An intelligent diagnosis system and method for potential safety hazards in production

Abstract

The present invention discloses an intelligent diagnosis system and method for potential safety hazards in production, belonging to the technical field of safety production management. Specifically, it includes: collecting on-site images through sensors and performing preprocessing, inputting the images into a multi-object recognition model to identify each target element in the images; using a preset production scenario knowledge graph to determine the production scenario containing all target elements, calling the corresponding specification documents, and outputting specification requirements through a large model; according to the specification requirements, dividing the images into large-scale information regions and small-scale information regions, and accordingly splitting the on-site images into large-scale sub-images and small-scale sub-images, and inputting these sub-images into the first deep learning model and the second deep learning model corresponding to the production scenario for hazard identification respectively; combining the output results of the first deep learning model and the second deep learning model to identify and record safety hazards; the present invention improves the accuracy and efficiency of safety hazard identification in the production site.

Description

Technical Field

[0001] The present invention belongs to the technical field of safety production management, and more particularly to an intelligent diagnosis system and method for potential safety hazards in production safety. Background Art

[0002] In the field of production safety, timely and accurate identification of potential hazards is crucial for preventing accidents. Traditional production safety management relies on manual inspections, which have problems such as low efficiency, easy omission of inspections, and slow response. With the development of image recognition and artificial intelligence technologies, it has become possible to use image recognition technology for intelligent identification of potential safety hazards in production safety. However, existing technologies are often limited to single-target recognition or simple scenario analysis and are difficult to comprehensively and accurately identify safety hazards in complex production environments.

[0003] For example, the Chinese patent application with the publication number CN109345148A discloses an intelligent diagnosis method for identifying potential safety hazards in natural gas pipelines and station process devices, including: data collection: collecting data on natural gas pipelines and station devices and establishing a ledger; establishing a database: inputting the data into a server, optimizing the selection, establishing a database, and performing data statistics and cloud computing analysis; identifying potential hazards: using phased array detection technology and injection-production string vibration fatigue life prediction technology to conduct a comprehensive inspection and detection of the integrity of key components of equipment and devices, pipeline networks, pipe fittings, and valve bodies for intrinsic safety issues, online intelligent remote monitoring technology, and on-site investigation to collect potential hazard data and perform test operation and processing. This technical solution understands the risk sources and digital construction status of each gas transmission pipeline through preliminary work, selects technologies, formulates implementation plans, and conducts project risk analysis to ensure the safe operation of natural gas production pipelines.

[0004] The above existing technologies all have the following problems: weak generalization ability and adaptability; poor real-time performance and efficiency; strong technical complexity and professionalism; optimized for specific application scenarios of natural gas pipelines and station process devices, but may be difficult to adapt to other types of potential safety hazard identification tasks. Summary of the Invention

[0005] In view of the deficiencies of the prior art, the present invention proposes an intelligent diagnosis system and method for potential safety hazards in production. By collecting on-site images through sensors and performing preprocessing, the images are input into a multi-object recognition model to identify each target element in the images. Using a preset production scenario knowledge graph, a production scenario containing all target elements is determined, and the corresponding specification file is called, and the specification requirements are output through a large model. According to the specification requirements, the image is divided into large-scale information regions and small-scale information regions, and accordingly, the on-site image is split into large-scale sub-images and small-scale sub-images, and these sub-images are respectively input into the first deep learning model and the second deep learning model corresponding to the production scenario for potential hazard identification. By combining the output results of the first deep learning model and the second deep learning model, potential safety hazards are identified and recorded. The present invention improves the accuracy and efficiency of identifying potential safety hazards in the production site.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent diagnosis method for potential safety hazards in production, comprising:

[0008] Obtain an on-site image of a sensor, the on-site image containing multiple target elements;

[0009] According to the target elements and a preset production scenario knowledge graph, determine a production scenario including all target elements;

[0010] According to the production scenario and at least one specification file corresponding to the production scenario, divide the on-site image into large-scale information regions and small-scale information regions;

[0011] According to the large-scale information regions and the small-scale information regions, perform potential safety hazard identification on the on-site image;

[0012] According to the divided large-scale information regions and small-scale information regions, split the preprocessed on-site image into at least one large-scale sub-image and at least one small-scale sub-image;

[0013] Input the large-scale sub-images and the large-scale information into the first deep learning model corresponding to the production scenario, and input the small-scale sub-images into the second deep learning model corresponding to the production scenario;

[0014] The first deep learning model and the second deep learning model include:

[0015] The first deep learning model is established based on edge recognition. The first deep learning model first performs edge recognition on the large-scale sub-image to obtain boundary contour information, and determines whether it meets the large-scale information according to the similarity of comparing the boundary contour information with the large-scale information;

[0016] The second deep learning model is used for material recognition to achieve the precise recognition and analysis of fine features in small-scale sub-images;

[0017] When performing boundary contour information recognition on the large-scale sub-image, it is to continue to strip out small-scale sub-images from the already segmented large-scale sub-images, which belongs to a dynamic verification process. The specific steps of the dynamic verification include:

[0018] Obtain the large-scale sub-image, calculate the texture complexity and gray-scale change rate indexes for each local area of the large-scale sub-image. When the texture complexity or gray-scale change rate index of any local area exceeds the preset threshold, it is considered that there are detailed hidden dangers in this local area that require further fine analysis;

[0019] According to the determined local areas with detailed hidden dangers, strip out the corresponding small-scale sub-images from the large-scale sub-image;

[0020] Establish an image area selection rule associated with the detailed hidden danger characteristics according to the detailed hidden danger characteristics in different production scenarios.

