Improved YOLOv8-based electric power insulation equipment infrared image defect detection method, system, equipment and medium
By improving the infrared image detection method for power insulation equipment in YOLOv8, utilizing the Ghost and DCFEM modules to enhance model adaptability, and combining data augmentation technology, the problems of low detection accuracy and slow speed are solved, providing a fast and accurate defect identification tool to ensure the safety and stability of the power system.
Patent Information
- Application Number
- CN202511483816.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-17
- Publication Date
- 2026-01-13
AI Technical Summary
Existing infrared image detection methods for power insulation equipment based on deep learning suffer from low detection accuracy and slow processing speed, making it difficult to quickly and accurately identify defects and thus limiting their practicality.
An improved YOLOv8 approach was adopted, replacing the CBS layer with the Ghost module and embedding the DCFEM and SOFEM modules to construct a defect detection model. Data augmentation processing, including CutMix and color dithering, was then performed to improve the model's adaptability and detection accuracy.
It achieves both increased detection speed and guaranteed detection accuracy, enabling rapid and accurate identification of defects in power insulation equipment and ensuring the safe and stable operation of the equipment.
Smart Images
Figure CN121329922A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of infrared defect image analysis, and in particular to an improved YOLOv8-based power insulation equipment infrared image defect detection method, system, device and medium. BACKGROUND
[0002] As a core component of the power transmission system, power insulation equipment plays a decisive role in the safety and reliable power supply of the power grid. However, it is easily affected by lightning, pollution, moisture and mechanical stress in the complex natural environment for a long time, and then defects such as flashover, cracks and damage occur. If these defects are not timely processed, the insulation performance will deteriorate and even serious power accidents will occur. Although traditional manual inspection can find some obvious defects, it has the disadvantages of low efficiency, dependence on subjective factors and difficulty in detecting hidden defects. With the development of science and technology, data-driven algorithms have become a research hotspot. Single-stage target detection algorithms have fast detection speed but limited accuracy, and two-stage target detection algorithms have better accuracy but are computationally complex and time-consuming. Existing detection methods cannot balance detection accuracy and efficiency. Under this background, image acquisition and analysis have become one of the main technical means for defect detection. However, the existing defect infrared image detection based on deep learning has the problems of low detection accuracy and slow processing speed due to complex shooting background, which cannot quickly and accurately identify existing defects, resulting in low practicality. Therefore, there is an urgent need for an efficient and accurate detection algorithm to meet the actual engineering needs. SUMMARY
[0003] In view of the above existing problems, the present application is proposed.
[0004] Therefore, the present application provides an improved YOLOv8-based power insulation equipment infrared image defect detection method, which can solve the problems of low detection accuracy, slow processing speed and inability to quickly and accurately identify existing defects.
[0005] To solve the above technical problems, the present application provides the following technical solutions: an improved YOLOv8-based power insulation equipment infrared image defect detection method, comprising: obtaining power insulation equipment infrared defect images and preprocessing; modifying the backbone network of YOLOv8, replacing part of the CBS layer with the Ghost module, embedding the DCFEM into the C2f module, connecting the small target feature enhancement module SOFEM after the neck network SPPF, and constructing a defect detection model; based on the preprocessed defect image and the defect detection model, inputting the training sample into the model for training and verification to obtain the best model; inputting the preprocessed defect infrared image into the best model to obtain the detection result.
[0006] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the preprocessing includes data enhancement processing on the infrared defect image.
[0007] The defects in the image are labeled.
[0008] The labeled image is divided into a training set, a validation set and a test set.
[0009] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the modification of the backbone network of YOLOv8 includes replacing part of the CBS layer with a Ghost module and embedding a C2f module in DCFEM.
[0010] Based on the improved backbone network, part of the CBS layer is replaced by a Ghost module in the early stage of the network to reduce the number of model parameters and maintain the feature extraction capability.
[0011] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the access to the small target feature enhancement module SOFEM includes supporting feature extraction with a parallel substructure and integrating three parallel paths in OFEM to extract attention weight descriptors from grouped feature maps.
[0012] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the data enhancement processing on the infrared defect image includes preprocessing the defect infrared image through the CutMix data enhancement method, represented as,
[0013]
[0014] wherein, represents the new image after mixing, represents the label after mixing, is a binary mask matrix, W is the width, and H is the height, represents the mixing ratio, is the original input image, represents the original label, i and j are indices, represents the width and height of the randomly cropped rectangle.
