Deep learning identification system for operation safety violation behaviors of oil field power distribution network

Abstract

The invention discloses an oil field power distribution network operation safety violation behavior deep learning recognition system. The system comprises a video acquisition module, a data preprocessing module, an improved YOLO model module, a violation behavior judgment module and a result output module. The video acquisition equipment is distributed in a key area of an oil field site and transmits video data to the server through a wireless network; the data preprocessing module adopts histogram equalization and sharpening operation to enhance the image contrast. The improved YOLO algorithm model is subjected to customized training, and behavior actions of operators and equipment can be accurately recognized. And the illegal behavior judgment module performs matching judgment on the result output by the model in combination with the safety specification of the oil field operation. According to the method, the difficult case mining technology and the multi-stage training strategy are adopted, the judgment capability of the model on illegal behaviors under complex conditions is improved, the OpenCV technology is combined, the capability of the system for coping with light changes and severe weather conditions is enhanced, and the recognition accuracy under different illumination intensities is ensured.

Description

Technical Field

[0001] The present invention relates to the field of safety monitoring technology, and in particular to a deep learning recognition system for safety violations in oilfield power distribution network operations. Background Art

[0002] As an important infrastructure in the process of oil extraction and processing, the safety and stability of the oilfield power distribution network are directly related to the smooth operation of the entire oilfield. However, due to the complexity and particularity of the oilfield operating environment, there are many safety hazards and violations in the power distribution network operation process.

[0003] With the continuous development of computer vision technology, video recognition technology has been widely used in many fields such as security monitoring, traffic management, industrial production, etc. The core of video recognition technology is to use algorithms to detect, classify and identify targets in the video to achieve monitoring and analysis of specific behaviors or events.

[0004] YOLO (You Only Look Once) is a deep learning-based object detection algorithm with the characteristics of fast detection speed and high accuracy. It regards the object detection task as a regression problem, directly predicts the category and location information of the object on multiple grid cells of an image or video, and realizes end-to-end object detection.

[0005] OpenCV is an open source computer vision library that provides a wealth of image processing and computer vision algorithms. In video processing, OpenCV can be used for video reading, preprocessing (such as grayscale, filtering, scaling, etc.), feature extraction and other operations, providing basic support for subsequent video recognition algorithms.

[0006] At the oilfield power distribution network site, the operation scenarios are complex and diverse, involving many professional equipment and operation procedures. Traditional video recognition algorithms cannot accurately distinguish between normal operation behaviors and illegal behaviors due to the lack of in-depth understanding and targeted optimization of the characteristics of oilfield power distribution network operations. For example, it is difficult to accurately distinguish between the normal inspection path of operators around the equipment and the illegal behavior of approaching dangerous areas, and normal equipment inspections may be misjudged as illegal intrusions. It is unable to provide reliable protection for the safety of oilfield power distribution network operations.

[0007] The light conditions in the oilfield power distribution network operating environment are complex and changeable. During the day, when the sun is shining directly, the equipment surface reflects strongly, forming highlights in some areas, while the shadows are dim. At night, the lighting conditions are limited and uneven, with a large number of alternating light and dark areas. Under such drastic changes in light, the models in traditional technologies cannot effectively adjust the recognition strategy, resulting in the loss of target details in the highlight area, such as the inability to see the facial features of the operator or the instrument readings of the equipment; in the low-light area, it is difficult to accurately identify the target object, such as the inability to identify the operator's movements in the shadow or the status of the equipment components, which greatly reduces the recognition effect and cannot stably and effectively identify the violations in the video, seriously affecting the reliability of the safety monitoring of the oilfield power distribution network operation.

[0008] The on-site background of the oilfield power distribution network is complex, with a large number of equipment, pipelines, protective facilities, warning signs, and natural environmental elements (such as grass, trees, etc.). Traditional models are easily disturbed by these complex backgrounds, misidentifying background objects as targets or ignoring real violation targets. For example, when identifying whether an operator is wearing a safety helmet, an object of similar color in the background may be misidentified as a safety helmet; when monitoring the operating status of the equipment, the logo on the equipment or the surrounding debris may be misidentified as a faulty part of the equipment, making it impossible to accurately identify the real violation and unable to provide effective technical support for the on-site safety management of the oilfield power distribution network. Summary of the invention

[0009] The technical problem to be solved by the present invention is to overcome the above technical defects and provide a deep learning identification system for safety violations in oilfield power distribution network operations.

