A method and related device for visual recognition of a compression joint of a power transmission line based on structural prior

Abstract

The application discloses a kind of based on structural prior transmission line crimp tube visual identification method and related equipment, method includes: constructing the structural prior transmission line crimp tube library and extracting structural prior vector;Structural prior vector is mapped, and prior embedding feature is obtained;Current transmission line image is input into stem network and is extracted, and original feature map is obtained, and original feature map is aligned, and candidate feature map is obtained;Prior embedding feature is combined, and candidate feature map is weighted and enhanced, and target enhancement feature map is obtained;Target enhancement feature map is input into detection head and is predicted, and target parameter prediction result is output;Structural parameter prediction result in target parameter prediction result is checked with structure consistency, and target parameter prediction result meeting the requirement of structure consistency is determined as target recognition result.The application can improve the recognition accuracy and recognition stability of crimp tube under complex background, and can be widely applied in artificial intelligence technical field.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a visual recognition method and related equipment for transmission line crimping pipes based on structural priors. Background Technology

[0002] Crimping pipes are key components in split conductors, enabling both mechanical and electrical connections. The quality of their internal crimping directly affects the conductor's current-carrying capacity, temperature rise, and long-term service safety. Currently, common methods for inspecting crimping pipes in the industry include manual power outages and tower climbing, fixed X-ray imaging, or ground-based visual inspection. Furthermore, with the development of drones and online inspection robots, the "drone + X-ray inspection equipment + robot" method for live-line high-altitude non-destructive testing is gradually becoming an important direction for industry development.

[0003] However, relying on manual inspection or conventional image processing methods is significantly affected by factors such as changes in lighting, wire swaying, and complex backgrounds, resulting in insufficient recognition stability. While traditional deep learning methods have improved detection capabilities, their generalization ability across different wire types and orientations is weak. Furthermore, multi-splitting wire scenarios present problems such as occlusion and random angle variations, making it difficult for current detection methods to guarantee recognition accuracy and meet the automation and reliability requirements of collaborative inspection between UAVs and robots.

[0004] In summary, the technical problems existing in the relevant technologies need to be improved. Summary of the Invention

[0005] The embodiments of this application aim to at least partially solve one of the technical problems in the related art. Therefore, the main objective of the embodiments of this application is to propose a visual recognition method and related equipment for transmission line crimped pipes based on structural priors, which can improve the recognition accuracy and stability of transmission line crimped pipes in complex backgrounds.

[0006] To achieve the above objectives, one aspect of this application proposes a visual recognition method for transmission line crimping pipes based on structural priors, the method comprising the following steps: Construct a structural prior library for transmission line press-fit pipes, and extract structural prior vectors from the structural prior library; The first prior embedding feature is obtained by performing a nonlinear mapping on the structural prior vector; The current transmission line image is acquired, and the current transmission line image is input into the backbone network for feature extraction processing to obtain the original feature map. The original feature map is then subjected to feature alignment processing to obtain the candidate feature map. The attention module guided by structural priors combines the first prior embedded features to perform weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map. The enhanced feature map of the target is input into the detection head for parameter prediction, and the target parameter prediction result is output. By combining the prior parameters of the compression pipe in the structural prior library, the structural parameter prediction results in the target parameter prediction results are checked for structural consistency, and the target parameter prediction results that meet the structural consistency requirements are determined as the target identification results.

[0007] In some embodiments, the structure-prior-guided attention module, in conjunction with the first prior embedded features, performs weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map, including: Global average pooling is performed on the candidate feature maps to obtain channel description vectors; The channel description vector is concatenated with the first prior embedded feature to obtain a joint vector; The joint vector is subjected to channel mapping and nonlinear activation processing to generate a channel attention weight vector; Based on the channel attention weight vector, the candidate feature map is recalibrated to obtain channel-enhanced features; Linear mapping is performed on the first prior embedded features to generate a prior space guidance map; The channel enhancement features are convolved to generate an initial spatial attention map; The prior spatial guidance map and the initial spatial attention map are fused to obtain the target spatial attention map; Based on the target spatial attention map, the channel enhancement features are spatially weighted to obtain the target enhancement feature map.

[0008] In some embodiments, the detection head includes a classification branch, a location regression branch, and a structural parameter prediction branch. The step of inputting the target enhanced feature map into the detection head for parameter prediction and outputting the target parameter prediction result includes: The enhanced feature map of the target is input into the detection head; The target enhanced feature map is classified using the classification branch to predict the category and obtain the category probability prediction result. The location regression branch is used to predict the location of the enhanced feature map of the target, and the spatial location parameter prediction result is obtained. The target enhanced feature map is structurally predicted through the structural parameter prediction branch to obtain the structural parameter prediction result. The target parameter prediction result is determined based on the category probability prediction result, the spatial location parameter prediction result, and the structural parameter prediction result.

[0009] In some embodiments, the step of combining the prior parameters of the compression pipe in the structural prior library to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results, and determining the target parameter prediction results that meet the structural consistency requirements as the target identification results, includes: Calculate the structural deviation between the predicted structural parameters and the prior parameters of the compression fitting; When the structural deviation is within a preset threshold range, the target parameter prediction result is determined to be a prediction result that meets the structural consistency requirement, and the target parameter prediction result that meets the structural consistency requirement is determined as the target identification result; When the structural deviation exceeds a preset threshold range, the target parameter prediction result is marked as an abnormal result, and the process returns to the previous steps of acquiring the current transmission line image, inputting the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and performing feature alignment processing on the original feature map to obtain the candidate feature map, until the target parameter prediction result that meets the structural consistency requirement is obtained, and the target parameter prediction result that meets the structural consistency requirement is determined as the target recognition result.

[0010] In some embodiments, the method further includes: If the current transmission line image is abnormal, the first prior embedded feature is corrected according to the candidate feature map to obtain the second prior embedded feature; wherein, the abnormality of the current transmission line image includes the fact that the type of crimp pipe corresponding to the current transmission line image is not in the structure prior library or the appearance change of the crimp pipe presented in the current transmission line image exceeds a preset change threshold. The second prior embedding feature is used as the first prior embedding feature. The process of returning to the structural prior-guided attention module and combining the first prior embedding feature to perform weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map continues until the target parameter prediction result that meets the structural consistency requirement is obtained. The target parameter prediction result that meets the structural consistency requirement is then determined as the target recognition result.