[0015] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the embedding of the C2f module in DCFEM includes enhancing the adaptability to local geometric deformation and restoring feature distortion caused by the aerial angle,
[0016]
[0017] wherein, is the value of the output feature map at position , is the offset, is the modulation coefficient, k is the relative offset coordinate inside the convolution kernel, ω(k) is the weight of the original convolution kernel at position k, which enhances the adaptability to local geometric deformation, and restores the feature distortion caused by the aerial angle.
[0018] As a preferred scheme of the power insulation equipment infrared image defect detection method based on the improved YOLOv8, the OFEM integrates three parallel paths, two paths are located in the 1*1 branch, and one path is located in the 3*3 branch.
[0019] The interdependence between all channels is captured, and in the 1*1 branch, the double 1D global average pooling operation encodes the channel information along two spatial dimensions; on the contrary, the 3*3 branch utilizes stacked 3*3 kernels to capture multi-scale feature representations:
[0020]
[0021] wherein, is the value of the cth channel and the Hth position in the output height direction feature descriptor, is the value of the cth channel and the Wth position in the output width direction feature descriptor.
[0022] Another object of the present application is to provide a power insulation equipment infrared image defect detection system based on an improved YOLOv8.
[0023] As a preferred scheme of the power insulation equipment infrared image defect detection system based on the improved YOLOv8, the system comprises a collection module, a construction module, a training module and an output module.
[0024] The collection module is configured to acquire and pre-process the infrared defect images of the power insulation equipment.
[0025] The construction module is configured to modify the backbone network of YOLOv8, wherein the modification replaces part of the CBS layer with a Ghost module, embeds a DCFEM in a C2f module, and connects a small target feature enhancement module SOFEM after the neck network SPPF to construct a defect detection model.
[0026] The training module is configured to input training samples into the model for training and verification based on the pre-processed defect images and the defect detection model, so as to obtain an optimal model.
[0027] The output module is configured to input the preprocessed defect infrared image into the optimal model to obtain a detection result.
[0028] The application provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the steps of the improved YOLOv8-based power insulation equipment infrared image defect detection method when executing the computer program.
[0029] The application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the improved YOLOv8-based power insulation equipment infrared image defect detection method when executed by a processor.
[0030] The application has the beneficial effects that: the defect infrared image is acquired and preprocessed, the preprocessed image is divided into training samples, verification samples and test samples, a defect detection model is constructed based on a YOLOv8 network, the model is trained through the training samples, after the model training is completed, the defect infrared image is input into the defect detection model, and corresponding detection results are obtained, so that the detection speed is improved while the detection accuracy is ensured, and accidents caused by defects are effectively prevented, and the safe and stable operation of equipment is ensured. BRIEF DESCRIPTION OF DRAWINGS
[0031] In order to more clearly illustrate the technical solutions of the embodiments of the application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.
[0032] Figure 1 A flowchart of an improved YOLOv8-based power insulation equipment infrared image defect detection method is provided for an embodiment of the application. DETAILED DESCRIPTION
[0033] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the specific embodiments of the application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should be within the protection scope of the application.
[0034] Embodiment 1, refer to Figure 1 For the first embodiment of the application, the embodiment provides an improved YOLOv8-based power insulation equipment infrared image defect detection method, including:
[0035] S1: Obtain an infrared defect image of power insulation equipment and perform preprocessing.
[0036] S2: Modify the backbone network of YOLOv8, replace part of the CBS layer with a Ghost module, embed a DCFEM in a C2f module, connect a small target feature enhancement module SOFEM after the neck network SPPF, and construct a defect detection model.
[0037] S3: Based on the preprocessed defect image and the defect detection model, input the training sample into the model for training and verification to obtain the best model.
[0038] S4: Input the preprocessed defect infrared image into the best model to obtain a detection result.
[0039] It should be noted that through the three core improvements of the lightweight model of the Ghost module, the enhancement of the geometric deformation adaptability of the DCFEM, and the strengthening of the small target detection capability of the SOFEM, and the deep fusion with the YOLOv8 framework, the key technical problems such as "small target detection difficulty, poor deformation target recognition, and large model calculation" in the infrared defect detection of power insulation equipment are effectively solved. The final effect is to provide a detection tool that can quickly respond and accurately identify, and to provide strong technical support for the safe, stable, and intelligent operation of the power system.
[0040] Embodiment 2 is an embodiment of the present application, based on the above embodiment, provides an improved YOLOv8-based power insulation equipment infrared image defect detection method.
