Insulator image segmentation method based on infrared enhanced image of unmanned aerial vehicle

Abstract

The invention relates to an insulator image segmentation method based on an infrared enhancement image of an unmanned aerial vehicle, and relates to the technical field of image processing, and the method comprises the steps: collecting an infrared image sequence through an infrared enhancement camera, obtaining a standard infrared image sequence, and carrying out the state recognition through an image perception prior engine, outputting a predicted insulator attention heat map and a predicted boundary confidence map; optimizing and adjusting the initial threshold segmentation algorithm and the initial region growing algorithm, and obtaining an adaptive threshold segmentation algorithm and an adaptive region growing algorithm; and carrying out image segmentation on the standard infrared image sequence to output an insulator image sequence, and carrying out early warning judgment on the insulator image sequence based on an adaptive early warning mechanism. The method solves the problems that a traditional insulator image segmentation method cannot effectively deal with the problems of low infrared image temperature contrast ratio, large noise interference and complex background, so that the insulator segmentation is easy to cause mistaken segmentation and missing segmentation, and the high-precision requirement of fault detection is difficult to meet.

Description

Technical Field

[0001] This application relates to the field of image processing, and in particular to an insulator image segmentation method based on UAV infrared enhanced images. Background Technology

[0002] In the operation and maintenance of power transmission lines, insulator fault detection is a key link in ensuring the safe and stable operation of the power grid. Accurate segmentation of insulator images is a prerequisite for efficient fault identification. The segmentation accuracy directly affects the accuracy of subsequent fault diagnosis, becoming an important technical requirement to support power grid operation and maintenance decisions.

[0003] Currently, traditional insulator image segmentation methods do not fully consider the inherent characteristics of infrared images, such as low temperature contrast, high noise interference, and complex backgrounds. They cannot specifically eliminate the interference of these factors on the segmentation process, which not only easily leads to missegmentation and omission of insulator areas, but also reduces the reliability of the segmentation results, making it difficult to meet the actual needs of insulator fault detection for high-precision segmentation. Summary of the Invention

[0004] This application provides an insulator image segmentation method based on UAV infrared enhanced images, which solves the shortcomings of traditional manual inspection and conventional image segmentation methods, and ensures the safe and stable operation of the power system.

[0005] The embodiments of this application disclose the following technical solutions: This application provides an insulator image segmentation method based on UAV infrared enhanced images, the method comprising: The infrared image sequence containing insulators is collected by an infrared enhancement camera mounted on a drone. The image preprocessing is performed to obtain a standard infrared image sequence. The state recognition is performed using an image perception prior engine, and the predicted insulator attention heatmap and predicted boundary confidence map are output. Based on the predicted insulator attention heatmap and predicted boundary confidence map, the initial threshold segmentation algorithm and the initial region growth algorithm are optimized and adjusted to obtain the adapted threshold segmentation algorithm and the adapted region growth algorithm. Using the adaptive threshold segmentation algorithm and the adaptive region growing algorithm, the standard infrared image sequence is segmented to output an insulator image sequence, and an early warning judgment is made on the insulator image sequence based on the adaptive early warning mechanism.

[0006] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application proposes an insulator image segmentation method based on UAV infrared enhanced images. Through a step-by-step process of constructing an initial segmentation algorithm, optimizing the adaptive segmentation algorithm, formulating an adaptive early warning mechanism, and making accurate early warning judgments, the method achieves accurate segmentation and defect early warning of insulator regions in UAV infrared enhanced images. First, the standard infrared images acquired by the UAV are preprocessed, including Gaussian filtering for noise reduction, histogram equalization, and normalization. Then, an initial threshold algorithm, including local adaptive threshold calculation and global threshold adjustment, and an initial region growth algorithm, including seed point selection and growth criterion definition, are constructed. Subsequently, based on prior information from the predicted insulator attention heatmap and predicted boundary confidence map, the initial threshold algorithm and initial region growth algorithm are optimized to obtain the adaptive threshold segmentation algorithm and adaptive region growth algorithm. Next, based on UAV flight speed, ambient temperature and humidity, and historical insulator performance data, the adaptive detection sensitivity coefficient is calculated, and multiple thresholds such as temperature and defect confidence are optimized and adjusted for the initial early warning mechanism to construct an adaptive early warning mechanism. Finally, the adaptive segmentation algorithm is used to segment the standard infrared image sequence to obtain an insulator image sequence, and multi-dimensional early warning judgments are performed on the sequence based on the adaptive early warning mechanism.

[0007] The technical solution of this application solves the problems of low segmentation accuracy caused by infrared image noise interference and complex background in traditional insulator image segmentation, the difficulty of adapting the fixed threshold early warning mechanism to different inspection scenarios, and the easy occurrence of false alarms and missed alarms in defect identification. It avoids the missegmentation of insulator regions and the missed detection of defects caused by insufficient generalization of segmentation algorithms or rigid early warning thresholds, and improves the accuracy of insulator segmentation and the timeliness of defect early warning under UAV infrared enhanced images. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating an insulator image segmentation method based on UAV infrared enhanced images, provided for an embodiment of this application; Figure 2 This is a schematic diagram illustrating the process of constructing an image perception prior engine including a shared encoder and dual decoders, as provided in an embodiment of this application. Detailed Implementation

[0010] This application provides an insulator image segmentation method based on UAV infrared enhanced images, which solves the technical problem that traditional insulator image segmentation methods in the prior art are prone to missegmentation and omissions due to their inability to effectively deal with the problems of low temperature contrast, large noise interference and complex background of infrared images, making it difficult to meet the high-precision requirements of fault detection.

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0013] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0014] Examples, as shown in the appendix Figure 1 As shown, this application provides an insulator image segmentation method based on UAV infrared enhanced images, the method comprising the following steps: S110: The UAV-mounted infrared enhancement camera collects infrared image sequences containing insulators, performs image preprocessing to obtain standard infrared image sequences, uses an image perception prior engine to perform state recognition, and outputs a predicted insulator attention heatmap and a predicted boundary confidence map. In this embodiment of the application, in the scenario of power system drone inspection and high-precision detection of insulator faults, in order to enable insulators that are difficult to accurately segment in infrared images due to low temperature contrast, large noise interference and complex background to be reliably segmented through prior guidance, it is necessary to first complete image acquisition and preprocessing, and then rely on the image perception prior engine to output accurate state recognition results, so as to provide clear positioning and boundary basis for subsequent segmentation algorithms.

[0015] Specifically, the image sequence of the power transmission line insulators is first acquired by using an infrared-enhanced camera on a drone to obtain the original infrared image sequence.

[0016] Furthermore, image preprocessing is carried out to eliminate inherent noise in infrared images, enhance temperature contrast, and unify pixel scale, laying a data foundation for subsequent state recognition and segmentation algorithm optimization of the image perception prior engine.

[0017] The method provided in this application embodiment includes at least Gaussian filtering for noise reduction, histogram equalization, and image pixel value normalization.

