WO2021042682A1 - Method, apparatus and system for recognizing transformer substation foreign mattter, and electronic device and storage medium - Google Patents

Abstract

A method for recognizing a transformer substation foreign matter. The method comprises the following steps: collecting a scene image of a transformer substation (110); and recognizing a foreign matter in the scene image according to a pre-built target detection model for a transformer substation foreign matter in order to obtain the position of the foreign matter in the scene image and the type of the foreign matter, wherein the target detection model for the transformer substation foreign matter is obtained by means of performing training according to a preset type of foreign matter in the transformer substation and a plurality of transformer substation foreign matter images (120).

Description

Substation foreign body identification method, device, system, electronic equipment and storage medium

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office with an application number of 201910826233.X on September 3, 2019. The entire content of this application is incorporated into this application by reference.

Technical field

This application relates to the fields of artificial intelligence and power transmission and transformation technologies, such as a method, device, system, electronic equipment, and storage medium for identifying foreign objects in a substation.

Background technique

The substation is a vital part of the long-distance power transmission architecture. It contains hundreds of devices, and each device includes dozens of components. In traditional substations, these devices and components are basically exposed. Protected by casing. On the other hand, substations are generally blocked from the external environment by low walls. Foreign objects in the environment, such as kites, balloons, films, etc., will inevitably break into the station and be easily entangled with the equipment in the station. If not found in time, It is easy to cause a short circuit of the equipment, and then damage the equipment or even cause more serious cascading disasters. Although fixed cameras have been set up in the substation so that personnel can observe the situation in the station through real-time video to find out the foreign objects that have entered, but in actual situations, it is impossible to require personnel to conduct 24-hour comprehensive and multi-angle monitoring. , And the human eyes are prone to fatigue, resulting in misjudgment. Patrol robots can be deployed in multiple substations, and the patrol robots take pictures of different locations in the station along a fixed route and perform automatic analysis to detect changes in equipment status and intrusion of foreign objects. However, the current target recognition technology used by robots has a low level of intelligence. It can only distinguish the similarities and differences between the current and historical images through image comparison, and cannot truly identify whether there are foreign objects in the images and the specific types of foreign objects. In addition, this method of image comparison is very susceptible to environmental images. When conditions such as external light change, it will cause the robot to misjudge.

Among the related technologies, the intelligent monitoring and inspection level of substations is low, and there is no ability to automatically and timely find foreign objects in the station to carry out research. Currently, fixed cameras have been deployed in substations to monitor the behavior of personnel and the conditions in the station. However, the monitoring results need to be analyzed and judged manually, which lacks intelligence. In addition, some substations have already used patrol robots. Although they can complete equipment status recognition and foreign body discovery to a certain extent, they often use simple methods of image comparison to achieve this. They have poor performance, low reliability, and are easily affected by the environment. Really distinguish the specific types of foreign objects. Foreign objects intruded into substations are often entangled with power equipment and grids. If they are not found in time, they will most likely cause damage to the equipment and even cause serious electrical accidents.

In order to solve these problems and improve the intelligence level of patrol and monitoring in substations, there is an urgent need for a method that enables robots and fixed cameras to automatically recognize foreign objects in the field of view.

Summary of the invention

In response to the above-mentioned problems in the related technologies, this application proposes a method, device, system, electronic equipment and storage medium for identifying foreign objects in a substation, which can intelligently identify foreign objects in the substation and improve the intelligent level of substation inspection and monitoring.

This application is realized through the following technical solutions:

A method for identifying foreign objects in a substation, including:

Collect scene images of substations;

Identify the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation, and obtain the location of the foreign body in the scene image and the type of the foreign body;

Wherein, the target detection model of foreign objects in the substation is obtained by training based on preset types of foreign objects in the substation and multiple images of foreign objects in the substation.

In one embodiment, the type of foreign body includes at least one of the following:

Bird nests, honeycombs, kites, balloons, plastic films and dust nets.

