X-ray-based portable high-voltage cable intelligent nondestructive testing method and device

Abstract

The invention discloses a portable intelligent nondestructive testing method for a high-voltage cable. Firstly, a portable X-ray imaging system is used for obtaining an internal structure image of a cable, and the image is sent to an intelligent image processing terminal through a wireless transmission system. According to the terminal, an FFDNet algorithm is adopted to carry out image denoising, an adaptive histogram equalization technology is utilized to carry out image enhancement, and then defect detection and classification are carried out through an improved YOLOv8 algorithm. Finally, the detection result can be automatically stored, classified and filed, and subsequent analysis and maintenance are facilitated. The system is compact in design, portable, high in adaptability and capable of coping with variable detection environments, and meanwhile the detection efficiency and accuracy are remarkably improved. In addition, the system has the characteristics of high automation, convenient data management and the like, and has a wide application prospect.

Description

Technical Field

[0001] The invention belongs to the technical field of cable detection, and in particular relates to a portable X-ray-based intelligent nondestructive detection method and device for high-voltage cables. Background Art

[0002] As an important medium for power transmission, high-voltage cables are widely used in all aspects of the power system. Their safe and stable operation is crucial to the reliability of the entire power network. However, during long-term operation, high-voltage cables may suffer internal structural damage, especially insulation damage and conductor damage, due to external environmental factors, overload operation or manufacturing defects. These internal defects are often difficult to detect with the naked eye, but may cause serious safety problems, such as cable insulation failure, short circuit and even fire, which seriously threaten the safe operation of the power system.

[0003] Existing high-voltage cable detection technologies are mostly based on traditional electrical or thermal methods, such as the bridge method, partial discharge detection method, infrared imaging, etc. Although these methods can detect some defects of the cable to a certain extent, it is often difficult to accurately judge the deep-seated defects inside the cable, especially the minor damage to the insulation layer and conductor. In addition, these methods usually require the cable to be powered off during the detection process, which not only affects the continuity of the power supply, but may also lead to low detection efficiency and increase maintenance costs.

[0004] On the other hand, previous X-ray nondestructive testing equipment is usually large in size, complicated to operate, and difficult to flexibly apply in changing testing sites. The bulkiness of this equipment limits its use in actual testing processes, especially in environments that require frequent movement or complex terrain. Traditional X-ray testing equipment is difficult to play its due role, further limiting testing efficiency and flexibility.

[0005] Secondly, traditional detection methods are also inadequate in identifying defect types and assessing the extent of damage. They cannot provide accurate imaging and analysis of defects inside cables, making it difficult to effectively assess the specific location, morphology, and severity of defects. This uncertainty poses challenges to cable maintenance and repair work, making it impossible to detect and address potential problems in a timely manner, increasing the risk of power system failures.

[0006] In summary, the existing technology has many deficiencies in the detection of internal defects in high-voltage cables, and it is difficult to meet the needs of modern power systems for efficient, accurate and non-contact detection. In addition, due to its large size and inconvenient operation, traditional X-ray detection equipment is even more unable to adapt to complex and changeable detection sites. Therefore, there is an urgent need for a new type of portable non-destructive testing device that can realize real-time monitoring during the daily maintenance of high-voltage cables, accurately identify insulation damage and conductor damage inside the cable, and accurately evaluate its type and degree, so as to ensure the safe and stable operation of the power system. Summary of the invention

[0007] The present invention provides a portable X-ray-based intelligent nondestructive testing method and device for high-voltage cables, which is specially designed for routine maintenance of high-voltage cables and can accurately detect defects inside cables, especially insulation damage and conductor damage, and evaluate the type and severity of defects. By combining X-ray imaging technology, wireless data transmission, advanced image processing algorithms and deep learning technology, the present invention solves the shortcomings of existing technologies in detection accuracy and efficiency, and realizes efficient, non-contact intelligent detection of high-voltage cables.