[0018] Specifically, Gaussian filtering is first used to remove noise from the image, such as noise caused by drone flight jitter or environmental electromagnetic interference, which is eliminated through the smoothing effect of Gaussian filtering. Then, histogram equalization is performed on the image to enhance the image contrast, especially in areas of temperature contrast between the insulator and the background, with significant effect on improving the contrast between locally overheated and normal areas of the insulator.

[0019] Furthermore, the image pixel values ​​are normalized to ensure they are within a uniform range, thereby avoiding the impact of image brightness differences on subsequent segmentation algorithms. For example, images taken at different times may have varying brightness due to differences in lighting or infrared device parameters. Normalization yields a standard infrared image sequence.

[0020] Furthermore, an image perception prior engine is constructed, which extracts multi-level features from standard infrared image sequences through a shared encoder, and then outputs the predicted insulator attention heatmap and the predicted boundary confidence map by dual decoders, respectively, to provide accurate region localization and boundary constraint basis for subsequent segmentation algorithms.

[0021] The method provided in this application embodiment utilizes an image perception prior engine for state recognition, outputting a predicted insulator attention heatmap and a predicted boundary confidence map, including: Construct an image perception prior engine that includes a shared encoder and dual decoders, wherein the dual decoders include an attention heatmap decoder and a boundary confidence decoder; The standard infrared image sequence is input into the image perception prior engine, and features are extracted by the shared encoder. The feature extraction results are then input into the attention heatmap decoder and the boundary confidence decoder for state recognition, and the predicted insulator attention heatmap and predicted boundary confidence map are output.

[0022] In this embodiment of the application, in order to enable insulators that are difficult to locate accurately to be reliably segmented by guiding subsequent segmentation algorithms through prior features, it is necessary to construct an image perception prior engine with dual decoding capabilities, which transforms the features of standard infrared images into explicit attention heatmaps and boundary confidence maps, so as to provide accurate regional positioning and boundary constraint basis for insulator segmentation.

[0023] As attached Figure 2 As shown, the method provided in this application embodiment constructs an image-aware prior engine including a shared encoder and dual decoders, comprising: A shared encoder is constructed using a deep convolutional neural network, wherein the shared encoder extracts multi-level feature representations from the input image through cascaded convolutional layers and pooling layers to form a shared feature map; Based on the shared feature map, two independent branch decoders are constructed in parallel. The first branch decoder is an attention heatmap decoder, which reconstructs spatial details through upsampling and feature fusion operations and outputs a predicted insulator attention heatmap. The second branch decoder is a boundary confidence decoder, which processes the shared features through an edge optimization module and outputs a predicted boundary confidence map. The shared encoder, attention heatmap decoder, and boundary confidence decoder are trained using an insulator image dataset with pixel-level annotations. The prediction accuracy of the attention heatmap and boundary confidence map is simultaneously optimized through a multi-task loss function until the loss function converges, thereby generating an image perception prior engine.

[0024] Specifically, a shared encoder is first constructed using a deep convolutional neural network. This network architecture contains multiple cascaded convolutional and pooling layers. Through a progressive feature extraction method, multi-level feature representations are mined from the input standard infrared image, ultimately forming a shared feature map that covers both local details and global structure of the image.

[0025] The convolutional layer uses a 3×3 kernel and a ReLU activation function to perform local correlation operations on image pixels. For example, for an infrared image containing insulators, the first convolutional layer can capture low-level features such as the fine textures and gradient information of the edge contours formed by temperature differences on the insulator surface.

[0026] Furthermore, the convolutional layer further integrates feature information from adjacent regions based on the low-level features, gradually constructing the local morphological features of the insulator. The pooling layer uses a 2×2 max pooling operation, which reduces the feature map resolution and computational load while preserving key features. At the same time, it extracts high-level features such as the overall morphology of the insulator and its spatial relationship with the background, ensuring that the shared feature map can reflect both detailed information and global structural relationships.

[0027] Furthermore, based on the shared feature map, two independent branch decoders are constructed in parallel. The two decoders share the underlying shared feature map and achieve different output targets through differentiated network structures.

[0028] The first branch decoder is an attention heatmap decoder. After obtaining the insulator's feature information from the shared feature map, it first restores the low-resolution feature map to the same size as the original image through an upsampling operation. Simultaneously, after upsampling at each layer, it performs feature fusion with the high-resolution feature map of the corresponding layer in the shared encoder to supplement spatial detail information and avoid detail loss during the upsampling process.

[0029] After multiple rounds of upsampling and feature fusion, a single-channel predicted insulator attention heatmap is finally output. The pixel value in the heatmap represents the confidence level that the area is an insulator. For example, in the attention heatmap corresponding to an infrared image containing a transmission line, the pixel value of the area where the insulator string is located is concentrated between 0.8 and 1.0, while the pixel value of the tower area in the background is about 0.3-0.5, and the pixel value of the sky area is less than 0.2. The core area of ​​the insulator can be clearly located by the difference in pixel value.

[0030] Furthermore, the second branch decoder is a boundary confidence decoder. After receiving the shared feature map, this boundary confidence decoder first processes the features through an edge optimization module. This edge optimization module contains a set of 5×5 convolutional kernels specifically designed to capture edge gradient features in the image, enhancing the boundary information between the insulator and the background.

[0031] Furthermore, by upsampling and feature fusion operations, spatial resolution is restored, and combined with the Sigmoid activation function, a single-channel prediction boundary confidence map is output. The higher the pixel value in the map, the greater the probability that the location is the boundary of the insulator. For example, at the edge contour of an insulator skirt, the pixel value of the confidence map can reach 0.95, while the pixel value in the internal region of the insulator is about 0.5-0.6, and the pixel value in the background region is less than 0.1. Through the continuous distribution of high confidence pixels, the edge contour of the insulator can be accurately delineated, and its boundary range with the surrounding environment can be clearly defined.

[0032] Furthermore, the shared encoder, attention heatmap decoder, and boundary confidence decoder were jointly trained using an insulator image dataset with pixel-level annotations to ensure that the parameters of the three network modules were optimized collaboratively.

[0033] The insulator image dataset contains 5,000 infrared images of insulators in different scenarios. Each image is labeled with two types of pixel-level labels: one is the attention heatmap label, and the other is the boundary confidence map label.

[0034] During training, a multi-task loss function is used to calculate the error between the prediction result and the label. This loss function consists of two parts: one is the cross-entropy loss of the attention heatmap, which measures the pixel-level difference between the predicted heatmap and the label heatmap; the other is the Dice loss of the boundary confidence map, which optimizes the prediction accuracy of edge regions and reduces the impact of class imbalance. The two parts of the loss are added together with a 1:1 weight to obtain the total loss.

[0035] During training, the Adam optimizer is used, with the learning rate initially set to 0.001. Every 10 training rounds, the learning rate is reduced to 0.5. The network parameters are continuously adjusted through iterative training until the total loss function remains stable for 5 consecutive rounds. At this point, the model is considered to have converged, training is stopped, and an image perception prior engine that can be directly used for state recognition is generated.

[0036] Through the above steps, an image perception prior engine that can accurately output insulator attention heatmaps and boundary confidence maps is constructed, providing key prior information for subsequent insulator image segmentation, ensuring the accuracy of the segmentation algorithm in low-contrast infrared images and complex backgrounds, and meeting the high-precision segmentation requirements of insulator fault detection.

