High-voltage cable major insulation surface defect detection method based on staggered sampling fusion and progressive feature pyramid

Abstract

The invention provides a high-voltage cable major insulation surface defect detection method based on staggered sampling fusion and a progressive feature pyramid. A detection model based on the staggered sampling fusion and the progressive feature pyramid for improving a YOLOv7 algorithm framework is used for performing defect detection on a high-voltage cable major insulation surface. According to the detection model, an interlaced sampling fusion module is introduced into the third layer, the fifth layer and the seventh layer of a network structure in a YOLOv7 algorithm framework, so that the loss of detail information of the model in the down-sampling process is reduced; a progressive feature pyramid network is introduced, features of different levels of the network are utilized, adaptive spatial feature fusion is utilized in the multi-level feature fusion process by adopting an asymptotic fusion strategy, upsample and concat operations in a YOLOv7 algorithm are replaced, and different spatial weights are given to the features of different levels.

Description

Technical Field

[0001] The present invention relates to the technical fields of high-voltage cable construction, machine learning, intelligent vision, image processing, target detection, and the like, and in particular to a method for detecting surface defects of main insulation of a high-voltage cable based on staggered sampling fusion and progressive feature pyramid. Background Art

[0002] During the installation of high-voltage cable accessories, the construction requirements and operating procedures must be strictly followed to ensure the installation quality and safe operation of the power system. The construction process includes: construction preparation, cable pretreatment, accessory installation and quality inspection. The cable pretreatment process includes:

[0003] 1. Cable cutting: Use professional cutting tools to ensure that the cable cut is smooth and burr-free; accurately measure and cut the cable according to the length requirements of the accessory installation.

[0004] 2. Cable stripping. According to the installation instructions of the accessories, correctly strip the outer sheath, steel armor, inner sheath, etc. of the cable; the stripping size should be accurate to avoid damaging the main insulation of the cable. 3. Main insulation grinding and cleaning. Including the rough grinding stage and the fine grinding stage. After grinding, use cleaning materials such as anhydrous alcohol, white cloth, etc. to clean the surface of the main insulation of the cable to ensure that there are no defects such as dirt, scratches, pits, etc. on the main insulation surface.

[0005] If the outer surface of the main insulation is polished with defects that cause it to be uneven or have dirt or scratches, an air gap will occur at the interface between the cable stress cone and the main insulation. Under the operating voltage, local discharge will exist for a long time, causing rapid aging of the insulation, the formation of electrical trees, and ultimately breakdown.

[0006] The detection of surface defects of high-voltage cable main insulation after grinding currently relies mainly on manual inspection and on-site verification. Professional technicians need to go to the high-voltage cable project site in person and use professional tools and rich experience to conduct detailed inspections of the installation process and finished products of cable accessories. They identify surface defects of high-voltage cables after grinding through visual inspection, measurement with measuring tools, and instrument testing when necessary. Problems with manual inspection:

[0007] 1. Special tools are expensive, complicated to operate, have high requirements for the testing environment, and are easily affected by external environmental factors.

[0008] 2. The labor intensity is high. When performing defect detection, technicians not only have to face heavy physical labor, but also need to pay attention to personal safety at all times. At the same time, due to the large number and dense distribution of cable accessories, manual inspection one by one is not only time-consuming and labor-intensive, but may also increase the risk of damage to equipment or accessories due to frequent contact.

[0009] 3. The detection is not real-time and accurate enough. The results of manual detection are affected by many subjective factors such as the detection personnel's experience, concentration and fatigue status, which may lead to reduced accuracy of the detection results, missed detection or false detection, and thus affect the quality and safety of the entire project.

[0010] 4. Poor consistency. The differences in judgment standards, operating habits and experience levels among different inspectors may result in inconsistent test results even when testing the same batch of cable accessories. This not only brings difficulties to subsequent quality analysis and improvement work, but also may affect customers' trust and satisfaction with the product. Summary of the invention

[0011] The principle of using machine vision detection technology to identify surface defects of the main insulation of high-voltage cables is to use a high-resolution camera or image sensor to shoot the surface of the main insulation of high-voltage cables, obtain image information, and then analyze and process the image through an image processing algorithm to identify defects on the surface of the main insulation. Its advantages are: it can realize automated detection, improve detection efficiency and accuracy; it can detect various defects on the surface of the main insulation, such as scratches, wear, cracks, etc.; the detection results can be saved in the form of images, which is convenient for subsequent analysis and tracing.

