Method, system, equipment and medium for AI online detection of unsafe behaviors

Abstract

The invention relates to the technical field of petrochemical engineering construction, and discloses an AI online unsafe behavior detection method, system, device and medium, and the method comprises the steps: constructing a neural network detection model taking the operation behavior of a petrochemical construction site as a detection target; training the neural network detection model through operation behavior training data; acquiring operation image data of a petrochemical construction site in real time; inputting the operation image data into a trained neural network detection model for detection, and detecting whether an unsafe operation behavior exists or not; and if yes, sending a detection result to the terminal equipment. And the timeliness and accuracy of unsafe behavior detection on the construction operation site are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of petrochemical construction, and particularly to a method, system, device and medium for AI online detection of unsafe behaviors. Background Art

[0002] The production environment in petrochemical enterprises is very complex, including multiple links such as production processes, equipment, and raw materials. At the same time, in a production environment with high temperature, high pressure, flammable and explosive substances, there are many potential accident hazards, such as equipment aging, improper operation, and extensive management.

[0003] Based on the above problems, most of the existing methods are based on video surveillance, on-site manual supervision, etc., and provide guidance and corrective measures immediately when unsafe behaviors are found. The above methods have obvious deficiencies: the limited energy of people leads to blind spots in monitoring, and unsafe operation behaviors are not discovered in time; there are problems such as interference from subjective factors in human judgment, insufficient guardianship skills, etc., making it difficult to accurately judge and remind corrections for unsafe operation behaviors; the lack of corresponding electronic records makes it difficult to conduct accident investigations and accountability.

[0004] Therefore, it is necessary to invent a method for identifying unsafe operations in petrochemical enterprises to conduct all-round real-time detection of possible unsafe operation behaviors during the construction process, reduce the occurrence of dangerous accidents, and improve work efficiency. Summary of the Invention

[0005] The technical problem to be solved by the present invention is: in the prior art, during the construction process of petrochemical enterprises, the subjective judgment of manual supervision leads to inaccurate identification, inconsistent standards among different supervisors, the impact on timeliness due to limited human energy when multiple construction operations are carried out simultaneously, and the problem of lack of electronic records during the project construction process.

[0006] To solve the above technical problems, in the first aspect, the present invention provides a method for AI online detection of unsafe behaviors, the method comprising:

[0007] Constructing a neural network detection model with the operation behaviors at the petrochemical construction site as the detection target;

[0008] Training the neural network detection model with operation behavior training data;

[0009] Real-time obtaining operation image data at the petrochemical construction site;

[0010] Inputting the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if so, sending the detection result to the terminal device.

[0011] Further, the neural network detection model adopts the YOLO algorithm.

[0012] Further, the neural network detection model sequentially includes a CSPDarkNet53 neural network, an SPP+PAN network, and a YOLO-Head module.

[0013] Further, the real-time acquisition of the operation video data of the petrochemical construction site includes:

[0014] Collecting the operation image data through a terminal device and a camera.

[0015] Further, the detection of the operation behaviors at the petrochemical construction site at least includes: operation personnel statistics, correct wearing of work clothes detection, correct wearing of safety belts detection, detection of no safety warning during hoisting or presence of people in the warning area, detection of spark splashing during hot work, detection of gas cylinder tipping during hot work, sun protection detection of oxygen and acetylene gas cylinders, detection of using mobile phones in the explosion-proof area, detection of hot work personnel wearing protective masks, positive pressure identification of hot work gas cylinders, leak test bottles and flame arresters detection, detection of team leaders and supervisors wearing identification signs, detection of operators wearing goggles, and detection of steel plates laid under the outriggers of the lifting crane.

