A low-altitude airspace electric wire detection method, device, equipment, medium and product

Abstract

The application discloses a low-altitude airspace electric wire detection method, device, equipment, medium and product, relates to the airworthiness field of an aircraft, and the method comprises the following steps: training an electric wire detection model by training a region rebalancing convolutional neural network based on an elongated target self-adaption according to a low-altitude airspace electric wire data set; the region rebalancing convolutional neural network based on the elongated target self-adaption comprises a backbone network and a segmentation head which are connected in sequence; the backbone network is an HRNet network in which 3x3 standard convolution in a bottleneck module is dynamic snake convolution; the segmentation head comprises a segmentation head network and a region rebalancing branch module; the input end of the region rebalancing branch module and the input end of the segmentation head network are both connected with the output end of the backbone network, and the second input end of the region rebalancing branch module is used for inputting a mask label, so that the application can improve the accuracy of low-altitude airspace electric wire detection and reduce the risk of wire collision when the aircraft is operating at a low altitude.

Description

Technical Field

[0001] The present application relates to the field of aircraft airworthiness, and in particular to a method, device, equipment, medium and product for detecting electric wires in low-altitude airspace. Background Art

[0002] Avoiding collisions with high-voltage power lines for low-altitude aircraft has become a critical issue for low-altitude safety and airworthiness. Sense and avoid (S&A) capability is essential for pilots to perceive the terrain and promptly avoid obstacles that could threaten flight safety. For low-altitude aircraft, improving S&A capabilities independently of the pilot can effectively enhance their safety and airworthiness. In S&A, perception (S) is the prerequisite and foundation for avoidance (A). However, due to the large number of low-altitude aircraft and their low cost and small payload, they lack reliable and reliable S&A capabilities. Large low-altitude aircraft primarily rely on primary and secondary air traffic control radars for terrain obstacle perception and avoidance, primarily through specialized equipment such as Automatic Dependent Surveillance (ADS) and Traffic Alert and Collision Avoidance System (TACAS). However, these devices are not suitable for light, low-altitude aircraft. Therefore, improving the autonomous sensing capability of low-altitude aircraft to detect low-altitude high-voltage wires and enhance their safety and airworthiness is of great significance. Therefore, research on a low-altitude airspace wire detection method that improves the accuracy of low-altitude airspace wire detection is of great significance to reducing the risk of wire collisions during low-altitude aircraft operations. Summary of the Invention

[0003] The purpose of this application is to provide a low-altitude airspace wire detection method, device, equipment, medium and product, which can improve the accuracy of low-altitude airspace wire detection and reduce the risk of aircraft hitting the wire during low-altitude operations.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a method for detecting electric wires in low-altitude airspace, comprising:

[0006] Acquire a low-altitude airspace wire dataset; the low-altitude airspace wire dataset includes multiple samples, each sample is an external view image of a sample low-altitude aircraft and a corresponding mask label; the external view image includes wires; the mask label is a mask image with wire boundaries;

[0007] According to the low-altitude airspace wire dataset, a regional rebalancing convolutional neural network based on slender target adaptation is trained to obtain a wire detection model, which is used to perform wire detection in the low-altitude airspace; the regional rebalancing convolutional neural network based on slender target adaptation includes a backbone network and a segmentation head connected in sequence; the backbone network is an HRNet network in which the 3x3 standard convolution in the Bottleneck module is a dynamic snake convolution; the segmentation head includes a segmentation head network and a regional rebalancing branch module; the first input end of the regional rebalancing branch module and the input end of the segmentation head network are both connected to the output end of the backbone network, and the second input end of the regional rebalancing branch module is used to input a mask label.

[0008] Optionally, the wire detection model is used to perform wire detection in low-altitude airspace, specifically including:

[0009] Acquire external visual images of the low-altitude aircraft to be detected;

[0010] The external visual image of the low-altitude aircraft to be detected is input into the wire detection model without the regional rebalancing branch module to obtain a mask image with wire boundaries corresponding to the external visual image of the low-altitude aircraft to be detected.

[0011] Optionally, the segmentation head network is a depth-wise separable convolution.