[0012] Consider the problems of existing surface defect models:

[0013] 1. It is necessary to ensure high resolution of the image and good lighting conditions. The shooting angle also needs to be strictly controlled. The adaptability to different environments and conditions is limited, and parameters need to be adjusted for specific situations.

[0014] 2. The model has weak detection capabilities for subtle defects and is prone to failure in details, resulting in false detection or missed detection.

[0015] 3. Complex image processing algorithms and deep models cause the detection process to take a long time and may not meet efficiency requirements.

[0016] In response to the above problems, the present invention aims to propose an innovative solution, mainly for the detection of surface defects after grinding of the main insulation of high-voltage cables. The defect detection model is trained by deep learning methods so that it can adapt to defect detection tasks under different angles and lighting conditions. At the same time, an interlaced sampling fusion module is proposed to effectively improve the model's detection ability for subtle defects and avoid missed detections and false detections. In addition, a progressive feature pyramid network is introduced to improve the model so that the model can adapt to defects of different scales. Based on the interlaced sampling fusion module and the progressive feature pyramid network, it has excellent detection performance for defects with large scale differences and weak feature details. It aims to achieve high efficiency, wide adaptability and high precision in the detection of surface defects of the main insulation of high-voltage cables.

[0017] The present invention specifically adopts the following technical solutions:

[0018] A method for detecting defects on the main insulation surface of a high-voltage cable based on staggered sampling fusion and progressive feature pyramid: a detection model improved on the YOLOv7 algorithm framework based on staggered sampling fusion and progressive feature pyramid is used to detect defects on the main insulation surface of a high-voltage cable; the detection model introduces staggered sampling fusion modules at the 3rd, 5th and 7th layers of the network structure in the YOLOv7 algorithm framework to reduce the loss of detail information of the model during the downsampling process; and a progressive feature pyramid network is introduced to utilize features at different levels of the network by adopting an asymptotic fusion strategy, and adaptive spatial feature fusion is used in the multi-level feature fusion process to replace the upsample and concat operations in the YOLOv7 algorithm, so as to give different spatial weights to features at different levels.

[0019] Furthermore, the interleaved sampling fusion module obtains four feature maps by interleaving the input feature map in four different ways, and then splices them along the channel dimension to reduce the scale of the input feature map; the input feature map x is represented as [C, H, W], where C represents the number of channels, H represents the height, and W represents the width, then the four sub-feature maps are represented as:

[0020]

[0021] The concatenated feature map is expressed as:

[0022] x cat =[x 1 ;x 2 ;x 3 ;x 4 ]

[0023] Then it is input into the SimAM module for attention mechanism processing and then output after a group convolution and a point convolution.

[0024] Furthermore, the SimAM module adopts an attention mechanism to determine the importance of neurons by evaluating their linear separability, and performs weighted processing on important neurons: first, an energy function is defined, which comprehensively considers the weight and bias of neurons, as well as the relationship between the estimated value and the true value of neurons to measure the importance of each neuron, and then multiplies the original feature matrix element by element after processing with the sigmoid function, thereby enhancing the feature map.

[0025] Furthermore, assuming that at position (i, j), the feature vector from the nth layer to the lth layer is expressed as The resulting feature vector obtained after the adaptive spatial feature fusion is It is expressed as:

[0026]

[0027] in Represents the spatial weights of the three level features of the lth layer, which are affected by restrictions.

[0028] Furthermore, the loss function of the detection model adopts the SIOU loss function.

[0029] And, a high-voltage cable main insulation surface defect detection system based on staggered sampling fusion and progressive feature pyramid is based on a computer system, including: a detection model improved by the YOLOv7 algorithm framework based on staggered sampling fusion and progressive feature pyramid, which is used for defect detection on the main insulation surface of the high-voltage cable; the detection model introduces staggered sampling fusion modules in the 3rd, 5th and 7th layers of the network structure in the YOLOv7 algorithm framework to reduce the loss of detail information of the model during the downsampling process; and introduces a progressive feature pyramid network to utilize features of different levels of the network by adopting an asymptotic fusion strategy, and utilizes adaptive spatial feature fusion in the multi-level feature fusion process to replace the upsample and concat operations in the YOLOv7 algorithm, so as to give different spatial weights to features of different levels.

