Augmented reality intelligent inspection device and method for power system

Abstract

The invention provides an augmented reality intelligent inspection device and method for a power system, and the method comprises the steps: capturing an image and a video of equipment in the power system through a high-definition camera at a front end, and carrying out the real-time analysis and processing of the image through an artificial intelligence algorithm, so as to recognize the abnormality and fault of the equipment; the glasses with the augmented reality display function are arranged on the safety helmet, the recognition result is displayed in real time, and visual and clear visual assistance is provided for inspection personnel. A cloud assistance module is arranged, cloud expert assistance is requested in case of special situations, and problems are synchronously analyzed and solved through real-time video data; the advantages of augmented reality and artificial intelligence technologies are integrated, the actual requirements of electric power system inspection are fully considered, the inspection efficiency and accuracy can be remarkably improved, safe production of workers is ensured, and the intelligent level of electric power system inspection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power inspection, and in particular to an augmented reality intelligent inspection device and method for an electric power system. Background Art

[0002] With the rapid increase in the scale and complexity of power network systems, the traditional inspection mode has problems such as high labor costs, long time consumption, and hidden faults that are difficult to detect, making it difficult to ensure the safe and efficient operation of the power grid. Improving the efficiency and accuracy of power system inspections with the help of advanced scientific and technological means is an important development direction in the current field of power inspection technology. The rapid development of augmented reality technology has provided a new idea for the intelligent transformation of power system inspections. By superimposing virtual information on real scenes, the visual perception of inspectors can be enhanced; in addition, the combination of augmented reality technology and artificial intelligence technology can accurately identify equipment anomalies through deep learning models in computer vision, which can further improve diagnostic efficiency and accuracy. Despite the support of these advanced technical theories, the current power system inspection still lacks corresponding intelligent devices as a carrier of technology. The device should not only comprehensively apply augmented reality and artificial intelligence technologies to give full play to the technical advantages; it also needs to adapt to the complex power system environment and ensure a certain degree of safety and convenience, so as to achieve efficient and accurate inspection tasks and provide strong guarantees for the stable operation of the power system. Summary of the invention

[0003] The purpose of the present invention is to address the deficiencies of the above-mentioned prior art and to provide an augmented reality intelligent inspection device and method for an electric power system, which can display the recognition results in real time and provide intuitive and clear visual assistance to the inspection personnel.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] The present invention provides an augmented reality intelligent inspection device for a power system, comprising a front end and a back end, wherein the front end comprises a data acquisition module and an augmented reality display module; the back end comprises an edge computing module and a cloud assistance module;

[0006] The data acquisition module includes a high-definition camera for real-time image data acquisition, wherein the high-definition camera integrates an image sensor and a lens module;

[0007] The augmented reality display module includes a lens capable of projecting a display image, and a micro-projection element is installed on the lens, which can project the display image from the rear end onto the lens, so that the real image observed by the human eye overlaps with the real field of view. The lens has two pieces, and each piece of the lens is 20 mm wide.

[0008] The edge computing module: uses the ROC-RK3328-CC development board as a hardware platform and is connected to the front end via a data cable; the edge computing module is equipped with a wireless LAN module to achieve communication and data sharing with the back-end cloud assistance module; the edge computing module has a built-in artificial intelligence image detection algorithm to process the image data collected by the front end, accurately detect whether there is an abnormality in the power transmission and transformation equipment, and mark the abnormal position in the image;

[0009] The cloud assistance module: communicates and shares data with the edge computing module through a wireless local area network of the TCP / IP protocol.

