Image data-based power work site violation recognition system and method

Abstract

The application relates to the technical field of violation detection, and particularly discloses a power operation site violation identification system and method based on image data, the method comprising determining installation parameters and collection parameters of a collection device; receiving image data acquired by each collection device, identifying the image data, and determining personnel features; identifying the personnel features, and determining a violation probability; and when the violation probability reaches a preset probability threshold, sending a close-range image acquisition instruction to a preset unmanned aerial vehicle end. The application installs a collection device, acquires image data based on the collection device, extracts personnel features of workers in the image data, identifies the personnel features based on a trained neural network model, determines the risk rate of the workers, and when the risk rate is high, acquires a close-range image or video through the preset unmanned aerial vehicle end, and further identifies the workers by a manual end.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of violation detection, and particularly relates to a power operation site violation identification system and method based on image data. BACKGROUND

[0002] A power operation site is a common engineering site, and front-line personnel mainly engage in daily maintenance operations and do not engage in production activities; because the consequences caused by risks of power equipment will be very serious, many rules and regulations are set for the daily maintenance operations of front-line personnel.

[0003] Personnel who have worked in the front line are well aware that these rules and regulations are very cumbersome and sometimes bring great inconvenience, and under the existing technical level, the probability of problems in daily maintenance operations is extremely low, so when working personnel perform daily maintenance operations, they are prone to habitual violations in pursuit of convenience; if these habitual violations are ticket violations and the like, they can be accepted, and at least will not affect the safety of the working personnel, and if they are habitual violations in the operation process, the risk of the working personnel will be greatly increased; for a single working personnel, the probability of problems is still very small, but when there are many power operation sites, some working personnel will have problems in the working process.

[0004] Therefore, how to monitor and identify the operation process of the working personnel is a technical problem that the technical solution of the present application wants to solve. SUMMARY

[0005] The present application aims to provide a power operation site violation identification system and method based on image data to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solution:

[0007] A power operation site violation identification method based on image data, the method comprising:

[0008] Obtaining a building model of a power operation site, determining a sub-region containing a function label according to the building model, determining installation parameters and collection parameters of a collection device based on the sub-region;

[0009] Receiving image data obtained by each collection device, identifying the image data to determine personnel features; the personnel features are used to represent the working state of the working personnel;

[0010] Querying the function label of the sub-region corresponding to the collection device, identifying the personnel features based on the function label to determine a violation probability;

[0011] The violation probability is compared with a preset probability threshold, and when the violation probability reaches the preset probability threshold, a close-range image acquisition instruction is sent to a preset drone end.

[0012] As a further scheme of the present application: the step of acquiring the building model of the power operation site, determining the sub-region containing the function label according to the building model, and determining the installation parameters and collection parameters of the collection device based on the sub-region comprises:

[0013] Acquiring a record model of the power operation site, and opening a model adjustment port; the top of the record model is in a hidden state;

[0014] Receiving an adjustment instruction of a worker based on the model adjustment port, and correcting the record model based on the adjustment instruction to obtain a building model of the power operation site;

[0015] Receiving a user inputted segmentation signal based on the building model, segmenting the building model according to the segmentation signal to obtain a sub-region, and receiving a function label inputted by a user based on the sub-region;

[0016] Determining the installation parameters and collection parameters of the collection device according to the sub-region.

[0017] As a further scheme of the present application: the step of determining the installation parameters and collection parameters of the collection device according to the sub-region comprises:

[0018] Acquiring the area range of the sub-region, and determining the type, quantity and installation position of the collection device based on the area range;

[0019] Reading the function label of the sub-region, and querying a risk parameter in a preset risk parameter library according to the function label; wherein, the risk parameter comprises a risk type and a risk position; the risk parameter in the risk parameter library is indexed by the function label;

[0020] Determining the collection angle of the collection device according to the installation position and the risk position;

[0021] Setting the collection clarity of the collection device as a value decreasing over time; when the collection device receives audio information, resetting the collection clarity according to the audio information.