[0011] In some embodiments, after performing structural consistency verification on the structural parameter prediction results in the target parameter prediction results by combining the pipe compression prior parameters in the structural prior library, and determining the target parameter prediction results that meet the structural consistency requirements as the target identification results, the method further includes: The target identification result is sent to the online robot control system, so that the online robot control system generates control instructions based on the target identification result, and controls the robot to perform control of the pressure pipe alignment and X-ray imaging according to the control instructions.

[0012] To achieve the above objectives, another aspect of this application proposes a visual recognition device for transmission line crimping pipes based on structural prior knowledge. The device includes the following modules: The structural prior construction and extraction module is used to construct a structural prior library for transmission line press-fit pipes and extract structural prior vectors from the structural prior library. The prior embedding feature acquisition module is used to perform a nonlinear mapping on the structural prior vector to obtain the first prior embedding feature; The candidate feature map acquisition module is used to acquire the current transmission line image, input the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and perform feature alignment processing on the original feature map to obtain the candidate feature map; The feature map enhancement processing module is used to perform weighted enhancement processing on the candidate feature map based on the structure prior-guided attention module, combined with the first prior embedded features, to obtain the target enhanced feature map; The feature map parameter prediction module is used to input the target enhanced feature map into the detection head for parameter prediction and output the target parameter prediction result. The structural consistency verification module is used to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results by combining the prior parameters of the compression pipe in the structural prior library, and to determine the target parameter prediction results that meet the structural consistency requirements as the target identification results.

[0013] To achieve the above objectives, another aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0014] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0015] To achieve the above objectives, another aspect of this application provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0016] The embodiments of this application include at least the following beneficial effects: This application provides a visual recognition method and related equipment for transmission line crimping pipes based on structural priors. This scheme constructs a structural prior library for transmission line crimping pipes and extracts structural prior vectors from the library; it performs nonlinear mapping on the structural prior vectors to obtain first prior embedding features; it acquires current transmission line images, inputs these images into a backbone network for feature extraction processing to obtain original feature maps, and performs feature alignment processing on the original feature maps to obtain candidate feature maps; based on a structural prior-guided attention module, and combined with the first prior embedding features, it performs weighted enhancement processing on the candidate feature maps to obtain target enhanced feature maps; it inputs the target enhanced feature maps into a detection head for parameter prediction and outputs the target parameter prediction results; it combines the crimping pipe prior parameters in the structural prior library to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results, and determines the target parameter prediction results that meet the structural consistency requirements as the target recognition results. This application constructs a structural prior library for transmission line crimping pipes and extracts structural prior vectors, quantifying the multi-morphological structural knowledge of crimping pipes in standardized vector form, providing a reliable prior constraint basis for subsequent detection processes. An attention module guided by structural priors, combined with prior embedded features, performs weighted enhancement processing on candidate feature maps. This allows the model to make targeted response adjustments based on the expected structural morphology and geometric constraints of the crimping pipe during feature weighting, thereby actively suppressing feature responses in regions inconsistent with structural priors. This ensures that the model can highlight regions consistent with the structural characteristics of the crimping pipe during feature enhancement, effectively improving the model's performance in complex backgrounds. The system enhances the target recognition capability, thereby improving the recognition accuracy of the crimped pipe in complex backgrounds. By inputting the enhanced feature map of the target into the detection head for parameter prediction, the detection head can perform parameter regression on feature representations with high target discrimination and low background interference, which can significantly improve the parameter prediction accuracy. By combining the prior parameters of the crimped pipe in the structural prior library to perform structural consistency verification on the structural parameter prediction results, the prediction results are effectively filtered from the dimensions of geometric rationality and physical feasibility, eliminating the prediction results with structural inconsistencies, and determining the target parameter prediction results that meet the structural consistency requirements as the target recognition results, thus realizing the stable recognition of the crimped pipe under complex working conditions. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of a visual recognition method for transmission line crimping pipes based on structural priors, provided in an embodiment of this application. Figure 2 This is a schematic diagram illustrating the construction process of a priori library and priori features for a press-fit pipe structure provided in an embodiment of this application; Figure 3This is a schematic diagram of an image feature extraction and structure prior-guided attention fusion process provided in an embodiment of this application; Figure 4 This is a schematic diagram of a process for testing and structural consistency constraints of a compression fitting, provided in an embodiment of this application. Figure 5 This is a schematic diagram of the structure of a visual recognition device for transmission line crimping pipes based on structural prior knowledge, provided in an embodiment of this application. Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application.

[0019] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”

[0020] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] Crimping pipes are crucial components in split conductors, enabling both mechanical and electrical connections. The quality of their internal crimping directly affects the conductor's current-carrying capacity, temperature rise, and long-term service safety. Currently, common crimping pipe inspection methods in the industry rely on manual power outages and tower climbing, fixed X-ray imaging, or ground-based visual inspections. These methods generally suffer from low efficiency, reliance on power outage windows, high manual risks, and limited coverage, making it difficult to meet the high efficiency and reliability requirements of daily inspections for ultra-high voltage, long-distance transmission lines. Especially in multi-split conductor structures such as four-split, six-split, and eight-split conductors, the small conductor spacing, large sway, and complex operating environment make it difficult for traditional inspection methods to achieve continuous, stable, and repeatable high-quality testing.

[0023] With the development of drone and online inspection robot technologies, the "drone + X-ray inspection equipment + robot" method for high-altitude non-destructive testing of live wires is gradually becoming an important direction for industry development. However, multi-splittered conductors will continuously oscillate under wind-induced effects, exhibiting multi-angle, non-rigid, and non-repetitive posture changes, resulting in highly unstable scale, texture, shape, and appearance of the crimped pipe in images. Current localization methods based on traditional image processing are difficult to adapt to factors such as complex backgrounds, dynamic lighting, conductor occlusion, and electromagnetic interference, resulting in insufficient recognition accuracy and stability. At the same time, most current deep learning methods rely on a large number of labeled samples for data-driven training, but for targets like crimped pipes with subtle appearance differences, weak texture, and strong dependence on structural features, simply using image features often fails to achieve high-precision localization and reliable recognition.

[0024] Furthermore, different models of crimped pipes exhibit significant differences in geometric dimensions, indentation pitch, crimp section length, and surface texture. This prior engineering information is often not effectively utilized in relevant identification methods, resulting in insufficient generalization ability of the model when faced with different models, split structures, and imaging angles. During flight, UAV-mounted equipment is also subject to interference from high-altitude wind fields, strong electromagnetic environments, and vibrations from the hoisting structure, easily causing blurring, jitter, and noise in visual images, further increasing the difficulty of automatic identification. Due to the lack of visual recognition technology for multi-split conductors, complex attitude changes, and structural priors, UAVs and X-ray robots are currently unable to achieve autonomous positioning, precise alignment, and imaging angle correction of crimped pipes, limiting the promotion and engineering application of high-altitude automated inspection technology under energized conditions.

