Power grid worker identity recognition method and system

Abstract

The invention discloses a power grid worker identity recognition method and system. The method comprises the following steps: constructing a target power distribution network worker identity recognition data set, a first feature data set, a second feature data set and a third feature data set of the target power distribution network worker identity recognition data set; establishing a first personnel identity recognition model based on the first feature, the second feature and the third feature; training the first personnel identity recognition model according to the first feature data set, the second feature data set and the third feature data set to obtain a second personnel identity recognition model; and performing power grid worker identity recognition according to the second worker identity recognition model. Through self-adaptive fusion of the face features and the gait features, the identity recognition accuracy can be greatly improved, and the identity recognition efficiency of the power grid workers is improved. Compared with a complex method, the recognition speed is higher, and compared with a traditional method, the recognition accuracy is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid staff identity recognition, and in particular to a power grid staff identity recognition method and system. Background Art

[0002] In the process of modern power grid construction and maintenance, ensuring accurate identification and supervision of on-site workers is the key to ensuring construction safety and stable operation of the power system. However, traditional identification methods such as access cards, fingerprints, and facial recognition face many challenges in real-time construction videos. First, the environment of power grid construction sites is complex and usually outdoors, affected by factors such as light and weather. Especially in strong light, shadows or bad weather conditions, the accuracy of traditional identification technology will be significantly reduced. Secondly, during the construction process, workers often wear protective equipment such as helmets and masks, which increase the difficulty of facial recognition. In addition, the changing postures of workers and the non-fixed video shooting angles further affect the recognition effect. Therefore, it is difficult for current technical means to achieve high-precision and real-time identity recognition of workers in construction videos in complex construction scenarios.

[0003] With the rapid development of deep learning and computer vision technology, video-based personnel recognition has shown good results in other fields. By introducing deep learning technology, real-time construction video data can be used to automatically detect and identify the identity of workers, and effectively deal with problems such as multiple angles, occlusions and environmental changes in complex scenes, providing a new technical direction for identity recognition in power grid construction scenes. The power grid worker identity recognition method based on deep learning can automatically extract identity features through large-scale data training models, adapt to complex and changing environments, and significantly improve the accuracy and robustness of power grid worker identity recognition. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method and system for identifying the identity of power grid personnel, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a method for identifying an identity of a power grid worker, comprising:

[0009] Constructing a target distribution network staff identity recognition data set, the target distribution network staff identity recognition data set comprising a first feature data set, a second feature data set and a third feature data set;

[0010] Establishing a first person identity recognition model based on the first feature, the second feature, and the third feature;

[0011] Training the first person identity recognition model according to the first feature data set, the second feature data set, and the third feature data set to obtain a second person identity recognition model;

[0012] The grid staff identity is identified according to the second personnel identity identification model.

[0013] As a preferred solution of the method for identifying the identity of power grid personnel of the present invention, the first personnel identification model includes:

[0014] The first personnel identity recognition model is any model that takes as input the first feature and the second feature and outputs the third feature or can directly or indirectly obtain personnel identity recognition related parameters of the third feature.

[0015] As a preferred solution of the method for identifying the identity of power grid personnel of the present invention, wherein: the first personnel identification model is trained according to the first feature data set, the second feature data set and the third feature data set to obtain the second personnel identification model comprises:

[0016] The first person identity recognition model and the second person identity recognition model are models of the same type or the same model;

[0017] The first person identity recognition model is an untrained model, and the second person identity recognition model is a trained first person identity recognition model.

[0018] As a preferred solution of the method for identifying the identity of power grid personnel described in the present invention, the first personnel identification model also includes a face extraction and enhancement module configured in the first personnel identification model.

[0019] As a preferred solution of the method for identifying the identity of power grid workers described in the present invention, the first personnel identification model also includes a gait multi-level feature extraction module configured in the first personnel identification model.

[0020] As a preferred solution of the method for identifying the identity of power grid workers described in the present invention, the first personnel identification model further includes a gait feature extraction module configured in the first personnel identification model.

[0021] As a preferred solution of the method for identifying the identity of power grid personnel described in the present invention, the first personnel identification model also includes a face and step fusion module configured in the first personnel identification model.

[0022] In a second aspect, the present invention provides a power grid staff identity recognition system, comprising:

[0023] A data set construction module, used to construct a target distribution network staff identity recognition data set, the target distribution network staff identity recognition data set includes a first feature data set, a second feature data set and a third feature data set;

[0024] A first model building module, used to build a first personnel identity recognition model based on the first feature, the second feature and the third feature;

[0025] A second model building module, used for training the first person identity recognition model according to the first feature data set, the second feature data set and the third feature data set to obtain a second person identity recognition model;

[0026] An identification module is used to identify the identity of the power grid staff according to the second personnel identity identification model.

