WO2022121886A1 - Method and apparatus for identifying dress code for electric power operations - Google Patents

Abstract

Provided are a method and apparatus for identifying a dress code for electric power operations, comprising: receiving a human body image to be recognized; performing image pre-processing of the human body image to be recognized to generate an image matrix to be recognized; inputting the image matrix to be recognized into a target human-body dressing feature matrix extraction network, and outputting a human dress feature matrix corresponding to the image matrix to be recognized; the target human dressing feature matrix extraction network is generated by means of a pre-programmed model training process; on the basis of a result of determination of whether the human dressing feature matrix satisfies a preset condition, determining whether a person corresponding to the human body image to be recognized is dressed in a standardized manner. Thus the accuracy of recognition of dress of personnel in complex environments is effectively improved, reducing the occurrence of safety hazards.

Description

Method and device for identifying dress code for electrical work

This application claims the priority of the Chinese patent application filed on December 11, 2020 with the application number 202011452940.6 and the invention titled "A method and device for identifying a dress code for electrical work", the entire contents of which are incorporated by reference in in this application.

technical field

The present application relates to the technical field of dress identification, and in particular, to a method and device for identifying a dress code for electrical work.

Background technique

With the rapid development of the social economy and the continuous improvement of the level of science and technology, my country's electric power industry has developed rapidly in recent years. In the process of electric power production, there are strict requirements on the clothing and clothing of operators, and it is necessary to monitor in real time whether the on-site operators are wearing appropriate clothing. Work clothes and safety helmets, and during the entire work process: do not open your chest and back, roll up your sleeves, roll up your trousers, keep your work clothes in good condition, your safety helmets must not be damaged, and even require that your work clothes be made of cotton, etc. . In the process of operation, if it is found that the on-site operators are wearing non-compliant clothes, an alarm should be issued in time and the images of the personnel should be recorded.

However, in the process of power production, most of the field operators work in the field. The lighting conditions on the site are fickle, the working environment varies widely, the working position changes all the time, and the posture of the human body changes constantly. requirements, especially when there are minor non-compliances in personnel wearing: open chest and back, rolled up sleeves, rolled up trousers, non-compliant fabrics of clothing, damaged helmets, etc., the existing target identification methods The recognition rate of the device is low, and it is difficult to accurately identify whether the personnel are dressed in compliance, which may lead to safety hazards.

SUMMARY OF THE INVENTION

The present invention provides a method and device for identifying the dress code for electrical work, which solves the problem that the existing target identification method has low identification rate and accuracy rate in complex environment, it is difficult to accurately identify whether personnel are dressed in compliance with regulations, which may lead to potential safety hazards. technical problem.

A method for identifying a dress code for electrical work provided by the present invention includes:

Receive the image of the human body to be recognized;

Perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

The to-be-recognized image matrix is input into the target human body clothing feature matrix extraction network, and the human body clothing feature matrix corresponding to the to-be-identified image matrix is output; wherein, the target human body clothing feature matrix extraction network is obtained through a preset model training process. generate;

Based on the judgment result of whether the human body clothing feature matrix satisfies the preset condition, it is determined whether the clothing of the person corresponding to the to-be-recognized human body image is standardized.

Optionally, the step of performing image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix includes:

converting the to-be-identified human body image into a first image matrix;

Perform an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

An image normalization operation and an image normalization operation are performed on the second image matrix to generate a to-be-identified image matrix.

Optionally, the model training process includes:

Respectively obtain human body images in various historical power operation scenarios;

performing image preprocessing on the human body image to generate a standardized image matrix;

Inputting the standardized image matrix into a preset initial human body clothing feature matrix extraction network to obtain a clothing feature matrix;

A preset dual-core optimization algorithm is used in combination with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.

Optionally, the preset dual-core optimization algorithm includes an overall comparison function and a patch comparison function, the clothing feature matrix includes an overall clothing feature matrix and a clothing patch feature matrix, and the preset dual-core optimization model is combined with the Describe the clothing feature matrix, the steps of training the initial human body clothing feature matrix extraction network to generate the target human body clothing feature matrix extraction network, including:

Get the standard overall dress feature matrix;

importing the standard overall clothing feature matrix and the overall clothing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

Based on the optimization target value and the tracking target value, a back-propagation algorithm is used to adjust the overall parameters of the initial human clothing feature matrix extraction network until the tracking target value is equal to the optimization target value.

Get the standard attire patch feature matrix;

importing the attire patch feature matrix and the standard attire patch feature matrix into the patch comparison function to obtain a patch optimization target value and a patch tracking target value;

Based on the patch optimization target value and the patch tracking target value, the back-propagation algorithm is used to adjust the patch parameters of the initial human clothing feature matrix extraction network until the patch tracking target value is equal to the Plaque optimization target value;

When the tracking target value is equal to the optimization target value, and the patch tracking target value is equal to the patch optimization target value, a target human body clothing feature matrix extraction network is generated.

