Power construction site target identification method and system based on deep learning network

Abstract

The invention relates to an electric power construction site target identification method and system based on a deep learning network. The method comprises the following steps: collecting photos of an electric power engineering construction site; the collected photos are input into a pre-constructed and trained image recognition network, target recognition is carried out on the electric power engineering construction site, a target recognition result is obtained, and the target recognition result comprises a bounding box, a confidence probability and a target category; and according to the identified bounding box, confidence probability and target category, performing target construction state judgment, and comparing with a pre-established construction safety knowledge graph to judge whether illegal operation or potential risk exists. Compared with the prior art, the real-time supervision capability of the construction site can be effectively improved, automatic identification of construction operation is realized, various conditions of the construction site are comprehensively evaluated and dealt with, the burden of manual supervision is relieved, and meanwhile, the objectivity and accuracy of judgment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of project construction, and in particular to a method and system for identifying targets at a power construction site based on a deep learning network. Background Art

[0002] At present, the judgment of illegal operations and construction safety at power engineering construction sites mainly relies on manual observation and supervision, which has a series of problems such as limited human resources, strong subjectivity, and lack of real-time performance, making it difficult to meet the growing needs of power engineering construction. In addition, the level of intelligent response at power engineering construction sites is low, often relying on the experience and subjective judgment of professionals, lacking automated and intelligent technical support, and unable to meet large-scale, high-density construction needs, and easily ignoring some potential illegal operations or safety risks. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and system for target recognition at a power construction site based on a deep learning network, so as to improve the level of real-time supervision at the construction site and realize automatic recognition of construction operations.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] A method for identifying targets at a power construction site based on a deep learning network comprises the following steps:

[0006] Collect photos of power engineering construction sites;

[0007] Input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include bounding boxes, confidence probabilities, and target categories;

[0008] The target construction status is judged based on the identified bounding box, confidence probability and target category, and compared with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0009] Furthermore, the training process of the image recognition network includes the following steps:

[0010] Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected;

[0011] Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity;

[0012] Network training steps: Use deep neural network as the learning framework to build an image recognition network. Input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters.

[0013] Furthermore, the deep neural network is a YOLOv5 network.

[0014] Furthermore, the image acquisition step also includes performing data enhancement on the acquired image by data rotation, flipping and scaling.

[0015] Furthermore, the target construction status determination is specifically as follows:

[0016] The confidence probability in each target recognition result is compared with a preset confidence threshold. If it is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted. In the same photo, distance and scene judgment are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status.

[0017] Furthermore, the obtained target construction status is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0018] Furthermore, the method also includes standardizing the collected photos of the power engineering construction site to obtain the standard image format required by the image recognition network.

[0019] The present invention also provides a power construction site target recognition system based on a deep learning network, comprising:

[0020] Data collection module, used to collect photos of power engineering construction sites;

[0021] An image recognition module is used to input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include a bounding box, a confidence probability, and a target category;

[0022] The risk judgment module is used to judge the target construction status based on the identified bounding box, confidence probability and target category, and compare it with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0023] Furthermore, the training process of the image recognition network includes the following steps:

[0024] Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected;

[0025] Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity;

[0026] Network training steps: Use deep neural network as the learning framework to build an image recognition network, input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters;

[0027] The deep neural network is a YOLOv5 network;

[0028] The image acquisition step also includes performing data enhancement on the acquired image by data rotation, flipping and scaling.

[0029] Furthermore, the target construction status determination is specifically as follows:

[0030] The confidence probability in each target recognition result is compared with a preset confidence threshold. If the confidence probability is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted; in the same photo, distance and scene judgments are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status;

[0031] The target construction status obtained is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0032] Compared with the prior art, the present invention has the following advantages:

[0033] (1) The present invention collects construction site image data in real time, uses a deep learning network for target recognition, realizes automatic recognition of various entities involved in the construction site, and extracts the identified bounding box, confidence probability and target category, thereby selecting the recognition result with a higher confidence probability to judge the relationship between different targets, and further judges whether there are illegal operations or potential risks. The present invention significantly improves the level of real-time supervision of the construction site, realizes the automatic recognition of construction operations, and thus reduces the burden of manual supervision.