[0025] In summary, current methods for identifying crimped conductors in transmission lines primarily rely on manual inspections or conventional image processing, which are significantly affected by factors such as lighting variations, conductor swaying, and complex backgrounds, resulting in insufficient identification stability. While traditional deep learning methods have improved detection capabilities, they fail to effectively utilize prior structural information such as the geometric dimensions and indentation pitch of the crimped conductor, leading to weak generalization ability across different models and orientations. Furthermore, multi-split conductor scenarios present challenges such as occlusion and random angle variations, making it difficult for current methods to guarantee identification accuracy and meet the automation and reliability requirements of collaborative inspection by UAVs and robots.

[0026] In view of this, this application provides a visual recognition method and related equipment for transmission line crimping pipes based on structural priors. This scheme constructs a structural prior library for transmission line crimping pipes and extracts structural prior vectors, quantifying the multi-morphological structural knowledge of the crimping pipes in standardized vector form, providing a reliable prior constraint basis for subsequent detection processes. An attention module guided by structural priors, combined with prior embedded features, performs weighted enhancement processing on candidate feature maps. This allows the model to make targeted response adjustments based on the expected structural morphology and geometric constraints of the crimping pipe during feature weighting, thereby actively suppressing regional feature responses inconsistent with structural priors. This ensures that the model can highlight features conforming to the crimping pipe during feature enhancement. By identifying structural features, the model's ability to recognize targets in complex environments is effectively enhanced, thereby improving the recognition accuracy of the press-fit pipe in such environments. By inputting the enhanced feature map of the target into the detection head for parameter prediction, the detection head can perform parameter regression on feature representations with high target discriminability and low background interference, which can significantly improve the accuracy of parameter prediction. By combining the prior parameters of the press-fit pipe in the structural prior library to perform structural consistency verification on the structural parameter prediction results, the prediction results are effectively filtered from the dimensions of geometric rationality and physical feasibility, eliminating prediction results with inconsistent structures, and determining the target parameter prediction results that meet the structural consistency requirements as the target recognition results, thus realizing the stable recognition of the press-fit pipe under complex working conditions.

[0027] This application provides a method for visual recognition of transmission line crimped pipes based on structural priors, relating to the field of artificial intelligence technology. This method can be applied to terminals, servers, or software running on either. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, or vehicle terminal, but is not limited thereto. The server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network. The software can be an application implementing the method for visual recognition of transmission line crimped pipes based on structural priors, but is not limited to the above forms.

[0028] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0029] Please see Figure 1 , Figure 1 This is an optional flowchart of a visual recognition method for transmission line crimped pipes based on structural prior knowledge, provided in an embodiment of this application. Figure 1 The method may include, but is not limited to, steps S101 to S106.

[0030] Step S101: Construct a structural prior library for transmission line press-fit pipes, and extract structural prior vectors from the structural prior library; In practical implementation, based on the crimping tube models used for conductors of different voltage levels and different split numbers, design parameters for each model of crimping tube are collected and organized to construct a structural prior knowledge library. This library uniformly stores the outer diameter of each type of crimping tube. Inner diameter or steel core diameter Total length Length of crimped section Indentation pitch Number of indentations and material type Design parameters, etc.

[0031] For structural prior vectors The structural prior vector is a one-dimensional numerical vector formed by preprocessing design parameters from a structural prior library and then concatenating them in a fixed order. Geometric quantities in the design parameters are normalized using mean subtraction and standardization methods, and material types are processed using one-hot encoding. The processed design parameters are then concatenated into the vector in a fixed order to form the structural prior vector. Specifically, structural prior vectors The calculation formula is as follows: ; In the formula, This indicates transpose.

[0032] In this embodiment, the geometric structure, topological relationship, and size distribution of the crimped pipe are solidified in a structured form to form a structural prior library, which provides clear physical constraints for subsequent reasoning, avoids unreasonable predictions by the network model without structural guidance, and thus improves the detection accuracy of the crimped pipe.

[0033] Step S102: Perform a nonlinear mapping on the structural prior vector to obtain the first prior embedding feature; The first prior embedding feature is a low-dimensional feature vector obtained by performing a nonlinear transformation on the structural prior vector using a multilayer perceptron.

[0034] In this embodiment of the application, in order to facilitate fusion with deep learning network features in the same feature space, the structural prior vector is... By performing nonlinear mapping using a multilayer perceptron, low-dimensional prior embedding features are obtained, which are the first prior embedding features. Among them, the first prior embedded features The calculation formula is as follows: ; In the formula, This represents a coding network with several fully connected layers and non-linear activation functions; Indicates the prior embedding dimension; The learnable parameters of the encoding network; first prior embedding features Used as a guiding signal for subsequent attention modules; This indicates the dimension of the feature space where the embedded feature resides (i.e., the length of the embedding vector).

[0035] In this embodiment, by performing nonlinear mapping dimensionality reduction on the structural prior vector, spatial alignment and dimensionality adaptation of prior features and image features are achieved, enabling structured prior knowledge to be effectively integrated into the network reasoning process, reducing computational redundancy, and significantly improving the processing efficiency and detection accuracy of the network model.

[0036] Step S103: Acquire the current transmission line image, input the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and perform feature alignment processing on the original feature map to obtain the candidate feature map; In this embodiment, the overall network structure consists of a backbone network, a structure-prior-guided attention module, and a detection head. The backbone network extracts basic visual features from the image of the transmission line crimping pipe, performing shallow and deep image information mining to provide raw image features for subsequent modules. The structure-prior-guided attention module uses prior embedded features as guidance to strengthen effective features and suppress useless interference, allowing the network to focus on the target area of ​​the crimping pipe and improve feature recognition. The detection head, based on the features optimized by the attention module, outputs the target location, category, and structural parameter prediction results of the crimping pipe.

[0037] Among them, the current transmission line images are real-world images of the transmission lines obtained by drones or online inspection robots, which are used as the raw input data for the network.

[0038] Additionally, it's important to note that during the network model training phase, the constructed image training dataset needs to fully encompass pipe images from different altitude regions (including plateau environments) and different voltage levels (110kV, 220kV, 500kV, etc.). These pipe image samples include, but are not limited to, various lighting, background, and pose variations for pipes at different voltage levels in plateau environments. This enables the trained network model to accurately identify pipes in plateau environments and at different voltage levels. Similarly, during the network training phase, the transmission line images in the image training dataset are directly used as input to the overall network, sequentially passing through the backbone network, the structure-prior-guided attention module, and the detection head to complete forward propagation. During training, classification loss, location regression loss, and structural consistency loss work together on the overall network parameters, achieving end-to-end optimization from image input to detection result output.

