High-altitude insulator detection method based on laser radar point cloud data decoding

Abstract

The invention relates to a high-altitude insulator detection method based on laser radar point cloud data decoding. The high-altitude insulator detection method comprises the following steps: converting laser radar point cloud data into a pseudo image; constructing an insulator detector based on a convolutional backbone network fused with a dynamic deformation convolution attention mechanism, and inputting the pseudo image into the insulator detector for insulator detection; distance detection is carried out on each insulator by calculating depth information of the point cloud, and when the detection distance of the insulator is smaller than a preset threshold value, an alarm is given; compared with an existing insulator detection method, the method has the advantages that the detection accuracy is improved, the insulator space position sensing capability is achieved, and safety guarantee is provided for insulator guiding unmanned aerial vehicle autonomous inspection flight; the high-altitude insulator detection method based on laser radar point cloud data decoding has a wide application prospect, and is particularly suitable for safety monitoring and maintenance of a high-voltage transmission line.

Description

Technical Field

[0001] The invention relates to a high-altitude insulator detection method based on laser radar point cloud data decoding, and belongs to the technical field of computer vision. Background Art

[0002] Insulators are an important part of power lines and one of the key inspection targets for drone inspections of power towers. Due to the importance of insulators in inspection tasks and the particularity of line installation locations, many research works use insulators as an important clue to assist drones in autonomous inspections of power lines and guide fully autonomous drones to understand the structure of mission scenarios. One of the important foundations is to accurately identify and locate insulators.

[0003] The invention patent with publication number "CN116612107A" proposes a training method for an insulator image detection model and an insulator image detection method. The training method for the insulator image detection model includes: constructing an original insulator image data set; the original insulator image data set includes: normal insulator images and faulty insulator images; preprocessing the original insulator image data set to obtain training samples of the original insulator images; dividing the training samples into a training set, a validation set, and a test set; using the training set and the validation set to train the insulator image detection model; using the test set to adjust the parameters of the trained insulator image detection model; the model solves the current problems of low efficiency and low inspection accuracy of insulator inspections.

[0004] However, due to the limitations of image data, the detection accuracy of this type of method is easily affected by factors such as lighting conditions and imaging background, and it does not contain visual depth information and does not have the ability to perceive the spatial information of insulators. It is difficult to support the needs of scene structure perception and fully autonomous inspection flight safety control of power autonomous inspection drone missions. Summary of the invention

[0005] In order to solve the above problems existing in the prior art, the present invention proposes a high-altitude insulator detection method based on laser radar point cloud data decoding.

[0006] The technical solution of the present invention is as follows:

[0007] On the one hand, the present invention provides a method for detecting high-altitude insulators based on laser radar point cloud data decoding, comprising the following steps:

[0008] Convert LiDAR point cloud data into pseudo images;

[0009] An insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional attention mechanism is constructed, and the pseudo image is input into the insulator detector for insulator detection.

[0010] The distance of each insulator is detected by calculating the depth information of the point cloud. When the detection distance of the insulator is less than the preset threshold, a warning is issued.

[0011] As a preferred embodiment of the present invention, the specific steps of converting the laser radar point cloud data into a pseudo image are:

[0012] The laser radar point cloud data is expanded by cylindrical projection, and the cylindrical projection expansion image is the pseudo image, which is shown in the following formula:

[0013] p=p(x,y,z)

[0014] σ=a tan 2(y,x)

[0015]

[0016]

[0017]

[0018] Where: p represents the coordinates of a point in space; σ represents the horizontal angular resolution of a single continuous laser beam emitter; represents the vertical angular resolution of a single continuous laser beam emitter; Δσ represents the horizontal angular resolution of all continuous laser beam emitters; represents the average vertical angular resolution of all continuous laser beam emitters; q represents the azimuth when observing the point; e represents the elevation when observing the point; (q, e) represents the mapping position of the point projected onto a two-dimensional plane, and the two-dimensional plane is a cylindrical expansion;

[0019] The two-dimensional projection point image of the point cloud data is divided into multiple two-dimensional grid units. For each two-dimensional grid unit, the relative coordinates of each point in the two-dimensional projection point image and its original three-dimensional coordinates are extracted and stored to generate the dimensional features of each point.