[0008] A portable X-ray-based intelligent nondestructive testing method for high-voltage cables, comprising:

[0009] (1) Acquire an initial X-ray image of the cross section of the high-voltage cable to be inspected;

[0010] (2) performing denoising on the obtained X-ray image to obtain an image that retains image details and edge information;

[0011] (3) Perform image enhancement on the denoised image;

[0012] (4) Anchor the enhanced image;

[0013] (5) Use the improved YOLOv8 model to locate the defect position and identify the defect type.

[0014] In step (1), a high-resolution cable X-ray image (i.e., X-ray image) can be acquired through a portable X-ray image acquisition system. The acquired X-ray image is then transmitted to an intelligent image processing terminal through a wireless transmission system. Through the wireless transmission system, the X-ray image can be quickly transmitted from the imaging system to the image processing terminal for subsequent processing and analysis.

[0015] Preferably, when acquiring an image, a microfocus X-ray emitter is used to acquire the initial X-ray image, and the nominal focus size of the emitter is less than or equal to 0.05 mm, thereby further improving the image precision and accuracy.

[0016] Furthermore, in the process of acquiring the initial X-ray image, a clear initial X-ray image is obtained by adjusting the output power of the transmitter and optimizing the incident angle of the ray. Specifically, by adjusting the output power of the transmitter to match the cable thickness (diameter), an overall clear X-ray image is obtained. Then, by optimizing the incident angle of the X-ray, the display accuracy of defects in the image can be further improved.

[0017] In the optional scheme, the wireless transmission system includes a wireless communication chip, a radio frequency power amplifier chip, a wireless receiving chip, a data processing chip, etc. The wireless communication chip is responsible for packaging and sending the processed image data, achieving high-speed and stable transmission through a specific frequency band and protocol, and extending the device operation time through a low-power design. The radio frequency power amplifier chip is used to enhance the signal strength so that the signal can be stably transmitted to the receiving end even in the presence of obstacles or at long distances. The wireless receiving chip accurately receives and demodulates the wireless signal through a high-sensitivity antenna and restores it to digital image data. The data processing chip decodes, verifies and compresses the data, and transmits it to the image processing end to ensure the integrity and efficiency of the transmission.

[0018] Furthermore, the FFDNet algorithm is used to denoise the cable X-ray images. FFDNet is an efficient image denoising algorithm that uses a deep convolutional neural network to achieve adaptive denoising. The core idea of ​​the algorithm is to estimate the noise level of the image in layers and apply different denoising filters to each layer for processing. FFDNet predicts the noise component based on the input image and the noise level; FFDNet uses a hierarchical structure to process images with different noise levels. Through multiple layers of convolution and activation functions, FFDNet can gradually extract useful information from the image and remove noise; finally, FFDNet uses a residual learning strategy to subtract the noise component output by the network from the original image to obtain a denoised image. This structure enables FFDNet to effectively remove noise while retaining image details, thereby improving the quality of X-ray images.

[0019] Furthermore, in order to avoid noise amplification that may be generated during local enhancement, the present invention adopts an adaptive histogram equalization method to perform the image enhancement.

[0020] Furthermore, the adaptive histogram equalization method includes:

[0021] (i) Divide the image into several sub-blocks;

[0022] (ii) for each pixel in the sub-block, calculate the cumulative distribution function of its grayscale value;

[0023] (iii) mapping the obtained cumulative distribution function value to a grayscale value after equalization;

[0024] (iv) The enhanced image is obtained by integration.

[0025] Adaptive histogram equalization is used to improve the contrast and detail visibility of the image. Adaptive histogram equalization effectively solves the problem of over-enhancement that may occur when traditional histogram equalization processes global images by performing histogram equalization on the local area of ​​the image. Through adaptive histogram equalization processing, the local contrast of the X-ray image is significantly improved, and the detail structure is clearer.

