Method and system for monitoring operation behaviors of construction personnel of power distribution network

Abstract

The invention discloses a power distribution network construction personnel operation behavior monitoring method and system, and the method comprises the steps: constructing a first data set and a second data set, the first data set being a personnel data set, and the second data set being a behavior data set; training a first recognition model and a second recognition model according to the first data set and the second data set; taking the output of the trained first recognition model as the input of a second recognition model, and establishing a new third recognition model; and monitoring the operation behavior of the power distribution network constructor according to the third identification model. Compared with the prior art, the method has the advantages that through multi-level feature map information extraction, the model can more accurately extract the characteristics of the power distribution network construction site, so that the construction personnel identification accuracy and the construction personnel behavior detection accuracy in a complex power distribution network site construction environment are effectively improved; and through continuous learning and training, the accuracy and efficiency of identification and behavior detection of the power distribution network constructors can be further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring the working behavior of distribution network construction personnel, and in particular to a method and system for monitoring the working behavior of distribution network construction personnel. Background Art

[0002] In the power system, the safe operation of distribution network construction workers is an important part of ensuring the stable operation of the power grid. However, traditional manual monitoring methods often have problems such as low efficiency and difficulty in ensuring accuracy. Especially in complex and changeable working environments, it is difficult to fully guarantee the safety of construction workers. Traditional manual inspection methods require inspectors to go to the site in person for inspection, which is not only time-consuming and labor-intensive, but also inefficient. In existing solutions, image acquisition equipment is first used for image acquisition, and subsequent image analysis and recognition work still requires manual participation, resulting in low inspection efficiency.

[0003] Existing image processing technology is not ideal for identifying distribution network construction personnel and detecting the behavior of distribution network personnel in complex environments. The reason is that the number of distribution network construction personnel is large, the structure is complex, and it may be affected by many factors. This makes the ability to accurately identify, classify and locate distribution network construction personnel limited, resulting in the inability to accurately monitor and inspect distribution network construction personnel. 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 monitoring the working behavior of distribution network construction 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 monitoring the operation behavior of distribution network construction personnel, comprising:

[0009] Constructing a first data set and a second data set, wherein the first data set is a personnel data set, and the second data set is a behavior data set;

[0010] Training a first recognition model and a second recognition model according to the first data set and the second data set;

[0011] Using the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model;

[0012] The operating behavior of distribution network construction personnel is monitored according to the third recognition model.

[0013] As a preferred solution of the method for monitoring the working behavior of distribution network construction personnel of the present invention, wherein: the training of the first recognition model and the second recognition model according to the first data set and the second data set includes:

[0014] The first recognition model is any model whose input is the first data set and whose output is the person recognition result or parameters related to the person recognition result;

[0015] The second recognition model is any model whose input is the second data set and whose output is the personnel behavior result or parameters related to the personnel behavior result.

[0016] As a preferred solution of the method for monitoring the operating behavior of distribution network construction personnel according to the present invention, the third recognition model includes:

[0017] The third recognition model is any model whose input is the output of the first recognition model and whose output is the personnel behavior result or parameters related to the personnel behavior result.

[0018] As a preferred solution of the method for monitoring the working behavior of distribution network construction personnel described in the present invention, the first recognition model includes a feature extraction module configured in the first recognition model.

[0019] As a preferred solution of the method for monitoring the working behavior of distribution network construction personnel described in the present invention, the first recognition model also includes a feature perception area enhancement module configured in the first recognition model.

[0020] As a preferred solution of the method for monitoring the working behavior of distribution network construction personnel described in the present invention, the first recognition model also includes a feature fusion module configured in the first recognition model.

[0021] As a preferred solution of the method for monitoring the operating behavior of distribution network construction personnel according to the present invention, the monitoring of the operating behavior of distribution network construction personnel according to the third recognition model includes:

[0022] Presetting a confidence threshold, and comparing the confidence threshold with an output of a third recognition model;

[0023] The operating behavior of distribution network construction personnel is monitored based on the comparison results.

[0024] In a second aspect, the present invention provides a distribution network construction personnel operation behavior monitoring system, comprising:

[0025] A data set construction module, used to construct a first data set and a second data set, wherein the first data set is a personnel data set and the second data set is a behavior data set;

[0026] A model training module, used for training a first recognition model and a second recognition model according to the first data set and the second data set;

[0027] A model building module, used to use the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model;

[0028] A monitoring module is used to monitor the operating behavior of distribution network construction personnel based on the third recognition model.

[0029] 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.

[0030] 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.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes a method and system for monitoring the work behavior of distribution network construction personnel, constructs a first data set and a second data set, the first data set is a personnel data set, and the second data set is a behavior data set; trains a first recognition model and a second recognition model according to the first data set and the second data set; uses the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model; monitors the work behavior of distribution network construction personnel according to the third recognition model. Compared with the prior art, the present application extracts multi-level feature map information, so that the model can more accurately extract the distribution network construction site features, thereby effectively improving the accuracy of construction personnel identification and the accuracy of construction personnel behavior detection in complex distribution network on-site construction environments, and through continuous learning and training, the accuracy and efficiency of distribution network construction personnel identification and behavior detection can be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] 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:

[0033] Figure 1 A method flow chart of a method and system for monitoring the working behavior of distribution network construction personnel provided by one embodiment of the present invention;

[0034] Figure 2 A network diagram of a distribution network construction personnel identification model of a distribution network construction personnel operation behavior monitoring method and system provided by an embodiment of the present invention;

[0035] Figure 3 A Gs_Block module structure diagram of a method and system for monitoring the operation behavior of construction personnel in a distribution network provided by an embodiment of the present invention;

[0036] Figure 4 A RFEM module structure diagram of a method and system for monitoring the working behavior of distribution network construction personnel provided by one embodiment of the present invention;

[0037] Figure 5 A FFM module structure diagram of a method and system for monitoring the operation behavior of construction personnel in a distribution network provided by an embodiment of the present invention;

[0038] Figure 6 A network diagram of a distribution network construction personnel behavior detection model of a distribution network construction personnel operation behavior monitoring method and system provided by an embodiment of the present invention;

[0039] Figure 7 An internal structural diagram of a computer device of a method and system for monitoring the working behavior of distribution network construction personnel provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0040] 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.