[0012] Optionally, a wire detection model is obtained by training a regional rebalancing convolutional neural network based on slender target adaptation according to the low-altitude airspace wire dataset, specifically comprising:

[0013] Using a data enhancement algorithm to perform data enhancement processing on the low-altitude airspace power line dataset to obtain an enhanced dataset;

[0014] With the goal of minimizing the loss function value, an enhanced dataset is used to train a regional rebalancing convolutional neural network based on slender target adaptation to obtain a wire detection model.

[0015] Optionally, the HRNet network is HRNet-w48.

[0016] Optionally, the loss function is:

[0017] in represents the loss function, represents the regional loss, represents pixel cross entropy loss, λ represents The coefficient of Calculated based on the output of the region rebalancing branch module and the mask label, Calculated based on the output of the segmentation head network and the mask labels.

[0018] In a second aspect, the present application provides a low-altitude airspace power line detection device, comprising:

[0019] An acquisition module is configured to acquire a low-altitude airspace power line dataset; the low-altitude airspace power line dataset comprises a plurality of samples, each sample being an external view image of a sample low-altitude aerial vehicle and a corresponding mask label; the external view image comprises a power line; and the mask label is a mask image with a power line boundary.

[0020] A training module is configured to train an elongated target adaptive region rebalancing convolutional neural network based on the low-altitude airspace power line dataset to obtain a power line detection model, which is used for power line detection in a low-altitude airspace; the elongated target adaptive region rebalancing convolutional neural network comprises a backbone network and a segmentation head connected in sequence; the backbone network is an HRNet network in which 3x3 standard convolution in a Bottleneck module is replaced by dynamic snake convolution; the segmentation head comprises a segmentation head network and a region rebalancing branch module; a first input end of the region rebalancing branch module and an input end of the segmentation head network are both connected to an output end of the backbone network, and a second input end of the region rebalancing branch module is configured to input a mask label.

[0021] In a third aspect, the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the low-altitude airspace power line detection method according to any one of the above.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program executable by a processor to implement the low-altitude airspace power line detection method according to any one of the above.

[0023] In a fifth aspect, the present application provides a computer program product, comprising a computer program executable by a processor to implement the low-altitude airspace power line detection method according to any one of the above.

[0024] According to the embodiments provided in the present application, the present application has the following technical effects:

[0025] The present application provides a low-altitude airspace power line detection method, device, equipment, medium and product, the HRNet fused with DSConv is used in the elongated target adaptive region rebalancing convolutional neural network of the present application, which ensures high resolution of images and pays more attention to features of elongated targets, thereby improving power line detection accuracy; the region rebalancing branch module is used to solve the imbalance between positive and negative samples, improve accuracy, reduce the rate of missed detection, improve the accuracy of low-altitude airspace power line detection, and reduce the risk of line collision when the aerial vehicle operates at low altitude. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 A flowchart of a low-altitude airspace wire detection method provided in one embodiment of the present application;

[0028] Figure 2 A flowchart of low-altitude airspace wire detection using a wire detection model provided in one embodiment of the present application;

[0029] Figure 3 This is a schematic diagram of the specific principle of the low-altitude airspace wire detection process;

[0030] Figure 4 This is the first data schematic diagram of the low-altitude airspace power line dataset;

[0031] Figure 5 This is a second data schematic diagram of the low-altitude airspace power line dataset;

[0032] Figure 6 This is the third data diagram in the low-altitude airspace power line dataset;

[0033] Figure 7 Schematic diagram of the regional rebalancing convolutional neural network structure based on slender target adaptation;

[0034] Figure 8 This is a schematic diagram of the backbone network structure in the wire detection model;

[0035] Figure 9 This is a schematic diagram of the structure of the segmentation head in the wire detection model;

[0036] Figure 10 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0038] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0039] In an exemplary embodiment, Figure 1 and Figure 3 As shown, a method for detecting electric wires in low-altitude airspace is provided, comprising the following steps, wherein:

[0040] Step 201: Obtain a low-altitude airspace wire dataset; the low-altitude airspace wire dataset includes multiple samples, each sample is an external view image of a sample low-altitude aircraft and a corresponding mask label (real mask); the external view image includes wires; the mask label is a mask image with wire boundaries. Specifically, a drone and a general aviation helicopter are used to obtain a low-altitude airspace wire dataset. Part of the data of the low-altitude airspace wire dataset is as follows Figures 4 to 6 As shown in Figure 2, datasets play a crucial role when it comes to deep learning models. High-quality datasets that reflect real-world scenarios will significantly improve the accuracy of deep learning models. Therefore, the low-altitude airspace powerline dataset to be used should accurately reflect the real-world scenarios of low-altitude aircraft operations.

[0041] Step 202: A wire detection model is obtained by training a slender target adaptive regional rebalancing convolutional neural network based on the low-altitude airspace wire dataset. The wire detection model is used for low-altitude airspace wire detection. The slender target adaptive regional rebalancing convolutional neural network includes a backbone network and a segmentation head connected in sequence. The backbone network is an HRNet network in which the 3x3 standard convolution in the Bottleneck module is a dynamic snake convolution (DSConv). The segmentation head includes a segmentation head network and a regional rebalancing branch module. The first input of the regional rebalancing branch module and the input of the segmentation head network are both connected to the output of the backbone network, and the second input of the regional rebalancing branch module is used to input a mask label. Specifically, the regional rebalancing branch module includes a cross product operation and a classifier. The cross product operation is used to cross-product the output of the backbone network with the mask label, and then the result of the cross product operation is input to the classifier.

[0042] In the above steps 201 to 202, HRNet is used to extract hierarchical features while ensuring high resolution of the image to obtain a feature map. DSConv is used to adapt to the slender features of the wires. The class imbalance problem is alleviated by the regional rebalancing branch module. The regional rebalancing branch module encourages the feature map to be located in a more balanced regional classification space and rebalances at the regional level rather than the pixel level. In the regional rebalancing branch module, the pixel features are cross-multiplied ( Figure 7 and Figure 9 Indicated in ) are averaged, grouping pixels belonging to the same category into regional features. The extracted regional features are then input into the region classifier (the classifier mentioned above). This approach uses techniques from long-tail image classification to enhance rebalancing to address regional imbalance. This approach does not incur additional testing costs, as the region rebalancing branch module is deleted after training is complete.

[0043] The feature encoding process uses HRNet integrated with DSConv to ensure high image resolution while paying more attention to the features of slender targets, thereby improving the accuracy of wire detection. The feature decoding process uses the regional rebalancing branch module to solve the imbalance problem of positive and negative samples, improve accuracy, and reduce the missed detection rate.

[0044] In an exemplary embodiment, Figure 2 As shown, the wire detection model is used to detect wires in low-altitude airspace, specifically including:

[0045] Acquire external visual images of the low-altitude aircraft to be detected.

[0046] The external visual image of the low-altitude aircraft to be detected is input into the wire detection model without the regional rebalancing branch module to obtain a mask image with wire boundaries corresponding to the external visual image of the low-altitude aircraft to be detected.

[0047] In an exemplary embodiment, the segmentation head network is a lightweight network segmentation head module, specifically a lightweight FCN, or depthwise separable convolution. Using a lightweight network segmentation head module ensures faster detection speeds. The FCN can accept input images of any size, uses bilinear interpolation for parameter initialization of the deconvolution layer, and upsamples the feature map output by the backbone network to the same size as the input image, thereby generating a prediction for each pixel while preserving the spatial information in the original input image. Finally, pixel-by-pixel classification is performed on the upsampled feature map.

[0048] In an exemplary embodiment, a wire detection model is obtained by training a regional rebalancing convolutional neural network based on slender target adaptation according to the low-altitude airspace wire dataset, specifically including:

[0049] The low-altitude airspace wire dataset is processed by using a data enhancement algorithm to obtain an enhanced dataset. Specifically, the low-altitude airspace wire dataset is processed by using a data enhancement library Albumentations to obtain the enhanced dataset. The data enhancement method includes pixel-level transformation and spatial-level transformation. The pixel-level transformation only changes the input image, such as blurring, color jitter, image compression, Gaussian noise, inversion, normalization, random rain, random brightness contrast, sharpening, hue saturation value, etc. The spatial-level transformation simultaneously acts on the image and the mask, such as horizontal rotation, vertical rotation, 90-degree rotation, center flipping, row-column transposition, scale zooming, etc.

