A method for locating the frame of power equipment components based on infrared images

Abstract

The present invention discloses a method for positioning the frames of power equipment components based on infrared images, comprising the following steps: S1. Use an infrared thermal imager to collect information on power transmission, transformation, and distribution equipment, obtain infrared images, and then extract temperature data from the infrared images to form a temperature visual matrix; S2. Perform type and position marking and normalization processing on the temperature visual matrix of the power equipment components to generate model data; S3. Build and train a power equipment component frame detection model; S4. Input the model data of known power equipment types into the power equipment component frame detection model to output component frames; S5. Screen and fill component frames: According to the number and relative distribution positions of actual power equipment components, screen and fill misdetected component frames and undetected component frames. The present invention has the characteristics of improving the detection rate and detection accuracy of power equipment components.

Description

Technical Field

[0001] The present invention relates to a method for positioning infrared images, in particular to a method for positioning component frames of power equipment based on infrared images. Background Art

[0002] Currently, in the work of power equipment fault diagnosis and detection, infrared images are mainly used as the research object, and it is necessary to judge whether there is abnormal heating in the power equipment by the temperature difference between various components of the power equipment. Due to the large variety of power equipment and the similarity between multiple devices, and the shooting environment is relatively complex, the captured infrared images or the temperature matrices extracted from the infrared images will contain various power equipment, or multiple power equipment of the same type and other interference factors. It is found that when using deep learning technology to establish a model to detect the component frames of a single device, there will be over-detection (detecting multiple component frames that do not belong to the same / group of power equipment), mis-detection (detecting other components with similar shapes to the components to be detected), and missed-detection. The detection accuracy is low, and the components in the power equipment cannot be accurately positioned, resulting in inaccurate diagnosis of abnormal heating of the power equipment and misjudgment. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for positioning component frames of power equipment based on infrared images. The present invention has the characteristics of improving the detection rate and detection accuracy of power equipment components.

[0004] The technical solution of the present invention: A method for positioning component frames of power equipment based on infrared images, comprising the following steps:

[0005] S1. Data acquisition: Use an infrared thermal imager to collect power transmission, transformation, and distribution equipment information, obtain infrared images, and then extract temperature data from the infrared images to form a temperature visual matrix;

[0006] S2. Data preprocessing: Perform type and position marking and normalization processing on the temperature visual matrix for power equipment components to generate model data;

[0007] S3. Build and train a power equipment component frame detection model;

[0008] S4. Equipment component frame detection inference: Input the model data of known power equipment types into the power equipment component frame detection model, output the component frames, and obtain the type, position, and confidence of the power equipment components;

[0009] S5. Screening and filling component frames:

[0010] S501. Screening component frames: According to the equipment frames of the power equipment, screen the mis-detected component frames, and screen out the component frames belonging to the same / group of power equipment;

[0011] S502. Fill component bounding boxes: Fill in the undetected component bounding boxes based on the quantity and relative distribution positions of the actual power equipment components.

[0012] In the aforementioned method for locating component bounding boxes of power equipment based on infrared images, the size of the temperature visual matrix in step S1 is 640×480×1. Among them, the value of each element in the matrix corresponds to the actual temperature value of the corresponding area in the environment.

[0013] In the aforementioned method for locating component bounding boxes of power equipment based on infrared images, the normalization process in step S2 is as follows: Scale the temperature visual matrix data to the range of 0 - 1, and then fill in zeros above and below the shorter side to generate model data of 640×640×1.

[0014] In the aforementioned method for locating component bounding boxes of power equipment based on infrared images, the method for establishing the power equipment component bounding box detection model in step S3 includes the following steps:

[0015] S301. Model construction: Use the YOLOv4 framework, and use MobileNet - V1 as the backbone feature extraction network;

[0016] S302. Model training: Input the model data into the model for model training and parameter setting to obtain the power equipment component bounding box detection model.

