visual acquisition information at the corresponding locations through the view acquisition nodes. The panoramic image fusion module is used to perform initial fusion processing on the visual acquisition information collected by each view acquisition node, obtain fusion coordinate points, and perform multi-dimensional non-overlapping stitching and fusion processing on the corresponding visual acquisition information based on the corresponding fusion coordinate points to obtain three-dimensional hose visual information. The simulation-driven analysis module is used to perform simulation-driven processing on the corresponding positions based on the obtained 3D hose visual information, obtain the driving parameter information of the corresponding oil hose, identify fault parameters from the obtained driving parameter information, and generate a fault parameter dataset. The multi-state parallel architecture module is used to perform target detection and target segmentation processing based on the obtained 3D hose visual information, obtain multiple target nodes according to the target processing results, perform running status analysis based on each target node, and mark them at the corresponding positions in the 3D hose visual image. The hose early warning analysis module is used to set up a fault analysis model for the corresponding fault parameter dataset, input the analysis results of the operating status of the corresponding target nodes in the 3D hose visual image into the fault analysis model for analysis and processing, and obtain the fault early warning information of the corresponding target nodes.

[0019] In specific implementation, the panoramic view acquisition module includes: The visual layout unit is used to acquire information on the corresponding ship loading and unloading areas within the storage tank and the dock, acquire oil hose operation information based on the ship loading and unloading area information, arrange the position of the optical imaging device based on the oil hose operation information, and set the view acquisition nodes based on the position layout results. It should be further explained that, in the specific implementation process, the storage tanks and wharves include basic information of the storage tank facilities, wharf layout information, wharf facility basic information, and ship loading and unloading area information; the ship loading and unloading area information includes the corresponding storage tank location information marker number, ship wharf location information marker number, and corresponding spatial layout information; the oil hose operation information includes the corresponding oil hose diameter information, basic information, storage tank marker number information, ship wharf marker number information, hazardous chemical loading and unloading information, and loading and unloading task information during the ship loading and unloading process; based on the corresponding ship loading and unloading area information and oil hose operation, the optical imaging devices involved in the corresponding oil hose loading and unloading process are positioned accordingly, wherein the optical imaging devices include corresponding light source devices and camera devices, and during the positioning process, the corresponding oil hoses are kept within the monitoring range of the optical imaging devices; The visual acquisition unit is used to acquire and process data based on the optical imaging device corresponding to the view acquisition node. The view acquisition node is used to dynamically adjust the acquisition parameters of the corresponding optical imaging device according to the corresponding oil hose operation information and acquire the corresponding visual acquisition information. It should be further explained that, in the specific implementation process, the corresponding view acquisition node is used to connect with the corresponding optical imaging device. The view acquisition node adjusts the light source and camera parameters of the light source and camera devices in the corresponding optical imaging device, and sets up the corresponding wireless terminal to obtain the image acquisition results obtained by the corresponding optical imaging device at each location. The corresponding image acquisition results are then marked according to the corresponding location layout information to obtain the corresponding visual acquisition information.

[0020] In specific implementation, the panoramic image fusion module includes: The feature recognition unit is used to acquire visual acquisition information obtained by each view acquisition node, perform target feature recognition on the visual acquisition information obtained by each view acquisition node, and perform labeling processing based on the target feature recognition results; It should be further explained that, in the specific implementation process, the corresponding oil hose operation information includes the pipe diameter information and basic information of the corresponding oil hose, which includes the pipe diameter marking information and pipe length marking information of the corresponding oil hose. During the target feature recognition process, the corresponding feature recognition unit identifies the pipe diameter marking information and pipe length marking information on the oil hose, determines the position information of the corresponding oil hose based on the corresponding marking information, extracts the corresponding oil hose position information in the corresponding visual acquisition information based on the visual analysis algorithm, and marks the oil hose in the identified visual acquisition information as the corresponding target feature recognition result. The initial fusion unit is used to acquire the target feature recognition results of each marker processing, perform matching processing on the target feature recognition results, perform initial fusion processing based on the matching processing results, and generate an initial fused image. It should be further explained that, in the specific implementation process, the initial fusion unit performs matching processing on the corresponding target feature recognition, and in the process of initial fusion, it sequentially marks and fuses the pipe diameter marking information and pipe length marking information corresponding to the corresponding target feature recognition results, fuses and marks the visual acquisition