[0041] Example 1

[0042] Reference Figure 1-Figure 7, which is the first embodiment of the present invention, provides a method and system for monitoring the operation behavior of distribution network construction personnel, including:

[0043] There are some shortcomings in the existing related technologies. For example, at the construction site of the distribution network, due to the complex and changeable environment and the diverse behaviors of the construction workers, the traditional monitoring methods often have difficulty in accurately identifying the behaviors of the construction workers, resulting in low monitoring efficiency and even possible false alarms or missed alarms. 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 distribution network construction worker operation behavior monitoring method;

[0044] Figure 1 A method flow chart of a method and system for monitoring the operation behavior of distribution network construction personnel is shown, including:

[0045] S101, constructing a first data set and a second data set, where the first data set is a personnel data set and the second data set is a behavior data set;

[0046] In an optional embodiment, the personnel data set can be obtained by collecting image information of personnel at the construction site and obtaining characteristic information of the personnel, such as height, body shape, clothing, etc., through manual labeling; the behavior data set can capture the behavior images of construction personnel in different working states through a video surveillance system and label them, including but not limited to the wearing of safety helmets, use of tools, working posture, etc.

[0047] In an optional embodiment, a camera may be used to collect images of the distribution network construction site to construct a distribution network construction personnel image dataset, i.e., a personnel dataset;

[0048] In an optional embodiment, specific information of construction personnel can also be directly retrieved from the network system of the distribution network construction site to obtain detailed personnel data, such as name, work number, position, etc., to enrich the content of the personnel data set.

[0049] In the embodiment of the present application, a high-definition surveillance camera is used to collect images of the distribution network construction site to construct a distribution network construction personnel image dataset, i.e., a personnel dataset. The constructed distribution network construction personnel dataset is manually selected to mark the distribution network construction personnel in the image;

[0050] In an optional embodiment, a deep learning-based annotation tool can be used to annotate the distribution network construction workers in the image. The tool can automatically identify the construction workers in the image and mark their key points, such as the head, hands, feet, etc., to assist manual more accurate annotation. In this way, the annotation efficiency and accuracy can be significantly improved, providing a higher quality data set for subsequent model training.

[0051] In an optional embodiment, an automatic annotation system based on machine learning can also be used to annotate the distribution network construction personnel in the image. The system first performs a preliminary analysis of the image through a pre-trained deep learning model, identifies the construction personnel in the image, and predicts the location of their key points. Then, the system compares these prediction results with the results of manual annotation, and continuously optimizes the model parameters by learning the accuracy of manual annotation to improve the accuracy of automatic annotation. Ultimately, the system can achieve rapid and accurate annotation of construction personnel in distribution network construction site images, providing efficient data support for subsequent monitoring methods.

[0052] In an optional embodiment, the distribution network construction personnel in the image may also be labeled using an image annotation tool, such as LabelMe image annotation tool and LabelImg, etc. These tools allow the user to draw a bounding box on the image and assign a label to each bounding box, thereby providing accurate annotation data for model training.

[0053] In an embodiment of the present application, the LabelMe image annotation tool is used to select the position of the distribution network construction personnel in the distribution network on-site construction image and mark its category as "distribution network construction personnel". LabelMe is a tool commonly used for image data annotation. Its annotation method is to select the position of the distribution network construction personnel in the distribution network on-site construction image and mark its category as "distribution network construction personnel".

[0054] In an optional embodiment, the behavior dataset can be captured by using a video surveillance system to capture and annotate the behavior images of construction workers in different working states. These behavior images include but are not limited to the wearing of safety helmets, the use of tools, working postures, etc. In this way, a dataset containing multiple behavior features can be constructed to provide rich training samples for the behavior recognition model.

[0055] In an optional embodiment, the behavior dataset can also obtain the behavior characteristics of construction workers in different working states by collecting existing distribution network construction workers' behavior image datasets and combining them with manual annotation. These datasets can include behavior images of construction workers in various environments and conditions, such as working in bad weather conditions, working in narrow spaces, and working at night. In this way, the diversity and comprehensiveness of the behavior dataset can be ensured, thereby improving the adaptability and accuracy of the monitoring system. In addition, the content of the behavior dataset can be further enriched by simulating different construction scenarios and using actors to play the role of construction workers for behavior simulation. Ultimately, these behavior datasets will be used to train and verify the behavior recognition model to ensure that the model can accurately identify and classify various work behaviors of construction workers.

[0056] In an embodiment of the present application, the behavior data set can be obtained by collecting existing distribution network construction personnel behavior image data sets, including but not limited to collecting the above data sets from various public channels, and then constructing the distribution network construction personnel behavior image data set of the present invention.

