Early warning method and device for chemical safety monitoring

Abstract

The invention discloses an early warning method and device for chemical safety monitoring. The early warning method for chemical safety monitoring comprises the following steps: acquiring image data of a chemical production site in real time; determining a coarse-grained image and a fine-grained image according to the image data; inputting the coarse-grained image into a pre-trained neural network model, and outputting a corresponding risk type, the risk type including an article risk and an operator risk; determining a specific risk type and a corresponding weight according to the risk type and the fine-grained image; and according to the specific danger type and the corresponding weight, risk early warning is sent out.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an early warning method and device for chemical safety monitoring. Background Art

[0002] In modern industrial production, the chemical industry has always been a key area of ​​safety management due to its high risk and complexity. With the development of science and technology, image processing and computer vision technology are increasingly used in chemical production safety, which can help us identify and analyze safety hazards and improve production safety efficiency. At the same time, the field of chemical engineering equipment and technology is also constantly developing, and new equipment and technology can provide stronger technical support for chemical production safety.

[0003] However, in the prior art, alarms can only be issued for obvious safety hazards, which may lead to the risk of untimely identification and omission of hazards.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present invention is to provide an early warning method and device for chemical safety monitoring, aiming to solve the problem that the prior art can only issue alarms for obvious safety hazards, which is prone to the risk of untimely identification and omission of hazards.

[0006] To achieve the above object, the present invention provides an early warning method for chemical safety monitoring, comprising:

[0007] Acquire image data of chemical production sites in real time;

[0008] Determine a coarse-grained image and a fine-grained image according to the image data;

[0009] Input the coarse-grained image into a pre-trained neural network model, and output the corresponding risk type, wherein the risk type includes object risk and operator risk;

[0010] Determining a specific risk type and a corresponding weight according to the risk type and the fine-grained image;

[0011] Issue risk warnings based on specific hazard types and corresponding weights.

[0012] Preferably, in the early warning method for chemical safety monitoring, determining the specific hazard type and the corresponding weight according to the risk type and the fine-grained image includes:

[0013] When the risk type is operator risk, extracting external features of the operator in the fine-grained image;

[0014] Input each external feature into the pre-trained first convolutional neural network and the second convolutional neural network, and output the predicted features of the position of each joint point and the predicted features of the connection relationship between each joint point respectively;

[0015] According to the predicted features of the positions of the joint points and the predicted features of the connection relationships between the joint points, a modal feature representation of the operator is generated. The modal feature representation of the operator is as follows:

[0016]

[0017] Where, e is the vector of a frame in the image data;

[0018] F i is the synthesized modal feature matrix;

[0019] v i j Represents the modal feature elements of the operator;

[0020] According to the generated modal feature representation, the preset third convolutional neural network is input to output the operator's danger type and the corresponding weight.

[0021] Preferably, in the early warning method for chemical safety monitoring, before the step of inputting the generated modal feature representation into a preset third convolutional neural network and outputting the hazard type and the corresponding weight, the early warning method includes:

[0022] Acquire image sample data of chemical production site risk type of operator risk;

[0023] Determine a coarse-grained sample image and a fine-grained sample image according to the image sample data;

[0024] Extracting external features of operators in the fine-grained sample image;

[0025] According to the features of each external part in the fine-grained sample image, a pre-trained first convolutional neural network and a pre-trained second convolutional neural network are input to respectively output predicted sample features of the position of each joint point and predicted sample features of the connection relationship between each joint point;

[0026] Generate a modal sample feature representation of the operator according to the predicted sample features of the positions of the joint points and the predicted sample features of the connection relationships between the joint points;

[0027] The third convolutional neural network is trained with the generated modal sample feature representation as input and various specific hazard types and corresponding weights as output; wherein, back propagation is used to optimize the loss function;

[0028] Among them, the loss function is:

[0029]

[0030] in,

[0031] p ci is the true label c of sample i i probability;

[0032] q ci is the model's predicted probability for the sample;

[0033] W ci is the weight;

[0034] N is the number of samples.

[0035] Preferably, in the early warning method for chemical safety monitoring, the calculation formula of the operator's modal characteristic element is as follows:

[0036]

[0037] Among them, (Δx1, Δy1) is the activity corresponding to the intersection of a joint and the trunk;

[0038] (Δx2, Δy2) is the amount of activity corresponding to the intersection of another joint and the trunk;

[0039] t1-t2 is the duration of the activities of the two joints;

[0040] atan2() is the arc tangent in radians.