information belonging to the same location information of the oil hose, sorts the location information of the corresponding visual acquisition information according to the corresponding oil hose and target feature recognition results, and obtains the initial fused image of the corresponding oil hose. The corresponding initial fused image is the approximate location of the oil hose to which the visual acquisition information belongs to the layout location information corresponding to the corresponding optical imaging device, that is, the obtained visual acquisition information is initially integrated; The multidimensional fusion unit is used to set fusion coordinate points for the initial fusion images corresponding to the matching processing results, and to perform multidimensional non-overlapping stitching and fusion processing on the initial fusion images according to the fusion coordinate points, including: The coordinate determination unit acquires the corresponding initial fused image, sets the corresponding feature fusion layer according to the target feature recognition results included in the initial fused image, acquires the fusion coordinate points and key coordinate points corresponding to the feature fusion layer in the corresponding initial fused image, marks each fusion coordinate point and key coordinate point, and sets them as the fusion coordinate points of the initial fused image to which the corresponding feature fusion layer belongs. It should be further explained that, in the specific implementation process, when setting feature fusion layers for the initial fused image, the corresponding feature fusion layers are the information related to the corresponding oil hoses collected by the corresponding view acquisition nodes. The information in each feature fusion layer is analyzed and processed first, and the corresponding feature fusion layers are analyzed and processed based on the neural network algorithm to obtain the corresponding layer feature data. Each feature fusion layer is compared and analyzed sequentially to obtain the corresponding layer feature data at the corresponding position in the initial fusion image. The corresponding layer feature data is then compared and analyzed. Based on the comparison results, the fusion coordinate points corresponding to the corresponding feature fusion layer are obtained and marked within the feature fusion layer. After all the feature fusion layers at the corresponding positions in the initial fusion image are marked, the corresponding layer feature data within each feature fusion layer is divided into fusion coordinate points and key coordinate points. The key coordinate points are layer feature data contained within the corresponding feature fusion layer that are not contained within other feature fusion layers. The stitching processing unit acquires the corresponding fusion coordinate points within the corresponding feature fusion layer, sequentially selects two or more fusion coordinate points within the corresponding feature fusion layer, performs image matching based on the selected fusion coordinate points, sets up a three-dimensional spatial coordinate system, maps the position information of the view acquisition nodes corresponding to the corresponding feature fusion layer and the corresponding fusion coordinate points to the three-dimensional spatial coordinate system, acquires the corresponding fusion pixels, and acquires the corresponding fusion pixels of each initial fusion image based on the mapping results of the corresponding fusion coordinate points and key coordinate points within each feature fusion layer. Based on the gradient-weighted class activation mapping technique Grad-CAM, the unit explores the role mechanism of the attention mechanism in the corresponding fusion pixels and key pixels within the corresponding feature fusion layer during the construction of the three-dimensional hose stitched image, and acquires the weight information and differential representation of the corresponding feature fusion layer in the construction of the three-dimensional hose stitched image. Based on the weight information and differential representation analysis results of the corresponding feature fusion layer, the unit performs fusion processing on the fusion pixels within the three-dimensional spatial coordinate system using an image fusion algorithm to acquire the three-dimensional hose stitched image. The deduplication processing unit is used to monitor the overlapping area of ​​the 3D hose spliced ​​image obtained by the splicing processing unit, obtain the overlapping area of ​​pixels in the 3D hose spliced ​​image, and perform deduplication processing on the overlapping area of ​​pixels by adaptively adjusting the fusion parameters based on the regional features to obtain the 3D hose visual information. It should be further explained that, in the specific implementation process, the stitching processing unit, in setting the three-dimensional hose stitching image, uses the corresponding fusion coordinate points as a basis, obtains the corresponding extreme points based on the self-attention mechanism algorithm, extracts the position, scale, and rotation-invariant feature information, sets the corresponding fusion pixel points, and performs matching processing on each fusion pixel point. Based on the matching processing result, the corresponding feature fusion layers are overlapped. Based on the overlap processing result, the corresponding feature fusion layers are displayed in the three-dimensional spatial coordinate system according to the corresponding fusion pixel points. Based on the display result of each feature fusion layer in the three-dimensional spatial coordinate system, the three-dimensional hose stitching image is obtained. During the monitoring of overlapping regions between corresponding feature fusion layers in a 3D hose spliced ​​image, grayscale values ​​are processed at the corresponding positions of each feature fusion layer in the 3D spatial coordinate system by matching corresponding pixel pairs. Grayscale similarity analysis is performed based on the grayscale processing results. The corresponding pixel overlap regions are then obtained based on the grayscale similarity analysis results. The obtained pixel overlap regions are deduplicated according to the weight information and differential representation of the corresponding feature fusion layers during the construction of the 3D hose spliced ​​image.