Optionally, the preset condition includes a plurality of sub-conditions, and the step of determining whether the clothing of the person corresponding to the to-be-recognized human body image is standardized based on the judgment result of whether the human body clothing feature matrix satisfies the preset condition includes:

If the human body clothing feature matrix satisfies any one of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image does not meet the specification;

If the human body clothing feature matrix does not satisfy all of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification.

The present invention also provides a device for identifying a dress code for electrical work, comprising:

A human body image receiving module to be recognized, used for receiving the human body image to be recognized;

A to-be-recognized image matrix generation module, configured to perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

The human body clothing feature matrix output module is used for inputting the to-be-recognized image matrix into the target human body clothing feature matrix extraction network, and outputting the human body clothing feature matrix corresponding to the to-be-identified image matrix; wherein, the target human body clothing feature matrix extraction The network is generated by the preset model training module;

A personnel dress code judging module is configured to determine whether the person corresponding to the to-be-recognized human body image is dressed according to the judgment result of whether the human body dress feature matrix satisfies a preset condition.

Optionally, the to-be-recognized image matrix generation module includes:

an image conversion submodule for converting the to-be-recognized human body image into a first image matrix;

a first image processing submodule, configured to perform an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

The second image processing submodule is configured to perform an image normalization operation and an image normalization operation on the second image matrix to generate an image matrix to be recognized.

Optionally, the model training module includes:

The historical human body image acquisition sub-module is used to separately acquire human body images in various historical power operation scenarios;

A standardized image matrix generation sub-module for performing image preprocessing on the human body image to generate a standardized image matrix;

Dressing feature matrix generation sub-module for inputting the standardized image matrix into a preset initial human body dressing feature matrix extraction network to obtain a dressing feature matrix;

The training sub-module is used for using a preset dual-core optimization algorithm in combination with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.

Optionally, the preset dual-core optimization algorithm includes an overall comparison function and a patch comparison function, the dressing feature matrix includes an overall dressing feature matrix and a dressing patch feature matrix, and the training submodule includes:

The standard overall dress feature matrix acquisition sub-module is used to obtain the standard overall dress feature matrix;

an overall comparison function processing submodule, used for importing the standard overall dressing feature matrix and the overall dressing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

an overall parameter adjustment sub-module, used for adjusting the overall parameters of the initial human clothing feature matrix extraction network based on the optimization target value and the tracking target value by using a back-propagation algorithm until the tracking target value is equal to the optimization target value.

Standard dress patch feature matrix acquisition sub-module, used to obtain standard dress patch feature matrix;

The patch comparison function processing sub-module is used to import the dressing patch feature matrix and the standard dressing patch feature matrix into the patch comparison function to obtain the patch optimization target value and the patch tracking target value ;

The patch parameter adjustment sub-module is used to adjust the patch parameters of the initial human clothing feature matrix extraction network based on the patch optimization target value and the patch tracking target value by using the back-propagation algorithm until all the the patch tracking target value is equal to the patch optimization target value;

The target body clothing feature matrix extraction network generation sub-module is used to generate the target body clothing feature matrix when the tracking target value is equal to the optimization target value and the patch tracking target value is equal to the patch optimization target value Extract the network.

Optionally, the preset condition includes multiple sub-conditions, and the personnel dress code judgment module includes:

A non-conformity determination sub-module, configured to determine that the clothing of the person corresponding to the to-be-recognized human body image does not conform to the specification if the human body attire feature matrix satisfies any one of the sub-conditions;

A specification-compliant determination sub-module, configured to determine that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification if the human body clothing feature matrix does not satisfy all of the sub-conditions.

As can be seen from the above technical solutions, the present invention has the following advantages:

The present invention receives the image of the human body to be recognized, performs image preprocessing on the image of the human body to be recognized, generates the image matrix to be recognized, and then uses the target human body clothing feature matrix extraction network to extract the human body clothing feature matrix corresponding to the to-be-recognized image matrix, and finally, based on the human body clothing feature Whether the matrix satisfies the preset conditions determines whether the dress code of the person corresponding to the human body image to be recognized is standardized. Therefore, the existing target recognition methods have low recognition rate and accuracy in complex environments, and it is difficult to accurately identify whether personnel are wearing compliance, which may lead to technical problems of potential safety hazards, and effectively improve the accuracy of personnel wearing identification in complex environments. to reduce the occurrence of security risks.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without any creative effort.

1 is a flowchart of steps of a method for identifying a dress code for electrical work provided by an embodiment of the present invention;

2 is a flow chart of steps of a method for identifying a dress code for electrical work provided by an optional embodiment of the present invention;

3 is a schematic flowchart of a network training process for extracting an initial human body dress feature matrix according to an embodiment of the present invention;

4 is a schematic flowchart of calculating an optimization target value and a tracking target value in an embodiment of the present invention;

5 is a schematic flowchart of calculating a patch optimization target value and a patch tracking target value in an embodiment of the present invention;

6 is a schematic structural diagram of a power operation supervision platform provided by an embodiment of the present invention;

FIG. 7 is a structural block diagram of an apparatus for identifying a dress code for electrical work according to an embodiment of the present invention.

