An image-based cable thermal fault detection method

Abstract

The present invention belongs to the technical field of cable thermal fault detection applications, and in particular, to an image-based cable thermal fault detection method. The fusion module based on improved guided filtering adopts an adaptive tolerance strategy to compensate for the transmittance of non-background areas, and the tolerance parameter is adaptively adjusted with the size of the background area, while other areas keep the original transmittance unchanged. In order to avoid the problems of distortion, image offset, perspective relationship error, etc. in the fused image, improved bilinear interpolation and homography transformation are performed to fuse the image. In order to realize the classification of cable thermal fault types and segmentation of the fault heating source range, a two-stage cable thermal fault intelligent detection model is constructed, in which the first stage is a classification model to classify the thermal fault type, and the second stage is to segment the fault heating source range. The improved Unet model combines 1D, 2D and 3D convolutions with the residual deconvolution module respectively to segment the fault heating source range.

Description

Technical Field

[0001] The invention belongs to the technical field of cable thermal fault detection applications, and in particular relates to an image-based cable thermal fault detection method. Background Art

[0002] During the operation of power equipment, various faults are inevitable, among which thermal fault is the most common type of fault. If the thermal fault point in the equipment is not detected in time, it will cause the local temperature to rise abnormally, which may easily cause the equipment to trip and power outage, bringing huge challenges to the safe and stable operation of power equipment.

[0003] In the current detection of thermal faults, the heat source is often inside the component. For the heat conduction problem of a single heat source and uniform medium in the structure, the internal temperature field change can generally be inverted based on the thermal radiation of the outer surface. However, for the complex internal heat source structure, it is difficult to determine the heat source distribution state through inversion. Current research is difficult to directly detect the internal heat source, which is not conducive to analyzing the heat distribution and tracing the heat source. In addition, the current process of thermal fault detection has the problem of low overall detection and fault warning accuracy. Summary of the invention

[0004] Aiming at the technical problems existing in cable thermal fault detection, the present invention proposes an image-based cable thermal fault detection method which has reasonable design, simple method, strong theoretical basis and can detect internal heat sources and thermal faults accurately.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: a cable thermal fault detection method based on an image, characterized in that it comprises the following steps:

[0006] S1, collecting infrared images, visible light images and X-ray transmission images of the detection position;

[0007] S2. Input the infrared image and the visible light image into a fusion module based on improved guided filtering to obtain a fused image 1. The fusion module based on improved guided filtering includes median filtering, Gaussian smoothing, improved guided filtering, edge feature map gradient, regional energy comparison and secondary morphological filtering. The improved guided filtering introduces adaptive tolerance, adopts guided filtering transmittance rough estimation, outputs refined transmittance estimation, introduces adaptive tolerance to correct transmittance, and defines tolerance parameter D:

[0008]

[0009] Among them, N I(x) represents the number of pixels corresponding to the bright area in the guidance image, C I(x) Represents the number of all pixels in the guide image, and the output corrected transmittance formula is:

[0010]

[0011] Among them, t n (x) represents the corrected transmittance of the output, I(x) represents the guide image, K represents the background area, and S represents the light value;

[0012] S3, the fused image 1 and the X-ray transmission image are subjected to improved bilinear interpolation and homography transformation to generate a fused image 2, which can be expressed as:

[0013]

[0014] Where W() represents the post-warping calculation function, V1 represents fused image 1, V2 represents fused image 2, V m represents the X-ray transmission image, x is a parameter whose value range is (0,1), γ∈N;

[0015] S4. Build a two-stage cable thermal fault intelligent detection model. The first stage is a classification model to classify the thermal fault type, and the second stage is to segment the fault heating source range. The classification model is a Resnet50 model. The ImageNet dataset is first used to pre-train the model, and then the model is migrated to thermal fault classification. A new loss function is proposed, and two methods for calculating weights are introduced. The two methods for calculating weights are linearly combined to balance the thermal fault types. The formula is:

[0016]

[0017] Where C is the number of thermal fault types, N i is the number of the i-th fault type, L D is the cross entropy damage function, is the equilibrium parameter, f(N i )、g(N i ) are two methods for calculating weights, where f(N i ) is calculated as:

[0018]

[0019] Among them, τ represents the hyperparameter, g(N i ) is:

[0020]

[0021] Among them, N i is the number of the i-th fault type, and the range of the fault heating source is segmented using the improved Unet model, wherein the improved Unet model combines 1D, 2D and 3D convolutions with the residual deconvolution module respectively to segment the range of the fault heating source;

[0022] S5. Train and test a two-stage cable thermal fault intelligent detection model to achieve thermal fault classification and segment the scope of the fault heating source.