[0010] In order to solve the above problems, the technical solution of the present invention is to include the following modules:

[0011] Video acquisition module, the video acquisition equipment is distributed in various key operation areas of the oil field power distribution network site, and is used to collect video data of the on-site operation;

[0012] A data preprocessing module, wherein the data preprocessing module performs preprocessing operations on the received video data, including image enhancement and size normalization, to improve data quality;

[0013] An improved YOLO model module, wherein the improved YOLO model module uses a YOLO algorithm that has undergone customized training and parameter optimization to analyze the preprocessed video data, identify the behavior of operators and equipment, and output corresponding recognition results;

[0014] A violation judgment module, which judges whether the current operation behavior is in violation of the regulations and determines the specific type of violation based on the recognition results of the improved YOLO model and the safety specifications and standards of the oilfield power distribution network on-site operation;

[0015] The result output module is used to output the violation judgment results in time, including real-time display of violation information in the monitoring center, and storage of violation records and violation pictures in the database and file picture server.

[0016] Furthermore, the data preprocessing module uses histogram equalization to enhance image contrast, highlight the features of operators and equipment, and enhance edge information through sharpening operations, and scales the video frame image to a uniform size required by the model.

[0017] Furthermore, the violation judgment module establishes a violation rule algorithm and matches the operator's actions, equipment status and other information in the recognition result with the rules in the algorithm to judge the violation.

[0018] Furthermore, the result output module uses a visual interface to display the on-site operation screen of the oilfield power distribution network in real time on the large screen of the monitoring center, highlights the identified violations with eye-catching logos, and marks the type of violation and the time of occurrence next to it.

[0019] Furthermore, it also includes a model training data enhancement alternative, a model structure alternative and a behavior sequence analysis module based on deep learning. The model training data enhancement alternative uses a generative adversarial network to generate synthetic data similar to real violation behavior data to expand the training data set and improve the generalization ability of the model. The model structure alternative adopts the EffectiveDet series of algorithms or combines a lightweight network structure with the YOLO algorithm to improve computing efficiency and reduce computing complexity. The behavior sequence analysis module based on deep learning uses a recurrent neural network or a long short-term memory network to model target behavior information of multiple consecutive frames, learns the spatiotemporal patterns of violations, and more accurately judges complex violations.

[0020] Furthermore, the video acquisition module transmits the video data to the server via a wireless method such as Wi-Fi or 4G / 5G network.

[0021] Further, the improved YOLO model module includes:

[0022] Difficult example mining technology, which is used to actively mine difficult samples that are easily misjudged or missed by the model during the training process, and give these samples higher weights;

[0023] A multi-stage training strategy that first performs pre-training on a general image dataset and then performs fine-tuning training on a specific dataset for oilfield power distribution network field operations;

[0024] Dataset construction: The data set construction collects a large amount of video data of oilfield power distribution network field operations, covering normal operation processes and various illegal behavior scenarios, and organizes professionals to annotate them;

[0025] Model training, based on the constructed data set, uses the stochastic gradient descent optimization algorithm combined with momentum and learning rate decay strategies to train the model. After multiple iterations of training, the performance of the model on the training set and validation set reaches a stable state.

[0026] Furthermore, the violations include common violations such as failure to wear insulating gloves, illegal use of open flames, and failure to wear safety belts. The objects involved are marked in detail, including insulating gloves, open flame sources, and safety belts.

[0027] Furthermore, the method also includes the step of performing image processing using OpenCV technology, wherein the step of performing image processing using OpenCV technology includes algorithms such as adaptive histogram equalization, light compensation, image denoising and defogging to cope with special light conditions and severe weather conditions at the oilfield power distribution network site.

[0028] The advantages of the present invention compared with the prior art are:

[0029] 1. The present invention provides a deep learning identification system for safety violations in oilfield power distribution network operations, which adopts difficult example mining technology and multi-stage training strategies. The difficult example mining technology enables the model to pay more attention to samples that are easily confused and difficult to identify during the training process, such as partially occluded operator actions under complex backgrounds, subtle changes in the early stages of equipment failures, etc. Multi-stage training first allows the model to learn basic image features and target recognition capabilities on general image data, and then conducts targeted training on specific data sets for on-site operations of oilfield power distribution networks, so that the model gradually adapts to complex operating behaviors and equipment feature changes in oilfield scenarios. In this way, the model's ability to distinguish violations in complex situations is further enhanced, ensuring that various types of violations can be accurately identified in actual applications, reducing the risk of increased safety management costs and reduced operating efficiency due to misjudgment, and improving the safety and reliability of oilfield power distribution network operations;

