Registration and accurate detection method and system for distribution network component image

Abstract

The application discloses a kind of registration and accurate detection method and system for distribution network component image, the method of the application includes obtaining distribution line pair visible light image and infrared image;According to infrared image, the visible light image is trimmed and registered, to obtain the visible light image after registration;Infrared image is interpolated to the same size of the visible light image after registration;The detection frame is obtained by sending the visible light image after registration into target detection network;Detection frame is drawn to the interpolated infrared image, and the temperature information of the detection target in the detection frame in infrared image is extracted to be used for the thermal fault discrimination of distribution network component.The application aims at making full and effective use of the beneficial information of two kinds of images of visible light image and infrared image, realizes the high-precision distribution network component detection, effectively reduces the occurrence of false detection and missed detection, provides a good foundation for subsequent temperature interpretation and thermal fault discrimination, effectively solves the problem of low precision of existing infrared detection method.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, specifically relating to a method and system for registration and accurate detection of images of power distribution network components. Background Technology

[0002] With the development of my country's power system, distribution network components have become an indispensable part of the power system. Distribution network components generate heat during operation, and when a component malfunctions, it causes temperature anomalies. Thermal mapping technology, by detecting the temperature distribution on the surface of a target object, can achieve rapid, non-contact detection of thermal faults in distribution network components, becoming an effective fault detection method. However, traditional thermal mapping technology suffers from problems such as narrow detection range, low detection accuracy, and slow detection efficiency, failing to meet the needs of practical applications.

[0003] In recent years, the rapid development of drone technology has provided new ideas for fault detection of power distribution network components. Drones, with their wide coverage, flexible flight, and portability, can effectively capture comprehensive, multi-angle images of power distribution network components, obtaining richer thermal image information. With the support of drone image processing technology, fault detection of power distribution network components can achieve more accurate and efficient results. However, due to the diverse sources and varying quality of images, some thermal fault information in drone-acquired images cannot be detected well, thus affecting the application effectiveness of drones. For example, identifying thermal fault information requires an infrared camera with thermal imaging capabilities to capture images of the target, and the captured infrared images contain temperature information. However, to accurately identify a thermal fault in a component, it is necessary to distinguish that component from other objects or the background in the image. Due to the low resolution and low contrast of infrared images, ordinary target detection methods often fail to meet the requirements. Therefore, a high-precision infrared image-based target detection method for typical spatial components in power distribution networks is needed to achieve real-time and accurate detection, thereby effectively improving the accuracy and efficiency of drone-based thermal fault detection of power distribution network components. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method and system for registration and accurate detection of distribution network component images, which addresses the above-mentioned problems in the prior art. This invention aims to make full and effective use of the beneficial information of both visible light and infrared images to achieve high-precision detection of distribution network components, effectively reduce the occurrence of false detections and missed detections, and provide a good foundation for subsequent temperature interpretation and thermal fault identification. It also effectively solves the problem of low accuracy in existing infrared detection methods.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A method for registration and accurate detection of images of power distribution network components includes:

[0007] S101, acquire visible light and infrared images of paired distribution network lines;

[0008] S102, the visible light image is cropped and registered according to the infrared image to obtain the registered visible light image;

[0009] S103, interpolate the infrared image to the same size as the registered visible light image;

[0010] S104, The registered visible light image is fed into the target detection network to obtain the detection box, target category and confidence score;

[0011] S105, draw the detection box onto the interpolated infrared image, and extract the temperature information of the detected target in the detection box in the infrared image for thermal fault identification of power distribution network components.

[0012] Optionally, step S102, which involves cropping and registering the visible light image based on the infrared image, includes: determining whether the visible light image is zoomed; if the visible light image is not zoomed, directly cropping a specified cropping area from the visible light image as the registered visible light image; otherwise, obtaining a zoomed cropping area based on the given cropping area (x1, y1, x2, y2) of the corresponding infrared image and the zoomed focal length f′ of the visible light image, where (x1, y1) is the upper left corner coordinate of the cropping area and (x2, y2) is the lower right corner coordinate of the cropping area, and then cropping the zoomed cropping area from the visible light image as the registered visible light image.

