Infrared insulator defect detection method and system

Abstract

The invention discloses an infrared insulator defect detection method and system. The method comprises the following steps: acquiring an insulator infrared image of a power transmission line; segmenting the insulator infrared image by using a trained image segmentation model to obtain an insulator region image; inputting the insulator area image into a trained heating defect detection model for detection to obtain a defect detection result; in the process of training the heating defect detection model, the condition diffusion model is utilized to expand the insulator area data set, the problem that defect images in the insulator area data set are insufficient is solved, the diversity of the insulator defect images in the data set is improved, the robustness of the heating defect detection model is improved, and the heating defect detection model has generalization performance. The method can adapt to changeable scenes; according to the method, the insulator region can be segmented from the acquired infrared image of the insulator, the defects of the insulator can be accurately detected, the detection precision is high, and the method can adapt to variable field environments.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an infrared insulator defect detection method and system. Background Art

[0002] With the rapid development of power systems, the safe operation of high-voltage transmission lines has become particularly important. As a key component in transmission lines, the performance of insulators directly affects the stability and reliability of power systems. The main function of insulators is to support and insulate conductors to prevent current leakage and arcing. However, due to environmental factors, mechanical stress, material aging, etc., insulators may have defects such as cracks, dirt, discharge marks, etc. These defects may lead to a decrease in insulation performance and even cause power accidents.

[0003] In order to detect and deal with these defects in a timely manner, traditional detection methods usually rely on manual inspections, which is not only inefficient but also easily affected by human factors and environmental conditions. In recent years, with the development of drone technology and infrared thermal imaging technology, the infrared cameras carried by drones can quickly and efficiently collect infrared images of transmission line insulators, providing a new technical means for insulator defect detection. Infrared thermal imaging technology uses the difference in temperature distribution on the surface of an object to detect defects in the object, which is particularly suitable for large-scale, automated detection needs.

[0004] Although the infrared images collected by drones provide a rich data source for insulator defect detection, how to accurately segment the insulator area from these images and further detect defects remains a technical challenge. Traditional image processing methods often have difficulty in achieving satisfactory results when processing images with complex backgrounds and changing conditions. In addition, existing image segmentation algorithms, such as threshold-based methods, edge detection methods, and region growing methods, usually require a lot of parameter adjustments, are sensitive to noise and complex backgrounds, and are difficult to adapt to changing field environments. Therefore, developing an insulator image segmentation and defect detection technology based on deep learning is of great significance for improving the automation level and accuracy of detection. Summary of the invention

[0005] The object of the present invention is to provide an infrared insulator defect detection method and system, which can segment the insulator area from the collected insulator infrared image and accurately detect the defects of the insulator, have high detection accuracy, and can adapt to changing on-site environments.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides an infrared insulator defect detection method, comprising:

[0008] Obtain infrared images of insulators on transmission lines;

[0009] Using the trained image segmentation model to segment the insulator infrared image to obtain an insulator region image;

[0010] Inputting the insulator region image into a trained heating defect detection model for detection to obtain a defect detection result;

[0011] The training steps of the heating defect detection model include:

[0012] Obtain infrared image dataset of insulators of transmission lines;

[0013] Using the trained image segmentation model to segment the insulator infrared image dataset to obtain a segmented insulator image dataset;

[0014] Preprocessing the segmented insulator image dataset to obtain an insulator region image dataset;

[0015] Inputting the insulator region image dataset into the conditional diffusion model for training to obtain a trained conditional diffusion model, and obtaining an expanded insulator region image dataset based on the trained conditional diffusion model;

[0016] The insulator region image dataset and the expanded insulator region image dataset are fused to obtain a training dataset;

[0017] Based on the training data set, the heating defect detection model is trained to determine the optimal model parameters, thereby obtaining a trained heating defect detection model.

[0018] Optionally, the using a trained image segmentation model to segment the insulator infrared image dataset to obtain a segmented insulator image dataset includes:

[0019] The images in the insulator infrared image dataset are sequentially input into the trained image segmentation model to obtain the corresponding mask images;

[0020] The image is fused with the corresponding mask image to obtain a segmented insulator image dataset.