[0017] In the aforementioned method for locating component bounding boxes of power equipment based on infrared images, step S501 specifically includes the following steps:

[0018] a. Determine the actual number of component bounding boxes of the power equipment as N according to the type of the power equipment;

[0019] b. Screen out n component bounding boxes whose centers fall within the equipment bounding box, and delete the remaining component bounding boxes;

[0020] c. If n≥N, then sort the n component bounding boxes in descending order of confidence, and select the first N as the final output;

[0021] d. If n<N, then fill the n component bounding boxes to N component bounding boxes by filling in component bounding boxes.

[0022] In the aforementioned method for locating component bounding boxes of power equipment based on infrared images, if the component bounding boxes in the power equipment are N component bounding boxes evenly linearly distributed, then the filling of component bounding boxes in step S502 includes the following steps:

[0023] a. Obtain the coordinates of the device frame: (left, top, right, bottom); where (left, top) are the coordinates of the upper left corner of the device frame, and (right, bottom) are the coordinates of the lower right corner of the device frame;

[0024] b. If n = 0, evenly generate N component frames in the device frame from left to right. The interval ww between component frames is ww = (right - left) / / 20, and the vertical interval hh between the component frame and the device frame is hh = (top - bottom) / / 10;

[0025] c. If n = 1, the coordinates of the component frame are ( l , t , r , b ). Calculate and determine the component frames to be filled on the left and right of this component frame in the device frame respectively according to the distance between the left side of the component frame and the left side of the device frame and the distance between the right side of the component frame and the right side of the device frame. Among them, ( l , t ) are the coordinates of the upper left corner of the component frame, and ( r , b ) are the coordinates of the lower right corner of the component frame;

[0026] d. If 2 ≤ n < N, use the position of the leftmost component frame and the device frame to judge the number and position of the component frames to be filled on the left; use the position of the rightmost component frame and the device frame to judge the number and position of the frames to be filled on the right; use the distance between two adjacent middle frames to judge the number and position of the frames to be filled in the middle; the width w of the filled component frame is the average value of the widths of n component frames, and the height h of the filled component frame is the average value of the heights of n component frames.

[0027] In the foregoing method for positioning component frames of a power device based on an infrared image, in step c, the calculation method for the component frames to be filled on the left side of the component frame is as follows:

[0028] c - 1. If l - left > (N - 1) × ( r - l ), then fill N - 1 frames on the left side of the component frame. The interval ww between component frames is ww = ( l - left - (N - 1) × ( r - l )) / / N, the width w = r-l of the filled component frame, and the height h = t-b ;

[0029] c - 2. If (N - k) × ( r - l ) <l -left < (N - k + 1) × ( r - l ), then fill N - k frames on the left side of the component frame, and the interval between component frames ww = ( l -left - (N - k) × ( r - l )) / / (N - k + 1), where N > k > 1 and k is an integer, and the width of the filled component frame w = r-l , and the height of the component frame h = t-b ;

[0030] c - 3. If l -left < r -l, then fill 0 frames on the left side of the component frame.

[0031] In the above - mentioned method for positioning component frames of power equipment based on infrared images, in step c, the calculation method for the component frames to be filled on the right side of the component frame is as follows:

[0032] c - 1. If right - r > (N - 1) × ( r - l ), then fill N - 1 frames on the right side of the component frame, and the interval between component frames ww = (right - r -(N - 1) × ( r - l )) / / N, the width of the filled component frame w = r - l , and the height of the component frame h = t-b ;

[0033] c - 2. If (N - k) × ( r - l ) < right - r < (N - k + 1) × ( r - l ), then fill N - k frames on the right side of the component frame, and the interval between component frames ww = (right - r -(N - k) × (r - l) / / (N - k + 1), where N > k > 1 and k is an integer, and the width of the filled component frame w = r - l , and the height of the component frame h = t-b ;

[0034] c - 3. If right - r < r - l , then fill 0 frames on the right side of the component frame.

