A method for judging infrared image quality of a power transmission line composite insulator

Abstract

This invention discloses a method for judging the quality of infrared images of composite insulators for transmission lines. It includes an infrared image acquisition module, a composite insulator detection module, a composite insulator quality feature extraction module, and a composite insulator quality judgment module. The composite insulator quality feature extraction module includes modules for multi-segment features, centering features, excessive distance features, and incompleteness features. The beneficial effects of this invention are: automatically, quickly, and objectively assessing whether the quality meets the requirements of subsequent analysis; fundamentally improving the effective utilization rate of UAV infrared inspection data, ensuring the accuracy of defect diagnosis results, and laying a solid data foundation for building an efficient and reliable intelligent assessment and early warning business process for transmission line insulator conditions. It has significant engineering application value and promising prospects for widespread application.

Description

Technical Field

[0001] This invention relates to the technical field of infrared image detection of power grid transmission lines, and in particular to a method for judging the quality of infrared images of composite insulators of transmission lines. Background Technology

[0002] With the continuous development of power systems towards higher voltage, larger capacity, and longer distances, the safe and stable operation of transmission lines is crucial to national energy security and economic and social development. Composite insulators, as a new type of insulating material, have gradually replaced or partially replaced traditional porcelain and glass insulators in transmission lines, especially ultra-high voltage, inter-regional power grids, and lines in harsh environments, due to their excellent hydrophobicity, resistance to flashover, light weight, high strength, and maintenance-free operation. Their operational reliability directly affects the safety of the entire line and even the local power grid. Typical faults of composite insulators include core rod brittle fracture, interface breakdown, sheath aging, and shed damage. These defects are often accompanied by abnormal localized heating during their development. Therefore, temperature monitoring of composite insulators is an effective technical means to diagnose early latent defects, assess their aging state, and prevent serious faults.

[0003] In recent years, the rapid development of intelligent inspection technology has brought revolutionary changes to the condition monitoring of power equipment. Among them, drone inspection systems equipped with infrared thermal imagers have become the mainstream method for detecting overheating in composite insulators on transmission lines, especially in complex terrain sections, due to their outstanding advantages such as high-altitude operation, overcoming geographical obstacles, close-range observation, flexibility, efficiency, and ensuring personnel safety. Through infrared thermal images collected by drones, maintenance personnel can obtain the temperature distribution field on the surface of composite insulators non-contactly and intuitively. Based on relevant standards or experience, they can then analyze characteristics such as temperature anomalies and temperature differences to determine whether overheating defects exist. This mode greatly improves the coverage and efficiency of inspections and is a key link in realizing condition-based maintenance and intelligent operation and maintenance of transmission lines.

[0004] However, in practical engineering applications, the effectiveness of infrared image-based thermal diagnosis is highly dependent on the quality of the acquired images. Because drone inspections are typically conducted outdoors in complex electromagnetic environments, variable weather conditions, and dynamic flight states, the infrared images captured often suffer from various quality problems, severely interfering with subsequent interpretation and analysis. Specifically: blurry images result in unclear key outlines and features of composite insulators, making it difficult to accurately locate hot spots; overexposed or underexposed images cause loss or distortion of temperature information, leading to discrepancies between the actual temperature value and the calculated temperature difference; improper shooting angles or excessive distances cause composite insulators to appear too small or deformed in the image, preventing effective imaging of key heating areas; furthermore, the presence of noise, artifacts, and overall color shifts due to inaccurate temperature calibration further complicates manual or automated identification. Low-quality images often force analysts to spend extra time on screening and filtering, and may even directly lead to missed or misjudged defects, thus creating potential safety hazards.

[0005] Currently, research and practice in the diagnostic analysis of infrared images of composite insulators largely focus on backend intelligent diagnostic models such as defect identification algorithms, temperature threshold setting, and fault classification. However, "input determines output." If the quality of the input model's data (infrared images) is unreliable, even the most advanced diagnostic algorithm will suffer a significant drop in the confidence level of its output. In building an automated and intelligent composite insulator heating detection workflow, the lack of a pre-emptive, automated, and standardized quality assessment step for input infrared images has become a key bottleneck in improving the reliability, accuracy, and efficiency of the entire diagnostic system. Therefore, it is urgent to research and establish a quality assessment method specifically for infrared images of composite insulators. Summary of the Invention

[0006] The present invention aims to overcome the aforementioned shortcomings in the prior art and provides a method for judging the infrared image quality of composite insulators for power transmission lines that can automatically, quickly, and objectively evaluate the quality of infrared images.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A method for judging the quality of infrared images of composite insulators for transmission lines includes an infrared image acquisition module, a composite insulator detection module, a composite insulator quality feature extraction module, and a composite insulator quality judgment module. The composite insulator quality feature extraction module includes a multi-segment feature module, a centering feature module, a distance-to-center feature module, and an incomplete feature module. The specific operation steps are as follows: (1) The infrared image acquisition module is responsible for acquiring raw data. This module automatically reads the infrared thermal image to be analyzed uploaded by the UAV inspection system from the designated cloud storage service. After reading, it performs basic format parsing and caching on the image to provide standardized input data for the subsequent process. (2) The composite insulator detection module performs target localization of composite insulators. It adopts the YOLO target detection algorithm based on deep learning to analyze the input infrared image in real time, identify the composite insulator as a whole in the image, and output the bounding box coordinates of its area. (3) The composite insulator quality feature extraction module performs multi-dimensional quality feature quantitative analysis. Among them, the composite insulator multi-segment feature module evaluates whether the ultra-high voltage line is photographed in multiple segments as required, while other lines cannot be photographed in multiple segments; the composite insulator centering feature module analyzes the relative position of the insulator bounding box and the image center to determine whether the target is located in the main area of ​​the image; the composite insulator too far feature module judges whether the composite insulator image is too small based on the ratio of the bounding box area to the whole image area; the composite insulator incomplete feature module detects whether the bounding box touches the image edge, thus causing the insulator component to be incomplete due to improper composition; (4) The composite insulator quality judgment module compares the quantitative values ​​of various features output by the composite insulator quality feature extraction module with the preset quality threshold rule library, performs a comprehensive score on the image quality through weighted summarization and classifies the level, and outputs the quality judgment result of the infrared image.

