Method, device and system for identifying hidden dangers of chemical equipment

Abstract

The invention relates to the technical field of chemical equipment hidden danger investigation, and particularly discloses a chemical equipment hidden danger identification method, device and system, and the method comprises the steps: obtaining to-be-identified chemical equipment image data; the to-be-recognized chemical equipment image data are input into a chemical industrial park field large model for hidden danger automatic recognition, a chemical equipment hidden danger recognition result is obtained, and the chemical industrial park field large model is obtained through self-supervised learning according to a pre-trained visual large model and a chemical equipment training data set. The chemical equipment hidden danger identification result at least comprises target chemical equipment information and a corresponding hidden danger type; and outputting and displaying the chemical equipment hidden danger identification result to a user. The chemical equipment hidden danger identification method provided by the invention can accurately and efficiently realize automatic identification of chemical equipment hidden dangers.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical equipment hidden danger detection, and in particular to a chemical equipment hidden danger identification method, a chemical equipment hidden danger identification device and a chemical equipment hidden danger identification system. Background Art

[0002] Existing chemical equipment hidden danger detection technologies mainly rely on manual inspections and traditional image processing methods. These methods often have problems such as low recognition accuracy, low efficiency, and high requirements for professional skills when dealing with chemical equipment in complex environments.

[0003] Traditional technologies have obvious deficiencies in processing large-scale data and real-time monitoring. For example, manual inspections are difficult to cover all equipment, and are less feasible in bad weather or dangerous environments. In addition, when identifying equipment hazards in complex backgrounds, traditional image processing technology is easily affected by factors such as lighting, shadows, and occlusions, resulting in false alarms and missed alarms. In addition, although there are some technical solutions in the prior art that attempt to improve the efficiency and accuracy of hazard detection through automation technology, these technical solutions still have some limitations, such as poor adaptability to specific types of equipment, or limited generalization capabilities when dealing with diverse hazards. Etc.

[0004] It can be seen that the chemical equipment in the prior art has the following main defects when conducting hidden danger inspections: (1) Low efficiency. Since traditional manual inspection methods require a lot of manpower and time, they are a huge burden for large chemical companies; (2) Lack of accuracy. Manual inspections are easily affected by the experience and subjective judgment of inspectors, which may lead to missed detections or misjudgments of hidden dangers; (3) Safety issues. Manual inspections in some high-risk areas may pose a threat to the safety of inspectors; (4) Limited data processing capabilities. The amount of data generated by manual inspections is limited, and it is difficult to conduct effective data analysis and historical comparisons; (5) Lack of real-time monitoring. Existing technologies cannot achieve real-time monitoring of chemical equipment and hidden danger warnings.

[0005] Therefore, how to provide a method for troubleshooting hidden dangers in chemical equipment that is efficient, accurate, and highly adaptable has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the invention

[0006] The present invention provides a chemical equipment hidden danger identification method, a chemical equipment hidden danger identification device and a chemical equipment hidden danger identification system, which solve the problem that the related technology cannot realize efficient and accurate chemical equipment hidden danger detection.

[0007] As a first aspect of the present invention, a method for identifying hidden dangers of chemical equipment is provided, which comprises:

[0008] Acquire image data of chemical equipment to be identified;

[0009] Inputting the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtaining chemical equipment hidden danger identification results, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification results at least include target chemical equipment information and corresponding hidden danger types;

[0010] The chemical equipment hidden danger identification result is output and displayed to the user.

[0011] Furthermore, the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, including:

[0012] Performing domain adjustment on the pre-trained visual large model to obtain an adjusted pre-trained visual large model, wherein the adjusted pre-trained visual large model at least includes: a multi-layer feature extraction unit, an adaptive alignment feature fusion unit, and a multi-task decoupled prediction head unit;

[0013] The chemical equipment training data set is input into the adjusted pre-trained visual large model to perform feature extraction, feature fusion and result prediction in sequence to complete the self-supervised learning process and obtain a chemical park field large model.

[0014] Furthermore, the pre-trained visual large model is domain adjusted, including:

[0015] The loss function of the pre-trained visual large model is adjusted based on the chemical park field to determine the loss function of the chemical park field large model, wherein the expression of the loss function of the chemical park field large model is:

[0016] L total =αL pretrained +βL domain ,

[0017] Among them, L total represents the loss function of the large model in the chemical park field, L pretrained represents the loss function of the pre-trained visual model, L domain represents the loss function corresponding to the hidden danger type, and α and β both represent weights.