[0050] The power line detection model is obtained by training the region rebalancing convolutional neural network based on the elongated target self-adaption using the enhanced dataset with the minimum loss function value as the target.

[0051] In an exemplary embodiment, the loss function is:

[0052] wherein represents the loss function, represents the region loss, represents the pixel cross-entropy loss, λ represents the coefficient of the pixel cross-entropy loss, which can be 0.5, is calculated according to the output result of the region rebalancing branch module (obtained by inputting the external view in the enhanced dataset into the region rebalancing convolutional neural network based on the elongated target self-adaption) and the mask label, is calculated according to the output result of the segmentation head network (obtained by inputting the external view in the enhanced dataset into the region rebalancing convolutional neural network based on the elongated target self-adaption) and the mask label.

[0053] Specifically, This formula aims to average the pixels belonging to the same class according to the ground-truth (X gt ) region map, group the pixels belonging to the same class as region features, and then input the extracted region features into the region classifier, wherein D is the enhanced dataset, represents the region frequency of the i-th class, R is all the extracted regions obtained in the cross multiplication process of the region rebalancing branch module, r y is the label mask of the region r in the enhanced dataset, r y represents the region value corresponding to the y-th class (the output result of the region rebalancing branch module), and Y is a label set including background, power lines and power towers.

[0054] Among them, P(i) is the true probability of the i-th class, which is obtained according to the label mask; Q(i) is the probability of the i-th class predicted by the model, which is the result output by the segmentation head network, and n represents the total number of classes.

[0055] In an exemplary embodiment, the HRNet network is HRNet-w48. Specifically, the present application uses the improved HRNet-w48 as the backbone network to maintain the high resolution of the image while downsampling and minimize the loss of wire features. Figure 7 and Figure 8 As shown in the figure, HRNet has four basic layers, Layer 1 to Layer 4, with the number of branches increasing in each layer. Each branch first passes through four repeatedly stacked convolutional blocks, Bottleneck modules or BasicBlock modules, each of which uses 48 channels, and then fuses information at different scales. The output of each branch in the next layer is obtained by fusing the output of the branches in the previous layer. For example, the output of the 4-fold downsampling branch in Layer 3 is obtained by adding the output of the 4-fold downsampling branch, the output of the 8-fold downsampling branch upsampled by 2, and the output of the 16-fold downsampling branch upsampled by 4, and then passing it through ReLU. The other branches are similar.

[0056] To adapt the model to slender targets such as wires, this application replaces the Bottleneck module in HRNet Layer 1 and uses DSConv to replace the original 3x3 standard convolution. Considering that tubular target segmentation is characterized by slenderness and complexity, standard convolution and dilated convolution cannot adjust the focus region according to the target characteristics. Deformable convolution can learn the region of interest more adaptively. However, for tubular targets, deformable convolution cannot limit the connectivity of the focus region, while DSConv limits the connectivity of the focus region, making it more suitable for wire segmentation scenarios.

[0057] To give the convolution kernel more flexibility and enable it to focus on the slender geometric features of the target, a deformation offset Δ is introduced, inspired by deformable convolution. However, if the model is completely free to learn the deformation offset, the receptive field will often deviate from the target, especially when dealing with slender tubular structures. Therefore, DSConv adopts an iterative strategy, sequentially selecting the next position of each target to be processed for observation, thereby ensuring the continuity of attention and preventing the receptive range from being spread too far due to large deformation offsets.