Detailed ways

The embodiments of the present invention provide a method and device for identifying a dress code for electrical work, which are used to solve the problem that the existing target identification method has a low recognition rate and accuracy rate in a complex environment, and it is difficult to accurately identify whether a person's clothing is compliant, which may lead to Technical issues with security risks.

In order to make the purpose, features and advantages of the present invention more obvious and understandable, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the following The described embodiments are only some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Please refer to FIG. 1 . FIG. 1 is a flowchart of steps of a method for identifying a dress code for electrical work provided by an embodiment of the present invention.

A method for identifying a dress code for electrical work provided by the present invention includes:

Step 101, receiving an image of a human body to be identified;

The to-be-recognized human image refers to an image in which an electric power operator exists in the electric power work scene.

In this embodiment of the present invention, there is usually a corresponding monitoring device such as a camera in the power work scene to ensure the safety of the power work. At this time, in order to determine whether the dress of the power worker is standard, target detection can be used from the monitoring video stream. The algorithm acquires the image of the human body to be recognized to determine the presence of the image of the operator.

Optionally, the predetermined period may be the acquisition time, and the image of the human body to be recognized at the corresponding time is acquired.

It is worth mentioning that the embodiment of the present invention is not limited to the target detection algorithm used to obtain the human body image to be recognized from the video stream.

Step 102, performing image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

After acquiring the image of the human body to be recognized, image preprocessing can be performed on the image of the human body to be recognized, so as to obtain, for example, a matrix of images to be recognized under different illuminations, different scenes and different angles.

Step 103, inputting the to-be-recognized image matrix into the target human body clothing feature matrix extraction network, and outputting the human body clothing feature matrix corresponding to the to-be-recognized image matrix;

The human clothing feature matrix includes but is not limited to the following features: overalls lining features, long-sleeved overalls features, clothing opening features, clothing damage features, helmet wearing characteristics, and helmet damage characteristics, etc.

In a specific implementation, the target human body clothing feature matrix extraction network is generated through a preset model training process, and the to-be-recognized image matrix is input into the target human body clothing feature matrix extraction network to obtain the human body clothing corresponding to the to-be-recognized image matrix. feature matrix.

It is worth mentioning that, in order to further improve the recognition accuracy, image matrices of different scales to be recognized may be generated in the process of step 102, so as to facilitate the determination of the human body clothing feature matrices under different scales.

Step 104 , based on the judgment result of whether the human body clothing feature matrix satisfies a preset condition, determine whether the clothing of the person corresponding to the to-be-recognized human body image is standard.

After obtaining the human body clothing feature matrix, it can be further determined whether the clothing of the person corresponding to the human body image to be recognized is standardized based on the judgment result of whether the human body clothing feature matrix satisfies the preset conditions, so as to determine whether a warning needs to be issued to reduce the potential safety hazards. occur.

In the embodiment of the present invention, by receiving the image of the human body to be recognized, performing image preprocessing on the image of the human body to be recognized, generating the image matrix to be recognized, and then using the target human body clothing feature matrix extraction network to extract the human body clothing feature matrix corresponding to the to-be-recognized image matrix, Finally, based on whether the human body clothing feature matrix satisfies the preset conditions, it is determined whether the clothing of the person corresponding to the human body image to be recognized is standardized. Therefore, the existing target recognition methods have low recognition rate and accuracy in complex environments, and it is difficult to accurately identify whether personnel are wearing compliance, which may lead to technical problems of potential safety hazards, and effectively improve the accuracy of personnel wearing identification in complex environments. to reduce the occurrence of security risks.

Please refer to FIG. 2 , which is a flowchart of steps of a method for identifying a dress code for electrical work provided by an optional embodiment of the present invention.

A method for identifying a dress code for electrical work provided by the present invention includes:

Step 201, receiving an image of a human body to be identified;

In this embodiment of the present invention, the specific implementation process of step 201 is the same as that of step 102 above, and details are not described herein again.

In an optional embodiment of the present invention, the technical feature in the above step 102 "to perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix" can be replaced with the following steps 202-204:

Step 202, converting the to-be-identified human body image into a first image matrix;

In this embodiment of the present invention, the first image matrix may be generated by performing a transformation operation of cropping or scaling on the image of the human body to be recognized.

Further, in order to improve the detection accuracy of the image, the image of the human body to be recognized may be converted into a multi-scale first image matrix to satisfy subsequent image processing of different scales.

Step 203, performing an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

In an example of the present invention, a corresponding image correction operation and an image enhancement operation may be performed on the first image matrix according to preset requirements, such as the image correction magnitude input by the user, the object of image enhancement and the enhancement magnitude, etc., to obtain A second image matrix more suitable for image recognition.