[0022] As a further scheme of the present application: the step of resetting the collection clarity according to the audio information when the collection device receives the audio information comprises:

[0023] When the collection device receives the audio information, inputting the audio information into a trained audio processing model, filtering the clutter in the audio information to obtain to-be-recognized audio;

[0024] Convert the to-be-identified audio into audio text, extract words in the audio text, and match the words with preset keywords;

[0025] Determine the number of occurrences of each keyword according to the matching result, and determine the feature matrix according to the number of occurrences of each keyword;

[0026] Input the feature matrix into the trained judgment model to determine whether to reset the collection clarity to the highest value.

[0027] As a further scheme of the present application: the step of receiving image data acquired by each acquisition device, identifying the image data, and determining personnel features comprises:

[0028] Receiving image data acquired by each acquisition device, color identifying the image data, and determining the safety helmet region and its color value according to the color identification result;

[0029] Determine the personnel type and the personnel contour according to the safety helmet region and its color value;

[0030] Obtain the center point of the personnel contour and calculate the distance between each personnel contour, and generate graph data according to the center point and the distance; the node of the graph data is determined by the personnel type, and the edge of the graph data is determined by the distance;

[0031] The graph data is used as the personnel feature.

[0032] As a further scheme of the present application: the step of querying the function label of the sub-region corresponding to the acquisition device, identifying the personnel feature based on the function label, and determining the probability of violation comprises:

[0033] Query the function label of the sub-region corresponding to the acquisition device, and obtain the standard working image according to the function label;

[0034] Extract the standard personnel feature in the standard working image to generate a sample set;

[0035] Train a neural network model based on the sample set;

[0036] Identify the acquired personnel feature according to the trained neural network model to determine the probability of violation.

[0037] As a further scheme of the present application: the step of comparing the probability of violation with a preset probability threshold, and sending a close-range image acquisition instruction to a preset unmanned aerial vehicle end when the probability of violation reaches the preset probability threshold comprises:

[0038] Compare the probability of violation with a preset probability threshold, and read the personnel type and the personnel contour when the probability of violation reaches the preset probability threshold.

[0039] determine the personnel level according to the personnel type, query the personnel contour corresponding to the highest personnel level;

[0040] take the personnel contour corresponding to the highest personnel level as an image acquisition center, and acquire a close-range image containing time information based on a preset detection path.

[0041] The technical scheme of the present application further provides a system for identifying violations in a power operation site based on image data, which comprises:

[0042] A parameter determination module is configured to acquire a building model of the power operation site, determine a sub-region containing a function label according to the building model, and determine installation parameters and acquisition parameters of a collection device based on the sub-region.

[0043] A feature extraction module is configured to receive image data acquired by each collection device, identify the image data, and determine personnel features; the personnel features are used to represent the working state of workers.

[0044] A probability determination module is configured to query the function label of the sub-region corresponding to the collection device, identify the personnel features based on the function label, and determine a violation probability.

[0045] A close-range detection module is configured to compare the violation probability with a preset probability threshold, and send a close-range image acquisition instruction to a preset unmanned aerial vehicle end when the violation probability reaches the preset probability threshold.

[0046] As a further scheme of the present application, the parameter determination module comprises:

[0047] A port opening unit is configured to acquire a record model of the power operation site, and open a model adjustment port; the top of the record model is in a hidden state.

[0048] A model correction unit is configured to receive an adjustment instruction of a worker based on the model adjustment port, correct the record model based on the adjustment instruction, and obtain a building model of the power operation site.

[0049] A model segmentation unit is configured to receive a segmentation signal input by a user based on the building model, segment the building model according to the segmentation signal, obtain a sub-region, and receive a function label input by the user based on the sub-region.

[0050] An execution unit is configured to determine installation parameters and acquisition parameters of a collection device according to the sub-region.

[0051] As a further scheme of the present application, the feature extraction module comprises:

[0052] The color recognition unit is configured to receive image data acquired by each acquisition device, recognize colors of the image data, and determine a safety helmet region and a color value of the safety helmet region according to a color recognition result.

[0053] The personnel marking unit is configured to determine a personnel type and a personnel contour according to the safety helmet region and the color value of the safety helmet region.