[0041] Preferably, in the early warning method for chemical safety monitoring, the external features of the operator include eye features;

[0042] The early warning method also includes:

[0043] determining a fatigue coefficient of the operator according to the eye characteristics;

[0044] According to the fatigue coefficient, the weight of the preset danger type is modified.

[0045] Preferably, in the early warning method for chemical safety monitoring, determining the fatigue coefficient of the operator according to the eye characteristics comprises:

[0046] Determine, based on the eye features, a change state of the operator's eyes within a preset time, wherein the change state includes an opening and closing frequency and a pupil diameter;

[0047] The fatigue coefficient is determined based on the frequency of eye opening and closing and the pupil diameter.

[0048] Preferably, in the early warning method for chemical safety monitoring, determining the fatigue coefficient according to the eye opening and closing frequency and the pupil diameter includes:

[0049] When the eye opening and closing frequency is less than a first preset value and the pupil diameter is greater than a second preset value, determining the eye fatigue value of the operator according to the eye opening and closing frequency and the pupil diameter;

[0050] Determine the fatigue coefficient based on the operator's eye fatigue value.

[0051] In order to achieve the above object, the present invention also provides an early warning device for chemical safety monitoring, comprising:

[0052] An acquisition unit, used for acquiring image data of a chemical production site in real time;

[0053] A determination unit, configured to determine a coarse-grained image and a fine-grained image according to the image data;

[0054] An output unit, used to input the coarse-grained image into a pre-trained neural network model, and output a corresponding risk type, wherein the risk type includes object risk and operator risk;

[0055] A generating unit, configured to determine a specific risk type and a corresponding weight according to the risk type and the fine-grained image;

[0056] The early warning unit issues risk warnings based on specific hazard types and corresponding weights.

[0057] In order to achieve the above object, the present invention further provides a computer device, characterized in that it includes:

[0058] at least one processor; and,

[0059] a memory communicatively connected to the at least one processor; wherein,

[0060] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned early warning method for chemical safety monitoring.

[0061] In order to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, the above-mentioned early warning method for chemical safety monitoring is implemented.

[0062] The present invention has at least the following beneficial effects:

[0063] The present invention acquires image data of a chemical production site in real time; determines a coarse-grained image and a fine-grained image based on the image data; inputs the coarse-grained image into a pre-trained neural network model, and outputs corresponding risk types, wherein the risk types include object risks and operator risks; determines specific hazard types and corresponding weights based on the risk types and the fine-grained images; and issues risk warnings based on the specific hazard types and corresponding weights, thereby accurately identifying risks and issuing reminders in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic flow chart of a first embodiment of the early warning method for chemical safety monitoring of the present invention;

[0065] Figure 2 It is a schematic flow chart of a second embodiment of the early warning method for chemical safety monitoring of the present invention;

[0066] Figure 3 A schematic flow chart of a third embodiment of the early warning method for chemical safety monitoring of the present invention;

[0067] Figure 4 FIG. 1 is a schematic diagram of a computer device according to a first embodiment of the present invention.

[0068] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0069] In the embodiments of the present invention, the term "and / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship.

[0070] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0071] In the embodiments of the present invention, the term "plurality" refers to two or more than two, and other quantifiers are similar.

[0072] In order to make the purpose, technical scheme and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. However, it can be understood by those skilled in the art that in the embodiments of the present invention, many technical details are proposed in order to enable the reader to better understand the present invention. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present invention can also be implemented. The division of the following embodiments is for the convenience of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0073] There are many risks in chemical scenes, mainly involving people and objects. Risks to operators include safety hazards caused by operators not wearing safety helmets and operators being too tired; risky items mainly include the risks of flammable, explosive, toxic and other items. These risks exist not only in the chemical production process, but also in multiple links such as storage and transportation. For example, problems such as storing flammable and explosive items too close to open flames.

[0074] In order to solve the above problems, the present embodiment relates to an early warning method for chemical safety monitoring, which can be applied to computer equipment. The computer equipment can be a desktop computer, a tablet computer, a notebook computer, and other electronic devices with data processing capabilities. In other embodiments, it can also be other electronic devices with data processing capabilities, and no specific restrictions are made here.

