Oil and gas field on-site video AI analysis system and method

Abstract

The invention discloses an oil and gas field on-site video AI analysis system and method, and belongs to the technical field of big data video AI analysis. The method comprises the following steps: collecting a field video image; the video image is preprocessed; obtaining a feature tag from the processed video image through an action model; integrating the feature tags in a preset time period to obtain a feature tag group, and performing similarity comparison on the feature tag group and a database to obtain risk information; and obtaining an early warning instruction according to the risk information, and sending the early warning instruction to the broadcast. On-site videos of the oil and gas field are collected, based on the artificial intelligence image recognition technology, intelligent recognition analysis of mass video monitoring of all operation areas of the oil and gas field is achieved, centralized visual safety supervision on key areas such as production sites, entrances and exits, crude oil storage areas and the like is achieved, accident potentials are early warned in the first time, and the video inspection pressure is relieved; the cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of big data video AI analysis, and in particular to an oil and gas field on-site video AI analysis system and method. Background Art

[0002] my country's oilfield industry has moved beyond its early stage of extensive growth and requires more refined, intelligent, and scientific operational management methods. Currently, oilfield production and management still face the following pain points:

[0003] Traditional security systems have drawbacks: high safety risk factors during oil production, transportation and refining, and no real-time warning; difficulty in manual management: with the continuous expansion of the production scale of large oil depots and the wide distribution of oil field production operations, staff safety management is difficult. Summary of the Invention

[0004] In order to solve the problems of the existing technology, the present invention provides an oil and gas field on-site video AI analysis system and method.

[0005] In one aspect, a method for AI analysis of oil and gas field on-site videos is provided, the method comprising:

[0006] Collect video images of the scene;

[0007] Preprocessing the video image;

[0008] The processed video image is passed through the action model to obtain feature labels;

[0009] Integrate the feature tags within a preset time period to obtain a feature tag group, compare the feature tag group with the database for similarity, and obtain risk information;

[0010] Obtain early warning instructions based on risk information and send them to the broadcaster.

[0011] Furthermore, the video images include: equipment operation video images, personnel video images and oil and gas field environment video images.

[0012] Furthermore, for the device operation video image, the feature tag includes the appearance and operation status of the device;

[0013] For the personnel video image, the feature tags include a safety helmet, work clothes, operator behavior and face image;

[0014] For the oil and gas field environment video image, the feature tag includes the surrounding environment.

[0015] Furthermore, the preprocessing includes: denoising, contrast enhancement, and brightness adjustment.

[0016] Furthermore, the action model model is a yolo model based on target search and a residual classifier model for image recognition.

[0017] Furthermore, the method further includes: saving the collected video images to facilitate historical data analysis and long-term data retention.

[0018] In another aspect, an oil and gas field on-site video AI analysis system is provided, the system comprising:

[0019] An acquisition module, used to acquire video images of the scene;

[0020] A preprocessing module, configured to preprocess the video image;

[0021] A feature label acquisition module is used to obtain feature labels from the processed video image through the action model model;

[0022] A database pre-stored with a set of security feature tags;

[0023] The comparison module is used to integrate the feature tags within a preset time period to obtain a feature tag group, compare the feature tag group with the database for similarity, and obtain risk information;

[0024] Early warning instruction extraction module, used to obtain early warning instructions based on risk information;

[0025] The broadcast module is used to broadcast warning instructions.

[0026] Furthermore, the system also includes: a saving module for saving the collected video images to facilitate historical data analysis and long-term data retention.

[0027] The beneficial effects of the technical solution provided by the embodiment of the present invention are as follows: by collecting on-site videos of oil and gas fields and based on artificial intelligence image recognition technology, the present invention realizes intelligent recognition and analysis of massive video surveillance in various operating areas of oil and gas fields, realizes centralized visual safety supervision of key areas such as production sites, entrances and exits, crude oil storage areas, etc., and warns of potential accidents in the first time, alleviates the pressure of video inspection, and reduces costs and increases efficiency.

[0028] Secondly, a safety supervision system for personnel, environment, and equipment is established. Feature tags are obtained through action models, and feature tag groups are compared to obtain risk information. Once a safety hazard is detected, an alarm is issued in real time to achieve efficient and intelligent supervision of on-site operation risks and personnel violations in oil and gas fields, thereby improving risk prevention capabilities.