[0027] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0028] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method described above when executed by a processor.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes a method and system for identifying power grid staff, constructs a target distribution network staff identification data set, the target distribution network staff identification data set includes a first feature data set, a second feature data set and a third feature data set; establishes a first personnel identification model based on the first feature, the second feature data set and the third feature data set; trains the first personnel identification model according to the first feature data set, the second feature data set and the third feature data set to obtain a second personnel identification model; and identifies power grid staff according to the second personnel identification model. By adaptively fusing facial features and gait features, the accuracy of identification can be greatly improved, a structure combining traditional three-dimensional convolution and pseudo three-dimensional convolution is introduced, and the efficiency of power grid staff identification is improved by combining attention mechanism and multi-scale feature extraction. Compared with complex methods, the recognition speed is faster, and compared with traditional methods, the recognition accuracy is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0031] Figure 1 A method flow chart of a method and system for identifying the identity of power grid personnel provided by one embodiment of the present invention;

[0032] Figure 2 A PGWINet network structure diagram of a method and system for identifying a power grid worker provided by an embodiment of the present invention;

[0033] Figure 3 A DCMPB (X) module structure diagram of a method and system for identifying a power grid worker provided by an embodiment of the present invention;

[0034] Figure 4 A FEEB module structure diagram of a method and system for identifying an identity of a power grid worker provided by an embodiment of the present invention;

[0035] Figure 5 A GMFEB module structure diagram of a method and system for identifying a power grid worker provided by an embodiment of the present invention;

[0036] Figure 6 A GFEB module structure diagram of a method and system for identifying a power grid worker provided by an embodiment of the present invention;

[0037] Figure 7 An internal structural diagram of a computer device of a method and system for identifying the identity of power grid personnel provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0038] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0039] Example 1

[0040] Reference Figure 1-Figure 7 , which is the first embodiment of the present invention, and provides a method and system for identifying the identity of a power grid worker, including:

[0041] There are some problems in the existing related technologies. For example, in complex working environments, such as power grid facilities, the identification of workers often faces many challenges. These challenges include but are not limited to changes in ambient light, protective equipment worn by workers, and identification of workers at different angles and distances.

[0042] This application provides a method that can effectively solve the above-mentioned problems. Next, we will combine multiple embodiments to explain in detail how to implement the grid staff identity recognition method;

[0043] Figure 1 A method flow chart of a method and system for identifying the identity of power grid staff is shown, including:

[0044] S101, constructing a target distribution network staff identity recognition data set, a first feature data set, a second feature data set and a third feature data set of the target distribution network staff identity recognition data set;

[0045] In an optional embodiment, the operation of identifying the identity of the target distribution network personnel can be judged by a variety of different features, such as facial information, human gait information, construction personnel identity ID, work clothes color, and the characteristics of the tools carried with them. The system can integrate these feature information through integrated multimodal recognition technology to improve the accuracy and robustness of recognition. For example, when changes in ambient light make it difficult to recognize facial information, the system can rely on human gait information or construction personnel identity ID for auxiliary recognition. In addition, the recognition of work clothes color and tool characteristics can be used as auxiliary information to further improve the recognition efficiency and accuracy in complex environments. Through this multi-feature fusion method, the system can effectively perform identity recognition even when the staff are wearing protective equipment or at different angles and distances.

[0046] In an embodiment of the present application, the construction worker identity ID is selected as the feature that ultimately needs to be identified, facial information and human gait information are selected as features to be identified, and a relationship between facial information, human gait information, and the construction worker identity ID is established.

[0047] In an optional embodiment, the relationship between face information, human gait information, and construction worker ID can be established by building a model. For example, a deep learning model can be built that can learn and extract face and gait features and associate these features with the construction worker ID. Through training, the model can learn the mapping relationship between different features, so as to accurately identify the identity of the worker in practical applications.

[0048] In an optional embodiment, the relationship between facial information, human gait information, and construction personnel identity ID can also be established by using a mapping relationship. For example, a mapping table can be created that records the correspondence between different facial information, gait features, and construction personnel identity ID. In actual operation, the system first captures the face and gait information of the staff member through a camera, and then compares this information with the data in the mapping table to quickly and accurately identify the identity of the staff member. This method not only improves the recognition speed, but also reduces the dependence on complex computing resources to a certain extent, making the system more suitable for scenarios with rapid on-site response.