[0018] Furthermore, the chemical equipment training data set is input into the adjusted pre-trained visual large model to perform feature extraction, feature fusion and result prediction in sequence to complete the self-supervised learning process, including:

[0019] Performing image processing on the chemical equipment training images in the chemical equipment training data set to obtain chemical equipment training processed images;

[0020] Performing hierarchical feature extraction on the chemical equipment training processing image based on multiple convolutional layers to obtain multi-layer chemical equipment training image extraction features;

[0021] Perform feature weighted fusion on multi-layer chemical equipment training image features to obtain chemical equipment training image fusion features;

[0022] Perform result prediction based on multiple prediction tasks on the fusion features of the chemical equipment training image;

[0023] Repeat the above feature extraction, feature fusion and result prediction to complete the self-supervised learning process.

[0024] Further, performing image processing on the chemical equipment training images in the chemical equipment training data set to obtain chemical equipment training processed images includes:

[0025] Predicting invisible image areas for visible image areas in the chemical equipment training image;

[0026] According to the predicted invisible image area and combined with the visible image area, a chemical equipment training processing image is obtained;

[0027] Among them, the prediction expression of the invisible image area is:

[0028]

[0029] Among them, L represents the difference loss function between the visible image area and the invisible area, x masked represents the invisible area, f represents the model function, x visible Represents the visible area, and N represents the number of samples of chemical equipment training images.

[0030] Furthermore, the expression of the hierarchical feature extraction is:

[0031] F l+1 =ReLU(W l ×F l +b l ),

[0032] Among them, F l represents the feature map of layer l, W l and b l Represent the weight and bias of the lth layer respectively;

[0033] The expression of the feature weighted fusion is:

[0034]

[0035] Among them, F fused represents the fusion features of chemical equipment training images, α l Represents the weight of the feature map of layer l;

[0036] The expression for result prediction based on multiple prediction tasks is:

[0037]

[0038] Among them, L multi-task Indicates result prediction based on multiple prediction tasks, L i represents the loss of the i-th prediction task, λ i represents the weight of the i-th prediction task, and M represents the number of prediction tasks.

[0039] Furthermore, it also includes:

[0040] Call the large language question answering model;

[0041] Inputting the chemical equipment hidden danger identification result into the large language question-answering model, and obtaining the question-answering result of the large language model, wherein the large language question-answering model can generate hidden danger content description, hidden danger cause analysis and treatment measure recommendation according to the chemical equipment hidden danger identification result;

[0042] The question-answering result of the large language model is output to the user.

[0043] Furthermore, it also includes:

[0044] Compare and analyze the chemical equipment hidden danger identification result with the preset warning level;

[0045] Generate early warning information based on the comparison and analysis results;

[0046] Send warning information to users.

[0047] As another aspect of the present invention, a chemical equipment hidden danger identification device is provided, which is used to implement the chemical equipment hidden danger identification method mentioned above, and includes:

[0048] An acquisition module, used for acquiring image data of chemical equipment to be identified;

[0049] A hidden danger identification module is used to input the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtain the chemical equipment hidden danger identification result, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification result at least includes the target chemical equipment information and the corresponding hidden danger type;

[0050] The output module is used to output and display the chemical equipment hidden danger identification result to the user.

[0051] As another aspect of the present invention, a chemical equipment hidden danger identification system is provided, which includes: an image acquisition device, a human-computer interaction device and the chemical equipment hidden danger identification device mentioned above, wherein the image acquisition device and the human-computer interaction device are both communicatively connected with the chemical equipment hidden danger identification device;

[0052] The image acquisition device is used to acquire image data of chemical equipment to be identified;

[0053] The chemical equipment hidden danger identification device is used to automatically identify hidden dangers of chemical equipment image data to be identified, and obtain chemical equipment hidden danger identification results;

[0054] The human-computer interaction device is used to realize the interaction between the chemical equipment hidden danger identification device and the user.

[0055] The chemical equipment hidden danger identification method provided by the present invention automatically identifies hidden dangers in chemical equipment image data to be identified through a chemical park field big model, and the chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set. Through the automatic hidden danger identification, target chemical equipment information and corresponding hidden danger types can be obtained. Since the chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set, it can solve the problems of low efficiency, insufficient accuracy and safety of the prior art in intelligent identification of chemical equipment and hidden danger identification, and realize high-precision, high-efficiency and safe intelligent identification of chemical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0057] Figure 1 The present invention provides a flowchart of the method for identifying hidden dangers in chemical equipment.

[0058] Figure 2 This is a flow chart of the self-supervised learning method for a large model in the chemical park field provided by the present invention.

[0059] Figure 3 This is a schematic diagram of the algorithm principle for self-supervised learning of a chemical equipment training data set provided by the present invention in an adjusted pre-trained visual large model.

[0060] Figure 4 This is a structural block diagram of the chemical equipment hidden danger identification device provided by the present invention.

[0061] Figure 5 This is a structural block diagram of the chemical equipment hidden danger identification system provided by the present invention.

[0062] Figure 6 A schematic diagram of the software architecture of the chemical equipment hidden danger identification system provided by the present invention.