[0057] In the embodiment of the present application, the data set contains images of the behavior of distribution network construction personnel, including the following three sub-data sets related to the behavior of distribution network construction personnel: a data set of images of safety helmets of distribution network construction personnel, a data set of images of work clothes of distribution network construction personnel, and a data set of images of bad construction behavior of distribution network construction personnel. The collected data set of behavior of distribution network construction personnel (i.e., behavior data set) is annotated according to the behavior of the construction personnel, and the specific annotation types include "distribution network construction personnel are not wearing safety helmets", "distribution network construction personnel are not wearing work clothes", and "distribution network construction personnel have bad construction behavior at the construction site".

[0058] It should be noted that in actual construction, public datasets can be collected through roboflow or other open source dataset platforms. Roboflow is a large-scale computer vision platform from which a large number of open source datasets can be obtained.

[0059] It should also be noted that the construction of the first data set and the second data set can provide more comprehensive and accurate information on personnel behavior, so that the monitoring system can more effectively identify and analyze the behavior patterns of construction personnel. By constructing a detailed data set containing personnel characteristics and behavior characteristics, the monitoring system can realize real-time monitoring and analysis of the work behavior of construction personnel, and timely discover potential safety risks and violations, thereby improving the safety management level and work efficiency of the construction site. In addition, by continuously updating and expanding the data set, the monitoring system can adapt to different construction environments and conditions to ensure the accuracy and effectiveness of monitoring. Ultimately, the construction and application of these data sets not only help to improve the level of intelligence in the monitoring of the work behavior of distribution network construction personnel, but also provide strong technical support for construction safety management and quality control.

[0060] S102, training a first recognition model and a second recognition model according to the first data set and the second data set;

[0061] In an embodiment of the present application, training the first recognition model and the second recognition model according to the first data set and the second data set includes:

[0062] The first recognition model is any model whose input is the first data set and whose output is the person recognition result or parameters related to the person recognition result;

[0063] The second recognition model is any model whose input is the second data set and whose output is the personnel behavior result or parameters related to the personnel behavior result.

[0064] In an optional embodiment, the first recognition model and the second recognition model can be constructed and trained by using a deep learning framework, such as TensorFlow or PyTorch. These frameworks provide a wealth of tools and libraries that can help researchers quickly implement complex neural network structures and perform efficient model training and optimization. During the training process, a variety of techniques can be used to improve the generalization ability of the model, such as data enhancement, regularization, and transfer learning using pre-trained models. Data enhancement technology transforms training images by means of rotation, scaling, cropping, etc. to increase data diversity and reduce the risk of overfitting. Regularization techniques such as dropout or weight decay help prevent the model from overfitting on the training data. The use of pre-trained models can utilize the feature representation obtained by pre-training on large-scale data sets to accelerate the convergence speed of the model on specific tasks and improve the final recognition performance. Through these methods, the first recognition model and the second recognition model can more accurately identify the identity and behavior of construction personnel, providing strong technical support for the distribution network construction personnel operation behavior monitoring system.

[0065] In an optional embodiment, the first recognition model and the second recognition model can further improve performance by integrating other advanced machine learning algorithms. For example, algorithms such as support vector machine (SVM), random forest or gradient boosting decision tree can be used in combination to enhance the decision-making ability of the model when dealing with complex scenarios. These algorithms can handle nonlinear problems and perform well in feature selection and classification, which helps to improve the accuracy and robustness of the monitoring system in practical applications.

[0066] In an optional embodiment, the first recognition model and the second recognition model can also be designed in other ways, but the present application designs similar models, that is, the first recognition model whose input is the first data set and whose output is the personnel recognition result or parameters related to the personnel recognition result; the second recognition model whose input is the second data set and whose output is the personnel behavior result or parameters related to the personnel behavior result, should all be within the scope of protection of our application.

[0067] In an optional embodiment, the parameters related to the personnel identification result may include the probability of personnel identification or other parameters that can directly or indirectly obtain the personnel identification result, such as personnel identification tags, personnel identification confidence, etc. Similarly, the parameters related to the personnel behavior results may include behavior identification tags, behavior identification confidence, behavior identification timestamp, etc. These parameters are crucial for the monitoring system because they not only provide the direct results of behavior identification, but also provide additional information, such as the accuracy of identification and the time when the behavior occurred, which is very useful for subsequent security analysis and decision support.

[0068] In an optional embodiment, the parameters related to the results of personnel behavior may include the probability of personnel behavior results or other parameters that can directly or indirectly obtain the results of personnel behavior, such as behavior recognition tags, behavior recognition confidence, behavior recognition timestamp, etc. These parameters are crucial for the monitoring system because they not only provide the direct results of behavior recognition, but also provide additional information, such as the accuracy of recognition and the time when the behavior occurs, which is very useful for subsequent safety analysis and decision support. Through these parameters, the monitoring system can record and analyze the behavior patterns of construction personnel, timely discover and warn potential safety risks, thereby achieving more effective risk management and intervention during the construction process. In addition, these parameters can also be used to generate detailed construction personnel behavior reports, provide data support for construction management, help managers better understand the actual situation on the construction site, optimize the operation process, and improve the overall construction efficiency and safety level.

[0069] In an embodiment of the present application, the first recognition model and the second recognition model can be designed by using a deep learning framework.

[0070] In the embodiment of the present application, the first recognition model includes a feature extraction module configured in the first recognition model.