[0054] The graph data generation unit is configured to acquire a center point of the personnel contour, calculate distances between the personnel contours, and generate graph data according to the center point and the distances; a node of the graph data is determined by the personnel type, and an edge of the graph data is determined by the distances.

[0055] The graph data storage unit is configured to store the graph data as personnel features.

[0056] Compared with the prior art, the present application has the following beneficial effects: the present application installs an acquisition device, acquires image data based on the acquisition device, extracts personnel features of workers in the image data, identifies the personnel features based on a trained neural network model, and determines a risk rate of the workers; when the risk rate is high, a close-range image or video is acquired by a preset unmanned aerial vehicle end, and the workers are further identified by a manual end. BRIEF DESCRIPTION OF DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only some embodiments of the present application.

[0058] Figure 1 The flowchart of the method for identifying violations in a power operation site based on image data.

[0059] Figure 2 The first sub-flowchart of the method for identifying violations in a power operation site based on image data.

[0060] Figure 3 The second sub-flowchart of the method for identifying violations in a power operation site based on image data.

[0061] Figure 4 The third sub-flowchart of the method for identifying violations in a power operation site based on image data.

[0062] Figure 5 The fourth sub-flowchart of the method for identifying violations in a power operation site based on image data.

[0063] Figure 6 The composition structure block diagram of the system for identifying violations in a power operation site based on image data.

[0064] Figure 7 A component structure block diagram of a parameter determination module in an image data-based power operation site violation recognition system.

[0065] Figure 8 A component structure block diagram of a feature extraction module in an image data-based power operation site violation recognition system. DETAILED DESCRIPTION

[0066] In order to make the technical problems to be solved by the present application, technical solutions and beneficial effects clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0067] Embodiment 1

[0068] Figure 1 A flow block diagram of an image data-based power operation site violation recognition method, in the embodiment of the present application, an image data-based power operation site violation recognition method, the method comprises steps S100 to S400:

[0069] Step S100: acquiring a building model of a power operation site, determining a sub-region containing a function label according to the building model, determining installation parameters and collection parameters of a collection device based on the sub-region;

[0070] The power operation site is an area, which generally does not change greatly; different areas have different functions, different functions correspond to different daily maintenance operations, and different daily maintenance operations have different regulations, such as, during most of the maintenance operations, the equipment should be locked and signed, and during the aerial operation, a safety belt should be provided; therefore, when the building model is divided, the division should be based on the function; after the sub-region is divided, the installation parameters and the collection parameters of each collection device can be determined.

[0071] Step S200: receiving image data acquired by each collection device, identifying the image data to determine personnel features; the personnel features are used to represent the working state of the workers;

[0072] The collection device needs to have image acquisition function and data communication function, according to the image data acquired by the collection device, some personnel features can be extracted, which can reflect the working state of the workers; for the maintenance operation of the power operation site, the workers generally do not change greatly, and their working methods and positions and the like should be similar; the working methods and positions and the like of the workers are detected to obtain the personnel features of the workers.

[0073] Step S300: query the function label of the sub-area corresponding to the collection device, identify the personnel feature based on the function label, and determine the violation probability;

[0074] From the above, it can be seen that the maintenance operation in the sub-area corresponding to the different function labels can be preset to be collected, in other words, some standard personnel features can be recorded according to the function label, and when a new personnel feature is received, comparison is performed, and the violation probability can be determined; of course, the accuracy of this determination process is not high, and many personnel features without violation behavior may also be identified as having violation, at this time, further violation identification is required; from the perspective of safety, this is actually a good thing.

[0075] Step S400: compare the violation probability with a preset probability threshold, and when the violation probability reaches the preset probability threshold, send a close-range image acquisition instruction to a preset unmanned aerial vehicle end;

[0076] When the violation probability reaches a certain degree, it means that the position of the worker is likely to have a violation phenomenon, at this time, further detailed images need to be acquired for subsequent processing, and the subsequent processing is mainly manual identification; it is worth mentioning that the unmanned aerial vehicle end needs to have a position adjustment function so that it can move to a specified position to acquire detailed images.