[0016] Further, the operation personnel statistics includes: detecting operation personnel from the operation video data and classifying and counting the operation personnel;

[0017] The correct wearing of work clothes detection includes: detecting whether the sleeves are rolled up or the upper clothes are open for personnel from the operation video data;

[0018] The correct wearing of safety belts detection includes: detecting whether the safety belt is hung low and used high from the operation video data;

[0019] The detection of no safety warning during hoisting or presence of people in the warning area includes: detecting whether there are people in the closed area connected by the coordinates of the barricades from the operation video data;

[0020] The detection of spark splashing during hot work includes: detecting the splashed sparks and the range of spark splashing from the operation video data;

[0021] The detection of gas cylinder tipping during hot work includes: detecting the placement angle of the gas cylinder from the operation video data;

[0022] The sun protection detection of oxygen and acetylene gas cylinders includes: detecting the gas cylinders in the image and detecting whether there is a covering above them from the operation video data;

[0023] The detection of using mobile phones in the explosion-proof area includes: detecting whether there are people using mobile phones in the explosion-proof area from the operation video data;

[0024] The detection of whether the hot work personnel wear sunscreen masks includes: determining the hot work personnel from the operation video data according to the spark position, and detecting whether they wear protective masks correctly;

[0025] The positive pressure identification of the hot work gas cylinder includes: detecting whether the pointer of the hot work gas cylinder is in the red area from the operation video data;

[0026] The detection of the leak test bottle and the flame arrester includes: detecting whether there are leak test bottles and flame arresters at the construction site from the operation video data;

[0027] The detection of whether the team leader and the supervisor of the operation wear identification includes: detecting the category of the personnel in the image and whether they wear armbands correctly from the operation video data;

[0028] The detection of whether the operation personnel wear safety goggles includes: detecting whether the operation personnel with operation actions wear safety goggles correctly from the operation video data;

[0029] The detection of the steel plates laid under the outriggers of the crane includes: detecting the outriggers of the crane in the image from the operation video data, and detecting whether iron plates are laid under the crane outriggers.

[0030] Further, the sending of the detection result to the terminal device includes:

[0031] If there is an unsafe behavior, send the image marking the unsafe behavior and the reason for determining the unsafe behavior to the terminal device.

[0032] In a second aspect, an AI online detection system for unsafe behaviors provided by an embodiment of the present invention includes:

[0033] A model construction module that constructs a neural network detection model with the operation behaviors at the petrochemical construction site as the detection target;

[0034] A model training module that trains the neural network detection model with operation behavior training data;

[0035] An image acquisition module that acquires the operation image data at the petrochemical construction site in real time;

[0036] An image detection module that inputs the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if there are, send the detection result to the terminal device.

[0037] In a third aspect, an embodiment of the present invention provides a computer device, which is characterized by comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the method for AI online detection of unsafe behaviors as described above is implemented.

[0038] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which is characterized in that the computer-readable storage medium includes a stored computer program; wherein, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for AI online detection of unsafe behaviors as described above.

[0039] Compared with the prior art, the method, system, device, and medium for AI online detection of unsafe behaviors in the embodiments of the present invention have the following beneficial effects: The present invention detects and trains various common unsafe operation behaviors in petrochemical enterprises, has diversity and adaptability, is applicable to different construction scenarios, and at the same time reduces the consumption of human resources and saves production costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 is a schematic flowchart of the method for AI online detection of unsafe behaviors provided by an embodiment of the present invention;

[0041] Figure 2 is a network structure diagram of the target detection algorithm for the method for AI online detection of unsafe behaviors provided by an embodiment of the present invention;

[0042] Figure 3 is a working process diagram of the system for AI online detection of unsafe behaviors provided by an embodiment of the present invention;

[0043] Figure 4 is a structural block diagram of the system for AI online detection of unsafe behaviors provided by an embodiment of the present invention;

[0044] Figure 5 is a structural diagram of the computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] The following combines the drawings and embodiments to further describe in detail the specific implementation manners of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0046] It should be noted that the step numbers in the text are only for the convenience of explaining specific embodiments and do not serve as a function of limiting the execution order of the steps. The method provided in this embodiment can be executed by a relevant server, and the following will take the server as the execution subject for illustration.