[0008] This invention proposes a method for analyzing the quality of infrared images of composite insulators based on factors such as location, length, and integrity. This method enables quality evaluation of infrared images of composite insulators in transmission lines. Before the infrared image enters the defect diagnosis process, it automatically, quickly, and objectively assesses whether its quality meets the requirements for subsequent analysis. By constructing a quality evaluation system from multiple dimensions such as occlusion, integrity, and distance, it achieves automatic screening and early warning of unqualified images, ensuring that only high-quality images are sent to subsequent analysis stages. The implementation of this method will fundamentally improve the effective utilization rate of UAV infrared inspection data, ensure the accuracy of defect diagnosis results, and lay a solid data foundation for building an efficient and reliable intelligent assessment and early warning business process for transmission line insulator conditions. It has significant engineering application value and promising prospects for widespread application.

[0009] Preferably, in step (1), the infrared image acquisition module realizes the automated reception and identification management of infrared images transmitted back by the UAV inspection system. This module connects with the enterprise-level object storage service interface to continuously monitor and acquire infrared image data files uploaded by the UAV inspection system platform. Each infrared image is stored with its file identifier MFID as a unique file name. After receiving the image file, the module directly parses the MFID from its file name. This MFID serves as the core identifier of the image in the entire system and is persisted in the metadata as a key index record.

[0010] Preferably, in step (2), the composite insulator detection module locates the target composite insulator from the complex inspection background and analyzes its key structural components; the specific operation is as follows: (21) The module receives infrared images from the infrared image acquisition module and processes the entire image using the YOLO target detection algorithm based on deep learning. The segmentation model is trained on the labeled infrared insulator images and identifies the composite insulator as a whole target in the image. The segmentation model outputs an axially aligned rectangular bounding box at the location of the composite insulator and records the coordinates of the upper left corner vertex of the box in pixel coordinates. Its width w and height h are used to achieve primary positioning and background separation of the target composite insulator; (22) Based on the overall rectangular bounding box, the instance segmentation algorithm is used to process the rectangular bounding box region. The segmentation model will distinguish each pixel belonging to the composite insulator at the pixel level and generate a binary mask. Based on this mask, the minimum bounding rectangle that encloses the composite insulator is calculated, which is the precise bounding box of the composite insulator. (23) Semantic segmentation is performed on the precise bounding box of the composite insulator obtained by segmentation. Using the trained segmentation model, the composite insulator is further parsed into three key functional components: the head of the composite insulator, the body of the composite insulator, and the high-voltage end of the composite insulator. At the same time, the file name of this infrared image is parsed, and the voltage level information corresponding to the insulator is automatically extracted. Finally, the information of this composite insulator is encapsulated as follows:

[0011] in, I Indicates composite insulator, V This indicates the voltage level corresponding to this composite insulator. M_whole This represents the overall matrix of the composite insulator obtained from the segmentation; P_head, P_shed, P_ring The matrix represents the three key components obtained from semantic segmentation, corresponding to the head, the umbrella skirt body, and the high-voltage equalization ring, respectively. R_head, R_ring This indicates whether the head and high-pressure equalizing ring are present.

[0012] Preferably, in step (3), the composite insulator quality feature extraction module is responsible for performing multi-dimensional shooting quality quantitative analysis on the composite insulator output by the composite insulator detection module. This module sets up sub-modules based on the physical structure of the composite insulator and the imaging geometry, focusing on continuity, composition, proportion, and integrity, to achieve automated quality evaluation. It includes a composite insulator multi-segment feature module to determine whether the composite insulator needs to be photographed as multiple segments, a composite insulator centering feature module to assess the degree of deviation of the overall position of the composite insulator in the image, a composite insulator too far feature module to determine whether the shooting distance is too far, resulting in the target being too small, based on the mapping relationship between the pixel size of the composite insulator and the theoretical size under the corresponding voltage, and a composite insulator imperfection feature module to verify whether key components are missing.

[0013] Preferably, in step (3), the composite insulator multi-segment feature module is used to automatically determine the rationality of the infrared image shooting method based on the voltage level of the composite insulator and the integrity of its key components, thereby identifying the problem that the main body of the composite insulator is cut into multiple segments or the single segment image is incomplete due to improper shooting composition angle; the module accepts the standardized instance object of the composite insulator detection module as input, and through the analysis of the input information, finally outputs a quantitative segmentation feature mark of whether it is reasonable; its processing flow is based on the voltage threshold of 500KV, specifically: comparing the voltage level of the composite insulator with the threshold to determine the category to which the composite insulator belongs, and combining the mark of the component. R_head and R_ring Perform the following judgment: (a) When V ≥ 500KV, it belongs to the high-voltage level insulator and requires segmented shooting; if R_head and R_ring A value of 1 indicates that a single image completely contains both the head and the high-voltage end, which is inconsistent with the expectation of segmented shooting and is therefore judged as abnormal; otherwise, it is judged as normal. (b) When V < 500KV, it belongs to the medium and low voltage level insulator, and a complete image is expected in a single shot; if R_head and R_ ring If both are 0, it is considered abnormal; otherwise, it is considered normal. The output segmentation feature label of this module is used. F_seg Indicate: .