[0039] Optionally, feature alignment processing may include, but is not limited to, multi-scale fusion or upsampling of the feature maps.

[0040] The original feature map consists of multi-scale visual features extracted from the transmission line image by the backbone network. It contains three-dimensional real data of channels, height, and width, and represents the initial feature representation of the image. The candidate feature map is a uniformly sized feature representation obtained after multi-scale fusion or upsampling of the original feature map, and is used for subsequent attention module processing.

[0041] In practice, the first step is to use images of power transmission lines collected by drones or online inspection robots. The input is fed into a deep learning backbone network (such as a convolutional neural network) for feature extraction processing to obtain multi-scale feature maps, which are the original feature maps. Suppose the backbone network outputs one or more feature maps. for: ; In the formula, Indicates the number of channels. and These represent the height and width of the feature map, respectively. This represents the real number space. Then, the original feature map... Multi-scale fusion or upsampling is performed to obtain a feature representation of uniform size, which is the candidate feature map. .

[0042] In this embodiment, by extracting visual features from the image and unifying the size and dimension of multi-scale features, the differences in feature specifications are eliminated, which can ensure the stable processing of the subsequent attention module.

[0043] Step S104: Based on the structure prior-guided attention module, the candidate feature map is weighted and enhanced by combining the first prior embedding features to obtain the target enhanced feature map; In some embodiments, step S104 may include: performing global average pooling on the candidate feature map to obtain a channel description vector; concatenating the channel description vector with a first prior embedded feature to obtain a joint vector; performing channel mapping and nonlinear activation processing on the joint vector to generate a channel attention weight vector; performing channel recalibration on the candidate feature map according to the channel attention weight vector to obtain channel enhancement features; performing linear mapping on the first prior embedded feature to generate a prior spatial guidance map; performing convolution processing on the channel enhancement features to generate an initial spatial attention map; fusing the prior spatial guidance map and the initial spatial attention map to obtain a target spatial attention map; and performing spatial weighting on the channel enhancement features according to the target spatial attention map to obtain a target enhancement feature map.

[0044] To enable the network to focus on regions conforming to the prior knowledge of the pipe fitting structure in both channel and spatial dimensions, this application embodiment designs a structure prior-guided attention module for weighted enhancement of image features. Specifically, the structure prior-guided steps of the attention module are as follows: The specific implementation process of prior guidance in channel attention is as follows: First, the candidate feature map is processed... Perform global average pooling to obtain the channel description vector. Channel description vector The calculation formula is as follows: ; In the formula, This represents the global average pooling operation function. Indicates the number of channels. Represents the space of real numbers.

[0045] Then, the channel description vector With the first prior embedding features Concatenate to obtain the joint vector joint vector The calculation formula is as follows: ; In the formula, This represents the prior embedding dimension.

[0046] Next, the joint vector is processed through two layers of fully connected mapping and a nonlinear activation function. Processing is performed to generate channel attention weight vectors. Channel attention weight vector The calculation formula is as follows: ; In the formula, and Both represent learnable weight matrices. Represents the ReLU function or other nonlinear functions. This represents the Sigmoid function.

[0047] Finally, using the channel attention weight vector For candidate feature maps Channel recalibration was performed to obtain channel enhancement features. Channel enhancement features The calculation formula is as follows: ; In the formula, Indicates the number of channels. and These represent the height and width of the feature map, respectively.

[0048] The specific implementation process of prior guidance in spatial attention is as follows: In order to highlight the area where there may be a pipe in the spatial dimension, the embodiments of this application first embed the first prior feature. A prior space guidance graph is generated through linear mapping. Prior space guidance diagram The calculation formula is as follows: ; In the formula, It is a linear mapping matrix. The operation is used to convert a vector into a spatial graph form.

[0049] At the same time, enhance the channel features pass Convolution generates data-driven initial spatial attention maps Initial spatial attention map The calculation formula is as follows: ; Then, the prior space guidance diagram With data-driven initial spatial attention graph By weighting and fusing, the final target space attention map is obtained. Target space attention map The calculation formula is as follows: ; In the formula, This indicates a learnable or preset balance coefficient.

[0050] Finally, using the target space attention map Channel enhancement features Spatial weighting is performed to obtain a target-enhanced feature map that simultaneously integrates structural priors and data features. Target Enhancement Feature Map The calculation formula is as follows: ; In this embodiment, prior guidance at both the channel and spatial levels makes it easier for the network to focus on regions that conform to the geometric and textural features of the pipe in the subsequent detection stage, thereby improving detection robustness.

[0051] In some embodiments, the method may further include: if an anomaly occurs in the current transmission line image, correcting the first prior embedded feature based on the candidate feature map to obtain a second prior embedded feature; wherein the anomaly in the current transmission line image includes the pipe fitting model corresponding to the current transmission line image not being in the structural prior library or the appearance change of the pipe fitting in the current transmission line image exceeding a preset change threshold; using the second prior embedded feature as the first prior embedded feature, returning to execute the attention module guided by structural prior, combining the first prior embedded feature to perform weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map, until a target parameter prediction result that meets the structural consistency requirement is obtained, and determining the target parameter prediction result that meets the structural consistency requirement as the target recognition result.

[0052] The second prior embedding feature is a corrected feature obtained by superimposing the prediction residual term on the first prior embedding feature, and is used to adapt to the detection of press-fit pipes of unknown models and with significant changes in appearance.

[0053] Anomalies in the current transmission line image include situations where the pipe fitting model corresponding to the current transmission line image is not in the structural prior library, or the appearance of the pipe fitting in the current transmission line image changes beyond a preset threshold. It is understood that, to further improve the network model's recognition performance when facing pipe fittings of unknown models or under special operating conditions, this application introduces a structural prior adaptive correction mechanism. That is, when the pipe fitting model corresponding to the transmission line image input to the backbone network is not in the structural prior library, or when the operating condition causes a significant change in the appearance of the pipe fitting, fixing the prior may introduce bias. Therefore, this application's embodiments design a prior correction module to dynamically adjust the first prior embedded features based on image features. To obtain the corrected second prior embedding features Specifically, the second prior embedding features The calculation formula is as follows: ; In the formula, This represents the original prior embedding feature, i.e., the first prior embedding feature; This represents the corrected prior embedding features, i.e., the second prior embedding features; Indicates the candidate feature map Through lightweight neural networks Predicted residuals; This represents a lightweight neural network, which can be implemented using residual blocks in a multilayer perceptron (MLP), lightweight convolutional neural network (Lightweight CNN), or residual network (ResNet), and is used to predict the correction amount of prior embedded features based on image features; This represents the image feature representation, i.e., the candidate feature map; This represents the learnable parameters of the lightweight neural network. The structure-prior adaptive correction mechanism enables the model to adapt to the unique appearance features of unknown models while maintaining structural prior constraints.