[0010] Furthermore, the artificial intelligence image detection algorithm includes data preprocessing and data reasoning, which can be expressed as:

[0011] X′=f PRE (X)

[0012] [u e ,v e ,d u ,d v ,θ,s]=f YOLO (X′,w) (1);

[0013] Among them, X is the image data of the power transmission and transformation equipment collected by the front-end data acquisition module, m and n are the height and width of the image data of power transmission and transformation equipment respectively; f PRE (X) is the image preprocessing function; X′ is the image after preprocessing, l and p are the height and width of the preprocessed image X′ respectively; f YOLO (X′,w) is the model inference function; w is the set of network model parameters obtained through training, w∈{w (1) ,w (2) ,…,w (L)}; where w (j) ,j∈[1,L] is the weight tensor of the jth layer in the network model;

[0014] L is the number of layers in the network model, the number of layers L and w (i) The dimensions of are pre-set in the network model; [u e ,v e ,d u ,d v ,θ,s] is the output of the algorithm, u e and v e They respectively represent the coordinates of the pixels with abnormal power transmission and transformation equipment in the power transmission and transformation equipment image data in the U-axis and V-axis directions of the image; d u and d vare the height and width of the abnormal annotation box in the image; θ is the angle value of the abnormal annotation box; s is the confidence of the abnormality.

[0015] Furthermore, the image preprocessing function f PRE (X) Specifically:

[0016] S101, normalize the image data of the power transmission and transformation equipment, convert the pixel values ​​of the power transmission and transformation equipment image to the range of [-1, 1], and the normalization calculation is:

[0017]

[0018] Among them, I(u,v) is the image data of the power transmission and transformation equipment at the pixel coordinate (u,v); I min is the minimum value of all pixel values ​​of the power transmission and transformation equipment image data; I max is the maximum value of all pixel values ​​of the power transmission and transformation equipment image data; I norm (u,v) is the normalized pixel value at the pixel coordinate (u,v);

[0019] S102, after performing the normalization calculation of formula (2) on all pixels of the power transmission and transformation equipment image data, perform a scale transformation to convert the image size from m×n to l×p; construct a preprocessed image of size l×p, and calculate the pixel coordinates (u, v) in the power transmission and transformation equipment image data corresponding to the pixel coordinates (u′, v′) of the preprocessed image as follows:

[0020]

[0021] Among them, floor() is an operation to take the integer part;

[0022] The pixel value I′(u′,v′) at the pixel coordinate (u′,v′) in the preprocessed image is obtained by bilinear interpolation and is calculated as:

[0023]

[0024] The pixel value of each pixel in the preprocessed image is calculated according to formula (3) and formula (4), and the preprocessed image after scale transformation is obtained, thus completing the preprocessing.

[0025] Furthermore, the network model parameter set w obtained through training is specifically:

[0026] S201, collect 1000 pictures of power transmission and transformation equipment in the power system in normal or abnormal state; the abnormal images in all pictures are X′ L(i) , i = 0, 1, ..., N, N is the total number of images with abnormalities; for image X′L(i) The abnormalities of the power transmission and transformation equipment are marked, and the marked information is expressed as follows:

[0027] Label i =[u eL(i) ,v eL(i) ,d uL(i) ,d vL(i) ,θ L(i) ,s L(i) ],i=0,1,…,N (5);

[0028] Among them, Label i is the abnormal annotation information of the i-th image with abnormality; u eL(i) 、v eL(i) d uL(i) d vL(i) ,θ L(i) is the coordinates of the U axis direction, the V axis direction, the standard frame height, the label frame width, and the label frame angle of the i-th abnormal image; s L(i) =1;

[0029] S202. Establish a loss function for training model parameters:

[0030]

[0031] Among them, λ CIoU , obj and λ angle are the manually set weight coefficients representing bounding box loss, confidence loss, and angle loss; L CIoU(i) is the loss of the abnormal bounding box in the i-th image with an abnormality, calculated as:

[0032]

[0033] Among them, IoU (i) is the intersection-over-union ratio of the abnormal detection box and the annotation box in the i-th image with abnormality; c i L is the diagonal distance between the minimum closure area of ​​the abnormal detection box and the annotation box in the i-th abnormal image; obj(i) is the confidence loss of the anomaly detection result in the i-th image with anomalies;

[0034] L obj(i) =(p i IoU (i) -s L(i) ) 2 ;

[0035] Among them, p i Indicates the probability value that the abnormal prediction box in the i-th image with abnormality contains the abnormality; is the model inference function fYOLO The intermediate variables generated in the calculation process of (X′,w); L angle(i) The angle loss of the anomaly detection box in the i-th image with anomalies is calculated as:

[0036] L angle(i) =(θ (i) -θ L(i) ) 2 ;

[0037] S203, using the gradient descent algorithm to calculate the loss function L(f YOLO (X i ',w),Label i ) is optimized and the model parameters when the loss function is minimized are calculated as the network model parameter set w.