[0014] Preferably, in step (3), the composite insulator centering feature module quantitatively evaluates the degree of deviation of the composite insulator's position in the infrared image. This module receives a standardized instance object from the composite insulator detection module as input, which includes the matrix of the entire composite insulator. M_wholeBy calculating the pixel ratio of the composite insulator body located in the center region of the image, a quantized centering feature marker is output. F_center Specifically: (i) Define a central region with the geometric center of the image as its core, based on the size of the input infrared image. C_ region The central region is defined as a rectangular area that is half the length and half the width of the image. Its pixel coordinates range from w / 4 to 3w / 4 and from h / 4 to 3h / 4. This region contains the core 50% of the pixel area of ​​the image. (ii) Extract the entire composite insulator M_whole The set of all pixels with a value of 1 P_whole By calculating the set P_whole It is also located in the central area C_region The number of pixels within the central region is used to obtain the set of composite insulator pixels located in the central region. P_center ; (iii) Calculate the percentage of pixels in the center region of the composite insulator by quantizing the centering degree. F_center:

[0015] in, Represents a set P_center The number of pixels, Represents a set P_whole The number of pixels; F_center The ratio reflects the degree of concentration of the composite insulator body in the central region of the image.

[0016] Preferably, in step (3), the composite insulator over-distance feature module evaluates whether the image shooting distance is appropriate. This module takes a standardized instance of the composite insulator detection module as input, and outputs its quantified over-distance feature by analyzing the absolute pixel size and shape ratio of the insulator; specifically: (A) For the main body of the umbrella skirt P_shed The analysis calculates the length of the region matrix along its principal axis and its width perpendicular to it in the image pixel space. By performing principal axis analysis on the mask or calculating its minimum bounding rectangle, two key pixel dimensions are obtained: the pixel length of the composite insulator subject area. L_pixel and pixel width W_pixel ; (B) Perform logical judgment, which is divided into two steps: the first step is to determine the absolute size, by using the absolute width threshold. W_th =10 pixels, if the calculated pixel width W_pixel < W_th If the composite insulator image is too small, the shooting distance is too far; if the first step is to determine... W_pixel > W_thThen, the second step, relative proportion judgment, is performed, taking into account the morphological characteristics of the composite insulator, and its width-to-length ratio is calculated.

[0017] in Ratio_wl Indicates the width-to-length ratio of the composite insulator; (C) Dynamic determination based on voltage level V, if the voltage level V ≥ 500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl < 0.15 If the voltage level is too high, the judgment is too far; if the voltage level is too high... V < 500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl < 0.1 If it is too far, then it is judged as too far; the judgment result F_far It is expressed as follows:

[0018] in, F_far = 1 This indicates that the shooting distance is appropriate. F_far = 0 This indicates that the image was taken too far away, resulting in substandard image quality.

[0019] Preferably, in step (3), the composite insulator incomplete feature module is used to quantitatively evaluate and determine the imaging integrity of the composite insulator in the infrared image, specifically as follows: (I) The main body of the composite insulator umbrella group P_shed Perform principal axis analysis to obtain its pixel length. L_pixel and pixel width W_pixel ;if W_pixel ≥ L_pixel If so, it is determined that the composite insulator image is incomplete; (II) If the pixel length of the composite insulator L_pixel < 30 This indicates that the target is too small in length in the image, and the thermal imaging information it contains is too sparse to make a reliable temperature determination. The results of the composite insulator incompleteness determination are as follows:

[0020] in, F_imcomplete = 1 This indicates that the image integrity check has passed.

[0021] Preferably, in step (4), the composite insulator quality judgment module is responsible for comprehensively evaluating the outputs of the four quality feature modules and generating the final infrared image quality assessment result; this module sequentially receives quantization markers from each sub-module: multi-segment feature markers F_seg , proportion of centered feature pixels P_center Too far feature markers F_far and incomplete feature markers F_imcomplete Make a comprehensive judgment: (41) Key item check: Check three features, if multiple feature segments are markedF_seg = 0 Too far feature markers F_far = 0 or incomplete feature markers F_imcomplete = 0 If any of the conditions are met, the module will directly determine that the composite insulator in the infrared image is unqualified. (42) Centering degree qualification judgment: If the proportion of centering feature pixels is judged... P_center If the proportion is less than 50%, the target is considered to be excessively off-center in the image and is judged as unqualified.

[0022] The beneficial effects of this invention are: it automatically, quickly, and objectively assesses whether the quality meets the requirements of subsequent analysis; it fundamentally improves the effective utilization rate of UAV infrared inspection data, ensures the accuracy of defect diagnosis results, and lays a solid data foundation for building an efficient and reliable intelligent assessment and early warning business process for power transmission line compliance, thus possessing significant engineering application value and promotion prospects. Attached Figure Description

[0023] Figure 1 This is a flowchart of the method of the present invention.