[0037] Furthermore, after inputting the standard infrared image sequence into the image perception prior engine, the trained shared encoder first performs automatic feature extraction on each frame of the sequence.

[0038] Taking a standard infrared image containing 5 insulators in series as an example, the first layer 3×3 convolution kernel of the shared encoder will first perform local correlation operations on the image pixels to capture low-level features such as point-like temperature textures formed by local heating on the surface of the insulator, gray-scale gradients between the edge of the skirt and the air background.

[0039] Furthermore, the convolutional layer integrates feature information from adjacent regions based on the aforementioned underlying features. For example, it combines the curved edge features of adjacent skirts with temperature texture features to gradually construct local morphological features such as the semi-circular shape and string-like gaps of the insulator skirts.

[0040] Meanwhile, after each set of convolutional layers, the 2×2 max pooling layer downsamples the feature map. While preserving key features such as the overall arrangement direction of the insulators, the resolution of the feature map is gradually reduced from the original 512×512 to 32×32 to reduce computational resource consumption. Finally, a shared feature map covering details and global structure is generated for the frame image.

[0041] After feature extraction is completed, the shared encoder will synchronously transmit the shared feature map of each frame to the attention heatmap decoder and the boundary confidence decoder to start dual-path state recognition.

[0042] For the attention-based heatmap decoder, after receiving the shared feature map, it performs an upsampling operation using bilinear interpolation. Specifically, the 32×32 shared feature map is first upsampled to 64×64, then to 128×128, until it is restored to its original size of 512×512. After each upsampling, it is fused with the high-resolution feature map of the corresponding layer in the shared encoder. For example, the 128×128 upsampled feature map is element-wise added to the skirt gap texture feature map output from the 128×128 resolution convolutional layer in the shared encoder to supplement the subtle structural information lost during upsampling.

[0043] Furthermore, after three rounds of upsampling and fusion, the decoder outputs a single-channel attention heatmap. In this heatmap, the pixel values ​​of the five insulator regions are concentrated between 0.83 and 0.96, the pixel values ​​of the tower regions in the background are approximately 0.28 to 0.42, and the pixel values ​​of the sky regions are below 0.16. The core range of the insulator can be directly locked by the difference in pixel values.

[0044] Meanwhile, after receiving the shared feature map, the boundary confidence decoder is first processed by the built-in 5×5 convolutional kernel edge optimization module. This module is used to capture edge gradient features, increasing the gradient value between the edge of the insulator skirt and the background from the original 15 to 27, thereby enhancing the boundary discrimination.

[0045] Furthermore, upsampling and feature fusion are performed to restore the feature map resolution. Finally, the output value is mapped to the 0-1 interval through the Sigmoid activation function to generate the boundary confidence map.

[0046] Taking an insulator image with slight rain obscuring it as an example, in the corresponding boundary confidence map, the pixel value of the outer edge of the insulator skirt reaches 0.90-0.94, and the pixel value of the inner edge is 0.76-0.83. Even if the rain obscuring causes local boundary blurring, the pixel value in this area can still be maintained at 0.62-0.68, while the pixel value of the background area is lower than 0.08. Through the continuous contour formed by high confidence pixels, the complete boundary of the insulator can be accurately delineated, and the range of the insulator, rainwater, and tower can be clearly distinguished.

[0047] Through the above application process, each frame of standard infrared image can quickly output the corresponding predicted insulator attention heatmap and predicted boundary confidence map, providing clear region localization and boundary constraints for subsequent segmentation algorithms, effectively solving the problem of blurred localization caused by low contrast and complex background of infrared images in traditional segmentation.

[0048] S120: Based on the predicted insulator attention heatmap and predicted boundary confidence map, optimize and adjust the initial threshold segmentation algorithm and the initial region growth algorithm to obtain the adapted threshold segmentation algorithm and the adapted region growth algorithm. In this embodiment of the application, in order to enable the segmentation algorithm to accurately locate the insulator region and avoid missegmentation or omission, it is necessary to rely on the prior information provided by the predicted insulator attention heatmap and the predicted boundary confidence map to optimize and adjust the key parameters and execution logic of the initial threshold segmentation algorithm and the initial region growth algorithm, so as to obtain a segmentation algorithm that adapts to the current infrared image features and ensures the accuracy and reliability of subsequent insulator image segmentation.

[0049] Specifically, the initial threshold segmentation algorithm was first optimized and adjusted. High-confidence regions were extracted from the predicted insulator attention heatmap, and the window size and weight distribution parameters for local threshold calculation in the initial threshold segmentation algorithm were corrected based on the gray-scale statistical characteristics of these high-confidence regions.

[0050] Meanwhile, during the global threshold calculation process, the weight of grayscale features in high-confidence regions is increased, making the threshold calculation more closely match the grayscale characteristics of the insulator region. In the region growing stage included in the initial threshold segmentation algorithm, a predicted boundary confidence map is introduced as a constraint. When a pixel region with a boundary confidence higher than a set threshold is detected, the similarity judgment threshold of the region growing is increased by a preset ratio.

[0051] Furthermore, based on the overall distribution characteristics of the predicted boundary confidence map, the appropriate morphological operation type is dynamically matched to optimize the initial threshold segmentation algorithm, resulting in an adapted threshold segmentation algorithm.

[0052] Furthermore, the initial region growing algorithm is optimized and adjusted. Pixels with the highest pixel values ​​are selected from the predicted insulator attention heatmap as candidate seed points, and then several pixels with the highest grayscale values ​​are selected as initial seed points.

[0053] During the region growing process, the prediction boundary confidence map is referenced in real time. When a region with a pixel value greater than the set value is encountered, the similarity threshold for region growing is increased. In each iteration step, the corresponding value of the pixel to be grown in the prediction boundary confidence map is detected. If the value exceeds the set standard, the expansion in the corresponding growth direction is terminated immediately.

[0054] Meanwhile, based on the average pixel value of the predicted boundary confidence map, morphological operation parameters are adaptively selected to optimize the initial region growing algorithm and obtain a suitable region growing algorithm.

[0055] This step involves targeted optimization of two initial segmentation algorithms, enabling the resulting adaptive algorithm to fully utilize prior information to avoid the inherent defects of infrared images. This provides algorithmic support for the accurate segmentation of subsequent standard infrared image sequences, ensuring that the segmentation results meet the requirements of insulator fault detection for region extraction accuracy.

[0056] Step S120 in the method provided in this application embodiment includes: An initial threshold algorithm is constructed, wherein the initial threshold algorithm includes local adaptive threshold calculation, global threshold adjustment, pixel similarity region growth, and region boundary evaluation optimization; Based on the predicted insulator attention heatmap and the predicted boundary confidence map, the initial threshold segmentation algorithm is optimized and adjusted to obtain the adapted threshold segmentation algorithm. The optimization and adjustment process of the initial threshold segmentation algorithm includes: Based on the predicted insulator attention heatmap, high-confidence regions are extracted. The window size and weight distribution parameters in the local threshold calculation are corrected according to the gray-level statistical characteristics of the high-confidence regions. The weight of the gray-level features of the high-confidence regions is increased in the global threshold calculation. During the region growing process, the predicted boundary confidence map is introduced as a constraint. In pixel regions where the boundary confidence is higher than a set threshold, the similarity judgment threshold of region growing is increased by a preset percentage, wherein the preset percentage is greater than or equal to 10% and less than or equal to 30%. The morphological operation type is dynamically matched based on the overall distribution characteristics of the predicted boundary confidence map.