[0071] For example, a feature extraction module Gs_Block is designed. Gs_Block uses a lightweight convolution module Gs-Conv and combines linear variation to quickly generate a mirror feature image, such as Figure 3 As shown in the figure, the number of model parameters is reduced, thereby improving the overall operation speed of the model, and Gs_Block expands the receptive field of the detail features of the distribution network construction site image, thereby better capturing the key information in the image and enhancing the model's ability to recognize construction personnel.

[0072] Specifically, the distribution network construction site feature map T1 with a size of H×W×C is input to the Gs-Conv layer with a convolution kernel number of C, a size of 3×3, a padding value of 2, and a step size of 1, and then the result is input to the BN batch normalization layer and the Swish activation function layer to obtain a distribution network construction site feature map T2 with a size of H×W×C. T2 is subjected to a linear transformation (linearvariation) operation to generate a mirror feature image T3 with a size of H×W×C; then T3 is input to the Gs-Conv layer with a convolution kernel number of C, a size of 3×3, a padding value of 2, and a step size of 1, and then input to the BN batch normalization layer and the Swish activation function layer to obtain a distribution network construction site feature map T4 with a size of H×W×C.

[0073] T2=Swish(BN(GC 3×3 (T1)))

[0074] T3=LNV(T2)

[0075] T4=Swish(BN(GC 3×3 (T3)))

[0076] Among them, GC 3×3 represents Gs-Conv of size 3×3, BN represents batch normalization, and Swish represents the activation function.

[0077] Furthermore, T3 and T4 are channel-superimposed to obtain a distribution network construction site feature map T5 of size H×W×2C. T5 is input into the Gs-Conv layer with 2C convolution kernels, size 3×3, padding value 2, and step size 2, and then input into the BN batch normalization layer and the Swish activation function layer to obtain and output a distribution network construction site feature map T6 of size H / 2×W / 2×2C.

[0078] T5=Concat(T2,T4)

[0079] T6=Swish(BN(GC 3×3 (T4)))

[0080] Among them, GC 3×3 represents a Gs-Conv layer of size 3×3, Concat represents a channel superposition operation, BN represents a batch normalization layer, and a Swish activation function layer.

[0081] It should be noted that an image of size H×W×C refers to an image with a length of H pixels, a width of W pixels, and a number of channels of C. The number of channels of the feature map obtained after the image is calculated by the convolution layer is the same as the number of convolution kernels in the convolution layer. GS-Conv is an existing lightweight deep learning convolution module, which is mainly used to reduce the computational complexity and parameter amount of the model while maintaining or improving the performance of the model. The concat operation is used to connect two or more feature matrices together in a certain dimension to generate a larger feature matrix. For example, a feature matrix of size H×W1 and a feature matrix of size H×W2 are concat-operated to a feature matrix of size H×(W1+W2).

[0082] In the embodiment of the present application, the first recognition model also includes a feature receptive area enhancement module configured in the first recognition model.

[0083] For example, a feature receptive field enhancement module RFEM (Receptive Field Enhancement Module) is designed to associate each pixel in the distribution network construction site image, improve the global receptive field of each pixel in the feature image, and thus enhance the model's ability to identify construction personnel. The specific structure of the RFEM module is as follows: Figure 4 shown.

[0084] Among them, the input size of the distribution network construction site characteristic map L1 to the three different branches is the first branch, the second branch, and the third branch. In the second branch, L1 is input into the Conv layer with the number of convolution kernels of C, the size of 3×3, the padding value of 2, and the step size of 1 to obtain the distribution network construction site characteristic map L2 with the size of H×W×C. In the third branch, L1 is input into the Conv layer with the number of convolution kernels of C, the size of 5×5, the padding value of 3, and the step size of 1 to obtain the distribution network construction site characteristic map L3 with the size of H×W×C.

[0085] L2=C 3×3 (L1)

[0086] L3=C 5×5 (L1)

[0087] Among them, C 3×3 represents a Conv layer of size 3×3, C 5×5 Represents a Conv layer of size 5×5.

[0088] Furthermore, L2 and L3 are channel-superimposed to obtain a distribution network construction site feature map L4 of size H×W×2C. Then L4 is input into a Conv layer with C convolution kernels, size 3×3, padding value 2, and step size 1 to obtain a distribution network construction site feature map L5 of size H×W×C.

[0089] L4=Concat(L2,L3)

[0090] L5=C 3×3 (L4)

[0091] Among them, Concat represents the channel superposition operation.

[0092] Furthermore, L1 and L5 are channel-superimposed, and the result is input into a CBS module with C convolution kernels, a size of 3×3, a padding value of 2, and a step size of 1 to obtain and output a distribution network construction site feature map L6 with a size of H×W×C.

[0093] L6=CBS 3×3 (Concat(L1,L5))

[0094] Among them: CBS 3×3 It means that the calculation is performed sequentially through a 3×3 Conv layer (convolution layer), a BN layer (normalization layer), and a Swish (activation function layer).

[0095] It should be noted that the CBS module is a commonly used deep learning module, which consists of Conv layer (convolution layer), BN layer (batch normalization layer) and Swish (activation function layer), and is mainly used for feature extraction and conversion. It extracts the features of the input data through convolution operation, uses BN layer for normalization, and provides nonlinear characteristics through Swish activation function.

[0096] In the embodiment of the present application, the first recognition model also includes a feature fusion module configured in the first recognition model.