In an embodiment, the process of establishing a target detection model for foreign objects in the substation includes:

Collect foreign body scene images of substations and simulated foreign body scene images of substations, mark each label in each image of the foreign body scene images of the substation and the simulated foreign body scene images of the substation with the label to which each label belongs, and mark The subsequent foreign body scene images in the substation and the simulated foreign body scene images in the substation are summarized to obtain a summarized image; wherein the label to which each label belongs is a text description of the type of each label;

The collected images are used for model training, and the target detection model of foreign objects in the substation is obtained.

In an embodiment, the collecting scene images of foreign objects in the substation includes:

Collect historical images of foreign objects in the substation, where the number of historical images is multiple;

According to the type of the foreign body in the historical image, the poor-quality historical image is eliminated, and the poor-quality historical image includes: the historical image whose pixel area occupied by the foreign body is less than 5% of the pixel area of the historical image, display In historical images where the surface area of the foreign object body is less than 10% of the actual surface area of the foreign object body and the historical image displayed in the historical image where the size of the foreign object body is less than 10% of the actual size of the foreign object body At least one

The simulated foreign body scene image of the substation includes: collecting the original foreign body image and the original substation scene image; cutting the foreign body in the original foreign body image to obtain the foreign body image; and performing the preset expansion method on the foreign body image Processing to expand the foreign body image; splicing each of the foreign body image and the expanded foreign body image with the original substation scene image to obtain a simulated foreign body scene image of the substation.

In an embodiment, the collecting an image of the original foreign body body includes:

Under the condition of ensuring that the foreign objects in the substation fall in the central area of the field of view of the camera equipment, the original objects with different brightness can be obtained by adjusting the aperture of the camera equipment and using the camera equipment to shoot the foreign objects from different angles and different distances. The image of the foreign body; and, downloading the image of the foreign body through the network to obtain the original foreign body image.

In one embodiment, cropping the foreign body in the original foreign body image to obtain the foreign body image includes: removing the background and cutting along the edge of each image in the original foreign body image to obtain the foreign body image;

The processing the foreign body image by a preset expansion method to expand the foreign body image includes: performing inversion, cropping, scaling, and noise processing on the foreign body image to obtain an expanded foreign body image;

The collection of scene images of the original substation includes: using a robot and a fixed camera needle to take pictures of the parts of the designated equipment and the parts of the designated parts in the substation, and adjust the shooting distance, angle, and aperture during the shooting process to obtain the original substation Scene image

The stitching of each of the foreign body image and the expanded foreign body image with the original substation scene image includes: stitching each of the foreign body image and the expanded foreign body image to the The location of the designated equipment and designated parts in the original substation scene image to simulate the scene image of foreign objects in the substation.

In one embodiment, reversing the image of the foreign object body includes: reversing the image of the foreign object body once every 30 degrees to obtain a reversed foreign object body image;

Cutting the foreign body image includes: cutting the foreign body image from top, bottom, left, and right to 20%, 40%, and 60% of the foreign body image to obtain the cut foreign body image. image;

The scaling of the foreign body image includes: adjusting the foreign body image to 25%, 50%, and 200% of the size of the foreign body image to obtain a scaled foreign body image;

Performing noise processing on the foreign body image includes: introducing image noise to the foreign body image to obtain a noise-processed foreign body image;

Wherein, the expanded foreign body image includes: the foreign body body image after inversion, the foreign body body image after cropping, the foreign body body image after scaling, and the foreign body body image after noise processing.

In an embodiment, the introducing image noise to the image of the foreign body body includes:

Introduce image noise to the foreign body image by the following formula:

f(x,y)=g(x,y)+q(.),

Among them, x and y respectively represent the horizontal and vertical coordinates of the pixels in the foreign body image, g(x,y) represents the true value of the pixels in the foreign body image, and for multi-channel images, g(x,y) represents the The vector corresponding to the pixel in the foreign body image, q(.) represents the noise function, f(x,y) represents the value of the pixel of the foreign body image after the noise is introduced; the noise function is derived from the probability density function of the Gaussian random variable Introduced, the noise function is expressed by the following formula:

Among them: z is a variable, z=g(x,y), q(z) is the probability density function of z, μ is the expectation of z, and σ is the variance of z.