[0047] As Figure 1As shown in the figure, an embodiment of the present invention provides a method for AI online detection of unsafe behaviors, including steps S11 to S14;

[0048] Step S11, construct a neural network detection model with the operation behaviors at the petrochemical construction site as the detection target;

[0049] As Figure 2 shown in the figure, adopt the YOLO algorithm model to construct a neural network detection model that sequentially includes a CSPDarkNet53 neural network, an SPP+PAN network, and a YOLO-Head module;

[0050] During the petrochemical construction process, the requirements and specifications for the safety standards of construction are extremely high, and there are very high standards for the real-time nature of the information for on-site supervision. Therefore, in this embodiment, the Darknet53 network with a CSP structure is preferably used as the BackBone of the algorithm model, and its strong image feature extraction ability is used to preliminarily process the image information collected at the petrochemical construction site. This network structure ensures the extraction effect of target features when the model structure is of moderate size, ensuring both the detection speed to meet the real-time nature required for detection and relatively high detection accuracy. At the same time, it can screen the duplicate information in the image library collected at the petrochemical construction site, reducing unnecessary calculation times to achieve a more efficient and rapid detection effect.

[0051] The Neck module of the neural network detection model uses an SPP+PAN network to extract feature maps at different levels, thereby expanding the receptive field of the detection algorithm to fuse more effective feature information and making the detection of unsafe behaviors at the petrochemical construction site more comprehensive.

[0052] The Head part uses the YOLO algorithm to perform target detection on the image using multi-scale features, and in the video surveillance system, it can detect and track the preset unsafe behavior features of petrochemical construction operations in real time, accurately detect the abnormal information during the construction process from the video, and give its position and category information.

[0053] Step S12, train the neural network detection model with operation behavior training data;

[0054] Collect image information of different unsafe behaviors at the petrochemical construction site through various means such as monitoring, and input it into the neural network detection model constructed in step S11 to perform iterative training on the model:

[0055] Divide the sample data set into multiple parts, input a part of the data into the model each time to obtain a predicted value, calculate the loss value between it and the true value, and then execute the function in reverse and change the parameters of the neural network detection model in real time according to the obtained results; after multiple rounds of iteration, obtain the trained neural network detection model.

[0056] The training data used for training the neural network detection model of the present invention can not only collect a large amount of image information of unsafe behaviors, but also adopt the training data in the existing general object recognition process.

[0057] Step S13: Obtain the operation image data of the petrochemical construction site in real time;

[0058] After the petrochemical construction operation starts, collect the real-time image information of the on-site work through each terminal and camera arranged at the construction site, and transmit the collected image information to the server side or the AI platform, and the server or the AI platform performs AI online detection of unsafe behaviors.

[0059] Step S14: Input the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if so, send the detection result to the terminal device.

[0060] Input the image data collected through step S13 into the neural network detection model obtained by training in step S12 for detection; as Figure 3 shown, the detection of unsafe behaviors includes:

[0061] Operation personnel statistics: Detect the operation personnel from the operation video data, classify and count the operation personnel, and respectively count the number of petrochemical staff and other personnel;

[0062] Correct work clothes wearing detection: Detect the personnel from the operation video data and detect whether their sleeves are rolled up or their upper clothes are open;

[0063] Correct safety belt wearing detection: Detect from the operation video data whether the safety belt is hung low and used high;

[0064] Detection of no safety warning during lifting or someone in the warning area: Identify the barricades in the picture, connect the coordinate positions of different barricades into a closed warning area, and detect whether there are personnel in the warning area;

[0065] Detection of spark splashing during hot work: Detect the splashed sparks and the spark splashing range from the operation video data;

[0066] Detection of gas cylinder tipping during hot work: Detect the placement angle of the gas cylinder from the operation video data to determine whether the gas cylinder is tipped;

[0067] Sun protection detection of oxygen and acetylene gas cylinders: Detect the gas cylinders in the image from the operation video data and detect whether there is a covering above them;

[0068] Detection of using mobile phones in the explosion-proof area: Detect from the operation video data whether there are personnel using mobile phones in the explosion-proof area;

[0069] Detection of sunscreen masks worn by hot work personnel. Determine the hot work personnel from the operation video data based on the position of the sparks, and detect whether they are wearing protective masks correctly;

[0070] Positive pressure identification of hot work cylinders. Detect whether the pointer of the hot work cylinder is in the red area from the operation video data;