[0063] Figure 7 This is a schematic diagram of the AI ​​recognition inspection function provided by the present invention.

[0064] Figure 8 This is a schematic diagram of the photo recognition function provided by the present invention.

[0065] Fig. 9 This is a schematic diagram of the video recording recognition function provided by the present invention.

[0066] Fig.10 Schematic diagram of the intelligent question-answering assistant provided by the present invention. DETAILED DESCRIPTION

[0067] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0068] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0069] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0070] Traditional technologies have obvious deficiencies in processing large-scale data and real-time monitoring. For example, manual inspections are difficult to cover all equipment, and are less feasible in bad weather or dangerous environments. In addition, traditional image processing technology is easily affected by factors such as lighting, shadows, and occlusions when identifying hidden dangers of equipment in complex backgrounds, resulting in false alarms and missed alarms.

[0071] Although there may have been some attempts in the prior art to improve the efficiency and accuracy of hidden danger detection through automation technology, these patents may not have fully utilized the advantages of deep learning and big data analysis, or have certain limitations in practical applications, such as poor adaptability to specific types of equipment, or limited generalization capabilities when dealing with diverse hidden dangers.

[0072] Based on this, a method for identifying hidden dangers of chemical equipment is provided in this embodiment. Figure 1 is a flow chart of a method for identifying hidden dangers of chemical equipment provided according to an embodiment of the present invention. Figure 1 As shown, including:

[0073] S100, obtaining image data of chemical equipment to be identified;

[0074] In the embodiment of the present invention, the image data of chemical equipment can be collected in real time by an image collection device installed in a chemical park. For example, the image of chemical equipment can be collected in real time by a camera device, or video information of chemical equipment can be collected.

[0075] Specifically, the information in the chemical equipment image data to be identified includes but is not limited to the appearance, operating status and environmental conditions of the chemical equipment.

[0076] S200, inputting the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtaining chemical equipment hidden danger identification results, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification results at least include target chemical equipment information and corresponding hidden danger types;

[0077] In an embodiment of the present invention, the above chemical equipment image data to be identified is input into a chemical park domain model for automatic identification of hidden dangers. The chemical park domain model specifically performs self-supervised learning based on a pre-trained visual model and a chemical equipment training data set. The specific chemical equipment training data set is a plurality of chemical equipment image data with labeled information. The chemical equipment image data with labeled information is input into the pre-trained data model for feature self-learning, thereby obtaining a chemical park domain model that can accurately identify potential hidden dangers of chemical equipment. Based on the obtained chemical park domain model, the hidden dangers of the chemical equipment image data to be identified are automatically identified, and the identification results of the hidden dangers of the chemical equipment can be output.

[0078] It should be noted that, in the embodiment of the present invention, the chemical equipment hidden danger identification result can specifically include the target chemical equipment information and the corresponding hidden danger type. For example, the chemical equipment image data to be identified includes chemical equipment A, chemical equipment B and chemical equipment C. The hidden danger types corresponding to chemical equipment A are a1, a2 and a3, the hidden danger types corresponding to chemical equipment B are b1, b2 and b3, and the hidden danger types corresponding to chemical equipment C are c1, c2 and c3. If the target chemical equipment automatically identified here is chemical equipment B, the hidden danger type corresponding to the chemical equipment B can also be output, such as the hidden danger type b1 corresponding to chemical equipment B.

[0079] It should also be noted that the pre-trained visual large model of the embodiment of the present invention can be specifically a TargetTransformer model, which is specifically a deep learning model based on the Transformer architecture, suitable for fault diagnosis of chemical processes. The Target Transformer model adopts an encoder-decoder structure, in which the encoder processes the input sequence and the decoder generates a fault diagnosis result. The core of the Target Transformer model is the self-attention mechanism, which allows the model to dynamically focus on different parts of the input sequence when processing data. Through multi-head attention, the Target Transformer model can learn in parallel and capture information from different subspaces. The introduction of position encoding enables the model to understand the sequential relationship of elements in the sequence, while residual connections and layer normalization help to stabilize the training of deep networks.

[0080] It should be understood that when the target chemical equipment is output as a chemical equipment hidden danger result, it can be specifically displayed in the form of annotation on the identified chemical equipment image, for example, it can be displayed by annotating with a red frame, or displaying coordinate information, etc., and annotating its hidden danger type.

[0081] S300: outputting and displaying the chemical equipment hidden danger identification result to the user.

[0082] In the embodiment of the present invention, the chemical equipment hidden danger identification result can be specifically displayed to the user through a human-computer interaction device.