[0075] The following is an explanation of the implementation details of the early warning method for chemical safety monitoring according to the first embodiment of the present invention. The following content is only the implementation details provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0076] The specific process of this implementation is as follows: Figure 1 As shown, specifically including:

[0077] Step S100, acquiring image data of a chemical production site in real time;

[0078] Specifically, image data of the chemical site can be acquired in real time through a camera device.

[0079] Step S200, determining a coarse-grained image and a fine-grained image according to the image data;

[0080] It should be noted that coarse-grained images are a type of image with relatively low resolution, less details, and relatively broad image features. Coarse-grained images may have specific uses in some cases, such as when quickly previewing an image, performing preliminary analysis of an image to determine its general content, or processing it in an environment with limited resources. Compared with high-resolution, fine-grained images, coarse-grained images usually have a smaller amount of data and may be more efficient to process.

[0081] Fine-grained images are the opposite of coarse-grained images. Fine-grained images have high resolution and rich detail information, and can clearly show the subtle texture, color changes and other features of objects. For example, when classifying and identifying plants and animals, fine-grained images can capture very subtle differences between species, such as the texture of bird feathers, the shape and color gradient of flower petals, etc. Processing fine-grained images usually requires more complex algorithms and higher computing resources because these subtle feature differences must be accurately identified and distinguished.

[0082] The present invention roughly determines the risk type through a coarse-grained image, and then uses a fine-grained image to further determine the specific risk type and the corresponding weight, which can greatly reduce the amount of calculation.

[0083] Step S300, inputting the coarse-grained image into a pre-trained neural network model, and outputting corresponding risk types, wherein the risk types include object risk and operator risk;

[0084] It should be noted that the neural network model is used to roughly classify the coarse-grained images. If it is a material risk, it is necessary to further determine the specific material risk through fine-grained images, such as whether it is flammable, explosive, or toxic, etc. The specific risk can be determined according to the actual chemical scene.

[0085] Assuming it is an operator risk, fine-grained images are needed to further determine which specific operator risk it belongs to, such as operator safety protection risks (not wearing a safety helmet, etc.), improper operator operation, or operator excessive fatigue, etc. The specific risks can be determined based on the actual chemical scenario.

[0086] Step S400, determining a specific risk type and a corresponding weight according to the risk type and the fine-grained image;

[0087] It should be noted that the specific danger types include danger A, danger B, and danger C. The corresponding weights of danger A, danger B, and danger C are determined based on the fine-grained image. If the weight of a certain danger type is too large and greater than a certain threshold, it is considered that the danger has occurred and an early warning needs to be issued.

[0088] Furthermore, the weight corresponding to the specific danger type is determined by fine-grained images. The specific danger type can be pre-stored, and only the weight of each danger type needs to be calculated during calculation to determine which danger type is more inclined to belong to. The specific danger type can also be obtained by continuous learning and addition during the process of identifying sample images.

[0089] Step S500: Issue a risk warning based on the specific risk type and the corresponding weight.

[0090] It should be understood that whether a risk warning needs to be issued can be determined based on whether the weight corresponding to a specific risk type is greater than a preset value. When there are multiple specific risk types whose weights are greater than the preset value, one or more risk types with the highest weights corresponding to the risk types can be screened out for warning, or all the risk types whose weights are greater than the preset value can be screened out for warning. The specific determination can be based on the accuracy of the risk warning.

[0091] For example, if the hazard types include hazard A, hazard B, and hazard C, and the weight corresponding to hazard A is 10, the weight corresponding to hazard B is 8, and the weight corresponding to hazard C is 7, then an early warning can usually be issued for the existence of hazard A. You can also select one or more with high weights for early warning according to the set accuracy requirements.

[0092] The present invention acquires image data of a chemical production site in real time; determines a coarse-grained image and a fine-grained image based on the image data; inputs the coarse-grained image into a pre-trained neural network model, and outputs corresponding risk types, wherein the risk types include object risks and operator risks; determines specific hazard types and corresponding weights based on the risk types and the fine-grained images; and issues risk warnings based on the specific hazard types and corresponding weights, thereby accurately identifying risks and issuing reminders in a timely manner.