[0097] For example, a feature fusion module (FFM) is designed. FFM extracts important features from the image by inputting the dual-branch distribution network construction site image, such as Figure 5 As shown in the figure, the dual-branch distribution network construction site features are fused through the Concat operation to improve the model's ability to identify distribution network construction personnel. The dual-branch input of FFM strengthens the redundancy of model recognition. When the distribution network construction site feature map input at one end is interfered by noise, the feature map input at the other end can still provide effective feature information support. This redundancy makes the model more robust and fault-tolerant when facing complex and changeable inputs, and can output construction personnel identification results more stably and reliably. In addition, FFM also improves the randomness of image features through shuffle operations, improves the generalization performance of the network, and avoids the gradient explosion problem when updating the model training weights due to regular data, thereby preventing the model from overfitting or underfitting.

[0098] Among them, the distribution network construction site feature maps A1 and A2 are input from the fourth branch and the fifth branch respectively, and the size of the feature maps A1 and A2 are both H×W×C. A1 is input into the CBS module with C convolution kernels, size 3×3, padding value 2, and step size 1, and the distribution network construction site feature map A3 with a size of H×W×C is obtained. A2 is input into the CBS module with C convolution kernels, size 3×3, padding value 2, and step size 1, and the distribution network construction site feature map A4 with a size of H×W×C is obtained.

[0099] A3=CBS 3×3 (A1)

[0100] A4=CBS 3×3 (A2)

[0101] A3 and A4 are first shuffled and then channel-superimposed to obtain a distribution network construction site feature map A5 of size H×W×2C. Finally, A5 is input into a CBS module with 2C convolution kernels, size 3×3, padding value 2, and step size 1 to obtain and output a distribution network construction site feature map A6 of size H×W×2C.

[0102] A5=Concat(shuffle(A3),shuffle(A4))

[0103] A6=CBS 3×3 (A5)

[0104] Among them: Concat represents the channel superposition operation, and shuffle represents the shuffle operation.

[0105] It should be noted that the shuffle operation is a deep learning operation that randomly shuffles the order of samples in the data set. The purpose of this is to ensure that the samples seen by the model in each iteration during the training process are random, thereby preventing the model from learning the sequential characteristics of the data and improving the generalization ability of the model.

[0106] In an optional embodiment, a distribution network construction personnel identification model is proposed based on the Gs_Block module, the RFEM module and the FFM module. The distribution network construction site image is input into the model, and the model automatically identifies and locates the construction personnel in the construction site image. The model network diagram is shown in Figure 2 As shown, the specific implementation process can be as follows:

[0107] Input a distribution network construction site image F1 to the CBS module to obtain the distribution network construction site feature map F2. Input F2 to the Gs_Block module for feature extraction to obtain the distribution network construction site feature map F3. Input F3 to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F4. Input F4 to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F5. Input F5 to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F6. Input F6 to the SPPF to obtain the distribution network construction site feature map F7.

[0108] Furthermore, F7 is interpolated linearly to obtain the distribution network construction site characteristic diagram F8. F5 and F8 are input into FFM for calculation, and then the calculation results are input into the RFEM module to obtain the distribution network construction site characteristic diagram F9. 10 . 10Input FFM to calculate, and then input the calculation results into RFEM module to obtain the distribution network construction site characteristic diagram F 11 . 11 Enter FFM for calculation, and then enter the calculation results into the RFEM module to obtain the distribution network construction site characteristic diagram F 12 . 12 Input FFM to calculate and obtain the distribution network construction site characteristic map F 13 .

[0109] Furthermore, the distribution network construction site characteristic map F 11 , feature map F 12 , feature map F 13 The image is input into the target detection head to obtain the distribution network construction personnel recognition result, which includes the category, regression box and confidence of the construction personnel in the image.

[0110] It should be noted that training the first recognition model and the second recognition model according to the first data set and the second data set enables the models to be trained on different data sets, thereby improving the accuracy and adaptability of the models in identifying distribution network construction personnel in different scenarios. The first data set may contain more types of construction environments and personnel postures, while the second data set may focus more on detailed features in a specific environment. By combining the models trained with the two data sets, the advantages of both can be combined, so that the model can maintain a high recognition rate when facing various complex scenarios.

[0111] S103, using the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model;

[0112] In this embodiment of the present application, the third recognition model includes:

[0113] The third recognition model is any model whose input is the output of the first recognition model and whose output is the human behavior results or parameters related to the human behavior results.

[0114] In an optional embodiment, the third recognition model can be implemented by deep learning, machine learning or neural network technology. For example, a convolutional neural network (CNN) can be used as the infrastructure of the third recognition model to learn the behavior patterns of distribution network construction personnel through training. During the training process, the back propagation algorithm and the gradient descent optimization method can be used to adjust the network weights to minimize the difference between the predicted results and the actual behavior results.

[0115] In the embodiment of the present application, a distribution network construction personnel behavior detection model (i.e., a trained second recognition model) is constructed, such as Figure 6As shown in the figure, the construction personnel identification results output by the distribution network construction personnel identification model are input into the model to detect whether the distribution network construction personnel have any behavior. The model uses the dilated convolution AD-Conv to extract image features. While maintaining the size of the feature map, it increases the receptive field of each convolution kernel, which helps to extract more contextual information from the input feature map, thereby improving detection accuracy and efficiency.