[0083] In summary, the chemical equipment hidden danger identification method provided by the present invention automatically identifies hidden dangers in the chemical equipment image data to be identified through a chemical park domain big model, and the chemical park domain big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set. Through the automatic identification of hidden dangers, the target chemical equipment information and the corresponding hidden danger type can be obtained. Since the chemical park domain big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set, it can solve the problems of low efficiency, insufficient accuracy and safety of the prior art in intelligent identification of chemical equipment and hidden danger identification, and realize high-precision, high-efficiency and safe intelligent identification of chemical equipment.

[0084] It should be understood that, in order to improve the accuracy and efficiency of subsequent processing, the acquired chemical equipment image data to be identified is preprocessed, for example, the chemical equipment image data to be identified can be denoised, contrast enhanced and size normalized.

[0085] In an embodiment of the present invention, the chemical park domain model is obtained by self-supervised learning based on a pre-trained visual model and a chemical equipment training data set. Figure 2 As shown, including:

[0086] S210, performing domain adjustment on the pre-trained visual large model to obtain an adjusted pre-trained visual large model, wherein the adjusted pre-trained visual large model at least includes: a multi-layer feature extraction unit, an adaptive alignment feature fusion unit, and a multi-task decoupling prediction head unit;

[0087] In an embodiment of the present invention, the pre-trained visual big model can specifically obtain an adjusted pre-trained time big model through domain adjustment, and self-supervised learning is performed on the chemical equipment training data set based on the adjusted pre-trained visual big model to obtain a chemical park domain big model.

[0088] Specifically, the pre-trained visual large model is domain adjusted, including:

[0089] The loss function of the pre-trained visual large model is adjusted based on the chemical park field to determine the loss function of the chemical park field large model, wherein the expression of the loss function of the chemical park field large model is:

[0090] L total =αL pretrained +βL domain ,

[0091] Among them, Ltotal represents the loss function of the large model in the chemical park field, L pretrained represents the loss function of the pre-trained visual model, L domain represents the loss function corresponding to the hidden danger type, and α and β both represent weights.

[0092] It should be noted that, in the embodiment of the present invention, the loss function corresponding to the hidden danger type specifically includes classification loss and detection loss, that is, for the classification task of different types of hidden dangers (such as leakage, corrosion, structural deformation, etc.), the cross loss function is used to optimize the model's ability to recognize the hidden danger type. Here, the model needs to locate the specific position of the hidden danger, and the bounding box regression loss (such as IoU loss) will be used to optimize the model's ability to detect the hidden danger position.

[0093] It should be understood that domain adaptation can preserve the general knowledge of the pre-trained visual large model while learning specific features of a specific domain. By adjusting the α and β weights, the importance of pre-trained knowledge and domain-specific knowledge can be balanced, thereby optimizing the performance of the model on a specific task.

[0094] S220, inputting the chemical equipment training data set into the adjusted pre-trained visual big model to perform feature extraction, feature fusion and result prediction in sequence to complete the self-supervised learning process and obtain a chemical park domain big model.

[0095] In the embodiment of the present invention, Figure 3 As shown, it is a schematic diagram of the algorithm principle for self-supervised learning of the chemical equipment training data set in the adjusted pre-trained visual large model. In the embodiment of the present invention, the improved pre-trained visual large model algorithm is used as the basic framework, and customized improvements are made according to the characteristics of chemical equipment. Combined with the powerful feature extraction capability of the convolutional neural network, feature learning is performed on the image of the chemical equipment. The model is trained by a large amount of annotated data, so that it can accurately identify chemical equipment and its potential hidden dangers.

[0096] In an embodiment of the present invention, the adjusted pre-trained large visual model performs feature extraction through a multi-layer feature extraction unit (corresponding to the backbone network backbone), performs feature fusion through an adaptive feature fusion unit (corresponding to the neck structure neck), and performs result prediction output through a multi-task decoupled prediction head unit (corresponding to the head structure head).

[0097] Specifically, the chemical equipment training data set is input into the adjusted pre-trained visual large model to perform feature extraction, feature fusion and result prediction in sequence to complete the self-supervised learning process, including:

[0098] 1) performing image processing on the chemical equipment training images in the chemical equipment training data set to obtain chemical equipment training processed images;

[0099] In self-supervised learning, the model learns by predicting parts of the data that are masked or perturbed during training. This approach does not require external annotations, but instead uses the structural information of the data itself as a supervisory signal.

[0100] Therefore, in image processing, the model needs to predict the masked image blocks. Specifically, performing image processing on the chemical equipment training image in the chemical equipment training data set to obtain the chemical equipment training processed image includes:

[0101] Predicting invisible image areas for visible image areas in the chemical equipment training image;

[0102] According to the predicted invisible image area and combined with the visible image area, a chemical equipment training processing image is obtained;

[0103] Among them, the prediction expression of the invisible image area is:

[0104]

[0105] Among them, L represents the difference loss function between the visible image area and the invisible area, x masked represents the invisible area, f represents the model function, x visible Represents the visible area, and N represents the number of samples of chemical equipment training images.