[0035] In the above-mentioned method for locating component frames of power equipment based on infrared images, step d is specifically as follows: If 2 ≤ n < N, according to step c, fill the component frames on both sides respectively, filling the left component frame of the leftmost component frame and the right component frame of the rightmost component frame; according to k = (horizontal distance between adjacent two component frames) / / w, fill the middle component frames with k as the number of middle component frames to be filled; the width w of the filled component frame is taken as the average value of the widths of n component frames, and the height h of the filled component frame is taken as the average value of the heights of n component frames.

[0036] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0037] Through model establishment and training, the power equipment component frame detection model obtained by the present invention can detect each component of different power equipment, and screen and fill the misdetected and undetected components, thereby improving the detection rate and detection accuracy of power equipment components, making the detection precision reach 94.20% and the detection recall rate reach 94.83%. Description of the Drawings

[0038] Figure 1 is the training loss curve graph of the component frame detection model in the present invention;

[0039] Figure 2 is the detection graph of the component frame detection model of the distribution network - pile head power equipment;

[0040] Figure 3 is the component frame screening graph of the distribution network - pile head power equipment;

[0041] Figure 4 is the component frame filling graph of the distribution network - pile head power equipment. Detailed Embodiments

[0042] The following further illustrates the present invention in conjunction with embodiments, but it is not used as a basis for limiting the present invention.

[0043] Embodiment 1:

[0044] A method for locating component frames of power equipment based on infrared images includes the following steps:

[0045] S1. Data acquisition: Use an infrared thermal imager to collect information on power transmission and distribution equipment, including information on insulators, pile heads, etc., obtain an infrared general data file storage format image, and then extract temperature data from the infrared image to form a temperature visual matrix; the size of the temperature visual matrix is 640×480×1. Among them, the value of each element in the temperature visual matrix corresponds to the actual temperature value of the corresponding area in the environment.

[0046] S2. Data preprocessing: Mark the type and position of the components of the power equipment on the temperature vision matrix and perform normalization processing, that is, scale the temperature vision matrix data to the range of 0-1, and then fill zeros at the top and bottom of the shorter side to generate model data of 640×640×1; randomly divide 10,000 pieces of model data into a training set and a test set according to a ratio of 9:1.

[0047] S3. Build and train a power equipment component frame detection model;

[0048] S301. Model building: Use the YOLOv4 framework, and use MobileNet-V1 for the backbone feature extraction network;

[0049] S302. Model training: Input the training set and the test set into the model for model training and parameter setting. The set parameters are shown in Table 1 to obtain a power equipment component frame detection model.

[0050] Table 1 Model network parameter settings

[0051] Parameter Name Parameter Value MobileNet-v1(alpha) 1 Initial Learning Rate 1e-2 Batch Size 16 Epoch 600 Optimizer SGD momentum 0.937 Max_boxes 5 nms_iou 0.01

[0052] The training loss curve is as Figure 1 shown. The initial training loss drops rapidly as the number of epochs increases. The subsequent training loss drops more slowly but still steadily decreases, indicating that the parameter settings are reasonable.

[0053] S4. Inference of equipment component frame detection: Input the model data of known power equipment types into the power equipment component frame detection model, output the component frame, and obtain the category, position, and confidence of the power equipment components.

[0054] S5. Screening and filling component frames: According to the actual number and relative distribution position of power equipment components, screen and fill the misdetected component frames and the undetected component frames.

[0055] S501. Screening component frames: According to the equipment frame of the power equipment, screen the misdetected component frames, and screen out the component frames that belong to the same / group of power equipment to reduce the misdetection rate; the equipment frame is obtained through the equipment frame detection model of the power equipment.

[0056] The specific steps for screening component frames are as follows:

[0057] a. According to the type of power equipment, determine the actual number of component frames of the power equipment as N;

[0058] b. Screen out n component frames whose centers fall within the equipment frame, and delete the remaining component frames;

[0059] c. If n ≥ N, sort the n component boxes in descending order of the predicted confidence level, and select the first N as the final output;

[0060] d. If n < N, fill in the n component boxes to N component boxes by filling in the missing component boxes.