[0024] Figure 2 This is the overall architecture diagram of the present invention. Detailed Implementation

[0025] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0026] like Figure 1 , Figure 2 In the aforementioned embodiment, a method for judging the quality of infrared images of composite insulators for transmission lines includes an infrared image acquisition module, a composite insulator detection module, a composite insulator quality feature extraction module, and a composite insulator quality judgment module. The composite insulator quality feature extraction module includes a multi-segment feature module, a centering feature module, a distance-to-center feature module, and an incomplete feature module. The specific operation steps are as follows: (1) The infrared image acquisition module is responsible for acquiring raw data. This module automatically reads the infrared thermal image to be analyzed uploaded by the UAV inspection system from the designated cloud storage service. After reading, it performs basic format parsing and caching on the image to provide standardized input data for the subsequent process.

[0027] The infrared image acquisition module automates the reception and identification management of infrared images transmitted from the UAV inspection system. This module interfaces with Enterprise Object Storage Service (OBS) to continuously monitor and acquire infrared image data files uploaded by the UAV inspection system platform. Each infrared image is stored with its file identifier (MFID) as a unique filename. Upon receiving an image file, the module directly parses the MFID from its filename. This MFID serves as the core identifier of the image within the entire system and is persisted as a key index record in the metadata.

[0028] (2) The composite insulator detection module performs target localization of composite insulators. It adopts the YOLO target detection algorithm based on deep learning to analyze the input infrared image in real time. It can accurately identify the composite insulator as a whole in the image and output the bounding box coordinates of its area. This module separates the target composite insulator from the complex inspection background, laying the foundation for subsequent targeted quality feature extraction.

[0029] The composite insulator detection module locates the target composite insulator from a complex inspection background and analyzes its key structural components, providing accurate input for subsequent quality feature extraction. The specific operation is as follows: (21) The module receives infrared images from the infrared image acquisition module and processes the entire image using the YOLO target detection algorithm based on deep learning. The segmentation model is trained on a large number of labeled infrared insulator images and can quickly identify composite insulator targets in the image. The segmentation model outputs an axially aligned rectangular bounding box at the location of the composite insulator, and records the coordinates of the top left vertex of the box in pixel coordinates. Its width w and height h enable primary positioning and background separation of the target composite insulator.

[0030] (22) Based on the overall rectangular bounding box, the precise external contour of the composite insulator is further obtained. The instance segmentation algorithm is used to refine the rectangular region. The segmentation model distinguishes each pixel belonging to the composite insulator at the pixel level and generates a precise binary mask. Based on this mask, the minimum bounding rectangle that encloses the composite insulator is calculated, which is the precise bounding box of the composite insulator.

[0031] (23) Semantic segmentation is performed on the precise bounding box of the composite insulator obtained from the segmentation. Using the trained segmentation model, the composite insulator is further parsed into three key functional components: the composite insulator head, the composite insulator body, and the high-voltage end of the composite insulator. At the same time, the file name of this infrared image is parsed, and the voltage level information corresponding to the insulator is automatically extracted. Finally, the information encapsulation of this composite insulator is as follows:

[0032] in, I Indicates composite insulator, V This indicates the voltage level corresponding to this composite insulator. M_whole This represents the overall matrix of the composite insulator obtained from the segmentation. P_head, P_shed, P_ring The matrix represents the three key components obtained from semantic segmentation, corresponding to the head, the umbrella skirt body, and the high-voltage equalization ring, respectively. R_head, R_ring This indicates whether the head and high-pressure equalizing ring are present.

[0033] (3) The composite insulator quality feature extraction module performs multi-dimensional quality feature quantitative analysis. Among them, the composite insulator multi-segment feature module evaluates whether the ultra-high voltage line is photographed in multiple segments as required, while other lines cannot be photographed in multiple segments; the composite insulator centering feature module analyzes the relative position of the insulator bounding box and the image center to determine whether the target is located in the main area of ​​the image and avoids excessive deviation; the composite insulator too far feature module judges whether the composite insulator image is too small based on the ratio of the bounding box area to the whole image area, resulting in blurred details; the composite insulator incomplete feature module detects whether the bounding box touches the image edge, thus causing the insulator component to be incomplete due to improper composition.

[0034] The composite insulator quality feature extraction module is responsible for performing multi-dimensional quantitative analysis of the composite insulator image quality output by the composite insulator detection module. Based on the physical structure and imaging geometry of the composite insulator, this module establishes sub-modules for continuity, composition, proportion, and integrity to achieve automated quality evaluation. It includes a multi-segment feature module to determine whether the composite insulator needs to be captured as multiple segments; a centering feature module to assess the degree of deviation of the overall position of the composite insulator in the image; an excessively far feature module to determine whether the shooting distance is too far, resulting in a small target and blurred details, based on the mapping relationship between the composite insulator pixel size and the theoretical size under the corresponding voltage; and an incomplete feature module to verify whether key components are missing.

[0035] The composite insulator multi-segment feature module is used to automatically determine the rationality of the infrared image shooting method based on the voltage level of the composite insulator and the integrity of its key components. This identifies problems such as the composite insulator being cropped into multiple segments or incomplete single-segment imaging due to improper shooting angles. The module accepts standardized instance objects from the composite insulator detection module as input. Through analysis of the input information, it ultimately outputs a quantified feature mark indicating whether the segmentation is reasonable. Its processing flow is based on a 500kV voltage threshold, specifically: comparing the composite insulator voltage level with the threshold to determine the category of the composite insulator (high voltage or medium / low voltage), and combining this with the markers indicating the presence of components. R_head and R_ring Perform the following judgment: (a) When V ≥ 500KV, it belongs to the high-voltage level insulator and requires segmented shooting. If R_head and R_ring A value of 1 indicates that a single image fully contains both the head and the high-voltage end, which is inconsistent with the expectation of segmented shooting and is therefore considered abnormal; otherwise, it is considered normal.