[0106] It should be noted that the purpose of the difference loss function L is to quantify the difference between the visible area and the invisible area.

[0107] 2) performing hierarchical feature extraction on the chemical equipment training processing image based on multiple convolutional layers to obtain multi-layer chemical equipment training image extraction features;

[0108] It should be understood that a multi-layer feature extractor, that is, a stack of multiple convolutional layers, is used to extract features at different levels in each layer.

[0109] Specifically, the expression of the hierarchical feature extraction is:

[0110] F l+1 =ReLU(W l ×F l +b l ),

[0111] Among them, F l represents the feature map of layer l, W l and b l Represent the weight and bias of the lth layer respectively;

[0112] It should be understood that this hierarchical feature extraction method can help the model capture features from low-level to high-level, providing rich information for subsequent tasks.

[0113] 3) Performing feature weighted fusion on multi-layer chemical equipment training image features to obtain chemical equipment training image fusion features;

[0114] With respect to the above-extracted features, in an embodiment of the present invention, an adaptive feature fusion method is used to enhance the features, which is achieved by weighted summation of feature maps of different layers.

[0115] Specifically, the expression of the feature weighted fusion is:

[0116]

[0117] Among them, F fused represents the fusion features of chemical equipment training images, α l Represents the weight of the feature map of layer l;

[0118] This fusion method of the embodiment of the present invention can dynamically adjust the contribution of features at different layers according to the requirements of the task, thereby improving the performance of the model.

[0119] 4) performing result prediction based on multiple prediction tasks by fusing features of the chemical equipment training image;

[0120] In the embodiment of the present invention, the prediction head may specifically include multiple ones, thereby being able to implement result prediction of multiple prediction tasks.

[0121] Specifically, the expression for result prediction based on multiple prediction tasks is:

[0122]

[0123] Among them, L multi-task Indicates result prediction based on multiple prediction tasks, L i represents the loss of the i-th prediction task, λ i represents the weight of the i-th prediction task, and M represents the number of prediction tasks.

[0124] In the embodiment of the present invention, through the prediction head structure, the model can learn knowledge that is helpful for other tasks while processing one task, thereby improving the generalization ability of the model.

[0125] 5) Repeat the above feature extraction, feature fusion and result prediction to complete the self-supervised learning process.

[0126] Therefore, the chemical park field big model of the present invention is specifically based on the self-supervised pre-trained visual big model combined with the chemical equipment training data set to achieve high-precision small-sample hidden danger inspection, and achieve the optimal efficiency and accuracy balance through the task target recognition architecture.

[0127] It should be noted that the chemical park domain model can automatically identify abnormal conditions of chemical equipment, such as leakage, corrosion, structural deformation, etc. The identified abnormal hazards can also be classified and prioritized according to severity and potential risks.

[0128] In the embodiment of the present invention, after obtaining the chemical equipment hidden danger identification result for the above-mentioned chemical park field large model, the processing measures for abnormal situations can also be obtained through the large language question-answering model. Therefore, the chemical equipment hidden danger identification method of the embodiment of the present invention also includes:

[0129] (1) Calling the large language question answering model;

[0130] (2) inputting the chemical equipment hidden danger identification result into a large language question-answering model, and obtaining a question-answering result of the large language model, wherein the large language question-answering model can generate a hidden danger content description, hidden danger cause analysis, and treatment measure recommendation according to the chemical equipment hidden danger identification result;

[0131] (3) Outputting the question-answering results of the large language model to the user.

[0132] It should be understood that in the embodiment of the present invention, a large model question-answering system can be integrated to assist in identification and provide decision support. The question-answering system can provide detailed descriptions of hidden dangers, possible causes, and recommended treatment measures based on the hidden danger identification results.

[0133] In order to facilitate users to obtain hidden danger identification results in a timely manner, the chemical equipment automatic identification method also includes:

[0134] Compare and analyze the chemical equipment hidden danger identification result with the preset warning level;

[0135] Generate early warning information based on the comparison and analysis results;

[0136] Send warning information to users.

[0137] It should be understood that the real-time monitoring of chemical equipment can be realized, and the model can update the recognition results in real time through continuous data flow input. The early warning mechanism is designed so that when potential hidden dangers are detected, the system can issue an alarm in time to notify relevant personnel to deal with it.

[0138] The embodiment of the present invention realizes the display of the hidden danger identification results of chemical equipment through a human-computer interaction device, and specifically allows the operator to view the identification results, hidden danger classification and system suggestions through a user interface. In addition, the operator can confirm or correct the identification results, and the system performs self-learning and optimization based on the feedback.