[0029] In addition, there are high safety risks in the oil and gas field production process, including fire, explosion, leakage, etc. These risks need to be reduced through effective monitoring and management methods; the skills of oil field operators vary, and there is a risk of illegal operations. Through AI analysis, the behavior of operators can be monitored in real time, and potential violations can be warned in time; oil and gas field production may have an impact on the environment. Through video AI analysis, environmental changes can be monitored and responded to in a timely manner; through the AI ​​video analysis system, the operating status of oil production equipment can be monitored in real time to prevent equipment failures and accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 This is a flow chart of an oil and gas field on-site video AI analysis method provided by the present invention;

[0032] Figure 2 This is a schematic diagram of an oil and gas field on-site video AI analysis system provided by the present invention;

[0033] Figure 3 It is the loss curve of a model provided by the present invention on the training set and the validation set;

[0034] Figure 4 It is a PR curve of a trained YOLOv5 model on a validation set provided by the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0036] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0037] Example 1

[0038] An AI analysis method for oil and gas field on-site videos includes the following steps:

[0039] Step (1): Collect video images of the scene.

[0040] It should be noted that first of all, a complete video surveillance system needs to be established to achieve comprehensive coverage and real-time monitoring of oil and gas field production sites. High-definition cameras and transmission equipment can transmit the on-site situation to the monitoring center in real time, helping managers to understand the production situation in a timely manner, detect anomalies and take corresponding measures.

[0041] Step (2): Preprocess the video image.

[0042] Through the standard RTSP\RTMP protocol, the video stream of the network camera and hard disk recorder is accessed, and the accessed video process is decoded, frame extracted, and image preprocessed. Image preprocessing includes denoising, contrast enhancement, brightness adjustment, etc. to improve image quality and provide better input for subsequent analysis.

[0043] Step (3): The processed video image is passed through the action model to obtain feature labels.

[0044] It should be noted that the image recognition processing technology based on deep learning is trained through a large amount of oil and gas field field data to train the corresponding scene AI model, which mainly includes the YOLO model (YOLOv5) based on target search and the residual classifier model (RESNET18) for image recognition. The specific training process is as follows:

[0045] Environment Configuration: First, you need to configure the hardware and software required for training. For hardware, you need a graphics card (GPU) with at least 4GB of video memory and at least 4GB of RAM. For software, you can choose a Windows or Linux operating system and install CUDA and CuDNN to accelerate model training. You will also need a Python IDE (such as PyCharm) and a Python package manager (such as Anaconda).

[0046] Install Yolov5: Download and install the open source Yolov5 code.

[0047] Data labeling: Use labelImg software to label data; this step is the basis of model training. The data needs to be labeled so that the model can learn the characteristics of the data.

[0048] Model training: In the configured environment, use PyCharm to open the Yolov5 project, select the virtual environment, test whether the code can run properly, then prepare your own dataset and write a data configuration file. These configuration files include parameters such as the dataset path, number of categories, and category names. After the configuration is complete, start training the model.

[0049] Model testing: After training is completed, the model needs to be tested to verify the performance of the model. This can be achieved by modifying the detect.py file and running the detect.py file to verify the detection results of the model.

[0050] Model deployment: If you need to deploy the model on other platforms, you may need to convert the model; for example, you can convert the model from PyTorch format to ncnn format to run on mobile devices or other platforms.

[0051] Effect evaluation:

[0052] YOLOv5 is a deep learning model for object detection that uses the YOLO (You Only Look Once) algorithm. Evaluation metrics are used to measure model performance. The following is an introduction to the commonly used evaluation metrics for YOLOv5:

[0053] mAP (mean Average Precision): Mean Average Precision. mAP is one of the most commonly used evaluation metrics in object detection, combining both precision and recall. mAP values ​​range from 0 to 1, with higher values ​​indicating better model performance.

[0054] mAP50: refers to the value of meanAverage Precision, which calculates the average precision of the detection model when the IoU (Intersection over Union) threshold is 0.5.