The image correction operation refers to automatically detecting the tilt direction and tilt angle of the image according to the image features, and further correcting the image position according to the detected information.

Image enhancement refers to the enhancement of useful information in an image, which can be a distortion process whose purpose is to improve the visual effect of an image for a given image application. Purposefully emphasize the overall or local characteristics of the image, make the original unclear image clear or emphasize some interesting features, expand the difference between the features of different objects in the image, suppress the uninteresting features, and improve the image. Quality, rich information, strengthen image interpretation and recognition effect, to meet the needs of some special analysis.

Step 204 , performing an image normalization operation and an image normalization operation on the second image matrix to generate a to-be-identified image matrix.

After the second image matrix is obtained, the second image matrix can be processed by zero mean normalization to realize the image normalization operation, and then the image normalization operation is performed on the image matrix after image normalization, and the image matrix is converted into [-1, 1 ] to obtain the matrix of images to be recognized.

Zero-mean canonical processing refers to taking each variable in the image matrix and subtracting their mean.

The image normalization operation refers to taking each variable in the image matrix minus their mean and dividing by the standard deviation.

Step 205, inputting the to-be-recognized image matrix into the target human body clothing feature matrix extraction network, and outputting the human body clothing feature matrix corresponding to the to-be-recognized image matrix;

Optionally, the to-be-recognized image matrix also includes frame coordinates and confidence levels of each person, and the to-be-recognized image matrix is input into the target human body clothing feature matrix extraction network to obtain a higher confidence level and is located within the frame coordinates. Matrix of Human Dressing Characteristics.

Wherein, the target human body clothing feature matrix extraction network is generated through a preset model training process;

Further, the model training process includes the following steps S1-S4:

S1. Respectively acquire human body images in various historical power operation scenarios;

S2, performing image preprocessing on the human body image to generate a standardized image matrix;

S3, inputting the standardized image matrix into a preset initial human body clothing feature matrix extraction network to obtain a clothing feature matrix;

In this embodiment of the present invention, human body images in various historical power work scenarios, such as human body images in power work scenarios with different lighting, different heights, or different angles, can be obtained respectively; image preprocessing is performed on the human body images to generating a standardized image matrix; and then inputting the standardized image matrix into a preset initial human body clothing feature matrix extraction network to obtain a clothing feature matrix in the standardized image matrix.

Optionally, the initial human clothing feature matrix extraction network can be constructed by using Convolutional Neural Network (CNN), and integrate data enhancement technology and anchor design strategy to improve human detection under different scales, backgrounds, and lighting. accuracy.

The specific implementation process of the image preprocessing is the same as the above steps 202-204, which will not be repeated here.

S4 , using a preset dual-core optimization algorithm combined with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.

Referring to FIG. 3 , FIG. 3 shows a schematic flowchart of a network training process for extracting an initial human clothing feature matrix in an embodiment of the present invention.

It includes inputting the clothing feature matrix into the overall comparison function calculation module 31 and the patch comparison function calculation module 32 respectively, and inputting the output of the overall comparison function calculation module into the overall comparison model parameter optimization module 33 to obtain the optimal Optimize the target value and the tracking target value; input the output of the patch comparison function calculation module to the patch comparison model parameter optimization module 34 to obtain the optimal patch optimization target value and patch tracking target value.

In an example of the present invention, the preset dual-kernel optimization algorithm includes an overall alignment function and a patch alignment function, the clothing feature matrix includes an overall clothing feature matrix and a clothing patch feature matrix, and step S4 may include the following sub-components Steps S41-S47:

S41. Obtain a standard overall dress feature matrix;

S42, importing the standard overall clothing feature matrix and the overall clothing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

S43. Based on the optimization target value and the tracking target value, use a back-propagation algorithm to adjust the overall parameters of the initial human body clothing feature matrix extraction network until the tracking target value is equal to the optimization target value.

Referring to FIG. 4, FIG. 4 shows a schematic flowchart of calculating an optimization target value and a tracking target value in an embodiment of the present invention;

In this embodiment, the obtained overall dressing feature matrix is sent to the first input of the first adder module 310, and the standard overall dressing feature matrix is sent to the second input of the first adder module 310; the first addition The output of the divider module 310 is sent to the second input terminal of the first divider module 311 , and the output of the first adder module 310 is also sent to the second input terminal of the second divider module 312 .

The overall dressing feature matrix is also sent to the first input terminal of the first divider module 311 , so that the matrix operation of dividing the overall dressing feature matrix by the output of the first adder module 310 is completed in the first divider module 311 . The division output value completed by the first divider module 311 is sent to the first natural logarithm module 313, and the logarithmic value calculated by the first natural logarithm module 313 is output to the first output of the overall comparison function to obtain the optimization target value .