[0041] Further, in the present application, step S1 obtains an infrared defect image of power insulation equipment and performs preprocessing, and the specific steps include S101-S102:
[0042] S101: Perform data enhancement processing on the defect infrared image, and label the defect infrared image, divide the labeled defect infrared image into training samples, verification samples and test samples, preferably, the labeling can be realized by using the labelImg labeling tool, or other labeling tools can be used, and the labeled defect infrared image is divided into training samples, verification samples and test samples according to the ratio of 8:1:1.
[0043] S102: Preprocess the defect infrared image by CutMix data enhancement method, denoted as:
[0044]
[0045] Among them, represents the mixed new image (output), Soft Label, a binary mask matrix (0-1 matrix) with the same shape as the image size (W for width, H for height), a mixing ratio determined by the area proportion of the rectangular region, is the original input image, is the original label, i and j are indices, is the width and height of the random cropping rectangle.
[0046] In this embodiment, the pre-processing of the image is realized by the CutMix data enhancement method;
[0047] In an alternative embodiment, the pre-processing can also be realized by combining random cropping and color jittering enhancement. Specifically, random scale cropping is performed on the training image, that is, after randomly scaling the image, a region of fixed size is cropped from the image, which can simulate the shooting of the target at different distances and force the model to pay attention to the local features of the target. Secondly, the image is subjected to limited color jittering. According to the characteristics of infrared images, small random adjustments are mainly made in brightness and contrast to simulate the image changes caused by different environmental lighting and device heating differences.
[0048] In another alternative embodiment, the pre-processing can also be realized by CutOut-based occlusion simulation enhancement. Specifically, one or more rectangular regions are randomly selected on the input image, and the pixel values in the region are set to zero (or filled with other constant values). This operation simulates the scenario where the target object is partially covered by dust, leaves or other temporary obstacles, forcing the model to not rely too much on a certain local feature of the target for judgment, but to learn more discriminative global or multiple local features.
[0049] Further, in the embodiments of the present application, step S2 modifies the backbone network of YOLOv8, replaces part of the CBS layer with a Ghost module, embeds a DCFEM in a C2f module, connects a small target feature enhancement module SOFEM after the neck network SPPF, and constructs a defect detection model. The specific steps include S201-S204:
[0050] S201: Replace part of the CBS layer with a Ghost module, and embed a DCFEM in a C2f module;
[0051] Based on the improved backbone network, part of the CBS layer is replaced by a Ghost module in the early stage of the network to reduce the model parameter quantity by 10% while maintaining the feature extraction capability;
[0052] The DCFEM is embedded in the C2f module to enhance the adaptability to local geometric deformation and restore the feature distortion caused by the aerial angle.
[0053]
[0054] in, To output feature map at position The value at that location, This is the offset. ω(k) is the modulation coefficient, k is the relative offset coordinate inside the convolution kernel, and ω(k) is the weight of the original convolution kernel at position k. This enhances the adaptability to local geometric deformation and restores the feature distortion caused by the aerial shooting angle.
[0055] In an optional embodiment, enhanced adaptability to local geometric deformations can also be achieved through depthwise separable convolutions and channel attention mechanisms. Specifically, in the early stages of the network, depthwise separable convolutions are used to replace some standard convolutional layers. This structure decomposes standard convolution into two steps: depthwise convolution and pointwise convolution. It is a recognized, efficient, and lightweight fundamental approach that directly reduces parameters and computational cost. A lightweight channel attention module is embedded in the C2f module. By modeling the interdependencies between channels, the weights of each channel's features are automatically calibrated, enabling the network to focus more on key feature channels related to defects.
[0056] In another alternative embodiment, enhanced adaptability to local geometric deformations can be achieved through enhancements based on grouped convolutions and spatial pyramid pooling. Specifically, in the early stages of the network, some standard convolutional layers are replaced with grouped convolutions. By processing the input channels in independent groups, grouped convolutions effectively reduce the number of model parameters and connection density. A miniature spatial pyramid pooling layer is introduced in or after the C2f module. This layer, by using multiple pooling kernels of different sizes in parallel, can capture multi-scale contextual information in local regions. This fusion of multi-scale features helps the network understand target structures at different scales and provides robustness to scale changes and slight deformations caused by distance or angle.
[0057] S202: Replacing part of the CBS layer with the Ghost module enables the generation of redundant features through linear transformations, significantly reducing the number of parameters and computational cost while maintaining feature expressiveness. For example, the Ghost module can decompose traditional convolution into backbone convolution and lightweight transformation operations, reducing computation by approximately 50%. Furthermore, the DCFEM (Dynamic Context Enhancement Module) embedded in the C2f module dynamically adjusts channel weights and strengthens key features, improving the model's ability to perceive complex backgrounds and small targets. The combination of these two technologies allows the Backbone to be lightweight while enhancing the representational power of multi-scale features, especially significantly improving localization accuracy in dense target scenes.