[0057] An initial region growth algorithm is constructed, wherein the initial region growth algorithm includes seed point selection, growth criterion definition, region expansion execution, growth process control, and growth region processing; Based on the predicted insulator attention heatmap and predicted boundary confidence map, the initial region growing algorithm is optimized and adjusted to obtain an adapted region growing algorithm. The optimization and adjustment process of the initial region growing algorithm includes: The top 5% of pixels in terms of pixel value are selected from the predicted insulator attention heatmap as candidate seed points, and then the highest grayscale values ​​of these pixels are selected as the initial seed points for region growth. During the region growing process, when a region with a pixel value greater than 0.7 is encountered in the prediction boundary confidence map, the similarity threshold for region growing is increased from the baseline value of 0.3 to 0.5. In each iteration of region growing, the corresponding value of the current pixel to be grown in the prediction boundary confidence map is detected in real time. When the corresponding value is greater than 0.8, the expansion in the current growing direction is terminated immediately. Morphological operation parameters are adaptively selected based on the average pixel value of the predicted boundary confidence map.

[0058] In this embodiment of the application, in order to avoid the problems of missegmentation and missed segmentation that occur in traditional segmentation algorithms, it is necessary to first construct an initial threshold algorithm and an initial region growth algorithm that include multiple steps. Then, based on the prior information provided by the predicted insulator attention heatmap and the predicted boundary confidence map, the key parameters and execution logic of the two initial algorithms are specifically optimized and adjusted to obtain a segmentation algorithm that is adapted to the current infrared image features, so as to ensure the accuracy of subsequent insulator image segmentation and meet the high-precision segmentation requirements of insulator fault detection.

[0059] Specifically, an initial threshold algorithm is first constructed, which includes local adaptive threshold calculation, global threshold adjustment, pixel similarity region growth, and region boundary evaluation and optimization.

[0060] In the method provided in this application embodiment, the local adaptive threshold calculation is used to divide the input image into multiple local regions, and calculate the adaptive threshold of each region based on the gray value distribution, temperature change characteristics and texture characteristics of each local region using the local weighted average method. The global threshold adjustment is used to dynamically adjust the global threshold based on the feature statistics of each local region; The pixel similarity region growing method is used to calculate the similarity between adjacent pixels based on the grayscale difference, temperature difference and texture features between pixels. When the similarity is greater than a preset threshold, the pixel is included in the current region. The region boundary evaluation optimization is used to evaluate the quality of the region boundary after each region expansion, determine whether to continue expansion based on the similarity of the boundary pixels, smooth the expanded region boundary, and use morphological operations to enhance the connectivity and integrity of the region.

[0061] Specifically, local adaptive threshold calculation is first performed. Taking a standard infrared image containing four insulators as an example, the local adaptive threshold calculation will divide the image into multiple local regions, such as a 100×100 pixel local region centered on each insulator.

[0062] For each local region, an adaptive threshold is calculated using a local weighted average method based on its internal grayscale distribution, temperature variation characteristics, and texture features. The center pixel of the region is used as the weight core, with pixels closer to the center having higher weights. For example, the center pixel has a weight of 0.8, and pixels 5 pixels away have a weight of 0.4. The adaptive threshold for this local region is calculated through a weighted average; for example, the adaptive threshold calculated for a certain local region is 105.

[0063] Furthermore, a global threshold adjustment is performed, dynamically adjusting the global threshold based on the statistical results of the features of each local region. Taking this frame image as an example, the adaptive thresholds of all local regions are statistically analyzed. If the adaptive thresholds of most insulator local regions are concentrated between 100 and 110, while the adaptive thresholds of background local regions are concentrated between 80 and 90, then the global threshold is adjusted to 95 to distinguish insulators from background regions at the global level.

[0064] Furthermore, pixel similarity region growing is performed, calculating the similarity of adjacent pixels based on grayscale differences, temperature differences, and texture features between pixels. For example, for a certain pixel and its adjacent pixels, the grayscale difference, temperature difference, and texture feature similarity between the two are calculated. These differences are then weighted according to a preset weight to calculate a comprehensive similarity. If the comprehensive similarity is greater than a preset threshold, the adjacent pixels are included in the current region.

[0065] After each region expansion, a region boundary evaluation and optimization is performed. The quality of the region boundary is evaluated, such as by calculating the average similarity of boundary pixels. If the average similarity of boundary pixels of a certain region is 0.6, which is lower than the set threshold of 0.7, then expansion in that direction is stopped.

[0066] Simultaneously, the expanded region boundaries are smoothed using Gaussian filtering to smooth boundary pixels and eliminate irregular edges caused by noise. Next, morphological operations are used to enhance the connectivity and integrity of the region. This involves dilating the insulator region to connect adjacent small regions into a complete insulator region, or performing erosion operations to remove isolated noise points within the insulator region, such as 1×1 pixel noise.

[0067] Furthermore, the initial threshold segmentation algorithm is optimized and adjusted based on the predicted insulator attention heatmap and the predicted boundary confidence map to obtain the adapted threshold segmentation algorithm.

[0068] For example, taking a standard infrared image containing two parallel insulators as an example, first extract the high confidence region from the predicted insulator attention heatmap, that is, the region where the pixel value corresponding to both insulators is above 0.85, and statistically analyze the gray-scale distribution characteristics of the region, such as an average gray-scale value of 118 and a gray-scale variance of 16.

[0069] Furthermore, based on the above characteristics, the window size and weight distribution parameters in the local threshold calculation are modified. The original 100×100 pixel local region window is adjusted to 90×90 pixels. At the same time, the weight distribution is adjusted so that the weight of the center pixel of the region is increased from 0.8 to 0.85, and the weight of the pixel 6 pixels away from the center is adjusted from 0.4 to 0.45.

[0070] By following the above correction steps, the adaptive threshold of the local region of the pixel is recalculated. For example, the adaptive threshold calculated after adjusting a local region is 110.

[0071] Meanwhile, in the calculation of the global threshold, the weight of the grayscale features of the high-confidence region is increased. If the original global threshold was obtained by combining the features of all local regions, the weight of the high-confidence region is now increased by 25%, so that the global threshold is more in line with the grayscale characteristics of the insulator region. For example, the global threshold is adjusted from 92 to 100.

[0072] Furthermore, during the region growing process, a predicted boundary confidence map is introduced as a constraint. When a pixel region with a boundary confidence score higher than a set threshold is detected, such as the boundary region between an insulator and a conductor, the similarity judgment threshold for region growing is increased by a preset percentage, which is between 10% and 30%, for example, by 20%. Previously, adjacent pixels with a comprehensive similarity score greater than 0.65 would be included in the current region. Now, in regions with high boundary confidence scores, only pixels with a comprehensive similarity score greater than 0.78 will be included, thereby avoiding misclassifying conductor pixels into insulator regions.