The standard overall dressing characteristic matrix is also sent to the first input terminal of the second divider module 312, so that the matrix operation of dividing the standard overall dressing characteristic matrix by the output of the first adder module 310 is completed in the second divider module 312. . The division calculation value completed by the second divider module 312 is sent to the second natural logarithm module 314, and the logarithmic value calculated by the second natural logarithm module 314 is output to the second output terminal of the overall comparison function to obtain the tracking target value .

Finally, the optimization target value and the tracking target value are substituted into the back-propagation algorithm to adjust the overall parameters of the initial human clothing feature matrix extraction network, calculate the output of the initial human clothing feature matrix extraction network, and continue to calculate the tracking target value until the tracking target value equals to Optimize the target value and complete the overall parameter adjustment.

The back-propagation algorithm can be, but is not limited to, a self-optimizing estimation method for a dual-constrained objective neural network.

S44. Obtain a standard dress patch feature matrix;

S45, importing the dressing patch feature matrix and the standard dressing patch feature matrix into the patch comparison function to obtain a patch optimization target value and a patch tracking target value;

S46. Based on the patch optimization target value and the patch tracking target value, use the back-propagation algorithm to adjust the patch parameters of the initial human clothing feature matrix extraction network until the patch tracking target value is equal to the plaque optimization target value;

Referring to FIG. 5 , FIG. 5 shows a schematic flowchart of calculating a target value for patch optimization and a target value for patch tracking in an embodiment of the present invention. Among them, the feature matrix of clothing patches includes negative patch array, positive patch array and complementary patch array.

In this embodiment, the standard dressing patch feature matrix is sent to the first input terminal of the first multiplier 320, to the first input terminal of the second multiplier 321, and to the first input terminal of the third multiplier 322. input.

The negative patch array is sent to the second input of the first multiplier 320, and is multiplied with the standard dressing patch feature matrix, and the operation result is sent to the first input of the first divider 323; the positive patch array is sent to the first The second input of the 2 multiplier 321 is multiplied with the standard dressing patch feature matrix, and the operation result is sent to the first input of the second divider 324; the patch array is sent to the second of the third multiplier 322. The input terminal is multiplied with the standard dressing patch feature matrix, and the operation result is sent to the first input terminal of the third divider 326; the comprehensive adjustment coefficient K given by the first setting module 325 is sent to the first divider 323 The second input terminal of , is also sent to the second input terminal of the second divider 324 and is also sent to the second input terminal of the third divider 326 . The value calculated by the first divider 323 is sent to the input terminal of the first exponent calculation module 327 to complete the exponent operation on the number divided by the 323 module; the value calculated by the second divider 324 is sent to the second exponent calculation module 328 The input terminal of 324 completes the exponent operation on the number divided by the 324 module; the value calculated by the third divider 326 is sent to the input terminal of the third exponent calculation module 329 to complete the exponent operation on the number divided by the 326 module. The output of the first index calculation module 327 is sent to the input end of the first accumulation calculation module 3210 to perform the accumulation calculation of the negative patch operation data; the output of the third index calculation module 329 is sent to the input end of the second accumulation calculation module 3211, Carry out the accumulation calculation of patch patch operation data; the output of the first accumulation calculation module 3210 is sent to the first input end of the first addition module 3211, and the output of the second index calculation module 328 is sent to the second input of the first addition module 3213 At the terminal, the negative patch accumulation operation data is superimposed on the positive patch operation data to form the negative patch comprehensive calculation value; the output of the second accumulation calculation module 3213 is also sent to the first input terminal of the second addition module 3212, the second The output of the exponent calculation module 328 is sent to the second input of the second addition module 3212, so that the patch patch accumulation operation data is superimposed on the positive patch operation data to form the patch patch comprehensive calculation value. The output of the second exponent calculation module 328 is also sent to the first input of the fourth divider module 3214, and the output of the first addition module 3211 is sent to the second input of the fourth divider module 3214, thus completing the positive patch The operation data is divided by the negative patch comprehensive calculation value; the output of the second exponent calculation module 328 is also sent to the first input of the fifth divider module 3215, and the output of the second addition module 3212 is sent to the fifth divider module 3215. In the second input terminal, the positive patch operation data is divided by the comprehensive calculation value of the patch patch. The output of the fourth divider module 3214 is sent to the first natural logarithm module 3216, which performs logarithmic operation on the comprehensive value calculated above; the output of the fifth divider module 3215 is sent to the second natural logarithm module 3217. The calculated composite value is logarithmic. The adjustment coefficient a given by the second setting module 3218 is sent to the first input terminal of the fourth multiplier 3219, and the output of the first natural logarithm module 3216 is sent to the second input terminal of the fourth multiplier 3219. The multiplied output value is sent to the first output terminal of the patch comparison function calculation to obtain the patch optimization target value.