[0026] Furthermore, an optimized anchor frame mechanism model is used for anchor frame selection. In the optimized anchor frame mechanism, the value of each anchor frame in the anchor frame set is obtained by multiplying the specific scale factor of the high-voltage cable defect by the aspect ratio of the defect shape. Assume that the anchor frame set is {a1, a2, ···, a k}, k is the number of anchor boxes, where each (i-th) anchor box a i =(w i ,h i ) represents the width and height of the anchor box. The optimized anchor box design can be expressed by the following formula:

[0027] a i =scale i aspect i

[0028] Among them, scale i is a specific scale factor for high voltage cable defects, aspect i is the aspect ratio of the defect shape.

[0029] The loss function part of the anchor box is usually related to the predicted bounding box The anchor box mechanism model training phase uses a bounding box regression loss function, which is obtained by adding the intersection-over-union term and the size change rate term, where the intersection-over-union term is the sum of the intersection-over-union ratios between all anchor boxes and the predicted box, and the size change rate term is the weight parameter multiplied by the sum of the width change rate and the height change rate. The specific bounding box regression loss function L bbox It can be expressed as:

[0030]

[0031] in, is the anchor box b i and prediction box The intersection over union ratio between them, λscale is a weight parameter used to adjust the contribution of the anchor box size error to the total loss.

[0032] By optimizing the size and shape of the anchor frame, the model can more accurately capture tiny defects in high-voltage cables, further improving the accuracy and robustness of detection.

[0033] Furthermore, in the backbone network of the improved YOLOv8 model, the EfficientNet module is used to extract the corresponding multi-scale feature map of the input anchor frame image; then the channel attention mechanism SE module is used to further enhance the correlation between channels; finally, the multi-scale feature fusion method is used to obtain defect information.

[0034] Defect detection is performed on X-ray images after image processing based on the improved YOLOv8 algorithm. In order to further improve the detection accuracy and efficiency, this paper optimizes the backbone network of YOLOv8, uses EfficientNet as a lightweight alternative, and integrates the channel attention mechanism SE module (Squeeze-and-Excitation) to enhance the network's feature expression ability. By introducing the combination of EfficientNet and SE modules, the network can adaptively adjust the importance of each channel, further improving the detection ability of cable defects.

[0035] Finally, multi-scale feature fusion technology is introduced to improve detection capabilities by combining feature maps of different layers. Through continuous online updates, the model can be adaptively adjusted to ensure efficient detection under different working conditions.

[0036] The detection algorithm can automatically identify various defects inside the cable (such as insulation damage, conductor damage, etc.), and classify and quantitatively analyze their location, type and severity.

[0037] The obtained test results can be displayed through the intelligent image processing terminal and automatically stored in the local database or cloud server to ensure the persistence and security of the data. The system automatically classifies and archives the detected defect data according to the preset classification standards. The stored information includes the original X-ray image, processed image, test results (defect type, specific location and severity) and other detailed data. The data of each test is accompanied by a timestamp and equipment parameters to ensure traceability. The system also supports the generation of analysis reports based on historical data to help operators identify potential problems and develop maintenance strategies, thereby significantly improving the detection efficiency and the long-term operation safety of the cable.

[0038] The present invention also provides a portable X-ray-based intelligent nondestructive testing device for high-voltage cables, comprising:

[0039] A portable X-ray transmitter for transmitting X-rays to the high-voltage cable to be inspected;

[0040] A flat panel detector, used to receive the attenuated X-ray signal and obtain an initial X-ray image;

[0041] An image processing module is used to obtain an initial X-ray image for processing and obtain corresponding defect detection results;

[0042] A wireless transmission system, used for transmitting the image obtained by the flat panel detector to the image processing module;

[0043] Optional options include:

[0044] A display for showing the results.

[0045] In summary, the present invention provides an efficient and reliable portable intelligent nondestructive testing method for high-voltage cables, which breaks through the limitations of existing testing technologies and realizes the automated detection of high-voltage cable defects by combining nondestructive testing technology with deep learning algorithms. The present invention is compact in design, lightweight, and has excellent portability. It can adapt to various complex testing sites and accurately identify and classify various defects inside the cable without affecting the normal operation of the cable, while performing quantitative analysis. Through automated testing and data management functions, the present invention provides a strong guarantee for the safe operation of high-voltage cables and significantly improves the efficiency, reliability, and on-site adaptability of daily maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a schematic diagram of the process flow of portable X-ray-based intelligent nondestructive testing of high-voltage cables in this method.