[0073] Furthermore, morphological operation types are dynamically matched based on the overall distribution characteristics of the predicted boundary confidence map. Specifically, if the boundary confidence of the insulator region is generally high in the boundary confidence map of the standard infrared image, an opening operation is used to eliminate isolated noise within the insulator region, such as 1×1 pixel noise caused by drone flight vibration.

[0074] Conversely, if the boundary confidence is low, a closing operation is used to fill the holes in the insulator region, such as a 2×2 pixel missing area caused by local temperature uniformity. Through the above optimizations, the resulting adaptive threshold segmentation algorithm can accurately segment the insulator region, providing a reliable foundation for subsequent fault detection.

[0075] Furthermore, an initial region growth algorithm is constructed, which includes seed point selection, growth criterion definition, region expansion execution, growth process control, and growth region processing.

[0076] In the method provided in this application embodiment, the seed point selection is used to automatically select the initial seed points for region growth in the preprocessed image based on gray-level extreme values ​​or significant temperature features. The growth criteria definition is used to define similarity criteria based on pixel grayscale differences, temperature differences, and texture features, and to set a similarity threshold for region growth. The region expansion is performed starting from the seed point and expanding the region to adjacent pixels according to the defined growth criteria, including pixels that meet the similarity conditions into the growth region. The growth process control is used to evaluate the boundary quality of the grown region after each region expansion, and to stop the expansion in that direction when the boundary pixel similarity is lower than a set threshold. The growth region processing is used to smooth the boundaries of the grown regions and to correct the segmentation results using morphological operations. The steps for correcting the segmentation results include using dilation operations to enhance region connectivity and erosion operations to remove isolated noise.

[0077] Specifically, seed point selection is performed first. For the preprocessed standard infrared image, pixels with extreme gray values ​​or significant temperature features are selected from the image as initial seed points.

[0078] For example, taking a standard infrared image containing three insulators as an example, after the standard infrared image is processed by Gaussian filtering for noise reduction, histogram equalization and normalization, the gray value of the core region of the insulator is significantly higher than that of the background region, with the highest gray value reaching 132. Moreover, the temperature of the insulator in this region is 6°C higher than that of the background. At this time, the pixels with the highest gray value and the most significant temperature are automatically selected as the initial seed points for region growth to ensure that the seed points can accurately fall on the core region of the insulator.

[0079] Furthermore, growth criteria are defined. Combining the characteristics of insulators in infrared images, similarity criteria are defined with pixel grayscale differences, temperature differences, and texture features as the core, and corresponding similarity thresholds are set.

[0080] For example, the allowable range of grayscale difference is set to ≤8, the allowable range of temperature difference is set to ≤2℃, and the allowable range of texture feature similarity is set to ≥0.75. The comprehensive similarity is obtained by weighted calculation, and the comprehensive similarity ≥0.6 is set as the similarity threshold for region growth to ensure that only pixels that are highly similar to the seed point features are included in the growth region.

[0081] Furthermore, a region expansion process is performed, which involves starting from the selected initial seed point and evaluating each adjacent pixel one by one according to the defined growth criteria. Taking a given initial seed point as an example, the grayscale value, temperature value, and texture features of its four adjacent pixels (up, down, left, and right) are detected sequentially, and the overall similarity with the seed point is calculated.

[0082] Specifically, if the grayscale value of an adjacent pixel differs from the seed point by 5, the temperature difference by 1℃, and the texture similarity is 0.8, the overall similarity calculation yields 0.72, which is greater than the set threshold of 0.6. Then, this pixel is included in the current growth region. If another adjacent pixel differs from the seed point by 12, the temperature difference by 3℃, and the texture similarity is 0.6, the overall similarity calculation yields 0.45, which is less than the set threshold of 0.6. Then, this pixel is not included. This point-by-point judgment achieves orderly expansion of the region.

[0083] Simultaneously, after each region expansion, growth process control is performed to assess the boundary quality of the current growth region.

[0084] Specifically, the comprehensive similarity between boundary pixels and core pixels within the region is calculated. If the average comprehensive similarity of boundary pixels in a certain direction is 0.52, which is lower than the set threshold of 0.6, it indicates that the pixels in that direction have deviated from the insulator features. The expansion in that direction should be stopped immediately to prevent the growth area from spreading excessively into the background area and to ensure that the growth area always expands around the insulator.

[0085] Finally, the growth region is processed, which involves smoothing the boundaries of the grown region by using Gaussian filtering to perform convolution operations on the boundary pixels, eliminating jagged and irregular edges caused by noise interference, and making the insulator boundaries smoother and more continuous.

[0086] Furthermore, morphological operations are used to correct the segmentation results. If isolated small areas caused by noise exist in the grown insulator region, erosion is used to remove these isolated noises. If small holes exist in the region due to uniform temperature, expansion is used to fill the holes and enhance the connectivity of the insulator region, ensuring that a complete and continuous insulator segmentation region is obtained, providing a clear and accurate image basis for subsequent fault detection.

[0087] Furthermore, the initial region growth algorithm is optimized and adjusted by first selecting candidate seed points and initial seed points from the attention heatmap of the predicted insulator.

[0088] Specifically, the values ​​of all pixels in the heat map are statistically analyzed, and the top 5% of pixels are selected as candidate seed points. These candidate seed points reflect the high-confidence areas of the insulator.

[0089] Next, several pixels with the highest grayscale values ​​are selected from the candidate seed points and determined as the initial seed points for region growth. This ensures that the initial seed points can accurately correspond to the core region of the insulator and have distinct grayscale features, providing a starting point for subsequent growth.

[0090] For example, taking the predicted insulator attention heatmap corresponding to a standard infrared image containing four series insulators as an example, the pixel value range in the heatmap is 0-1, containing a total of 262,144 pixels. After sorting the pixel values ​​from high to low, the top 13,107 pixels (i.e. 5%) are selected as candidate seed points. These candidate seed points are mainly distributed in the central area of ​​the skirts of the four insulators.

[0091] Furthermore, the grayscale values ​​of the candidate seed points are statistically analyzed, and the 20 pixels with the highest grayscale values ​​(e.g., pixels with grayscale values ​​of 130-135) are selected as initial seed points. These seed points are evenly distributed in the core position of each insulator, effectively avoiding growth shift caused by deviation of a single seed point.

[0092] Meanwhile, during the region growth process, the similarity threshold is dynamically adjusted based on the prediction boundary confidence map. Specifically, when a region with a pixel value greater than 0.7 is detected in the prediction boundary confidence map during the growth process, it indicates that the region is close to the insulator boundary, and the similarity judgment standard needs to be increased to avoid erroneous growth across the boundary. At this time, the similarity threshold for region growth is increased from the baseline value of 0.3 to 0.5.