[0021] Optionally, the training process of the image segmentation model includes:

[0022] Take the pre-trained MobileNetV4-DeepLabV3+ network as the starting point and freeze the parameters of its base network layer;

[0023] The pre-trained MobileNetV4-DeepLabV3+ network is trained using the insulator infrared image dataset to obtain the MobileNetV4-DeepLabV3+ network parameters;

[0024] The attention mechanism is introduced, and the MobileNetV4-DeepLabV3+ network is retrained based on the MobileNetV4-DeepLabV3+ network parameters and the insulator infrared image dataset to determine the optimal network parameters and obtain a trained image segmentation model.

[0025] Optionally, the segmented insulator image dataset is preprocessed to obtain an insulator region image dataset, including:

[0026] The segmented insulator image dataset is standardized to obtain a standardized dataset;

[0027] Image segmentation algorithm is used to extract image data containing only insulator information from the standardized data set;

[0028] Screening the image data to obtain a normal insulator data set and an insulator defect data set;

[0029] Based on the normal insulator dataset and the defective insulator dataset, an insulator area image dataset is constructed.

[0030] Optionally, the conditional diffusion model includes a forward diffusion unit and a reverse sampling unit; the reverse sampling unit is based on a U-Net network, including a downsampling part, an intermediate layer and an upsampling part connected in sequence; the downsampling part includes 4 sequentially connected residual downsampling modules and 1 residual attention module; the intermediate layer includes 2 residual attention modules; the upsampling part includes 4 sequentially connected residual upsampling modules and 1 residual attention module; wherein there is a jump connection between the residual downsampling module and the residual upsampling module; the residual upsampling module and the residual downsampling module are both composed of a residual attention module and a two-dimensional convolutional layer; the residual attention module includes a first block module, a noise addition module, a second block module and a self-attention module connected in sequence; the first block module and the second block module are both composed of a normalization layer, a Swish activation function layer, a Dropout layer and a two-dimensional convolutional layer connected in sequence.

[0031] Optionally, the step of inputting the insulator region image dataset into a conditional diffusion model for training to obtain a trained conditional diffusion model includes:

[0032] The insulator region image dataset is input into the forward diffusion unit for noise processing to obtain a noise sequence; the time step t and the two forms of labels, normal and defective, are input into the reverse sampling unit as conditions, and the noise sequence is used as input to train and optimize the reverse sampling unit using diffusion loss and optimizer;

[0033] When the reverse sampling unit is trained to a more appropriate position, a new loss function is obtained based on the weighting of the original diffusion loss and perceptual loss;

[0034] The reverse sampling unit is continued to be trained based on the new loss function to obtain the trained conditional diffusion model.

[0035] Optionally, the new loss function is expressed as:

[0036]

[0037] in, Indicates new losses; Indicates perceived loss; represents diffusion loss; Represents a dynamic weight parameter between 0 and 0.5.

[0038] Optionally, the expanded insulator region image dataset obtained based on the trained conditional diffusion model includes:

[0039] Taking random normally distributed noise as the starting point, and based on the forward diffusion time step t and the two forms of labels, normal and defective, in the insulator area image dataset, the content of the noise sequence corresponding to the time step t is restored to generate an expanded insulator area image dataset.

[0040] Optionally, the heating defect detection model is trained based on the training data set to determine the optimal model parameters, thereby obtaining a trained heating defect detection model, including:

[0041] Taking the training data set as input, training is performed using a pre-trained YOLOv8 model to obtain optimal model parameters;

[0042] Based on the optimal model parameters, a trained heating defect detection model is determined.

[0043] In a second aspect, the present invention provides a system, comprising a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the method described in the first aspect.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention provides an infrared insulator defect detection method and system. The method can segment an insulator area from a collected insulator infrared image by using a trained heating defect detection model, and accurately detect insulator defects based on the trained heating defect detection model. The method has high detection accuracy and can adapt to changing on-site environments.

[0046] The accuracy and robustness of the detection neural network model are closely related to the breadth and depth of its training data. However, it is very difficult to obtain a large number of training samples of infrared insulator defects. In view of this, the present invention chooses to use the conditional diffusion model to overcome the problem of insufficient insulator area data set in the process of training the heating defect detection model, and improves the diversity of insulator defect images in the data set. Therefore, the heating defect detection model trained based on the expanded data set has improved robustness and more generalized performance, and can adapt to changing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Shown is a schematic flow chart of a defect detection method in an embodiment of the present invention;

[0048] Figure 2 The figure is a schematic diagram of the training process of the heating defect detection model in one embodiment of the present invention;

[0049] Figure 3 FIG. 1 is a schematic diagram of a process for acquiring an expanded insulator region image data set in an embodiment of the present invention;