[0036] (b) When V < 500KV, it belongs to the medium and low voltage level insulator, and a complete image is expected in a single shot. If R_head and R_ ring If both are 0, it is considered abnormal; otherwise, it is considered normal.

[0037] The output segmentation feature label of this module is used. F_seg Indicate: .

[0038] The composite insulator centering feature module quantitatively evaluates the degree of deviation of the composite insulator's position in the infrared image. Its core function is to determine whether the main body of the composite insulator is located in the center region of the image. A centered image usually means that the target subject is prominent and the background interference is small, which is beneficial for subsequent detailed observation and heat point analysis. This module receives a standardized instance object from the composite insulator detection module as input, which contains the matrix of the entire composite insulator. M_ whole By calculating the pixel ratio of the composite insulator body located in the center region of the image, a quantized centering feature marker is output. F_center Specifically: (i) Define a central region with the geometric center of the image as its core, based on the size of the input infrared image. C_ region The central region is defined as a rectangular area that is half the length and half the width of the image. Its pixel coordinates range from w / 4 to 3w / 4 on the horizontal axis and from h / 4 to 3h / 4 on the vertical axis. This region contains the core 50% of the image's pixel area.

[0039] (ii) Extract the entire composite insulator M_whole The set of all pixels with a value of 1 P_whole By calculating the set P_whole It is also located in the central area C_region The number of pixels within the central region is used to obtain the set of composite insulator pixels located in the central region. P_center .

[0040] (iii) Calculate the percentage of pixels in the center region of the composite insulator by quantizing the centering degree. F_center:

[0041] in, Represents a set P_center The number of pixels, Represents a set P_whole The number of pixels; F_center The ratio reflects the degree of concentration of the composite insulator body in the central region of the image.

[0042] The composite insulator over-distance feature module evaluates whether the image shooting distance is appropriate, identifying problems such as the composite insulator appearing too small and lacking detail resolution in the image due to an excessively long shooting distance. This module takes a standardized instance of the composite insulator detection module as input, analyzes the absolute pixel size and morphological proportions of the insulator, and outputs its quantified over-distance features. Specifically: (A) For the main body of the umbrella skirt P_shed The analysis calculates the length of the region matrix along its principal axis and its width perpendicular to it in the image pixel space. By performing principal axis analysis on the mask or calculating its minimum bounding rectangle, two key pixel dimensions are obtained: the pixel length of the composite insulator subject area. L_pixel and pixel width W_pixel .

[0043] (B) Perform logical judgment, which is divided into two steps: the first step is to determine the absolute size, by using the absolute width threshold. W_th =10 pixels. If the calculated pixel width... W_pixel < W_th If the image of the composite insulator is too small, it can be directly determined that the imaging distance is too far. If the first step is to determine... W_pixel > W_th Then, the second step, relative proportion judgment, is performed, taking into account the morphological characteristics of the composite insulator, and its width-to-length ratio is calculated.

[0044] in Ratio_wl Indicates the width-to-length ratio of the composite insulator; (C) Dynamic determination based on voltage level V, if the voltage level V ≥ 500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl < 0.15 If the voltage level is too high, the judgment is too far; if the voltage level is too high... V < 500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl < 0.1 If so, it is determined to be too far. The determination result is... F_far It is expressed as follows:

[0045] in, F_far = 1 This indicates that the shooting distance is appropriate. F_far = 0 This indicates that the image was taken too far away, resulting in substandard image quality.

[0046] The composite insulator incompleteness feature module is used to quantitatively evaluate and determine the imaging integrity of the composite insulator in infrared images. If the composite insulator imaging is incomplete, the number of effective temperature pixels extracted subsequently will be insufficient, making it impossible to accurately analyze whether it has heating defects. Specifically: (I) The main body of the composite insulator umbrella group P_shed Perform principal axis analysis to obtain its pixel length. L_pixel and pixel width W_pixel .if W_pixel ≥ L_pixel If so, it is determined that the composite insulator is not photographed completely.

[0047] (II) If the pixel length of the composite insulator L_pixel < 30 This indicates that the target is too small in length in the image, and the thermal imaging information it contains is too sparse to make a reliable temperature determination. The results of the composite insulator incompleteness determination are as follows:

[0048] in, F_imcomplete = 1 This indicates that the image integrity check has passed.

[0049] (4) The composite insulator quality judgment module compares the quantitative values ​​of various features (centering, proportion, integrity, number of segments) output by the composite insulator quality feature extraction module with the preset quality threshold rule library, performs a comprehensive score on the image quality through weighted summarization and classifies the level, and outputs the quality judgment result of the infrared image.

[0050] The composite insulator quality assessment module is responsible for comprehensively evaluating the outputs of the four quality feature modules and generating the final infrared image quality assessment result. This module sequentially receives quantization markers from each submodule: multi-segment feature markers. F_seg , proportion of centered feature pixels P_center Too far feature markers F_far and incomplete feature markers F_ imcomplete Make a comprehensive judgment: (41) Key item check: Check three features, if multiple feature segments are marked F_seg = 0 Too far feature markers F_far = 0 or incomplete feature markers F_imcomplete = 0 If any of the conditions are met, the module will directly determine that the composite insulator in the infrared image is unqualified.