[0049] In an optional embodiment, the relationship between facial information, human gait information, and construction personnel identity ID can also be established by using a tree structure. For example, a tree structure model can be constructed, which uses facial information and gait features as leaf nodes and the construction personnel identity ID as the root node. In the tree structure, each node represents a feature or a combination of features. By comparing and matching layer by layer, the root node is finally reached to determine the identity of a specific staff member. This method effectively reduces the search space and improves the recognition efficiency in a hierarchical manner, and is particularly suitable for large-scale power grid staff identity recognition scenarios. In addition, the tree structure model also has good scalability, and can easily add new features or update existing features to adapt to the ever-changing working environment and personnel characteristics.

[0050] In an embodiment of the present application, a first personnel identity recognition model based on the first feature, the second feature and the third feature is established through deep learning to perform feature recognition.

[0051] In an optional embodiment, the grid staff identification dataset can be constructed by obtaining an existing public grid staff identification dataset, for example, by collecting relevant images and identity information through web crawler technology, or by cooperating with the grid company to directly collect images and identity data of staff from the field. The collected data needs to be preprocessed, including operations such as cropping, scaling, and normalization of the image to ensure the quality and consistency of the dataset. In addition, the data needs to be labeled, that is, the correct identity label is assigned to each image. This step is crucial for training an accurate recognition model.

[0052] In an optional embodiment, the construction of the power grid staff identity recognition data set can also enhance the diversity of the data set and the reliability of the recognition system by collecting biometric data of field staff, such as fingerprints, irises, or voices. These biometric data can be combined with existing image data to form a more comprehensive multimodal data set. In practical applications, the system can flexibly select different feature combinations for identity verification to adapt to different working environments and recognition requirements. For example, in an environment with insufficient light, the system can give priority to iris recognition technology; in a noisy environment, it can rely on voice recognition technology. Through this fusion of multimodal biometric data, the power grid staff identity recognition system can provide more stable and accurate recognition results.

[0053] In an optional embodiment, the construction of the power grid staff identity recognition data set can also be carried out by collecting the staff's behavioral data in different work scenarios, such as the movements of operating equipment, walking paths, working hours, etc. These behavioral data can reflect the daily behavior patterns of the staff and provide additional verification information for identity recognition. For example, the system can analyze the walking paths and operating equipment patterns of the staff and match them with known staff behavior patterns to assist in confirming the identity. In addition, by collecting and analyzing these behavioral data over a long period of time, the system can also learn the behavioral habits of the staff and further improve the accuracy and security of recognition. In practical applications, this recognition method based on behavioral patterns can be used as a supplement to other feature recognition methods to enhance the robustness of the entire power grid staff identity recognition system.

[0054] In the embodiment of the present application, a grid staff identity recognition dataset is constructed, and an existing public grid staff identity recognition dataset is used. Grid construction video data is collected (such as on-site shooting or using an existing public dataset), and the collected video data is converted into frame-by-frame pictures. The pictures are grouped according to a fixed number of frames T, and each group of pictures is annotated by LabelMe in frame order. LabelMe is used to annotate the face area, gait contour area, and identity ID of the staff in the image. After the annotation is completed, the grid staff identity recognition dataset is obtained.

[0055] For example, any element in the data set is represented by a triple<P,G,Y> Represents, where the face sequence P is the face area position of the staff member with identity ID Y arranged in frame order in a set of pictures, the gait sequence G is the gait contour area position of the staff member with identity ID Y arranged in frame order in a set of pictures, and Y is the corresponding staff member's identity ID.

[0056] In an optional embodiment, LabelMe can be used as a labeling tool, and other labeling tools, such as CVAT (Computer Vision Annotation Tool) or VGG Image Annotator, etc., can also be used. These tools can also provide accurate image labeling functions. When using these tools, the operator can manually mark the face and gait areas and associate these areas with the corresponding identity IDs. After the labeling is completed, the data set will contain rich labeling information, providing a solid foundation for subsequent model training. In addition, these tools usually support batch labeling and automatic labeling functions, which can significantly improve the efficiency of data set construction. In actual operation, the most suitable labeling tool can be selected according to specific needs and resource conditions to achieve the best labeling effect and work efficiency.