[0139] Therefore, the chemical equipment hidden danger identification method provided by the present invention can automatically identify chemical equipment and detect potential hidden dangers. By introducing a large model question-and-answer system to assist in identification, the present invention not only improves the accuracy and efficiency of hidden danger identification, but also enhances the security and real-time monitoring capabilities of the system. In addition, the present invention can also process and analyze a large amount of data, and can provide chemical companies with deeper insights and decision support through file upload and question-and-answer methods.

[0140] As another embodiment of the present invention, a chemical equipment hidden danger identification device 100 is provided, which is used to implement the chemical equipment hidden danger identification method described above, wherein: Figure 4 As shown, including:

[0141] An acquisition module 110 is used to acquire image data of chemical equipment to be identified;

[0142] The hidden danger identification module 120 is used to input the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtain the chemical equipment hidden danger identification result, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification result at least includes the target chemical equipment information and the corresponding hidden danger type;

[0143] The output module 130 is used to output and display the chemical equipment hidden danger identification result to the user.

[0144] The chemical equipment hidden danger identification device provided by the present invention automatically identifies hidden dangers of chemical equipment image data to be identified through a chemical park field big model, and the chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set. Through the automatic identification of hidden dangers, target chemical equipment information and corresponding hidden danger types can be obtained. Since the chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set, it can solve the problems of low efficiency, insufficient accuracy and safety of the prior art in intelligent identification of chemical equipment and hidden danger identification, and realize high-precision, high-efficiency and safe intelligent identification of chemical equipment.

[0145] The working principle of the chemical equipment hidden danger identification device provided by the present invention can be referred to the description of the chemical equipment hidden danger identification method in the previous text, and will not be repeated here.

[0146] As another embodiment of the present invention, a chemical equipment hidden danger identification system 10 is provided, wherein: Figure 5 As shown, it includes: an image acquisition device 200, a human-computer interaction device 300 and the chemical equipment hidden danger identification device 100 mentioned above, and the image acquisition device 200 and the human-computer interaction device 300 are both connected to the chemical equipment hidden danger identification device 100 for communication;

[0147] The image acquisition device 200 is used to acquire image data of chemical equipment to be identified;

[0148] The chemical equipment hidden danger identification device 100 is used to automatically identify hidden dangers of chemical equipment image data to be identified, and obtain chemical equipment hidden danger identification results;

[0149] The human-computer interaction device 300 is used to realize the interaction between the chemical equipment hidden danger identification device and the user.

[0150] The chemical equipment hidden danger identification system provided by the present invention adopts the chemical equipment hidden danger identification device mentioned above, and automatically identifies hidden dangers of chemical equipment image data to be identified through a chemical park field big model. The chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set. Through the automatic identification of hidden dangers, the target chemical equipment information and the corresponding hidden danger type can be obtained. Since the chemical park field big model is obtained by self-supervised learning based on a pre-trained visual big model and a chemical equipment training data set, it can solve the problems of low efficiency, insufficient accuracy and safety of the prior art in intelligent identification of chemical equipment and hidden danger identification, and realize high-precision, high-efficiency and safe intelligent identification of chemical equipment.

[0151] Combine the following Figure 6 The software architecture of the chemical equipment hidden danger identification system of the present invention is described in detail.

[0152] like Figure 6 As shown in the figure, the software module of the chemical equipment hidden danger identification system mainly consists of the following parts:

[0153] Application layer: including three modules: daily hidden danger inspection, regular hidden danger inspection and special hidden danger inspection for chemical enterprises. This layer is directly facing users and provides customized services;

[0154] Service layer: includes intelligent device identification, intelligent hidden danger identification, Chat Q&A system, Knowledge intelligent Q&A and equipment risk point library. These modules are responsible for processing requests from the application layer and performing specific identification, Q&A and risk assessment tasks;

[0155] Functional layer: It consists of three modules: intelligent recognition based on a large visual model, intelligent question and answer based on a large natural language processing model, and guided investigation. This layer provides the core algorithms and logic to support the functional implementation of the service layer;

[0156] Data layer: including videos, pictures, policy documents, equipment model library and equipment risk point library. This layer is responsible for storing and managing data and providing data support for the functional layer;

[0157] Support layer: It is composed of OpenCV, TensorFlow, PyTorch, NVIDIA, Android, Hadoop, Kubernetes and other technologies. This layer provides the technical foundation and operating environment to ensure the stable operation of the entire system.

[0158] The application layer calls the algorithms and logic of the functional layer through the service layer. The functional layer uses the data of the data layer for processing, and the entire system relies on the technology of the support layer to be implemented.

[0159] Based on the above software architecture, the implementation process specifically includes:

[0160] User request: The user initiates a hidden danger troubleshooting request through the application layer;

[0161] Service layer processing: After receiving the request, the service layer calls the corresponding intelligent recognition or question-answering system according to the request type (daily, regular or special inspection);

[0162] Functional layer execution: The intelligent recognition module of the functional layer uses the visual basic model to analyze videos or pictures to identify equipment and potential hazards; the intelligent question-answering module uses the natural language processing basic model to answer users' questions;

[0163] Data layer support: During the identification and question-answering process, the functional layer will obtain necessary data from the data layer, such as device models, risk point information, etc.