[0077] Figure 2 The first sub-flow block diagram of the method for identifying violation in the power operation site based on image data, the steps of acquiring the building model of the power operation site, determining the sub-area containing the function label based on the building model, and determining the installation parameters and collection parameters of the collection device based on the sub-area include steps S101 to S104:

[0078] Step S101: acquire a record model of the power operation site, and open a model adjustment port; the top of the record model is in a hidden state;

[0079] Step S102: receive an adjustment instruction of a worker based on the model adjustment port, correct the record model based on the adjustment instruction, and obtain a building model of the power operation site;

[0080] Step S103: receive a user inputted segmentation signal based on the building model, segment the building model according to the segmentation signal to obtain a sub-area, and receive a function label inputted by a user based on the sub-area;

[0081] Step S104: determine the installation parameters and collection parameters of the collection device according to the sub-area.

[0082] The above describes the generation process of the building model in detail. First, a record model of the power operation site is obtained, which can be directly extracted from the database during the design process. During the extraction of the record model, all the top parts need to be hidden because the collection equipment is generally installed on the top and will not be blocked by the top. The model adjustment port is also inserted in the record model, and the staff can adjust the record model according to the actual situation to make the record model more consistent with the actual situation. Then, the building model receives the splitting signal input by the user, and the building model is split into different sub-regions according to the function of the splitting signal. Finally, the installation parameters and collection parameters of the collection equipment are determined according to the sub-regions obtained by splitting.

[0083] Further, the step of determining the installation parameters and collection parameters of the collection equipment according to the sub-regions comprises:

[0084] Obtaining the area range of the sub-region, determining the model, number and installation position of the collection equipment based on the area range;

[0085] Reading the function label of the sub-region, and querying the risk parameter in the preset risk parameter library according to the function label; wherein the risk parameter comprises a risk type and a risk position; the risk parameter in the risk parameter library is indexed by the function label;

[0086] Determining the collection angle of the collection equipment according to the installation position and the risk position;

[0087] Setting the collection clarity of the collection equipment to a value that decreases over time; when the collection equipment receives audio information, resetting the collection clarity according to the audio information.

[0088] The sub-region has an area range, and the installation height of the collection equipment is generally determined, that is, it is installed on the top in the actual site. The installation requirement is that the union of the collection equipment range should be greater than the area range, so the wide angle, number and installation position of the collection equipment need to be set. The wide angle is related to the model. Further, there are equipment areas and open activity areas in the sub-region, and the probability of risk in the equipment area is greater, so the daily working angle of the collection equipment should be aimed at the equipment area. Specifically, for the clarity of the collection equipment, since the power operation site is working on its own most of the time, at this time, the collection equipment mainly monitors the equipment to prevent accidents such as fire, and the identification process of the fire and other accidents is not difficult, which is very obvious in the image data, so the clarity does not need to be required. Therefore, when there is no staff in the collection range of the collection equipment, the clarity of the equipment can be lowered to save energy and improve energy utilization.

[0089] As a preferred embodiment of the technical scheme of the present application, when the audio information is received by the collection device, the step of resetting the collection clarity according to the audio information comprises:

[0090] When the audio information is received by the collection device, the audio information is input into the trained audio processing model, the clutter in the audio information is filtered, and the audio to be recognized is obtained;

[0091] The audio to be recognized is converted into audio text, the words in the audio text are extracted, and the words are matched with the preset keywords;

[0092] The occurrence times of each keyword are determined according to the matching result, and the feature matrix is determined by counting the occurrence times of each keyword;

[0093] The feature matrix is input into the trained determination model to determine whether to reset the collection clarity to the highest value.

[0094] The above content specifically describes the clarity adjustment process of the collection device. When the collection device does not receive audio information or does not receive valid audio information, the clarity itself is a gradually decreasing value. When valid audio is received, the clarity is reset to a different value, that is, the clarity is increased.

[0095] Specifically, the judgment process of whether the audio information is valid needs to use existing speech recognition technology to convert the audio information into text, extract the text, and determine the working state of the staff. For example, if there are many complaining keywords in the audio information issued by the staff, it can be considered that the staff is in an irritable state, which indirectly increases the probability of violation of operation rules. Therefore, the collection clarity needs to be increased.