The adjustment coefficient b given by the third setting module 3221 is sent to the first input of the fifth multiplier 3220, and the output of the second natural logarithm module 3217 is sent to the second input of the fifth multiplier 3220. The multiplied output value is sent to the second output terminal of the patch comparison function calculation to obtain the patch tracking target value.

Finally, the patch optimization target value and patch tracking target value are substituted into the back-propagation algorithm to adjust the patch parameters of the initial human clothing feature matrix extraction network, calculate the output of the initial human clothing feature matrix extraction network, and continuously calculate the patch tracking target. until the patch tracking target value is equal to the patch optimization target value, and the patch parameter adjustment is completed.

S47. When the tracking target value is equal to the optimization target value, and the patch tracking target value is equal to the patch optimization target value, generate a target human body clothing feature matrix extraction network.

Step 206 , based on the judgment result of whether the human body clothing feature matrix satisfies a preset condition, determine whether the clothing of the person corresponding to the to-be-recognized human body image is standard.

Optionally, the preset condition includes multiple sub-conditions, and step 206 may include the following sub-steps:

If the human body clothing feature matrix satisfies any one of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image does not meet the specification;

If the human body clothing feature matrix does not satisfy all of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification.

In this embodiment of the present invention, multiple sub-conditions may include, but are not limited to, not wearing short clothes, long-sleeved clothes are not work clothes, clothes are open, and clothes are damaged; the wearing of safety helmets does not meet the specification requirements and the safety helmets are damaged, etc. When the human body clothing feature matrix satisfies any one of the above sub-conditions, it is determined that the clothing of the person corresponding to the human body image to be recognized does not meet the specification; otherwise, if all the sub-conditions are not satisfied, it is determined that the clothing of the person corresponding to the human body image to be recognized conforms to the specification.

Optionally, when the clothing of the person corresponding to the human body image to be recognized does not conform to the specification, an alarm may be output to inform the monitoring personnel to perform further processing.

In the embodiment of the present invention, the image of the human body to be recognized is received, the image of the human body to be recognized is preprocessed, the image matrix to be recognized is generated, the target human body clothing feature matrix extraction network is used to extract the human body clothing feature matrix corresponding to the to-be-recognized image matrix, and finally based on the human body Whether the clothing feature matrix satisfies the preset conditions, it is determined whether the clothing of the person corresponding to the human body image to be recognized is standardized. Therefore, the existing target recognition methods have low recognition rate and accuracy in complex environments, and it is difficult to accurately identify whether personnel are wearing compliance, which may lead to technical problems of potential safety hazards, and effectively improve the accuracy of personnel wearing identification in complex environments. to reduce the occurrence of security risks.

Please refer to FIG. 6. FIG. 6 is a schematic structural diagram of an electric power operation supervision platform according to an embodiment of the present invention.

It includes the physical resource layer 601, the scheduling management layer 602, the training environment layer 603, and the business application layer 604 to ensure the normal and stable operation of the intelligent identification algorithm, and transplant it to the edge side terminal to provide users with model training, prediction, evaluation and deployment. full life cycle management capabilities.

The physical resource layer 601 includes heterogeneous computing hardware (CPU, GPU), storage, network equipment, and security protection equipment.

The scheduling management layer 602 is developed based on Kubernetes and docker, and includes cluster management, resource virtualization, and task scheduling.

The training environment layer 603 provides services in the form of docker, including the mainstream learning frameworks TensorFlow, PyTorch, Caffe, scikit-learn, XGBoost and other machine learning/deep learning environments, while integrating JupyterHub and other interactive code debugging notebooks and MPI parallel programming interfaces . The system operating environment and learning environment are iteratively managed through the docker repository.

The business application layer 604 includes data processing, data labeling, model training, and model publishing.

The model training module is based on various machine learning and deep learning training environments, and suspends training after configuring parameters through pre-written training scripts. The entire training process is automatically completed by the pipeline built in the background, and model production is carried out around data processing, data labeling, training, and model management processes. Model training presets TensorFlow, PyTorch, Caffe, scikit-learn, XGBoost and other learning environments through docker. Using the task scheduling system, users can submit learning task codes to the cluster, and the task management system will allocate resources to users according to the user's quota, create an environment specified by the user, and add learning tasks to the task queue. When the resources are free, the learning program will be run. Users can submit code with one click, generate distributed tasks, and greatly reduce development costs and resource occupation.

Please refer to FIG. 7, which is a flowchart of steps of a method for identifying a dress code for electrical work provided by an embodiment of the present invention.

A device for identifying a dress code for electrical work provided by the present invention includes:

The human body image receiving module 701 to be recognized is configured to receive the human body image to be recognized;

A to-be-recognized image matrix generation module 702, configured to perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

The human body clothing feature matrix output module 703 is used to input the to-be-recognized image matrix into the target human body clothing feature matrix extraction network, and output the human body clothing feature matrix corresponding to the to-be-recognized image matrix; wherein, the target human body clothing feature matrix The extraction network is generated by the preset model training module 704;

The personnel dress code judgment module 705 is configured to determine whether the person corresponding to the to-be-recognized body image is dressed according to the judgment result of whether the human body dress feature matrix satisfies a preset condition.