[0013] Optionally, the cropped area after zooming includes:

[0014] S201, determine the center coordinates (x) of the cutting area according to the following formula. c ,y c ):

[0015] x c =(x 1+ x2) / 2,y c =(y 1+ y2) / 2,

[0016] S202, the center coordinates (x) of the cropping area c ,y c Move in place to the center point of the image; so that the center coordinates (x, y) of the visible light image are... c ,y c The coordinates of the cropping region are given with the image center as the origin.

[0017] x c =x c-W / 2,y c =y c -H / 2,

[0018] In the above formula, W and H are the width and height of the unzoomed visible light image, respectively;

[0019] S203, obtain the center coordinates (x′) of the cropped area after zooming according to the following formula. c ,y c ′):

[0020] x′ c =x c *z,y′ c =y c *z,

[0021] In the above formula, z is the zoom factor, and z = f′ / f, where f′ is the focal length after zooming and y is the focal length before zooming;

[0022] S204, Move the origin of the coordinate system to the upper left corner of the image according to the following formula:

[0023] x′ c =x′ c +w′ / 2,y′ c =y′ c +H′ / 2,

[0024] In the above formula, W′ is the width of the zoomed visible light image, H′ is the height of the zoomed visible light image, and when the zoom method is optical zoom, the width W′ of the zoomed visible light image is the same as the width W of the unzoomed visible light image, and the height H′ of the zoomed visible light image is the same as the height H of the unzoomed visible light image.

[0025] S205, calculate the coordinates (x1′, y1′, x′2, y2′) of the cropped area after zooming according to the following formula, where (x1′, y1′) are the coordinates of the upper left corner of the cropped area after zooming, and (x′2, y2′) are the coordinates of the lower right corner of the cropped area after zooming:

[0026] x′1=x′ c -w′ / 2,y′1=y′ c -h′ / 2,x′2=x′ c +w′ / 2,y2′=y′ c -h′ / 2,

[0027] In the above formula, h′ is the height of the cropped area after zooming, w′ is the width of the cropped area after zooming, and w′=w*z,h′=h*z,where w is the width of the cropped area before zooming, and h is the height of the cropped area before zooming.

[0028] Optionally, determining whether a visible light image is zoomed includes: obtaining the zoomed focal length f′ from the metadata of the visible light image, combining it with the un-zoomed focal length f obtained in advance from the un-zoomed visible light image, calculating the zoom factor z according to z=f′ / f, and determining that the visible light image is zoomed if the zoom factor z is not equal to 1, otherwise determining that the visible light image is not zoomed.

[0029] Optionally, in step S103, the infrared image is interpolated using bilinear interpolation.

[0030] Optionally, the target detection network in step S104 is a YOLOv5 convolutional neural network.

[0031] Optionally, step S101, obtaining the visible light and infrared images of the distribution network lines in pairs, refers to obtaining the visible light and infrared images of the distribution network lines in pairs collected by the UAV.

[0032] Furthermore, the present invention also provides a drone, including a drone body with a visible light camera and an infrared camera, wherein the drone body is provided with a microprocessor and a memory interconnected thereto, the visible light camera and the infrared camera are respectively connected to the microprocessor, and the microprocessor is programmed or configured to execute the registration and accurate detection method for images of power distribution components.

[0033] Furthermore, the present invention also provides a registration and accurate detection system for images of distribution network components, including a computer device with a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the registration and accurate detection method for images of distribution network components.

[0034] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program that is programmed or configured by a microprocessor to execute the registration and accurate detection method for images of distribution network components.

[0035] Compared with existing technologies, the present invention has the following advantages: The present invention involves cropping and registering the visible light image based on the infrared image to obtain a registered visible light image; interpolating the infrared image to the same size as the registered visible light image; feeding the registered visible light image into a target detection network to obtain the detection box, target category, and confidence level; drawing the detection box onto the interpolated infrared image, and extracting the temperature information of the detected target in the detection box from the infrared image for thermal fault identification of power distribution network components. The present invention achieves high accuracy by utilizing the high-resolution visible light image for target detection. Furthermore, outputting the detection box to the infrared image yields the temperature information of the detected target within the detection box. This effectively utilizes the beneficial information from both visible light and infrared images, achieving high-precision detection of power distribution network components, effectively reducing false positives and false negatives, and providing a good foundation for subsequent temperature interpretation and thermal fault identification. It effectively solves the problem of low accuracy in existing infrared detection methods. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention. Detailed Implementation