[0030] Furthermore, the interleaved sampling fusion module obtains four feature maps by interleaving the input feature map in four different ways, and then splices them along the channel dimension to reduce the scale of the input feature map; the input feature map x is represented as [C, H, W], where C represents the number of channels, H represents the height, and W represents the width, then the four sub-feature maps are represented as:

[0031]

[0032] The concatenated feature map is expressed as:

[0033] x cat =[x 1 ;x 2 ;x 3 ;x 4 ]

[0034] Then it is input into the SimAM module for attention mechanism processing and then output after a group convolution and a point convolution.

[0035] Furthermore, assuming that at position (i, j), the feature vector from the nth layer to the lth layer is expressed as The resulting feature vector obtained after the adaptive spatial feature fusion is It is expressed as:

[0036]

[0037] in Represents the spatial weights of the three level features of the lth layer, which are affected by restrictions.

[0038] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for detecting surface defects of the main insulation of a high-voltage cable based on interleaved sampling fusion and progressive feature pyramid are implemented as described above.

[0039] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for detecting surface defects of the main insulation of a high-voltage cable based on interleaved sampling fusion and progressive feature pyramid as described above.

[0040] Compared with the prior art, the present invention and its preferred solution aim at the problems of cumbersome workload and low efficiency in the traditional method of detecting defects of high-voltage cable engineering accessories. By designing an interlaced sampling fusion module and introducing a progressive feature pyramid network, the model can accurately detect both subtle defects and defects with significant scale differences. The model's detection capability for complex defect types is significantly improved, and missed detections and false detections are reduced, thereby ensuring the reliability and accuracy of the detection results. Furthermore, the designed detection solution enables technicians to quickly and easily obtain surface defect images of high-voltage cable engineering accessories through industrial cameras. The use of defect detection models and the design of host computer software enables the statistics and recording of surface defects of high-voltage cable engineering accessories to be automated. The problems of low efficiency, poor consistency and low precision of traditional methods are effectively solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0042] Figure 1 This is the YOLOv7 algorithm framework diagram;

[0043] Figure 2 This is a diagram of an interleaved sampling fusion module according to an embodiment of the present invention;

[0044] Figure 3 This is a module diagram of SimAM according to an embodiment of the present invention;

[0045] Figure 4 It is a progressive feature pyramid network diagram of an embodiment of the present invention;

[0046] Figure 5 This is a diagram of the ASFF network architecture of an embodiment of the present invention;

[0047] Figure 6 This is a network structure diagram of a model according to an embodiment of the present invention;

[0048] Figure 7 The following is a flowchart of the software system corresponding to the embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the features and advantages of the present invention more clearly understood, the following embodiments are specifically described in detail with reference to the accompanying drawings:

[0050] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0051] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0052] In the defect detection stage, the present invention improves the YOLOv7 algorithm framework to greatly improve its defect detection performance for high-voltage cable engineering accessories.

[0053] YOLOv7 algorithm framework Figure 1 As shown in FIG. 1 , the main improvement of the present invention is that the 3rd, 5th, and 7th layers of the network structure reduce the loss of detail information in the downsampling process of the model by introducing the staggered sampling fusion module SFD, so that the feature expression ability of the model is effectively improved. The core of the staggered sampling fusion module is to perform staggered sampling of the input feature map in four different ways, and then splice the obtained four feature maps along the channel dimension, thereby reducing the scale of the input feature map. Its structure diagram is shown in FIG. Figure 2 shown.

[0054] Specifically, the core of the interleaved sampling fusion module is to obtain four parts of the feature map by interleaving the input feature map in four different ways, and then splice them along the channel dimension to reduce the scale of the input feature map. If the input feature map x is represented as [C, H, W], where C represents the number of channels, H represents the height, and W represents the width, the four sub-feature maps can be expressed by the following formula:

[0055]

[0056] The concatenated feature map can be expressed as:

[0057] x cat =[x 1 ;x 2 ;x 3 ;x 4 ]

[0058] Figure 2 The SimAM module in the example is an attention mechanism module, which is used to focus the model on key information. Its network structure is as follows: Figure 3 shown.

[0059] The SimAM attention mechanism determines the importance of neurons by evaluating their linear separability and performs weighted processing on important neurons. The mechanism first defines an energy function that comprehensively considers the weights, biases, and relationship between the estimated values ​​and true values ​​of neurons. Neurons with low energy values ​​usually mean greater differences and higher importance. Based on this energy value, the importance of each neuron can be measured, and after being processed by the sigmoid function, it is multiplied element by element with the original feature matrix to enhance the feature map. The calculation formula is as follows:

[0060]

[0061] In this way, the SimAM attention mechanism can focus on the more important neurons and improve their contribution in the feature map accordingly, thereby improving the performance of the overall model. At the same time, a group convolution and a point convolution are introduced at the end of the interleaved sampling fusion module to ensure that the output feature map can be seamlessly connected with the subsequent modules to maintain information fluency and network effectiveness.