[0054] It should be noted that the residual term The size of the residual term is adaptively determined by the image features. Specifically, when the input image sample has a high degree of matching with the prior structure library, the residual term... Approaching zero, at this time The network model uses the original prior embedding features; when the input image sample is of unknown model or has significant appearance variations, the residual term... The original prior embedding features will be significantly adjusted to achieve adaptive modeling for complex working conditions, resulting in corrected prior embedding features. In such abnormal situations, the subsequent feature fusion and detection processes will uniformly adopt the corrected prior embedding features. Participate in the calculation.

[0055] Step S105: Input the target enhanced feature map into the detection head for parameter prediction, and output the target parameter prediction result; The detection head is a module used for target detection and fine-grained structure prediction of the crimping pipe. Specifically, the detection head, as the end prediction module of the overall neural network, is mounted after the backbone feature extraction network (referred to as the backbone network) and the structure prior-guided attention module, and receives the target enhancement feature map as input. Optionally, the detection head can be implemented using a target detection structure based on a convolutional neural network, such as a detection head similar to YOLO, and form an end-to-end unified network structure with the backbone network and the structure prior-guided attention module. The detection head includes a classification branch, a location regression branch, and a structural parameter prediction branch. The classification branch outputs the category probability of the crimping pipe, the location regression branch predicts the spatial location parameters of the crimping pipe, and the structural parameter prediction branch outputs fine-grained structural features such as indentation pitch and indentation number.

[0056] Optionally, the target parameter prediction results include the category probability prediction results, spatial location parameter prediction results, and structural parameter prediction results corresponding to the press pipe.

[0057] In some embodiments, step S105 may include: inputting the target augmented feature map into the detection head; performing category prediction on the target augmented feature map through a classification branch to obtain a category probability prediction result; performing location prediction on the target augmented feature map through a location regression branch to obtain a spatial location parameter prediction result; performing structure prediction on the target augmented feature map through a structure parameter prediction branch to obtain a structure parameter prediction result; and determining the target parameter prediction result based on the category probability prediction result, the spatial location parameter prediction result, and the structure parameter prediction result.

[0058] For classification and location regression predictions, the specific implementation includes: enhancing the target feature map. Candidate points or candidate boxes are generated on the grid, and the class probability of each location is output through convolutional or fully connected layers. and bounding box parameters The specific calculation formula is as follows: ; In the formula, This represents the probability distribution of whether it is a press-fit pipe and its model category. This indicates the center position and width / height or rotation frame parameters of the compression fitting. This represents the detection header function.

[0059] For structural parameter prediction, the specific implementation includes: to enhance the consistency between the identification results and the prior knowledge of the engineering structure, this embodiment adds a structural parameter prediction branch to the detection head, outputting the indentation pitch. Number of indentations The specific calculation formula for the predicted value of the isostructural parameters is as follows: ; In the formula, This indicates an estimate that includes the structural characteristics of the press-fit pipe. This represents the branch function for predicting structural parameters.

[0060] In this embodiment, the detection head performs synchronous prediction of category, location, and structural parameters through multiple branches in parallel. The fusion of multi-dimensional outputs can improve the completeness and credibility of the results, providing complete data support for subsequent structural consistency verification.

[0061] Step S106: Combine the prior parameters of the compression pipe in the structural prior library to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results, and determine the target parameter prediction results that meet the structural consistency requirements as the target identification results.

[0062] In some embodiments, step S106 may include: calculating the structural deviation between the predicted structural parameters and the prior parameters of the press-fit pipe; when the structural deviation is within a preset threshold range, determining the predicted target parameter as a prediction that meets the structural consistency requirements, and determining the predicted target parameter as a target recognition result; when the structural deviation exceeds the preset threshold range, marking the predicted target parameter as an abnormal result, and returning to the step of acquiring the current transmission line image, inputting the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and performing feature alignment processing on the original feature map to obtain the candidate feature map, until the predicted target parameter that meets the structural consistency requirements is obtained, and determining the predicted target parameter that meets the structural consistency requirements as the target recognition result.

[0063] Among them, the a priori parameter of the press-fit pipe is the outer diameter in the structural prior library. Inner diameter or steel core diameter Total length Length of crimped section Indentation pitch Number of indentations Material type Design parameters, etc.

[0064] In some embodiments, after step S106, the method may further include: sending the target recognition result to the online robot control system, so that the online robot control system generates control instructions based on the target recognition result, and controls the robot to perform control of the pressure pipe alignment and X-ray imaging based on the control instructions.

[0065] In the specific implementation, the predicted structural parameters output by the detection head are compared with the prior structural parameters of the corresponding type of crimping pipe in the prior structural library. The structural deviation between the predicted structural parameters and the prior structural parameters in the library is calculated. When the structural deviation is within a preset threshold range, the predicted structural parameter is deemed to meet the structural consistency requirement and is output as a valid result. This means that when the predicted structural parameter meets the structural consistency requirement, the predicted target parameter corresponding to it also meets the structural consistency requirement, and this target parameter prediction is identified as the target recognition result. When the structural deviation exceeds the preset threshold range, the predicted structural parameter is marked as an abnormal or low-confidence result, and the corresponding target parameter prediction is also marked as an abnormal or low-confidence result. This triggers a resampling, angle adjustment, or manual review mechanism, thereby preventing abnormal recognition results from being directly used for robot alignment and X-ray imaging control. The final target recognition result can be provided to the UAV or online robot control system for crimping pipe alignment, X-ray imaging window selection, and subsequent defect assessment, achieving automated and intelligent detection of crimping pipes in multi-split conductor scenarios.

[0066] In this embodiment of the application, to evaluate the recognition reliability of the network model under unknown operating conditions, an uncertainty estimation branch is introduced to output the uncertainty of each detection result. The specific calculation formula is as follows: ; In the formula, Represents an uncertainty prediction network; This indicates its learnable parameters; Represents the Sigmoid function; This represents the input feature representation, i.e., the target augmented feature map. When uncertainty... When the threshold is exceeded, it indicates that the current recognition result has low credibility. The system can trigger a manual review or supplementary sampling mechanism to ensure the safety of operations under special working conditions.