[0093] For example, taking the predicted boundary confidence map of a frame of insulator infrared image as an example, when the region grows to near the edge of the insulator skirt, the boundary confidence value of the region reaches 0.75, exceeding the set threshold of 0.7. The similarity threshold is immediately increased from 0.3 to 0.5. Originally, the combined similarity between a neighboring pixel and the seed point was 0.38, which could be included in the growing region when the threshold was 0.3. After adjustment, because 0.38 is less than 0.5, this pixel is no longer included, effectively preventing the growing region from spreading to the towers or conductors in the background.

[0094] Meanwhile, in each iteration of the region growth process, the growth is controlled in real time by correlating the prediction boundary confidence map. For each pixel to be grown, its corresponding value in the prediction boundary confidence map is first queried. If the value is greater than 0.8, it means that the pixel is already outside the insulator boundary, and the expansion in the current growth direction must be terminated immediately to avoid the growth area exceeding the actual range of the insulator.

[0095] Finally, morphological operation parameters are adaptively selected based on the average pixel value of the predicted boundary confidence map. Specifically, the average value of all pixel values ​​in the predicted boundary confidence map is first calculated. If the average pixel value is greater than 0.6, it indicates that the insulator boundary is generally clear. A 3×3 pixel structuring element is then used for boundary refinement to further optimize the smoothness of the edge contour.

[0096] Conversely, if the average pixel value is less than or equal to 0.6, it indicates that there are blurry areas at the insulator boundary. A 5×5 pixel structuring element is used to optimize the region integrity, enhance the connectivity of the insulator region, and fill the boundary gaps.

[0097] For example, for a certain frame of predicted boundary confidence map, its average pixel value is calculated to be 0.65, which is greater than 0.6. A 3×3 pixel structuring element is used to refine the boundary of the grown insulator region. The small noise of 1-2 pixels at the boundary is eliminated by morphological opening operation, making the arc edge of the insulator skirt more regular.

[0098] In addition, for another frame with an average pixel value of 0.58, a morphological closing operation was performed using a 5×5 pixel structuring element to fill the 3-4 pixel holes in the insulator area, while connecting the broken areas caused by boundary blurring to ensure the integrity of the insulator segmentation results.

[0099] Through the above steps, the adaptive threshold segmentation algorithm and the adaptive region growing algorithm are finally obtained. Both algorithms, relying on prior information, effectively solve the problems of misclassification and omission in traditional segmentation, and can accurately extract complete insulator regions from infrared images, providing a reliable foundation for subsequent insulator fault detection.

[0100] S130: Using the adaptive threshold segmentation algorithm and the adaptive region growth algorithm, the standard infrared image sequence is segmented to output an insulator image sequence, and an early warning judgment is made on the insulator image sequence based on the adaptive early warning mechanism.

[0101] In this embodiment of the application, in order to enable insulator defects to be accurately identified and timely warned, the standard infrared image sequence is first finely segmented using the adaptive threshold segmentation algorithm and the adaptive region growth algorithm to obtain a complete insulator image sequence. Then, the insulator status in the sequence is judged in a multi-dimensional way by relying on the adaptive warning mechanism, thereby ensuring the accuracy and timeliness of insulator defect detection in power inspection.

[0102] Specifically, the adaptive threshold segmentation algorithm is first used to perform preliminary segmentation on each frame of the standard infrared image sequence.

[0103] For example, taking a standard infrared image containing three insulators as an example, the adaptive threshold segmentation algorithm first extracts high-confidence regions (regions where the pixel values ​​corresponding to the three insulators are all above 0.8) based on the predicted insulator attention heatmap, corrects the window size and weight distribution parameters of the local threshold calculation, then adjusts the similarity threshold of the region growth by combining the predicted boundary confidence map, and dynamically matches the morphological operation type to quickly separate the insulator region from the background and obtain the preliminary insulator segmentation region.

[0104] Furthermore, the preliminary segmentation results are finely optimized using an adaptive region growing algorithm. Specifically, candidate seed points with the top 5% pixel values ​​are selected from the predicted insulator attention heatmap, and several pixels with the highest grayscale values ​​are selected as initial seed points. During the region growing process, the similarity threshold is dynamically adjusted and the growth direction is controlled by combining the predicted boundary confidence map.

[0105] Finally, morphological operation parameters are selected based on the average pixel value of the boundary confidence map to refine the boundary of the insulator region, correct holes and noise within the region, and finally output a complete and accurate insulator image sequence in each frame.

[0106] Furthermore, after obtaining the insulator image sequence, an early warning judgment is made based on the adaptation early warning mechanism to promptly identify abnormal temperature, defect areas, texture abnormalities, and other conditions of the insulator, providing a basis for power inspection personnel to accurately detect and handle insulator faults.

[0107] The method provided in this application embodiment includes the following process for formulating the adaptation early warning mechanism: The appropriate detection sensitivity coefficient is obtained based on the current detection status assessment. The initial warning mechanism is optimized and adjusted based on the adaptation detection sensitivity coefficient to construct an adaptation warning mechanism. The initial warning mechanism includes at least a temperature anomaly warning threshold, a defect identification confidence threshold, a minimum effective defect area threshold, a texture anomaly detection threshold, and a warning duration frame threshold.

[0108] In this embodiment of the application, in order to enable the insulator defect early warning to accurately adapt to different detection scenarios, it is necessary to first evaluate and calculate the adaptation detection sensitivity coefficient based on the current detection status of the UAV, and then linearly adjust the various thresholds of the initial early warning mechanism according to the adaptation detection sensitivity coefficient, thereby constructing an adaptation early warning mechanism that can accurately match the current detection scenario.

[0109] In the method provided in this application embodiment, the adaptive detection sensitivity coefficient is obtained by weighted fusion calculation based on the real-time flight speed of the UAV, environmental temperature and humidity data and historical performance status data of the insulator.

[0110] For example, the weights for flight speed, ambient temperature and humidity, and historical performance status of insulators are set to 0.3. Specifically, the scoring rules for real-time UAV flight speed are: when the flight speed is ≤5m / s, the slower the speed, the higher the score; 5m / s corresponds to 0.5, and the score increases by 0.1 for every 1m / s decrease. The scoring rules for ambient temperature and humidity data are: ambient temperature is based on 25℃, corresponding to 0.5, and the score changes by 0.2 for every 5℃ deviation; humidity is based on 50%, corresponding to 0.5, and the score changes by 0.1 for every 15% deviation. The final ambient temperature and humidity score is the average of the temperature and humidity scores. The scoring rules for historical performance status data of insulators are: no historical defect records correspond to 1.0, sporadic defect records correspond to 0.7, and frequent defects correspond to 0.4.

[0111] Specifically, the formula for calculating the adaptation detection sensitivity coefficient can be expressed as: Adaptation detection sensitivity coefficient = Flight speed score × 0.3 + Average ambient temperature and humidity score × 0.4 + Historical performance status score × 0.3. When the real-time flight speed of the UAV is 4 m / s, the ambient temperature is 30℃, the humidity is 65%, and the historical performance status of the insulator is generally good, the current adaptation detection sensitivity coefficient is obtained by multiplying each score by its corresponding weight and summing the results: (0.6 × 0.3) + ((0.7 + 0.6) / 2 × 0.4) + (0.7 × 0.3) = 0.18 + 0.26 + 0.21 = 0.65.