[0164] Result feedback: The service layer feeds back the processing results of the functional layer to the application layer, and the application layer displays the results to the user;

[0165] User interaction: Users can confirm the results or make further inquiries. This interaction process will trigger processing at the service layer and function layer again.

[0166] System optimization: The system will continuously optimize the recognition algorithm and question-answering system based on user feedback and new data to improve accuracy and efficiency.

[0167] The above process covers the entire process from user request to system response, demonstrating how various modules work together to realize the function of hidden danger detection of chemical equipment.

[0168] In the embodiment of the present invention, Figure 7 Taking the example shown, the AI ​​recognition and inspection function supports users to intelligently identify key target equipment by taking photos or real-time videos, automatically obtain equipment risk points, assist in hidden danger identification image acquisition, record inspection results in real time, and simplify user operation processes. It not only improves the efficiency of safety management, but also provides solid technical support for safe production.

[0169] like Figure 8 As shown, photo recognition can intelligently identify the target equipment by taking photos, automatically obtain the risk points of the equipment to be checked, and further inspect the hidden dangers of the equipment.

[0170] like Fig. 9 As shown in the figure, the video recognition function supports real-time acquisition of video streams, and intelligently identifies target devices through technologies such as image processing, image segmentation, image classification, OCR recognition, voiceprint recognition, and audio noise reduction. It automatically obtains risk points of the equipment to be checked, provides decision-making support for hidden danger detection, and greatly improves inspection efficiency and accuracy.

[0171] In the embodiment of the present invention, Fig.10 As shown in Figure 1, the intelligent question-answering assistant is an interactive tool that integrates advanced natural language processing technology, designed to optimize the user experience by accurately understanding user queries and providing instant, accurate, and helpful responses. It can handle complex instructions, conduct multiple rounds of conversations, and provide users with comprehensive answers based on the latest search results, while complying with security and compliance requirements to ensure the suitability and professionalism of the content.

[0172] The intelligent question-and-answer assistant has a powerful fuzzy query function that can understand and process query requests that are not completely clear or ambiguous, significantly improving the user experience and enabling users to obtain useful and timely information feedback even when their expressions are not precise enough, thereby enhancing the flexibility of interaction and the practicality of the assistant.

[0173] In summary, the present invention effectively solves the efficiency and accuracy problems in the hidden danger inspection of chemical equipment by introducing intelligent recognition technology based on a large visual model and a large model question-answering system based on natural language processing. This technical solution uses a deep learning algorithm to automatically and intelligently identify and troubleshoot chemical equipment, significantly improving the recognition speed and accuracy while reducing the interference of human factors. Compared with traditional manual inspection methods, the system of the present invention can identify potential safety hazards more quickly and issue early warnings in a timely manner, thereby avoiding accidents and significantly improving the safety management level of chemical companies.

[0174] In addition, the present invention has achieved a significant improvement in system performance. Through real-time monitoring and early warning mechanisms, the present invention can issue an alarm before hidden dangers occur, allowing relevant personnel to respond quickly, thereby improving the safety of chemical companies. This preventive measure is lacking in the prior art, and it can significantly reduce the incidence of safety accidents. At the same time, the present invention provides a user-friendly interface and interactive mode, allowing operators to easily use the system to check for hidden dangers, thereby improving user experience.

[0175] In terms of economic and environmental benefits, the present invention also shows significant beneficial effects. By improving the efficiency and accuracy of hidden danger detection, the present invention can help enterprises reduce economic losses caused by safety accidents. In the long run, this economic benefit is significant and helps to improve the competitiveness of enterprises. At the same time, the present invention helps to reduce environmental pollution and ecological damage by improving the safety of chemical enterprises, reflecting the positive impact of the social value and environmental benefits of the present invention.

[0176] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying hidden dangers of chemical equipment, characterized in that: include: Acquire image data of chemical equipment to be identified; Inputting the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtaining chemical equipment hidden danger identification results, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification results at least include target chemical equipment information and corresponding hidden danger types; The chemical equipment hidden danger identification result is output and displayed to the user.

2. The method for identifying hidden dangers of chemical equipment according to claim 1, characterized in that: The chemical park domain model is obtained by self-supervised learning based on the pre-trained visual model and the chemical equipment training data set, including: Performing domain adjustment on the pre-trained visual large model to obtain an adjusted pre-trained visual large model, wherein the adjusted pre-trained visual large model at least includes: a multi-layer feature extraction unit, an adaptive alignment feature fusion unit, and a multi-task decoupled prediction head unit; The chemical equipment training data set is input into the adjusted pre-trained visual large model to perform feature extraction, feature fusion and result prediction in sequence to complete the self-supervised learning process and obtain a chemical park field large model.