[0021] Specifically, the process of determining the production scenario including all target elements includes:

[0022] Input the on-site image into the multi-target recognition model, and the multi-target recognition model outputs each target element in the image;

[0023] Input all the recognized target elements into the preset production scenario knowledge graph, and determine the production scenario including all target elements according to the association relationships in the production scenario knowledge graph.

[0024] Specifically, the process of dividing the large-scale information area and the small-scale information area includes:

[0025] According to the production scenario, call at least one specification file corresponding to the production scenario, and call the large model to perform semantic output on all specification files to obtain multiple specification requirements;

[0026] Based on the specification requirements, divide the image into a large-scale information area and a small-scale information area.

[0027] Specifically, the specific steps of the dynamic verification further include:

[0028] After stripping out the small-scale sub-image from the large-scale sub-image, establish a real-time feedback mechanism between the first deep learning model and the second deep learning model;

[0029] Use the first deep learning model to perform preliminary analysis on the stripped area, and transfer the position, size, and preliminary analysis results of the stripped area to the second deep learning model;

[0030] After receiving the information transmitted by the first deep learning model, the second deep learning model conducts a detailed analysis of the small-scale sub-images, detects and identifies specific detailed potential hazard information, where the detailed potential hazard information includes the specific location, form, and the associated situation with the large-scale information of the potential hazard.

[0031] Specifically, the specific steps of the dynamic verification further include:

[0032] Feed back the detected detailed potential hazard information to the first deep learning model;

[0033] During multiple image stripping and model analysis processes, adjust the verification criteria and model parameters according to the comparison between each verification result and the actual potential hazard situation;

[0034] Output the finally analyzed potential hazard information, location, and severity to the staff, and record the process and results of each analysis.

[0035] Specifically, the specific process of stripping the corresponding small-scale sub-images from the large-scale sub-images includes:

[0036] Determine the local area with detailed potential hazards in the large-scale sub-image, where the local area with detailed potential hazards is a bounding box composed of multiple pixel coordinates;

[0037] According to the determined local area of detailed potential hazards, calculate the parameters required to strip the small-scale sub-images, including the starting coordinates, width, and height of the small-scale sub-images;

[0038] Use the calculated stripping parameters to strip the corresponding small-scale sub-images from the large-scale sub-images through image cropping operations;

[0039] Check whether the stripped small-scale sub-images contain the complete potential hazard area and do not contain unnecessary additional areas.

[0040] Specifically, when combining the output results of the first deep learning model and the second deep learning model, a weighted average method is adopted, and different weights are assigned to them according to the accuracies of the first deep learning model and the second deep learning model and .

[0041] An intelligent diagnosis system for potential hazards in safe production includes: an image processing module, a production scenario determination module, a sub-image generation module, a deep learning model analysis module, and a potential hazard identification and recording module;

[0042] The image processing module collects on-site images through sensors and conducts preprocessing;

[0043] The production scenario determination module is used to identify each target element from the pre - processed image by using a multi - target recognition model, and determine the current production scenario according to the identified target elements in combination with a preset production scenario knowledge graph;

[0044] The sub - image generation module is used to divide the image into regions according to the determined specification requirements, and split the pre - processed on - site image into large - scale sub - images and small - scale sub - images according to the divided large - scale information regions and small - scale information regions;

[0045] The deep - learning model analysis module is used to input the large - scale sub - images and large - scale information into the first deep - learning model corresponding to the production scenario, and input the small - scale sub - images into the second deep - learning model corresponding to the production scenario for a preliminary analysis of potential safety hazards;

[0046] The hazard identification and recording module is used to identify potential safety hazards at the production site by combining the output results of the two deep - learning models and record the results of each hazard identification.

[0047] Specifically, the sub - image generation module includes: a specification file calling unit, a specification semantics output unit, a region division unit, and a sub - image cropping unit;

[0048] The specification file calling unit is used to call the corresponding specification file according to the determined production scenario;

[0049] The specification semantics output unit is used to perform semantic analysis on the specification file through a large - model and output multiple specification requirements;

[0050] The region division unit is used to divide the image into large - scale information regions and small - scale information regions based on the specification requirements;

[0051] The sub - image cropping unit is used to split the pre - processed on - site image into large - scale sub - images and small - scale sub - images according to the divided large - scale information regions and small - scale information regions.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] 1. The present invention proposes an intelligent diagnosis system for potential safety hazards in production, and has optimized improvements in terms of architecture, operation steps and processes. The system has the advantages of simple process, low investment and operation costs, and low production cost.