[0112] Furthermore, the initial early warning mechanism is optimized and adjusted based on the adaptive detection sensitivity coefficient. Specifically, the temperature anomaly early warning threshold is first linearly adjusted, following the rule that for every 0.1 increase in the sensitivity coefficient, the temperature early warning threshold decreases by 1.5℃.

[0113] For example, if the initial temperature anomaly warning threshold is 85℃, when the adaptation detection sensitivity coefficient is 0.6, it increases by 0.6 compared to coefficient 0 (initial state). Then the temperature warning threshold decreases by 1.5℃ × (0.6 / 0.1) = 9℃, and the adjusted temperature anomaly warning threshold is 85℃ - 9℃ = 76℃.

[0114] Furthermore, the defect identification confidence threshold is adjusted according to the rule that for every 0.1 increase in the sensitivity coefficient, the defect identification confidence threshold decreases by 0.08. Assuming the initial defect identification confidence threshold is 0.8, when the adaptation detection sensitivity coefficient is 0.6, the defect identification confidence threshold decreases by 0.08 × (0.6 / 0.1) = 0.48, and the adjusted defect identification confidence threshold is 0.8 - 0.48 = 0.32.

[0115] Furthermore, the minimum effective defect area threshold is dynamically set, following the rule that the area threshold decreases by 15% for every 0.1 increase in the sensitivity coefficient. If the initial minimum effective defect area threshold is 100 pixels, when the adaptation detection sensitivity coefficient is 0.6, the area threshold decreases by 15% × (0.6 / 0.1) = 90%, and the adjusted minimum effective defect area threshold is 100 pixels × (1-90%) = 10 pixels.

[0116] Furthermore, the texture anomaly detection threshold is adjusted according to the rule that for every 0.1 increase in the sensitivity coefficient, the texture anomaly detection threshold decreases by 0.1. For example, if the initial texture anomaly detection threshold is 1.0, when the adaptation detection sensitivity coefficient is 0.6, the threshold decreases by 0.1 × (0.6 / 0.1) = 0.6, and the adjusted texture anomaly detection threshold is 1.0 - 0.6 = 0.4.

[0117] Finally, a warning duration frame threshold is set, following the rule that for every 0.1 increase in the sensitivity coefficient, the warning duration frame threshold decreases by 2 frames. Assuming the initial warning duration frame threshold is 10 frames, when the adaptation detection sensitivity coefficient is 0.6, the threshold decreases by 2 frames × (0.6 / 0.1) = 12 frames. The adjusted warning duration frame threshold is 10 - 12 = -2 frames (in practice, the threshold is not less than 1 frame, so we take 1 frame).

[0118] Furthermore, after obtaining the adaptation early warning mechanism, the insulator image sequence is judged based on the adaptation early warning mechanism to identify insulator defects in a timely manner, providing a basis for power inspection personnel to handle faults.

[0119] Specifically, taking a sequence containing 10 frames of insulator images as an example, each frame is first segmented using the adaptive threshold segmentation algorithm and the adaptive region growing algorithm to obtain the complete insulator region in each frame.

[0120] Furthermore, the warning judgment is carried out frame by frame. For the third frame image, the temperature of its insulator area is detected. If the measured local temperature reaches 78℃, which is higher than the adjusted temperature anomaly warning threshold of 76℃, a temperature anomaly warning is triggered.

[0121] Further, the defect identification confidence score for this region is calculated. If the defect identification confidence score is 0.31, which is lower than the adjusted threshold of 0.32, it is determined to be a defective region. Then, the area of ​​the defective region is calculated. If the area is 12 pixels, which is greater than the adjusted threshold of 10 pixels, it is confirmed as a valid defect.

[0122] Furthermore, the degree of insulator texture abnormality is analyzed. If the texture abnormality detection value is 0.39, which is lower than the adjusted threshold of 0.4, it is determined to be a texture abnormality. At the same time, the number of consecutive frames in the sequence where the abnormality occurs is monitored. If the abnormality occurs for one consecutive frame starting from the 3rd frame, an early warning is immediately triggered.

[0123] By comprehensively judging the temperature, defect confidence, area, texture, and number of consecutive frames of each insulator image through the above steps, accurate early warning of insulator image sequences is achieved, enabling power inspection personnel to promptly detect potential defects in insulators and ensure the safe and stable operation of the power system.

[0124] The embodiments of this application, through the specific implementation methods described above, achieve the following technical effects: This application proposes an insulator image segmentation method based on UAV infrared enhanced images. First, an infrared image sequence containing insulators is acquired using an infrared enhanced camera mounted on a UAV. This sequence is then preprocessed with Gaussian filtering for noise reduction, histogram equalization, and normalization to obtain a standard infrared image sequence. Next, an image perception prior engine containing a shared encoder and dual decoders is constructed to extract features and identify the state of the standard infrared image sequence, outputting a predicted insulator attention heatmap and a predicted boundary confidence map. Based on these two prior maps, the initial threshold algorithm (including local adaptive threshold calculation) and the initial region growing algorithm (including seed point selection) are optimized to obtain an adaptive threshold segmentation algorithm and an adaptive region growing algorithm. Then, by combining UAV flight speed, ambient temperature and humidity, and historical insulator performance data, a weighted fusion calculation of the adaptive detection sensitivity coefficient is performed. Based on this, the temperature and defect confidence thresholds of the initial warning mechanism are optimized to construct an adaptive warning mechanism. Finally, the adaptive segmentation algorithm is used to segment the standard infrared image sequence to obtain an insulator image sequence. Based on the adaptive warning mechanism, multi-dimensional warning judgments are performed on the sequence to achieve accurate identification and timely warning of insulator defects.

[0125] The method provided in this application, through the technical solution of "image acquisition and preprocessing - prior image generation - segmentation algorithm optimization - early warning mechanism construction - segmentation and early warning judgment", solves the problems of low segmentation accuracy, missegmentation and omission caused by low temperature contrast of infrared images and complex background in traditional insulator image segmentation, as well as the difficulty of adapting fixed early warning thresholds to different inspection scenarios and the easy false alarms and omissions in defect identification. It improves the pertinence and timeliness of defect early warning, and finally provides reliable technical support for insulator condition monitoring and fault handling in power system UAV inspection.

[0126] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0127] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0128] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. An insulator image segmentation method based on an unmanned aerial vehicle infrared enhanced image, characterized in that, The methods include: Infrared image sequences containing insulators are acquired using an infrared-enhanced camera mounted on a drone. Image preprocessing yields a standard infrared image sequence. An image perception prior engine is used for state recognition, outputting a predicted insulator attention heatmap and a predicted boundary confidence map. Based on these maps, the initial threshold segmentation algorithm and initial region growing algorithm are optimized to obtain an adapted threshold segmentation algorithm and an adapted region growing algorithm. Using these algorithms, the standard infrared image sequence is segmented to output an insulator image sequence, and an early warning mechanism is used to determine the appropriate insulator image sequence.