[0037] like Figure 1 As shown, the registration and accurate detection method for distribution network component images in this embodiment includes:

[0038] S101, acquire visible light and infrared images of paired distribution network lines;

[0039] S102, the visible light image is cropped and registered according to the infrared image to obtain the registered visible light image;

[0040] S103, interpolate the infrared image to the same size as the registered visible light image;

[0041] S104, The registered visible light image is fed into the target detection network to obtain the detection box, target category and confidence score;

[0042] S105, draw the detection box onto the interpolated infrared image, and extract the temperature information of the detected target in the detection box in the infrared image for thermal fault identification of power distribution network components.

[0043] In this embodiment, when obtaining the visible light image and infrared image of the distribution network line pair in step S101, the resolution of the infrared image is 640*512 and the resolution of the visible light image is 4000*3000.

[0044] In this embodiment, step S102, which involves cropping and registering the visible light image based on the infrared image, includes: determining whether the visible light image is zoomed; if the visible light image is not zoomed, directly cropping a specified cropping area from the visible light image as the registered visible light image; otherwise, obtaining the zoomed cropping area based on the given cropping area (x1, y1, x2, y2) of the corresponding infrared image and the zoomed focal length f′ of the visible light image, where (x1, y1) is the upper left corner coordinate of the cropping area and (x2, y2) is the lower right corner coordinate of the cropping area, and then cropping the zoomed cropping area from the visible light image as the registered visible light image. Specifically, the cropped region (x1, y1, x2, y2) corresponding to the infrared image can be obtained in advance by manually registering a pair of infrared and visible light images, cropping the visible light image, and recording the cropping coordinates. Thus, the zoomed cropped region is obtained for the given cropped region (x1, y1, x2, y2) of the corresponding infrared image, so that the cropped visible light image has completed the registration purpose.

[0045] The transformation of the center coordinates and the width and height of the cropped area can be calculated by changing the focal length and using a mathematical model. For ease of code calculation, the top left corner of the image is set as the origin, (x1, y1) is the top left corner coordinate of the cropped area, (x2, y2) is the bottom right corner coordinate of the cropped area, and the coordinates of the cropped area before zooming are (x1, y1, x2, y2). The image size is H×W. In this embodiment, obtaining the cropped area after zooming includes:

[0046] S201, determine the center coordinates (x) of the cutting area according to the following formula. c ,y c ):

[0047] x c =(x 1+ x2) / 2,y c =(t 1+ y2) / 2,

[0048] S202, the center coordinates (x) of the cropping area c ,y c Move in place to the center point of the image; so that the center coordinates (x, y) of the visible light image are... c ,y c The coordinates of the cropping region are given with the image center as the origin.

[0049] x c =x c -W / 2,y c =y c -H / 2,

[0050] In the above formula, W and H are the width and height of the unzoomed visible light image, respectively;

[0051] S203, since optical zoom does not change the image resolution, the center coordinates (x′) of the cropped area after zooming can be obtained according to the following formula. c ,y c ′):

[0052] x′ c =x c *z,y′ c =y c *z,

[0053] In the above formula, z is the zoom factor, and z = f′ / f, where f′ is the focal length after zooming and f is the focal length before zooming;

[0054] S204, Move the origin of the coordinate system to the upper left corner of the image according to the following formula:

[0055] x′ c =x′ c +W′ / 2,y′ c =y′ c +H′ / 2,

[0056] In the above formula, W′ is the width of the zoomed visible light image, H′ is the height of the zoomed visible light image, and when the zoom method is optical zoom, the width W′ of the zoomed visible light image is the same as the width W of the unzoomed visible light image, and the height H′ of the zoomed visible light image is the same as the height H of the unzoomed visible light image.

[0057] S205, calculate the coordinates (x′1, y′1, x′2, y′2) of the cropped area after zooming according to the following formula, where (x′1, y′1) is the coordinate of the upper left corner of the cropped area after zooming, and (x′2, y′2) is the coordinate of the lower right corner of the cropped area after zooming:

[0058] x′1=x′ c -w′ / 2,y′1=y′ c -h′ / 2,x′2=x′ c +w′ / 2,y′2=y′ c -h′ / 2,

[0059] In the above formula, h′ is the height of the cropped area after zooming, w′ is the width of the cropped area after zooming, and w′=w*z,h′=h*z,where w is the width of the cropped area before zooming, and h is the height of the cropped area before zooming. Through this method, adaptive cropping and registration of visible light images can be achieved.