[0054] 2. The present invention proposes an intelligent diagnosis method for potential safety hazards in production. By integrating multi-object recognition, knowledge graph application, and deep learning model analysis, it realizes the intelligence and automation of safety hazard recognition in the production site, improves the recognition efficiency and accuracy, reduces manual intervention, and enhances the level of production safety management. At the same time, this method demonstrates high efficiency and flexibility in detail processing. By dividing the large-scale information area and the small-scale information area, and applying targeted deep learning models for analysis respectively, it not only improves the accuracy of hazard recognition, but also can effectively cope with complex and changeable production scenarios, ensuring that safety hazards are identified and recorded without omission. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Schematic diagram of an intelligent diagnosis method for potential safety hazards in production according to the present invention;

[0056] Figure 2 Principle flowchart of an intelligent diagnosis method for potential safety hazards in production according to the present invention;

[0057] Figure 3 Schematic diagram of the lamp room of an intelligent diagnosis method for potential safety hazards in production according to the present invention;

[0058] Figure 4 Schematic diagram of the main step-down control room of an intelligent diagnosis method for potential safety hazards in production according to the present invention;

[0059] Figure 5 Schematic diagram of the ventilation window grille of the 10KV capacitor room of an intelligent diagnosis method for potential safety hazards in production according to the present invention;

[0060] Figure 6 System architecture diagram of an intelligent diagnosis method for potential safety hazards in production according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] Embodiment 1

[0062] Please refer to Figures 1 to 5 , an embodiment provided by the present invention: an intelligent diagnosis method for potential safety hazards in production, the method includes steps S101 to S104, where:

[0063] S101: Obtain the on-site image of the sensor, the on-site image includes multiple target elements;

[0064] Further, the specific steps of S101 include:

[0065] (1) Obtain the on-site image collected by the sensor (such as a CMOS image sensor). The pixel array on the CMOS sensor consists of many micro-photodiodes. These photodiodes generate charges when irradiated by light, complete the photoelectric conversion process, and the generated charges accumulate in the storage area of each pixel for integration;

[0066] After integration is completed, the charge signals of each pixel are read through the row and column scan registers. Among them, the read analog charge signals are amplified one by one and then converted into digital signals through an analog-to-digital converter.

[0067] The digital signals are then sent to the digital signal processor at the back end for processing, including noise removal, white balance adjustment, and color interpolation.

[0068] Images with different resolutions or different spectra are fused to improve the quality and information content of the images, and a preprocessed on-site image is obtained.

[0069] S102: Determine the production scenario including all target elements based on the target elements and the preset production scenario knowledge graph.

[0070] The process of determining the production scenario including all target elements includes:

[0071] A1: Input the on-site image into a multi-target recognition model, and the multi-target recognition model outputs each target element in the image. Among them, the target elements include but are not limited to equipment, personnel, and materials.

[0072] A2: Input all the identified target elements into the preset production scenario knowledge graph, and determine the production scenario including all target elements according to the association relationships in the production scenario knowledge graph.

[0073] Among them, the multi-target recognition model refers to a deep learning model that can recognize multiple target objects in an image, which needs to be pre-loaded, and the deep learning model is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here.

[0074] Further, the specific steps of A2 include:

[0075] (1) Extract all the target elements identified in the image from the output of the multi-target recognition model. The specific target elements depend on the production scenario and the design of the recognition model.

[0076] (2) Match the extracted target elements with the preset production scenario knowledge graph. Among them, the knowledge graph is a graph-structured database containing entities, attributes, and relationships. The entities represent the objects in the production scenario, and the relationships represent the interactions or connections between these objects.

[0077] Among them, the specific process of matching the extracted target elements with the preset production scenario knowledge graph includes:

[0078] a. Ensure that the preset production scenario knowledge graph is complete and accurate, containing all relevant entities, attributes, relationships, and their hierarchical and associative information. The knowledge graph usually represents knowledge in the form of triples: entity - relationship - entity;

[0079] b. Use natural language processing methods to extract target elements from the data source to be matched. These elements may be instances of entities, attributes, or relationships. Among them, natural language processing methods are the existing technical content in this field and not the creative solution of this application, so they will not be elaborated here;

[0080] c. Select a graph - based matching algorithm according to the characteristics of the target elements and the knowledge graph;

[0081] d. Compare the extracted target elements with the entities, attributes, or relationships in the knowledge graph one by one, and calculate the similarity between the target elements and each element in the knowledge graph according to the selected graph - based matching algorithm. Among them, the similarity calculation formula is the existing technical content in this field and not the creative solution of this application, so it will not be elaborated here;

[0082] e. According to the preset threshold, select the element with the highest matching degree as the matching result, and verify the matching result to ensure the accuracy and reliability of the matching result;

[0083] f. Adjust and optimize the matching algorithm according to the verification result to improve the matching performance and accuracy.