[0047] Figure 2 It is a wireless transmission system in the portable X-ray-based intelligent nondestructive testing method for high-voltage cables.

[0048] Figure 3 This is the improved YOLOv8 model structure diagram.

[0049] Figure 4 for Figure 1 The corresponding enlarged image in . DETAILED DESCRIPTION

[0050] See also Figure 1 to Figure 4 In order to describe the present invention in more detail, the technical solution of the present invention is described in detail below in conjunction with specific embodiments. However, the scope of protection of the present invention is not limited thereto.

[0051] The present invention discloses a portable intelligent nondestructive testing method for high-voltage cables. First, a portable X-ray imaging system is used to obtain the internal structure image of the cable, and the image is sent to an intelligent image processing terminal through a wireless transmission system. The terminal uses the FFDNet algorithm for image denoising and the adaptive histogram equalization technology for image enhancement, and then uses the improved YOLOv8 algorithm for defect detection and classification. Finally, the test results will be automatically stored and classified for archiving, which is convenient for subsequent analysis and maintenance.

[0052] (1) Obtaining X-ray images

[0053] The portable X-ray transmitter adopts micro-focus technology with a nominal focus size of 0.01mm to ensure high-precision detection. The X-ray direction is set perpendicular to the cable axis to scan the cable cross section. According to the material thickness of the cable to be tested, the output power of the transmitter is adjusted to achieve the best transmission effect, and the ray incident angle is optimized through the 12° adjustable target angle to improve imaging accuracy and detail clarity. The imaging device adopts a split design with a compact structure and a weight of only 6KG, which is convenient for rapid installation and adjustment in various complex field environments. During operation, the X-ray beam penetrates the high-voltage cable and produces different degrees of attenuation due to the density and thickness differences of the internal materials. The attenuated ray signal is captured by the flat panel detector (FPD) in the receiving board and converted into a high-resolution electrical signal. These signals are processed by the internal image acquisition module and converted into a visual X-ray image.

[0054] (2) Wireless transmission system

[0055] The high-resolution X-ray images collected by the imaging system are transmitted to the computer in real time through the wireless transmission module for processing and analysis, as shown in the schematic diagram. Figure 2 The wireless transmission system is composed of multiple key chips, and each chip cooperates with each other during the data transmission process to ensure efficient and stable data transmission.

[0056] In the wireless transmission module, the Nordic nRF52840 wireless communication chip is responsible for packaging and sending the processed image data, achieving high-speed and stable transmission through the 2.4GHz frequency band and BLE protocol, and extending the device operation time through low-power design. To ensure the quality of data transmission in complex environments, the system is equipped with a Skyworks SE2431L RF power amplifier chip to enhance signal strength so that the signal can be stably transmitted to the receiving end even in the presence of obstacles or long distances. The receiving end uses the TI CC2592 wireless receiving chip to accurately receive and demodulate the wireless signal through a high-sensitivity antenna and restore it to digital image data. Subsequently, the ESP32 data processing chip decodes, verifies and compresses the data and transmits it to the image processing end to ensure the integrity and efficiency of the transmission.

[0057] Through this wireless transmission system, X-ray images can be quickly transmitted from the imaging system to the image processing end for subsequent processing and analysis.

[0058] (3) FFDNet denoising

[0059] The present invention adopts the FFDNet algorithm to denoise the cable X-ray image. FFDNet is an efficient image denoising algorithm that implements adaptive denoising through a deep convolutional neural network. The core idea of ​​the algorithm is to estimate the noise level of the image in layers and apply different denoising filters to each layer for processing. For the input noise image y, FFDNet first assumes that the noise n follows an independent and identically distributed Gaussian distribution, that is, n~Ν(0,σ 2 ), where σ is the standard deviation of the noise.