[0096] Figure 3 The second sub-flow block diagram of the power operation site violation identification method based on image data comprises steps S201 to S204 of receiving image data obtained by each collection device, identifying the image data, and determining personnel characteristics.

[0097] Step S201: receiving image data obtained by each collection device, color identifying the image data, and determining the safety hat region and its color value according to the color identification result;

[0098] Step S202: determining the personnel type and personnel contour according to the safety hat region and its color value;

[0099] Step S203: obtaining the center point of the personnel contour and calculating the distance between each personnel contour, and generating graph data according to the center point and the distance; the node of the graph data is determined by the personnel type, and the edge of the graph data is determined by the distance;

[0100] Step S204: taking the graph data as the personnel feature.

[0101] Steps S201 to S204 provide a specific personnel feature extraction process. First, the safety helmet region is located according to the image recognition technology, and the type of the worker can be determined from the safety helmet region, such as judging whether there is a safety officer (yellow hat). Then, the center point of each personnel contour is determined, and the distribution of each personnel can be determined from the center point. The distribution and the type can be represented by the graph data. The graph data can be taken as the personnel feature.

[0102] Figure 4 The third sub-process flow chart of the method for identifying violations in the power operation site based on image data is shown in FIG. 3. The steps of querying the function label of the sub-region corresponding to the collection device, identifying the personnel feature based on the function label, and determining the violation probability include steps S301 to S304.

[0103] Step S301: query the function label of the sub-region corresponding to the collection device, and obtain the standard working image according to the function label;

[0104] Step S302: extract the standard personnel feature in the standard working image, and generate a sample set;

[0105] Step S303: train a neural network model based on the sample set;

[0106] Step S304: identify the personnel feature obtained according to the trained neural network model, and determine the violation probability.

[0107] Steps S301 to S304 specifically limit the application process of the personnel feature. The focus is to count some historical images, and manually judge whether it is a standard operation. If it is a standard operation, the personnel feature is obtained in the same way as the standard personnel feature. The standard personnel feature is used to train a neural network model. The newly obtained personnel feature is identified according to the neural network model, and a matching degree can be determined, which represents the violation probability.

[0108] Figure 5 The fourth sub-process flow chart of the method for identifying violations in the power operation site based on image data is shown in FIG. 4. The steps of comparing the violation probability with the preset probability threshold, and sending a close-range image acquisition instruction to the preset unmanned aerial vehicle end when the violation probability reaches the preset probability threshold include steps S401 to S403.

[0109] Step S401: compare the violation probability with the preset probability threshold, and read the personnel type and personnel contour when the violation probability reaches the preset probability threshold.

[0110] Step S402: determining the personnel level according to the personnel type, and querying the personnel contour corresponding to the highest personnel level;

[0111] Step S403: taking the personnel contour corresponding to the highest personnel level as the image acquisition center, and acquiring the close-range image containing time information based on the preset detection path.

[0112] When the violation probability is high, the personnel level is determined according to the personnel type, the personnel level is determined by the safety helmet area, and the personnel with the highest level is generally the person in charge. The close-range image is acquired with him as the center, which is more in line with the actual situation.

[0113] The detection path needs to be explained, which is specifically set by the R&D personnel. For example, when a certain worker is located, a zoom operation is performed to obtain a video from far to near. During the acquisition process, the shooting angle of the unmanned aerial vehicle end is adjusted to obtain the situation around the worker. This process is the detection path.

[0114] Embodiment 2

[0115] Figure 6 The composition structure block diagram of the power operation site violation identification system based on image data is shown in the embodiment of the present application. A power operation site violation identification system based on image data, the system 10 comprises:

[0116] The parameter determination module 11 is used to acquire the building model of the power operation site, determine the sub-area containing the function label according to the building model, and determine the installation parameters and acquisition parameters of the collection device based on the sub-area.

[0117] The feature extraction module 12 is used to receive the image data acquired by each collection device, identify the image data, and determine the personnel feature. The personnel feature is used to represent the working state of the worker.

[0118] The probability determination module 13 is used to query the function label of the sub-area corresponding to the collection device, identify the personnel feature based on the function label, and determine the violation probability.