[0021] In specific implementation, the simulation-driven analysis module includes: The simulation driving unit is used to perform ship loading and unloading simulation driving processing on the obtained three-dimensional hose visual information, obtain hose curvature change data of the corresponding oil hose during loading and unloading, and synchronously map the hose curvature change data with the corresponding three-dimensional hose visual information to obtain the corresponding driving parameter information, and set the corresponding three-dimensional hose driving visual model according to the driving parameter information. The fault integration unit is used to perform simulation-driven processing on different types of hazardous chemicals and loading and unloading data according to the three-dimensional hose driving vision model, obtain hose curvature change data corresponding to the corresponding oil hose, identify fault parameters on the hose curvature change data, and obtain the corresponding fault parameter dataset. It should be further explained that, in the specific implementation process, the corresponding fault parameter dataset is the hose curvature change data obtained by the corresponding three-dimensional hose driving vision model based on the simulation of the corresponding fault operation state during the ship loading and unloading process. Specifically, it includes fault types such as pipe collapse, pipe expansion, pipe running, pipe twisting, and multiple types of damage.

[0022] In practical implementation, the multi-state parallel architecture module includes: The target processing unit acquires loading and unloading hose operation information, obtains corresponding 3D hose visual information based on the loading and unloading hose operation information, performs target detection on the 3D hose visual information, obtains corresponding hose identification area information, performs bend recognition on the hose identification area information, obtains corresponding hose bend recognition area information, performs target segmentation based on the corresponding hose identification area information and hose bend recognition area information, and sets target nodes for the corresponding hose bend recognition area information and other areas respectively. The operation processing unit is used to analyze and process the operating status of multiple target nodes based on the hose bending identification area information, including: The statistical processing unit is used to analyze the running status of the corresponding identification area by the corresponding target node, obtain the view acquisition node corresponding to the corresponding identification area, set the time series axis according to the corresponding view acquisition node, and the target node maps the hose curvature change data of the corresponding identification area per unit time to the time series axis according to the corresponding view acquisition node. The unit performs change statistical analysis on the corresponding hose curvature change data according to the time series axis to obtain the sequence change data. The analysis and processing unit is used to extract features from the sequence change data corresponding to each time series axis within the corresponding target node, and obtain the state features corresponding to the corresponding time series axis respectively. It should be further explained that, in the specific implementation process, during the extraction of state features from the corresponding time series axes, a corresponding state feature monitoring model is constructed based on the target monitoring network. A historical sequence change dataset is set according to the corresponding historical sequence change data within the corresponding historical time series axes. The historical sequence change dataset is divided into a training set and a validation set. The training set is analyzed and processed based on the target monitoring network to construct the corresponding state feature monitoring model and obtain the loss function model of the corresponding state feature monitoring model. The loss function model is as follows: ; ; Where K is the number of samples, and y is the average of the squared differences between the predicted and actual values; until the corresponding loss function model tends to stabilize, the corresponding state feature monitoring model is output, and the obtained state feature monitoring model is validated and analyzed based on the validation set. The state feature monitoring model that has completed the validation analysis is analyzed and processed on the corresponding sequence change data to obtain the corresponding state features. The integration processing unit acquires the state features corresponding to the time series axes within the target node, integrates and statistically analyzes the frequency of occurrence of the corresponding state features, determines the type of the corresponding state features, sets the state confidence level according to the type of the corresponding state features, and integrates the state features of each time series axis within the corresponding target node according to the corresponding state confidence level to generate target state data. It should be further explained that, in the specific implementation process, the corresponding state credibility includes the frequency and duration of occurrence of the corresponding state feature type. The corresponding weight coefficient is set according to the corresponding state feature type, and the corresponding frequency and duration of occurrence are analyzed and processed according to the corresponding weight coefficient to obtain the corresponding state credibility.