[0038] Furthermore, an enhanced intelligent inspection method for a power system is implemented by using the augmented reality intelligent inspection device for a power system, and includes:

[0039] S301, through the data acquisition module, simulate the perspective of the inspection worker to collect real-time image and video data;

[0040] S302, transmitting the collected image and video data to an edge computing module, executing an artificial intelligence image detection algorithm in the edge computing module to detect abnormal conditions of power transmission and transformation equipment in the image;

[0041] S303, the edge computing module marks the position of the detected abnormal equipment, and transmits the picture containing the marked information to the augmented reality display module, so as to assist the inspection workers to quickly locate and judge the abnormal status of the equipment in the display environment;

[0042] S304. When the auxiliary information of 303 differs from the inspection worker's judgment, the inspection worker calls the cloud assistance module through the inspection device to request the assistance of the cloud expert. The cloud expert makes judgments and analyses based on the video and image data transmitted by the device and provides corresponding maintenance guidance.

[0043] The beneficial effects of the present invention are as follows: it can be divided into two parts, the front end and the back end; the front end integrates a high-definition camera and glasses with augmented reality display function, which can be directly installed on a safety helmet, capturing images of power transmission and transformation equipment in the power system in real time and intuitively displaying abnormal detection information of the equipment for inspection personnel; the back end deploys the artificial intelligence detection algorithm in the edge computing module, and combines it with a cloud assistance module to achieve real-time data processing and remote expert assistance, which can cope with a variety of on-site problems and ensure that inspection workers can quickly obtain professional guidance and support in complex situations. The present invention combines the advantages of augmented reality and artificial intelligence technology, and its design fully considers the actual needs of power system inspections, which can significantly improve inspection efficiency and accuracy, ensure workers' safe production, and improve the intelligence level of power system inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of an augmented reality intelligent inspection method for a power system;

[0045] Figure 2 A schematic diagram of workers using an augmented reality intelligent inspection device for a power system;

[0046] Figure 3 A module architecture diagram of an augmented reality intelligent inspection device for a power system;

[0047] Figure 4 A physical picture of an augmented reality intelligent inspection device for a power system;

[0048] Figure 5 This is the data set of some backend artificial intelligence algorithms and the detection effect diagram. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0050] See also Figure 1 , an augmented reality intelligent inspection device for a power system, comprising a front end and a back end, wherein the front end comprises a data acquisition module and an augmented reality display module; the back end comprises an edge computing module and a cloud assistance module;

[0051] The data acquisition module includes a high-definition camera for real-time image data acquisition, wherein the high-definition camera integrates an image sensor and a lens module;

[0052] The augmented reality display module includes a lens capable of projecting a display image, and a micro-projection element is installed on the lens, which can project the display image from the rear end onto the lens, so that the real image observed by the human eye overlaps with the real field of view. The lens has two pieces, and each piece of the lens is 20 mm wide.

[0053] In a specific embodiment, the lens is integrated with a high-definition camera at the front end, which is compact, has a diameter of less than 10 mm, weighs less than 2 g, and can collect image data in real time. The high-definition camera and the lens are integrated into the form of conventional glasses, wherein the high-definition camera is located above the lens to simulate the user's perspective for image data collection;

[0054] The device has an open Type C interface, which is connected to the back-end edge computing module through a USB 3.0 standard data cable, and is connected to the data acquisition module and augmented reality display module through an internal hub. The glasses also include a clamp that can be directly clamped on a safety helmet, and weigh only 150g, which is easy to install and carry.

