A power insulator string detection method and system for multi-scale occlusion suppression

Abstract

The application provides a power insulator string detection method and system for multi-scale occlusion suppression, which comprises the following steps: acquiring a visible light image of a power transmission line containing an insulator string; inputting the visible light image into a trained insulator string detection model to obtain an insulator string detection result; the model training process comprises the following steps: acquiring training data for visible light power transmission line insulator string detection; constructing an insulator string detection model, which processes the input visible light image through a double-main-stem multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map; suppressing invalid responses in the occlusion area through an occlusion suppression attention module; adaptively fusing four levels of features output by the occlusion suppression module through a four-branch adaptive scale fusion detection head to obtain an insulator string rotation frame detection result; training the insulator string detection model according to the training data and outputting the trained insulator string detection model.

Description

Technical Field

[0001] This invention belongs to the field of power insulator string detection, specifically relating to a method and system for multi-scale shading suppression. Background Technology

[0002] Insulators, as critical components in power transmission lines, directly affect the electrical isolation and mechanical safety of the lines. With the widespread adoption of drone inspection technology, visible light images have become a crucial data source for power equipment inspection. In real-world engineering scenarios, the scale of insulator strings in visible light images varies dramatically: long-distance shooting makes insulators appear as small-scale targets with indistinct edge details, while close-up or side-view shooting may reveal large-scale, slender structures with significant tilt angles. Furthermore, complex interference such as forest backgrounds, tower components, conductors, and tangled vegetation can cause various types of occlusion, making the acquisition of effective features even more difficult. Although existing deep learning-based methods incorporate attention, multi-scale pyramid, or rotating box strategies to enhance expressive power, they still struggle to achieve effective occlusion recovery and cross-scale feature consistency in visible light inspection environments with large scale spans and complex occlusion types, limiting the accuracy and robustness of detection. Summary of the Invention

[0003] To overcome the problem that existing deep learning-based insulator string detection methods struggle to achieve effective shading recovery and cross-scale feature consistency in visible light inspection environments with large scale spans and complex shading types, resulting in insufficient detection accuracy, this invention provides a power insulator string detection method and system oriented towards multi-scale shading suppression. By enhancing the network's robust feature modeling capability for shading areas and constructing a cross-scale adaptive fusion mechanism, it achieves more accurate and stable insulator localization and orientation estimation in complex environments.

[0004] According to one aspect of the present invention, a method for detecting power insulator strings for multi-scale shading suppression is provided, comprising:

[0005] Acquire visible light images of a transmission line containing insulator strings;

[0006] The visible light image is input into the trained insulator string detection model to obtain the insulator string detection result;

[0007] The training process of the insulator string detection model includes:

[0008] To obtain training data for visible light transmission line insulator string detection;

[0009] An insulator string detection model is constructed. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. An occlusion suppression attention module is used to suppress invalid responses in occluded regions in the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result.

[0010] Train an insulator string detection model based on the training data, and output the trained insulator string detection model.

[0011] As a further technical solution, the dual-backbone multi-level auxiliary feature extraction structure includes two parts: a main branch and an auxiliary branch. The main branch uses a YOLO11 backbone network with a four-level downsampling structure to process the input inspection image and extract features layer by layer to obtain multi-level semantic feature maps. The auxiliary branch uses a multi-level auxiliary feature pyramid composed of several CBLinear modules and several CBFuse modules to enhance the local texture information of the multi-level semantic feature maps across levels and output a multi-level semantic-texture fusion feature map.

[0012] As a further technical solution, a multi-level auxiliary feature pyramid enhances the local texture information across multiple levels of the semantic feature map, including:

[0013] Starting from the second layer, the semantic feature map extracted by the backbone network is subjected to lightweight convolutional mapping through the CBLinear module, and the auxiliary features corresponding to different layers are obtained by splitting them according to the channel dimension. The CBFuse module aligns the auxiliary features of different layers to the spatial size of the semantic feature map of the current layer, and sums and fuses them with the semantic features of the current layer to obtain the semantic-texture fusion feature map of the current layer.