[0023] In practical implementation, the hose early warning analysis module includes: The fault analysis unit is used to acquire the fault parameter dataset corresponding to the 3D hose visual image, analyze and train the fault parameter dataset based on machine learning algorithm, construct a fault analysis model, input the target state data obtained from the corresponding target node into the fault analysis model for analysis and processing, and obtain fault analysis data. The fault verification unit is used to acquire fault analysis data corresponding to each target node, sort the corresponding marked target nodes according to the 3D hose visual image, acquire the correlation data of the fault analysis data corresponding to each target node, perform verification processing according to the corresponding correlation data, and output the fault verification data corresponding to the target node in the 3D hose visual image. It should be further explained that, in the specific implementation process, the historical fault analysis data of the corresponding oil hoses are obtained by using features such as area intersection-union ratio and curvature to characterize the characteristics of hose expansion (overpressure), collapse, runaway, and damage, respectively. The degree of fault is characterized by specific numerical quantification, and a fault prediction model is established based on recurrent neural networks such as long short-term memory. Predictions are made by collecting conveyor belt status data in real time. At the same time, based on process knowledge and expert experience, a fault knowledge graph of the corresponding oil hoses is constructed. The fault knowledge graph contains cause-result-oriented correlation data corresponding to the historical fault analysis data of the corresponding oil hoses. The fault analysis data of the corresponding three-dimensional hose visual images and the corresponding marked target nodes are verified according to the corresponding fault knowledge graph to obtain the corresponding fault verification data. The fault warning unit is used to distinguish the corresponding fault verification data in the three-dimensional hose visual image, generate fault warning data based on the distinction results of the fault verification data, and perform warning processing on the identification area corresponding to the corresponding target node based on the generated fault warning data.

[0024] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A ship loading and unloading hose curvature intelligent monitoring system, characterized in that, include: The panoramic view acquisition module is used to set up corresponding optical imaging devices based on the information of the ship loading and unloading area in the storage tank and the dock, set up view acquisition nodes based on the optical imaging devices, and acquire visual acquisition information at the corresponding locations through the view acquisition nodes. The panoramic image fusion module is used to perform initial fusion processing on the visual acquisition information collected by each view acquisition node, obtain fusion coordinate points, and perform multi-dimensional non-overlapping stitching and fusion processing on the corresponding visual acquisition information based on the corresponding fusion coordinate points to obtain three-dimensional hose visual information. The simulation-driven analysis module is used to perform simulation-driven processing on the corresponding positions based on the obtained 3D hose visual information, obtain the driving parameter information of the corresponding oil hose, identify fault parameters from the obtained driving parameter information, and generate a fault parameter dataset. The multi-state parallel architecture module is used to perform target detection and target segmentation processing based on the obtained 3D hose visual information, obtain multiple target nodes according to the target processing results, perform running status analysis based on each target node, and mark them at the corresponding positions in the 3D hose visual image. The hose early warning analysis module is used to set up a fault analysis model for the corresponding fault parameter dataset, input the operation status analysis results of the corresponding target nodes in the 3D hose visual image into the fault analysis model for analysis and processing, and obtain the fault early warning information of the corresponding target nodes. The panoramic view acquisition module includes: The visual layout unit is used to acquire information on the corresponding ship loading and unloading areas within the storage tank and the dock, acquire oil hose operation information based on the ship loading and unloading area information, arrange the position of the optical imaging device based on the oil hose operation information, and set the view acquisition nodes based on the position layout results. The visual acquisition unit is used to acquire and process data based on the optical imaging device corresponding to the view acquisition node. The view acquisition node is used to dynamically adjust the acquisition parameters of the corresponding optical imaging device according to the corresponding oil hose operation information and acquire the corresponding visual acquisition information. The panoramic image fusion module includes: The feature recognition unit is used to acquire visual acquisition information obtained by each view acquisition node, perform target feature recognition on the visual acquisition information obtained by each view acquisition node, and perform labeling processing based on the target feature recognition results; The initial fusion unit is used to acquire the target feature recognition results of each marker processing, perform matching processing on the target feature recognition results, perform initial fusion processing based on the matching processing results, and generate an initial fused image. The multidimensional fusion unit is used to set fusion coordinate points for the initial fusion image corresponding to the corresponding matching processing result, and to perform multidimensional non-overlapping stitching and fusion processing on each initial fusion image according to the fusion coordinate points. The multidimensional fusion unit includes: The coordinate determination unit acquires the corresponding initial fused image, sets the corresponding feature fusion layer according to the