[0084] (3) In the knowledge graph, according to the associative relationships between target elements, such as spatial position, functional relationship, process flow, etc., analyze and determine the production scenario to which they belong. This involves traversing and searching the nodes and edges in the knowledge graph to find the production scenario associated with the target elements;

[0085] (4) According to the result of the associative relationship analysis, determine the production scenario that contains all target elements. This step involves evaluating and comparing multiple candidate scenarios to select the scenario that best matches the associative relationship of the target elements.

[0086] S103: Divide the on - site image into large - scale information regions and small - scale information regions according to the production scenario and at least one specification file corresponding to the production scenario;

[0087] It should be noted that in a production scenario, the requirements for decision-making and monitoring of different information are different. Large-scale information may focus more on the overall structure and macro environment, while small-scale information may focus more on detailed features and minor changes. By dividing an image into information regions of different scales, key information can be captured and analyzed more accurately, improving the accuracy and precision of the analysis. For information regions of different scales, different processing methods and algorithms can be adopted to optimize the entire processing flow and improve efficiency and effectiveness. On the other hand, by dividing information regions of different scales, changes and potential risks can be captured more sensitively.

[0088] The process of dividing the large-scale information region and the small-scale information region includes:

[0089] B1: According to the production scenario, call at least one specification file corresponding to the production scenario, and call a large model to perform a canonical semantic output on all specification files to obtain multiple specification requirements;

[0090] Among them, the specification file refers to an enterprise specification or an operation manual; the large model adopts a natural language processing model and uses the output function of the natural language processing model to convert the text content in the specification file into structured specification requirements, and the specification requirements include operation steps, quality standards, and safety specifications.

[0091] Exemplarily, let the production scenario be S, the specification file library be F, and the large model be M;

[0092] For a given production scenario S, find the corresponding set of specification files in the specification file library F , where, represents the nth specification file;

[0093] For each specification file , call the large model M for semantic analysis to obtain a set of specification requirements , where, represents the mth specification requirement of the ith specification file;

[0094] Merge the sets of specification requirements of all specification files to obtain the final set of specification requirements , where, represents the Nth specification requirement in the final set of specification requirements;

[0095] Sort out and verify the final set of specification requirements to obtain the final available specification requirements.

[0096] B2: Based on the specification requirements, divide the image into a large-scale information region and a small-scale information region.

[0097] S104: Identify potential safety hazards in on-site images based on the large-scale information region and the small-scale information region.

[0098] The process of identifying potential safety hazards in on-site images based on the large-scale information region and the small-scale information region includes:

[0099] C1: Split the preprocessed on-site image into at least one large-scale sub-image and at least one small-scale sub-image according to the divided large-scale information region and small-scale information region; the large-scale sub-image can overlap with the small-scale sub-image region, that is, the relationship between the large-scale sub-image and the small-scale sub-image region includes: intersection, inclusion, and irrelevance;

[0100] In the present invention, splitting the on-site image into large-scale sub-images and small-scale sub-images is to better analyze and utilize the multi-scale information in the image. Moreover, since the information in the image often does not exist in isolation, but is interrelated and interacts with each other, through the overlapping region, the information connection between different scales in the image can be better captured and retained. Among them, the large-scale sub-image usually contains the overall structure and trend information, while the small-scale sub-image contains more details and texture information.

[0101] C2: Input the large-scale sub-image and the large-scale information into the first deep learning model corresponding to the production scenario, and input the small-scale sub-image into the second deep learning model corresponding to the production scenario;

[0102] Furthermore, the specific steps of C2 include:

[0103] (1) Organize the preprocessed large-scale sub-image and the large-scale information into a data format suitable for input into the deep learning model. Similarly, organize the small-scale sub-image into a data format suitable for input into the second deep learning model;

[0104] (2) Load the pre-constructed first deep learning model and second deep learning model according to the requirements of the production scenario and the image features. Among them, the first deep learning model is used to process the large-scale sub-image and the large-scale information, and the second deep learning model is used to process the small-scale sub-image;

[0105] Among them, the first deep learning model and the second deep learning model are constructed based on the convolutional neural network model architecture, and the convolutional neural network model is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here.

[0106] (3) Use the training dataset to train the first deep learning model and the second deep learning model respectively, so that they can accurately identify and process large-scale sub-images, large-scale information, and small-scale sub-images. Among them, the training set needs to collect historical large-scale sub-images, large-scale information, and small-scale sub-images in advance to obtain training set data;

[0107] (4) Deploy the trained and qualified first deep learning model and second deep learning model to the production environment for processing large-scale sub-images, large-scale information, and small-scale sub-images;

[0108] (5) In the production environment, input the newly obtained large-scale sub-images and large-scale information into the first deep learning model for prediction, and input the newly obtained small-scale sub-images into the second deep learning model for prediction.