[0014] As a further technical solution, the occlusion suppression attention module extracts local structural features from the acquired multi-level semantic-texture fusion feature maps through the CSMM cross-stage multi-level aggregation module to obtain local enhanced feature maps at each level; global average pooling and fully connected mapping are performed on the local enhanced feature maps to generate channel recalibration weights for the corresponding level; the channel recalibration weights are used to adjust the channel-by-channel weights of the semantic-texture fusion feature maps at the corresponding level to obtain multi-level semantic-texture fusion feature maps after suppressing interference responses in occluded regions.

[0015] As a further technical solution, the CSMM cross-stage multi-level aggregation module performs depthwise separable convolution on the high-dimensional local feature sequences of each level through parallel CSMM branches to extract spatial local features, and performs cross-channel fusion through pointwise convolution. After normalization and activation processing, it outputs the local structural feature maps of each level.

[0016] As a further technical solution, the four-branch adaptive scale fusion detection module includes a multi-scale feature alignment module and a four-branch adaptive scale fusion detection head. The multi-scale feature alignment module receives multi-level semantic-texture fusion feature maps that suppress occlusion and aligns the features at different levels with spatial resolution. The four-branch adaptive scale fusion detection head processes the aligned features at the four levels respectively and performs dynamic fusion based on scale response weights to output the insulator string rotation frame detection result.

[0017] As a further technical solution, the four-branch adaptive scale fusion detection head includes a first scale branch, a second scale branch, a third scale branch, a fourth scale branch, and a detection network. The four scale branches generate scale response weights for corresponding levels through lightweight convolution modules, and dynamically weight and fuse the output features of the four scale branches according to the scale response weights to obtain the target fusion features. The detection network outputs the rotation bounding box detection results of the insulator string based on the target fusion features, including the center point coordinates, width, height, and rotation angle.

[0018] According to another aspect of this specification, a power insulator string detection system for multi-scale shading suppression is provided, characterized in that it includes:

[0019] The input module is used to acquire visible light images of transmission lines containing insulator strings;

[0020] The insulator string detection module is used to input visible light images into the trained insulator string detection model to obtain insulator string detection results;

[0021] The training process of the insulator string detection model includes:

[0022] To obtain training data for visible light transmission line insulator string detection;

[0023] An insulator string detection model is constructed. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. An occlusion suppression attention module is used to suppress invalid responses in occluded regions in the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result.

[0024] Train an insulator string detection model based on the training data, and output the trained insulator string detection model.

[0025] According to another aspect of this specification, an electronic device is provided, including a memory and a processor, the memory storing program instructions executed by the processor, the processor invoking the program instructions to perform a method for detecting power insulator strings for multi-scale shading suppression.

[0026] According to another aspect of this specification, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute a method for detecting power insulator strings oriented towards multi-scale occlusion suppression.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] A dual-backbone multi-level auxiliary feature extraction structure is adopted to achieve cross-level local texture enhancement and fusion, which can preserve the local morphological features of individual insulator pieces and capture the macroscopic arrangement rules of insulator strings, thereby improving the model's adaptability to changes in insulator scale at different shooting distances.

[0029] By introducing an occlusion suppression attention module to extract multi-scale local structural features, the expression of the repeatability of insulator disc edges, the geometric constraints of hardware, and the axial continuity of the string is enhanced, and the interference response of occluded areas and complex background areas is suppressed.

[0030] By designing a four-branch adaptive scale fusion detection head, dynamic weighted fusion of multi-scale features is achieved, enabling the model to adaptively adjust the contribution of features at different levels according to the target scale. Combined with a rotation box regression strategy, the rotation box positioning results adapted to insulator strings with different tilt angles are output, thereby improving the direction estimation and positioning accuracy. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating a method for detecting power insulator strings with multi-scale shading suppression, provided in an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of the dual-backbone multi-level auxiliary feature extraction structure in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of a four-branch adaptive scale fusion detection head in an embodiment of the present invention;

[0035] Figure 4 This is an example diagram showing the detection results of power insulator strings in an embodiment of the present invention;

[0036] Figure 5 This is a schematic diagram of a power insulator string detection system for multi-scale occlusion suppression provided in an embodiment of the present invention;

[0037] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0038] It should be noted that:

[0039] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0040] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0041] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0042] likeFigure 1 As shown, this invention provides a method for detecting power insulator strings for multi-scale shading suppression, comprising the following steps:

[0043] Step 1: Obtain a visible light image of the transmission line containing insulator strings;

[0044] Step 2: Input the visible light image into the trained insulator string detection model to obtain the insulator string detection results;

[0045] The training process of the insulator string detection model includes:

[0046] Step 2-1: Obtain training data for visible light transmission line insulator string detection;

[0047] Step 2-2: Construct an insulator string detection model. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. The occlusion suppression module suppresses invalid responses in the occlusion region of the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result.