[0093] like Figure 2 As shown, Figure 2 The second embodiment of the early warning method for chemical safety monitoring of the present invention is illustrated. Based on the first embodiment, step S400 includes:

[0094] Step S410, when the risk type is operator risk, extracting external features of the operator in the fine-grained image;

[0095] It should be noted that the external features of the operator usually include features of the whole body, such as features of the arms and head, and so on.

[0096] Step S420, inputting each external feature into a pre-trained first convolutional neural network and a second convolutional neural network, and outputting prediction features of the position of each joint point and prediction features of the connection relationship between each joint point respectively;

[0097] It should be understood that two branches are used to predict the position of each joint point and the connection relationship between each joint point, for example, the first convolutional neural network is used to predict the position of each joint point, and the second convolutional neural network is used to predict the connection relationship between each joint point.

[0098] Human joints include shoulder joints, elbow joints, wrist joints, etc. The key points of the operator can be effectively extracted through the first convolutional neural network and the second convolutional neural network.

[0099] Step S430, generating a modal feature representation of the operator based on the predicted features of the positions of the joint points and the predicted features of the connection relationships between the joint points. The modal feature representation of the operator is as follows:

[0100]

[0101] Where, e is the vector of a frame in the image data;

[0102] F i is the synthesized modal feature matrix;

[0103] v i j Represents the operator's modal feature element.

[0104] Step S440, according to the generated modal feature representation, input a preset third convolutional neural network to output the operator's danger type and the corresponding weight.

[0105] It should be noted that, before step S440, the early warning method includes:

[0106] Step S510, obtaining image sample data of the chemical production site risk type of operator sexual risk;

[0107] Step S520, determining a coarse-grained sample image and a fine-grained sample image according to the image sample data;

[0108] Step S530, extracting the external features of the operator in the fine-grained sample image;

[0109] Step S540, according to the features of each external part in the fine-grained sample image, input the pre-trained first convolutional neural network and the second convolutional neural network, and respectively output the predicted sample features of the position of each joint point and the predicted sample features of the connection relationship between each joint point;

[0110] Among them, the calculation formula of the operator's modal characteristic element is as follows:

[0111]

[0112] Among them, (Δx1, Δy1) is the activity amount corresponding to the intersection of a joint and the trunk. Suppose the coordinates of joint 1 are (x1, y1), and the intersection of joint 1 and the trunk is O(x, y), then (Δx1, Δy1) = (x1-x, y1-y); (Δx2, Δy2) is the activity amount corresponding to the intersection of another joint and the trunk. Suppose the coordinates of joint 2 are (x2, y2), and the intersection of joint 2 and the trunk is O(x, y), then (Δx2, Δy2) = (x2-x, y2-y); t1-t2 is the duration of the activity of the two joints;

[0113] atan2() is the arc tangent in radians.

[0114] Step S550, generating a modal sample feature representation of the operator according to the predicted sample features of the positions of the joint points and the predicted sample features of the connection relationships between the joint points;

[0115] Step S560, taking the generated modal sample feature representation as input and various specific hazard types and corresponding weights as output, training the third convolutional neural network; wherein back propagation is used to optimize the loss function;

[0116] Among them, the loss function is:

[0117]

[0118] in,

[0119] p ci is the true label c of sample i i probability;

[0120] q ci is the model's predicted probability for the sample;

[0121] W ci is the weight;

[0122] N is the number of samples.

[0123] like Figure 3 As shown, Figure 3 The third embodiment of the early warning method for chemical safety monitoring of the present invention is illustrated, and the external features of the operator include eye features. Based on the first embodiment, the early warning method further includes:

[0124] Step S610, determining the fatigue coefficient of the operator according to the eye features;

[0125] Specifically, based on the eye characteristics, the change state of the operator's eyes within a preset time is determined, and the change state includes the opening and closing frequency and the pupil diameter; based on the eye opening and closing frequency and the pupil diameter, the fatigue coefficient is determined.

[0126] When the eye opening and closing frequency is less than a first preset value and the pupil diameter is greater than a second preset value, determining the eye fatigue value of the operator according to the eye opening and closing frequency and the pupil diameter;

[0127] Determine the fatigue coefficient based on the operator's eye fatigue value.