[0050] Figure 4 Shown is a schematic diagram of the network structure of a conditional diffusion model in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0052] Example 1

[0053] like Figure 1 As shown, the embodiment of the present invention introduces an infrared insulator defect detection method, comprising the following steps:

[0054] S1: Acquire infrared images of insulators of transmission lines;

[0055] S2: using the trained image segmentation model to segment the insulator infrared image to obtain an insulator region image;

[0056] S3: inputting the insulator region image into a trained heating defect detection model for detection to obtain a defect detection result;

[0057] Specifically, the infrared insulator defect detection method provided by the present invention can accurately detect defects of insulators based on a trained heating defect detection model, has high detection accuracy, and can adapt to changing on-site environments.

[0058] In this embodiment, if Figure 2 As shown in FIG. 1 , the training steps of the heating defect detection model specifically include:

[0059] S01: Obtain infrared image dataset of insulators of transmission lines;

[0060] S02: using the trained image segmentation model to segment the insulator infrared image dataset to obtain a segmented insulator image dataset;

[0061] S03: preprocessing the segmented insulator image dataset to obtain an insulator region image dataset;

[0062] S04: inputting the insulator region image dataset into the conditional diffusion model for training to obtain a trained conditional diffusion model, and obtaining an expanded insulator region image dataset based on the trained conditional diffusion model;

[0063] S05: fusing the insulator region image dataset and the expanded insulator region image dataset to obtain a training dataset;

[0064] S06: Based on the training data set, the heating defect detection model is trained to determine the optimal model parameters, thereby obtaining a trained heating defect detection model.

[0065] In this embodiment, the infrared image data set of the insulator of the transmission line in step S01 is collected by an infrared camera carried by the power inspection drone; in addition, the infrared image of the insulator of the transmission line is also collected by the infrared camera during the defect detection process;

[0066] In this embodiment, in step S02, the trained image segmentation model is used to segment the insulator infrared image dataset to obtain a segmented insulator image dataset, which specifically includes:

[0067] S021: input the images in the insulator infrared image dataset into the trained image segmentation model in sequence to obtain the corresponding mask image;

[0068] S022: Fusing the image with the corresponding mask image to obtain a fused image;

[0069] S023: A dataset based on fused images, that is, a segmented insulator image dataset.

[0070] Specifically, Figure 3 As shown, Figure 3 The 01 data set is the infrared image data set of the insulator in step S02, and the 02 data set is the segmented insulator image data set in step S02;

[0071] Specifically, the training process of the image segmentation model includes:

[0072] Use the pre-trained MobileNetV4-DeepLabV3+ network as a starting point and freeze the parameters of its base network layers to preserve the learned features;

[0073] On the basis of the frozen basic network layer, the pre-trained MobileNetV4-DeepLabV3+ network is trained using the insulator infrared image dataset, that is, the unfrozen layers in the network are trained to obtain the MobileNetV4-DeepLabV3+ network parameters so that the model can better adapt to new image features, accelerate the convergence speed of the model and improve the segmentation performance;

[0074] In order to further improve the segmentation accuracy of the model, the attention mechanism is introduced, and the MobileNetV4-DeepLabV3+ network is retrained based on the MobileNetV4-DeepLabV3+ network parameters and the insulator infrared image dataset, so that the attention mechanism can learn the key areas in the image, achieve optimized segmentation effects, and improve the model's ability to identify insulator defects;

[0075] Based on the above retraining results, the optimal network parameters are determined to obtain a trained image segmentation model.

[0076] In this embodiment, in step S03, the segmented insulator image dataset is preprocessed to obtain an insulator region image dataset, which specifically includes:

[0077] The segmented insulator image data set is standardized to obtain a standardized data set; wherein the standardization step is to crop the image in the segmented insulator image data set to 256×256 pixels and completely retain the size containing the insulator part data information;

[0078] Image segmentation algorithm is used to extract image data containing only insulator information from the standardized data set;

[0079] Screening the image data to obtain a normal insulator data set and an insulator defect data set;

[0080] Based on the normal insulator dataset and the defective insulator dataset, an insulator area image dataset is constructed.