[0048] Steps 2-3: Train the insulator string detection model based on the training data, and output the trained insulator string detection model.

[0049] For example, in step 1, the acquisition of visible light images of transmission lines containing insulator strings in this embodiment of the invention is supported by drone aerial inspection.

[0050] Specifically, in step 2-1, the training data includes multiple visible light images of transmission lines containing insulator strings, as well as rotating bounding box annotation information of the insulator strings corresponding to the visible light images.

[0051] Firstly, such as Figure 2 As shown, in step 2-2, a dual-backbone structure is constructed with the YOLO11 backbone network as the main branch and the CBLinear / CBFuse module superposition as the auxiliary branch. This achieves cross-layer compensation and multi-scale feature enhancement of deep semantics and shallow texture, providing multi-scale and hierarchical consistent basic feature representation for the occlusion suppression module and the detection head.

[0052] The dual-backbone multi-level auxiliary feature extraction structure includes a main branch and an auxiliary branch. The main branch uses a YOLO11 backbone network with a four-level downsampling structure to process the input visible light inspection image. It extracts features layer by layer through convolutional modules and C3k2 modules to obtain multi-level semantic feature maps. The auxiliary branch uses several CBLinear modules and several CBFuse modules to form a multi-level auxiliary feature pyramid. It performs channel adjustment, spatial size alignment and cross-level local texture information enhancement on the multi-level semantic feature maps to output a multi-level semantic-texture fusion feature map.

[0053] Furthermore, the multi-level auxiliary feature pyramid enhances the local texture information across multiple levels of the multi-level semantic feature map, including:

[0054] Starting from the second layer, the semantic feature map extracted by the backbone network is subjected to lightweight convolutional mapping through the CBLinear module, and the auxiliary features corresponding to different layers are obtained by splitting them according to the channel dimension. The CBFuse module aligns the auxiliary features of different layers to the spatial size of the semantic feature map of the current layer, and sums and fuses them with the semantic features of the current layer to obtain the semantic-texture fusion feature map of the current layer.

[0055] Specifically, in this embodiment of the invention, the main branch adopts a four-level downsampling YOLOv11 backbone network, which captures the longitudinal arrangement pattern, overall directionality, and macroscopic structure of the insulator's disc sequence through progressive convolution and feature compression. The auxiliary branch uses a multi-level auxiliary feature pyramid composed of CBLinear and CBFuse modules to enhance the expression of local morphological features of individual insulator discs.

[0056] The CBLinear module performs lightweight convolutional mapping (usually 1×1 convolution) on the semantic feature map to obtain auxiliary features at different levels. These auxiliary features are used to provide local texture enhancement information to the current level to supplement visual cues such as disc edges, cracks, or surface stains in insulator detection.

[0057] The CBFuse module is used for cross-branch alignment and fusion: It receives auxiliary features distributed from CBLinear and semantic feature maps of the current level's main branch. It fuses the auxiliary and main features through spatial size alignment and summation, outputting a semantic-texture fusion feature map for that level to improve the consistency and expressive strength of multi-scale features. This enables interactive compensation between features at different levels across different directions, allowing deep semantic information and shallow detail information to complement each other, thus constructing feature sets with multi-scale expressive capabilities. This provides input features for subsequent occlusion suppression attention modules and adaptive scale fusion detection heads. Additionally, the CBFuse module includes spatial size alignment functionality. When there is a spatial size difference between the auxiliary and main features of the current level, the CBFuse module uses interpolation to adjust the spatial size of the auxiliary features to match that of the main features of the current level.

[0058] Secondly, in step 2-2, in order to solve the problem of local occlusion and edge loss of insulator strings under the interference of vegetation, tower components and conductors, this invention introduces an occlusion suppression attention module in the high-resolution fine-grained layer and deep semantic layer to suppress invalid responses in occluded areas in the multi-level semantic-texture fusion feature map.