[0071] Detection of leak test bottles and flame arresters. Detect whether there are leak test bottles and flame arresters at the construction site from the operation video data;

[0072] Detection of identification badges worn by team leaders and supervisors. Detect the personnel category in the image and whether they are wearing armbands correctly from the operation video data;

[0073] Detection of safety goggles worn by operators. Detect whether the operators with operation actions are wearing safety goggles correctly from the operation video data;

[0074] Detection of steel plates laid under the outriggers of lifting cranes. Detect the outriggers of the lifting crane in the image from the operation video data, and detect whether iron plates are laid under the crane outriggers.

[0075] If unsafe operation behaviors including the above situations are detected, immediately send the images marking the unsafe behaviors and the reasons for judging the unsafe behaviors to the on-site terminal device, so that on-site supervisors can be notified in time and rectify and handle them, avoid potential safety hazards, and electronically archive the records. Of course, the detection of unsafe behaviors in the petrochemical construction process by the present invention is not limited to the above detection items. It is only used to explain that the present invention performs real-time identification and risk warning on video data during the construction process or acceptance process through a detection model based on a neural network. For other items in the petrochemical construction process, the video data can also be identified through this model.

[0076] Use the video and image information corresponding to the unsafe behaviors detected in the petrochemical construction process in step S14 as training data, incorporate it into the training data set, and optimize and iterate the neural network detection model obtained in step S12 to make it more suitable for the detection of the on-site construction environment.

[0077] This method constructs a target detection network model with fast recognition speed, high real-time detection network of mAP, and avoiding misidentification of objects by using the YOLO algorithm model; collects image information of unsafe behaviors at different petrochemical construction sites through various means such as monitoring, and iterates the target detection network model by using the training data in the existing general target recognition process to make it gradually more adaptable to the target detection at the construction operation site; identifies the image information collected at the petrochemical construction site to judge whether there are unsafe operation behaviors, and at the same time uses the illegal images determined to be unsafe behaviors obtained during the execution process as training data to gradually improve the neural network detection model.

[0078] As shown in Figure 4 the figure, an embodiment of the present invention further provides a system for AI online detection of unsafe behaviors, and the system includes:

[0079] A model construction module 21 that constructs a neural network detection model with the operation behaviors at a petrochemical construction site as the detection target;

[0080] Adopt the YOLO algorithm model to construct a neural network detection model that sequentially includes a CSPDarkNet53 neural network, an SPP+PAN network, and a YOLO-Head module;

[0081] A model training module 22 that trains the neural network detection model with operation behavior training data;

[0082] Collect the image information of unsafe behaviors at the petrochemical construction site, input it into the neural network detection model constructed by the model construction module 21 for model training. Each time a small batch of samples is extracted and input into the neural network detection model to obtain a predicted value, calculate the loss function value between its predicted value and the true value, then perform gradient backpropagation, and optimize and adjust the parameters of the model according to the obtained results, continuously update the model, and make it gradually adapt to the detection of unsafe behaviors at the petrochemical construction site.

[0083] An image acquisition module 23 that obtains the operation image data of the petrochemical construction site in real time;

[0084] After the petrochemical construction operation starts, the handheld terminals, other sensors, and cameras at the construction site start to collect the on-site construction video in real time and transmit the data to the server side.

[0085] An image detection module 24 that inputs the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if so, send the detection result to the terminal device.

[0086] Preferably, the video and image information corresponding to the unsafe behaviors at the petrochemical construction site identified by the specification recognition module 24 can be used as training data, incorporated into the training data set, and input into the neural network detection model obtained by training in the model training module for iterative training.

[0087] Next, taking the AI platform integrating the above system of the present invention as an example, the AI online detection process will be described in detail. As Figure 3 shown, a working process of the present invention is:

[0088] The on-site construction operation process starts, and at the same time, this system starts to work; during the operation process of the system, the information exchanges and pushes between the AI platform and the camera terminal are recorded throughout the process.