In an embodiment, the collecting scene images of the substation includes:

Collecting scene videos of the substation and extracting frames from the scene videos to obtain scene images;

Performing image processing on the scene image to obtain a scene image of the substation;

The image processing includes linear grayscale enhancement and logarithmic function nonlinear transformation.

In an embodiment, the method for linear grayscale enhancement includes:

Where g(x,y) is the true value of the pixel of the scene image, [a,b] is the grayscale range of the scene image, and [c,d] is the grayscale of the image after linear grayscale enhancement The expected range of f(x,y) is the pixel value of the image after linear grayscale enhancement;

The nonlinear transformation formula of the logarithmic function is as follows:

Among them, h(x, y) is the pixel value of the image after the nonlinear transformation of the logarithmic function, and a, b, and c are adjustable parameters. After iterative correction, the gray range of the newly generated image is within the specified range.

In an embodiment, when the scene image is a color image, the image processing further includes: gradation, and the gradation is implemented by a built-in method of opencv.

A device for identifying foreign objects in a substation includes:

The acquisition module is set to collect scene images of the substation;

The recognition module is configured to recognize the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation, and obtain the location of the foreign body in the scene image and the type of the foreign body;

Wherein, the target detection model of the foreign body in the substation is obtained by training according to a preset foreign body type and foreign body image.

A substation foreign body recognition system, comprising a substation patrol robot and a substation foreign body recognition device as claimed in claim 13, wherein the substation patrol robot includes a camera device, and the substation patrol robot is set to pass through The camera device collects scene images of the substation.

An electronic device includes a processor and a memory, where a computer program is stored in the memory, and when the computer program is executed by the processor, any one of the methods described above is implemented.

A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, any one of the methods described above is implemented.

Description of the drawings

In order to facilitate those of ordinary skill in the art to understand and implement the application, the application will be described below in conjunction with the accompanying drawings and specific implementations, but it should be understood that the protection scope of the application is not limited by the specific implementations.

FIG. 1 is a flowchart of a method for identifying foreign objects in a substation according to an embodiment of the application;

Fig. 2 is a schematic diagram of the overall framework provided by an embodiment of the present application.

detailed description

Example 1

FIG. 1 is a flowchart of a method for identifying foreign objects in a substation according to an embodiment of the present application, and FIG. 2 is a schematic diagram of the overall framework provided by an embodiment of the present application.

As shown in Figure 1 and Figure 2, this application provides a method for identifying foreign objects in a substation. The method includes the following steps:

Step 110: Collect scene images of the substation.

Step 120: Identify the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation, and obtain the location of the foreign body in the scene image and the type of the foreign body;

Wherein, the target detection model of foreign objects in the substation is obtained by training based on preset types of foreign objects in the substation and multiple images of foreign objects in the substation.

The following is an explanation of the application method:

Step 10: Combing foreign objects. Sort out the types of foreign objects with high frequency and high degree of damage in the substation. After investigation and research, it is concluded that the typical types of foreign objects include one or more of the following: bird nests, honeycombs, kites, balloons, plastic films and dust-proof nets. On this basis, the types of foreign objects in the substation are expanded.

Step 20: Image acquisition. Image collection is divided into the following two situations, including the collection of scene images of foreign objects in substations and the simulation of scene images of foreign objects (ie, simulation of scenes of foreign objects in substations).

Step 20-10: Collection of foreign object scene images. Collect historical images of foreign objects in the substation, classify historical images according to the type of foreign objects, such as bird nests, beehives, kites, balloons, plastic films, dust-proof nets, etc., and remove poor-quality historical images. The historical image includes: (1) the example where the area of the pixel area occupied by the foreign body in the scene image is less than 5% of the pixel area of the historical image, and (2) the example where only less than 10% of the foreign body is partially displayed in the historical image.

Step 20-20: Simulation of the foreign object scene image. The simulation of the foreign body scene image mainly includes: shooting and collection of the original foreign body image, shooting of the scene in the original substation, partial clipping of the foreign body body, image expansion, changing the image noise and image splicing.

