Training efficiency improving method and system for target recognition model of power transmission equipment

Abstract

The embodiment of the invention provides a training efficiency improving method and system for a target recognition model of power transmission equipment, and belongs to the technical field of target detection. The training efficiency improving method comprises the following steps: acquiring a data image of a historical dial plate through an unmanned aerial vehicle; performing data processing on the data image of the historical dial plate; according to the Faster-RCNN, constructing a target identification model; inputting the data image of the historical dial plate into the target recognition model to recognize a data image of a target dial plate in the data image of the historical dial plate; constructing a damage identification model according to the YOLOv3; and inputting the data image of the target dial plate into the damage identification model to obtain a damaged dial plate in the target dial plate. According to the invention, the data images of all target dial plates are obtained through the target identification model, and then the normal dial plate and the damaged dial plate are distinguished through the damage identification model, so that the analysis accuracy is improved, and the workload of manual analysis processing is greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to a training efficiency improvement method and system for a target recognition model of power transmission equipment. Background Art

[0002] Transmission equipment inspection can detect hidden dangers and faults in the operation of equipment in advance, deal with defects in time, and effectively prevent the occurrence of power grid failures, reduce threats to residents' personal safety, property safety and industrial generation, among which the inspection of dials plays a pivotal role in the daily maintenance of transmission equipment. At present, in the daily monitoring of dials, there are often situations where the shooting angle is not good or the shooting distance is far, which makes it difficult to detect the damage of transmission equipment dials. Summary of the invention

[0003] The purpose of the embodiments of the present invention is to provide a method and system for improving the training efficiency of a target recognition model for power transmission equipment, which solves the problem of a large number of images and unclear images during dial detection.

[0004] In order to achieve the above-mentioned object, an embodiment of the present invention provides a method for improving the training efficiency of a target recognition model of a power transmission equipment, and the training efficiency improvement method includes:

[0005] Acquire data images of historical dials through drones;

[0006] Performing data processing on the data image of the historical dial;

[0007] Build a target recognition model based on Faster-RCNN;

[0008] Inputting the data image of the historical dial into the target recognition model to recognize the data image of the target dial in the data image of the historical dial;

[0009] Build a damage recognition model based on YOLOv3;

[0010] The data image of the target watch dial is input into the damage recognition model to obtain a damaged watch dial in the target watch dial.

[0011] Optionally, data processing is performed on the data image of the historical dial, including:

[0012] Arranging the data images of the historical dial;

[0013] Performing a data enhancement operation on the data image of the historical dial to increase the data volume;

[0014] The data image of the historical dial is labeled.

[0015] Optionally, arranging the data image of the historical dial includes:

[0016] storing the data image of the historical dial in a unified format;

[0017] removing blurred, duplicated or damaged data images of the historical dial;

[0018] Rename the data image of the historical dial;

[0019] The data images of the historical dials are resized to the same size.

[0020] Optionally, a target recognition model is constructed based on the Faster-RCNN network, including:

[0021] Using the candidate box selection network RPN as the candidate box of the data image of the historical dial;

[0022] Using a convolutional neural network (CNN) to extract features from the data image of the historical dial in the candidate frame;

[0023] Determine whether the image in the candidate frame includes a watch face;

[0024] When it is determined that the image within the candidate frame includes a dial, the candidate frame is made to be large enough to accommodate the included dial and has a minimum area through non-maximum suppression.

[0025] Optionally, a damage recognition model is constructed based on YOLOv3, including:

[0026] Dividing the data image of the target dial into a training set and a validation set;

[0027] Inputting the training set into the damage recognition model for training;

[0028] Determine whether the training is finished;

[0029] On the premise of judging that the training is finished, calculating the performance index of the damage recognition model;

[0030] Adjusting parameters according to the performance indicators of the damage identification model;

[0031] Using the validation set to validate the damage recognition model, and optimizing the performance of the damage recognition model in actual scenarios;

[0032] The damage identification model is saved.

[0033] Optionally, a damage recognition model is constructed based on YOLOv3, including:

[0034] The images in the training set and the validation set are classified according to the state of the dial, and the normal dial label is set to 1 and the damaged dial label is set to 0.

[0035] Optionally, inputting the data image of the target watch dial into the damage recognition model to obtain a damaged watch dial in the target watch dial comprises:

[0036] The prediction box with a confidence level greater than the first set threshold is highlighted, saved and displayed, and the prediction box is a damaged dial.

[0037] Optionally, the first set threshold is set to 0.7.

[0038] On the other hand, the present invention provides a training efficiency improvement system for a target recognition model of power transmission equipment, wherein the training efficiency improvement system includes a processor for executing any of the training efficiency improvement methods described above.

[0039] On the other hand, the present invention provides a machine-readable storage medium, wherein the computer-readable storage medium stores instructions, and the instructions are used to be read by a machine so that the machine executes any of the training efficiency improvement methods described above.