[0058] It should be noted that in an optional embodiment, in the C2f module, an extrusion excitation module and a spatial attention module can also be sequentially embedded. The SE module models the channel importance through a global average pooling and a fully connected layer, dynamically calibrating the channel weight; the spatial attention module generates a spatial weight map through a simple pooling and convolution operation, highlighting important regions.
[0059] S203: Adding the SOFEM module includes: using a parallel substructure to support feature extraction, and the OFEM integrates three parallel paths to extract attention weight descriptors from grouped feature maps. Among them, two paths are in the 1x1 branch, and one is in the 3x3 branch.
[0060] The architecture aims to capture the interdependence between all channels while effectively managing computing resources. Within the 1x1 branch, double 1D global average pooling operations encode channel information along two spatial dimensions. In contrast, the 3x3 branch utilizes stacked 3x3 kernels to capture multi-scale feature representations.
[0061]
[0062] wherein, is the value of the cth channel and the Hth position in the height direction feature descriptor of the output, is the value of the cth channel and the Wth position in the width direction feature descriptor of the output.
[0063] S204: In the Neck part, SPPF (Fast Spatial Pyramid Pooling) captures different features through multi-scale pooling operations, but the traditional design is inefficient for cross-layer feature fusion. After introducing the SOFEM (Semantic Optimized Feature Enhancement Module), the module fuses feature maps of different levels through an adaptive weighting mechanism, such as nonlinear interaction of shallow detail features and deep semantic features, thereby optimizing the integration efficiency of multi-scale information. Experiments show that SOFEM can effectively improve the recall rate of small target detection, while suppressing false detection of complex backgrounds, so that the Neck part enhances the discriminability of cross-scale features while maintaining real-time performance.
[0064] In an optional embodiment, discriminating cross-scale features can also be achieved through adaptive spatial feature fusion, specifically by introducing an adaptive spatial feature fusion mechanism. This mechanism learns an adaptive weight parameter for each spatial position (i.e., each pixel point) from different levels, and then performs weighted summation on the features of these levels at the same position. ASFF also implements the core idea of "adaptive weighted fusion", allowing the network to decide whether to rely more on shallow detail information or deep semantic information according to the specific image content.
[0065] In an optional embodiment, the discriminative cross-scale feature can also be implemented by enhancing the path aggregation network, specifically, enhancing the standard path aggregation network structure. First, the bottom-up path is strengthened to ensure that the accurate positioning information at the shallow layer can be effectively transmitted to the deep layer. Second, a learnable scalar weight is introduced for each layer before the splicing or adding operation of the feature maps at different levels, and the network automatically adjusts these weights during the training process to balance the contribution of different levels of features during fusion.
[0066] Further, in the embodiment of the application, step S3 inputs the training sample into the model for training and verification based on the preprocessed defect image and the defect detection model to obtain the best model, and the specific content includes:
[0067] After training the defect model by using the training sample and adjusting the model weight and other parameters, the model is verified by using the verification sample to obtain the best model.
[0068] Further, in the embodiment of the application, step S4 inputs the preprocessed defect infrared image into the best model to obtain a detection result, and the specific steps include S401-S402:
[0069] S401: The power insulating equipment infrared image to be detected, which has been preprocessed (such as size normalization), is prepared and input into the model.
[0070] The model performs forward propagation on the input image. The improved backbone network (Ghost module and DCFEM enhanced C2f module) extracts multi-scale features, which significantly reduces the computational burden while ensuring the feature expression capability. Subsequently, the neck network (Neck integrated with SOFEM) fuses and enhances the extracted features, especially the feature discrimination ability for small targets and complex backgrounds.
[0071] The model predicts the position and category of the possible defect target in the image according to the fused high-level features at the output layer. The position information is usually represented in the form of a bounding box, and the category information corresponds to a specific defect type.
[0072] S402: The model outputs a detection result image with defect category labels and positioning bounding boxes, and can generate a structured data report containing defect position, category and confidence, etc.
[0073] Embodiment 3 is the third embodiment of the application, which is different from the first two embodiments:
[0074] The embodiment also provides an improved YOLOv8-based power insulating equipment infrared image defect detection system, which includes a collection module, a construction module, a training module and an output module.