[0060] The denoising process can be expressed as:

[0061]

[0062] in, is the denoised image, and f(y,σ) is the output of the FFDNet network, i.e., the noise component predicted based on the input image y and the noise level σ.

[0063] FFDNet uses a hierarchical structure to process images with different noise levels. The output of each layer can be expressed as:

[0064] h i =Φ(W i *h i-1 +b i ) (1-2)

[0065] Among them, h i is the output feature map of the i-th layer, W i is the convolution kernel, b i is the bias term, and Φ is a nonlinear activation function (such as ReLU). Through multiple layers of convolution and activation functions, FFDNet can gradually extract useful information from the image and remove noise.

[0066] Finally, FFDNet uses a residual learning strategy to subtract the noise component of the network output from the original image to obtain a denoised image:

[0067]

[0068] This structure enables FFDNet to effectively remove noise while preserving image details, thereby improving the quality of X-ray images.

[0069] (4) Image enhancement

[0070] The present invention adopts contrast-limited adaptive histogram equalization (CLAHE) to improve the contrast and detail visibility of the image. Adaptive histogram equalization effectively solves the over-enhancement problem that may occur when traditional histogram equalization processes the global image by performing histogram equalization on the local area of ​​the image.

[0071] First, assume that the input grayscale image is I(x,y) and its grayscale range is [0,L-1]. The first step of AHE is to divide the image into several m×n blocks, and the pixels in each sub-block are locally contrast enhanced by histogram equalization.

[0072] For a pixel I in a sub-block block (x,y), calculate the cumulative distribution function (CDF) of its gray value:

[0073]

[0074] Among them, H(i) represents the histogram frequency of gray value i, k is the gray level, and mn is the total number of pixels in the sub-block.

[0075] Next, the grayscale value of the pixel is mapped to the equalized grayscale value I'(x,y):

[0076] I'(x,y)=(L-1)×CDF(I block (x,y)) (1-5)

[0077] Through the above steps, CLAHE can make the dark areas in the image brighter, while the bright areas will not be overexposed due to global enhancement. However, in order to avoid the noise amplification that may occur during local enhancement, the improved version of AHE, Contrast Limited Adaptive Histogram Equalization (CLAHE), further limits the cumulative frequency of the histogram of each sub-block, making the image enhancement more natural and smooth.

[0078] Finally, the enhanced image can be expressed as:

[0079]

[0080] Through adaptive histogram equalization processing, the local contrast of X-ray images is significantly improved, and the detailed structure is clearer.

[0081] (5) Optimize the anchor frame mechanism

[0082] The traditional YOLOv8 algorithm estimates the target position and size through a set of predefined anchor boxes, which are usually based on the statistical characteristics of the data set. However, for the high-voltage cable detection task, this paper optimizes the design of the anchor box to make it more adaptable to the characteristics of cable defects. Assume that the anchor box set is {a1, a2, ···, a k}, k is the total number of anchor boxes, where any i-th anchor box a i =(w i ,h i ) represents the width and height of the anchor box. The optimized anchor box design can be expressed by the following formula:

[0083] a i =scale i aspect i (1-11)

[0084] Among them, scale i It is a specific scale factor for high-voltage cable defects, expressed as (a, b), where the value ranges of a and b are independently 65 to 85. i is the aspect ratio of the defect shape.

[0085] According to the size characteristics of common defects in high-voltage cables, in this embodiment, (75,75) is selected as the scale factor in the high-voltage cable defect detection task. Other selections and optimizations can also be made according to the balance between accuracy and computational efficiency required by the detection task. There are various types of defects in high-voltage cables, which may include tiny cracks, insulation damage, conductor breakage, etc. After analyzing and testing a large amount of data, it was found that the sizes of most defects are concentrated in a certain range, and the anchor frame size of 75×75 can well cover the areas of these common defects. This scale factor is large enough to include larger defects, but not too large to affect the detection capability of small defects, thereby ensuring a high-precision detection effect.