[0057] It should be noted that constructing the target distribution network staff identity recognition dataset can ensure the accuracy and generalization ability of model training. In the process of constructing the dataset, it is also necessary to consider the diversity and representativeness of the data to ensure that the dataset can cover the various behaviors and appearance changes of power grid staff in different environments and conditions. For example, it is necessary to include images taken under different lighting conditions and at different angles, as well as images of staff wearing different work clothes, to ensure that the model can accurately identify the identity of the staff in practical applications.

[0058] S102, establishing a first person identity recognition model based on the first feature, the second feature and the third feature;

[0059] In the embodiment of the present application, the first person identity recognition model includes:

[0060] The first personnel identity recognition model is any model that takes as input the first feature and the second feature and outputs the third feature or can directly or indirectly obtain personnel identity recognition related parameters of the third feature.

[0061] In an optional embodiment, the first personnel identification model can be constructed by a deep learning framework, such as using a convolutional neural network (CNN) to extract features in the image. CNN can automatically learn and extract spatial hierarchical features in the image, which is crucial for identifying the identity of the staff.

[0062] In an optional embodiment, the first personnel identification model can also be constructed by other machine learning algorithms, such as support vector machine (SVM) or random forest. These algorithms can process features extracted from images and identify the identity of the staff through training. In practical applications, the most appropriate algorithm can be selected according to the size and complexity of the data set to achieve the best recognition effect.

[0063] In an optional embodiment, the first personnel identity recognition model can also realize the hierarchical representation of features by establishing a tree structure. The tree structure model can construct a decision tree based on the relationship between features, and identify the identity of the staff through layer-by-layer decision-making. This model is particularly suitable for processing data with hierarchical characteristics, and can effectively improve the accuracy and efficiency of recognition. When constructing a tree structure model, algorithms such as ID3, C4.5 or CART can be used, which can automatically construct a decision tree based on the feature information in the data set and quickly and accurately classify new samples. In addition, the tree structure model also has good interpretability, which can help understand which features are more important for identity recognition, thereby providing guidance for the optimization of the model. In practical applications, the tree structure model can be used in combination with other models to form a hybrid model to further improve the performance of the power grid staff identity recognition system.

[0064] In the embodiment of the present application, a deep learning-based power grid staff identity recognition network PGWINet is designed, such as Figure 2 As shown:

[0065] The staff face sequence P is input into PGWINet, and P is first processed by the DCMPB (64) module to obtain a face feature map F1, and the resolution of F1 is half of P. F1 is processed by the DCMPB (128) module to obtain a face feature map F2, and the resolution of F2 is half of F1.

[0066] It should be noted that DCMPB(64) refers to a DCMPB(X) module (Double Conv Max Pooling Block) with 64 convolution kernels in the conv3(X) convolution layer, and DCMPB(128) refers to a DCMPB(X) module with 128 convolution kernels in the conv3(X) convolution layer.

[0067] In an optional embodiment, a DCMPB(X) module is designed to extract global facial features, where X refers to the number of convolution kernels in the conv3(X) convolution layer in the module. That is, the conv3(X) convolution layer refers to a convolution layer with a convolution kernel size of 3×3 and a number of convolution kernels of X. The scale of the input feature map is gradually reduced after being processed by the DCMPB(X) module. Two conv3(X) convolution layers are used to extract facial features. MaxPooling is added after the convolution layer to reduce the size of the feature map while retaining important features. The application of the conv3(X) convolution layer can reduce the number of parameters while maintaining a certain scale of receptive field. For the input facial features, the DCMPB(X) module mainly focuses on the overall contour of the person's face, and by changing the number of channels, PGWINet learns richer features of the entire face. The module structure is as follows: Figure 3 shown.

[0068] In an optional embodiment, after being processed by the DCMPB (256) module, F2 is input into the FEEB (Facial Extraction Enhancement Block) module and the DCMPB (512) module to refine and enhance the facial features, thereby obtaining a facial feature map F3, the resolution of which is half of that of F2.

[0069] It should be noted that DCMPB(256) refers to a DCMPB(X) module in which the number of convolution kernels of the conv3(X) convolution layer is 256, and DCMPB(512) refers to a DCMPB(X) module in which the number of convolution kernels of the conv3(X) convolution layer is 512.

[0070] In an embodiment of the present application, the first person identity recognition model also includes a face extraction enhancement module (FEEB module) configured in the first person identity recognition model, and the face extraction enhancement module is used to implement a lightweight spatial attention mechanism (LSAM). The application of the FEEB module allows local features to receive higher attention, and local features often have rich identity information; unimportant features will be further ignored, thereby improving the performance of facial feature extraction. The FEEB module allows the network to pay more attention to image parts such as eyes, nose, mouth, etc. that are more meaningful in representing identity and more difficult to extract. The structure of the FEEB module is as follows: Figure 4 shown.