[0055] mAP50-95: mAP50-95 means that when calculating the average precision, the IoU threshold is calculated from 0.5 to 0.95, and then the average is taken; this indicator can more comprehensively evaluate the performance of the detection model at different IoU thresholds.

[0056] Precision: Precision refers to the proportion of samples predicted by the model as positive that are actually positive. The calculation formula for precision is the number of samples predicted as positive and correct divided by the number of samples predicted as positive.

[0057] Recall: Recall refers to the proportion of true positive examples that are correctly predicted as positive by the model. The calculation formula for Recall is the number of samples predicted as positive and correct divided by the number of samples that are true positive.

[0058] F1-score: The F1-score is the harmonic mean of precision and recall, taking into account the performance of both. The formula for calculating F1-score is 2*(Precision*Recall) / (Precision+Recall).

[0059] AP (Average Precision): Average precision; AP is a component of mAP, which is the average of the precision values ​​calculated under different confidence thresholds.

[0060] IoU (Intersection over Union): Intersection over Union ratio; IoU is the ratio of the overlapping parts between the predicted box and the true box, which is used to judge the degree of matching between the predicted box and the true box; generally, when IoU is greater than a certain threshold, it is considered that the predicted box and the true box are matched successfully.

[0061] In the experimental results and analysis section, we use indicators such as precision and recall to evaluate the performance of the model, and also analyze the training process through loss curves and PR curves. During the training phase, we used the oil and gas field dataset introduced earlier for training, and used the YOLOv5 algorithm to train the dataset for a total of 300 epochs. During the training process, we used tensorboard to record the loss curves of the model on the training set and validation set. Figure 3 It can be seen that as the number of training times increases, the training loss and validation loss of the model gradually decrease, indicating that the model continues to learn more accurate features.

[0062] Figure 4 The PR curve of the trained YOLOv5 model on the validation set is shown. As can be seen from the figure, the model has achieved high recall and precision, and performed well overall.

[0063] Step (4): Integrate the feature tags within the preset time period to obtain a feature tag group, compare the feature tag group with the database for similarity, obtain risk information, obtain warning instructions based on the risk information, and send the warning instructions to the broadcast.

[0064] It should be noted that the video images include: equipment operation video images, personnel video images and oil and gas field environment video images; for equipment operation video images, the feature tags include the appearance and operating status of the equipment. By monitoring the appearance of the equipment, it can be found whether the equipment is leaking, and by monitoring the equipment operating status, it can be found whether there is a fault; for personnel video images, the feature tags include safety helmets, work clothes, operator behavior and facial images. By monitoring safety helmets and work clothes, it can be determined whether the staff are wearing work clothes and safety helmets correctly in accordance with regulations when working. By monitoring the behavior of operators, it can be determined whether there are any illegal operations. Face recognition can be performed through facial images, and identity verification and management of personnel entering and leaving the oil and gas field can be carried out to ensure production safety; for oil and gas field environment video images, the feature tags include the surrounding environment. By monitoring the surrounding environment, it can be determined whether there are abnormal conditions such as smoke and flames.

[0065] Various safety feature labels can be pre-stored in the database. For example, after taking a video image of a person, the feature label of the head area will be obtained, and the feature label group of the safety helmet will be obtained. The safety feature label in the database is the feature label of correctly wearing a safety helmet. The obtained feature label group of the safety helmet will be compared with the safety feature label for similarity. If it is compared that there is no safety helmet on the head, an early warning instruction of not wearing a safety helmet is obtained, and the early warning instruction is broadcasted through broadcasting.

[0066] It should also be noted that the warning instructions corresponding to each type of risk information in each area are different; each camera has an independent device number, and the area photographed by each camera has an independent area code. The device number of each camera, the area code of the shooting area, the area location information, each type of risk information, and the warning instructions corresponding to each risk information are associated and saved. When broadcasting the warning instructions, the staff can clearly know which risk has occurred in which area.

[0067] In addition, the captured video images can be saved to facilitate historical data analysis and long-term data retention.

[0068] It is worth noting that the present invention is based on artificial intelligence image recognition technology to realize intelligent recognition and analysis of massive video surveillance in various operating areas of oil and gas fields, and realize centralized visual safety supervision of key areas such as production sites, entrances and exits, crude oil storage areas, etc., to warn of potential accidents in the first time, alleviate the pressure of video inspection, and reduce costs and increase efficiency.