[0055] The edge computing module: uses the ROC-RK3328-CC development board as a hardware platform and is connected to the front end via a data cable; the edge computing module is equipped with a wireless LAN module to achieve communication and data sharing with the back-end cloud assistance module; the edge computing module has a built-in artificial intelligence image detection algorithm to process the image data collected by the front end, accurately detect whether there is an abnormality in the power transmission and transformation equipment, and mark the abnormal position in the image;

[0056] In a specific implementation, the edge computing module is equipped with an Android 7.1 operating system and has an open Type C interface. It is connected to the modified special glasses (composed of a high-definition camera and lenses) through a USB 3.0 standard data cable to achieve fast data transmission. At the same time, the edge computing module has a built-in wireless local area network WLAN module that complies with the IEEE 802.11 standard, and supports wireless communication and data sharing with the cloud assistance module. The edge computing module adopts a compact design with dimensions of 140 mm in length, 60 mm in width, and 15 mm in thickness. It weighs about 250 g, and its operation is similar to that of a mobile phone, making it easy to carry and operate.

[0057] The cloud assistance module: communicates and shares data with the edge computing module through a wireless local area network of the TCP / IP protocol.

[0058] The artificial intelligence image detection algorithm includes data preprocessing and data reasoning, which can be expressed as:

[0059] X′=f PRE (X)

[0060] [u e ,v e ,d u ,d v ,θ,s]=f YOLO (X′,w) (1);

[0061] Among them, X is the image data of the power transmission and transformation equipment collected by the front-end data acquisition module, m and n are the height and width of the image data of power transmission and transformation equipment respectively; f PRE (X) is the image preprocessing function; X′ is the image after preprocessing, l and p are the height and width of the preprocessed image X′ respectively; f YOLO (X′,w) is the model inference function; w is the set of network model parameters obtained through training, w∈{w (1) ,w (2) ,…,w (L)}; where w (j) ,j∈[1,L] is the weight tensor of the jth layer in the network model;

[0062] L is the number of layers in the network model, the number of layers L and w (i) The dimensions of are pre-set in the network model; [u e ,v e ,d u ,d v ,θ,s] is the output of the algorithm, u e and v e They respectively represent the coordinates of the pixels with abnormal power transmission and transformation equipment in the power transmission and transformation equipment image data in the U-axis and V-axis directions of the image; d u and d v are the height and width of the abnormal annotation box in the image; θ is the angle value of the abnormal annotation box; s is the confidence of the abnormality.

[0063] The image preprocessing function f PRE (X) Specifically:

[0064] S101, normalize the image data of the power transmission and transformation equipment, convert the pixel values ​​of the power transmission and transformation equipment image to the range of [-1, 1], and the normalization calculation is:

[0065]

[0066] Among them, I(u,v) is the image data of the power transmission and transformation equipment at the pixel coordinate (u,v); I min is the minimum value of all pixel values ​​of the power transmission and transformation equipment image data; I maxis the maximum value of all pixel values ​​of the power transmission and transformation equipment image data; I norm (u,v) is the normalized pixel value at the pixel coordinate (u,v);

[0067] S102, after performing the normalization calculation of formula (2) on all pixels of the power transmission and transformation equipment image data, perform a scale transformation to convert the image size from m×n to l×p; construct a preprocessed image of size l×p, and calculate the pixel coordinates (u, v) in the power transmission and transformation equipment image data corresponding to the pixel coordinates (u′, v′) of the preprocessed image as follows:

[0068]

[0069] Among them, floor() is an operation to take the integer part;

[0070] The pixel value I′(u′,v′) at the pixel coordinate (u′,v′) in the preprocessed image is obtained by bilinear interpolation and is calculated as:

[0071]

[0072] The pixel value of each pixel in the preprocessed image is calculated according to formula (3) and formula (4), and the preprocessed image after scale transformation is obtained, thus completing the preprocessing.

