Knife switch on-off state identification method based on convolutional neural network

Abstract

The invention discloses a method for identifying the on-off state of a disconnecting link based on a convolutional neural network, and belongs to the field of disconnecting links. The method comprises the following steps: training and optimizing a yov5s lightweight identification model through a large number of disconnecting link on-off state video images, then carrying out transfer learning on the model by adopting a marked actual scene image of disconnecting link application, and training again through an image training sample of a disconnecting link in an actual scene to obtain a disconnecting link on-off identification model for the actual scene. The opening and closing type and the opening and closing state of the disconnecting link in the actual scene are identified, and the accuracy of automatic identification of the opening and closing state of the disconnecting link is improved.

Description

Technical Field

[0001] The present invention relates to the field of disconnecting switches, and in particular to a method for identifying the opening and closing states of disconnecting switches based on a convolutional neural network. Background Art

[0002] When opening and closing the disconnecting switches in a 220 kV substation, a flashlight needs to be used to irradiate one observation hole, and whether the opening and closing are in place is observed in the observation hole on the opposite side. At present, the method of observing the opening and closing states of disconnecting switches by operating personnel not only takes time but is also difficult to operate, and the observation results are not accurate enough due to the influence of human factors. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for identifying the opening and closing states of disconnecting switches based on a convolutional neural network, which can improve the accuracy of automatic identification of the opening and closing states of disconnecting switches.

[0004] To achieve the above purpose, the present invention provides the following solutions:

[0005] A method for identifying the opening and closing states of disconnecting switches based on a convolutional neural network includes:

[0006] Labeling the opening and closing types and states of disconnecting switch images, and forming a first training set with the disconnecting switch images; the disconnecting switch images are images of different types of disconnecting switches taken from different shooting angles;

[0007] Using the first training set to train the yolov5s lightweight recognition model to obtain an initial disconnecting switch opening and closing recognition model;

[0008] For the actual scenarios where disconnecting switches are applied, obtaining training samples of disconnecting switches in the actual scenarios; the training samples include images of disconnecting switches in the actual scenarios and labels;

[0009] Performing augmentation of the training samples to form a second training set;

[0010] Using the second training set to train the initial disconnecting switch opening and closing recognition model to obtain the disconnecting switch opening and closing recognition model for the actual scenarios;

[0011] Inputting the image of the disconnecting switch to be recognized taken in the actual scenario into the disconnecting switch opening and closing recognition model for the actual scenario, and outputting the opening and closing types and states of the disconnecting switch in the actual scenario.

[0012] A system for identifying the opening and closing states of disconnecting switches based on a convolutional neural network includes:

[0013] An image acquisition module, configured to acquire images of different types of disconnecting switches taken from different shooting angles;

[0014] An annotation module for annotating the opening / closing type and opening / closing state of a disconnecting switch image and forming a first training set together with the disconnecting switch image; the disconnecting switch image is an image of different types of disconnecting switches taken from different shooting angles;

[0015] A primary training module for training a yolov5s lightweight recognition model using the first training set to obtain an initial disconnecting switch opening / closing recognition model;

[0016] A training sample acquisition module for obtaining training samples of disconnecting switches in an actual scenario for the application of disconnecting switches; the training samples include images of disconnecting switches in the actual scenario and annotations;

[0017] An augmentation module for augmenting the training samples to form a second training set;

[0018] A secondary training module for training the initial disconnecting switch opening / closing recognition model using the second training set to obtain the disconnecting switch opening / closing recognition model for the actual scenario;

[0019] An application module for inputting an image of a disconnecting switch to be recognized taken in the actual scenario into the disconnecting switch opening / closing recognition model for the actual scenario and outputting the opening / closing type and opening / closing state of the disconnecting switch in the actual scenario.

[0020] An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the method for recognizing the opening / closing state of a disconnecting switch based on a convolutional neural network as described above.

[0021] A computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the method for recognizing the opening / closing state of a disconnecting switch based on a convolutional neural network as described above.

[0022] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0023] The present invention discloses a method for recognizing the opening / closing state of a disconnecting switch based on a convolutional neural network. The yolov5s lightweight recognition model is trained and optimized through a large number of video images of the opening / closing state of disconnecting switches. Further, for the actual scenario of the application of disconnecting switches, transfer learning is performed on the model using the annotated images of the actual scenario of the application of disconnecting switches to obtain a disconnecting switch opening / closing recognition model for this actual scenario, and the opening / closing type and opening / closing state of the disconnecting switch in this actual scenario are recognized, improving the accuracy of automatic recognition of the opening / closing state of the disconnecting switch. Description of the Drawings

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0025] Figure 1 It is a flowchart of a method for identifying the opening and closing states of a disconnecting switch based on a convolutional neural network provided by an embodiment of the present invention. Detailed implementation manners

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0027] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0028] As Figure 1 shown, an embodiment of the present invention provides a method for identifying the opening and closing states of a disconnecting switch based on a convolutional neural network, including:

[0029] Step 1: Label the opening and closing types and states of the disconnecting switch images, and form a first training set with the disconnecting switch images.