[0040] Through the above technical scheme, the present invention first obtains the data images of all target dials through the target recognition model, and then distinguishes the normal dials from the damaged dials through the damage recognition model. During the drone inspection of power transmission equipment, there are often situations where the shooting angle is not good or the shooting distance is far, which makes it difficult to detect the damage of the power transmission equipment dial. By using relevant algorithms such as target detection in machine learning, the target dial in the drone shot image can be better extracted, and then the damaged dial can be detected. The present invention can uniformly and intelligently manage a large number of dial images, improve the accuracy of analysis, and greatly reduce the workload of manual analysis and processing.

[0041] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0043] Figure 1 is a flowchart of a training efficiency improvement method according to an embodiment of the present invention;

[0044] Figure 2 is a flow chart of data processing according to one embodiment of the present invention;

[0045] Figure 3is a flow chart for collating data images of historical dials according to one embodiment of the present invention;

[0046] Figure 4 is a flowchart of constructing a target recognition model according to one embodiment of the present invention;

[0047] Figure 5 is a flowchart of training a damage recognition model according to one embodiment of the present invention. DETAILED DESCRIPTION

[0048] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0049] Figure 1 is a flow chart of a training efficiency improvement method according to an embodiment of the present invention. In the figure, the training efficiency improvement method includes:

[0050] In step S1, a data image of a historical dial is acquired by a drone.

[0051] In step S2, data processing is performed on the data image of the historical dial.

[0052] In step S3, a target recognition model is constructed according to Faster-RCNN.

[0053] In step S4 , the data images of the historical dials are input into the target recognition model to recognize the data images of the target dial in the data images of the historical dials.

[0054] In step S5, a damage recognition model is constructed according to YOLOv3.

[0055] In step S6 , the data image of the target watch dial is input into the damage recognition model to obtain damaged watch dials in the target watch dial.

[0056] Compared with the prior art, the present invention first obtains the data images of all target dials through the target recognition model, and then distinguishes the normal dials from the damaged dials through the damage recognition model. During the drone inspection of power transmission equipment, there are often situations where the shooting angle is not good or the shooting distance is far, which makes it difficult to detect the damaged dials of the power transmission equipment. By utilizing relevant algorithms such as target detection in machine learning, the target dials in the drone image can be better extracted, and then the damaged dials can be detected. The present invention can uniformly and intelligently manage a large number of dial images, improve the accuracy of the analysis, and greatly reduce the workload of manual analysis and processing.

[0057] In this implementation, in order to effectively use the acquired historical dial data image for the target recognition model, the data image needs to be further processed. The specific steps of the processing may be various ones known to those skilled in the art. In one example of the present invention, for example Figure 2 As shown, it may include:

[0058] In step S21, the data images of the historical dials are sorted.

[0059] In step S22, a data enhancement operation is performed on the data image of the historical dial to increase the data volume.

[0060] In step S23, the data image of the historical dial is annotated.

[0061] In this embodiment, in order to remove irrelevant or low-quality image data, the specific steps for sorting the data images of the historical dial can be various steps known to those skilled in the art. In one example of the present invention, for example Figure 3 As shown, it may include:

[0062] In step S211 , the data image of the historical dial is stored in a unified format.

[0063] In step S212, blurred, duplicated or damaged data images of historical dials are removed.

[0064] In step S213, the data image of the historical dial is renamed.

[0065] In step S214, the data images of the historical dial are adjusted to the same size.

[0066] Through this process, irrelevant or low-quality images can be removed, ensuring that the image data for subsequent processing is consistent and high-quality, thereby improving the processing efficiency and accuracy of the entire target recognition process.

[0067] In this implementation, the step of constructing the target recognition model can be a variety of methods known to those skilled in the art, provided that the target dial is recognized. In one example of the present invention, Figure 4 As shown, it may include:

[0068] In step S31, the candidate box selection network RPN is used as the candidate box of the data image of the historical dial. The candidate box reduces the time of exhaustive search on the entire image, thereby improving the efficiency of target detection.

[0069] In step S32, a convolutional neural network (CNN) is used to extract features from the data image of the historical dial within the candidate frame.

[0070] In step S33, it is determined whether the image in the candidate frame includes a dial. If it is determined that the image in the candidate frame includes a dial, step S34 is executed; otherwise, step S32 is executed.

[0071] In step S34, the candidate box is reduced to a minimum area by non-maximum suppression so that the candidate box is large enough to accommodate the included dial.

[0072] By determining whether the image in the candidate frame contains the target dial, and adjusting the size and position of the candidate frame through non-maximum suppression when the dial is contained, it is ensured that the candidate frame is large enough to accommodate the contained dial and has the smallest area, thereby accurately locating the position of the target dial.

[0073] In this implementation, in order to obtain a complete damage recognition model, it is also necessary to input the data image of the target dial into the damage recognition model for training. The steps for training the damage recognition model may be various steps known to those skilled in the art. In one example of the present invention, the specific training steps may be as follows: Figure 5 As shown. Figure 5 In the training step, the training step may include:

[0074] In step S51 , the data image of the target dial is divided into a training set and a validation set.