[0081] In this embodiment, the conditional diffusion model includes a forward diffusion unit and a reverse sampling unit; the reverse sampling unit is based on a U-Net network, including a downsampling part, an intermediate layer and an upsampling part connected in sequence; the downsampling part includes 4 residual downsampling modules and 1 residual attention module connected in sequence; the intermediate layer includes 2 residual attention modules; the upsampling part includes 4 residual upsampling modules and 1 residual attention module connected in sequence; wherein there is a jump connection between the residual downsampling module and the residual upsampling module; the residual upsampling module and the residual downsampling module are both composed of a residual attention module and a two-dimensional convolutional layer; the residual attention module includes a first block module, a noise addition module, a second block module and a self-attention module connected in sequence; the first block module and the second block module are both composed of a normalization layer, a Swish activation function layer, a Dropout layer and a two-dimensional convolutional layer connected in sequence.

[0082] Specifically, Figure 4 As shown, in the forward diffusion process, the image Together with the time series (time step) t, the encoder part (NLM, noise-level-MLP) in the diffusion network, that is, the forward diffusion unit, performs an overall encoding operation on the input content to obtain the diffusion input Z, which is embedded in the latent space for diffusion. This approach can better preserve the internal information of the image and reduce the computational complexity of the model, which is conducive to the rapid convergence of the model.

[0083] In the reverse sampling process, the downsampling part first extracts features of different dimensions in turn by four residual downsampling modules, and then integrates the information by one residual attention module (ResnetBlockWithAttention). The middle layer is composed of two residual attention modules. The upsampling part integrates features of different dimensions by four residual upsampling modules, and integrates the information by one residual attention module. It is worth noting that there is a jump connection from the downsampling module to the upsampling module, which can realize the fusion of feature maps of the same dimension to retain more spatial information and improve the accuracy of feature extraction. Among them, the residual downsampling module and the residual upsampling module are both composed of a residual attention module and a two-dimensional convolutional layer (Conv2d); the residual attention module is composed of two blocks, a noise module (FeatureWise) and a self-attention module; the block is composed of group normalization (Group Norm), a Swish activation function layer, a Dropout layer and a two-dimensional convolutional layer (Conv2d);

[0084] The output of the diffusion model is In order to restore it from the latent space to the pixel space, it needs to go through a decoder D that is the opposite of the encoder. Perform decoding operation to obtain the final output image .

[0085] In this embodiment, in step S04, the insulator region image dataset is input into the conditional diffusion model for training to obtain a trained conditional diffusion model, which specifically includes:

[0086] S041: Input the insulator region image data set into the forward diffusion unit for noise processing to obtain a noise sequence; input the time step t and the two forms of labels, normal and defective, into the reverse sampling unit as conditions, and use the noise sequence as input to train and optimize the reverse sampling unit using diffusion loss and optimizer;

[0087] Among them, diffusion loss By calculating The corresponding time t in the forward noise process The difference between for arrive The product of, T is the total number of diffusion steps, is the noise value at each step; diffusion loss .

[0088] Specifically, Figure 3 As shown in FIG. 1 , Diffusion represents the processing step of forward diffusion, that is, the process of adding noise. Specifically, the processing step of forward diffusion is performed using the forward diffusion unit in the conditional diffusion model. The image in the insulator area image data set is fused with the random standard Gaussian distribution noise to simulate the diffusion process, so that the image is gradually transformed into a noise state close to the standard normal distribution, forming a complete forward adding noise process.

[0089] Specifically, in the forward noise addition process, the output form of each noise addition satisfies ,in, The image of the previous stage, is a noise that satisfies the standard normal distribution; the noise addition process is achieved by adjusting the weight of the noise, and the weight satisfies ,in The value of is a set of parameters that are set artificially and increase linearly with the input time step t from 0.0001 to 0.02. The corresponding Parameter value, to achieve a gradual increase in noise with time step t.

[0090] S042: When the reverse sampling unit is trained to a more appropriate position, a new loss function is obtained based on the weighting of the original diffusion loss and the perception loss;

[0091] Specifically, the new loss function is expressed as:

[0092]

[0093] in, Indicates new losses; Indicates perceived loss; represents diffusion loss; Represents a dynamic weight parameter between 0 and 0.5.

[0094] S043: Continue to train the reverse sampling unit based on the new loss function to obtain a trained conditional diffusion model.

[0095] Specifically, at the beginning of enabling the new loss function, diffusion loss is mainly used to guide the model to generate clearer images. As the model performance improves, the weight of perceptual loss is gradually increased to ensure that the generated image is more similar to the target image in visual features. Through this combination, the generative model can generate high-quality images while ensuring the visual similarity of the image to the target image.