[0062] In addition, in order to enhance the model's detection performance for defects with large scale differences, this embodiment introduces a progressive feature pyramid network to utilize the features of different layers of the network to improve the model's detection capability for targets with large scale differences. The network structure is as follows: Figure 4 shown.

[0063] By adopting a gradual fusion strategy, the progressive feature pyramid network successfully solves the problem of insufficient feature fusion between non-adjacent levels in the ordinary feature pyramid network. Under this fusion mechanism, features from different levels gradually converge in the process of gradual fusion, effectively narrowing the semantic gap between them and ensuring the full retention of key information. In addition, the progressive feature pyramid network uses adaptive spatial feature fusion (ASFF) to assign different spatial weights to features at different levels in the process of multi-level feature fusion. The network architecture of ASFF is as follows: Figure 5As shown in the figure, it is used to replace the upsample and concat operations in the YOLOv7 algorithm. ASFF assigns different spatial weights to features at different levels by learning weight parameters. These weight parameters control the importance of each feature map in the fusion process, so that the features of key levels are strengthened while suppressing contradictory information from different objects.

[0064] Assume that at position (i, j), the feature vector from layer n to layer l is expressed as The resulting feature vector obtained after ASFF fusion is It can be expressed as:

[0065]

[0066] in Represents the spatial weights of the three level features of the lth layer, which are affected by restrictions.

[0067] Through the method of gradual fusion, the progressive feature pyramid network successfully solves the problem of poor fusion effect of non-adjacent level features in the feature pyramid network. In this feature fusion method, features from different levels gradually approach each other during the fusion process, effectively bridging the semantic gap between them, thereby better retaining key information. This embodiment combines the design concept of the progressive feature pyramid and comprehensively optimizes and reconstructs the multi-scale feature fusion network, aiming to further enhance the feature representation ability and detection accuracy of the model, thereby improving the overall performance.

[0068] In addition, the model of this embodiment further adopts SIOU to improve the loss function, and optimizes the learning ability and detection performance of the model by introducing the loss of angle, distance and shape. The calculation formula of angle loss is as follows:

[0069]

[0070] Where x is the sine value of α, σ is the distance between the predicted box and the midpoint of the true box, and C h is the height difference between the predicted box and the true box.

[0071] The distance loss is calculated as follows:

[0072]

[0073] Where γ=2-Λ,ρ x C w With C x The square of the ratio, ρ y C h With C yThe square of the ratio.

[0074] The shape loss is calculated as follows:

[0075]

[0076] Where θ is the shape loss weight, ω w is the ratio of the absolute value of the width difference between the predicted box and the real box to the maximum value, ω h It is the ratio of the absolute value of the height difference between the predicted box and the true box to the maximum value.

[0077] In summary, the mathematical expression of SIoU is as follows:

[0078]

[0079] Compared with the CIoU loss function, the SIoU loss function comprehensively considers the matching direction, distance and shape when evaluating the matching degree between the real box and the predicted box. And when SIoU measures the width and height relationship between the two, it adopts a more sophisticated independent calculation method, rather than simply comparing their relative proportions. This avoids the problem that the penalty term of the original CIOU may return to zero when the aspect ratio of the predicted box is consistent with that of the real box, and achieves a more accurate and stable loss evaluation. The final model network structure of the present invention is as follows Figure 6 shown.

[0080] Based on the above algorithm design, this embodiment further considers that in order to complete the actual defect detection task, the software is designed to decode the data collected by the camera and detect the target, and the information results identified during the target detection process are displayed, and recorded in the form of a table file to the host computer. The core workflow of the system is as follows Figure 7 shown.