[0071] In an optional embodiment, in the FEEB module, a Conv(1x1) operation is first performed on the input feature map Input with a dimension of C×H×W to obtain a feature map with a dimension of The feature map S1 is then activated by the ReLU function to obtain S2, and the Conv(1x1) operation is performed on S2 to obtain the spatial weight map S3 with a dimension of 1×H×W. Finally, the spatial weighting of S3 and Input is calculated by Spatial-wise to obtain the output Output of the FEEB module. The calculation process of the FEEB module output Output can be described by the following formula.

[0072]

[0073] In an optional embodiment, for a spatial weight map S3 of dimension 1×H×W and an input feature map Input of dimension C×H×W, the spatial-wise operation first broadcasts the 1×H×W spatial weight map S3 to each channel to generate an S4 of dimension C×H×W. Specifically, each channel uses the same H×W attention map, that is, H×W is copied C times along the channel. Then S4 is element-wise multiplied by Input to obtain Output. The formula is:

[0074] O c,i,j =s c,i,j ×I c,i,j

[0075] In the formula, O c,i,j is the value of the i-th row and j-th column on the c-th channel of Output, s c,i,j is the value of the i-th row and j-th column on the c-th channel of S4, I c,i,j The value of the i-th row and j-th column on the c-th channel of Input.

[0076] In an optional embodiment, F1 is input into the FEEB module, the facial features are refined and enhanced, and a facial feature map F4 is obtained, and the resolution of F4 is the same as that of F1. F2 is input into the FEEB module, the facial features are refined and enhanced, and a facial feature map F5 is obtained, and the resolution of F5 is the same as that of F2.

[0077] In an optional embodiment, the obtained facial feature maps F3, F4, and F5 are restored to the resolution of the original input face sequence P through upsampling operations, and then concatenated (Concat) in the channel dimension. The concatenated result is converted into a single channel after a Conv (3x3) convolution operation to obtain a facial feature vector output R.

[0078] In an optional embodiment, the staff gait sequence G is input into PGWINet, and after being processed by Conv(3x3), G is input into the gait multilevel feature extraction module GMFEB (Gait Multilevel Feature Extraction Block) module to extract and refine the gait features to obtain a gait feature map F7, and the resolution of F7 is the same as that of G.

[0079] In an embodiment of the present application, the first personnel identity recognition model also includes a gait multi-level feature extraction module configured in the first personnel identity recognition model. It is used to deeply extract and refine the gait features of the input feature map. The GMFEB module is composed of a traditional three-dimensional convolution (Conv3D) and a pseudo three-dimensional convolution (Conv2D+Conv1D) structure. The present invention uses the traditional three-dimensional convolution for extracting multi-dimensional features of gait sequences as the backbone network, which has solved the problem that two-dimensional neural networks cannot effectively capture spatiotemporal features. In addition, in order to solve the overfitting problem caused by the stacking of three-dimensional convolution blocks, the present invention introduces a pseudo three-dimensional convolution as a branch, so as to enhance the backbone network's temporal feature extraction capability while also being able to extract spatial features. The module structure is as follows Figure 5 shown.

[0080] In an embodiment of the present application, in the GMFEB module, a Conv3D operation is first performed on the input feature map Input with a dimension of C×H×W to extract the spatiotemporal features of the Input and obtain a feature map C1. In order to refine the spatial features, a Conv2D operation is performed on C1 to continue to capture the spatial features and obtain a feature map C2. At the same time, a Conv1D operation is performed on C1, focusing on the local dynamic changes in the time dimension to obtain a feature map C3. A Conv2D operation is performed on the Input to continue to capture the spatial features and obtain a feature map C4. At the same time, a Conv1D operation is performed on the Input to focus on the local dynamic changes in the time dimension to obtain a feature map C5. A Concat operation is performed on C4 and C5, and a Conv3D operation is performed on the result of the Concat splicing to further extract the spatiotemporal features and obtain a feature map C6. A Concat operation is performed on C2, C3, and C6, and the result of the Concat splicing is activated by the ReLU function to obtain the output feature map Output of the GMFEB module. The calculation process of the output feature map Output of the GMFEB module can be described by the following formula.