[0096] In this embodiment, the expanded insulator region image dataset is obtained based on the trained conditional diffusion model in step S04, including:

[0097] Taking random normally distributed noise as the starting point, and based on the forward diffusion time step t and the two forms of labels, normal and defective, in the insulator area image dataset, the content of the noise sequence corresponding to the time step t is restored to generate an expanded insulator area image dataset.

[0098] Specifically, Figure 3 As shown, Sample represents the process of image restoration generation, which is performed using the reverse sampling unit in the conditional diffusion model, including:

[0099] The diffusion step code t of the insulator image and the two forms of labels label, normal and defective, are input into the conditional diffusion model as conditions. The noise sequence after forward diffusion is used as input, and the reverse sampling unit is used to predict the noise and calculation formula of each round to complete the reverse diffusion process, that is, the image generation process, so as to obtain the expanded insulator area image dataset, that is, Figure 3 The 03 dataset in .

[0100] In this embodiment, in step S06, the heating defect detection model is trained based on the training data set to determine the optimal model parameters, thereby obtaining a trained heating defect detection model, including:

[0101] The training data set is used as input, and the pre-trained YOLOv8 model is used for training to obtain the optimal model parameters; wherein the YOLOv8 model is an existing detection model, and is not described in detail here because it is not an innovation of the present invention;

[0102] Based on the optimal model parameters, a trained heating defect detection model is determined.

[0103] Example 2

[0104] An embodiment of the present invention introduces a system, including a storage medium and a processor; the storage medium is used to store instructions; the processor is used to operate according to the instructions to execute the method described in embodiment 1.

[0105] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0106] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0108] Example 3

[0109] The present invention introduces in detail an infrared insulator defect detection method, which aims to improve the efficiency and accuracy of insulator defect detection in high-voltage transmission lines in power systems. As a key component of transmission lines, the performance of insulators is directly related to the stability and reliability of power systems. However, due to environmental factors, mechanical stress, material aging and other reasons, insulators may have defects such as cracks, dirt, and discharge marks, which may lead to reduced insulation performance and even cause power accidents. Therefore, the development of an effective defect detection technology is crucial to ensure the safe operation of power systems.

[0110] The core of the present invention is to use deep learning technology to collect infrared images of transmission line insulators through infrared cameras carried by power inspection drones, and use the insulator image segmentation model based on MobileNetV4-DeepLabV3+ to segment the insulator target from the image, and then use the conditional diffusion model to expand the infrared insulator image data set, solving the problem of insufficient data sets in traditional detection methods. The conditional diffusion model simulates the gradual "diffusion" process of data distribution to simple noise distribution, learns the inverse process to reconstruct high-quality infrared insulator image data from noise, and thus generates an expanded insulator data set. This process includes image data preprocessing, segmentation, forward diffusion of the diffusion model, training of the conditional diffusion model, and the creation of a new loss function. Through these steps, infrared insulator image data containing a variety of different defects can be generated, greatly enriching the scale and diversity of the data set.

[0111] On the basis of data set expansion, the present invention adopts an image detection method based on the YOLOv8 model to process the generated infrared insulator image data to realize defect detection. The method adopts strategies such as transfer learning, hyperparameter adjustment, network structure optimization, and multi-task learning to ensure that the model has good learning and generalization capabilities for the expanded data set. In addition, the method can also identify other important information in the image, such as the defect type and location of the insulator, and provide richer detection results.

[0112] In summary, the present invention improves the detection efficiency of insulator defects, provides strong technical support for the safe operation of the power system, and has important practical application value and broad market prospects. With the continuous improvement of the automation and intelligence level requirements of the power system, the infrared insulator defect detection method proposed in the present invention is expected to be more widely used in the future.

[0113] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.

Claims

1. An infrared insulator defect detection method, characterized in that: include: Obtain infrared images of insulators on transmission lines; Using the trained image segmentation model to segment the insulator infrared image to obtain an insulator region image; Inputting the insulator region image into a trained heating defect detection model for detection to obtain a defect detection result; The training steps of the heating defect detection model include: Obtain infrared image dataset of insulators of transmission lines; Using the trained image segmentation model to segment the insulator infrared image dataset to obtain a segmented insulator image dataset; Preprocessing the segmented insulator image dataset to obtain an insulator region image dataset; Inputting the insulator region image dataset into the conditional diffusion model for training to obtain a trained conditional diffusion model, and obtaining an expanded insulator region image dataset based on the trained conditional diffusion model; The insulator region image dataset and the expanded insulator region image dataset are fused to obtain a training dataset; Based on the training data set, the heating defect detection model is trained to determine the optimal model parameters, thereby obtaining a trained heating defect detection model.