3. The method for identifying hidden dangers of chemical equipment according to claim 2, characterized in that: The pre-trained visual model is domain adjusted, including: The loss function of the pre-trained visual large model is adjusted based on the chemical park field to determine the loss function of the chemical park field large model, wherein the expression of the loss function of the chemical park field large model is: L total =αL pretrained +βL domain , Among them, L total represents the loss function of the large model in the chemical park field, L pretrained represents the loss function of the pre-trained visual model, L domain represents the loss function corresponding to the hidden danger type, and α and β both represent weights.

4. The method for identifying hidden dangers of chemical equipment according to claim 2, characterized in that: Inputting the chemical equipment training data set into the adjusted pre-trained visual large model to sequentially perform feature extraction, feature fusion and result prediction to complete the self-supervised learning process, including: Performing image processing on the chemical equipment training images in the chemical equipment training data set to obtain chemical equipment training processed images; Performing hierarchical feature extraction on the chemical equipment training processing image based on multiple convolutional layers to obtain multi-layer chemical equipment training image extraction features; Perform feature weighted fusion on multi-layer chemical equipment training image features to obtain chemical equipment training image fusion features; Perform result prediction based on multiple prediction tasks on the fusion features of the chemical equipment training image; Repeat the above feature extraction, feature fusion and result prediction to complete the self-supervised learning process.

5. The method for identifying hidden dangers of chemical equipment according to claim 4, characterized in that: Performing image processing on the chemical equipment training images in the chemical equipment training data set to obtain chemical equipment training processed images, including: Predicting invisible image areas for visible image areas in the chemical equipment training image; According to the predicted invisible image area and combined with the visible image area, a chemical equipment training processing image is obtained; Among them, the prediction expression of the invisible image area is: Among them, L represents the difference loss function between the visible image area and the invisible area, x masked represents the invisible area, f represents the model function, x visible Represents the visible area, and N represents the number of samples of chemical equipment training images.

6. The method for identifying hidden dangers of chemical equipment according to claim 4, characterized in that: The expression of the hierarchical feature extraction is: F l+1 =ReLU(W l ×F l +b l ), Among them, F l represents the feature map of layer l, W l and b l Represent the weight and bias of the lth layer respectively; The expression of the feature weighted fusion is: Among them, F fused represents the fusion features of chemical equipment training images, α l Represents the weight of the feature map of layer l; The expression for result prediction based on multiple prediction tasks is: Among them, L multi-task Indicates result prediction based on multiple prediction tasks, L i represents the loss of the i-th prediction task, λ i represents the weight of the i-th prediction task, and M represents the number of prediction tasks.

7. The method for identifying hidden dangers of chemical equipment according to any one of claims 1 to 6, characterized in that: Also includes: Call the large language question answering model; Inputting the chemical equipment hidden danger identification result into the large language question-answering model, and obtaining the question-answering result of the large language model, wherein the large language question-answering model can generate hidden danger content description, hidden danger cause analysis and treatment measure recommendation according to the chemical equipment hidden danger identification result; The question-answering result of the large language model is output to the user.

8. The method for identifying hidden dangers of chemical equipment according to any one of claims 1 to 6, characterized in that: Also includes: Compare and analyze the chemical equipment hidden danger identification result with the preset warning level; Generate early warning information based on the comparison and analysis results; Send warning information to users.

9. A chemical equipment hidden danger identification device, used to implement the chemical equipment hidden danger identification method according to any one of claims 1 to 8, characterized in that: include: An acquisition module, used for acquiring image data of chemical equipment to be identified; A hidden danger identification module is used to input the chemical equipment image data to be identified into the chemical park domain big model for automatic hidden danger identification, and obtain the chemical equipment hidden danger identification result, wherein the chemical park domain big model is obtained by self-supervised learning based on the pre-trained visual big model and the chemical equipment training data set, and the chemical equipment hidden danger identification result at least includes the target chemical equipment information and the corresponding hidden danger type; The output module is used to output and display the chemical equipment hidden danger identification result to the user.

10. A chemical equipment hidden danger identification system, characterized in that: include: An image acquisition device, a human-computer interaction device, and the chemical equipment hidden danger identification device according to claim 9, wherein the image acquisition device and the human-computer interaction device are both communicatively connected to the chemical equipment hidden danger identification device; The image acquisition device is used to acquire image data of chemical equipment to be identified; The chemical equipment hidden danger identification device is used to automatically identify hidden dangers of chemical equipment image data to be identified, and obtain chemical equipment hidden danger identification results; The human-computer interaction device is used to realize the interaction between the chemical equipment hidden danger identification device and the user.

Images