[0060] Because the visible light camera on a drone zooms during shooting, making cropping coordinates unavailable, the focal length of each captured visible light image is obtained by acquiring its metadata. The metadata of a visible light image stores information such as the camera model and focal length used to capture that image. Obtaining the focal length from the metadata of each image allows us to determine whether the image has been zoomed relative to the initial image. Obtaining the focal length of a visible light image from its metadata can be represented as:

[0061] f = exifread(I v ),

[0062] In the above formula, f is the focal length, exifread is a known function for reading EXIF ​​information of an image, and I v This is a visible light image.

[0063] n = z / f,

[0064] Where n is approximately a constant and z is the zoom factor, the ratio of focal length to zoom factor is constant for the same optical zoom camera. Therefore, the presence of zoom can be determined by obtaining the focal length. In this embodiment, determining whether a visible light image is zoomed includes: obtaining the zoomed focal length f′ from the metadata of the visible light image, combining it with the unzoomed focal length f obtained beforehand from the unzoomed visible light image, and calculating the zoom factor z according to z = f′ / f. If the zoom factor z is not equal to 1, it is determined that the visible light image is zoomed; otherwise, it is determined that the visible light image is not zoomed. The unzoomed focal length f can be obtained from the metadata of the unzoomed visible light image and used as a parameter.

[0065] In this embodiment, the initial size of the visible light image is 4000*3000, and the size of the infrared image is 640*512. After cropping and registration, the size of the registered visible light image becomes 2230*1750. To achieve subsequent detection box migration, the infrared image needs to be linearly interpolated to 2230*1750 to ensure consistent resolution, thus aligning the detection box coordinates to the same position. It should be noted that the infrared image interpolation in step S103 can be performed using any interpolation method as needed; for example, bilinear interpolation is used in this embodiment.

[0066] The target detection network in step S104 can be a convolutional neural network as needed. For example, as an optional implementation, the target detection network in step S104 of this embodiment is a YOLOv5 convolutional neural network. The registered visible light image, resized to 2230*1750, is fed into the YOLOv5 convolutional neural network to obtain the detection results (including detection boxes, target categories, and confidence scores). The detection boxes are then plotted onto the infrared image using coordinates. Detection is performed on visible light in this embodiment because visible light has high spatial resolution, achieving high accuracy. The detection boxes are output to the infrared image because the final thermal fault determination will be performed through temperature interpretation. If the infrared image is used directly for target detection, the detection accuracy will be very low, and many false positives and false negatives will occur.

[0067] The YOLOv5 convolutional neural network processes registered visible light images in the following ways:

[0068] 1. Image preprocessing: Preprocessing the input image, such as scaling, cropping, normalization, etc.

[0069] 2. Feature Extraction: Deep convolutional neural networks (CNNs) extract features at different levels of the image, which can be represented as:

[0070] F = CNN(x),

[0071] Where x is the input image, and F represents the extracted features at multiple levels, i.e.: F = {F1, F2, ..., F...} n}

[0072] 3. Feature Fusion: This involves fusing features from different levels to detect objects of different scales. It can be represented as:

[0073] m = merge(F1, F2, ..., F n ),

[0074] Where m represents the fusion feature and merge represents the fusion operation;

[0075] 4. Target Prediction: Target prediction is performed on the feature map, including target location, category, and confidence level. A detection result can be represented as:

[0076] p c ,p c1 ,p c2 ,...,p cn ,b x ,b y ,b w ,b h =predict(m)

[0077] Where, p cIt is the confidence level (the probability that the target exists), p c1 ,p c2 ,...,p cn These are the probabilities of n distinct categories, (b x ,b y ,b w ,b h ) refers to the position and size of the detection frame, (b x ,b y (b) represents the coordinates of the center point. w ,b h ) represents the width and height, and predict represents the target prediction;

[0078] 5. Non-maximum suppression (NMS): Filters the prediction results, removing overlapping boxes and boxes with low confidence.