[0075] a collection module configured to acquire an infrared defect image of the power insulation equipment and perform preprocessing;
[0076] a construction module configured to modify a backbone network of YOLOv8, the modification replacing part of CBS layers with Ghost modules, embedding DCFEM into a C2f module, and connecting a small target feature enhancement module SOFEM after a neck network SPPF to construct a defect detection model;
[0077] a training module configured to input training samples into the model for training and verification based on the preprocessed defect image and the defect detection model, and obtain an optimal model;
[0078] an output module configured to input the preprocessed defect infrared image into the optimal model to obtain a detection result.
[0079] The embodiment also provides an electronic device, including a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the power insulation equipment infrared image defect detection method based on the improved YOLOv8.
[0080] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the power insulation equipment infrared image defect detection method based on the improved YOLOv8.
[0081] The storage medium provided in the embodiment and the power insulation equipment infrared image defect detection method based on the improved YOLOv8 provided in the above embodiment belong to the same inventive concept, and the technical details not described in detail in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.
[0082] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present application.
[0083] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, and all the modifications and equivalents should be included in the scope of the claims of the present application.
Claims
1. A method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8, characterized in that: include, Acquire infrared defect images of electrical insulation equipment and perform preprocessing; Modify the YOLOv8 backbone network by replacing part of the CBS layer with the Ghost module, embedding DCFEM into the C2f module, and connecting the small target feature enhancement module SOFEM after the neck network SPPF to build a defect detection model. Based on the preprocessed defect images and the defect detection model, the training samples are input into the model for training and validation to obtain the optimal model; The preprocessed infrared image of the defect is input into the optimal model to obtain the detection result.
2. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 1, characterized in that: The preprocessing includes performing data enhancement processing on the infrared defect image; Annotate defects in the image; The labeled images are divided into training, validation, and test sets.
3. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 2, characterized in that: The modification to the YOLOv8 backbone network includes replacing part of the CBS layer with a Ghost module and embedding a C2f module into DCFEM. Based on the improved backbone network, some CBS layers are replaced by the Ghost module in the early stage of the network to reduce the number of model parameters while maintaining the feature extraction capability.
4. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 3, characterized in that: The access small target feature enhancement module SOFEM includes a parallel substructure to support feature extraction. OFEM integrates three parallel paths to extract attention weight descriptors from the grouped feature maps.
5. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 4, characterized in that: The data augmentation processing of the infrared defect image includes preprocessing the defect infrared image using the CutMix data augmentation method, as shown below. ， in, This represents the new image after blending. Indicates the mixed label, This is a binary mask matrix, where W is the width and H is the height. Indicates the mixing ratio. The original input image, This represents the original label, where i and j are indices. This represents the width and height of the randomly cropped rectangle.
6. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 5, characterized in that: The DCFEM embedded C2f module includes features that enhance adaptability to local geometric deformations and recover feature distortions caused by aerial photography angles. in, To output feature map at position The value at that location, This is the offset. ω(k) is the modulation coefficient, k is the relative offset coordinate inside the convolution kernel, and ω(k) is the weight of the original convolution kernel at position k. This enhances the adaptability to local geometric deformation and restores the feature distortion caused by the aerial shooting angle.
7. The method for detecting defects in power insulation equipment using infrared images based on an improved YOLOv8 as described in claim 6, characterized in that: The OFEM integration includes three parallel paths, two of which are located in a 1×1 branch and one in a 3×3 branch; To capture the interdependencies between all channels, the 1×1 branch uses a dual 1D global average pooling operation to encode channel information along two spatial dimensions; conversely, the 3×3 branch utilizes stacked 3×3 kernels to capture multi-scale feature representations. in, This refers to the value at position H, in the c-th channel of the output height orientation feature descriptor. This is the value of the c-th channel and the W-th position in the output width-direction feature descriptor.
8. A system for detecting infrared image defects in power insulation equipment based on an improved YOLOv8, employing the method for detecting infrared image defects in power insulation equipment based on an improved YOLOv8 as described in any one of claims 1 to 7, characterized in that, include: The module consists of a collection module, a construction module, a training module, and an output module. The collection module is used to acquire infrared defect images of power insulation equipment and perform preprocessing. The construction module is used to modify the backbone network of YOLOv8. The modification replaces part of the CBS layer with the Ghost module, embeds the DCFEM into the C2f module, and adds the small target feature enhancement module SOFEM after the neck network SPPF to build a defect detection model. The training module is used to train and validate the model by inputting training samples into the model based on the preprocessed defect image and the defect detection model, so as to obtain the best model. The output module is used to input the preprocessed defect infrared image into the optimal model to obtain the detection result.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of any one of claims 1 to 7 of the infrared image defect detection method for power insulation equipment based on improved YOLOv8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the infrared image defect detection method for power insulation equipment based on the improved YOLOv8 as described in any one of claims 1 to 7.