[0081] C1=Conv3D(Input)

[0082] C2=Conv2D(C1)

[0083] C3=Conv1D(C2)

[0084] C4=Conv2D(Input)

[0085] C5=Conv1D(Input)

[0086] C6=Conv3D(Concat(C4,C5))

[0087] Output=ReLU(Concat(C2,C3,C6))

[0088] Conv3D in the above text is a three-dimensional convolution with a convolution kernel size of 3×3×3. Conv2D represents a two-dimensional convolution with a convolution kernel size of 3×3. Conv1D represents a one-dimensional convolution with a convolution kernel size of 1×1.

[0089] It should be noted that Concat represents the channel dimension concatenation operation in deep learning. When concatenating in the channel dimension, the length and width of the feature map must be the same, but the number of channels can be different. For example, if there are two feature maps A and B, their shapes are H×W×32 and H×W×64 respectively, then they can be concatenated in the channel dimension to obtain a new feature map with a shape of H×W×96. The size of the convolution kernel refers to its size in each dimension. The size of a three-dimensional convolution kernel is the "depth×height×width" of the convolution kernel, and the size of a two-dimensional convolution kernel is the "height×width" of the convolution kernel.

[0090] In an optional embodiment, after F7 is subjected to 3DMaxPooling to extract main features and reduce parameters and calculation amount, it is input into the GMFEB module to extract and refine the gait features of F7 to obtain a gait feature graph F8, and the resolution of F8 is the same as that of F7.

[0091] In an optional embodiment, F7 is processed by the GFEB (Gait Feature Extraction Block) module, and the gait features of F7 are extracted at multiple scales to obtain a gait feature graph F9, and the resolution of F9 is the same as that of F7. F8 is processed by the GFEB module, and the gait features of F8 are extracted at multiple scales to obtain a gait feature graph F 10 , F 10 The resolution is the same as F8. 10 Concatenation (Concat) is performed on the channel dimension, and the concatenated result is converted into a single channel after the convolution operation of Conv(3x3) to obtain the gait feature vector output B.

[0092] In an embodiment of the present application, the first personnel identity recognition model also includes a gait feature extraction module configured in the first personnel identity recognition model, which is used to perform multi-scale extraction of gait features, and then assign different weights to the obtained multi-scale features to generate features with significant discriminative properties. The reason for doing this is that in complex scenes, affected by the objects carried and occlusions, the use of a single scale to analyze the gait profile will affect the recognition accuracy. Therefore, the present invention analyzes image features from multiple scales, uses the GFEB module to decompose multiple features from different semantic levels, and extracts features with significant discriminative properties. Networks of different depths will respond to features at different levels, and features at different levels have the same importance. Therefore, in order to extract fine-grained features with significant discriminative properties, the present invention applies the GFEB module to different network depths respectively. The module structure is as follows: Figure 6 shown.

[0093] In an embodiment of the present application, in the GFEB module, the input feature map Input with a dimension of C×H×W is processed by the Conv(5x5)_BN_LReLU module, that is, the Conv layer with a convolution kernel size of 5×5, the BN layer, and the LReLU layer in sequence to extract the initial features and obtain a feature map output M1, and the resolution of M1 is half of that of Input. M1 is processed by the Conv(3x3)_BN_LReLU module, that is, the Conv layer with a convolution kernel size of 3×3, the BN layer, and the LReLU layer in sequence to extract features and downsample to obtain a feature map output M2, and the resolution of M2 is half of that of M1. M2 is processed by the Conv(3x3)_BN_LReLU module to extract features and downsample to obtain a feature map output M3, and the resolution of M3 is half of that of M2. M3 is processed by the Conv(1x1)_BN_LReLU module, i.e., the Conv layer with a convolution kernel size of 1×1, the BN layer, and the LReLU layer in sequence, extracting global features and adjusting the number of channels to obtain the feature map output M4, whose resolution is half of M3. M1, M2, M3, and M4 are respectively input into the CE_Upsampling module, upsampled to a resolution of H×W, and the upsampled results are concat-operated to obtain the feature map output M5. A global average pooling operation is performed on M5 to obtain a feature map with a dimension of The feature vector M6 is then Conv(1x1) operated on M6 to obtain a dimension of The feature vector M7 of M7 is converted into a conv(1x1) feature vector and a sigmoid activation feature vector. The channel weight vector M8 is obtained, and finally the channel weights of M5 and M8 are calculated by Channel-wise to obtain the output feature map Output of the GFEB module. The calculation process of the output feature map Output of the GFEB module can be described by the following formula.