Optionally, the to-be-recognized image matrix generation module 702 includes:

an image conversion submodule for converting the to-be-recognized human body image into a first image matrix;

a first image processing submodule, configured to perform an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

The second image processing submodule is configured to perform an image normalization operation and an image normalization operation on the second image matrix to generate an image matrix to be recognized.

Optionally, the model training module 704 includes:

The historical human body image acquisition sub-module is used to separately acquire human body images in various historical power operation scenarios;

A standardized image matrix generation sub-module for performing image preprocessing on the human body image to generate a standardized image matrix;

Dressing feature matrix generation sub-module for inputting the standardized image matrix into a preset initial human body dressing feature matrix extraction network to obtain a dressing feature matrix;

The training sub-module is used for using a preset dual-core optimization algorithm in combination with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.

Optionally, the preset dual-core optimization algorithm includes an overall comparison function and a patch comparison function, the clothing feature matrix includes an overall clothing feature matrix and a clothing patch feature matrix, and the training submodule includes:

The standard overall dress feature matrix acquisition sub-module is used to obtain the standard overall dress feature matrix;

an overall comparison function processing submodule, used for importing the standard overall dressing feature matrix and the overall dressing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

an overall parameter adjustment sub-module, used for adjusting the overall parameters of the initial human clothing feature matrix extraction network based on the optimization target value and the tracking target value by using a back-propagation algorithm until the tracking target value is equal to the optimization target value.

Standard dress patch feature matrix acquisition sub-module, used to obtain standard dress patch feature matrix;

The patch comparison function processing sub-module is used to import the dressing patch feature matrix and the standard dressing patch feature matrix into the patch comparison function to obtain the patch optimization target value and the patch tracking target value ;

The patch parameter adjustment sub-module is used to adjust the patch parameters of the initial human clothing feature matrix extraction network based on the patch optimization target value and the patch tracking target value by using the back-propagation algorithm until all the the patch tracking target value is equal to the patch optimization target value;

The target body clothing feature matrix extraction network generation sub-module is used to generate the target body clothing feature matrix when the tracking target value is equal to the optimization target value and the patch tracking target value is equal to the patch optimization target value Extract the network.

Optionally, the preset condition includes multiple sub-conditions, and the personnel dress code judgment module 705 includes:

A non-conformity determination sub-module, configured to determine that the clothing of the person corresponding to the to-be-recognized human body image does not conform to the specification if the human body attire feature matrix satisfies any one of the sub-conditions;

A specification-compliant determination sub-module, configured to determine that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification if the human body clothing feature matrix does not satisfy all of the sub-conditions.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process of the above-described devices and units, reference may be made to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

As mentioned above, the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand: The technical solutions described in the embodiments are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for identifying a dress code for electrical work, comprising:

   Receive the image of the human body to be recognized;

   Perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

   The to-be-recognized image matrix is input into the target human body clothing feature matrix extraction network, and the human body clothing feature matrix corresponding to the to-be-identified image matrix is output; wherein, the target human body clothing feature matrix extraction network is obtained through a preset model training process. generate;

   Based on the judgment result of whether the human body clothing feature matrix satisfies the preset condition, it is determined whether the clothing of the person corresponding to the to-be-recognized human body image is standardized.
2. The method for recognizing dress code for electrical work according to claim 1, wherein the step of performing image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix comprises:

   converting the to-be-identified human body image into a first image matrix;

   Perform an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

   An image normalization operation and an image normalization operation are performed on the second image matrix to generate a to-be-identified image matrix.
3. The method for identifying a dress code for electrical work according to claim 1, wherein the model training process comprises:

   Respectively obtain human body images in various historical power operation scenarios;

   performing image preprocessing on the human body image to generate a standardized image matrix;

   Inputting the standardized image matrix into a preset initial human body clothing feature matrix extraction network to obtain a clothing feature matrix;

   A preset dual-core optimization algorithm is used in combination with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.
4. The method for identifying dress codes for electrical work according to claim 3, wherein the preset dual-core optimization algorithm includes an overall comparison function and a patch comparison function, and the dress feature matrix includes an overall dress feature matrix and a dress spot block feature matrix, the steps of using the preset dual-core optimization model combined with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate the target human body clothing feature matrix extraction network include:

   Get the standard overall dress feature matrix;

   importing the standard overall clothing feature matrix and the overall clothing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

   Based on the optimization target value and the tracking target value, a back-propagation algorithm is used to adjust the overall parameters of the initial human clothing feature matrix extraction network until the tracking target value is equal to the optimization target value;

   Get the standard attire patch feature matrix;

   importing the attire patch feature matrix and the standard attire patch feature matrix into the patch comparison function to obtain a patch optimization target value and a patch tracking target value;