[0081] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0082] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0083] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0084] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0085] The present invention is not limited to the above-mentioned optimal implementation mode. Anyone can derive other various forms of high-voltage cable main insulation surface defect detection methods based on staggered sampling fusion and progressive feature pyramid under the inspiration of the present invention. All equal changes and modifications made according to the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A method for detecting surface defects of main insulation of high-voltage cables based on staggered sampling fusion and progressive feature pyramid, characterized in that: The main insulation surface of the high-voltage cable is defect-detected using a detection model that improves the YOLOv7 algorithm framework based on staggered sampling fusion and progressive feature pyramid. The detection model introduces staggered sampling fusion modules at the 3rd, 5th and 7th layers of the network structure in the YOLOv7 algorithm framework to reduce the loss of detail information in the downsampling process of the model. A progressive feature pyramid network is introduced to utilize features at different levels of the network by adopting an asymptotic fusion strategy. In the multi-level feature fusion process, adaptive spatial feature fusion is used to replace the upsample and concat operations in the YOLOv7 algorithm, and different spatial weights are given to features at different levels.

2. The method for detecting surface defects of main insulation of high-voltage cables based on staggered sampling fusion and progressive feature pyramid according to claim 1 is characterized in that: The interlaced sampling fusion module obtains four feature maps by interlacing the input feature map in four different ways, and then splices them along the channel dimension to reduce the scale of the input feature map; the input feature map x is represented as [C, H, W], where C represents the number of channels, H represents the height, and W represents the width, then the four sub-feature maps are represented as: The concatenated feature map is expressed as: x cat =[x1;x2;x3;x4] Then it is input into the SimAM module for attention mechanism processing and then output after a group convolution and a point convolution.

3. The method for detecting surface defects of high-voltage cable main insulation based on staggered sampling fusion and progressive feature pyramid according to claim 2 is characterized in that: The SimAM module adopts the attention mechanism to determine the importance of neurons by evaluating their linear separability, and performs weighted processing on important neurons: first, an energy function is defined, which comprehensively considers the weight and bias of neurons, as well as the relationship between the estimated value and the true value of neurons to measure the importance of each neuron, and then multiplies the original feature matrix element by element after processing with the sigmoid function, thereby enhancing the feature map.

4. The method for detecting surface defects of high-voltage cable main insulation based on staggered sampling fusion and progressive feature pyramid according to claim 1 is characterized in that: Assume that at position (i, j), the feature vector from layer n to layer l is expressed as The resulting feature vector obtained after the adaptive spatial feature fusion is It is expressed as: in Represents the spatial weights of the three level features of the lth layer, which are affected by restrictions.

5. The method for detecting surface defects of main insulation of high-voltage cables based on staggered sampling fusion and progressive feature pyramid according to claim 4 is characterized in that: The loss function of the detection model adopts the SIOU loss function.

6. A high-voltage cable main insulation surface defect detection system based on staggered sampling fusion and progressive feature pyramid, based on a computer system, characterized in that: include: A detection model based on staggered sampling fusion and progressive feature pyramid is used to improve the YOLOv7 algorithm framework for defect detection on the main insulation surface of a high-voltage cable. The detection model introduces staggered sampling fusion modules in the 3rd, 5th and 7th layers of the network structure of the YOLOv7 algorithm framework to reduce the loss of detail information in the downsampling process of the model. A progressive feature pyramid network is introduced to utilize features at different levels of the network by adopting an asymptotic fusion strategy. Adaptive spatial feature fusion is used in the multi-level feature fusion process to replace the upsample and concat operations in the YOLOv7 algorithm, and different spatial weights are given to features at different levels.

7. A high-voltage cable main insulation surface defect detection system based on staggered sampling fusion and progressive feature pyramid according to claim 6, characterized in that: The interlaced sampling fusion module obtains four feature maps by interlacing the input feature map in four different ways, and then splices them along the channel dimension to reduce the scale of the input feature map; the input feature map x is represented as [C, H, W], where C represents the number of channels, H represents the height, and W represents the width, then the four sub-feature maps are represented as: The concatenated feature map is expressed as: x cat =[x1;x2;x3;x4] Then it is input into the SimAM module for attention mechanism processing and then output after a group convolution and a point convolution.

8. A high-voltage cable main insulation surface defect detection system based on staggered sampling fusion and progressive feature pyramid according to claim 6, characterized in that: Assume that at position (i, j), the feature vector from layer n to layer l is expressed as The resulting feature vector obtained after the adaptive spatial feature fusion is It is expressed as: in Represents the spatial weights of the three level features of the lth layer, which are affected by restrictions.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the high-voltage cable main insulation surface defect detection method based on staggered sampling fusion and progressive feature pyramid are implemented as described in any one of claims 1 to 5.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for detecting surface defects of the main insulation of a high-voltage cable based on staggered sampling fusion and progressive feature pyramid as described in any one of claims 1 to 5 are implemented.

Images