Steps 20-20-10: shooting and collecting the image of the original foreign body. When taking real-time shooting of foreign objects in the substation, it is necessary to ensure that the foreign objects fall in the central area of the field of view of the camera equipment, shooting from different distances and angles, and to obtain images with different brightness as the original foreign object images by adjusting the device aperture to increase the number of foreign object images And diversity; download pictures of foreign bodies, including bird nests, beehives, kites, balloons, plastic films and dust-proof nets, from the Internet or other channels as the original foreign body images.

Steps 20-20-20: Cut the foreign body part. For each original foreign body image, the background is removed and the image of the foreign body part is cut out along the edge to obtain the foreign body image.

Step 20-20-30: Image expansion. The foreign body images are processed by flipping, cropping and zooming respectively to increase the number of foreign body images. The method of processing the image of the foreign object body through the inversion method is as follows: the image of the foreign object body is inverted every 30 degrees to obtain a new image (that is, the image of the foreign object body after the inversion); the image of the foreign object body is processed by cropping The method is as follows: randomly crop the obtained foreign body image, and cut out 20%, 40%, and 60% of the foreign body image from the top, bottom, left and right of the foreign body image. Finally, pass each The foreign body image acquires 12 new foreign body images (that is, the cropped foreign body images); the method of processing the foreign body images by scaling is as follows, and the foreign body images are changed to 25%, 50%, and 25% of the original size. 200% of the size, a new image (ie, the zoomed foreign body image) is acquired.

Step 20-20-40: Change the image noise. For the obtained foreign body image, image noise is introduced to expand the number of foreign body images. The specific method is as follows. The image is processed by adding noise. The formula is as follows: f(x,y)=g(x,y)+q (.), where x and y represent the horizontal and vertical coordinates of the pixels in the image of the foreign body, g(x,y) represents the true value of each pixel, for multi-channel images, g(x,y) represents each pixel The corresponding vector, q(.) represents the noise function, f(x,y) represents the value of each pixel of the foreign body image after the noise is introduced, where the noise is introduced by the probability density function of the Gaussian random variable, and the noise function is The following formula represents:

Among them: z is a variable, z=g(x,y), q(z) is the probability density function of z, μ is the expectation of z, and σ is the variance of z.

Steps 20-20-50: shooting scenes in the original substation. Use robots and fixed cameras to take pictures of the parts of the equipment and components that are easy to be entangled with foreign objects in the substation and the parts that are easily affected by foreign objects and cause failures. By changing the shooting distance, angle, and aperture changes, the diversity of the images is improved, and the retention image.

Steps 20-20-60: image stitching. The image of the foreign object body is spliced to the part that is easy to be entangled or affected by the foreign object in the scene image in the substation taken by the robot and the fixed camera to simulate the intrusion of the foreign object in the substation.

Step 20-30: Image summary. Collect and number the images collected and designed in steps 20-10 and 20-20-60.

Step 30: Image annotation. For each image, when the target (ie foreign body) is completely displayed in the image, use the smallest rectangle to mark all the targets, including bird nests, honeycombs, kites, balloons, plastic films, dust-proof nets, etc.; for targets When only a part of the object is displayed in the image, use the smallest rectangle to mark this part. For the case where the target is partially occluded, use the smallest rectangle to mark the visible part of the target. Label each tag in the image, including bird’s nest, kite, balloon, plastic film, dust net, etc.

Step 40: Model deployment; if the model is obtained, perform model training and deploy the trained model to the hardware device to provide services for the later substation foreign body identification; if the model is not obtained, perform step 40-10- again Step 40-40;

The model training is an iterative cycle process, and each iteration includes main steps such as data splitting, algorithm parameter adjustment, training model, and model verification.

Step 40-10: Data splitting. Divide all images into three parts, which are training set, validation set, and test set, so that the number of images in the three subsets accounts for 70%, 20%, and 10% of all images, respectively.

Step 40-20: Algorithm tuning. Select the target detection algorithm based on the deep learning algorithm, and make preliminary adjustments to the parameters of the algorithm.