[0034] (2) The present invention can more comprehensively evaluate and respond to various situations at the construction site by combining the relationships between different targets identified based on the construction site image data with the knowledge graph, thus providing a scientific basis for construction management.

[0035] (3) The present invention adopts a deep learning network, which significantly improves the objectivity and accuracy of judgment, providing a new intelligent solution for power engineering construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic flow chart of a method for identifying targets at a power construction site based on a deep learning network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0038] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0040] Example 1

[0041] like Figure 1 As shown, this embodiment provides a method for identifying targets at a power construction site based on a deep learning network, comprising the following steps:

[0042] S1: Collect photos of power engineering construction sites;

[0043] S2: Input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include bounding boxes, confidence probabilities, and target categories;

[0044] S3: According to the identified bounding box, confidence probability and target category, the target construction status is judged and compared with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0045] Specifically, in step S2, the training process of the image recognition network includes the following steps:

[0046] Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected;

[0047] Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity;

[0048] Network training steps: Use deep neural network as the learning framework to build an image recognition network. Input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters.

[0049] Optionally, the deep neural network is a YOLOv5 network, that is, the image recognition network is built based on the YOLOv5 framework.

[0050] Preferably, the image acquisition step also includes data enhancement of the acquired images by data rotation, flipping and scaling, that is, comprehensive application of data enhancement techniques, such as rotation, flipping, scaling, etc., to increase the diversity of the data set and improve the generalization ability of the model so that it can better adapt to specific target entities at the power engineering construction site.

[0051] In this embodiment, during the image data collection stage at the power engineering construction site, high-resolution cameras and sensor devices are used to ensure comprehensive coverage of all dimensions and angles of the construction site; subsequently, professionals use image annotation tools such as LabelImg to annotate the collected image data in detail, indicating the bounding box and related information of each target entity in the image, to ensure the quality and accuracy of the data set; next, a deep neural network based on YOLOv5 is used as a learning framework, and the model is constructed and trained based on the Python programming language and the PyTorch deep learning library. During the training process, the back propagation algorithm and optimizer are used to adjust the model parameters, and data enhancement techniques such as rotation, flipping, scaling, etc. are comprehensively applied to increase the diversity of the data set and improve the generalization ability of the model, so that it can better adapt to the specific target entities at the power engineering construction site; after the model training is completed, the optimized deep learning network parameters are saved.

[0052] In step S3, the target construction status is determined as follows:

[0053] The confidence probability in each target recognition result is compared with the preset confidence threshold. If it is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted. In the same photo, distance and scene judgment are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status.

[0054] The target construction status obtained is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0055] The identified targets are common entities involved in the construction process of power engineering projects.

[0056] Specifically, the target results identified in step S3 include a JPG format image in which the identified target is marked with a bounding box and a confidence probability, and an XML format file in which the bounding box coordinates, the confidence probability and the identified target category are indicated in the form of data items; after the target entity is identified, a subsequent judgment is made based on the construction safety knowledge graph as to whether there are any illegal operations or potential risks, and a response is made.

[0057] Preferably, the method further comprises standardizing the collected photos of the power engineering construction site to obtain a standard image format required by the image recognition network.

[0058] In this embodiment, in actual application, the image data collected in real time is standardized through image preprocessing technology to ensure that the input data meets the input requirements of the model, and then the standardized image data is input into the trained YOLOv5 model, and the forward propagation mechanism of the model is used to obtain the recognition result of the target entity; the recognition result of the target entity is combined with the knowledge graph technology, and the recognition result of the target entity is compared with the previously established knowledge graph to further determine whether there is any illegal operation or potential risk.