[0119] The close-range detection module 14 is used to compare the violation probability with the preset probability threshold value, and send a close-range image acquisition instruction to the preset unmanned aerial vehicle end when the violation probability reaches the preset probability threshold value.

[0120] Figure 7 The composition structure block diagram of the parameter determination module in the power operation site violation identification system based on image data is shown in the embodiment of the present application. The parameter determination module 11 comprises:

[0121] The port opening unit 111 is configured to obtain a record model of a power operation site, and open a model adjustment port; the top of the record model is in a hidden state;

[0122] The model correction unit 112 is configured to receive an adjustment instruction of a worker based on the model adjustment port, correct the record model based on the adjustment instruction, and obtain a building model of the power operation site;

[0123] The model cutting unit 113 is configured to receive a cutting signal input by a user based on the building model, cut the building model according to the cutting signal, obtain a sub-region, and receive a function label input by the user based on the sub-region;

[0124] The execution unit 114 is configured to determine an installation parameter and a collection parameter of a collection device according to the sub-region.

[0125] Figure 8 A feature extraction module of a power operation site violation identification system based on image data is provided, and the feature extraction module 12 comprises:

[0126] The color recognition unit 121 is configured to receive image data obtained by each collection device, perform color recognition on the image data, and determine a safety helmet region and a color value thereof according to a color recognition result;

[0127] The personnel marking unit 122 is configured to determine a personnel type and a personnel contour according to the safety helmet region and the color value thereof;

[0128] The graph data generation unit 123 is configured to obtain a center point of the personnel contour and calculate a distance between each personnel contour, and generate graph data according to the center point and the distance; a node of the graph data is determined by the personnel type, and an edge of the graph data is determined by the distance;

[0129] The graph data storage unit 124 is configured to store the graph data as personnel features.

[0130] The above merely describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement and improvement within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An image data-based power work site violation recognition method, characterized by, The method comprises: acquiring a building model of a power operation site, determining a sub-region containing a function label according to the building model, determining installation parameters and collection parameters of a collection device based on the sub-region; receiving image data acquired by each collection device, identifying the image data, and determining personnel features; the personnel features are used to represent the working state of workers; inquiring the function label of the sub-region corresponding to the collection device, identifying the personnel features based on the function label, and determining a violation probability; comparing the violation probability with a preset probability threshold, and when the violation probability reaches the preset probability threshold, sending a close-range image acquisition instruction to a preset unmanned aerial vehicle end; wherein the step of receiving image data acquired by each collection device, identifying the image data, and determining personnel features comprises: receiving image data acquired by each collection device, color identifying the image data, and determining a safety helmet region and its color value according to the color identification result; determining personnel types and personnel contours according to the safety helmet region and its color value; acquiring the center points of the personnel contours and calculating the distances between the personnel contours, and generating graph data according to the center points and the distances; the nodes of the graph data are determined by the personnel types, and the edges of the graph data are determined by the distances; using the graph data as personnel features.

2. The method of claim 1, wherein, The step of acquiring a building model of a power operation site, determining a sub-region containing a function label according to the building model, and determining installation parameters and collection parameters of a collection device based on the sub-region comprises: acquiring a record model of a power operation site, and opening a model adjustment port; the top of the record model is in a hidden state; receiving adjustment instructions of workers based on the model adjustment port, correcting the record model based on the adjustment instructions, and obtaining a building model of a power operation site; receiving a user inputted segmentation signal based on the building model, segmenting the building model according to the segmentation signal to obtain a sub-region, and receiving a function label inputted by a user based on the sub-region; determining installation parameters and collection parameters of a collection device according to the sub-region.

3. The method of claim 2, wherein, The step of determining installation parameters and collection parameters of a collection device according to the sub-region comprises: acquiring the area range of the sub-region, determining the type, quantity and installation position of the collection device based on the area range; reading the function label of the sub-region, and querying risk parameters in a preset risk parameter library according to the function label; wherein the risk parameters include risk types and risk positions; the risk parameters in the risk parameter library are indexed by function labels; determining the collection angle of the collection device according to the installation position and the risk position; setting the collection clarity of the collection device to a value that decreases over time; when the collection device receives audio information, resetting the collection clarity according to the audio information.