[0051] (42) Centering degree qualification judgment: If the proportion of centering feature pixels is judged... P_center If the proportion is less than 50%, the target is considered to be excessively off-center in the image and is judged as unqualified.

[0052]

[0053] Combination Figure 1 (Method Flowchart) and Figure 2 As shown in the system architecture diagram, the specific implementation process of the infrared image quality judgment method for composite insulators of transmission lines provided by this invention is as follows: Step S1: Infrared image acquisition and data preprocessing. The system automatically listens for and acquires infrared thermal image files uploaded by the UAV inspection platform via the Object Storage Service (OBS) interface. Each image file is named with its unique File Identifier (MFID). The acquisition module parses the filename to obtain the MFID, which is stored as the core index of the image in the system. Subsequently, the module performs basic format parsing and standardization caching of the images to provide input data in a unified format for subsequent processing.

[0054] Step S2: Composite insulator target detection and case analysis.

[0055] Coarse target localization: A pre-trained YOLO target detection model is used to process the entire input infrared image to quickly identify the composite insulator as a whole and output its initially axially aligned rectangular bounding box. BBox_detect = (x_ (min, y_min, w, h) .

[0056] Instance segmentation: Within the initial bounding box region, an instance segmentation algorithm (such as...) is applied. Mask R-CNN Pixel-level segmentation is performed to generate a binary mask of the entire insulator. M_whole And calculate its minimum bounding box.

[0057] Component parsing and information encapsulation: for M_whole The region was further semantically segmented and parsed into three key components: the insulator head, the shed body, and the high-voltage end equalizing ring, and their masks were obtained respectively. P_head, P_shed, P_ ring Simultaneously, the rated voltage level of the insulator is parsed from the image file name or associated metadata. V Based on the above information, a standardized composite insulator instance object is constructed. I : I = {V, M_whole, P_head, P_shed, P_ring, R_head, R_ring} in, R_head and R_ring These are binary flags, indicating the presence of the head and high-voltage end components, respectively. Area > 0 ).

[0058] Step S3: Extraction and quantification of multi-dimensional quality features.

[0059] Multi-segment feature extraction: based on voltage level Vand component markings R_head, R_ring Judge the appropriateness of the shooting method. If V ≥ 500kV The plan is to shoot in segments; if at this time... R_head and R_ring Both are 1 (i.e., ( R_head ∧ R_ ring)=1 If ), then it is determined that it does not meet the segmentation expectation ( F_seg=0 ), otherwise it meets the requirements ( F_seg=1 ).like V < 500kV The expected result is a single complete image; if at this time... R_head and R_ring If both are 1, then it meets the expectation ( F_seg=1 Otherwise it does not meet the requirements. F_seg=0 ).

[0060] Centered feature extraction: Calculate the pixel ratio of the insulator body located in the central region of the image (defined as a central rectangle that is half the length and half the width of the image). P_center .

[0061]

[0062] Where |P_center| is the number of insulator pixels located in the center region, and |P_whole| is the total number of insulator pixels. The P_center value is directly used for subsequent comprehensive scoring.

[0063] Feature extraction for excessive distances: based on umbrella skirt body mask P_shed Calculate its pixel width W_pixel and pixel length L_pixel .

[0064] like W_pixel < 10 It was directly determined that the shooting distance was too far ( F_far=0 Otherwise, calculate the width-to-length ratio. Ratio_wl = W_pixel / L_pixel According to voltage level V Set threshold: If V ≥ 500kV and Ratio_wl < 0.15 If it is too far ( F_far=0 );like V < 500kV and Ratio_wl < 0.10 If it is too far ( F_far=0 Other situations are normal. F_far=1 ).

[0065] Incomplete feature extraction: also based on P_shed calculate W_pixel and L_pixel .

[0066] like W_pixel > L_pixel (morphological abnormality) or L_pixel < 30 (If the size is too small), it is considered incomplete. F_incomplete=0 Otherwise, it is judged as complete. F_incomplete=1 ).

[0067] Step S4: Comprehensive quality assessment and result output. Key veto item check: First, conduct a veto check. If there are multiple features ( F_seg=0 ), excessively far characteristics ( F_ far=0 ) or incomplete features ( F_incomplete=0 If any item in the process is "unqualified", the module will directly determine that the overall quality of the infrared image is unqualified and terminate the process.

[0068] Centering degree compliance judgment: After passing the rejection items check, the centering characteristics are judged. If P_center < 0.5 If the percentage is less than 50%, the target is considered to be excessively off-center from the image and is deemed to be unqualified overall.

[0069] Through the above four steps, this invention achieves full automation of the entire process, from automatic acquisition of UAV infrared images, accurate identification of composite insulators, quantification of multi-dimensional quality characteristics, to intelligent comprehensive evaluation. This method closely integrates the structural characteristics of composite insulators with the imaging requirements for thermal diagnosis, effectively filtering out high-quality images and laying a solid data foundation for subsequent reliable identification of thermal defects. This significantly improves the operational efficiency and reliability of UAV intelligent inspection.

[0070] In summary, this method aims to propose a quality assessment approach for infrared images of composite insulators. This method can automatically, quickly, and objectively evaluate whether the quality of infrared images meets the requirements of subsequent analysis before they enter the defect diagnosis process. By constructing a quality evaluation system from multiple dimensions such as occlusion, integrity, and distance, it achieves automatic screening and early warning of unqualified images, thereby ensuring that only high-quality images are sent to subsequent analysis stages. The implementation of this method will fundamentally improve the effective utilization rate of UAV infrared inspection data, ensure the accuracy of defect diagnosis results, and lay a solid data foundation for building an efficient and reliable intelligent assessment and early warning business process for transmission line insulator conditions. It has significant engineering application value and promising prospects for widespread application.