[0086] Secondly, the scale factor of (75,75) can also effectively reduce the computational complexity in practical applications. For the anchor frame mechanism, the size of the anchor frame directly affects the computational overhead. Through experiments, it is found that the scale of 75×75 can reduce the computational burden of the system while ensuring the detection accuracy and improve the real-time performance of the detection. It is very suitable for high-voltage cables, which require fast response.

[0087] The loss function part of the anchor box is usually related to the predicted bounding box The loss function L for bounding box regression is related to the difference between the bounding box b = (x, y, w, h) and the true bounding box b = (x, y, w, h). bbox It can be expressed as:

[0088]

[0089] in, is the anchor box b i and prediction box The intersection over union ratio between them, λscale is a weight parameter used to adjust the contribution of the anchor box size error to the total loss. is the prediction box width, is the height of the prediction box.

[0090] By optimizing the size and shape of the anchor frame, the model can more accurately capture tiny defects in high-voltage cables, further improving the accuracy and robustness of detection.

[0091] (6) Cable defect detection based on improved YOLOv8 algorithm

[0092] See also Figure 3 , based on the improved YOLOv8 algorithm, defect detection is performed on the processed X-ray images. In order to further improve the detection accuracy and efficiency, this paper optimizes the backbone network of YOLOv8, uses EfficientNet as a lightweight alternative, and integrates the channel attention mechanism SE module (Squeeze-and-Excitation) to enhance the network's feature expression ability.

[0093] First, EfficientNet optimizes the depth, width, and resolution of the network through a composite scaling strategy. The feature map output by its backbone network can be expressed as:

[0094] F=EfficientNet(I) (1-7)

[0095] Where I represents the input X-ray image, and F represents the multi-scale feature map extracted by EfficientNet. The lightweight design of EfficientNet not only reduces the number of model parameters, but also maintains a high feature extraction capability, which is suitable for real-time detection in resource-constrained environments.

[0096] Next, the SE module is combined to enhance the correlation between channels. The SE module first performs global average pooling on the feature map F to obtain the statistical information z of each channel:

[0097]

[0098] Among them, H and W are the height and width of the feature map respectively, and F c (i, j) represents the pixel value of feature map F on channel c.

[0099] Then, two fully connected layers are used to perform nonlinear mapping on the channel information to generate the weight vector sc :

[0100] s c =σ(W2·δ(W1·z)) (1-9)

[0101] Among them, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the Sigmoid activation function. Finally, use the weight vector s c Weight the original feature map:

[0102] F′ c =s c ·F c (1-10)

[0103] Among them, F c That is F c (i,j).

[0104] By introducing the combination of EfficientNet and SE modules, the network can adaptively adjust the importance of each channel, further improving the ability to detect cable defects.

[0105] Then, the multi-scale feature fusion technology is introduced to improve the detection capability by combining feature maps of different layers. Assume that the feature maps are F1, F2, ···, F n , where F i Represents the feature map of the i-th layer. The fused feature map F fused It can be expressed as:

[0106]

[0107] Among them, α i It is the weight coefficient of each layer of feature map, which ensures that the feature information of different scales is reasonably integrated. Through multi-scale feature fusion, the model can be more robust in detecting defects of different sizes and shapes.

[0108] In the model training phase, the online learning module continuously trains the real-time acquired data to update the model parameters. Assume that the weight parameter of the model is θ and the input data is x t , the loss function is L t (x t ; θ), the update process of online learning can be expressed as:

[0109]

[0110] Among them, θ t is the model parameter at the current moment, η is the learning rate, is the gradient of the current loss function.

[0111] Through continuous online updates, the model can be adaptively adjusted to ensure efficient detection under different working conditions.

[0112] The detection algorithm can automatically identify various defects inside the cable (such as insulation damage, conductor damage, etc.), and classify and quantitatively analyze their location, type and severity.