[0067] It should be noted that during the training phase, the target augmentation feature map The input detection head, along with classification loss, location regression loss, and structural consistency loss, participates in end-to-end training; during the inference phase, the target augmentation feature map... The trained detection head is directly input to output the predicted position, category, and structural parameters of the pressure pipe.

[0068] The specific content regarding structural consistency loss includes: using the true parameters in the structural prior library. For reference, construct structural consistency loss. The specific calculation formula is as follows: ; In the formula, Indicates the predicted pressure pipe section Item structure characteristic parameters; Taking into account the classification loss Location regression loss With structural consistency loss Among them, classification loss For the prediction results of constrained pipe fitting categories, cross-entropy loss or Focal Loss can be used for calculation; location regression loss The position prediction results for constrained pipe boundary boxes or rotated boxes can be calculated using Smooth L1 loss, IoU loss, or GIoU loss, defining the overall loss function. The following calculation formula: ; In the formula, To balance the weighting coefficients of the structural consistency term and the detection term, end-to-end training is employed. This allows the network to actively learn the correspondence between the structural features of the press-fit pipe and prior parameters while performing the target detection task, thereby improving the recognition accuracy under different models, different postures, and complex working conditions.

[0069] It is worth noting that the aforementioned structural consistency constraint only evaluates the credibility of the structural parameter prediction results. In contrast, the uncertainty estimation branch evaluates the credibility of the overall prediction results comprised of location, category, and structural parameters, rather than being used for manual modification of enhanced features. When the uncertainty is below a preset threshold, the system sends the detection results to the robot control system; when the uncertainty is above the preset threshold, the system triggers resampling, adjustment of the shooting angle, or manual verification mechanisms to ensure on-site operational safety and the reliability of the recognition results.

[0070] Steps S101 to S106 as illustrated in this embodiment construct a structural prior library for transmission line crimping pipes and extract structural prior vectors. This quantifies the multi-morphological structural knowledge of the crimping pipe in standardized vector form, providing a reliable prior constraint basis for subsequent detection processes. An attention module guided by structural priors, combined with prior embedded features, performs weighted enhancement processing on candidate feature maps. This allows the model to make targeted response adjustments based on the expected structural morphology and geometric constraints of the crimping pipe during feature weighting, thereby actively suppressing regional feature responses inconsistent with structural priors. This ensures that the model can highlight regions conforming to the structural characteristics of the crimping pipe during feature enhancement, effectively improving... The model enhances target recognition capabilities in complex backgrounds, thereby improving the recognition accuracy of press-fit pipes in such environments. By inputting the enhanced target feature map into the detection head for parameter prediction, the detection head can perform parameter regression on feature representations with high target discriminability and low background interference, significantly improving parameter prediction accuracy. By combining the press-fit pipe prior parameters in the structural prior library to perform structural consistency verification on the structural parameter prediction results, the prediction results are effectively filtered from the dimensions of geometric rationality and physical feasibility, eliminating structurally inconsistent prediction results, and determining the target parameter prediction results that meet the structural consistency requirements as the target recognition results, thus achieving stable recognition of press-fit pipes under complex working conditions.

[0071] To explain in detail the principles of the technical solution of this application, the overall process of this application will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principles of this application and should not be regarded as a limitation of this application.

[0072] This application proposes a visual recognition method for transmission line crimped pipes based on structural priors. By introducing the geometric parameters, structural features, indentation period, and material information of the crimped pipe, and constructing a deep learning recognition model that can adapt to complex high-altitude environments, it achieves high-precision positioning, attitude judgment, and fine-grained recognition of the crimped pipe. This provides reliable detection guidance for UAVs and online inspection robots, and provides key technical support for the automation and intelligent operation of live X-ray inspection of transmission lines.

[0073] In its specific implementation, the visual recognition method for transmission line crimped pipes based on structural priors proposed in this application mainly consists of three parts: The first part involves constructing a structural prior library for transmission line crimped pipes and uniformly encoding parameters such as geometric dimensions and indentation pitch in the library to form structural prior features that can be used for network input; the second part uses a deep learning network to extract features from field-acquired images and integrates structural prior features with image features in a structural prior-guided attention module to highlight the target area of ​​the crimped pipe; the third part performs crimped pipe detection and structural consistency constraints based on the fused features, outputting the location, category, and consistency evaluation with the structural priors of the crimped pipe, achieving stable recognition under complex operating conditions. The specific implementation process of the visual recognition method for transmission line crimped pipes based on structural priors proposed in this application includes the following: Please see Figure 2 , Figure 2 This is a schematic diagram illustrating the construction process of a priori library and priori features for a press-fit pipe structure provided in an embodiment of this application, as shown below. Figure 2 As shown, the specific implementation process of constructing a structural prior library for transmission line crimping pipes and uniformly encoding parameters such as geometric dimensions and indentation pitch in the structural prior library to form structural prior features that can be used for network input includes: First, based on the crimping pipe models used for conductors of different voltage levels and different split numbers, the design parameters of each model of crimping pipe are collected and organized to construct a structural prior library; then, the design parameters in the structural prior library are preprocessed and concatenated into structural prior vectors in a fixed order; then, the structural prior vectors are nonlinearly mapped through a multilayer perceptron to obtain low-dimensional prior embedding features, i.e., the first prior embedding features.

[0074] Please see Figure 3 , Figure 3 This is a schematic diagram of an image feature extraction and structure prior-guided attention fusion process provided in an embodiment of this application, such as... Figure 3 As shown, the specific implementation process of feature extraction from on-site acquired images based on a deep learning network, and the fusion of structural prior features and image features in the structure prior-guided attention module, includes: Figure 2Based on the first prior embedded features, the images of power transmission lines collected by drones or online inspection robots are first input into the deep learning backbone network for feature extraction to obtain the original feature map. The original feature map is then subjected to multi-scale fusion or upsampling to obtain candidate feature maps of uniform size. Subsequently, global average pooling is performed on the candidate feature maps to obtain channel description vectors. Then, the channel description vectors are concatenated with the first prior embedded features to obtain a joint vector. Next, channel mapping and nonlinear activation are performed on the joint vector to generate channel attention weight vectors. Based on the channel attention weight vectors, the candidate feature maps are recalibrated to obtain channel enhancement features. Based on the obtained channel enhancement features, the first prior embedded features are linearly mapped to generate a prior spatial guidance map. Further, the channel enhancement features are convolved to generate an initial spatial attention map. Then, the prior spatial guidance map and the initial spatial attention map are fused to obtain the target spatial attention map. Finally, based on the target spatial attention map, the channel enhancement features are spatially weighted to obtain the target enhancement feature map. By using prior guidance at both the channel and spatial levels, the network can more easily focus on regions that conform to the geometric and textural features of the pipe in the subsequent detection stage, thereby improving detection robustness.