[0079] boxes = apply_nms(p c ,p c1 ,p c2 ,...,p cn ,b x ,b y ,b w ,b h )

[0080] Here, boxes is the list of filtered target boxes, and apply_nms represents nonmaximum suppression.

[0081] In the YOLOv5 convolutional neural network, the predicted detection box is represented by three parameters: center point coordinates, width, and height. It also needs to predict the probability of the target class. Specifically, for each cell, the model predicts three anchor boxes of different sizes. For each anchor box, the model outputs an object presence score and a class probability distribution vector. Based on the threshold of the object presence score, it can determine which anchor boxes contain objects and convert them into detection boxes.

[0082] The center coordinates, width, and height of the detection frame can be calculated using the following formula:

[0083] b x = (sigmoid(t_x) + c_x) * stride

[0084] b y = (sigmoid(t_y)+c_y)*stride

[0085] b w =p_w*e t _w*anchor_w

[0086] bh =p_h*e t _h*anchor_h

[0087] In the above formula, t_x, t_y, t_w, and t_h are the four parameters predicted by the model, c_x and c_y are the center coordinates of the current cell on the feature map (with the top left corner as the origin), stride is the downsampling factor of the feature map relative to the original image, p_w and p_h are the normalized width and height, and anchor_w and anchor_h are the width and height of the current anchor box.

[0088] The probabilities p of n distinct classes c1 ,p c2 ,...,p cn It can be calculated using the following formula:

[0089] Pr(class_i|object)=sigmoid(t_i),

[0090] Here, t_i is the probability of the i-th class predicted by the model. Sigmoid represents the sigmoid function, and Pr(class_i|object) represents the probability that the target object belongs to class_i. Since each anchor box is only responsible for predicting the probability of one class, the class probabilities of the three anchor boxes need to be merged. Typically, the class with the highest probability value is selected as the final predicted class. Finally, the three anchor boxes and their predicted scores and class probabilities are merged to obtain the detection box and predicted class for each object. Detection boxes of the same position and size are drawn on the infrared image using the center coordinates, width, and height of the detection boxes, and different class information is distinguished by the color of the detection boxes. It should be noted that the YOLOv5 convolutional neural network is an existing convolutional neural network. This embodiment only involves the application of the YOLOv5 convolutional neural network and does not involve any improvement to the YOLOv5 convolutional neural network. In this embodiment, although the infrared image is not directly detected, a class prediction map of the infrared image is obtained, which has the same effect as direct detection of the infrared image but greatly improves the detection accuracy and effectively reduces the occurrence of false positives and false negatives.

[0091] As an optional implementation, step S101 of this embodiment, acquiring paired visible light and infrared images of distribution network lines, refers to acquiring paired visible light and infrared images of distribution network lines collected by a drone. Undoubtedly, step S101 of this embodiment, acquiring paired visible light and infrared images of distribution network lines, does not depend on any specific acquisition method.

[0092] In summary, this embodiment's registration and accurate detection method for power distribution network component images performs cross-modal registration of visible light and infrared images using an adaptive cropping registration method. The registered visible light image is then fed into a detection network. By obtaining the center coordinates, width, height, and category information of the detection box predicted by the detection network and plotting it on the infrared image, it achieves target detection using high-resolution visible light images and temperature interpretation using infrared images containing temperature information. This efficiently utilizes the beneficial information from both types of images. It enables real-time and accurate plotting of the detection box on the infrared image, achieving high-quality temperature interpretation, and thus accurately identifying thermal faults in typical power distribution network components.

[0093] In addition, this embodiment also provides a drone, including a drone body with a visible light camera and an infrared camera. The drone body is provided with a microprocessor and a memory that are interconnected. The visible light camera and the infrared camera are respectively connected to the microprocessor. The microprocessor is programmed or configured to execute the registration and accurate detection method for distribution network component images described above, thereby enabling real-time inspection and detection of the distribution network.

[0094] Furthermore, this embodiment also provides a registration and accurate detection system for images of distribution network components, including a computer device with interconnected microprocessors and memory. The microprocessor is programmed or configured to execute the aforementioned registration and accurate detection method for images of distribution network components, thereby providing an image processing solution independent of image acquisition devices. The computer device can be deployed offline, networked, or even in the cloud as needed.