[0113] Figure 4 Among them, (a) is the original image obtained by steps (1) and (2); (b) is the image obtained after processing in step (3); (c) is the image obtained after processing in step (4); and (d) is the image obtained after processing in steps (5) and (6).

[0114] (7) Result storage and classification

[0115] The test results are displayed on the intelligent image processing terminal and automatically stored in the local database or cloud server to ensure data persistence and security. The system automatically classifies and archives the detected defect data based on the preset classification standards. The stored information includes detailed data such as the original X-ray image, processed image, test results, defect type, specific location and severity. Each test data is accompanied by a timestamp and equipment parameters to ensure traceability. The system also supports the generation of analysis reports based on historical data to help operators identify potential problems and develop maintenance strategies, thereby significantly improving detection efficiency and the long-term operation safety of cables.

[0116] Finally, based on the detected defect data, the present invention can automatically classify and archive relevant information of high-voltage cables, such as defect type, specific location and severity. The device is not only compact and portable, highly adaptable, and can be flexibly applied to various complex detection environments, but also significantly improves detection efficiency and accuracy, providing strong technical support for the intelligent maintenance of high-voltage cables. In addition, the system has the characteristics of high automation and convenient data management, and has broad application prospects.

Claims

1. A portable high-voltage cable intelligent nondestructive testing method based on X-rays, characterized in that: include: (1) Acquire an initial X-ray image of the cross section of the high-voltage cable to be inspected; (2) performing denoising on the obtained X-ray image to obtain an image that retains image details and edge information; (3) Perform image enhancement on the denoised image; (4) Anchor the enhanced image; (5) Use the improved YOLOv8 model to locate the defect position and identify the defect type.

2. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 is characterized in that: The FFDNet algorithm is used to perform the denoising process on the image.

3. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 is characterized in that: The image enhancement is performed using an adaptive histogram equalization method.

4. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 3 is characterized in that: The adaptive histogram equalization method comprises: (i) Divide the image into several sub-blocks; (ii) for each pixel in the sub-block, calculate the cumulative distribution function of its grayscale value; (iii) mapping the obtained cumulative distribution function value to a grayscale value after equalization; (iv) The enhanced image is obtained by integration.

5. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 is characterized in that: An optimized anchor frame mechanism is used for anchoring frames. In the optimized anchor frame mechanism, the value of each anchor frame in the anchor frame set is obtained by multiplying the specific scale factor of the high-voltage cable defect by the aspect ratio of the defect shape.

6. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 5 is characterized in that: During the training phase of the anchor box mechanism model, a bounding box regression loss function is used, which is obtained by adding an intersection-over-union term and a size change rate term, wherein the intersection-over-union term is the sum of the intersection-over-union ratios between all anchor boxes and prediction boxes, and the size change rate term is the weight parameter multiplied by the sum of the width change rate and the height change rate.

7. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 or 5, characterized in that: In the backbone network of the improved YOLOv8 model, the EfficientNet module is used to extract the corresponding multi-scale feature map of the input anchor frame image; then the channel attention mechanism SE module is used to further enhance the correlation between channels; finally, the multi-scale feature fusion method is used to obtain defect information.

8. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 is characterized in that: The initial X-ray image is acquired using a micro-focus X-ray emitter, the nominal focus size of the emitter being less than or equal to 0.05 mm.

9. The portable X-ray-based intelligent nondestructive testing method for high-voltage cables according to claim 1 is characterized in that: In the process of acquiring the initial X-ray image, a clear initial X-ray image is obtained by adjusting the output power of the transmitter and optimizing the ray incident angle.

10. A portable intelligent nondestructive testing device for high-voltage cables based on X-rays, characterized in that include: A portable X-ray transmitter for transmitting X-rays to the high-voltage cable to be inspected; A flat panel detector, used to receive the attenuated X-ray signal and obtain an initial X-ray image; An image processing module is used to obtain an initial X-ray image for processing and obtain corresponding defect detection results; A wireless transmission system, used for transmitting the image obtained by the flat panel detector to the image processing module; Optional options include: A display for showing the results.

Images