[0075] In step S52, the training set is input into the damage recognition model for training.

[0076] In step S53, it is determined whether the training is completed. If the training is completed, step S54 is executed; otherwise, step S52 is executed.

[0077] In step S54, the performance index of the damage identification model is calculated.

[0078] In step S55, parameters are adjusted according to the performance index of the damage identification model.

[0079] In step S56, the damage recognition model is verified using the verification set, and the performance of the damage recognition model in the actual scenario is optimized.

[0080] In step S57, the damage recognition model is saved.

[0081] In this implementation, in order to facilitate the distinction between a normal dial and a damaged dial, the images in the training set and the validation set may be classified according to the state of the dial, with the normal dial label being set to 1 and the damaged dial label being set to 0.

[0082] In this implementation, in order to further reduce false detection and improve detection accuracy, the prediction box with a confidence level greater than a first set threshold value may be highlighted, saved and displayed, and the prediction box is a damaged dial.

[0083] In this implementation, the value of the first set threshold value may be any of a variety of values ​​known to those skilled in the art. In one example of the present invention, the value of the first set threshold value may be 0.7.

[0084] On the other hand, the present invention provides a training efficiency improvement system for a target recognition model of power transmission equipment, wherein the training efficiency improvement system includes a processor for executing any of the above-mentioned training efficiency improvement methods.

[0085] On the other hand, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, wherein the instructions are used to be read by a machine so that the machine executes any of the above-mentioned training efficiency improvement methods.

[0086] Through the above technical scheme, the present invention first obtains the data images of all target dials through the target recognition model, and then distinguishes the normal dials from the damaged dials through the damage recognition model. During the drone inspection of power transmission equipment, there are often situations where the shooting angle is not good or the shooting distance is far, which makes it difficult to detect the damage of the power transmission equipment dial. By using relevant algorithms such as target detection in machine learning, the target dial in the drone shot image can be better extracted, and then the damaged dial can be detected. The present invention can uniformly and intelligently manage a large number of dial images, improve the accuracy of analysis, and greatly reduce the workload of manual analysis and processing.

[0087] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0088] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0089] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0090] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for improving the training efficiency of a power transmission equipment target recognition model, characterized in that: The training efficiency improvement method comprises: Acquire data images of historical dials through drones; Performing data processing on the data image of the historical dial; Build a target recognition model based on Faster-RCNN; Inputting the data image of the historical dial into the target recognition model to recognize the data image of the target dial in the data image of the historical dial; Build a damage recognition model based on YOLOv3; The data image of the target watch dial is input into the damage recognition model to obtain a damaged watch dial in the target watch dial.

2. The training efficiency improvement method according to claim 1, characterized in that: The data image of the historical dial is processed, including: Arranging the data images of the historical dial; Performing a data enhancement operation on the data image of the historical dial to increase the data volume; The data image of the historical dial is labeled.

3. The training efficiency improvement method according to claim 2, characterized in that: The data image of the historical dial is sorted, including: storing the data image of the historical dial in a unified format; removing blurred, duplicated or damaged data images of the historical dial; Rename the data image of the historical dial; The data images of the historical dials are resized to the same size.

4. The training efficiency improvement method according to claim 3, characterized in that: The target recognition model is constructed based on the Faster-RCNN network, including: Using the candidate box selection network RPN as the candidate box of the data image of the historical dial; Using a convolutional neural network (CNN) to extract features from the data image of the historical dial in the candidate frame; Determine whether the image in the candidate frame includes a watch face; When it is determined that the image within the candidate frame includes a dial, the candidate frame is made to be large enough to accommodate the included dial and has a minimum area through non-maximum suppression.

5. The training efficiency improvement method according to claim 4, characterized in that: Build a damage recognition model based on YOLOv3, including: Dividing the data image of the target dial into a training set and a validation set; Inputting the training set into the damage recognition model for training; Determine whether the training is finished; On the premise of judging that the training is finished, calculating the performance index of the damage recognition model; Adjusting parameters according to the performance indicators of the damage identification model; Using the validation set to validate the damage recognition model, and optimizing the performance of the damage recognition model in actual scenarios; The damage identification model is saved.

6. The training efficiency improvement method according to claim 5, characterized in that: Build a damage recognition model based on YOLOv3, including: The images in the training set and the validation set are classified according to the state of the dial, and the normal dial label is set to 1 and the damaged dial label is set to 0.

7. The training efficiency improvement method according to claim 6, characterized in that: Inputting the data image of the target watch dial into the damage recognition model to obtain a damaged watch dial in the target watch dial includes: The prediction box with a confidence level greater than the first set threshold is highlighted, saved and displayed, and the prediction box is a damaged dial.

8. The training efficiency improvement method according to claim 7, characterized in that: The first set threshold is set to 0.

7.

9. A training efficiency improvement system for a power transmission equipment target recognition model, characterized in that: The training efficiency improvement system includes a processor for executing the training efficiency improvement method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and the instructions are used to be read by a machine so that the machine executes the training efficiency improvement method as described in any one of claims 1-8.

Images