[0030] The opening and closing states of the disconnecting switch are judged by the flag bits pointed to by the pointer. The types (categories) of disconnecting switches produced by different manufacturers are also different. Some disconnecting switches use text to distinguish the flag bits, and some use colors to distinguish the flag bits.

[0031] Take pictures of the pointer pointing by video, and then perform intelligent recognition on the pictures, and judge the opening and closing states of the disconnecting switch according to the recognition results.

[0032] Pre-label the disconnecting switch samples, including the opening and closing states. At this time, different opening and closing states can be refined and classified. The advantage of this is that it can be determined whether the sample data is balanced, and the number of network neurons can be increased to make the classification (identification) ability stronger.

[0033] Step 2: Use the first training set to train the yolov5s lightweight recognition model to obtain an initial disconnecting switch opening and closing recognition model.

[0034] Using the lightweight recognition model of YOLOv5s, the input data is processed by a CNN convolutional neural network. Mainly, it captures features such as the texture, structure, and relative position of the image. During the training phase, the model parameters can be updated to enable the model to have the ability to "learn".

[0035] After extracting the features, the specified switch classification and position information are output through the model classification head. Thus, the purpose of detection and classification is achieved.

[0036] Step 3: For the actual scenario of the switch application, obtain the training samples of the switch in the actual scenario; the training samples include the images of the switch in the actual scenario and the annotations.

[0037] Step 4: Augment the training samples to form a second training set.

[0038] Since the number of training samples is small, manual data augmentation is performed in the early stage and various parameters are adjusted during training to increase the sample diversity. Augmentation is used to prevent overfitting.

[0039] The augmentation specifically includes:

[0040] Flip, perspective, shear, scale, and change the angle of the training samples to obtain augmented training samples;

[0041] Add noise to the augmented training samples; the noise includes salt-and-pepper noise, Gaussian noise, and grayscale noise. Adding noise simulates authenticity.

[0042] Step 5: Use the second training set to train the initial switch opening / closing recognition model to obtain the switch opening / closing recognition model for the actual scenario.

[0043] Use Adam instead of SGD to accelerate convergence, and at the same time adjust batch, LR, and WD accordingly.

[0044] During the calculation of OpenCL (Open Computing Language), most calculations are performed in the data layout of cl::Image2D. Therefore, it is necessary to adjust the input spatial according to the specific operating environment to accelerate the inference without significantly reducing the accuracy.

[0045] Inference stage:

[0046] Adjust the conf-threshold (confidence threshold) and IOU-threshold (IOU threshold) to better remove duplicates and ensure certainty according to the actual scenario.

[0047] Introduce the dedicated inference framework of Baidu Paddle-Lite to infer the opening / closing type and opening / closing state of the switch.

[0048] Step 6: Input the disconnector image to be recognized captured in the actual scenario into the disconnector opening / closing recognition model of the actual scenario, and output the opening / closing type and opening / closing state of the disconnector in the actual scenario.

[0049] The present invention performs Fine-tunning transfer learning based on the conventional detection scheme yolov5s model to achieve the functions of fast detection and classification of small targets.

[0050] This intelligent recognition algorithm adopts a deep learning algorithm model constructed based on a convolutional neural network, and trains and optimizes the model through a large number of video images of the disconnector opening / closing state. Finally, the algorithm model converges into an intelligent agent that can recognize the disconnector opening / closing state.

[0051] To execute the method corresponding to the above embodiment to achieve the corresponding functions and technical effects, the following provides an identification system for the disconnector opening / closing state based on a convolutional neural network, including: an image acquisition module, a labeling module, a primary training module, a training sample acquisition module, an augmentation module, a secondary training module, and an application module.

[0052] The labeling module is used to label the opening / closing type and opening / closing state of the disconnector image, and form a first training set with the disconnector image. The disconnector image is an image of different types of disconnectors captured from different shooting angles.

[0053] The primary training module is used to train the yolov5s lightweight recognition model using the first training set to obtain an initial disconnector opening / closing recognition model.

[0054] The training sample acquisition module is used to obtain training samples of the disconnector in the actual scenario for the actual scenario where the disconnector is applied; the training samples include images and labels of the disconnector in the actual scenario.