[0116] Specifically, input the distribution network construction personnel feature map K1 of size H×W×C into two different branches, upper and lower. In the upper branch, K1 is input into the AD-Conv dilated convolution layer with C convolution kernels, size 3×3, padding value 2, and step length 1, and then the result is subjected to the Avg-Pooling average pooling operation to obtain the distribution network construction personnel feature map K2 of size H×W×C. In the lower branch, K1 is input into the AD-Conv dilated convolution layer with C convolution kernels, size 3×3, padding value 2, and step length 1, and then the result is subjected to the Max-Pooling maximum pooling operation to obtain the distribution network construction personnel feature map K3 of size H×W×C.

[0117] K2=ADConv 3×3 (AvgPooling(K1))

[0118] K3=ADConv 3×3 (MaxPooling(K1))

[0119] Where: ADConv 3×3 represents an AD-Conv hole convolution layer of size 3×3, AvgPooling represents average pooling, and MaxPooling represents maximum pooling.

[0120] The distribution network construction personnel feature maps K2 and K3 are channel-superimposed, and the results are input into the AD-Conv hole convolution layer with the number of convolution kernels C, size 3×3, padding value 2, and step size 1, to obtain the distribution network construction personnel feature map K4 with a size of H×W×C. K4 is input into the Full-connection layer to calculate and classify the distribution network construction personnel behavior results. Finally, the distribution network construction personnel behavior detection results are output.

[0121] It should be noted that the second recognition model and the third recognition model can be the same model or use the same model, because the input and output of the second recognition model and the third recognition model are the same, except that the input of the third recognition model is provided by the output of the first recognition model. The second recognition model and the third recognition model are used in this application to distinguish the two models. The essence of the third recognition model is to take the output of the first recognition model as input and input it into the second recognition model. It can also be considered as a combination of the first recognition model and the second recognition model.

[0122] S104: Monitor the operation behavior of the distribution network construction personnel according to the third recognition model.

[0123] In the embodiment of the present application, monitoring the operation behavior of distribution network construction personnel according to the third recognition model includes:

[0124] Presetting a confidence threshold, and comparing the confidence threshold with an output of a third recognition model;

[0125] The operating behavior of distribution network construction personnel is monitored based on the comparison results.

[0126] In an optional embodiment, after the distribution network construction personnel identification model and the distribution network construction personnel behavior detection model are trained, the two models are applied to the monitoring and early warning of the distribution network construction personnel behavior. First, ensure that the model can receive the construction personnel behavior video frame data from the distribution station in real time, and then set reasonable identification and detection thresholds in the system to ensure that the confidence of the construction personnel identity recognition and behavior detection meets the requirements. Once the model identifies the distribution network construction personnel or detects the behavior, and the confidence exceeds the preset alarm threshold, the system will automatically trigger the alarm mechanism. The mechanism will record the specific time of the behavior and the key details such as the identity and location of the construction personnel involved in the distribution station, and then immediately inform the on-duty personnel, relevant team managers and security personnel through multiple channels such as emails, text messages or system notifications to ensure timely identification and response to potential safety risks. This can effectively improve the supervision of the behavior of distribution network construction personnel, avoid safety problems that may be caused by behavior, and further ensure the stable operation and safety of the distribution network.

[0127] In an optional embodiment, a distribution network construction personnel data set is allocated to train and verify the distribution network construction personnel identification model. The constructed distribution network construction personnel data set is divided into a training set and a verification set in a ratio of 4:1, and the distribution network construction personnel identification model is trained. First, all parameters of the network are initialized and training-related hyperparameters are configured. The training set and the verification set will be divided into several small batches, and a batch of training data will be input into the model for learning each time, and the training loss value of the batch will be calculated and recorded. Then, the batch data of the verification set is input into the model, and the corresponding verification loss value is calculated. The algorithm self-adjusts and optimizes according to the loss situation each time. As the training proceeds, the loss value gradually converges, and finally completes the training of the distribution network construction personnel identification model.

[0128] In an optional embodiment, a distribution network construction personnel behavior data set is allocated to train and verify the distribution network construction personnel behavior detection model. The constructed distribution network construction personnel behavior data set is divided into a training set and a verification set in a ratio of 4:1, and the distribution network construction personnel behavior detection model is trained. First, all parameters of the network are initialized, and training-related hyperparameters are configured. The training set and the verification set will be divided into several small batches, and each time a batch of training data is input into the model for learning, and the training loss value of the batch is calculated and recorded. Then, the batch data of the verification set is input into the model, and the corresponding verification loss value is calculated. The algorithm self-adjusts and optimizes according to the loss situation each time. As the training proceeds, the loss value gradually converges, and finally completes the training of the distribution network construction personnel behavior detection model.

[0129] In summary, the present invention proposes a method for monitoring the work behavior of distribution network construction personnel, constructing a first data set and a second data set, wherein the first data set is a personnel data set, and the second data set is a behavior data set; training a first recognition model and a second recognition model according to the first data set and the second data set; using the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model; and monitoring the work behavior of distribution network construction personnel according to the third recognition model. Compared with the prior art, this application extracts multi-level feature map information, so that the model can more accurately extract the distribution network construction site features, thereby effectively improving the accuracy of construction personnel identification and the accuracy of construction personnel behavior detection in a complex distribution network on-site construction environment, and through continuous learning and training, the accuracy and efficiency of distribution network construction personnel identification and behavior detection can be further improved.

[0130] Example 2

[0131] This embodiment also provides a distribution network construction personnel operation behavior monitoring system, including:

[0132] A data set construction module, used to construct a first data set and a second data set, the first data set is a personnel data set, and the second data set is a behavior data set;

[0133] A model training module, used for training a first recognition model and a second recognition model according to the first data set and the second data set;

[0134] A model building module, used to use the output of the first recognition model after training as the input of the second recognition model to establish a new third recognition model;

[0135] The monitoring module is used to monitor the operating behavior of distribution network construction personnel according to the third recognition model.