[0059] The occlusion suppression attention module extracts local structural features from the acquired multi-level semantic-texture fusion feature maps through the CSMM cross-stage multi-level aggregation module, obtaining local enhanced feature maps for each level. Global average pooling and fully connected mapping are then applied to these local enhanced feature maps to generate channel recalibration weights for the corresponding levels. These channel recalibration weights are then used to adjust the channel-by-channel weights of the semantic-texture fusion feature maps for the corresponding levels, resulting in multi-level semantic-texture fusion feature maps after suppressing interference responses in occluded regions.

[0060] The CSMM cross-stage multi-level aggregation module extracts spatial local features by performing depthwise separable convolutions on the high-dimensional local feature sequences of each level through parallel CSMM branches, and performs cross-channel fusion through pointwise convolutions. After normalization and activation processing, it outputs the local structural feature maps of each level.

[0061] As a supplementary explanation, the occlusion suppression attention module performs local convolution mapping on multi-level semantic-texture fusion feature maps to obtain local enhancement feature maps at each level. Insulators are composed of multiple similarly shaped discs strung together. Local features may include one or more disc edges. This local feature extraction allows the model to capture the periodic texture changes at the disc edges, highlighting the repetitiveness of the insulator disc edges. Fittings typically have specific geometric shapes, and local feature extraction can enhance the edge responses of these specific shapes, forming geometric constraint signals. The CSMM module processes the local enhancement feature maps at each level through depthwise separable convolution and pointwise convolution, aggregating local features into local structural features that reflect spatial neighborhood relationships. This learns the axial continuity of the insulator string, maintains the axial continuity prediction, and reduces the risk of detection box offset or breakage due to local occlusion. Therefore, the multi-level local structural features output by the CSMM cross-stage multi-level aggregation module can highlight the repetitiveness of insulator disc edges, the geometric constraints of fitting connections, and the axial continuity of the string, enhancing the model's ability to perceive occluded areas.

[0062] The multi-scale outputs are processed by global average pooling and fully connected mapping to generate channel recalibration weights, and the weights of each channel in the fused feature map are adjusted accordingly. This recalibration operation suppresses invalid responses in occluded areas while enhancing effective features in unoccluded areas, such as insulator disc edges, hardware connections, and longitudinal axial structures. During this process, the model adaptively adjusts the weights based on the response differences of different channels, thus maintaining the continuity of the insulator string along its long axis and avoiding the risk of directional prediction shifts due to local missing features.

[0063] Third, in step 2-2, the four-branch adaptive scale fusion detection module includes a multi-scale feature alignment module and a four-branch adaptive scale fusion detection head. Through the multi-scale feature alignment module, multi-level semantic-texture fusion feature maps with suppressed occlusion are received, and the features at different levels are aligned in terms of spatial resolution. The four-branch adaptive scale fusion detection head processes the aligned features at the four levels respectively, and performs dynamic fusion based on the scale response weights to output the insulator string rotation frame detection result.

[0064] Among them, such as Figure 3As shown, the four-branch adaptive scale fusion detection head includes a first scale branch, a second scale branch, a third scale branch, and a fourth scale branch, corresponding to the small target detection layer (level 1), medium target detection layer (level 2), auxiliary target detection layer (level 3), and large target detection layer (level 4) in the figure. The first scale branch, second scale branch, third scale branch, and fourth scale branch respectively receive the occlusion suppression feature maps of the four levels after spatial resolution alignment, and are used to extract the response features of the insulator string targets within different scale ranges. The four scale branches generate scale response weights for the corresponding levels through lightweight convolution modules, and dynamically weight and fuse the output features of the four scale branches according to the scale response weights to obtain the target fusion features; the detection network (including auxiliary feature fusion detection heads corresponding to each scale branch) outputs the rotation bounding box detection results of the insulator string based on the target fusion features, and the rotation bounding box detection results include the center point coordinates, width, height, and rotation angle.