Step 40-30: Train the model. The algorithm is trained on the basis of the training set images to obtain the target detection model of foreign objects in the substation.

Step 40-40: Model verification. On the basis of the verification set, the accuracy of the target detection model is evaluated. If the accuracy does not meet the requirements, the parameters of the target detection algorithm are adjusted, and the training set is used for model training, and the verification set is used for the accuracy of the model Evaluation of the degree of accuracy; if it is verified to meet the requirements, proceed to the next step. If it does not meet the requirements, return to the algorithm tuning step and repeat the previous step until the accuracy meets the conditions. The accuracy of the model is again determined on the basis of the test set. For evaluation, if the accuracy meets the requirements, the model is retained. If the accuracy cannot meet the requirements, the quality of the sample is re-expanded and optimized through the step 20, and the model training is completed in sequence according to the process until the model finally meets the accuracy requirements.

The model deployment is to perform model training on the acquired model, and then deploy the trained model to the hardware device. The model deployment methods are divided into the following four types. According to the actual hardware configuration in the substation, they are respectively deployed to the substation inspection robot body , Fixed camera body, remote control system of substation patrol robot and fixed camera remote control system.

Step 50: Image detection. The main steps of image detection include real-time collection of robot patrol video and fixed camera video, video frame extraction, image processing, and image detection.

Step 50-10: Real-time collection of robot patrol video and fixed camera video. In the actual inspection and monitoring process, both the robot and the fixed camera will take real-time video shooting, and they need to pass these real-time videos to the intelligent recognition module.

Step 50-20: Video frame extraction. The module will frame the video at a fixed interval. The length of the interval depends on the performance of the module and the computing power of the hardware carrier.

Step 50-30: Image processing. The frame image needs to be image enhanced, and the image enhancement method includes linear grayscale enhancement and logarithmic function nonlinear transformation. Due to the images subject to shooting conditions, the above-mentioned frame images may have poor quality, which affects the recognition accuracy of the intelligent recognition module. Therefore, each frame of image needs to be enhanced. The specific methods include linear grayscale enhancement and logarithmic function. Non-linear transformation.

The method of linear gray scale enhancement is as follows,

Among them, g(x,y) is the true value of each pixel, [a,b] is the grayscale range of the original image, [c,d] is the expected grayscale range of the new image, f(x,y) is The pixel value of the image after the change.

The nonlinear transformation formula of logarithmic function is as follows,

Among them, h(x,y) is the pixel value of the image after the nonlinear transformation of the logarithmic function, and a, b, and c are adjustable parameters. After iterative correction, the grayscale range of the newly generated image can be within the ideal range. . If the processed image is a color image, the color to grayscale conversion is required before the image is enhanced. Here, the built-in method of opencv can be used to achieve. The enhanced grayscale image can also be converted back to a color image by the method provided by opencv.

Step 50-40: Detect the image. Send the corrected picture to the image recognition module, which will detect whether the image contains the target object, the specific pixel position of the object, and the type of the object.

Step 60: Carry out an alarm and provide alarm information.

This application proposes a method, device, system, electronic equipment and storage medium for intelligent recognition of foreign objects in substations based on end-to-end deep learning algorithms. The intelligent foreign object recognition module based on this method, referred to as the recognition module, can be integrated into robots and cameras. The real-time and intelligent identification of foreign objects in the main body or their remote control system not only improves the accuracy of foreign object detection, but also can specifically distinguish the types of foreign objects. During the inspection and monitoring process, the robot and the smart camera will continuously take images and feed back the video data to the foreign object intelligent recognition module in real time. This module will split the video into pictures, extract frames from them at fixed intervals, analyze them, and then detect them. Foreign objects in the current monitoring screen. This application can also play a role in assisting substation inspection and monitoring.

The fixed camera in the current substation does not yet have the ability to intelligently recognize the intrusion of foreign objects. Although the robot has a preliminary foreign object recognition function during the inspection process, the technology adopted is based on image comparison, resulting in low recognition accuracy. Poor reliability and lack of intelligence. This application proposes a method for identifying foreign objects in substations. The modules developed based on this method can be integrated into fixed cameras and robot bodies and their remote control systems, giving them intelligent recognition of foreign objects in substations. Ability to improve the intelligent level of substation inspection and monitoring.