target feature recognition results included in the initial fused image, acquires the fusion coordinate points and key coordinate points corresponding to the feature fusion layer in the corresponding initial fused image, marks each fusion coordinate point and key coordinate point, and sets them as the fusion coordinate points of the initial fused image to which the corresponding feature fusion layer belongs. The stitching processing unit acquires the corresponding fusion coordinate points within the corresponding feature fusion layer, sequentially selects two or more fusion coordinate points within the corresponding feature fusion layer, performs image matching based on the selected fusion coordinate points, sets a three-dimensional spatial coordinate system, maps the position information of the view acquisition nodes corresponding to the corresponding feature fusion layer and the corresponding fusion coordinate points to the three-dimensional spatial coordinate system, acquires the corresponding fusion pixels, acquires the corresponding fusion pixels of each initial fusion image based on the mapping results of the corresponding fusion coordinate points and key coordinate points within each feature fusion layer, and performs fusion processing on the fusion pixels in the three-dimensional spatial coordinate system based on the image fusion algorithm to acquire a three-dimensional hose stitched image; The deduplication processing unit is used to monitor the overlapping area of ​​the 3D hose spliced ​​image obtained by the splicing processing unit, obtain the overlapping area of ​​pixels in the 3D hose spliced ​​image, and perform deduplication processing on the overlapping area of ​​pixels by adaptively adjusting the fusion parameters based on the regional features to obtain the 3D hose visual information. The simulation-driven analysis module includes: The simulation driving unit is used to perform ship loading and unloading simulation driving processing on the obtained three-dimensional hose visual information, obtain hose curvature change data of the corresponding oil hose during loading and unloading, and synchronously map the hose curvature change data with the corresponding three-dimensional hose visual information to obtain the corresponding driving parameter information, and set the corresponding three-dimensional hose driving visual model according to the driving parameter information. The fault integration unit is used to perform simulation-driven processing on different types of hazardous chemicals and loading and unloading data according to the three-dimensional hose driving vision model, obtain the hose curvature change data corresponding to the oil hose, identify fault parameters on the hose curvature change data, and obtain the corresponding fault parameter dataset. The multi-state parallel architecture module includes: The target processing unit acquires oil hose operation information, obtains corresponding 3D hose visual information based on the oil hose operation information, performs target detection on the 3D hose visual information, obtains corresponding hose identification area information, performs bend recognition on the hose identification area information, obtains corresponding hose bend recognition area information, performs target segmentation based on the corresponding hose identification area information and hose bend recognition area information, and sets target nodes for the corresponding hose bend recognition area information and other areas respectively. The operation processing unit is used to analyze and process the operating status of multiple target nodes based on the hose bending identification area information. The operation processing unit includes: The statistical processing unit is used to analyze the running status of the corresponding identification area by the corresponding target node, obtain the view acquisition node corresponding to the corresponding identification area, set the time series axis according to the corresponding view acquisition node, and the target node maps the hose curvature change data of the corresponding identification area per unit time to the time series axis according to the corresponding view acquisition node. The unit performs change statistical analysis on the corresponding hose curvature change data according to the time series axis to obtain the sequence change data. The analysis and processing unit is used to extract features from the sequence change data corresponding to each time series axis within the corresponding target node, and obtain the state features corresponding to the corresponding time series axis respectively. The integration processing unit acquires the state features corresponding to the time series axes within the target node, integrates and statistically analyzes the frequency of occurrence of the corresponding state features, determines the type of the corresponding state features, sets the state confidence level according to the type of the corresponding state features, and integrates the state features of each time series axis within the corresponding target node according to the corresponding state confidence level to generate target state data. The hose early warning analysis module includes: The fault analysis unit is used to acquire the fault parameter dataset corresponding to the 3D hose visual image, analyze and train the fault parameter dataset based on machine learning algorithm, construct a fault analysis model, input the target state data obtained from the corresponding target node into the fault analysis model for analysis and processing, and obtain fault analysis data. The fault verification unit is used to acquire fault analysis data corresponding to each target node, sort the corresponding marked target nodes according to the 3D hose visual image, acquire the correlation data of the fault analysis data corresponding to each target node, perform verification processing according to the corresponding correlation data, and output the fault verification data corresponding to the target node in the 3D hose visual image. The fault warning unit is used to distinguish the corresponding fault verification data in the three-dimensional hose visual image, generate fault warning data based on the distinction results of the fault verification data, and perform warning processing on the identification area corresponding to the corresponding target node based on the generated fault warning data.

Images