[0089] The system terminal camera collects the work ticket status information of different HSE (Health, Safety and Environment Management System) terminals at the site. The work ticket status includes start, pause, and close. After the collection, the work ticket status, work type, work ticket ID, HSE ID, and camera ID are transmitted to the target system: the AI platform.

[0090] Ten minutes after the work starts, the AI platform determines whether the pointer of the gas cylinder for hot work is in the red area by uploading the work type and image recognition of the hot work gas cylinder through the camera terminal, so as to judge whether the hot work gas cylinder is under positive pressure. At the same time, the status of the leak test bottle and the flame arrester in the image is recognized. And the recognition results, work type, work ticket ID, HSE ID, and camera ID are output together.

[0091] When problems occur during the construction work and the work stops, the status of the work ticket is switched to pause, and the camera transmits the collected work ticket status, work type, work ticket ID, HSE ID, and camera ID to the AI platform.

[0092] After the problem troubleshooting is completed, the work status of the work ticket is switched back to start, and the camera terminal returns the camera abnormal information, work type, work ticket ID, HSE ID, and camera ID to the AI platform.

[0093] During the work process, the supervisor can manually trigger a temporary detection at any time. At this time, the AI platform returns the work type, work ticket ID, HSE ID, camera ID, and the corresponding recognition results in real time: the number of staff, the number of others, the working status of each item, etc.

[0094] The AI platform analyzes the images and real-time identifies the following violations during the work process: staff not wearing work clothes correctly, making phone calls in the explosion-proof area, not wearing safety belts correctly, no safety warning during lifting or someone in the warning area, sparks splashing during hot work, the hot work gas cylinder toppling, no sun protection for oxygen or acetylene gas cylinders, hot work personnel not wearing protective masks, the team leader and the supervisor not wearing identification marks, work personnel not wearing goggles, and the outriggers of the crane not being padded with steel plates. After identifying the violation behavior, the violation reason, the violation image, work type, work ticket ID, HSE ID, and camera ID are output together immediately.

[0095] After the work is completed, the status of the work ticket is switched to close, and it is uploaded to the AI platform together with the work type, work ticket ID, HSE ID, and camera ID. The work process of the work ticket ends, and the system stops the operation of this work.

[0096] The technical features and effects of the system proposed in the embodiments of the present invention are the same as those of the method proposed in the embodiments of the present invention, and will not be elaborated here. Each module in the above system can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above modules.

[0097] As Figure 5 shown, the embodiments of the present invention further provide a computer device. The figure is a structural block diagram of a preferred embodiment of a computer device provided by the present invention. The computer device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for AI online detection of unsafe behaviors as described above.

[0098] Preferably, the computer program can be divided into one or more modules / units (such as computer program 1, computer program 2,...). The one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the computer device.

[0099] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor can also be any conventional processor. The processor is the control center of the computer device and connects various parts of the computer device through various interfaces and circuits.

[0100] The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory can be a high-speed random access memory, or a non-volatile memory, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory can also be other volatile solid-state storage devices.

[0101] It should be noted that the above computer device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 5 The structural block diagram is only an example of the computer device, and does not constitute a limitation on the computer device. It may include more or fewer components than those shown in the figure, or combine some components, or different components.

[0102] In summary, the embodiments of the present invention provide a method, a system, a device and a medium for AI online detection of unsafe behaviors, which detect and train various common unsafe operation behaviors in petrochemical enterprises, have diversity and adaptability, are applicable to different construction scenarios, and at the same time reduce the consumption of human resources and save production costs.

[0103] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the counting principle of the present invention, several improvements and replacements can still be made, and these improvements and replacements should also be regarded as the protection scope of the present invention.

[0104] In the description of this specification, the description of terms such as "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0105] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for AI online detection of unsafe behaviors, characterized in that, the method includes: constructing a neural network detection model with the operation behaviors at a petrochemical construction site as the detection target; training the neural network detection model with operation behavior training data; obtaining the operation image data at the petrochemical construction site in real time; inputting the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if so, sending the detection result to the terminal device.

2. The method for AI online detection of unsafe behaviors according to claim 1, characterized in that, the neural network detection model adopts the YOLO algorithm.