[0128] It should be noted that the eye opening and closing frequency is the frequency of eye blinking. When the operator is tired, the blinking frequency will decrease and the pupil diameter will increase significantly. Therefore, the fatigue coefficient of the operator can be evaluated by the blinking frequency and the pupil diameter. Among them, the first preset value and the second preset value can be determined according to the different types of work of the operator. For jobs that require high concentration and general concentration, different first preset values ​​and second preset values ​​can be set.

[0129] The eyes being in a closed or open state within a preset time is also a representation of fatigue of the subject. For example, if the number of openings counted by each frame image within the preset time is greater than a preset value, the eyes are judged to be in an open state; if the number of closings counted by each frame image within the preset time is greater than another preset value, the eyes are judged to be in a closed state, and the eyes are considered to be relatively fatigued. In other embodiments, whether the eyes are fatigued can be judged based on the opening and closing state of the eyes. When the eyes are in an open state, the eyes are considered to be not in a fatigued state; when the eyes are in a closed state, the eyes are considered to be in a fatigued state.

[0130] Whether the eyes are open can be determined based on the eye opening. For example, assuming that there are 20 frames of images within a preset time, when the eye opening in each frame is greater than 0, the value is set to 1. If 15 frames of images have an eye opening greater than 0, the total opening value is 15. When the total opening value is greater than the preset threshold, it is determined that the eyes are open, so that occasional blinks can be ignored.

[0131] It should be noted that the energy of operators varies significantly in different time periods. At night, due to poor visibility, operators need to concentrate more and feel more tired after working all day. Therefore, setting the fatigue correction value can fully take this into account.

[0132] In addition, the fatigue coefficient can be corrected according to different time periods. Usually, the fatigue degree of each stage will be different in the evening (19:00-22:00), late night (22:00-2:00 the next day), dawn (2:00-4:00), and dawn (4:00 to 6:00). The fatigue correction value can be set according to each stage, for example, the fatigue correction value at night is 7, and the fatigue correction value at midnight is 10; or the fatigue correction value can be increased proportionally from 19:00 to dawn the next day.

[0133] Step S620: modifying the weight of the preset risk type according to the fatigue coefficient.

[0134] In order to achieve the above-mentioned purpose, the present invention also provides an early warning device for chemical safety monitoring, which includes an acquisition unit, a determination unit, an output unit, a generation unit, and an early warning unit.

[0135] The acquisition unit is used to acquire image data of the chemical production site in real time;

[0136] The determining unit is used to determine a coarse-grained image and a fine-grained image according to the image data;

[0137] The output unit is used to input the coarse-grained image into a pre-trained neural network model and output the corresponding risk type, wherein the risk type includes object risk and operator risk;

[0138] The generating unit is used to determine a specific danger type and a corresponding weight according to the risk type and the fine-grained image;

[0139] The early warning unit issues risk warnings based on specific hazard types and corresponding weights.

[0140] It should be noted that each unit of the early warning device for chemical safety monitoring corresponds to the content of the above method, and the above method embodiment can be applied to the early warning device for chemical safety monitoring, which will not be described in detail here.

[0141] In order to achieve the above object, the present invention also provides a computer device, such as Figure 4 As shown, the computer device includes at least one processor 701; and a memory 702 that is communicatively connected to the at least one processor 701; wherein the memory 702 stores instructions that can be executed by the at least one processor 701, and the instructions are executed by the at least one processor 701 so that the at least one processor 701 can execute the above-mentioned early warning method for chemical safety monitoring.

[0142] The memory 702 and the processor 701 are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors 701 and the memory 702 together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor 701 is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor 701.

[0143] The processor 701 is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management and other control functions. The memory 702 can be used to store data used by the processor 701 when performing operations.

[0144] In order to achieve the above-mentioned purpose, the present invention provides a computer-readable storage medium storing a computer program, which implements the above-mentioned early warning method for chemical safety monitoring when executed by a processor.

[0145] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned implementation method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each implementation of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0146] Obviously, the embodiments described above are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can make other different forms of changes or modifications without creative work, which should fall within the scope of protection of the present invention.

Claims

1. A chemical safety monitoring early warning method, characterized in that: include: Acquire image data of chemical production sites in real time; Determine a coarse-grained image and a fine-grained image according to the image data; Input the coarse-grained image into a pre-trained neural network model, and output the corresponding risk type, wherein the risk type includes object risk and operator risk; Determining a specific risk type and a corresponding weight according to the risk type and the fine-grained image; Issue risk warnings based on specific hazard types and corresponding weights.