[0075] Please see Figure 4 , Figure 4 This is a schematic diagram illustrating the process of pipe fitting inspection and structural consistency constraints provided in an embodiment of this application. Figure 4 As shown, based on the fusion features, the crimping nozzle is detected and structural consistency constraints are applied. The location and type of the crimping nozzle, as well as the consistency evaluation with the prior structure, are output. The specific implementation process of the stable identification step under complex working conditions includes: Figure 3Based on the obtained target augmented feature map, it is input into the detection head. The classification branch performs category prediction to obtain the category probability prediction result, the location regression branch performs location prediction to obtain the spatial location parameter prediction result, and the structure parameter prediction branch performs structure prediction to obtain the structure parameter prediction result. Then, the category probability prediction result, the spatial location parameter prediction result, and the structure parameter prediction result are combined to form the target parameter prediction result. After obtaining the target parameter prediction result, the structure parameter prediction result output by the detection head is compared with the corresponding model of the crimped pipe in the structure prior library, and the relationship between the structure parameter prediction result and the corresponding model of the crimped pipe in the structure prior library is calculated. Structural deviation is assessed. When the structural deviation is within a preset threshold range, the predicted structural parameter is deemed to meet the structural consistency requirement and output as a valid result. This means that when the predicted structural parameter meets the structural consistency requirement, the predicted target parameter also meets the requirement, and this target parameter prediction is identified as the target recognition result. When the structural deviation exceeds the preset threshold range, the predicted structural parameter is marked as abnormal or low-confidence, and the corresponding target parameter prediction is also marked as abnormal or low-confidence. This triggers resampling, angle adjustment, or manual verification mechanisms, preventing abnormal recognition results from being directly used for robot alignment and X-ray imaging control. The final target recognition result can be provided to the UAV or online robot control system for pipe alignment, X-ray imaging window selection, and subsequent defect assessment, enabling automated and intelligent detection of pipes in multi-split conductor scenarios.

[0076] It should be noted that this embodiment is only a brief illustrative description of the overall process of a visual recognition method for transmission line crimping pipes based on structural priors. Detailed descriptions of each step can be found in the relevant content of the foregoing embodiments, and will not be repeated here. It is understood that this application does not impose any limitations on this.

[0077] In summary, the key advantages of the visual recognition method for transmission line crimping pipes based on structural prior knowledge provided in this application are as follows: (1) When constructing the prior library of press-fit pipe structure, it is necessary to collect key parameters such as outer diameter, length, and indentation pitch for different models of press-fit pipe, and perform unified normalization and coding processing to provide a standardized benchmark for subsequent constraint prediction results.

[0078] (2) In the prior-guided attention fusion process, both channel attention and spatial attention need to introduce structural prior features to ensure that the model feature enhancement process can highlight the regions that conform to the structural features of the press pipe.

[0079] (3) In the crimping pipe detection stage, fine-grained features such as branch output indentation pitch and indentation number are predicted by structural parameters, and consistency constraints are imposed by combining structural prior parameters, which is conducive to improving the identification stability of crimping pipes of different models and different postures.

[0080] This application combines prior structural information such as the geometric dimensions and indentation pitch of the crimped connector with deep learning features, achieving an organic fusion of data-driven training and structural priors. This effectively improves the recognition accuracy and stability of the crimped connector. It is understood that this application can improve the recognition accuracy of crimped connectors in complex backgrounds, reduce interference from environmental changes, enhance adaptability to different types of crimped connectors, improve recognition stability in multi-split conductor scenarios, and output recognition results consistent with structural features, providing reliable visual basis for robot alignment and X-ray imaging. The method provided in this application can achieve reliable positioning and fine-grained recognition of crimped connectors in environments with multi-split conductors, complex lighting, and conductor swaying, providing precise visual guidance for UAVs and inspection robots, and meeting the application requirements of automated inspection of transmission lines under energized conditions.

[0081] Please see Figure 5 This application also provides a visual recognition device 500 for transmission line crimping pipes based on structural prior knowledge, which can implement the above-mentioned method. The device includes the following modules: The structural prior construction and extraction module 501 is used to construct a structural prior library for transmission line press pipes and extract structural prior vectors from the structural prior library. The prior embedding feature acquisition module 502 is used to perform a nonlinear mapping on the structural prior vector to obtain the first prior embedding feature; The candidate feature map acquisition module 503 is used to acquire the current transmission line image, input the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and perform feature alignment processing on the original feature map to obtain the candidate feature map; The feature map enhancement processing module 504 is used to perform weighted enhancement processing on the candidate feature map by combining the first prior embedded features with the structure prior guided attention module to obtain the target enhanced feature map; The feature map parameter prediction module 505 is used to input the target enhanced feature map into the detection head for parameter prediction and output the target parameter prediction result. The structural consistency verification module 506 is used to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results by combining the prior parameters of the compression pipe in the structural prior library, and to determine the target parameter prediction results that meet the structural consistency requirements as the target identification results.

[0082] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0083] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0084] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0085] Please see Figure 6 , Figure 6 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 601 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 602 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 602 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 602 and is called and executed by the processor 601 using the methods described in the embodiments of this application. The input / output interface 603 is used to implement information input and output; The communication interface 604 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 605 transmits information between various components of the device (e.g., processor 601, memory 602, input / output interface 603, and communication interface 604); The processor 601, memory 602, input / output interface 603, and communication interface 604 are connected to each other within the device via bus 605.