[0058] In dynamic snake convolution, the standard convolution kernel is linearized in both the x-axis and y-axis directions. Consider a convolution kernel of size 9:

[0059] K={(x-1,y-1), (x-1,y),…, (x+1,y+1)}

[0060] Taking the x-axis direction as an example, the specific position of each grid in K is expressed as:

[0061]

[0062] Where c represents the horizontal distance from the center grid. Each grid position K in the convolution kernel K i±c The selection of is a cumulative process. i Initially, the position of the grid away from the center depends on the position of the previous grid: K i+1 Relative to K i An offset Δ={δ|δ∈[-1, 1]} is added. Therefore, the offset needs to be accumulated Σ to ensure that the convolution kernel conforms to the linear morphology structure.

[0063] This application uses a regional rebalancing network that is adaptive to slender targets to improve the accuracy of mask convolutional neural networks in detecting wires in low-altitude airspace, reducing the missed detection rate of wires while achieving the optimal balance between detection speed and detection accuracy, significantly improving model performance and reducing the risk of aircraft hitting wires during low-altitude operations.

[0064] The HRNet network architecture in this application continuously performs multi-scale fusion during the downsampling process, ensuring high resolution of the image when extracting features and avoiding the loss of slender features such as wires, thereby providing a solid foundation for the precise semantic segmentation task in the regional rebalancing convolutional neural network structure based on slender object adaptation.

[0065] This application uses a Bottleneck module of dynamic snake convolution so that during the downsampling process, the convolution kernel in the residual structure can adaptively deform to fit the slender topological structure, making it easier to capture the slender features of the wires for further classification and positioning.

[0066] After the backbone network extracts the feature map of the image, this application uses the fully connected layer of the lightweight network segmentation head module, uses depthwise separable convolution to reduce the computational complexity, and then performs pixel-by-pixel classification after upsampling the feature map to the original image size.

[0067] The regional rebalancing branch module of this application belongs to the auxiliary segmentation head, which aggregates pixels within the same mask label into one region, solves the imbalance problem of positive and negative samples from a regional perspective, and improves the accuracy of wire detection. This module only participates in supervision during the model training process and does not introduce additional inference costs.

[0068] Based on the same inventive concept, the present application also provides a low-altitude airspace wire detection device for implementing the aforementioned low-altitude airspace wire detection method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more embodiments of the low-altitude airspace wire detection device provided below can be found in the above-mentioned limitations of the low-altitude airspace wire detection method and will not be further elaborated here.

[0069] In an exemplary embodiment, a low-altitude airspace wire detection device is provided, comprising:

[0070] An acquisition module is used to acquire a low-altitude airspace wire dataset; the low-altitude airspace wire dataset includes multiple samples, each sample is an external view image of a sample low-altitude aircraft and a corresponding mask label; the external view image includes wires; the mask label is a mask image with wire boundaries.

[0071] A training module is used to train a regional rebalancing convolutional neural network based on slender target adaptation according to the low-altitude airspace wire dataset to obtain a wire detection model, and the wire detection model is used to perform wire detection in the low-altitude airspace; the regional rebalancing convolutional neural network based on slender target adaptation includes a backbone network and a segmentation head connected in sequence; the backbone network is an HRNet network in which the 3x3 standard convolution in the Bottleneck module is a dynamic snake convolution; the segmentation head includes a segmentation head network and a regional rebalancing branch module; the first input end of the regional rebalancing branch module and the input end of the segmentation head network are both connected to the output end of the backbone network, and the second input end of the regional rebalancing branch module is used to input a mask label.

[0072] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 10As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store low-altitude airspace electric line detection data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with the terminal outside through network connection. The computer program is executed by the processor to realize a low-altitude airspace electric line detection method.

[0073] Those skilled in the art can understand that, Figure 10 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement. In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to realize the above-mentioned method embodiments.

[0074] In one exemplary embodiment, a computer readable storage medium is provided, storing a computer program, which is executed by a processor to realize the above-mentioned method embodiments.

[0075] In one exemplary embodiment, a computer program product is provided, including a computer program, which is executed by a processor to realize the above-mentioned method embodiments.

[0076] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0077] In the present application, all actions of obtaining signals, information or data are carried out in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the corresponding device owner.