[0020] As a preferred embodiment of the present invention, the convolutional backbone network includes a feature enhancement network and a feature splicing network;

[0021] The input data passes through an initial convolutional layer to extract features of initial size;

[0022] Input the initial size features into the feature enhancement network, which is composed of multiple layers of convolutional layers with gradually increasing output channels and gradually decreasing length and width. Each convolutional layer outputs feature maps of different scales, and residual connections are used between each convolution layer. At the same time, ReLU activation function is used for output after the convolution operation of each convolutional layer;

[0023] Feature maps of different scales are fused through the feature concatenation network to output a three-dimensional tensor of integrated features.

[0024] As a preferred embodiment of the present invention, the dynamic deformation convolutional attention mechanism adds a set of offset sampling convolutional layers corresponding to pedestrian features according to the output features of the convolutional backbone network. The offset sampling convolutional layer adds a coordinate offset during the convolution parsing process to analyze the offset of the convolution relative sampling position and the grid position.

[0025] As a preferred embodiment of the present invention, the initial loss function Loss of the insulator detector is constructed based on the joint complete cross loss function. CIoU , as shown in the following formula:

[0026]

[0027]

[0028]

[0029] Where: 2 (b,b gt ) represents the Euclidean distance between the center point of the prediction box and the center point of the true value box; IoU represents the intersection over union ratio of the prediction box and the true value box; b represents the prediction box; b gt represents the true value box; c represents the diagonal distance of the minimum closed interval that contains both the prediction box and the true value box; w represents the width of the prediction box; w gt represents the width of the true value box; h represents the height of the prediction box; h gt Indicates the height of the truth box;

[0030] Then use the prior frame to enhance the detection, and design the amplification factor to amplify the regression loss of the prior frame, as shown in the following formula:

[0031] Loss Enl arg eIoU =μLoss CIoU

[0032]

[0033] Where: w g Indicates the width of the minimum prior box; H g represents the height of the minimum prior box; μ represents the magnification factor;

[0034] Introduce the dynamic proportional factor and combine it with the above formula to form the final loss function Loss of the insulator detector. DynamicIoU , as shown in the following formula:

[0035]

[0036] Where: D represents the gradient gain; θ represents the dynamic scaling factor; δ and γ represent hyperparameters.

[0037] As a preferred embodiment of the present invention, the distance detection step of the insulator is:

[0038] Each point in the point cloud data represents a reflection point on the surface of the insulator, and the reflection point contains three-dimensional coordinates;

[0039] The Euclidean distance from the sensor position to the insulator point cloud is calculated to obtain the depth information of the insulator. For an insulator, there are multiple reflection points. The depth information of all points is calculated, and the shortest distance in the depth information is selected as the final detection result.

[0040] On the other hand, the present invention also provides a high altitude insulator detection system based on laser radar point cloud data decoding, including a pseudo image conversion module, an insulator detection module, a distance detection module and a warning module;

[0041] The pseudo image conversion module is used to convert the laser radar point cloud data into a pseudo image;

[0042] The insulator detection module is used to construct an insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional layer, and input the pseudo image into the insulator detector for insulator detection;

[0043] The distance detection module is used to perform distance detection on each insulator by calculating the depth information of the point cloud;

[0044] The warning module is used to issue a warning when the detection distance of the insulator is less than a preset threshold.

[0045] On the other hand, the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present invention when executing the program.

[0046] In yet another aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0047] The present invention has the following beneficial effects:

[0048] 1. The present invention significantly improves the accuracy and robustness of insulator detection by converting point cloud data into pseudo images through cylindrical expansion, and then combining the convolution backbone network with the dynamic deformation convolution attention mechanism. This multi-stage data processing and feature extraction mechanism can better cope with complex background environments and reduce false detections and missed detections.

[0049] 2. The laser radar used in the present invention is not affected by lighting and weather conditions. Therefore, it is more stable than traditional image-based methods and is particularly suitable for detection in adverse conditions such as low-light environments, rainy and snowy weather. The cylindrical expansion technology makes full use of the 360-degree all-round viewing angle of the laser radar, further enhancing environmental adaptability.