[0086] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0087] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0088] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0089] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0090] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0091] This application provides a visual recognition method and related equipment for transmission line crimped pipes based on structural priors. It constructs a structural prior library for transmission line crimped pipes and extracts structural prior vectors, quantifying the multi-morphological structural knowledge of the crimped pipes in standardized vector form, providing reliable prior constraints for subsequent detection processes. An attention module guided by structural priors, combined with prior embedded features, performs weighted enhancement processing on candidate feature maps. This allows the model to make targeted response adjustments based on the expected structural morphology and geometric constraints of the crimped pipe during feature weighting, thereby actively suppressing feature responses in regions inconsistent with structural priors. This ensures that the model can highlight features consistent with the crimped pipe structure during feature enhancement. In areas where the model is in use, the target recognition capability of the model under complex backgrounds is effectively improved, thereby increasing the recognition accuracy of the press-fit pipe under complex backgrounds. By inputting the target enhanced feature map into the detection head for parameter prediction, the detection head can perform parameter regression on feature representations with high target discrimination and low background interference, which can significantly improve the parameter prediction accuracy. By combining the press-fit pipe prior parameters in the structural prior library to perform structural consistency verification on the structural parameter prediction results, the prediction results are effectively filtered from the dimensions of geometric rationality and physical feasibility, eliminating the prediction results with inconsistent structures, and determining the target parameter prediction results that meet the structural consistency requirements as the target recognition results, thus realizing the stable recognition of the press-fit pipe under complex working conditions.

[0092] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0093] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0094] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0095] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0096] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0097] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0098] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0099] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0100] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0101] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A visual recognition method for transmission line crimped pipes based on structural priors, characterized in that, The method includes the following steps: Construct a structural prior library for transmission line press-fit pipes, and extract structural prior vectors from the structural prior library; The first prior embedding feature is obtained by performing a nonlinear mapping on the structural prior vector; The current transmission line image is acquired, and the current transmission line image is input into the backbone network for feature extraction processing to obtain the original feature map. The original feature map is then subjected to feature alignment processing to obtain the candidate feature map. The attention module guided by structural priors combines the first prior embedded features to perform weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map. The enhanced feature map of the target is input into the detection head for parameter prediction, and the target parameter prediction result is output. By combining the prior parameters of the compression pipe in the structural prior library, the structural parameter prediction results in the target parameter prediction results are checked for structural consistency, and the target parameter prediction results that meet the structural consistency requirements are determined as the target identification results.

2. The method according to claim 1, characterized in that, The structure-prior-guided attention module, combined with the first prior embedded features, performs weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map, including: Global average pooling is performed on the candidate feature maps to obtain channel description vectors; The channel description vector is concatenated with the first prior embedded feature to obtain a joint vector; The joint vector is subjected to channel mapping and nonlinear activation processing to generate a channel attention weight vector; Based on the channel attention weight vector, the candidate feature map is recalibrated to obtain channel-enhanced features; Linear mapping is performed on the first prior embedded features to generate a prior space guidance map; The channel enhancement features are convolved to generate an initial spatial attention map; The prior spatial guidance map and the initial spatial attention map are fused to obtain the target spatial attention map; Based on the target spatial attention map, the channel enhancement features are spatially weighted to obtain the target enhancement feature map.

3. The method according to claim 1, characterized in that, The detection head includes a classification branch, a location regression branch, and a structural parameter prediction branch. The step of inputting the target enhanced feature map into the detection head for parameter prediction and outputting the target parameter prediction result includes: The enhanced feature map of the target is input into the detection head; The target enhanced feature map is classified using the classification branch to predict the category and obtain the category probability prediction result. The location regression branch is used to predict the location of the enhanced feature map of the target, and the spatial location parameter prediction result is obtained. The target enhanced feature map is structurally predicted through the structural parameter prediction branch to obtain the structural parameter prediction result. The target parameter prediction result is determined based on the category probability prediction result, the spatial location parameter prediction result, and the structural parameter prediction result.

4. The method according to claim 1, characterized in that, The step of combining the prior parameters of the compression pipe in the structural prior library to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results, and determining the target parameter prediction results that meet the structural consistency requirements as the target identification results, includes: Calculate the structural deviation between the predicted structural parameters and the prior parameters of the compression fitting; When the structural deviation is within a preset threshold range, the target parameter prediction result is determined to be a prediction result that meets the structural consistency requirement, and the target parameter prediction result that meets the structural consistency requirement is determined as the target identification result; When the structural deviation exceeds a preset threshold range, the target parameter prediction result is marked as an abnormal result, and the process returns to the previous steps of acquiring the current transmission line image, inputting the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and performing feature alignment processing on the original feature map to obtain the candidate feature map, until the target parameter prediction result that meets the structural consistency requirement is obtained, and the target parameter prediction result that meets the structural consistency requirement is determined as the target recognition result.

5. The method according to claim 1, characterized in that, The method further includes: If the current transmission line image is abnormal, the first prior embedded feature is corrected according to the candidate feature map to obtain the second prior embedded feature; wherein, the abnormality of the current transmission line image includes the fact that the type of crimp pipe corresponding to the current transmission line image is not in the structure prior library or the appearance change of the crimp pipe presented in the current transmission line image exceeds a preset change threshold. The second prior embedding feature is used as the first prior embedding feature. The process of returning to the structural prior-guided attention module and combining the first prior embedding feature to perform weighted enhancement processing on the candidate feature map to obtain the target enhanced feature map continues until the target parameter prediction result that meets the structural consistency requirement is obtained. The target parameter prediction result that meets the structural consistency requirement is then determined as the target recognition result.

6. The method according to claim 1, characterized in that, After combining the prior parameters of the compression pipe in the structural prior library, performing structural consistency verification on the structural parameter prediction results in the target parameter prediction results, and determining the target parameter prediction results that meet the structural consistency requirements as the target identification results, the method further includes: The target identification result is sent to the online robot control system, so that the online robot control system generates control instructions based on the target identification result and controls the robot to perform the alignment of the pressure pipe and X-ray imaging according to the control instructions.

7. A visual recognition device for transmission line crimping pipes based on structural priors, characterized in that, The device includes the following modules: The structural prior construction and extraction module is used to construct a structural prior library for transmission line press-fit pipes and extract structural prior vectors from the structural prior library. The prior embedding feature acquisition module is used to perform a nonlinear mapping on the structural prior vector to obtain the first prior embedding feature; The candidate feature map acquisition module is used to acquire the current transmission line image, input the current transmission line image into the backbone network for feature extraction processing to obtain the original feature map, and perform feature alignment processing on the original feature map to obtain the candidate feature map; The feature map enhancement processing module is used to perform weighted enhancement processing on the candidate feature map based on the structure prior guided attention module, combined with the first prior embedded features, to obtain the target enhanced feature map; The feature map parameter prediction module is used to input the target enhanced feature map into the detection head for parameter prediction and output the target parameter prediction result. The structural consistency verification module is used to perform structural consistency verification on the structural parameter prediction results in the target parameter prediction results by combining the prior parameters of the compression pipe in the structural prior library, and to determine the target parameter prediction results that meet the structural consistency requirements as the target identification results.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images