[0023] Preferably, the thermal fault types include short circuit fault, ground fault, open circuit fault, flashover fault and mixed fault, and the mixed fault includes two or more of the short circuit fault, ground fault, open circuit fault and flashover fault.

[0024] Preferably, the secondary morphological filtering in step S2 includes filling and erosion.

[0025] Compared with the prior art, the advantages and positive effects of the present invention are:

[0026] The present invention proposes an image-based cable thermal fault detection method. The fusion module based on the improved guided filter adopts an adaptive tolerance strategy to compensate for the transmittance of the non-background area, and the tolerance parameter is adaptively adjusted with the size of the background area, while other areas keep the original transmittance unchanged. Therefore, the algorithm can not only compensate for the transmittance of the non-background area, but also ensure that the transmittance of the background area is not miscompensated or over-compensated, which better overcomes the regional distortion problem and obtains image features with better processing effect. In order to avoid the problems of distortion, image offset, perspective relationship error, etc. in the fused image, improved bilinear interpolation and homography transformation are performed to fuse the image. In order to realize the classification of cable thermal fault types and segmentation of the fault heating source range, a two-stage cable thermal fault intelligent detection model is built. The first stage is a classification model to classify the thermal fault type, and the second stage is to segment the fault heating source range. The classification model is a Resnet50 model. The ImageNet dataset is first used to pre-train the model, and then the model is migrated to thermal fault classification. A loss function is constructed to solve the category imbalance, and the improved Unet model is used to segment the range of the fault heating source. The improved Unet model combines 1D, 2D and 3D convolutions with residual deconvolution modules respectively to segment the range of the fault heating source. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 A flow chart of a cable thermal fault detection method based on an image is provided for an embodiment of the present invention;

[0029] Figure 2A schematic diagram of the structure of a fusion module based on improved guided filtering provided by an embodiment of the present invention;

[0030] Figure 3 A schematic diagram of the structure of an improved Unet model provided in an embodiment of the present invention;

[0031] Figure 4 A diagram of a residual deconvolution module provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0033] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0034] Embodiment: During the operation of power equipment, various faults are inevitable, among which thermal fault is the most common type of fault. Once the thermal fault point in the equipment is not detected in time, it will cause abnormal increase in local temperature, which may easily cause equipment tripping and power outage, bringing huge challenges to the safe and stable operation of power equipment. In the current detection of thermal faults, the heating element is often inside the component. For the heat conduction problem of a single heat source and a uniform medium in the structure, the internal temperature field change can generally be inverted based on the thermal radiation of the outer surface. However, for the construction of a complex internal heat source structure, it is difficult to determine the heat source distribution state through inversion. Current research makes it difficult to directly detect the internal heat source, which is not conducive to analyzing the heat distribution and tracing the heating element. In addition, in the current process of thermal fault detection, there is a problem of low accuracy of overall detection and fault warning. For this reason, if Figure 1 As shown, the present invention proposes an image-based cable thermal fault detection method.

[0035] Considering that infrared images have rich thermal radiation information, clear target outlines, little impact from weather, and strong anti-interference ability, they have disadvantages such as low contrast and resolution, little detail information, and poor visual effects. Visible light images have high clarity and contrast, contain rich detail information, and have good visual effects, but are easily affected by factors such as weather, lighting conditions, and distance, and have poor anti-interference ability. X-rays have high penetrating power and can penetrate many objects that are opaque to visible light and detect their interior. Therefore, infrared images, visible light images, and X-ray transmission images of the detection location are collected.

[0036] In order to obtain a better fusion effect of visible light image and infrared image, an adaptive tolerance strategy is used to compensate the transmittance of non-background area, and the tolerance parameter is adaptively adjusted with the size of the background area, while other areas keep the original transmittance unchanged. Therefore, the algorithm can not only compensate the transmittance of non-background area, but also ensure that the transmittance of background area is not miscompensated or over-compensated, which better overcomes the regional distortion problem and obtains image features with better processing effect.