[0050] 3. The present invention calculates the depth information of the laser radar point cloud data, and the method can measure the distance between the insulator and the laser radar in real time and accurately. By setting the distance threshold, when the distance of the insulator is less than the set threshold, the system can automatically issue a warning to avoid potential risks. This automated distance measurement function reduces the reliance on manual intervention and greatly improves efficiency.

[0051] 4. The present invention introduces a dynamic non-monotonic bounding box regression loss function, which effectively solves the confidence imbalance problem, stabilizes the gradient update during training, and significantly improves the efficiency and stability of pseudo image training. This method ensures that the system can maintain good detection performance when facing a variety of complex situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a flow chart of the method of the present invention;

[0053] Figure 2 A pseudo image generation flow chart of the present invention;

[0054] Figure 3 This is a structural diagram of the convolutional backbone network described in the present invention;

[0055] Figure 4 This is a structural diagram of the dynamic deformation convolutional attention mechanism described in the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.

[0058] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0059] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0060] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.

[0061] Embodiment 1:

[0062] See also Figure 1 , a high altitude insulator detection method based on laser radar point cloud data decoding, comprising the following steps:

[0063] Convert LiDAR point cloud data into pseudo images;

[0064] An insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional layer is constructed, and the pseudo image is input into the insulator detector for insulator detection.

[0065] The distance of each insulator is detected by calculating the depth information of the point cloud. When the detection distance of the insulator is less than the preset threshold, a warning is issued.

[0066] As a preferred implementation of this embodiment, see Figure 2 The specific steps of converting the laser radar point cloud data into a pseudo image are:

[0067] The laser radar point cloud data is expanded by cylindrical projection, and the point cloud is evenly discretized in the xy plane. The cylindrical projection expansion image is the pseudo image, as shown in the following formula:

[0068] p=p(x,y,z)

[0069] σ=a tan 2(y,x)

[0070]

[0071]

[0072]

[0073] Where: p represents the coordinates of a point in space; σ represents the horizontal angular resolution of a single continuous laser beam emitter; represents the vertical angular resolution of a single continuous laser beam emitter; Δσ represents the horizontal angular resolution of all continuous laser beam emitters; represents the average vertical angular resolution of all continuous laser beam emitters; q represents the azimuth when observing the point; e represents the elevation when observing the point; (q, e) represents the mapping position of the point projected onto a two-dimensional plane, and the two-dimensional plane is a cylindrical expansion; according to the reasoning of the above formula, the point cloud data of the laser radar can be projected into a pseudo-image projection point diagram of digital signal integration. The resulting projection point diagram can be compared to a cylindrical expansion. This data conversion application fully expands the inherent advantages of the laser radar hardware and brings image output of a full range of viewing angles that is not affected by light or rain.

[0074] The two-dimensional projection point image of the point cloud data is divided into multiple two-dimensional grid units. For each two-dimensional grid unit, the relative coordinates of each point in the two-dimensional projection point image and its original three-dimensional coordinates are extracted and stored to generate the dimensional features of each point. After expanding the feature dimensions, each point has 6 dimensions (x, y, z, xd, yd, zd). Among them, xd, yd, zd represent the relative positions of the points on the cylinder.

[0075] As a preferred implementation of this embodiment, see Figure 3 , the convolutional backbone network includes a feature enhancement network and a feature splicing network;

[0076] The input data passes through an initial convolutional layer to extract features of initial size;

[0077] The initial size feature is input into the feature enhancement network. The feature enhancement network is composed of multiple layers of convolutional layers with gradually increasing output channels and gradually decreasing length and width, forming a structure similar to a spatial pyramid. The shape of the output feature map is assumed to be (C, W, H), where C is the number of channels, W is the image width, and H is the image height. Each layer of convolution operation can capture information of different scales to ensure the comprehensive extraction of details and context information. Each convolution layer outputs feature maps of different scales. Residual connections are used between each convolution layer. At the same time, ReLU activation function is used for output after the convolution operation of each convolution layer to help introduce nonlinearity and reduce the possibility of gradient disappearance while obtaining features;

[0078] Feature maps of different scales are fused through the feature splicing network, and a three-dimensional tensor of integrated features is output. The output result retains the rich features extracted from different spatial scales.