[0061] S502. Filling in the missing component boxes: Based on the number and relative distribution positions of the actual power equipment components, fill in the missing component boxes to reduce the missed detection rate.

[0062] If the component boxes in the power equipment are N component boxes evenly linearly distributed, the method for filling in the missing component boxes includes the following steps:

[0063] a. Obtain the device box coordinates: (left, top, right, bottom); where (left, top) are the upper left coordinates of the device box, and (right, bottom) are the lower right coordinates of the device box;

[0064] b. If n = 0, in the device box, generate N component boxes evenly from left to right. The interval ww between the component boxes = (right - left) / / 20, and the vertical interval hh between the component box and the device box = (top - bottom) / / 10;

[0065] c. If n = 1, the coordinates of this component box are ( l , t , r , b ), where (l, t) are the upper left coordinates of the component box, and (r, b) are the lower right coordinates of the component box; respectively calculate and determine the missing component boxes on the left and right of this component box in the device box according to the distance between the left side of the component box and the left side of the device box and the distance between the right side of the component box and the right side of the device box.

[0066] For the component box on the left of the component box, fill it in according to the distance between the left side of the component box and the left side of the device box. The calculation method for the missing component box on the left of the component box is:

[0067] c - 1. If l - left > (N - 1) × ( r - l ), then fill in N - 1 boxes on the left of the component box. The interval ww between the component boxes = ( l - left - (N - 1) × ( r - l )) / / N, the width w = r-l of the filled component box, and the height h = t-b ;

[0068] c-2. If (N - k) × ( r - l ) < l -left < (N - k + 1) × ( r - l ), then fill N - k frames on the left side of the component frame, and the interval between component frames ww = ( l -left - (N - k) × ( r - l ) / / (N - k + 1), where N > k > 1 and k is an integer, the width of the filled component frame w = r-l , and the height of the component frame h = t-b ;

[0069] c-3. If l -left < r -l, then fill 0 frames on the left side of the component frame.

[0070] For the component frame on the right side of the component frame, fill it according to the distance between the right side of the component frame and the right side of the device frame. The calculation method for the component frame to be filled on the right side of the component frame is as follows:

[0071] c-1. If right - r > (N - 1) × ( r - l ), then fill N - 1 frames on the right side of the component frame, and the interval between component frames ww = (right - r - (N - 1) × ( r - l )) / / N, the width of the filled component frame w = r - l , and the height of the component frame h = t-b ;

[0072] c-2. If (N - k) × ( r - l ) < right - r < (N - k + 1) × ( r - l ), then fill N - k frames on the right side of the component frame, and the interval between component frames ww = (right - r - (N - k) × (r - l) / / (N - k + 1), where N > k > 1 and k is an integer, the width of the filled component frame w = r - l , and the height of the component frame h = t-b ;

[0073] c-3. If right - r< r -[[]] l If so, fill 0 frames on the right side of the component frame.

[0074] d. If 2 ≤ n < N, use the position of the leftmost component frame and the device frame to determine the number and position of the component frames to be filled on the left; use the position of the rightmost component frame and the device frame to determine the number and position of the frames to be filled on the right; use the distance between two adjacent frames in the middle to determine the number and position of the frames to be filled in the middle. The width w of the filled component frame is the average value of the widths of n component frames, and the height h of the filled component frame is the average value of the heights of n component frames.

[0075] Specifically:

[0076] 1) Filling of component frames on both sides: According to step c, fill the left component frames of the leftmost component frame and the right component frames of the rightmost component frame respectively.

[0077] 2) Filling of component frames in the middle: k = (horizontal distance between two adjacent component frames) / / w, where k is the number of component frames filled in the middle.

[0078] S6. Evaluate the regression model: After the above processing, use Precision and Recall to evaluate the prediction results.

[0079] The evaluation formulas are: Precision P = TP / (TP + FP), Recall R = TP / (TP + FN); where TP represents the number of correctly recognized component frames, FP represents the number of wrongly recognized component frames, and FN represents the number of wrongly recognized non-component frames.