[0055] The augmentation module is used to augment the training samples to form a second training set.

[0056] The secondary training module is used to train the initial disconnector opening / closing recognition model using the second training set to obtain the disconnector opening / closing recognition model of the actual scenario.

[0057] The application module is used to input the disconnector image to be recognized captured in the actual scenario into the disconnector opening / closing recognition model of the actual scenario, and output the opening / closing type and opening / closing state of the disconnector in the actual scenario.

[0058] The identification system for the disconnector opening / closing state based on a convolutional neural network provided in the embodiment of the present invention has a working principle and beneficial effects similar to those of the identification method for the disconnector opening / closing state based on a convolutional neural network described in the above embodiment, so it will not be elaborated here. For specific content, refer to the introduction of the above method embodiment.

[0059] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for identifying the opening and closing states of a disconnecting switch based on a convolutional neural network as described above is implemented.

[0060] In addition, when the computer program in the above-mentioned memory is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories, random access memories, magnetic disks, or optical discs that can store program codes.

[0061] Furthermore, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed, the method for identifying the opening and closing states of a disconnecting switch based on a convolutional neural network as described above is implemented.

[0062] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other.

[0063] Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The descriptions of the above embodiments are only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A recognition method for the opening and closing states of a knife switch based on a convolutional neural network, characterized in that, Including: Label the opening / closing type and opening / closing state of the disconnecting switch image, and form the first training set together with the disconnecting switch image; the disconnecting switch image is an image of different types of disconnecting switches taken from different shooting angles; Use the first training set to train the yolov5s lightweight recognition model to obtain an initial disconnecting switch opening / closing recognition model; For the actual scenario where the disconnecting switch is applied, obtain the training samples of the disconnecting switch in the actual scenario; the training samples include the image and label of the disconnecting switch in the actual scenario; Augment the training samples to form the second training set; Use the second training set to train the initial disconnecting switch opening / closing recognition model to obtain the disconnecting switch opening / closing recognition model for the actual scenario; Input the image of the disconnecting switch to be recognized taken in the actual scenario into the disconnecting switch opening / closing recognition model for the actual scenario, and output the opening / closing type and opening / closing state of the disconnecting switch in the actual scenario.

2. The method for identifying the opening and closing state of a disconnecting switch based on a convolutional neural network according to claim 1, wherein The augmentation of the training samples specifically includes: Flip, perspective, shear, scale, and change the angle of the training samples to obtain augmented training samples; Add noise to the augmented training samples; the noise includes salt-and-pepper noise, gauss noise, and grayscale noise.

3. The method for identifying the opening and closing state of a knife switch based on a convolutional neural network according to claim 1, wherein, During the process of training the initial disconnecting switch opening / closing recognition model, use the Adam optimization algorithm to optimize the initial disconnecting switch opening / closing recognition model to accelerate convergence.

4. The method for identifying the opening and closing state of a knife switch based on a convolutional neural network according to claim 1, wherein During the process of training the initial disconnecting switch opening / closing recognition model, introduce the Baidu paddle-lite dedicated inference framework to infer the opening / closing type and opening / closing state of the disconnecting switch.

5. An identification system for the opening and closing states of a disconnecting switch based on a convolutional neural network, characterized in that, Including: A labeling module for labeling the opening / closing type and opening / closing state of the disconnecting switch image and forming the first training set together with the disconnecting switch image; the disconnecting switch image is an image of different types of disconnecting switches taken from different shooting angles; A first training module for using the first training set to train the yolov5s lightweight recognition model to obtain an initial disconnecting switch opening / closing recognition model; A training sample obtaining module for obtaining the training samples of the disconnecting switch in the actual scenario for the actual scenario where the disconnecting switch is applied; The training samples include the image and label of the disconnecting switch in the actual scenario; An augmentation module for augmenting the training samples to form the second training set; A second training module for using the second training set to train the initial disconnecting switch opening / closing recognition model to obtain the disconnecting switch opening / closing recognition model for the actual scenario; An application module for inputting the image of the disconnecting switch to be recognized taken in the actual scenario into the disconnecting switch opening / closing recognition model for the actual scenario and outputting the opening / closing type and opening / closing state of the disconnecting switch in the actual scenario.

6. An electronic device, characterized in that, Including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, it implements the method for recognizing the opening / closing state of a disconnecting switch based on a convolutional neural network as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, Stored thereon is a computer program, and when the computer program is executed, it implements the method for recognizing the opening / closing state of a disconnecting switch based on a convolutional neural network as described in any one of claims 1 to 4.

Images