[0079] As a preferred implementation of this embodiment, see Figure 4In order to adapt to the high aspect ratio characteristics of pseudo images, the dynamic deformation convolution layer adds a set of offset sampling convolution layers corresponding to pedestrian features according to the output characteristics of the convolution backbone network. The offset sampling convolution layer adds a coordinate offset in the convolution parsing process, and flexibly offsets the convolution relative sampling position and the grid position. Taking two-dimensional convolution as an example, offset access is performed in the x and y directions, and the relative coordinates of the current offset convolution kernel are calculated. Subsequently, the boundary of the current convolution kernel is calculated using the bilinear interpolation method to obtain a convolution operation with a stride of 3. The expansion of the original convolution from a fixed value to a dynamic value increases the receptive field of the feature, realizes the improvement from a fixed convolution receptive field to a dynamic receptive field, and significantly improves the range and flexibility of convolution to obtain information.

[0080] As a preferred implementation of this embodiment, in the training loss part, an integrated module of a dynamic non-monotonic bounding box regression loss function for pseudo image training is proposed to solve the training gradient explosion problem caused by the imbalance of confidence discrimination during the training process, thereby improving the stability of pseudo image training and prediction to a certain extent, as follows:

[0081] Based on the joint complete intersection-over-Union (CIoU) loss function, the initial loss function Loss of the insulator detector is constructed. CIoU , as shown in the following formula:

[0082]

[0083]

[0084]

[0085] Where: 2 (b,b gt ) represents the Euclidean distance between the center point of the prediction box and the center point of the true value box; IoU represents the intersection over union ratio of the prediction box and the true value box; b represents the prediction box; b gt represents the true value box; c represents the diagonal distance of the minimum closed interval that contains both the prediction box and the true value box; w represents the width of the prediction box; w gt represents the width of the true value box; h represents the height of the prediction box; h gt Indicates the height of the truth box;

[0086] Then the prior frame is used to enhance the detection, and the amplification factor is designed to amplify the regression loss of the prior frame, as shown in the following formula:

[0087] Loss Enl arg eIoU =μLoss CIoU

[0088]

[0089] Where: w g Indicates the width of the minimum prior box; H g represents the height of the minimum prior box; μ represents the magnification factor;

[0090] Introduce the dynamic proportional factor and combine it with the above formula to form the final loss function Loss of the insulator detector. DynamicIoU , as shown in the following formula:

[0091]

[0092] Where: D represents the gradient gain; θ represents the dynamic scaling factor; δ and γ represent hyperparameters, and the mapping of the gradient gain D is controlled by δ and γ;

[0093] The dynamic scaling factor is defined as the outlier feature of the prior box, that is, the ratio of the current loss to the current average loss. A smaller outlier feature means a higher quality prior box, giving it a smaller gradient gain so that the bounding box loss is concentrated on the preset prior box.

[0094] In addition, assigning smaller gradient gains to prior boxes with larger outlier features can effectively prevent gradient explosion caused by low-quality images;

[0095] By adjusting the dynamic scaling factor, the weights of easily distinguishable samples can be dynamically reduced during training, so that more difficult to distinguish samples can be focused on more quickly.

[0096] As a preferred implementation of this embodiment, the distance detection step of the insulator is:

[0097] Each point in the point cloud data represents a reflection point on the surface of the insulator, and the reflection point contains three-dimensional coordinates (X, Y, Z);

[0098] The Euclidean distance from the sensor position to the insulator point cloud is calculated to obtain the depth information of the insulator. For an insulator, there may be multiple reflection points. In this case, the depth information of all points is calculated and the shortest distance in the depth information is selected as the final detection result.