3. The method for AI online detection of unsafe behaviors according to claim 2, characterized in that, the neural network detection model sequentially includes a CSPDarkNet53 neural network, an SPP+PAN network, and a YOLO-Head module.

4. The method for AI online detection of unsafe behaviors according to claim 1, characterized in that, the obtaining the operation video data at the petrochemical construction site in real time includes: collecting the operation image data through a terminal device and a camera.

5. The method for AI online detection of unsafe behaviors according to claim 1, characterized in that, the detection of the operation behaviors at the petrochemical construction site at least includes: operation personnel statistics, correct work clothes wearing detection, correct safety belt wearing detection, detection of no safety warning during hoisting or someone in the warning area, detection of spark splashing during hot work, detection of gas cylinder tipping during hot work, sun protection detection of oxygen and acetylene gas cylinders, detection of using mobile phones in the explosion-proof area, detection of hot work personnel wearing protective masks, positive pressure identification of hot work gas cylinders, detection of leak test bottles and flame arresters, detection of team leaders and supervisors wearing identification marks during operation, detection of operation personnel wearing safety goggles, and detection of steel plates laid under the outriggers of lifting cranes.

6. The method for AI online detection of unsafe behaviors according to claim 5, characterized in that, the operation personnel statistics includes: detecting operation personnel from the operation video data and classifying and counting the operation personnel; the correct work clothes wearing detection includes: detecting whether the sleeves of the personnel are rolled up or the upper clothes are open from the operation video data; the correct safety belt wearing detection includes: detecting whether the safety belt is hung low and used high from the operation video data; the detection of no safety warning during hoisting or someone in the warning area includes: detecting whether there are personnel in the closed area connected by the coordinates of the barricades from the operation video data; the detection of spark splashing during hot work includes: detecting the splashed sparks and the spark splashing range from the operation video data; the detection of gas cylinder tipping during hot work includes: detecting the placement angle of the gas cylinder from the operation video data; the sun protection detection of oxygen and acetylene gas cylinders includes: detecting the gas cylinders in the image from the operation video data and detecting whether there is a covering above them; the detection of using mobile phones in the explosion-proof area includes: detecting whether there are personnel using mobile phones in the explosion-proof area from the operation video data; The detection of whether the hot work personnel wear sunscreen masks includes: determining the hot work personnel from the operation video data according to the spark position, and detecting whether they correctly wear protective masks; The positive pressure identification of the hot work gas cylinders includes: detecting from the operation video data whether the gauge needles of the hot work gas cylinders are in the red area; The detection of leak test bottles and flame arresters includes: detecting from the operation video data whether there are leak test bottles and flame arresters at the construction site; The detection of whether the team leader and the supervisor wear identification includes: detecting from the operation video data the types of personnel in the image and whether they correctly wear armbands; The detection of whether the operators wear safety goggles includes: detecting from the operation video data whether the personnel with operation actions correctly wear safety goggles; The detection of the steel plates laid under the outriggers of the crane includes: detecting the outriggers of the crane in the image from the operation video data, and detecting whether iron plates are laid under the crane outriggers.

7. The method for AI online detection of unsafe behaviors according to claim 1, wherein, The sending of the detection results to the terminal device includes: If there are unsafe behaviors, sending the images marking the unsafe behaviors and the reasons for determining the unsafe behaviors to the terminal device.

8. An AI online detection system for unsafe behaviors, wherein, The system includes: A model construction module for constructing a neural network detection model with the operation behaviors at the petrochemical construction site as the detection target; A model training module for training the neural network detection model with operation behavior training data; An image acquisition module for real-time acquisition of the operation image data at the petrochemical construction site; An image detection module for inputting the operation image data into the trained neural network detection model for detection to detect whether there are unsafe operation behaviors; if there are, sending the detection results to the terminal device.

9. A computer device, wherein, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the method for AI online detection of unsafe behaviors according to any one of claims 1 to 7.

10. A computer-readable storage medium, wherein, The computer-readable storage medium includes a stored computer program; wherein, when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the method for AI online detection of unsafe behaviors according to any one of claims 1 to 7.

Images