[0078] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0079] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0080] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0081] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for detecting electric wires in low-altitude airspace, characterized in that: The low-altitude airspace wire detection method comprises: Acquire a low-altitude airspace wire dataset; the low-altitude airspace wire dataset includes multiple samples, each sample is an external view image of a sample low-altitude aircraft and a corresponding mask label; the external view image includes wires; the mask label is a mask image with wire boundaries; According to the low-altitude airspace wire dataset, a regional rebalancing convolutional neural network based on slender target adaptation is trained to obtain a wire detection model, which is used to perform wire detection in the low-altitude airspace; the regional rebalancing convolutional neural network based on slender target adaptation includes a backbone network and a segmentation head connected in sequence; the backbone network is an HRNet network in which the 3x3 standard convolution in the Bottleneck module is replaced by a dynamic snake convolution; the segmentation head includes a segmentation head network and a regional rebalancing branch module; the first input end of the regional rebalancing branch module and the input end of the segmentation head network are both connected to the output end of the backbone network, and the second input end of the regional rebalancing branch module is used to input a mask label; the regional rebalancing branch module includes: a cross product operation and a classifier; the cross product operation is used to perform a cross product operation on the output of the backbone network and the mask label, and then the result of the cross product operation is input into the classifier.

2. The low-altitude airspace wire detection method according to claim 1, characterized in that: The wire detection model is used to detect wires in low-altitude airspace, specifically including: Acquire external visual images of the low-altitude aircraft to be detected; The external visual image of the low-altitude aircraft to be detected is input into the wire detection model without the regional rebalancing branch module to obtain a mask image with wire boundaries corresponding to the external visual image of the low-altitude aircraft to be detected.

3. The low-altitude airspace wire detection method according to claim 1, characterized in that: The segmentation head network is a depth-wise separable convolutional network.

4. The low-altitude airspace wire detection method according to claim 1, characterized in that: The wire detection model is obtained by training a regional rebalancing convolutional neural network based on slender target adaptation according to the low-altitude airspace wire dataset, specifically including: Using a data enhancement algorithm to perform data enhancement processing on the low-altitude airspace power line dataset to obtain an enhanced dataset; With the goal of minimizing the loss function value, an enhanced dataset is used to train a regional rebalancing convolutional neural network based on slender target adaptation to obtain a wire detection model.

5. The low-altitude airspace wire detection method according to claim 1, characterized in that: The HRNet network is HRNet-w48.

6. The low-altitude airspace wire detection method according to claim 4, characterized in that: The loss function is: in represents the loss function, represents the regional loss, represents pixel cross entropy loss, λ represents The coefficient of Calculated based on the output of the region rebalancing branch module and the mask label, Calculated based on the output of the segmentation head network and the mask labels.

7. A low-altitude airspace wire detection device, characterized in that: The low-altitude airspace wire detection device comprises: An acquisition module is configured to acquire a low-altitude airspace wire dataset; the low-altitude airspace wire dataset includes a plurality of samples, each sample being an exterior view image of a sample low-altitude aircraft and a corresponding mask label; the exterior view image includes wires; and the mask label is a mask image with wire boundaries. A training module is used to train a regional rebalancing convolutional neural network based on slender target adaptation according to the low-altitude airspace wire dataset to obtain a wire detection model, and the wire detection model is used to perform wire detection in the low-altitude airspace; the regional rebalancing convolutional neural network based on slender target adaptation includes a backbone network and a segmentation head connected in sequence; the backbone network is an HRNet network in which the 3x3 standard convolution in the Bottleneck module is replaced with a dynamic snake convolution; the segmentation head includes a segmentation head network and a regional rebalancing branch module; the first input end of the regional rebalancing branch module and the input end of the segmentation head network are both connected to the output end of the backbone network, and the second input end of the regional rebalancing branch module is used to input a mask label; the regional rebalancing branch module includes: a cross product operation and a classifier; the cross product operation is used to perform a cross product operation on the output of the backbone network and the mask label, and then input the result of the cross product operation into the classifier.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the low-altitude airspace wire detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the low-altitude airspace wire detection method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the low-altitude airspace wire detection method according to any one of claims 1 to 6 is implemented.

Images