[0059] Example 2

[0060] This embodiment provides a power construction site target recognition system based on a deep learning network, including:

[0061] Data collection module, used to collect photos of power engineering construction sites;

[0062] An image recognition module is used to input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include a bounding box, a confidence probability, and a target category;

[0063] The risk judgment module is used to judge the target construction status based on the identified bounding box, confidence probability and target category, and compare it with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0064] The training process of an image recognition network consists of the following steps:

[0065] Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected;

[0066] Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity;

[0067] Network training steps: Use deep neural network as the learning framework to build an image recognition network, input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters;

[0068] The image acquisition step also includes data augmentation of the acquired images through data rotation, flipping and scaling.

[0069] The specific steps for judging the target construction status are as follows:

[0070] The confidence probability of each target recognition result is compared with the preset confidence threshold. If it is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted. In the same photo, the distance and scene judgment are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status.

[0071] The target construction status obtained is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

[0072] The specific content and beneficial effects of the above system can be found in the above method embodiment, which will not be repeated here.

[0073] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for identifying targets at power construction sites based on deep learning networks, characterized in that: The following steps are involved: Collect photos of power engineering construction sites; Input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include bounding boxes, confidence probabilities, and target categories; The target construction status is judged based on the identified bounding box, confidence probability and target category, and compared with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

2. According to claim 1, a method for identifying targets at power construction sites based on a deep learning network is characterized in that: The training process of the image recognition network includes the following steps: Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected; Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity; Network training steps: Use deep neural network as the learning framework to build an image recognition network. Input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters.

3. A method for identifying targets at power construction sites based on a deep learning network according to claim 2, characterized in that: The deep neural network is a YOLOv5 network.

4. A method for identifying targets at power construction sites based on a deep learning network according to claim 2, characterized in that: The image acquisition step also includes performing data enhancement on the acquired image by data rotation, flipping and scaling.

5. The method for identifying targets at power construction sites based on a deep learning network according to claim 1, characterized in that: The target construction status determination is specifically as follows: The confidence probability in each target recognition result is compared with a preset confidence threshold. If it is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted. In the same photo, distance and scene judgment are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status.

6. A method for identifying targets at power construction sites based on a deep learning network according to claim 5, characterized in that: The target construction status obtained is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

7. The method for identifying targets at power construction sites based on a deep learning network according to claim 1, characterized in that: The method also includes standardizing the collected photos of the power engineering construction site to obtain the standard image form required by the image recognition network.

8. A power construction site target recognition system based on deep learning network, characterized in that: include: Data collection module, used to collect photos of power engineering construction sites; An image recognition module is used to input the collected photos into a pre-built and trained image recognition network to perform target recognition on the power engineering construction site and obtain target recognition results, which include a bounding box, a confidence probability, and a target category; The risk judgment module is used to judge the target construction status based on the identified bounding box, confidence probability and target category, and compare it with the pre-established construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

9. The power construction site target recognition system based on deep learning network according to claim 8 is characterized in that: The training process of the image recognition network includes the following steps: Image acquisition steps: multi-dimensional and multi-angle image acquisition of the power engineering construction site to be inspected; Data annotation step: Use image annotation tools to annotate the collected images, indicating the bounding box and category information of each target entity; Network training steps: Use deep neural network as the learning framework to build an image recognition network, input the collected images and corresponding annotation information into the image recognition network for training. During the training process, use the back propagation algorithm and optimizer to adjust the model parameters; The deep neural network is a YOLOv5 network; The image acquisition step also includes performing data enhancement on the acquired image by data rotation, flipping and scaling.

10. The power construction site target recognition system based on deep learning network according to claim 8, characterized in that: The target construction status determination is specifically as follows: The confidence probability in each target recognition result is compared with a preset confidence threshold. If the confidence probability is greater than the confidence threshold, the corresponding target category and bounding box are retained, otherwise the corresponding target recognition result is deleted; in the same photo, distance and scene judgments are performed based on different bounding boxes and corresponding target categories to obtain the relationship between different targets as the target construction status; The target construction status obtained is matched with the corresponding information of the construction safety knowledge graph to determine whether there are any illegal operations or potential risks.

Images