4. The image data-based power work site violation identification method of claim 3, wherein, The step of resetting the collection clarity according to the audio information when the collection device receives the audio information comprises: when the collection device receives audio information, inputting the audio information into a trained audio processing model, filtering noise in the audio information, and obtaining to-be-identified audio. Convert the to-be-identified audio into audio text, extract words in the audio text, and match the words with preset keywords; Determine the number of occurrences of each keyword according to the matching result to determine a feature matrix. Input the feature matrix into a trained judgment model to determine whether to reset the collection clarity to the highest value.

5. The image data-based power work site violation identification method of claim 1, wherein, The step of querying the function label of the sub-region corresponding to the collection device, identifying the personnel feature based on the function label, and determining the probability of violation includes: Querying the function label of the sub-region corresponding to the collection device, and obtaining a standard working image according to the function label; Extracting standard personnel features in the standard working image to generate a sample set; Training a neural network model based on the sample set; Identifying the personnel feature according to the trained neural network model to determine the probability of violation.

6. The image data-based power work site violation identification method of claim 1, wherein, The step of comparing the probability of violation with a preset probability threshold and sending a close-range image acquisition instruction to a preset drone end when the probability of violation reaches the preset probability threshold includes: Comparing the probability of violation with a preset probability threshold, and reading the personnel type and personnel contour when the probability of violation reaches the preset probability threshold; Determining the personnel level according to the personnel type, and querying the personnel contour corresponding to the highest personnel level; Taking the personnel contour corresponding to the highest personnel level as an image acquisition center, and acquiring a close-range image containing time information based on a preset detection path.

7. An image data-based power work site violation recognition system, characterized by, The system includes: A parameter determination module configured to obtain a building model of a power operation site, determine a sub-region containing a function label according to the building model, and determine installation parameters and collection parameters of a collection device based on the sub-region; A feature extraction module configured to receive image data acquired by each collection device, identify the image data, and determine a personnel feature; the personnel feature is used to represent the working state of a worker; A probability determination module configured to query the function label of the sub-region corresponding to the collection device, identify the personnel feature based on the function label, and determine a probability of violation; A close-range detection module configured to compare the probability of violation with a preset probability threshold, and send a close-range image acquisition instruction to a preset drone end when the probability of violation reaches the preset probability threshold. The feature extraction module is further configured to receive image data acquired by each collection device, perform color identification on the image data, and determine a safety helmet region and a color value thereof according to a color identification result; Determine a personnel type and a personnel contour thereof according to the safety helmet region and the color value thereof; Obtain a center point of the personnel contour and calculate distances between the personnel contours, and generate graph data according to the center point and the distances; the nodes of the graph data are determined by the personnel type, and the edges of the graph data are determined by the distances; The graph data is used as the personnel feature.

8. The image data-based power work site violation identification system of claim 7, wherein, The parameter determination module includes: A port opening unit configured to obtain a record model of a power operation site, and open a model adjustment port; the top of the record model is in a hidden state. The model correction unit is configured to receive an adjustment instruction of the port adjustment worker based on the model, correct the filed model based on the adjustment instruction, and obtain a building model of the power operation site; The model cutting unit is configured to receive a cutting signal input by a user based on the building model, cut the building model according to the cutting signal to obtain a sub-region, and receive a function label input by the user based on the sub-region; The execution unit is configured to determine installation parameters and collection parameters of the collection device according to the sub-region.

9. The image data-based power work site violation identification system of claim 7, wherein, The feature extraction module comprises: The color recognition unit is configured to receive image data acquired by each collection device, perform color recognition on the image data, and determine a safety helmet region and a color value thereof according to a color recognition result; The personnel marking unit is configured to determine a personnel type and a personnel contour according to the safety helmet region and the color value thereof; The graph data generation unit is configured to acquire a center point of the personnel contour and calculate a distance between each personnel contour, and generate graph data according to the center point and the distance; a node of the graph data is determined by the personnel type, and an edge of the graph data is determined by the distance; The graph data storage unit is configured to store the graph data as personnel features.

Images