[0073] The network model parameter set w obtained through training is specifically:

[0074] S201, collect 1000 pictures of power transmission and transformation equipment in the power system in normal or abnormal state; the abnormal images in all pictures are X′ L(i) , i = 0, 1, ..., N, N is the total number of images with abnormalities; for image X′ L(i) The abnormalities of the power transmission and transformation equipment are marked, and the marked information is expressed as follows:

[0075] Label i =[u eL(i) ,v eL(i) ,d uL(i) ,d vL(i) ,θ L(i) ,s L(i) ],i=0,1,…,N (5);

[0076] Among them, Label i is the abnormal annotation information of the i-th image with abnormality; u eL(i) 、v eL(i) d uL(i) d vL(i) ,θ L(i)is the coordinates of the U axis direction, the V axis direction, the standard frame height, the label frame width, and the label frame angle of the i-th abnormal image; s L(i) =1;

[0077] S202. Establish a loss function for training model parameters:

[0078]

[0079] Among them, λ CIoU , obj and λ angle are the manually set weight coefficients representing bounding box loss, confidence loss, and angle loss; L CIoU(i) is the loss of the abnormal bounding box in the i-th image with an abnormality, calculated as:

[0080]

[0081] Among them, IoU (i) is the intersection-over-union ratio of the abnormal detection box and the annotation box in the i-th image with abnormality; c i L is the diagonal distance between the minimum closure area of ​​the abnormal detection box and the annotation box in the i-th abnormal image; obj(i) is the confidence loss of the anomaly detection result in the i-th image with anomalies;

[0082] L obj(i) =(p i IoU (i) -s L(i) ) 2 ;

[0083] Among them, p i Indicates the probability value that the abnormal prediction box in the i-th image with abnormality contains the abnormality; is the model inference function f YOLO The intermediate variables generated in the calculation process of (X′,w); L angle(i) The angle loss of the anomaly detection box in the i-th image with anomalies is calculated as:

[0084] L angle(i) =(θ (i) -θ L(i) ) 2 ;

[0085] S203, using the gradient descent algorithm to calculate the loss function L(f YOLO (X i ',w),Label i ) is optimized and the model parameters when the loss function is minimized are calculated as the network model parameter set w.

[0086] An enhanced intelligent inspection method for a power system is implemented by using an augmented reality intelligent inspection device for the power system, comprising:

[0087] S301, through the data acquisition module, simulate the perspective of the inspection worker to collect real-time image and video data;

[0088] S302, transmitting the collected image and video data to an edge computing module, executing an artificial intelligence image detection algorithm in the edge computing module to detect abnormal conditions of power transmission and transformation equipment in the image;

[0089] S303, the edge computing module marks the position of the detected abnormal equipment, and transmits the picture containing the marked information to the augmented reality display module, so as to assist the inspection workers to quickly locate and judge the abnormal status of the equipment in the display environment;

[0090] S304. When the auxiliary information of 303 differs from the inspection worker's judgment, the inspection worker calls the cloud assistance module through the inspection device to request the assistance of the cloud expert. The cloud expert makes judgments and analyses based on the video and image data transmitted by the device and provides corresponding maintenance guidance.

[0091] Embodiment 1:

[0092] 1500 pictures containing normal and abnormal equipment were collected from a substation of State Grid Enshi Power Supply Co., Ltd., of which 1000 were used to train the model of the artificial intelligence detection algorithm based on YOLOv5, and the remaining 500 were used as a test set. The test set was input into the trained model for offline anomaly detection. The results show that the model successfully identified 93.9% of the abnormal equipment pictures, proving the high reliability of the AI ​​algorithm. Specifically, for pictures marked as abnormal, the detection accuracy was 97.9% and the false detection rate was 1.2%; while for pictures marked as normal, the detection accuracy was 98.9% and the false detection rate was 1.4%. After the algorithm was verified, the operational reliability of the inspection device itself was verified on-site at a substation of State Grid Enshi Power Supply Co., Ltd., for abnormal detection of circuit breakers and disconnectors. On the premise of ensuring the safe operation of the equipment, the abnormal state of disconnection of the circuit breaker and the disconnector was artificially created, and online real-time detection was performed by wearing the intelligent inspection device of the present invention. The experimental results show that the device can operate normally and accurately detect abnormal conditions of circuit breakers and disconnectors in real time, and promptly prompt the experimenters with abnormal conditions of the equipment through augmented reality glasses.