[0080] Precision = 94.20%; Recall = 94.83%

[0081] Without processing the component frames, its Precision = 72.51% and Recall = 93.31%.

[0082] Therefore, this method greatly improves the Precision and Recall of the model.

[0083] Example 2:

[0084] A method for positioning component frames of power equipment based on infrared images. Taking the distribution network - pile head equipment as an example, there are four pile heads in a group of pile heads of the distribution network, that is, there are four component frames in one pile head equipment frame and they are linearly distributed, including the following steps:

[0085] ​S1. Data acquisition: Use an infrared thermal imager to collect information on the distribution network pile heads, obtain infrared images in the general data file storage format, and then extract temperature data from the infrared images to form a temperature vision matrix; the size of the temperature vision matrix is 640×480×1. Among them, the value of each element in the temperature vision matrix corresponds to the actual temperature value of the corresponding area in the environment.

[0086] S2. Data preprocessing: Mark the type and position of the power equipment components on the temperature vision matrix and perform normalization processing, that is, scale the temperature vision matrix data to the 0-1 interval, and then fill zeros at the top and bottom of the short side to generate model data of 640×640×1; randomly divide 10,000 pieces of model data into a training set and a test set according to a ratio of 9:1.

[0087] S3. Establish a power equipment component frame detection model;

[0088] S301. Model construction: Use the YOLOv4 framework, and the backbone feature extraction network uses MobileNet-V1;

[0089] S302. Model training: Input the training set and the test set into the model for model training and parameter setting to obtain a power equipment component frame detection model.

[0090] S4. Device component frame detection inference: Input the model data of the distribution network - pile head power equipment into the power equipment component frame detection model, and the model output results are as Figure 2 shown, to obtain the component category, confidence level and device frame position of the power equipment.

[0091] The detection results of the component frames are marked with light - colored rectangular frames, while the dark - colored rectangular frames are the original manual marking results. For Figure 2 the pile head detection results, it can be seen that a total of 4 component frames (the four light - colored frames) are detected, among which 3 are correct (the 3rd - 5th from left to right, the light - colored frame coincides with the dark - colored frame), 1 is misdetected (the first one on the left, the single light - colored frame), and 1 is undetected (the second single dark - colored frame from left to right).

[0092] S5. Screen and fill the component frames: According to the actual number and relative distribution position of the power equipment components, screen and fill the misdetected component frames and undetected component frames.

[0093] S501. Screen the component frames: According to the device frame of the distribution network - pile head, screen the misdetected component frames, and screen out the component frames belonging to the distribution network - pile head power equipment to reduce the misdetection rate; among them, the device frame of the distribution network - pile head is obtained through the device frame detection model of the power equipment.

[0094] The specific steps for screening the component frames are as follows:

[0095] a. Determine that the power equipment is a distribution network - pile head device, and the actual number of component frames is N = 4;

[0096] b. Screen out n = 3 component frames whose centers fall within the equipment frame, and delete the remaining component frames; as Figure 3 shown, the largest large square frame is the distribution network - pile head equipment frame, and the 3 small square frames within the large square frame are the screened - out component frames.

[0097] c. Since n < N, fill the 3 component frames to 4 component frames by the method of filling component frames.

[0098] S502. Fill the component frames of the distribution network - pile head equipment: According to the actual distribution of the distribution network - pile head component frames, the number of components is 4, and the relative distribution positions are linearly distributed. According to this rule, fill the missed - detected component frames to reduce the missed - detection rate.

[0099] Specifically, it includes the following steps:

[0100] a. Obtain the coordinates of the equipment frame: (left, top, right, bottom); where (left, top) is the upper - left corner coordinates of the equipment frame, and (right, bottom) is the lower - right corner coordinates of the equipment frame;

[0101] b. Since n = 3 and N = 4, only one frame needs to be filled. The coordinates of the known three component frames from left to right are respectively ( l 1, t 1, r 1, b 1), ( l 2, t 2, r 2, b 2); ( l 3, t 3, r 3, b 3); where ( l , t ) is the upper - left corner coordinates of the component frame, and ( r , b ) is the lower - right corner coordinates of the component frame; calculate and determine the component frame to be filled in the equipment frame respectively.