2. The infrared insulator defect detection method according to claim 1, characterized in that: The step of using the trained image segmentation model to segment the insulator infrared image dataset to obtain a segmented insulator image dataset includes: The images in the insulator infrared image dataset are sequentially input into the trained image segmentation model to obtain the corresponding mask images; The image is fused with the corresponding mask image to obtain a segmented insulator image dataset.

3. The infrared insulator defect detection method according to claim 2, characterized in that: The training process of the image segmentation model includes: Take the pre-trained MobileNetV4-DeepLabV3+ network as the starting point and freeze the parameters of its base network layer; The pre-trained MobileNetV4-DeepLabV3+ network is trained using the insulator infrared image dataset to obtain the MobileNetV4-DeepLabV3+ network parameters; The attention mechanism is introduced, and the MobileNetV4-DeepLabV3+ network is retrained based on the MobileNetV4-DeepLabV3+ network parameters and the insulator infrared image dataset to determine the optimal network parameters and obtain a trained image segmentation model.

4. The infrared insulator defect detection method according to claim 3, characterized in that: The preprocessing of the segmented insulator image dataset to obtain the insulator region image dataset includes: The segmented insulator image dataset is standardized to obtain a standardized dataset; Image segmentation algorithm is used to extract image data containing only insulator information from the standardized data set; Screening the image data to obtain a normal insulator data set and an insulator defect data set; Based on the normal insulator dataset and the defective insulator dataset, an insulator area image dataset is constructed.

5. The infrared insulator defect detection method according to claim 4, characterized in that: The conditional diffusion model includes a forward diffusion unit and a reverse sampling unit; the reverse sampling unit is based on a U-Net network, and includes a downsampling part, an intermediate layer and an upsampling part connected in sequence; the downsampling part includes 4 residual downsampling modules and 1 residual attention module connected in sequence; the intermediate layer includes 2 residual attention modules; the upsampling part includes 4 residual upsampling modules and 1 residual attention module connected in sequence; wherein there is a jump connection between the residual downsampling module and the residual upsampling module; the residual upsampling module and the residual downsampling module are both composed of a residual attention module and a two-dimensional convolutional layer; the residual attention module includes a first block module, a noise addition module, a second block module and a self-attention module connected in sequence; the first block module and the second block module are both composed of a normalization layer, a Swish activation function layer, a Dropout layer and a two-dimensional convolutional layer connected in sequence.

6. The infrared insulator defect detection method according to claim 5, characterized in that: The step of inputting the insulator region image dataset into the conditional diffusion model for training to obtain a trained conditional diffusion model comprises: The insulator region image dataset is input into the forward diffusion unit for noise processing to obtain a noise sequence; the time step t and the two forms of labels, normal and defective, are input into the reverse sampling unit as conditions, and the noise sequence is used as input to train and optimize the reverse sampling unit using diffusion loss and optimizer; When the reverse sampling unit is trained to a more appropriate position, a new loss function is obtained based on the weighting of the original diffusion loss and perceptual loss; The reverse sampling unit is continued to be trained based on the new loss function to obtain the trained conditional diffusion model.

7. The infrared insulator defect detection method according to claim 6, characterized in that: The new loss function is expressed as: ； in, Indicates new losses; Indicates perceived loss; represents diffusion loss; Represents a dynamic weight parameter between 0 and 0.

5.

8. The infrared insulator defect detection method according to claim 7, characterized in that: The expanded insulator region image dataset obtained based on the trained conditional diffusion model includes: Taking random normally distributed noise as the starting point, and based on the forward diffusion time step t and the two forms of labels, normal and defective, in the insulator area image dataset, the content of the noise sequence corresponding to the time step t is restored to generate an expanded insulator area image dataset.

9. The infrared insulator defect detection method according to claim 7, characterized in that: The method of training the heating defect detection model based on the training data set and determining the optimal model parameters to obtain the trained heating defect detection model includes: Taking the training data set as input, training is performed using a pre-trained YOLOv8 model to obtain optimal model parameters; Based on the optimal model parameters, a trained heating defect detection model is determined.

10. A system, characterized in that: including storage media and processors; The storage medium is used to store instructions; The processor is configured to operate according to the instructions to perform the method according to any one of claims 1-9.

Images