2. The early warning method for chemical safety monitoring according to claim 1, characterized in that: The determining of a specific danger type and a corresponding weight according to the risk type and the fine-grained image includes: When the risk type is operator risk, extracting external features of the operator in the fine-grained image; Input each external feature into the pre-trained first convolutional neural network and the second convolutional neural network, and output the predicted features of the position of each joint point and the predicted features of the connection relationship between each joint point respectively; According to the predicted features of the positions of the joint points and the predicted features of the connection relationships between the joint points, a modal feature representation of the operator is generated. The modal feature representation of the operator is as follows: Where, e is the vector of a frame in the image data; F i is the synthesized modal feature matrix; v i j Represents the modal feature elements of the operator; According to the generated modal feature representation, the preset third convolutional neural network is input to output the operator's danger type and the corresponding weight.

3. The early warning method for chemical safety monitoring according to claim 2, characterized in that: Before the step of inputting the generated modal feature representation into a preset third convolutional neural network and outputting the danger type and the corresponding weight, the early warning method includes: Acquire image sample data of chemical production site risk type of operator risk; Determine a coarse-grained sample image and a fine-grained sample image according to the image sample data; Extracting external features of operators in the fine-grained sample image; According to the features of each external part in the fine-grained sample image, a pre-trained first convolutional neural network and a pre-trained second convolutional neural network are input to respectively output predicted sample features of the position of each joint point and predicted sample features of the connection relationship between each joint point; Generate a modal sample feature representation of the operator according to the predicted sample features of the positions of the joint points and the predicted sample features of the connection relationships between the joint points; The third convolutional neural network is trained with the generated modal sample feature representation as input and various specific hazard types and corresponding weights as output; wherein, back propagation is used to optimize the loss function; Among them, the loss function is: in, p ci is the true label c of sample i i probability; q ci is the model's predicted probability for the sample; W ci is the weight; N is the number of samples.

4. The early warning method for chemical safety monitoring according to claim 2, characterized in that: The calculation formula of the operator's modal characteristic element is as follows: Among them, (Δx1, Δy1) is the activity corresponding to the intersection of a joint and the trunk; (Δx2, Δy2) is the amount of activity corresponding to the intersection of another joint and the trunk; t1-t2 is the duration of the activities of the two joints; atan2() is the arc tangent in radians.

5. The early warning method for chemical safety monitoring according to claim 2, characterized in that: The external features of the operator include eye features; The early warning method also includes: determining a fatigue coefficient of the operator according to the eye characteristics; According to the fatigue coefficient, the weight of the preset danger type is modified.

6. The early warning method for chemical safety monitoring according to claim 5, characterized in that: Determining the fatigue coefficient of the operator according to the eye characteristics includes: Determine, based on the eye features, a change state of the operator's eyes within a preset time, wherein the change state includes an opening and closing frequency and a pupil diameter; The fatigue coefficient is determined based on the frequency of eye opening and closing and the pupil diameter.

7. The early warning method for chemical safety monitoring according to claim 6, characterized in that: Determining the fatigue coefficient according to the eye opening and closing frequency and the pupil diameter includes: When the eye opening and closing frequency is less than a first preset value and the pupil diameter is greater than a second preset value, determining the eye fatigue value of the operator according to the eye opening and closing frequency and the pupil diameter; Determine the fatigue coefficient based on the operator's eye fatigue value.

8. An early warning device for chemical safety monitoring, characterized in that: include: An acquisition unit, used for acquiring image data of a chemical production site in real time; A determination unit, configured to determine a coarse-grained image and a fine-grained image according to the image data; An output unit, used to input the coarse-grained image into a pre-trained neural network model, and output a corresponding risk type, wherein the risk type includes object risk and operator risk; A generating unit, configured to determine a specific risk type and a corresponding weight according to the risk type and the fine-grained image; The early warning unit issues risk warnings based on specific hazard types and corresponding weights.

9. A computer device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the early warning method for chemical safety monitoring as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the early warning method for chemical safety monitoring as described in any one of claims 1 to 7 is implemented.

Images