[0065] Preferably, the multi-scale feature alignment module jointly performs scale transformation through a differentiable spatial transformation operator and a convolution adjustment module, thereby achieving accurate alignment while preserving the feature geometry. Compared with traditional interpolation methods, the differentiable mapping method can reduce feature misalignment and boundary discontinuity problems, ensuring that the spatial correspondence of the insulator disk's directional texture, fine-grained edges, and axial structure is consistent across scales, providing a more stable foundation for subsequent scale fusion.

[0066] Specifically, based on the output of the multi-scale feature alignment module, this invention constructs a four-branch adaptive scale fusion detection head, such as... Figure 3 As shown, the four-branch adaptive scale fusion detection head uses a lightweight convolutional module to estimate the scale fusion weights for features at different levels, and adaptively calculates the contribution weight of each level based on the response differences of features at different levels. This weight can enhance the feature response that matches the current target scale and suppress the interference response of mismatched scale levels, thus achieving scale optimization. Subsequently, the features at each level are dynamically weighted and fused according to the learned spatial attention weights, making deep semantic features and shallow detail features complementary.

[0067] Specifically, in steps 2-3, the training and loss strategy employs a joint optimization approach combining classification loss and rotating frame regression loss. The rotating frame regression loss, composed of probIoU, DFL, and repulsion loss, is used to improve the regression accuracy of the insulator string orientation angle, center position, and major axis length. probIoU helps maintain stable overlap estimation under partial occlusion conditions, DFL improves the regression accuracy of positioning parameters such as center position and width / height, and repulsion loss is used to avoid frame overlap and erroneous association between adjacent insulator discs.

[0068] This invention employs a rotation box optimization strategy incorporating probIoU, DFL, and repulsion losses to achieve compact regression of the insulator string's attitude. This loss combination effectively reduces regression jitter caused by orientation shifts, scale inconsistencies, or occlusions, allowing the predicted bounding box to better fit the longitudinal arrangement of the insulator string, thus improving orientation estimation accuracy and positioning robustness. Finally, the features, after multi-scale adaptive fusion and optimization, are input to the detection head, outputting the rotation box positioning result of the insulator string, achieving high-precision orientation detection in complex scenarios.

[0069] Preferably, by setting a classification prediction branch in the detection head and configuring insulator category labels in the training data, the model proposed in this invention can output the category of the insulator string while outputting the rotation frame positioning result of the insulator string. Figure 4 The figure shows the detection results of insulator strings according to a preferred embodiment of the present invention. The various types of insulators identified in the image are classified and labeled by different colored bounding boxes and labels: glass insulator, fuhe insulator, and taoci insulator.

[0070] The implementation of the various embodiments of the present invention is based on programmed processing through a system with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a power insulator string detection system for multi-scale shading suppression. This system is used to execute a power insulator string detection method for multi-scale shading suppression from the above method embodiments.

[0071] See Figure 5 The system includes:

[0072] The input module is used to acquire visible light images of transmission lines containing insulator strings;

[0073] The insulator string detection module is used to input visible light images into the trained insulator string detection model to obtain insulator string detection results;

[0074] The training process of the insulator string detection model includes:

[0075] To obtain training data for visible light transmission line insulator string detection;

[0076] An insulator string detection model is constructed. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. An occlusion suppression attention module is used to suppress invalid responses in occluded regions in the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result.

[0077] Train an insulator string detection model based on the training data, and output the trained insulator string detection model.

[0078] It should be noted that the system embodiments provided by the present invention are used not only to implement the methods in the above method embodiments, but also to implement the methods in other method embodiments provided by the present invention. The only difference is that corresponding functional modules are set. The principle is basically the same as that of the above system embodiments provided by the present invention. As long as those skilled in the art can improve the modules in the above system embodiments by referring to the specific technical solutions in other method embodiments and combining technical features to obtain corresponding technical means and technical solutions composed of these technical means, on the basis of the above system embodiments, and on the premise of ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments.

[0079] The method described in this invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of this invention provide an electronic device, such as... Figure 6 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0080] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0081] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0082] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] Based on the same technical concept as the foregoing embodiments, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions that cause the computer to execute a method.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting power insulator strings for multi-scale shading suppression, characterized in that, include: Acquire visible light images of a transmission line containing insulator strings; The visible light image is input into the trained insulator string detection model to obtain the insulator string detection result; The training process of the insulator string detection model includes: To obtain training data for visible light transmission line insulator string detection; An insulator string detection model is constructed. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. An occlusion suppression attention module is used to suppress invalid responses in occluded regions in the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result. Train an insulator string detection model based on the training data, and output the trained insulator string detection model.

2. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 1, characterized in that, The dual-backbone multi-level auxiliary feature extraction structure includes two parts: a main branch and an auxiliary branch. The main branch uses a four-level downsampling YOLO11 backbone network to process the input inspection image and extracts features layer by layer to obtain a multi-level semantic feature map. The auxiliary branch uses a multi-level auxiliary feature pyramid composed of several CBLinear modules and several CBFuse modules to enhance the local texture information of the multi-level semantic feature map across levels and output a multi-level semantic-texture fusion feature map.

3. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 2, characterized in that, Multi-level auxiliary feature pyramids enhance cross-level local texture information in multi-level semantic feature maps, including: Starting from the second layer, the semantic feature map extracted by the backbone network is subjected to lightweight convolutional mapping through the CBLinear module, and the auxiliary features corresponding to different layers are obtained by splitting them according to the channel dimension. The CBFuse module aligns the auxiliary features of different layers to the spatial size of the semantic feature map of the current layer, and sums and fuses them with the semantic features of the current layer to obtain the semantic-texture fusion feature map of the current layer.

4. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 1, characterized in that, The occlusion suppression attention module extracts local structural features from the acquired multi-level semantic-texture fusion feature maps through the CSMM cross-stage multi-level aggregation module to obtain local enhanced feature maps for each level. Global average pooling and fully connected mapping are then applied to the local enhanced feature maps to generate channel recalibration weights for the corresponding levels. These channel recalibration weights are then used to adjust the channel-by-channel weights of the semantic-texture fusion feature maps for the corresponding levels, resulting in multi-level semantic-texture fusion feature maps after suppressing interference responses in occluded regions.

5. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 4, characterized in that, The CSMM cross-stage multi-level aggregation module performs depthwise separable convolution on the semantic-texture fusion feature maps of each level through parallel CSMM branches to extract local spatial features, and performs cross-channel fusion through pointwise convolution. After normalization and activation processing, it outputs the local structural feature maps of each level.

6. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 1, characterized in that, The four-branch adaptive scale fusion detection module includes a multi-scale feature alignment module and a four-branch adaptive scale fusion detection head. The multi-scale feature alignment module receives multi-level semantic-texture fusion feature maps that suppress occlusion and aligns the features at different levels with spatial resolution. The four-branch adaptive scale fusion detection head processes the aligned features at the four levels respectively and performs dynamic fusion based on scale response weights to output the insulator string rotation frame detection result.

7. The method for detecting power insulator strings for multi-scale shading suppression as described in claim 6, characterized in that, The four-branch adaptive scale fusion detection head includes a first scale branch, a second scale branch, a third scale branch, a fourth scale branch, and a detection network. The four scale branches generate scale response weights for corresponding levels through a lightweight convolution module, and the output features of the four scale branches are dynamically weighted and fused according to the scale response weights to obtain the target fusion features. The detection network outputs the rotational bounding box detection results of the insulator string based on the target fusion features, including the center point coordinates, width, height and rotation angle.

8. A power insulator string detection system for multi-scale shading suppression, characterized in that, include: The input module is used to acquire visible light images of transmission lines containing insulator strings; The insulator string detection module is used to input visible light images into the trained insulator string detection model to obtain insulator string detection results; The training process of the insulator string detection model includes: To obtain training data for visible light transmission line insulator string detection; An insulator string detection model is constructed. The insulator string detection model processes the input visible light image through a dual-main-level multi-level auxiliary feature extraction structure to obtain a multi-level semantic-texture fusion feature map. An occlusion suppression attention module is used to suppress invalid responses in occluded regions in the multi-level semantic-texture fusion feature map. The output of the occlusion suppression module is processed by a four-branch adaptive scale fusion detection head to obtain the insulator string detection result. Train an insulator string detection model based on the training data, and output the trained insulator string detection model.

9. An electronic device, characterized in that, The method includes a memory and a processor, the memory storing program instructions that are executed by the processor, the processor invoking the program instructions to perform the method according to any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the method described in any one of claims 1 to 7.

Images