[0094] M1=Conv(5×5)_BN_LReLU(Input)

[0095] M2=Conv(3×3)_BN_LReLU(M1)

[0096] M3=Conv(3×3)_BN_LReLU(M2)

[0097] M4=Conv(1×1)_BN_LReLU(M3)

[0098] M5=Concat(CE_Upsampling(M1,M2,M3,M4))

[0099] M8=Sig(Conv(Conv(GVPooling(M5))))

[0100] Output=Channel-wise(M5,M8)

[0101] For the dimension The channel weight vector M8 and dimension is In the channel-wise operation, the value at the cth position of M8 is the weight of channel c, and then the 1×1 weight W of each channel c is calculated. c Broadcast to each position of H×W to get Output. The formula is:

[0102] O c,i,j =W c ×M c,i,j

[0103] In the formula, O c,i,j is the value of the i-th row and j-th column on the c-th channel of Output, M c,i,j is the value of the i-th row and j-th column on the c-th channel of M5.

[0104] S103, training a first person identity recognition model according to the first feature data set, the second feature data set, and the third feature data set to obtain a second person identity recognition model;

[0105] In an embodiment of the present application, the first person identity recognition model is trained according to the first feature data set, the second feature data set, and the third feature data set to obtain the second person identity recognition model, including:

[0106] The first person identification model and the second person identification model are models of the same type or the same model;

[0107] The first person identity recognition model is an untrained model, and the second person identity recognition model is the trained first person identity recognition model.

[0108] In an optional embodiment, the constructed power grid staff identification network PGWINet is trained using the power grid staff identification dataset. The hyperparameters required for training the PGWINet network are initialized, such as the size of a batch sample during training, the number of training rounds, and the initial learning rate. During the training process, PGWINet optimizes the loss function by adjusting the weight values ​​in the network using the Adam algorithm, so that the loss function converges and the network reaches a balanced state.

[0109] S104: Perform identity recognition of a power grid worker according to a second personnel identity recognition model.

[0110] In an optional embodiment, after PGWINet training is completed, PGWINet is used to identify the identity of power grid workers. In actual applications, when accessing real-time power grid construction video, the target detection algorithm is used to detect the worker target frame by frame in the video, and the obtained face sequence and gait sequence are input into PGWINet, which processes the input sequence and outputs the personnel identification result.

[0111] It should be noted that mature target detection algorithms such as YOLO can be used to realize staff target detection in the video frame by frame. This is a common method in the field of deep learning. Since the present invention only involves staff identity recognition, the details of the staff target detection algorithm will not be described in detail here.

[0112] In summary, the present invention proposes a method for identifying power grid staff, constructs a target distribution network staff identification data set, the target distribution network staff identification data set includes a first feature data set, a second feature data set and a third feature data set; establishes a first personnel identification model based on the first feature, the second feature and the third feature; trains the first personnel identification model according to the first feature data set, the second feature data set and the third feature data set to obtain a second personnel identification model; and identifies power grid staff according to the second personnel identification model. By adaptively fusing facial features and gait features, the accuracy of identity recognition can be greatly improved, a structure combining traditional three-dimensional convolution and pseudo three-dimensional convolution is introduced, and the efficiency of power grid staff identification is improved by combining attention mechanism and multi-scale feature extraction. Compared with complex methods, the recognition speed is faster, and compared with traditional methods, the recognition accuracy is higher.

[0113] Example 2

[0114] This embodiment also provides a CE_Upsampling upsampling module that combines nearest neighbor interpolation and convolution operations, which is used to restore the resolution of the output feature map to the original input size. The upsampling formula of the CE_Upsampling module is as follows:

[0115] F up =σ(W*BilinearInterpolation(F s ,H t ,W t )+b)

[0116] First, bilinear interpolation is used to transform the feature map from low resolution Upsample to target resolution H t ×W t :

[0117] Finterp =BilinearInterpolation(F s ,H t ,W t )

[0118] In the formula, H t and W t are the width and height of the target.

[0119] Then, a convolutional layer is used to perform further feature learning on the upsampled feature map to enhance the details and feature expression capabilities:

[0120] F up =σ(W*F interp +b)

[0121] Where W is the convolution kernel, b is the bias term, and σ is the activation function (such as ReLU) used to introduce nonlinearity.

[0122] Furthermore, the facial feature vector output R and the gait feature vector output B are input into the FGFBB module, and the two features are adaptively fused to obtain the output of the power grid staff identity recognition result.