[0030] 2. The present invention provides a deep learning identification system for safety violations in oilfield power distribution network operations. With the help of OpenCV's powerful image processing function, it adopts adaptive histogram equalization, light compensation and other technologies for the special light conditions at the oilfield power distribution network site. In the case of drastic light changes (such as the transition from strong light during the day to weak light at night), it can effectively enhance the image contrast and highlight the characteristics of the target object, so that the model can accurately identify the status of operators and equipment under different light intensities, and reduce the recognition errors caused by light problems. Under severe weather conditions (such as sandstorms, rain and snow), image denoising, defogging and other algorithms are used to pre-process the video images to reduce the impact of weather factors on video quality and ensure that the model can stably identify violations in the video. For example, in sandstorm weather, the image denoising algorithm is used to remove the interference of sand and dust particles on the picture, so that the model can still accurately identify whether the operators comply with safety regulations and whether the equipment is operating normally, which enhances the adaptability and robustness of the system in the special environment of the oilfield and ensures the continuity and effectiveness of safety monitoring of oilfield power distribution network operations;

[0031] 3. The present invention provides a deep learning identification system for safety violations in oilfield power distribution network operations. It uses OpenCV's image segmentation, target extraction and other technologies, combined with the improved YOLO algorithm, to effectively distinguish between operators and equipment targets in the complex background of the oilfield power distribution network site. By modeling and suppressing the background, the impact of interferences such as protective facilities, warning signs, and vegetation on target recognition is reduced. For example, when identifying whether an operator has entered a dangerous area, it can accurately eliminate the interference of surrounding protective facilities and warning signs, and accurately determine the location of the personnel; when monitoring the operating status of the equipment, temporary signs and debris on the surface of the equipment can be ignored, and the state changes of key components of the equipment can be accurately identified, improving the model's recognition stability of violations in complex backgrounds, ensuring that the system can accurately monitor the real safety hazards at the oilfield power distribution network site, and effectively prevent the occurrence of safety accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of the deep learning identification system for safety violations in oilfield power distribution network operations of the present invention.

[0033] Figure 2 It is a structural diagram of the deep learning identification system for safety violations in oilfield power distribution network operations of the present invention. DETAILED DESCRIPTION

[0034] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices consistent with some aspects of the present disclosure as detailed in the appended claims.

[0035] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0036] like Figure 1 and Figure 2 As shown, this embodiment proposes a deep learning identification system for safety violations in oilfield power distribution network operations, including the following modules:

[0037] Video acquisition module: Video acquisition equipment is distributed in various key operation areas of the oilfield power distribution network site and is used to collect video data of on-site operations;

[0038] A data preprocessing module performs preprocessing operations on the received video data, including image enhancement and size normalization, to improve data quality;

[0039] Improve the YOLO model module. The improved YOLO model module uses the customized trained and parameter-optimized YOLO algorithm to analyze the pre-processed video data, identify the behavior of operators and equipment, and output the corresponding recognition results;

[0040] Illegal behavior judgment module, based on the recognition results of the improved YOLO model and the safety specifications and standards of oilfield power distribution network field operations, the illegal behavior judgment module judges whether the current operation behavior is illegal and determines the specific type of illegal behavior;

[0041] The result output module is used to output the violation judgment results in a timely manner, including real-time display of violation information in the monitoring center, and storage of violation records and violation images in the database and file image server.

[0042] Furthermore, the video acquisition module transmits the video data to the server via wireless methods such as Wi-Fi or 4G / 5G networks.

[0043] Furthermore, the data preprocessing module uses histogram equalization to enhance image contrast, highlight the features of operators and equipment, and enhances edge information through sharpening operations, scaling the video frame images to a uniform size required by the model.

[0044] Further, the improved YOLO model modules include:

[0045] Difficult example mining technology is used to actively mine difficult samples that are easily misjudged or missed by the model during the training process, and give these samples higher weights;

[0046] Multi-stage training strategy,The multi-stage training strategy is first pre-trained on a general image dataset, and then fine-tuned on a specific dataset for oilfield power distribution network field operations;

[0047] Dataset construction: The data set construction collects a large amount of video data of oilfield power distribution network field operations, covering normal operation processes and various illegal behavior scenarios, and organizes professionals to annotate them;

[0048] Model training, based on the constructed data set, uses the stochastic gradient descent optimization algorithm combined with momentum and learning rate decay strategies to train the model. After multiple iterations of training, the performance of the model on the training set and validation set reaches a stable state.