[0109] In the present invention, because the information in the image often has multi-scale characteristics, that is, information at different scales plays different roles in image understanding and analysis. By processing large-scale and small-scale information separately, the complementarity of these information can be fully utilized to improve the comprehensiveness and accuracy of processing. Therefore, the practice of inputting large-scale sub-images and large-scale information into the first deep learning model corresponding to the production scenario and inputting small-scale sub-images into the second deep learning model corresponding to the production scenario is based on considerations such as the complementarity of multi-scale information, model adaptability, and reasonable utilization of computing resources, etc., which can more effectively capture the details and global features in the image, thereby improving the processing accuracy and precision.

[0110] C3: Combine the output results of the first deep learning model and the second deep learning model to identify potential safety hazards at the production site and record the results of each hazard identification.

[0111] When combining the output results of the first deep learning model and the second deep learning model, a weighted average method is adopted, and different weights are assigned to them according to the accuracy rates of the first deep learning model and the second deep learning model and .

[0112] The first deep learning model and the second deep learning model include:

[0113] The first deep learning model is established based on edge recognition. The first deep learning model first performs edge recognition on large-scale sub-images to obtain boundary contour information, and determines whether it meets the large-scale information according to the similarity between the comparison of the boundary contour information and the large-scale information. Among them, edge recognition uses an edge detection algorithm, and the edge detection algorithm is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here;

[0114] In the present invention, large-scale information includes the identified chemical dosing room, main step-down and main control room, and window grille information, while small-scale information includes chemical dosing room lamp wires, curtains, and window grille corrosion area information.

[0115] The second deep learning model is used for material identification to achieve the precise identification and analysis of fine features in small-scale sub-images.

[0116] Exemplarily, as Figure 3 shown, in Figure 3 after the identification by the first deep learning model, the large-scale sub-image is determined to be the chemical dosing room. Then, after the identification of the large-scale sub-image, detailed identification is carried out to obtain small-scale sub-images, including but not limited to light bulbs, wires, and load-bearing walls, as Figure 3 in which A in

[0117] is the wiring diagram of the chemical dosing room lamp wires and is a small-scale sub-image. When identifying the connection details of the chemical dosing room lamp wires, it can focus on features that can only be clearly presented at a small scale, such as the connection points between the wires and the lamps, the tiny damage to the wire insulation layer, and the screw tightening degree. Through the learning of a large number of such small-scale hidden danger samples, the model can judge whether these details meet the small-scale information requirements in the safety specifications, such as whether the wire connections are tight and whether the insulation layer is intact. Figure 4 Figure 4 shown, in after the identification by the first deep learning model, the large-scale sub-image is determined to be the main step-down and main control room. Then, small-scale analysis is carried out in the main step-down and main control room. As

[0118] in which B in Figure 5 is the curtain. For details related to the combustion performance of the curtain, the second deep learning model can analyze its microscopic features such as fiber structure and weaving method, and compare them with the small-scale information in the fire prevention specifications, such as the combustion characteristic indexes of specific fiber materials, to determine whether there are fire hazards. Figure 5 shown, in

[0119] When identifying the boundary contour information of the large-scale sub-image, it is to continue to peel off the small-scale sub-images from the already divided large-scale sub-images, which belongs to a dynamic verification process. The specific steps of the dynamic verification include:

[0120] Obtain a large-scale sub-image. For each local area of the large-scale sub-image, calculate its texture complexity and gray-scale change rate indicators. When the texture complexity or gray-scale change rate indicator of any local area exceeds a preset threshold, it is considered that there are detailed potential hazards in this local area that require further refined analysis;

[0121] It should be noted that the calculation process of the texture complexity and gray-scale change rate indicators includes:

[0122] (1) Intercept or extract the large-scale sub-image from the preprocessed image;

[0123] (2) Divide the large-scale sub-image into multiple overlapping or non-overlapping local areas, and the size and step of the division can be adjusted according to actual needs;

[0124] (3) For each local area, calculate its texture complexity using the gray-level co-occurrence matrix method;

[0125] Among them, the gray-level co-occurrence matrix is a matrix obtained by calculating the frequency of pixel pairs appearing at a specific distance and direction in the image. Then, multiple texture features, such as contrast, energy, homogeneity, and correlation, can be extracted from this matrix, and these features can be comprehensively used to measure the texture complexity of the image. Further, when extracting multiple texture features from this matrix, the ready-made image processing library scikit-image is used to calculate these features, and scikit-image is the prior art content in this field and not the creative solution of this application, so it will not be elaborated here.

[0126] (4) For each local area, calculate its gray-scale change rate , to measure the speed of gray-scale change, where represents the gradient of the image gray value in the horizontal direction, represents the gradient of the image gray value in the vertical direction.