[0069] Establish a safety supervision system for personnel, environment, and equipment, and issue real-time alarms once safety hazards are detected, thereby achieving efficient and intelligent supervision of on-site operation risks and personnel violations in oil and gas fields, and improving risk prevention capabilities.

[0070] A visual AI early warning management center will be established, where alarm results can be viewed and analyzed in real time, transforming the traditional post-verification mode under security protection into an intelligent management mode of pre-warning, thereby improving the intelligent management level of oil and gas fields.

[0071] Secondly, by unifying the access of surveillance cameras and other equipment in various operating areas of the oil and gas fields, full coverage of video visualization of the control area can be achieved. Video monitoring, recording and playback, video storage, cloud storage, alarm reporting, cascade sharing, etc. can also be carried out in key areas. It supports rapid access or integration into the superior monitoring center through standard API interfaces or standard protocols.

[0072] In addition, intelligent prevention and control links multiple subsystems such as access control and voice intercom to achieve multi-system perception of early warning and multi-system linkage of alarm response, meet the needs of regional intrusion detection and security risk detection, provide more reliable security protection, and improve management efficiency and operational efficiency.

[0073] Example 2

[0074] An AI analysis system for oil and gas field on-site videos includes: an acquisition module, a preprocessing module, a feature label acquisition module, a database, a comparison module, a warning instruction extraction module, a broadcast module and a storage module.

[0075] The acquisition module is used to acquire video images of the scene; the preprocessing module is used to preprocess the video images; the feature label acquisition module is used to obtain feature labels from the processed video images through the action model model; the database pre-stores a security feature label group; the comparison module is used to integrate the feature labels within a preset time period, obtain a feature label group, compare the feature label group with the database for similarity, and obtain risk information; the early warning instruction extraction module is used to obtain early warning instructions based on risk information; the broadcast module is used to broadcast the early warning instructions; the storage module is used to save the acquired video images to facilitate historical data analysis and long-term data retention.

[0076] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An AI analysis method for oil and gas field on-site video, characterized in that: The method comprises: Collect video images of the scene; Preprocessing the video image; The processed video image is passed through the action model to obtain feature labels; Integrate the feature tags within a preset time period to obtain a feature tag group, compare the feature tag group with the database for similarity, and obtain risk information; Obtain early warning instructions based on risk information and send them to the broadcast.

2. The AI ​​analysis method for oil and gas field on-site video according to claim 1 is characterized in that: The video images include: equipment operation video images, personnel video images and oil and gas field environment video images.

3. The AI ​​analysis method for oil and gas field on-site video according to claim 2 is characterized in that: For the device operation video image, the feature tag includes the appearance and operation status of the device; For the personnel video image, the feature tags include a helmet, work clothes, operator behavior and a face image; For the oil and gas field environment video image, the feature tag includes the surrounding environment.

4. The AI ​​analysis method for oil and gas field on-site video according to claim 1 is characterized in that: The preprocessing includes: denoising, contrast enhancement, and brightness adjustment.

5. The AI ​​analysis method for oil and gas field on-site video according to claim 1 is characterized in that: The action model model is a YOLO model based on target search and a residual classifier model for image recognition.

6. The AI ​​analysis method for oil and gas field on-site video according to claim 1 is characterized in that: The method further includes: saving the collected video images to facilitate historical data analysis and long-term data retention.

7. An oil and gas field on-site video AI analysis system, characterized in that: The system comprises: A collection module, used to collect video images of the scene; A preprocessing module, used for preprocessing the video image; A feature label acquisition module is used to obtain feature labels from the processed video images through the action model model; A database pre-stored with a set of security feature tags; A comparison module is used to integrate the feature tags within a preset time period, obtain a feature tag group, compare the feature tag group with the database for similarity, and obtain risk information; The early warning instruction extraction module is used to obtain early warning instructions based on risk information; The broadcast module is used to broadcast the warning instructions.

8. The oil and gas field on-site video AI analysis system according to claim 7, characterized in that: The system also includes: a storage module, which is used to store the collected video images to facilitate historical data analysis and long-term data retention.

Images