[0102] The width w of the filled component frame takes the average value of the widths of the three component frames, and the height h takes the average value of the heights of the three component frames.

[0103] Since l 1 - left > w: then at the coordinate of ( l 1, t 1, r 1, bFill a box on the left side of the component box in (1).

[0104] Since right- r 3 < w, so there is no need to fill on the right side of the component box with coordinates ( l 3, t 3, r 3, b 3).

[0105] And the distance between adjacent component boxes is less than w, so there is no need to fill in the middle.

[0106] The filling result is as Figure 4 shown. The large box is the distribution network pile head equipment box. The three small boxes on the right side in the large box are the screened component boxes, and the leftmost one in the large box is the filled component box.

[0107] S6. Evaluate the regression model: After the above processing, use Precision and Recall to evaluate the prediction results.

[0108] The evaluation formulas are: Precision P = TP / (TP + FP), Recall R = TP / (TP + FN); where, TP represents the number of correctly identified component boxes, FP represents the number of mis-identified component boxes, and FN represents the number of mis-identified non-component boxes.

[0109] After evaluation, for the distribution network - pile head, the Precision of the initial model is 72.51%, and the Recall is 93.13%.

[0110] After being processed by the component box screening and filling method, Precision = 94.20%; Recall = 94.83%

[0111] For the substation - disconnector, the Precision of the initial model is 96.24%, and the Precision is 65.18%.

[0112] After being processed by the component box screening and filling method, Precision = 94.65%, and Precision = 93.17%.

[0113] Therefore, this method improves the Precision and Recall of the detection of power equipment component boxes.

Claims

1. A method for positioning the frame of a power equipment component based on an infrared image, characterized in that: It includes the following steps: S1. Data acquisition: Use an infrared thermal imager to collect information on power transmission, transformation, and distribution equipment, obtain infrared images, and then extract temperature data from the infrared images to form a temperature vision matrix; S2. Data preprocessing: Perform type and position marking and normalization processing on the temperature vision matrix for power equipment components to generate model data; S3. Build and train a power equipment component box detection model; S4. Equipment component box detection inference: Input the model data of known power equipment types into the power equipment component box detection model, output the component boxes, and obtain the type, position, and confidence of the power equipment components; S5. Screen and fill the component boxes: S501. Screen the component boxes: According to the equipment box of the power equipment, screen the misdetected component boxes and screen out the component boxes belonging to the same / group of power equipment; S502. Fill the component boxes: Based on the actual number and relative distribution positions of power equipment components, fill the undetected component boxes.

2. The method for positioning the frame of a power equipment component based on an infrared image according to claim 1, wherein: The size of the temperature vision matrix in step S1 is 640×480×1, where the value of each element in the matrix corresponds to the actual temperature value of the corresponding area in the environment.

3. A method for positioning the frame of a power equipment component based on an infrared image according to claim 1, characterized in that: The normalization processing in step S2 is as follows: Scale the temperature vision matrix data to the 0-1 interval, and then fill zeros above and below the short side to generate model data of 640×640×1.

4. A method for positioning a frame of a power equipment component based on an infrared image according to claim 1, characterized in that: The method for establishing the power equipment component box detection model in step S3 includes the following steps: S301. Model building: Use the YOLOv4 framework, and use MobileNet-V1 for the backbone feature extraction network; S302. Model training: Input the model data into the model for model training and parameter setting to obtain the power equipment component box detection model.

5. A method for positioning the frame of a power equipment component based on an infrared image according to claim 1, characterized in that: Step S501 specifically includes the following steps: a. Determine the actual number of component boxes of the power equipment as N according to the type of the power equipment; b. Screen out n component boxes whose centers fall within the equipment box, and delete the remaining component boxes; c. If n≥N, sort the n component boxes in descending order of confidence and select the first N as the final output; d. If n<N, fill the n component boxes to N component boxes by filling the component boxes.