Example 2

This application also provides a device for identifying foreign objects in a substation. The principle of the device for solving a technical problem is similar to that of a method for identifying foreign objects in a substation, and the repetition will not be repeated.

The foreign body identification device of the substation includes:

The acquisition module is set to collect scene images of the substation;

The recognition module is configured to recognize the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation to obtain the location of the foreign body in the scene image and the type of the foreign body; wherein, the target of the foreign body in the substation The detection model is obtained by training according to preset foreign body types and foreign body images.

Example 3

This application also provides a substation foreign body recognition system, including a substation patrol robot and a substation foreign body recognition device as described in Embodiment 2. The power substation patrol robot includes a camera device, and the power substation patrol robot is set up To collect scene images of the substation through the camera device.

The present application also provides an electronic device, which includes a processor and a memory, and the memory stores a computer program. When the computer program is executed by the processor, the method for identifying foreign objects in a substation provided by any embodiment of the application is implemented.

The embodiment of the present application also provides a computer-readable storage medium. In an embodiment, the foregoing storage medium stores a computer program, and the computer program, when executed by a processor, implements the method for identifying foreign objects in a substation as provided in any embodiment of the present application.

Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, this application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this application may use one or more computer-usable storage media containing computer-usable program codes (including but not limited to disk storage, compact Disc Read-Only Memory, CD-ROM), optical storage, etc. ) In the form of a computer program product implemented on it.

This application is described with reference to flowcharts and/or block diagrams of methods, devices (systems), and computer program products according to embodiments of this application. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment are generated It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.

Claims (15)

A method for identifying foreign objects in a substation, including: Collect scene images of substations; Identify the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation, and obtain the location of the foreign body in the scene image and the type of the foreign body; Wherein, the target detection model of foreign objects in the substation is obtained by training based on preset types of foreign objects in the substation and multiple images of foreign objects in the substation. The method according to claim 1, wherein the type of the foreign body includes at least one of the following: Bird nests, honeycombs, kites, balloons, plastic films and dust nets. The method according to claim 1, wherein the process of establishing a target detection model of foreign objects in the substation comprises: Collect foreign body scene images of substations and simulated foreign body scene images of substations, mark each label in each image of the foreign body scene images of the substation and the simulated foreign body scene images of the substation with the label to which each label belongs, and mark The subsequent foreign body scene images in the substation and the simulated foreign body scene images in the substation are summarized to obtain a summarized image; wherein the label to which each label belongs is a text description of the type of each label; The collected images are used for model training to obtain the target detection model of foreign objects in the substation. The method according to claim 3, wherein the collecting a scene image of a foreign object in a substation comprises: Collect historical images of foreign objects in the substation, where the number of historical images is multiple; According to the type of the foreign body in the historical image, the poor-quality historical image is eliminated. The poor-quality historical image includes: the historical image with the pixel area occupied by the foreign body less than 5% of the pixel area of the historical image, display In historical images where the surface area of the foreign body is less than 10% of the actual surface area of the foreign body, and in historical images where the size of the foreign body displayed in the historical image is less than 10% of the actual size of the foreign body At least one The simulated foreign body scene image of the substation includes: collecting the original foreign body image and the original substation scene image; cutting the foreign body in the original foreign body image to obtain the foreign body image; and performing the preset expansion method on the foreign body image Processing to expand the foreign body image; splicing each of the foreign body image and the expanded foreign body image with the original substation scene image to obtain a simulated foreign body scene image of the substation. The method according to claim 4, wherein said collecting an image of the original foreign body body comprises: Under the condition of ensuring that the foreign objects in the substation fall in the central area of the field of view of the camera equipment, the original objects with different brightness can be obtained by adjusting the aperture of the camera equipment and using the camera equipment to shoot the foreign objects from different angles and different distances. The image of the foreign body; and, downloading the image of the foreign body through the network to obtain the original foreign body image. The method of claim 4, wherein: Cropping the foreign body in the original foreign body image to obtain the foreign body image includes: removing the background and cutting along the edge of each image in the original foreign body image to obtain the foreign body image; The processing the foreign body image by a preset expansion method to expand the foreign body image includes: performing inversion, cropping, scaling, and noise processing on the foreign body image to obtain an expanded foreign body image; The collection of scene images of the original substation includes: using a robot and a fixed camera needle to take pictures of the parts of the designated equipment and the parts of the designated parts in the substation, and adjust the shooting distance, angle and aperture during the shooting process to obtain the original substation Scene image Stitching each of the foreign body image and the expanded foreign body image with the original substation scene image includes: stitching each of the foreign body image and the expanded foreign body image to the The location of the designated equipment and the location of the designated component in the original substation scene image are used to simulate the scene image of foreign objects in the substation. The method according to claim 6, wherein: Reversing the image of the foreign object body includes: reversing the image of the foreign object body once every 30 degrees to obtain a reversed foreign object body image; Cutting the foreign body image includes: cutting the foreign body image from top, bottom, left, and right to 20%, 40%, and 60% of the foreign body image to obtain the cut foreign body image. image; The scaling of the foreign body image includes: adjusting the foreign body image to 25%, 50%, and 200% of the size of the foreign body image to obtain a scaled foreign body image; Performing noise processing on the foreign body image includes: introducing image noise to the foreign body image to obtain a noise-processed foreign body image; Wherein, the expanded foreign body image includes: the foreign body body image after inversion, the foreign body body image after cropping, the foreign body body image after scaling, and the foreign body body image after noise processing. The method according to claim 7, wherein: The introducing image noise to the image of the foreign body body includes: Introduce image noise to the foreign body image by the following formula: f(x,y)=g(x,y)+q(.), Among them, x and y respectively represent the horizontal and vertical coordinates of the pixels in the foreign body image, g(x,y) represents the true value of the pixels in the foreign body image, and for multi-channel images, g(x,y) represents the The vector corresponding to the pixel in the foreign body image, q(.) represents the noise function, f(x,y) represents the value of the pixel of the foreign body image after the noise is introduced; the noise function is derived from the probability density function of the Gaussian random variable Introduced, the noise function is expressed by the following formula:

Among them: z is a variable, z=g(x,y), q(z) is the probability density function of z, μ is the expectation of z, and σ is the variance of z. The method according to claim 1, wherein said collecting a scene image of a substation comprises: Collecting scene videos of the substation and extracting frames from the scene videos to obtain scene images; Performing image processing on the scene image to obtain a scene image of the substation; The image processing includes linear grayscale enhancement and logarithmic function nonlinear transformation. The method according to claim 9, wherein the method of linear grayscale enhancement comprises:

Where g(x,y) is the true value of the pixel of the scene image, [a,b] is the grayscale range of the scene image, and [c,d] is the grayscale of the image after linear grayscale enhancement The expected range of f(x,y) is the pixel value of the image after linear grayscale enhancement; The nonlinear transformation formula of the logarithmic function is as follows:

Among them, h(x, y) is the pixel value of the image after the nonlinear transformation of the logarithmic function, and a, b, and c are adjustable parameters. After iterative correction, the gray range of the newly generated image is within the specified range. The method according to claim 10, wherein, when the scene image is a color image, the image processing further comprises: grayscale, and the grayscale is implemented by a built-in method of opencv. A device for identifying foreign objects in a substation includes: The acquisition module is set to collect scene images of the substation; The recognition module is configured to recognize the foreign body in the scene image according to the pre-established target detection model of the foreign body in the substation, and obtain the location of the foreign body in the scene image and the type of the foreign body; Wherein, the target detection model of the foreign body in the substation is obtained by training according to a preset foreign body type and foreign body image. A substation foreign body recognition system, comprising a substation patrol robot and a substation foreign body recognition device according to claim 12, wherein the substation patrol robot includes a camera device, and the substation patrol robot is set to pass through The camera device collects scene images of the substation. An electronic device comprising a processor and a memory, the memory storing a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1 to 11 is implemented. A computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images