[0049] Furthermore, the violation judgment module establishes a violation rule algorithm and matches the operator's actions, equipment status and other information in the recognition results with the rules in the algorithm to judge the violation.

[0050] Furthermore, common violations such as failure to wear insulating gloves, illegal use of open flames, and failure to wear safety belts are marked in detail for the objects involved, which include insulating gloves, open flame sources, and safety belts.

[0051] Furthermore, the result output module uses a visual interface to display the on-site operation screen of the oilfield power distribution network in real time on the large screen of the monitoring center, highlights the identified violations with eye-catching logos, and marks the type of violation and the time of occurrence next to it.

[0052] Furthermore, the step of performing image processing using OpenCV technology is also included. The step of performing image processing using OpenCV technology includes algorithms such as adaptive histogram equalization, light compensation, image denoising and defogging to cope with special light conditions and severe weather conditions at the oil field power distribution network site.

[0053] Furthermore, it also includes model training data enhancement alternatives, model structure alternatives and a behavior sequence analysis module based on deep learning. The model training data enhancement alternative uses a generative adversarial network to generate synthetic data similar to real violation data to expand the training data set and improve the generalization ability of the model. The model structure alternative adopts the EffectiveDet series of algorithms or combines a lightweight network structure with the YOLO algorithm to improve computing efficiency and reduce computing complexity. The behavior sequence analysis module based on deep learning uses a recurrent neural network or a long short-term memory network to model target behavior information of multiple consecutive frames, learns the spatiotemporal patterns of violations, and more accurately judges complex violations.

[0054] Furthermore, the enhanced alternative uses a GAN network to generate synthetic data similar to real violation data to expand the training data set. The generator in the GAN network learns the distribution of real data and generates new samples, while the discriminator is used to distinguish between real samples and generated samples. Through continuous training, the generator can generate high-quality synthetic data, increase data diversity, improve the generalization ability of the model, obtain labeled data from other related fields (such as traffic monitoring data in other cities, security data of similar scenarios, etc.), and use transfer learning technology to transfer the knowledge of these data to the video violation identification task of the present invention. By pre-training the model on the source data set and then fine-tuning it on the target data set (the violation data set required by the present invention), the need for a large amount of labeled data can be reduced, while speeding up the model training and improving performance.

[0055] Furthermore, the EffectiveDet series of algorithms can effectively identify video violations by adjusting the network parameters of EffectiveDet (such as depth and width factors) and adding specific violation detection heads, and to a certain extent reduce the consumption of computing resources and improve the processing speed.

[0056] Furthermore, lightweight network structures (such as MobileNet, ShuffleNet, etc.) are combined with the YOLO algorithm. These lightweight networks significantly reduce the computational complexity and parameter count of the model while ensuring a certain feature extraction capability. By building a YOLO model based on a lightweight network and conducting targeted training, video violation identification can be achieved on resource-constrained devices (such as mobile devices and edge computing devices) while maintaining good performance.

[0057] Furthermore, the behavior sequence analysis based on deep learning uses recurrent neural network (RNN) or long short-term memory network (LSTM) to model the target behavior information of multiple consecutive frames and learn the spatiotemporal patterns of illegal behaviors. By analyzing the dynamic changes of the behavior sequence, complex illegal behaviors (such as continuous illegal lane changes of vehicles at intersections, pedestrians wandering on the road, and other illegal behaviors that may involve time series) can be judged more accurately.

[0058] For specific use, refer to Figure 1 and Figure 2 As shown in the figure, the high-definition camera collects the on-site operation video and transmits the video data to the back-end server in real time through the 4G network. After receiving the video data, the data preprocessing module performs histogram equalization on each frame of the image to enhance the contrast, making the color and outline of the operator's work clothes, safety helmets and equipment clearer. Finally, the image size is uniformly adjusted to 640×380 pixels for input into the improved YOLO model.

[0059] Load the trained improved YOLO algorithm model, which has been optimized in parameters and adds markings for objects related to the identification of violations in the oilfield power distribution network. Through the analysis of a large amount of video data of on-site operations of the oilfield power distribution network, the model parameters are carefully adjusted to make it more suitable for the target detection task in this specific scenario. At the same time, objects (such as insulating gloves, open flame sources, seat belts, etc.) involved in common violations such as not wearing insulating gloves, illegal use of open flames, and not wearing seat belts are marked in detail, so that the model can more accurately locate and judge the status of these key objects during the identification process, effectively improving the ability to identify violations of the oilfield power distribution network.