2. The insulator image segmentation method based on the infrared enhanced image of the unmanned aerial vehicle according to claim 1, characterized in that, The image preprocessing steps include at least Gaussian filtering for noise reduction, histogram equalization, and image pixel value normalization.

3. The insulator image segmentation method based on the infrared enhanced image of the unmanned aerial vehicle according to claim 1, characterized in that, State recognition is performed using an image-aware prior engine, outputting a predicted insulator attention heatmap and a predicted boundary confidence map, including: Construct an image perception prior engine that includes a shared encoder and dual decoders, wherein the dual decoders include an attention heatmap decoder and a boundary confidence decoder; The standard infrared image sequence is input into the image perception prior engine, and features are extracted by the shared encoder. The feature extraction results are then input into the attention heatmap decoder and the boundary confidence decoder for state recognition, and the predicted insulator attention heatmap and predicted boundary confidence map are output.

4. The insulator image segmentation method based on the UAV infrared enhanced image according to claim 3, characterized in that, Construct an image-aware prior engine that includes a shared encoder and dual decoders, including: A shared encoder is constructed using a deep convolutional neural network, wherein the shared encoder extracts multi-level feature representations from the input image through cascaded convolutional layers and pooling layers to form a shared feature map; Based on the shared feature map, two independent branch decoders are constructed in parallel. The first branch decoder is an attention heatmap decoder, which reconstructs spatial details through upsampling and feature fusion operations and outputs a predicted insulator attention heatmap. The second branch decoder is a boundary confidence decoder, which processes the shared features through an edge optimization module and outputs a predicted boundary confidence map. The shared encoder, attention heatmap decoder, and boundary confidence decoder are trained using an insulator image dataset with pixel-level annotations. The prediction accuracy of the attention heatmap and boundary confidence map is simultaneously optimized through a multi-task loss function until the loss function converges, thereby generating an image perception prior engine.

5. The insulator image segmentation method based on the UAV infrared enhanced image according to claim 1, characterized in that, Based on the predicted insulator attention heatmap and the predicted boundary confidence map, the initial threshold segmentation algorithm is optimized and adjusted to obtain a suitable threshold segmentation algorithm, including: An initial threshold algorithm is constructed, wherein the initial threshold algorithm includes local adaptive threshold calculation, global threshold adjustment, pixel similarity region growth, and region boundary evaluation optimization; Based on the predicted insulator attention heatmap and the predicted boundary confidence map, the initial threshold segmentation algorithm is optimized and adjusted to obtain the adapted threshold segmentation algorithm. The optimization and adjustment process of the initial threshold segmentation algorithm includes: Based on the predicted insulator attention heatmap, high-confidence regions are extracted. The window size and weight distribution parameters in the local threshold calculation are corrected according to the gray-level statistical characteristics of the high-confidence regions. The weight of the gray-level features of the high-confidence regions is increased in the global threshold calculation. During the region growing process, the predicted boundary confidence map is introduced as a constraint. In pixel regions where the boundary confidence is higher than a set threshold, the similarity judgment threshold of region growing is increased by a preset percentage, wherein the preset percentage is greater than or equal to 10% and less than or equal to 30%. The morphological operation type is dynamically matched based on the overall distribution characteristics of the predicted boundary confidence map.

6. The insulator image segmentation method based on the UAV infrared enhanced image according to claim 5, characterized in that, The local adaptive threshold calculation is used to divide the input image into multiple local regions. Based on the gray value distribution, temperature change characteristics and texture characteristics of each local region, the local weighted average method is used to calculate the adaptive threshold of each region. The global threshold adjustment is used to dynamically adjust the global threshold based on the feature statistics of each local region; The pixel similarity region growing method is used to calculate the similarity between adjacent pixels based on the grayscale difference, temperature difference and texture features between pixels. When the similarity is greater than a preset threshold, the pixel is included in the current region. The region boundary evaluation optimization is used to evaluate the quality of the region boundary after each region expansion, determine whether to continue expansion based on the similarity of the boundary pixels, smooth the expanded region boundary, and use morphological operations to enhance the connectivity and integrity of the region.

7. The insulator image segmentation method based on the UAV infrared enhanced image according to claim 1, characterized in that, Based on the predicted insulator attention heatmap and predicted boundary confidence map, the initial region growing algorithm is optimized and adjusted to obtain a suitable region growing algorithm, including: An initial region growth algorithm is constructed, wherein the initial region growth algorithm includes seed point selection, growth criterion definition, region expansion execution, growth process control, and growth region processing; Based on the predicted insulator attention heatmap and predicted boundary confidence map, the initial region growing algorithm is optimized and adjusted to obtain an adapted region growing algorithm. The optimization and adjustment process of the initial region growing algorithm includes: The top 5% of pixels in terms of pixel value are selected from the predicted insulator attention heatmap as candidate seed points, and then the highest grayscale values ​​of these pixels are selected as the initial seed points for region growth. During the region growing process, when a region with a pixel value greater than 0.7 is encountered in the prediction boundary confidence map, the similarity threshold for region growing is increased from the baseline value of 0.3 to 0.

5. In each iteration of region growing, the corresponding value of the current pixel to be grown in the prediction boundary confidence map is detected in real time. When the corresponding value is greater than 0.8, the expansion in the current growing direction is terminated immediately. Morphological operation parameters are adaptively selected based on the average pixel value of the predicted boundary confidence map.

8. The insulator image segmentation method based on UAV infrared enhanced image according to claim 7, characterized in that, The seed point selection is used to automatically select initial seed points for region growth in the preprocessed image based on gray-level extreme values ​​or significant temperature features. The growth criteria definition is used to define similarity criteria based on pixel grayscale differences, temperature differences, and texture features, and to set a similarity threshold for region growth. The region expansion is performed starting from the seed point and expanding the region to adjacent pixels according to the defined growth criteria, including pixels that meet the similarity conditions into the growth region. The growth process control is used to evaluate the boundary quality of the grown region after each region expansion, and to stop the expansion in that direction when the similarity of the boundary pixels is lower than a set threshold. The growth region processing is used to smooth the boundaries of the grown regions and to correct the segmentation results using morphological operations. The steps for correcting the segmentation results include using dilation operations to enhance region connectivity and erosion operations to remove isolated noise.

9. The insulator image segmentation method based on UAV infrared enhanced image according to claim 1, characterized in that, The process of developing the adaptive early warning mechanism includes: The appropriate detection sensitivity coefficient is obtained based on the current detection status assessment. The initial warning mechanism is optimized and adjusted based on the adaptation detection sensitivity coefficient to construct an adaptation warning mechanism. The initial warning mechanism includes at least a temperature anomaly warning threshold, a defect identification confidence threshold, a minimum effective defect area threshold, a texture anomaly detection threshold, and a warning duration frame threshold.

10. The insulator image segmentation method based on UAV infrared enhanced images according to claim 1, characterized in that, Based on the real-time flight speed of the UAV, environmental temperature and humidity data, and historical performance status data of the insulators, the adaptive detection sensitivity coefficient is obtained through weighted fusion calculation.

Images