[0127] When making a judgment, it should be understood that by comparing the texture complexity and gray-scale change rate of each local area with the preset threshold, if any indicator exceeds the threshold, it is considered that there are detailed potential hazards in this local area that require further refined analysis.

[0128] Exemplarily, when analyzing the large-scale sub-image of a 10kV capacitor room, if the gray-scale changes violently and the texture is complex in the area near the ventilation window, it may imply that there are situations such as corrosion or foreign object attachment on the window grille. At this time, the system automatically peels off a small-scale sub-image in this area for more detailed detection by the second deep learning model.

[0129] According to the judged local areas with detailed potential hazards, peel off the corresponding small-scale sub-images from the large-scale sub-image;

[0130] According to the detailed hidden danger characteristics in different production scenarios, establish image region selection rules associated with the detailed hidden danger characteristics;

[0131] Furthermore, the process of establishing image region selection rules associated with the detailed hidden danger characteristics includes:

[0132] (1) Analyze each production scenario to understand the production process, equipment layout, and operation specifications;

[0133] (2) Identify and record the detailed hidden danger characteristics that may occur in each scenario, such as equipment aging, abnormal sounds, vibrations, leaks, and improper operations;

[0134] (3) According to the identified detailed hidden danger characteristics, formulate corresponding image region selection criteria, which may include characteristics such as region size, shape, position, color, and texture;

[0135] (4) Combine the specific situation of the production scenario and the image region selection criteria to establish detailed image region selection rules. Among them, the rules should clarify which regions need to be selected for further analysis, as well as the specific conditions and methods for selecting these regions;

[0136] Among them, the image region selection rules refer to a set of rules or criteria that determine which image regions need to be selected or concerned during the process of image processing and analysis. These rules are usually based on certain features or attributes in the image, such as color, texture, shape, size, etc., and the association between these features and specific goals or requirements.

[0137] (5) Apply the image region selection rules in the actual production environment and collect feedback data;

[0138] (6) Adjust and optimize the rules according to the feedback data to ensure their accuracy and effectiveness.

[0139] Exemplarily, in the analysis of large-scale sub-images in the chemical dosing room, if abnormal shadows or blurred areas are detected around the lamps and are associated with electrical fault hidden danger characteristics, the system will preferentially select this region for small-scale sub-image stripping and focus on inspecting details such as lamp wiring and bulb status. This active image stripping based on hidden danger characteristics can improve the pertinence and efficiency of detection.

[0140] When small-scale sub-images are stripped from the large-scale sub-images, establish a real-time feedback mechanism between the first deep learning model and the second deep learning model;

[0141] Use the first deep learning model to conduct a preliminary analysis of the stripped region and transfer the position, size, and preliminary analysis results of the stripped region to the second deep learning model;

[0142] After receiving the information transmitted by the first deep learning model, the second deep learning model conducts a detailed analysis of the small-scale sub-image, detects and identifies specific detailed potential hazard information, where the detailed potential hazard information includes the specific location, form, and association with large-scale information of the potential hazard;

[0143] Feed back the detected detailed potential hazard information to the first deep learning model;

[0144] Exemplarily, in the analysis of the large-scale sub-image of the curtain in the main step-down control room, if the first deep learning model finds that the overall area of the curtain is large but the color or texture of some local areas is abnormal, after stripping the small-scale sub-image, the second deep learning model detects that there are details such as loose fibers in this area that may affect the combustion performance, and feeds this information back to the first deep learning model, enabling the first deep learning model to comprehensively judge the overall safety status of the curtain and realizing the collaborative work and dynamic verification between the models.

[0145] During multiple image stripping and model analysis processes, adjust the verification standard and model parameters according to the comparison between each verification result and the actual potential hazard situation;

[0146] Exemplarily, after processing a series of images of electrical equipment rooms, if it is found that some types of potential hazards are omitted or misjudged under specific conditions, the system automatically adjusts relevant parameters to improve the accuracy of subsequent detections, enabling the entire dynamic verification process to continuously adapt to different production scenarios and potential hazard changes.

[0147] Output the finally analyzed potential hazard information, location, and severity to the staff, and record the process and results of each analysis.

[0148] The specific process of stripping the corresponding small-scale sub-image from the large-scale sub-image includes:

[0149] Determine the local area with detailed potential hazards in the large-scale sub-image, where the local area with detailed potential hazards is a bounding box composed of multiple pixel coordinates;

[0150] According to the determined local area of detailed potential hazards, calculate the parameters required to strip the small-scale sub-image, including the starting coordinates, width, and height of the small-scale sub-image;

[0151] Among them, when calculating the parameters required to strip the small-scale sub-image, functions provided by the image processing library OpenCV are used for calculation. OpenCV is the prior art content in this field and is not the creative solution of this application, so it will not be elaborated here.