[0060] The preprocessed video frames are input into the model, and the model outputs the detected target information, including operators, various tools and equipment, and predicts their categories, locations and confidences. For example, the model detects an operator operating in front of a power distribution cabinet, and its category is "operator", the confidence is 0.90, and the location information is the bounding box coordinates (x1, y1, x2, y2).

[0061] The violation judgment module makes judgments based on the model output and the pre-set safety rules for oilfield power distribution network operations. If the model detects that the operator is not wearing a safety helmet when operating the distribution cabinet, and the confidence level is greater than 0.85, the system determines it as a violation, and the violation type is "not wearing a safety helmet to operate".

[0062] The result output module immediately displays the area where the violation occurred in the monitoring center, and uses a yellow warning box to explain the type of illegal operation. The system records the detailed information of the violation (including the time of violation, the location of the distribution cabinet area, the information of the violator (assuming that the specific employee is associated with the face recognition), the type of violation, etc.) in the database.

[0063] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0064] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0065] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A deep learning identification system for safety violations in oilfield power distribution network operations, characterized by: Includes the following modules: Video acquisition module, the video acquisition equipment is distributed in various key operation areas of the oil field power distribution network site, and is used to collect video data of the on-site operation; A data preprocessing module, wherein the data preprocessing module performs preprocessing operations on the received video data, including image enhancement and size normalization, to improve data quality; An improved YOLO model module, wherein the improved YOLO model module uses a YOLO algorithm that has undergone customized training and parameter optimization to analyze the preprocessed video data, identify the behavior of operators and equipment, and output corresponding recognition results; A violation judgment module, which judges whether the current operation behavior is in violation of the regulations and determines the specific type of violation based on the recognition results of the improved YOLO model and the safety specifications and standards of the oilfield power distribution network on-site operation; The result output module is used to output the violation judgment results in time, including real-time display of violation information in the monitoring center, and storage of violation records and violation pictures in the database and file picture server.

2. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The video acquisition module transmits the video data to the server via a wireless method such as Wi-Fi or 4G / 5G network.

3. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The data preprocessing module uses histogram equalization to enhance image contrast, highlight the features of operators and equipment, and enhance edge information through sharpening operations. The video frame image is scaled to a uniform size required by the model.

4. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized in that: The improved YOLO model module includes: Difficult example mining technology, which is used to actively mine difficult samples that are easily misjudged or missed by the model during the training process, and give these samples higher weights; A multi-stage training strategy that first performs pre-training on a general image dataset and then performs fine-tuning training on a specific dataset for oilfield power distribution network field operations; Dataset construction: The data set construction collects a large amount of video data of oilfield power distribution network field operations, covering normal operation processes and various illegal behavior scenarios, and organizes professionals to annotate them; Model training, based on the constructed data set, uses the stochastic gradient descent optimization algorithm combined with momentum and learning rate decay strategies to train the model. After multiple iterations of training, the performance of the model on the training set and validation set reaches a stable state.

5. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The violation judgment module establishes a violation rule algorithm and matches the operator's actions, equipment status and other information in the recognition result with the rules in the algorithm to judge the violation.

6. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The violations include common violations such as failure to wear insulating gloves, illegal use of open flames, and failure to wear safety belts. The objects involved are marked in detail, including insulating gloves, open flame sources, and safety belts.

7. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The result output module uses a visual interface to display the on-site operation screen of the oilfield power distribution network in real time on the large screen of the monitoring center, highlights the identified violations with eye-catching logos, and marks the type of violation and the time of occurrence next to it.

8. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 1 is characterized by: The method also includes the step of performing image processing using OpenCV technology, wherein the step of performing image processing using OpenCV technology includes algorithms such as adaptive histogram equalization, light compensation, image denoising and defogging to cope with special light conditions and severe weather conditions at the oil field power distribution network site.

9. The deep learning identification system for safety violations in oilfield power distribution network operations according to claim 8 is characterized by: It also includes a model training data enhancement alternative, a model structure alternative, and a behavior sequence analysis module based on deep learning. The model training data enhancement alternative uses a generative adversarial network to generate synthetic data similar to real violation behavior data to expand the training data set and improve the generalization ability of the model. The model structure alternative uses the EfficientDet series of algorithms or combines a lightweight network structure with the YOLO algorithm to improve computing efficiency and reduce computing complexity. The behavior sequence analysis module based on deep learning uses a recurrent neural network or a long short-term memory network to model target behavior information of multiple consecutive frames, learns the spatiotemporal patterns of violations, and more accurately judges complex violations.

Images