[0136] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0137] 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 monitoring the work behavior of distribution network construction personnel 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 button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0138] 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:

[0139] Constructing a first data set and a second data set, wherein the first data set is a personnel data set, and the second data set is a behavior data set;

[0140] Training a first recognition model and a second recognition model according to the first data set and the second data set;

[0141] The output of the first recognition model after training is used as the input of the second recognition model to establish a new third recognition model;

[0142] The operating behavior of distribution network construction personnel is monitored based on the third identification model.

[0143] Example 3

[0144] Reference Figure 1-Figure 6 , which is an embodiment of the present invention, provides a method and system for monitoring the working behavior of distribution network construction personnel. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0145] In this embodiment, a high-definition surveillance camera is used to collect 550 distribution network on-site construction images for constructing a distribution network construction personnel dataset, and the LabelMe image annotation tool is used to annotate the coordinate frame information of the distribution network construction personnel positions in the distribution network on-site construction images. Each image contains one or more distribution network construction personnel position targets, and the distribution network construction personnel targets need to be anchored multiple times. After all distribution network construction personnel targets are annotated, 550 annotation files in XML format can be obtained. The name of each annotation file is consistent with the name of the corresponding distribution network on-site construction image file, and each annotation file can contain annotation anchor frame information and category information of multiple targets.

[0146] Public data sets were collected through roboflow to construct a distribution network construction personnel behavior data set. The data set contains distribution network construction personnel behavior images, including the following sub-datasets: 376 images of distribution network construction personnel safety helmet data set, 234 images of distribution network construction personnel work clothes data set, and 1,700 images of distribution network construction personnel on-site bad construction behavior data set, a total of 2,310 images. "Distribution network construction personnel did not wear safety helmets", "Distribution network construction personnel did not wear work clothes", "Distribution network construction personnel had bad construction behaviors at the construction site".

[0147] The constructed distribution network construction personnel data set is divided into a training set and a validation set at a ratio of 4:1, that is, a training set of 440 distribution network construction personnel data and a test set of 110. The distribution network construction personnel recognition model designed according to this scheme is used to train the distribution network on-site construction images to complete the construction of the distribution network construction personnel identification model. First, initialize the various parameters and hyperparameters of the algorithm training. This embodiment describes several important parameters in the algorithm training: the root initialized optimizer is the AdaGrad optimizer, the initialization training epoch is 500, the batch_size is 32, and the initial learning rate is 0.0015. These parameters need to be adaptively adjusted according to the results of multiple algorithm training until the algorithm modeling achieves the optimal effect.

[0148] The constructed distribution network construction personnel behavior data set is divided into a training set and a validation set at a ratio of 4:1, that is, 1848 distribution network construction personnel behavior training sets and 462 test sets. The distribution network construction personnel classification model designed according to this scheme is used to train the distribution network on-site construction images to complete the construction of the distribution network construction personnel classification model. First, initialize the various parameters and hyperparameters of the algorithm training. This embodiment describes several important parameters in the algorithm training: the root initialization optimizer is the Adam optimizer, the initialization training epoch is 500, the batch_size is 32, the initial learning rate is 0.001, and the weight initialization method is He initialization. These parameters need to be adaptively adjusted according to the results of multiple algorithm training until the algorithm modeling achieves the optimal effect.

[0149] After completing the initialization of the basic parameters of the algorithm training, the training of the distribution network construction personnel identification model is started. The distribution network on-site construction image is input into the distribution network construction personnel identification model of the present invention, assuming that the input is a distribution network construction site feature map F1. The distribution network construction site feature map F1 is input to the CBS module to obtain the distribution network construction site feature map F2. F2 is input to the Gs_Block module for feature extraction to obtain the distribution network construction site feature map F3, and F3 is input to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F4. F4 is input to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F5. F5 is input to the Gs_Block module and the RFEM module to obtain the distribution network construction site feature map F6. F6 is input to the SPPF to obtain the distribution network construction site feature map F7. F7 is subjected to a linear interpolation operation to obtain the distribution network construction site feature map F8. F5 and F8 are used as the input of FFM and the results are input to the RFEM module to obtain the distribution network construction site feature map F9. The distribution network construction site characteristic map F is obtained by linear interpolation of F9. 10 . 10 As input to FFM, the results are input to the RFEM module to obtain the distribution network construction site characteristic diagram F 11 . 11 As input to FFM, the results are input to the RFEM module to obtain the distribution network construction site characteristic diagram F 12 . 12 As the input of FFM, the distribution network construction site characteristic map F 13 . The distribution network construction site characteristic map F 11 , feature map F 12 , feature map F 13And input into the target detection head head, the training of the detection algorithm of the present invention updates the parameters inside the algorithm through the back propagation of the loss function, and the confidence loss of the bounding box adopts the binary cross entropy loss. The above loss function is updated round by round in the gradient back propagation process of the algorithm training. The algorithm gradually converges in the training round by round, and the detection accuracy is continuously improved. After all rounds of training are completed, the algorithm will perform dual verification of precision and recall rate on the divided verification data, and use mAP50 as the evaluation standard for testing the effectiveness of algorithm training. Finally, the recognition result of distribution network construction personnel is obtained, and the recognition result of distribution network construction personnel includes the category, regression box, and confidence of the construction personnel in the image.