[0093] like Figure 2As shown in the figure, the inspection workers can obtain two parts of auxiliary information: the first part is the abnormal situation in the on-site image identified by the artificial intelligence detection algorithm; the second part is the on-site picture and voice shared with cloud experts, so as to obtain expert analysis and guidance. All this information can be directly and conveniently displayed in the field of vision of the inspection workers through the augmented reality display module to improve the efficiency and accuracy of the inspection work.

[0094] like Figure 3 and Figure 4 As shown, for inspection workers, the device of the present invention mainly includes two parts of hardware: one is the front-end augmented reality glasses integrated with a high-definition camera and a special augmented reality display lens; the other is the edge computing device developed and packaged with the ROC-RK3328-CC development board as the main body;

[0095] like Figure 5 As shown in the figure, the algorithm can detect abnormalities in power transmission and transformation equipment. After detecting an abnormality, a red box will be drawn in the image to mark the abnormal location. At the same time, the probability of the algorithm's judgment of the abnormality will be noted. Inspection workers can use this probability as an aid to judge equipment abnormalities. If there is a dispute between the algorithm result and the inspection worker's own judgment, the assistance of cloud experts can be requested to ensure the accuracy and safety of the inspection.

[0096] The above-mentioned embodiments only express the implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be based on the attached claims.

Claims

1. An augmented reality intelligent inspection device for a power system, characterized in that: It includes a front end and a back end, wherein the front end includes a data acquisition module and an augmented reality display module; the back end includes an edge computing module and a cloud assistance module; The data acquisition module includes a high-definition camera for real-time image data acquisition, wherein the high-definition camera integrates an image sensor and a lens module; The augmented reality display module includes a lens capable of projecting a display image, and a micro-projection element is installed on the lens, which can project the display image from the rear end onto the lens, so that the real image observed by the human eye overlaps with the real field of view. The lens has two pieces, and each piece of the lens is 20 mm wide. The edge computing module: uses the ROC-RK3328-CC development board as a hardware platform and is connected to the front end via a data cable; the edge computing module is equipped with a wireless LAN module to achieve communication and data sharing with the back-end cloud assistance module; the edge computing module has a built-in artificial intelligence image detection algorithm to process the image data collected by the front end, accurately detect whether there is an abnormality in the power transmission and transformation equipment, and mark the abnormal position in the image; The cloud assistance module: communicates and shares data with the edge computing module through a wireless local area network of the TCP / IP protocol.

2. The augmented reality intelligent inspection device for a power system according to claim 1, characterized in that: The artificial intelligence image detection algorithm includes data preprocessing and data reasoning, which can be expressed as: X′=f PRE (X) [u e ,v e ,d u ,d v ,θ,s]=f YOLO (X′,w) (1); Among them, X is the image data of the power transmission and transformation equipment collected by the front-end data acquisition module, m and n are the height and width of the image data of power transmission and transformation equipment respectively; f PRE (X) is the image preprocessing function; X′ is the image after preprocessing, l and p are the height and width of the preprocessed image X′ respectively; f YOLO (X′,w) is the model inference function; w is the set of network model parameters obtained through training, w={w (1) ,w (2) ,…,w (L) }; where w (j) ,j∈[1,L] is the weight tensor of the jth layer in the network model; L is the number of layers in the network model, the number of layers L and w (j) The dimensions of are pre-set in the network model; [u e ,v e ,d u ,d v ,θ,s] is the output of the algorithm, u e and v e They respectively represent the coordinates of the pixels with abnormal power transmission and transformation equipment in the power transmission and transformation equipment image data in the U-axis and V-axis directions of the image; d u and d v are the height and width of the abnormal annotation box in the image; θ is the angle value of the abnormal annotation box; s is the confidence of the abnormality.