[0123] In the embodiment of the present application, the first person identification model further includes a face and gait fusion module (FGFBB module) configured in the first person identification model, which is used to adaptively fuse facial features and gait features. The feature fusion formula is as follows:

[0124] O=w1R+w2B

[0125] In the formula, O is the feature fusion output, is the face feature weight, is the gait feature weight, where μ is the mean and σ is the standard deviation.

[0126] Furthermore, w1 and w2 are obtained by solving the feature fusion optimization equation, and the equation definition is as follows, where E(X) represents the expectation of calculating X. According to the conditional extreme value Lagrangian theory, in the formula, σ 2 For w i The first-order partial derivative of is 0, and the total mean square error σ of feature fusion can be obtained. 2 Minimum, we can get w1 and w2.

[0127]

[0128] After obtaining the feature fusion output O, O is input into the final fully connected layer to obtain the output of the power grid staff identity recognition result.

[0129] Example 3

[0130] This embodiment also provides a power grid staff identity recognition system, including:

[0131] A data set construction module, used to construct a target distribution network staff identity recognition data set, a first feature data set, a second feature data set and a third feature data set of the target distribution network staff identity recognition data set;

[0132] A first model building module, used to build a first personnel identity recognition model based on the first feature, the second feature and the third feature;

[0133] A second model building module is used to train the first person identity recognition model according to the first feature data set, the second feature data set and the third feature data set to obtain a second person identity recognition model;

[0134] The identification module is used to identify the identity of the power grid staff according to the second personnel identity identification model.

[0135] The above-mentioned unit modules may be embedded in or independent of the processor in the computer device in the form of hardware, or may be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0136] This embodiment also provides a computer device, which may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for identifying the identity of a power grid worker is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0137] This embodiment further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0138] Constructing a target distribution network staff identity recognition data set, a first feature data set, a second feature data set and a third feature data set of the target distribution network staff identity recognition data set;

[0139] Establishing a first person identity recognition model based on the first feature, the second feature, and the third feature;

[0140] Training the first person identity recognition model according to the first feature data set, the second feature data set, and the third feature data set to obtain a second person identity recognition model;

[0141] The identity of the power grid staff is identified according to the second personnel identity identification model.

[0142] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0143] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0144] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0145] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0147] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0148] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for identifying the identity of a power grid worker, characterized in that: include: Constructing a target distribution network staff identity recognition data set, the target distribution network staff identity recognition data set comprising a first feature data set, a second feature data set and a third feature data set; Establishing a first person identity recognition model based on the first feature, the second feature, and the third feature; Training the first person identity recognition model according to the first feature data set, the second feature data set, and the third feature data set to obtain a second person identity recognition model; The grid staff identity is identified according to the second personnel identity identification model.

2. The method for identifying the identity of a power grid worker according to claim 1, characterized in that: The first personnel identity recognition model includes: The first personnel identity recognition model is any model that takes the first feature and the second feature as input and outputs the third feature or can directly or indirectly obtain personnel identity recognition related parameters of the third feature.

3. The method for identifying the identity of a power grid worker according to claim 2, characterized in that: The training of the first person identity recognition model according to the first feature data set, the second feature data set, and the third feature data set to obtain the second person identity recognition model includes: The first person identity recognition model and the second person identity recognition model are models of the same type or the same model; The first person identity recognition model is an untrained model, and the second person identity recognition model is a trained first person identity recognition model.

4. The method for identifying the identity of a power grid worker according to claim 3, characterized in that: The first person identity recognition model also includes a face extraction enhancement module configured in the first person identity recognition model.

5. The method for identifying the identity of a power grid worker according to claim 4, characterized in that: The first personnel identity recognition model also includes a gait multi-level feature extraction module configured in the first personnel identity recognition model.

6. The method for identifying the identity of a power grid worker according to claim 5, characterized in that: The first personnel identity recognition model also includes a gait feature extraction module configured in the first personnel identity recognition model.

7. The method for identifying the identity of a power grid worker according to claim 6, characterized in that: The first personnel identity recognition model also includes a face and step fusion module configured in the first personnel identity recognition model.

8. A power grid staff identification system, characterized in that: include: A data set construction module, used to construct a target distribution network staff identity recognition data set, the target distribution network staff identity recognition data set includes a first feature data set, a second feature data set and a third feature data set; A first model building module, used to build a first personnel identity recognition model based on the first feature, the second feature and the third feature; A second model building module, used for training the first person identity recognition model according to the first feature data set, the second feature data set and the third feature data set to obtain a second person identity recognition model; An identification module is used to identify the identity of the power grid staff according to the second personnel identity identification model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images