   Based on the patch optimization target value and the patch tracking target value, the back-propagation algorithm is used to adjust the patch parameters of the initial human clothing feature matrix extraction network until the patch tracking target value is equal to the Plaque optimization target value;

   When the tracking target value is equal to the optimization target value, and the patch tracking target value is equal to the patch optimization target value, a target human body clothing feature matrix extraction network is generated.
5. The method for recognizing dress code for electrical work according to claim 1, wherein the preset condition includes a plurality of sub-conditions, and the determination is made based on a judgment result of whether the human body dress characteristic matrix satisfies the preset condition. The steps of identifying whether the person corresponding to the human body image is dressed properly, including:

   If the human body clothing feature matrix satisfies any one of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image does not meet the specification;

   If the human body clothing feature matrix does not satisfy all of the sub-conditions, it is determined that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification.
6. A device for identifying a dress code for electrical work, characterized in that it includes:

   A human body image receiving module to be recognized, used for receiving the human body image to be recognized;

   A to-be-recognized image matrix generation module, configured to perform image preprocessing on the to-be-recognized human body image to generate a to-be-recognized image matrix;

   The human body clothing feature matrix output module is used for inputting the to-be-recognized image matrix into the target human body clothing feature matrix extraction network, and outputting the human body clothing feature matrix corresponding to the to-be-identified image matrix; wherein, the target human body clothing feature matrix extraction The network is generated by the preset model training module;

   A personnel dress code judging module is configured to determine whether the person corresponding to the to-be-recognized human body image is dressed according to the judgment result of whether the human body dress feature matrix satisfies a preset condition.
7. The apparatus for recognizing dress code for electrical work according to claim 6, wherein the image matrix generation module to be recognized comprises:

   an image conversion submodule for converting the to-be-recognized human body image into a first image matrix;

   a first image processing submodule, configured to perform an image correction operation and an image enhancement operation on the first image matrix according to preset requirements to obtain a second image matrix;

   The second image processing submodule is configured to perform an image normalization operation and an image normalization operation on the second image matrix to generate an image matrix to be recognized.
8. The apparatus for identifying electrical work dress codes according to claim 6, wherein the model training module comprises:

   The historical human body image acquisition sub-module is used to separately acquire human body images in various historical power operation scenarios;

   a standardized image matrix generation sub-module for performing image preprocessing on the human body image to generate a standardized image matrix;

   Dressing feature matrix generation sub-module for inputting the standardized image matrix into a preset initial human body dressing feature matrix extraction network to obtain a dressing feature matrix;

   The training sub-module is used for using a preset dual-core optimization algorithm in combination with the clothing feature matrix to train the initial human body clothing feature matrix extraction network to generate a target human body clothing feature matrix extraction network.
9. The apparatus for recognizing clothing code for electrical work according to claim 8, wherein the preset dual-core optimization algorithm includes an overall comparison function and a patch comparison function, and the clothing feature matrix includes an overall clothing feature matrix and a clothing patch Block feature matrix, the training submodule includes:

   The standard overall dress feature matrix acquisition sub-module is used to obtain the standard overall dress feature matrix;

   an overall comparison function processing submodule, used for importing the standard overall dressing feature matrix and the overall dressing feature matrix into the overall comparison function to obtain an optimization target value and a tracking target value;

   An overall parameter adjustment sub-module for adjusting the overall parameters of the initial human clothing feature matrix extraction network based on the optimization target value and the tracking target value using a back-propagation algorithm until the tracking target value is equal to the optimization target value;

   Standard dress patch feature matrix acquisition sub-module, used to obtain standard dress patch feature matrix;

   The patch comparison function processing sub-module is used to import the dressing patch feature matrix and the standard dressing patch feature matrix into the patch comparison function to obtain the patch optimization target value and the patch tracking target value ;

   The patch parameter adjustment sub-module is used to adjust the patch parameters of the initial human clothing feature matrix extraction network based on the patch optimization target value and the patch tracking target value by using the back-propagation algorithm until all the patch parameters are extracted. the patch tracking target value is equal to the patch optimization target value;

   The target body clothing feature matrix extraction network generation sub-module is used to generate the target body clothing feature matrix when the tracking target value is equal to the optimization target value and the patch tracking target value is equal to the patch optimization target value Extract the network.
10. The apparatus for identifying the dress code for electrical work according to claim 6, wherein the preset condition includes a plurality of sub-conditions, and the personnel dress code judgment module includes:

    A non-conformity determination sub-module, used for determining that the clothing of the person corresponding to the to-be-recognized human body image does not conform to the specification if the human body clothing feature matrix satisfies any one of the sub-conditions;

    The specification-compliant determination sub-module is configured to determine that the clothing of the person corresponding to the to-be-recognized human body image conforms to the specification if the human body clothing feature matrix does not satisfy all of the sub-conditions.

Images