[0152] Use the calculated stripping parameters to strip the corresponding small-scale sub-image from the large-scale sub-image through image cropping operations;

[0153] Check whether the extracted small-scale sub-images contain the complete hidden danger area and do not contain unnecessary additional areas.

[0154] Embodiment 2

[0155] Please refer to Figure 6 , another embodiment provided by the present invention: an intelligent diagnosis system for potential safety hazards in production, including:

[0156] An image processing module, a production scenario determination module, a sub-image generation module, a deep learning model analysis module, and a potential hazard identification and recording module;

[0157] The image processing module collects on-site images through sensors and performs preprocessing, including denoising, enhancement, and scaling, to ensure that the image quality meets the requirements of subsequent processing;

[0158] The production scenario determination module is used to identify each target element from the preprocessed image by using a multi-target recognition model, and determine the current production scenario according to the identified target elements in combination with the preset production scenario knowledge graph;

[0159] The sub-image generation module is used to divide the image into regions according to the determined specification requirements, and split the preprocessed on-site image into large-scale sub-images and small-scale sub-images according to the divided large-scale information regions and small-scale information regions;

[0160] The deep learning model analysis module is used to input the large-scale sub-images and large-scale information into the first deep learning model corresponding to the production scenario, and input the small-scale sub-images into the second deep learning model corresponding to the production scenario for preliminary analysis of potential safety hazards;

[0161] The potential hazard identification and recording module is used to combine the output results of the two deep learning models to identify the potential safety hazards at the production site and record the results of each potential hazard identification.

[0162] The image processing module includes: an image acquisition unit and a preprocessing unit;

[0163] The image acquisition unit is used to collect on-site images through sensors;

[0164] The preprocessing unit is used to perform preprocessing operations such as denoising, enhancement, and scaling on the collected on-site images.

[0165] The production scenario determination module includes: a target recognition unit and a knowledge graph matching unit;

[0166] The target recognition unit is used to identify each target element from the preprocessed image by using a multi-target recognition model;

[0167] A knowledge graph matching unit, configured to determine the current production scenario according to the identified target elements in combination with a preset production scenario knowledge graph.

[0168] The sub-image generation module includes: a specification file calling unit, a specification semantics output unit, a region division unit, and a sub-image cropping unit;

[0169] The specification file calling unit is configured to call the corresponding specification file according to the determined production scenario;

[0170] The specification semantics output unit is configured to perform semantic analysis on the specification file through a large model and output multiple specification requirements;

[0171] The region division unit is configured to divide the image into a large-scale information region and a small-scale information region based on the specification requirements;

[0172] The sub-image cropping unit is configured to split the preprocessed on-site image into a large-scale sub-image and a small-scale sub-image according to the divided large-scale information region and small-scale information region.

[0173] The potential hazard identification and recording module includes: a potential hazard identification unit and a result recording unit;

[0174] The potential hazard identification unit is configured to integrate the output results of two deep learning models to identify potential safety hazards in the production site;

[0175] The result recording unit is configured to record the results of each potential hazard identification for subsequent analysis and processing.

[0176] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments without departing from the purpose and scope of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. An intelligent diagnosis method for potential safety hazards in production, characterized in that, Including: Obtain the on-site image of the sensor, where the on-site image contains multiple target elements; Determine the production scenario including all target elements according to the target elements and the preset production scenario knowledge graph; Divide the on-site image into large-scale information regions and small-scale information regions according to the production scenario and at least one specification file corresponding to the production scenario; Identify potential safety hazards in the on-site image according to the large-scale information region and the small-scale information region; The process of dividing the large-scale information region and the small-scale information region includes: According to the production scenario, call at least one specification file corresponding to the production scenario, and call a large model to perform a standardized semantic output on all specification files to obtain multiple specification requirements; Based on the specification requirements, divide the image into large-scale information regions and small-scale information regions; The process of identifying potential safety hazards in the on-site image according to the large-scale information region and the small-scale information region includes: According to the divided large-scale information region and small-scale information region, split the preprocessed on-site image into at least one large-scale sub-image and at least one small-scale sub-image; Input the large-scale sub-image and the large-scale information into the first deep learning model corresponding to the production scenario, and input the small-scale sub-image into the second deep learning model corresponding to the production scenario; Combine the output results of the first deep learning model and the second deep learning model to identify potential safety hazards at the production site and record the results of each hazard identification; The first deep learning model and the second deep learning model include: The first deep learning model is established based on edge recognition. The first deep learning model first performs edge recognition on the large-scale sub-image to obtain boundary contour information, and determines whether it meets the large-scale information according to the similarity between the comparison boundary contour information and the large-scale information; The large-scale information includes the identified chemical addition room, main step-down and main control room, and window grille information; the small-scale information includes the chemical addition room lamp wires, curtains, and window grille corrosion area information; The second deep learning model is used for material identification to achieve the precise identification and analysis function of fine features in the small-scale sub-image; When performing boundary contour information recognition on the large-scale sub-image, it is a process of continuing to peel off the small-scale sub-image from the already divided large-scale sub-image, which belongs to a dynamic verification process. The specific steps of the dynamic verification include: Obtain the large-scale sub-image, calculate the texture complexity and gray-scale change rate indexes for each local area of the large-scale sub-image. When the texture complexity or gray-scale change rate index of any local area exceeds the preset threshold, it is considered that there are detailed hidden hazards in this local area that require further detailed analysis; According to the determined local areas with detailed hidden hazards, peel off the corresponding small-scale sub-images from the large-scale sub-image; Establish an image region selection rule associated with the detailed hidden hazard characteristics according to the detailed hidden hazard characteristics under different production scenarios.