Claims

1. A method for judging the quality of infrared images of composite insulators for transmission lines, characterized in that, It includes an infrared image acquisition module, a composite insulator detection module, a composite insulator quality feature extraction module, and a composite insulator quality judgment module. The composite insulator quality feature extraction module includes modules for multi-segment composite insulator features, composite insulator centering features, composite insulator excessive distance features, and composite insulator incompleteness features. The specific operation steps are as follows: (1) The infrared image acquisition module is responsible for acquiring raw data. This module automatically reads the infrared thermal image to be analyzed uploaded by the UAV inspection system from the designated cloud storage service. After reading, it performs basic format parsing and caching on the image to provide standardized input data for the subsequent process. (2) The composite insulator detection module performs target localization of composite insulators. It adopts the YOLO target detection algorithm based on deep learning to analyze the input infrared image in real time, identify the composite insulator as a whole in the image, and output the bounding box coordinates of its area. (3) The composite insulator quality feature extraction module performs multi-dimensional quality feature quantitative analysis. Among them, the composite insulator multi-segment feature module evaluates whether the ultra-high voltage line is photographed in multiple segments as required, while other lines cannot be photographed in multiple segments; the composite insulator centering feature module analyzes the relative position of the insulator bounding box and the image center to determine whether the target is located in the main area of ​​the image; the composite insulator too far feature module judges whether the composite insulator image is too small based on the ratio of the bounding box area to the whole image area; the composite insulator incomplete feature module detects whether the bounding box touches the image edge, thus causing the insulator component to be incomplete due to improper composition; (4) The composite insulator quality judgment module compares the quantitative values ​​of various features output by the composite insulator quality feature extraction module with the preset quality threshold rule library, performs a comprehensive score on the image quality through weighted summarization and classifies the level, and outputs the quality judgment result of the infrared image.

2. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 1, characterized in that, in In step (1), the infrared image acquisition module realizes the automated reception and identification management of infrared images transmitted back by the UAV inspection system. This module connects with the enterprise-level object storage service interface to continuously monitor and acquire infrared image data files uploaded by the UAV inspection system platform. Each infrared image is stored with its file identifier MFID as a unique file name. After receiving the image file, the module directly parses the MFID from its file name. This MFID serves as the core identifier of the image in the entire system and is persisted in the metadata as a key index record.

3. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 1, characterized in that, In step (2), the composite insulator detection module locates the target composite insulator from the complex inspection background and analyzes its key structural components; the specific operation is as follows: (21) The module receives infrared images from the infrared image acquisition module and processes the entire image using the YOLO target detection algorithm based on deep learning. The segmentation model is trained on the labeled infrared insulator images and identifies the composite insulator as a whole target in the image. The segmentation model outputs an axially aligned rectangular bounding box at the location of the composite insulator and records the coordinates of the upper left corner vertex of the box in pixel coordinates. Its width w and height h are used to achieve primary positioning and background separation of the target composite insulator; (22) Based on the overall rectangular bounding box, the instance segmentation algorithm is used to process the rectangular bounding box region. The segmentation model will distinguish each pixel belonging to the composite insulator at the pixel level and generate a binary mask. Based on this mask, the minimum bounding rectangle that encloses the composite insulator is calculated, which is the precise bounding box of the composite insulator. (23) Semantic segmentation is performed on the precise bounding box of the composite insulator obtained by segmentation. Using the trained segmentation model, the composite insulator is further parsed into three key functional components: the head of the composite insulator, the body of the composite insulator, and the high-voltage end of the composite insulator. At the same time, the file name of this infrared image is parsed, and the voltage level information corresponding to the insulator is automatically extracted. Finally, the information of this composite insulator is encapsulated as follows: in, I Indicates composite insulator, V This indicates the voltage level corresponding to this composite insulator. M_whole This represents the overall matrix of the composite insulator obtained from the segmentation; P_head、P_ shed, P_ring The matrix represents the three key components obtained from semantic segmentation, corresponding to the head, the umbrella skirt body, and the high-voltage equalization ring, respectively. R_head, R_ring This indicates whether the head and high-pressure equalizing ring are present.

4. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 1, characterized in that, in In step (3), the composite insulator quality feature extraction module is responsible for performing multi-dimensional imaging quality quantification analysis on the composite insulator output by the composite insulator detection module. Based on the physical structure of the composite insulator and the imaging geometry, this module sets up sub-modules from continuity, composition, proportion and integrity to realize automated quality evaluation. The system includes a multi-segment feature module for composite insulators to determine whether the composite insulator needs to be captured as multiple segments; a centering feature module for composite insulators to assess the degree of deviation of the overall position of the composite insulator in the image; a feature module for composite insulators that are too far away to determine whether the shooting distance is too far, resulting in the target being too small, based on the mapping relationship between the pixel size of the composite insulator and the theoretical size under the corresponding voltage; and a feature module for composite insulators that are incomplete to verify whether key components are missing.

5. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 3, characterized in that, in In step (3), the composite insulator multi-segment feature module is used to automatically determine the rationality of the infrared image shooting method based on the voltage level of the composite insulator and the integrity of its key components, thereby identifying the problem that the main body of the composite insulator is cut into multiple segments or the single segment image is incomplete due to improper shooting composition angle; the module accepts the standardized instance object from the composite insulator detection module as input, and through the analysis of the input information, finally outputs a quantitative segmentation feature mark of whether it is reasonable; The processing flow is based on a voltage threshold of 500KV. Specifically, it involves comparing the voltage level of the composite insulator with the threshold to determine the category of the composite insulator, and then combining this with the markings on the components. R_head and R_ring Perform the following judgment: (a) When V ≥ 500KV, it belongs to the high-voltage level insulator and requires segmented shooting; if R_head and R_ring A value of 1 indicates that a single image completely contains both the head and the high-voltage end, which is inconsistent with the expectation of segmented shooting and is therefore judged as abnormal; otherwise, it is judged as normal. (b) When V < 500KV, it belongs to the medium and low voltage level insulator, and a complete image is expected in a single shot; if R_head and R_ring If both are 0, it is considered abnormal; otherwise, it is considered normal. The output segmentation feature label of this module is used. F_seg Indicate: 。 6. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 3, characterized in that, in In step (3), the composite insulator centering feature module quantitatively evaluates the degree of deviation of the composite insulator's position in the infrared image. This module receives a standardized instance object from the composite insulator detection module as input, which contains the matrix of the entire composite insulator. M_whole By calculating the pixel ratio of the composite insulator body located in the center region of the image, a quantized centering feature marker is output. F_center Specifically: (i) Define a central region with the geometric center of the image as its core, based on the size of the input infrared image. C_ region The central region is defined as a rectangular area that is half the length and half the width of the image. Its pixel coordinates range from w / 4 to 3w / 4 and from h / 4 to 3h / 4. This region contains the core 50% of the pixel area of ​​the image. (ii) Extract the entire composite insulator M_whole The set of all pixels with a value of 1 P_whole By calculating the set P_whole It is also located in the central area C_region The number of pixels within the central region is used to obtain the set of composite insulator pixels located in the central region. P_center ; (iii) Calculate the percentage of pixels in the center region of the composite insulator by quantizing the centering degree. F_center: in, Represents a set P_center The number of pixels, Represents a set P_whole The number of pixels; F_center The ratio reflects the degree of concentration of the composite insulator body in the central region of the image.

7. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 3, characterized in that, In step (3), the composite insulator too far feature module evaluates whether the shooting distance of the image is appropriate. This module takes the standardized instance object of the composite insulator detection module as input, and outputs its quantified too far feature by analyzing the absolute pixel size and shape ratio of the insulator. Specifically: (A) For the main body of the umbrella skirt P_shed The analysis calculates the length of the region matrix along its principal axis and its width perpendicular to it in the image pixel space. By performing principal axis analysis on the mask or calculating its minimum bounding rectangle, two key pixel dimensions are obtained: the pixel length of the composite insulator subject area. L_pixel and pixel width W_pixel ; (B) Perform logical judgment, which is divided into two steps: the first step is to determine the absolute size, by using the absolute width threshold. W_th =10 pixels, if the calculated pixel width W_pixel <W_th If the composite insulator image is too small, the shooting distance is too far; if the first step is to determine... W_pixel>W_th Then, the second step, relative proportion judgment, is performed, taking into account the morphological characteristics of the composite insulator, and its width-to-length ratio is calculated. in Ratio_wl Indicates the width-to-length ratio of the composite insulator; (C) Dynamic determination based on voltage level V, if the voltage level V≥500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl<0.15 If the voltage level is too high, the judgment is too far; if the voltage level is too high... V<500KV If the calculated width-to-length ratio of the composite insulator Ratio_wl<0.1 If it is too far, then it is judged as too far; the judgment result F_far It is expressed as follows: in, F_far=1 This indicates that the shooting distance is appropriate. F_far=0 This indicates that the image was taken too far away, resulting in substandard image quality.

8. The method for judging the infrared image quality of composite insulators for transmission lines according to claim 7, characterized in that, in In step (3), the composite insulator incomplete feature module is used to quantitatively evaluate and determine the imaging integrity of the composite insulator in the infrared image, specifically as follows: (I) The main body of the composite insulator umbrella group P_shed Perform principal axis analysis to obtain its pixel length. L_pixel and pixel width W_pixel ;if W_pixel≥L_pixel If so, it is determined that the composite insulator image is incomplete; (II) If the pixel length of the composite insulator L_pixel<30 This indicates that the target is too small in length in the image, and the thermal imaging information it contains is too sparse to make a reliable temperature determination. The results of the composite insulator incompleteness determination are as follows: in, F_imcomplete=1 This indicates that the image integrity check has passed.

9. A method for judging the infrared image quality of composite insulators for transmission lines according to claim 5, 6, 7, or 8, characterized in that, in In step (4), the composite insulator quality judgment module is responsible for comprehensively evaluating the outputs of the four quality feature modules and generating the final infrared image quality assessment result; this module sequentially receives quantization markers from each sub-module: multi-segment feature markers F_seg , proportion of centered feature pixels P_center Too far feature markers F_far and incomplete feature markers F_imcomplete Make a comprehensive judgment: (41) Key item check: Check three features, if multiple feature segments are marked F_seg=0 Too far feature markers F_far=0 or incomplete feature markers F_imcomplete=0 If any of the conditions are met, the module will directly determine that the composite insulator in the infrared image is unqualified. (42) Centering degree qualification judgment: If the proportion of centering feature pixels is judged... P_center If the proportion is less than 50%, the target is considered to be excessively off-center in the image and is judged as unqualified.

Images