[0150] After completing the initialization of the basic parameters of the algorithm training, the distribution network construction personnel behavior detection model is trained. Assume that the input is a distribution network construction personnel feature map F1, and input the distribution network construction site feature map K1 of size H×W×C into the upper and lower branches. In the upper branch, K1 is input into the AD-Conv hole convolution layer with C convolution kernels, size 3×3, padding value 2, and step length 1, and then the result is subjected to the Avg-Pooling average pooling operation to obtain the distribution network construction personnel feature map K2 of size H×W×C. In the lower branch, K1 is input into the AD-Conv hole convolution layer with C convolution kernels, size 3×3, padding value 2, and step length 1, and then the result is subjected to the Max-Pooling maximum pooling operation to obtain the distribution network construction personnel feature map K3 of size H×W×C. The distribution network construction personnel feature maps K2 and K3 are channel-superimposed, and the results are input into the AD-Conv hole convolution layer with C convolution kernels, size 3×3, padding value 2, and step length 1, to obtain the distribution network construction personnel feature map K4 with size H×W×C. K4 is input into the Full-connection layer for result calculation and classification. Finally, the distribution network construction personnel behavior detection results are output.

[0151] The distribution network construction personnel behavior detection model training of the present invention updates internal parameters through the back propagation mechanism of the loss function, wherein the classification loss adopts the cross entropy loss function. During the algorithm training process, the model weights are gradually updated through gradient back propagation in each iteration. As the rounds increase, the model gradually converges and the classification accuracy continues to improve. After all training rounds are completed, the algorithm will use the divided validation set data for dual verification of accuracy and F1 score, and use the overall classification accuracy and average F1 score as evaluation criteria for testing the algorithm training effect. Finally, the model can output behavior classification results, including the category labels of the behaviors in the image and their corresponding confidence levels.

[0152] After the distribution network construction personnel identification model and the distribution network construction personnel behavior detection model are trained, the models are connected. The process is as follows: the distribution network on-site construction image is input into the distribution network construction personnel identification model to identify the distribution network construction personnel, and then the distribution network construction personnel identification results are input into the distribution network construction personnel behavior detection model to detect the behavior of the distribution network construction personnel.

[0153] The model is integrated into the existing distribution network field monitoring system to ensure that it can receive image or video data from the distribution network field area in real time. In the system, the alarm threshold is set to 0.6 in the system to capture the confidence of the distribution network construction personnel's behavior. Once the model detects that the distribution network construction personnel have bad behavior and the confidence exceeds the preset alarm threshold, the system will automatically trigger the alarm mechanism. The mechanism will record key details such as the specific time of the behavior and the identity and location of the construction personnel involved in the distribution station, and then immediately inform the on-duty personnel, relevant team managers and security personnel through multiple channels such as emails, text messages or system notifications to ensure timely identification and response to potential safety risks. This can effectively improve the supervision of the behavior of distribution network construction personnel, avoid safety problems that may be caused by behavior, and further ensure the stable operation and safety of the distribution network.

[0154] 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.

[0155] 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.

[0156] 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.

[0157] 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.

[0158] 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.

[0159] 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.

[0160] 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 monitoring the operation behavior of distribution network construction personnel, characterized in that: include: Constructing a first data set and a second data set, wherein the first data set is a personnel data set, and the second data set is a behavior data set; Training a first recognition model and a second recognition model according to the first data set and the second data set; Using the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model; The operating behavior of distribution network construction personnel is monitored according to the third recognition model.

2. The method for monitoring the operation behavior of distribution network construction personnel according to claim 1, characterized in that: The training of the first recognition model and the second recognition model according to the first data set and the second data set comprises: The first recognition model is any model whose input is the first data set and whose output is the person recognition result or parameters related to the person recognition result; The second recognition model is any model whose input is the second data set and whose output is the personnel behavior result or parameters related to the personnel behavior result.

3. The method for monitoring the operation behavior of distribution network construction personnel according to claim 2, characterized in that: The third recognition model includes: The third recognition model is any model whose input is the output of the first recognition model and whose output is the personnel behavior result or parameters related to the personnel behavior result.

4. The method for monitoring the operation behavior of distribution network construction personnel according to claim 3, characterized in that: The first recognition model includes a feature extraction module configured in the first recognition model.

5. The method for monitoring the operation behavior of distribution network construction personnel according to claim 4, characterized in that: The first recognition model also includes a feature receptive area enhancement module configured in the first recognition model.

6. The method for monitoring the operation behavior of distribution network construction personnel according to claim 5, characterized in that: The first recognition model also includes a feature fusion module configured in the first recognition model.

7. The method for monitoring the operation behavior of distribution network construction personnel according to claim 6, characterized in that: The monitoring of the operation behavior of the distribution network construction personnel according to the third recognition model includes: Presetting a confidence threshold, and comparing the confidence threshold with an output of a third recognition model; The operating behavior of distribution network construction personnel is monitored based on the comparison results.

8. A distribution network construction personnel operation behavior monitoring system, characterized in that: include: A data set construction module, used to construct a first data set and a second data set, wherein the first data set is a personnel data set and the second data set is a behavior data set; A model training module, used for training a first recognition model and a second recognition model according to the first data set and the second data set; A model building module, used to use the output of the trained first recognition model as the input of the second recognition model to establish a new third recognition model; A monitoring module is used to monitor the working behavior of distribution network construction personnel according to the third recognition 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