[0095] In addition, this embodiment also provides a computer-readable storage medium storing a computer program, which is used to be programmed or configured by a microprocessor to execute the registration and accurate detection method for distribution network component images described above.

[0096] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for registration and accurate detection of images of power distribution network components, characterized in that, include: S101, acquire visible light and infrared images of paired distribution network lines; S102, the visible light image is cropped and registered according to the infrared image to obtain the registered visible light image; S103, interpolate the infrared image to the same size as the registered visible light image; S104, The registered visible light image is fed into the target detection network to obtain the detection box, target category and confidence score; S105, draw the detection box onto the interpolated infrared image, and extract the temperature information of the detected target in the detection box in the infrared image for thermal fault identification of distribution network components; Step S102, which involves cropping and registering the visible light image based on the infrared image, includes: determining whether the visible light image is zoomed; if the visible light image is not zoomed, directly cropping a specified cropping region from the visible light image as the registered visible light image; otherwise, using the given cropping region (x1, y1, x2, y2) of the corresponding infrared image and the zoomed focal length of the visible light image... Obtain the zoomed cropped area, where (x1, y1) is the upper left corner coordinate of the cropped area and (x2, y2) is the lower right corner coordinate of the cropped area. Then, crop the zoomed cropped area from the visible light image to obtain the registered visible light image. Determining whether a visible light image has been zoomed includes: obtaining the zoomed focal length from the metadata of the visible light image. Combined with the unfocused focal length obtained from the previously unfocused visible light image. ,according to Calculate the zoom ratio z If zoom magnification z If the value is not equal to 1, the visible light image is determined to be zoomed; otherwise, the visible light image is determined not to be zoomed. The cropped area obtained after zooming includes: S201, determine the center coordinates (x) of the cutting area according to the following formula. c ,y c ): ， S202, the center coordinates (x) of the cropping area c ,y c Move in place to the center point of the image; so that the center coordinates (x, y) of the visible light image are... c ,y c The coordinates of the cropping region are given with the image center as the origin. ， In the above formula, W and H These represent the width and height of the unzoomed visible light image, respectively. S203, obtain the center coordinates of the cropped area after zooming according to the following formula ( , ): ， In the above formula, z is the zoom factor, and we have ,in The focal length after zooming. This is the focal length without zooming; S204, Move the origin of the coordinate system to the upper left corner of the image according to the following formula: ， In the above formula, The width of the zoomed visible light image. The height of the zoomed visible light image, and the width of the zoomed visible light image when the zoom method is optical zoom. The zoomed visible light image has the same width W as the unzoomed visible light image, but the zoomed visible light image has the same height. The height H is the same as that of the unfocused visible light image; S205, calculate the coordinates of the cropped area after zooming according to the following formula ( , , , ),in( , ) represents the coordinates of the top-left corner of the cropped area after zooming. , The coordinates of the bottom right corner of the cropped area after zooming are: ， ， In the above formula, This represents the height of the cropped area after zooming. The width of the cropped area after zooming, and has , h' = h * z ,in w The width of the cropping area before zooming. h The height of the cropped area before zooming.

2. The registration and accurate detection method for images of power distribution network components according to claim 1, characterized in that, In step S103, the infrared image is interpolated using bilinear interpolation.

3. The registration and accurate detection method for images of power distribution network components according to claim 1, characterized in that, The target detection network in step S104 is a YOLOv5 convolutional neural network.

4. The registration and accurate detection method for images of power distribution network components according to claim 1, characterized in that, Step S101, acquiring paired visible light and infrared images of distribution network lines, refers to acquiring paired visible light and infrared images of distribution network lines collected by the UAV.

5. A drone, comprising a drone body equipped with a visible light camera and an infrared camera, wherein the drone body has a microprocessor and a memory interconnected therewith, and the visible light camera and the infrared camera are respectively connected to the microprocessor, characterized in that, The microprocessor is programmed or configured to execute the registration and accurate detection method for distribution network component images as described in any one of claims 1 to 4.

6. A registration and accurate detection system for images of power distribution network components, comprising a computer device with interconnected microprocessors and memory, characterized in that, The microprocessor is programmed or configured to execute the registration and accurate detection method for distribution network component images as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, The computer program is used to be programmed or configured by a microprocessor to execute the registration and accurate detection method for images of power distribution network components as described in any one of claims 1 to 4.

Images