2. The intelligent diagnosis method for potential safety hazards in production as claimed in claim 1, wherein, The process of determining the production scenario including all target elements includes: Input the on-site image into a multi-target recognition model, and the multi-target recognition model outputs each target element in the image; Input all the identified target elements into a preset production scenario knowledge graph, and determine the production scenario including all target elements according to the association relationships in the production scenario knowledge graph.

3. The intelligent diagnosis method for potential safety hazards in production as claimed in claim 1, characterized in that, The specific steps of the dynamic verification further include: After stripping the small-scale sub-image from the large-scale sub-image, establish a real-time feedback mechanism between the first deep learning model and the second deep learning model; Use the first deep learning model to conduct a preliminary analysis on the stripped area, and transmit the position, size, and preliminary analysis results of the stripped area to the second deep learning model; After receiving the information transmitted by the first deep learning model, the second deep learning model conducts a refined analysis on the small-scale sub-image, detects and identifies specific detail hidden danger information, where the detail hidden danger information includes the specific position, form, and association with the large-scale information of the hidden danger.

4. The intelligent diagnosis method for potential production safety hazards according to claim 3, characterized in that The specific steps of the dynamic verification further include: Feed back the detected detail hidden danger information to the first deep learning model; During multiple image stripping and model analysis processes, adjust the verification criteria and model parameters according to the comparison between each verification result and the actual hidden danger situation; Output the finally analyzed hidden danger information, position, and severity to the staff, and record the process and results of each analysis.

5. The intelligent diagnosis method for potential safety hazards in production according to claim 1, characterized in that, The specific process of stripping the corresponding small-scale sub-image from the large-scale sub-image includes: Determine the local area with detail hidden dangers in the large-scale sub-image, and the local area with detail hidden dangers is a bounding box composed of multiple pixel coordinates; According to the determined local area of detail hidden dangers, calculate the parameters required to strip the small-scale sub-image, including the starting coordinates, width, and height of the small-scale sub-image; Use the calculated stripping parameters to strip the corresponding small-scale sub-image from the large-scale sub-image through image cropping operations; Check whether the stripped small-scale sub-image contains a complete hidden danger area and does not contain unnecessary additional areas.

6. The intelligent diagnosis method for potential safety hazards in production according to claim 1, characterized in that, When combining the output results of the first deep learning model and the second deep learning model, a weighted average method is adopted, and different weights are assigned to them according to the accuracy rates of the first deep learning model and the second deep learning model and .

7. An intelligent diagnosis system for potential safety hazards in production, which is used to implement the intelligent diagnosis method for potential safety hazards in production according to any one of claims 1-6, and is characterized in that Include: An image processing module, a production scenario determination module, a sub-image generation module, a deep learning model analysis module, and a hidden danger identification and recording module; The image processing module collects on-site images through sensors and performs preprocessing; The production scenario determination module is used to identify each target element from the preprocessed image by using a multi-target recognition model, and determine the current production scenario according to the identified target elements in combination with a preset production scenario knowledge graph; The sub-image generation module is used to divide the image according to the determined specification requirements, and split the preprocessed on-site image into a large-scale sub-image and a small-scale sub-image according to the divided large-scale information area and small-scale information area; The deep learning model analysis module is used to input the large-scale sub-image and large-scale information into the first deep learning model corresponding to the production scenario, and input the small-scale sub-image into the second deep learning model corresponding to the production scenario for a preliminary analysis of safety hidden dangers; The hidden danger identification and recording module is used to identify the safety hidden dangers at the production site in combination with the output results of the two deep learning models, and record the results of each hidden danger identification.

8. The intelligent diagnosis system for potential safety hazards in production according to claim 7, characterized in that, The sub-image generation module includes: a specification file calling unit, a specification semantics output unit, a region division unit, and a sub-image cropping unit; The specification file calling unit is used to call the corresponding specification file according to the determined production scenario; The specification semantics output unit is used to perform semantic analysis on the specification file through a large model and output multiple specification requirements; The region division unit is used to divide the image into a large-scale information region and a small-scale information region based on the specification requirements; The sub-image cropping unit is used to split the preprocessed on-site image into a large-scale sub-image and a small-scale sub-image according to the divided large-scale information region and small-scale information region.

Images