3. The augmented reality intelligent inspection device for a power system according to claim 2, characterized in that: The image preprocessing function f PRE (X) Specifically: S101, normalize the image data of the power transmission and transformation equipment, convert the pixel values ​​of the power transmission and transformation equipment image to the range of [-1, 1], and the normalization calculation is: Among them, I(u,v) is the image data of the power transmission and transformation equipment at the pixel coordinate (u,v); I min is the minimum value of all pixel values ​​of the power transmission and transformation equipment image data; I max is the maximum value of all pixel values ​​of the power transmission and transformation equipment image data; I norm (u,v) is the normalized pixel value at the pixel coordinate (u,v); S102, after performing the normalization calculation of formula (2) on all pixels of the power transmission and transformation equipment image data, perform a scale transformation to convert the image size from m×n to l×p; construct a preprocessed image of size l×p, and calculate the pixel coordinates (u, v) in the power transmission and transformation equipment image data corresponding to the pixel coordinates (u′, v′) of the preprocessed image as follows: Among them, floor() is the operation of taking the integer part; The pixel value I′(u′,v′) at the pixel coordinate (u′,v′) in the preprocessed image is obtained by bilinear interpolation and is calculated as: The pixel value of each pixel in the preprocessed image is calculated according to formula (3) and formula (4), and the preprocessed image after scale transformation is obtained, thus completing the preprocessing.

4. The augmented reality intelligent inspection device for a power system according to claim 3, characterized in that: The network model parameter set w obtained through training is specifically: S201, collect 1000 pictures of power transmission and transformation equipment in the power system in normal or abnormal state; the abnormal images in all pictures are X′ L(i) , i = 0, 1, ..., N, N is the total number of images with abnormalities; for image X′ L(i) The abnormalities of the power transmission and transformation equipment are marked, and the marked information is expressed as follows: Label i J[u eL(i) ,v eL(i) ,d uL(i) ,d vL(i) ,θ L(i) ,sh L(i) ],iS0,1,...,N (5) Among them, Label i is the abnormal annotation information of the i-th image with abnormality; u eL(i) 、v eL(i) ,d uL(i) ,d vL(i) ,θ L(i) is the U-axis coordinate, V-axis coordinate, standard frame height, label frame width, and label frame angle of the abnormality in the i-th image with abnormality; s L(i) =1; S202. Establish a loss function for training model parameters: Among them, λ CIoU , obj and λ angle are the manually set weight coefficients representing bounding box loss, confidence loss, and angle loss; L CIoU(i) is the loss of the abnormal bounding box in the i-th image with an abnormality, calculated as: Among them, IoU (i) is the intersection-over-union ratio of the abnormal detection box and the annotation box in the i-th image with abnormality; c i L is the diagonal distance between the minimum closure area of ​​the abnormal detection box and the annotation box in the i-th abnormal image; obj(i) is the confidence loss of the anomaly detection result in the i-th image with anomalies; L obj(i) =(p i ·IoU (i) -s L(i) ) 2 ; Among them, p i Indicates the probability value that the abnormal prediction box in the i-th image with abnormality contains the abnormality; is the model inference function f YOLO The intermediate variables generated in the calculation process of (X′,w); L angle(i) The angle loss of the anomaly detection box in the i-th image with anomalies is calculated as: L angle(i) =(θ (i) -θ L(i) ) 2 ; S203, using the gradient descent algorithm to calculate the loss function L(f YOLO (X i ',w),Label i ) is optimized and the model parameters when the loss function is minimized are calculated as the network model parameter set w.

5. A method for enhancing intelligent inspection of a power system, characterized in that: The method is implemented by using an augmented reality intelligent inspection device for a power system according to any one of claims 1 to 4, comprising: S301, through the data acquisition module, simulate the perspective of the inspection worker to collect real-time image and video data; S302, transmitting the collected image and video data to an edge computing module, executing an artificial intelligence image detection algorithm in the edge computing module to detect abnormal conditions of power transmission and transformation equipment in the image; S303, the edge computing module marks the position of the detected abnormal equipment, and transmits the picture containing the marked information to the augmented reality display module, so as to assist the inspection workers to quickly locate and judge the abnormal status of the equipment in the display environment; S304. When the auxiliary information of 303 differs from the inspection worker's judgment, the inspection worker calls the cloud assistance module through the inspection device to request the assistance of the cloud expert. The cloud expert makes judgments and analyses based on the video and image data transmitted by the device and provides corresponding maintenance guidance.

Images