6. A method for positioning the frame of a power equipment component based on an infrared image according to claim 1, characterized in that: If the component boxes in the power equipment are N component boxes evenly linearly distributed, then filling the component boxes in step S502 includes the following steps: a. Obtain the equipment box coordinates: (left, top, right, bottom); where, (left, top) is the upper left coordinate of the equipment box, and (right, bottom) is the lower right coordinate of the equipment box; b. If n = 0, evenly generate N component boxes in the equipment box from left to right. The interval ww between the component boxes is (right - left) / / 20, and the vertical interval hh between the component boxes and the equipment box is (top - bottom) / / 10; c. If n = 1, the coordinates of the component frame are ( l , t , r , b ). Calculate and determine the component frames to be filled on the left and right of the component frame in the device frame respectively according to the distance between the left side of the component frame and the left side of the device frame and the distance between the right side of the component frame and the right side of the device frame, where ( l , t ) are the upper left coordinates of the component frame, and ( r , b ) are the lower right coordinates of the component frame; d. If 2 ≤ n < N, use the position of the leftmost component frame and the device frame to determine the number and positions of the component frames to be filled on the left; use the position of the rightmost component frame and the device frame to determine the number and positions of the frames to be filled on the right; use the distance between two adjacent frames in the middle to determine the number and positions of the frames to be filled in the middle; the width w of the filled component frame is taken as the average value of the widths of n component frames, and the height h of the filled component frame is taken as the average value of the heights of n component frames.

7. A method for positioning the frame of a power equipment component based on an infrared image according to claim 6, characterized in that: In step c, the calculation method for the component frames to be filled on the left side of the component frame is: c-1. If l -left>(N - 1)×( r - l ), then fill N - 1 frames on the left side of the component frame, and the interval ww between component frames is ww = ( l -left - (N - 1)×( r - l )) / / N, the width of the filled component frame w = r-l , and the height of the component frame h = t - b ; c-2. If (N - k) × ( r - l ) < l -left < (N - k + 1) × ( r - l ), then fill N - k frames on the left side of the component frame, and the interval ww between component frames is ww = ( l -left - (N - k) × ( r - l ) / / (N - k + 1), where N > k > 1 and k is an integer, the width of the filled component frame w = r-l , and the height of the component frame h = t - b ; c-3. If l -left< r -l, then 0 frames are filled on the left side of the component frame.

8. A method for positioning the frame of a power equipment component based on an infrared image according to claim 6, characterized in that: In step c, the calculation method for the component frames to be filled on the right side of the component frame is: c-1. If right - r >(N - 1)×( r - l ), then fill N - 1 frames on the right side of the component frame. The interval between component frames is ww = (right - r -(N - 1)×( r - l )) / / N. The width of the filled component frame w = r - l , and the height of the component frame h = t - b ; c-2. If (N-k)×( r - l ) < right - r < (N-k+1)×( r - l ), then fill N-k frames on the right side of the component frame, and the interval between component frames ww = (right - r -(N-k)×(r-l) / / (N-k+1), where N>k>1 and k is an integer, the width w of the filled component frame = r - l , and the height of the component frame h = t - b ; c-3. If right- r < r - l , then 0 frames are filled on the right side of the component frame.

9. The method for positioning the frame of a power equipment component based on an infrared image according to claim 6, wherein: Specifically, in step d, if 2 ≤ n < N, according to step c, fill the component frames on both sides respectively, filling the left component frame of the leftmost component frame and the right component frame of the rightmost component frame; according to k = (the horizontal distance between two adjacent component frames) / / w, use k as the number of component frames to be filled in the middle for filling the middle component frames; the width w of the filled component frame is taken as the average value of the widths of n component frames, and the height h of the filled component frame is taken as the average value of the heights of n component frames.

Images