[0037] So, if Figure 2 As shown, the infrared image and the visible light image are input into the fusion module based on the improved guided filtering to obtain the fused image 1. The fusion module based on the improved guided filtering includes median filtering, Gaussian smoothing, improved guided filtering, edge feature map gradient, regional energy comparison and secondary morphological filtering. The improved guided filtering introduces adaptive tolerance, adopts the guided filtering transmittance rough estimation, outputs the refined transmittance estimation, introduces adaptive tolerance to correct the transmittance, and defines the tolerance parameter D:

[0038]

[0039] Among them, N I(x) represents the number of pixels corresponding to the bright area in the guidance image, C I(x) Represents the number of all pixels in the guide image, and the output corrected transmittance formula is:

[0040]

[0041] Among them, t n (x) represents the corrected transmittance of the output, I(x) represents the guide image, K represents the background area, S represents the light value, and the secondary morphological filtering includes filling and erosion.

[0042] Considering the characteristics of X-ray transmission images, due to the distortion of the fused image, the conflict between the distribution and acquisition schemes, an erroneous fused image will appear, resulting in image offset, perspective relationship errors and other problems, affecting the final fusion quality. The bilinear interpolation method re-interpolates the embedded points to create a larger space for fusion. The fused image 1 and the X-ray transmission image are subjected to improved bilinear interpolation and homography transformation to generate the fused image 2, which can be expressed as:

[0043]

[0044] Where W() represents the post-warping calculation function, V1 represents fused image 1, V2 represents fused image 2, V m Represents the X-ray transmission image, x is a parameter whose value range is (0,1), γ∈N.

[0045] In order to realize the classification of cable thermal fault types and segmentation of the fault heating source range, a two-stage cable thermal fault intelligent detection model is built. The first stage is the classification model to classify the thermal fault type, and the second stage is to segment the fault heating source range. The classification model is the Resnet50 model. The ImageNet dataset is first used to pre-train the model, and then the model is transferred to the thermal fault for classification. The improved Unet model is used to segment the range of the fault heating source. The improved Unet model combines 1D, 2D and 3D convolutions to segment the range of the fault heating source. Considering that the ResNet50 network is a deep residual network, it has many significant advantages, which make it perform well in many deep learning tasks. First, ResNet50 effectively solves the common gradient vanishing problem of deep neural networks by introducing residual connections, allowing the number of network layers to increase significantly without reducing the training quality, so that more rich and abstract features can be extracted. Secondly, the network structure of ResNet50 has good generalization ability. It can achieve excellent performance on a variety of image recognition tasks, and also shows strong transfer learning ability in adapting to new data sets. In addition, due to its deeper network structure and residual block design, ResNet50 can improve the learning efficiency and accuracy of the model while maintaining computational efficiency, which makes it the preferred model for many complex tasks, such as image classification, object detection, and face recognition. In summary, the advantages of ResNet50 are that its deep network structure can effectively extract features, residual connections solve gradient problems, strong generalization and transfer learning capabilities, and high-precision performance while maintaining computational efficiency. These features together constitute the basis for ResNet50 to be widely used in many fields. The prediction accuracy of deep learning is based on its huge data set training. This condition often limits the use of neural networks, especially when there are fewer data samples, the effect of neural networks is poor or even overfitting occurs. Transfer learning solves problems such as data scarcity, high annotation costs, and poor model generalization ability by applying knowledge and experience in the source field to the target field, thereby accelerating the learning process and improving learning performance. First, use a network trained on a large public data set to transfer the pre-trained parameters of the internal layer of the network to a small data set, and fine-tune the parameters when training the layers without transferred parameters. The thermal fault types in the current dataset have a class imbalance problem, where the class with a large number of samples is called the majority class, and the other classes are called the minority class. Class imbalance will cause the classifier to tend to judge all samples as the majority class, and the classification effect of the minority class is poor, which seriously affects the overall classification performance. Consider adding the residual block to the residual deconvolution module. The two batch normalization layers introduced in the residual block can normalize the input of each batch, alleviate the problem of gradient vanishing after residual connection and feature fusion, and provide a smoother input to the ReLu activation function.The introduction of residual connections will alleviate the gradient vanishing problem and help the network better capture the details and complex features in the image. Therefore, the classification model is the Resnet50 model. The ImageNet dataset is first used to pre-train the model, and then the model is migrated to thermal fault classification. A new loss function is proposed, and two methods for calculating weights are introduced. The two methods for calculating weights are linearly combined to balance the types of thermal faults. The formula is:

[0046]

[0047] Where C is the number of thermal fault types, N i is the number of the i-th fault type, L D is the cross entropy damage function, is the equilibrium parameter, f(N i )、g(N i ) are two methods for calculating weights, where f(N i ) is calculated as:

[0048]

[0049] Among them, τ represents the hyperparameter, g(N i ) is:

[0050]

[0051] Among them, N i is the number of the i-th fault type, and the improved Unet model is used to segment and predict the range of harmful gases. In order to increase the network's ability to represent the appearance and movement of harmful gases. The task of the Unet model encoder is to extract features, and the decoder projects the feature semantics identified by the former into the pixel space to obtain pixel-level classification to adapt to the detection of the range of the fault heating source. Therefore, Figure 3 , Figure 4 As shown in the figure, the improved Unet model combines 1D, 2D and 3D convolutions with residual deconvolution modules respectively, and uses layer normalization, LeakyRELU, maximum pooling, upsampling and full connection to segment the range of fault heating sources. The thermal fault types include short circuit faults, ground faults, open circuit faults, flashover faults and mixed faults. The mixed faults include two or more of short circuit faults, ground faults, open circuit faults and flashover faults. Finally, the two-stage cable thermal fault intelligent detection model is trained and tested to achieve thermal fault classification and segmentation of the range of fault heating sources.

[0052] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A cable thermal fault detection method based on image, characterized in that: The steps include: S1, collecting infrared images, visible light images and X-ray transmission images of the detection position; S2. Input the infrared image and the visible light image into a fusion module based on improved guided filtering to obtain a fused image 1. The fusion module based on improved guided filtering includes median filtering, Gaussian smoothing, improved guided filtering, edge feature map gradient, regional energy comparison and secondary morphological filtering. The improved guided filtering introduces adaptive tolerance, adopts guided filtering transmittance rough estimation, outputs refined transmittance estimation, introduces adaptive tolerance to correct transmittance, and defines tolerance parameter D: Among them, N I(x) represents the number of pixels corresponding to the bright area in the guidance image, C I(x) Represents the number of all pixels in the guide image, and the output corrected transmittance formula is: Among them, t n (x) represents the corrected transmittance of the output, I(x) represents the guide image, K represents the background area, and S represents the light value; S3, the fused image 1 and the X-ray transmission image are subjected to improved bilinear interpolation and homography transformation to generate a fused image 2, which can be expressed as: Where W() represents the post-warping calculation function, V1 represents fused image 1, V2 represents fused image 2, V m represents the X-ray transmission image, x is a parameter whose value range is (0,1), γ∈N; S4. Build a two-stage cable thermal fault intelligent detection model. The first stage is used as a classification model to classify the thermal fault type. The second stage is to segment the fault heating source range. The classification model is a Resnet50 model. The ImageNet dataset is first used to pre-train the model, and then the model is migrated to the thermal fault type for classification. A new loss function is proposed, and two weight calculation methods are introduced. The two weight calculation methods are linearly combined to balance the thermal fault types. The formula is: Where C is the number of thermal fault types, N i is the number of the i-th fault type, L D is the cross entropy damage function, is the equilibrium parameter, f(N i )、g(N i ) are two methods for calculating weights, where f(N i ) is calculated as: Among them, τ represents the hyperparameter, g(N i ) is: Among them, N i is the number of the i-th fault type, and the range of the fault heating source is segmented using the improved Unet model, wherein the improved Unet model combines 1D, 2D and 3D convolutions with the residual deconvolution module respectively to segment the range of the fault heating source; S5. Train and test a two-stage cable thermal fault intelligent detection model to achieve thermal fault classification and segment the scope of the fault heating source.

2. The image-based cable thermal fault detection method according to claim 1, characterized in that: The thermal fault types include short circuit fault, ground fault, open circuit fault, flashover fault and mixed fault. The mixed fault includes two or more of the short circuit fault, ground fault, open circuit fault and flashover fault.

3. The image-based cable thermal fault detection method according to claim 1, characterized in that: The secondary morphological filtering in step S2 includes filling and erosion.

Images