[0099] Embodiment 2:

[0100] A high altitude insulator detection system based on laser radar point cloud data decoding, comprising a pseudo image conversion module, an insulator detection module, a distance detection module and a warning module;

[0101] The pseudo image conversion module is used to convert the laser radar point cloud data into a pseudo image;

[0102] The insulator detection module is used to construct an insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional layer, and input the pseudo image into the insulator detector for insulator detection;

[0103] The distance detection module is used to perform distance detection on each insulator by calculating the depth information of the point cloud;

[0104] The warning module is used to issue a warning when the detection distance of the insulator is less than a preset threshold.

[0105] The system is used to implement the method in Example 1, which will not be described in detail here.

[0106] Embodiment three:

[0107] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present invention when executing the program.

[0108] Embodiment 4:

[0109] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in any embodiment of the present invention is implemented.

[0110] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0111] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0112] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0113] In several embodiments provided in the present application, if any function is implemented in the form of 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 the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0114] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for detecting high-altitude insulators based on laser radar point cloud data decoding, characterized in that: The following steps are involved: Convert LiDAR point cloud data into pseudo images; An insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional attention mechanism is constructed, and the pseudo image is input into the insulator detector for insulator detection. The distance of each insulator is detected by calculating the depth information of the point cloud. When the detection distance of the insulator is less than the preset threshold, a warning is issued.

2. The high-altitude insulator detection method based on laser radar point cloud data decoding according to claim 1 is characterized in that: The specific steps of converting the laser radar point cloud data into a pseudo image are: The laser radar point cloud data is expanded by cylindrical projection, and the cylindrical projection expansion image is the pseudo image, which is shown in the following formula: p=p(x,y,z) σ=a tan2(y,x) Where: p represents the coordinates of a point in space; σ represents the horizontal angular resolution of a single continuous laser beam emitter; represents the vertical angular resolution of a single continuous laser beam emitter; Δσ represents the horizontal angular resolution of all continuous laser beam emitters; represents the average vertical angular resolution of all continuous laser beam emitters; q represents the azimuth when observing the point; e represents the elevation when observing the point; (q, e) represents the mapping position of the point projected onto a two-dimensional plane, and the two-dimensional plane is a cylindrical expansion; The two-dimensional projection point image of the point cloud data is divided into multiple two-dimensional grid units. For each two-dimensional grid unit, the relative coordinates of each point in the two-dimensional projection point image and its original three-dimensional coordinates are extracted and stored to generate the dimensional features of each point.

3. The high-altitude insulator detection method based on laser radar point cloud data decoding according to claim 1 is characterized in that: The convolutional backbone network includes a feature enhancement network and a feature splicing network; The input data passes through an initial convolutional layer to extract features of initial size; Input the initial size features into the feature enhancement network, which is composed of multiple layers of convolutional layers with gradually increasing output channels and gradually decreasing length and width. Each convolutional layer outputs feature maps of different scales, and residual connections are used between each convolution layer. At the same time, ReLU activation function is used for output after the convolution operation of each convolutional layer; Feature maps of different scales are fused through the feature concatenation network to output a three-dimensional tensor of integrated features.

4. The high-altitude insulator detection method based on laser radar point cloud data decoding according to claim 3 is characterized in that: The dynamic deformation convolution attention mechanism adds a set of offset sampling convolution layers corresponding to pedestrian features according to the output features of the convolution backbone network. The offset sampling convolution layer adds a coordinate offset during the convolution parsing process to parse the offset of the convolution relative sampling position and the grid position.

5. The high-altitude insulator detection method based on laser radar point cloud data decoding according to claim 3 is characterized in that: The initial loss function Loss of the insulator detector is constructed based on the joint complete cross loss function CIoU , as shown in the following formula: Where: 2 (b,b gt ) represents the Euclidean distance between the center point of the prediction box and the center point of the true value box; IoU represents the intersection over union ratio of the prediction box and the true value box; b represents the prediction box; b gt represents the true value box; c represents the diagonal distance of the minimum closed interval that contains both the prediction box and the true value box; w represents the width of the prediction box; w gt represents the width of the true value box; h represents the height of the prediction box; h gt Indicates the height of the truth box; Then use the prior frame to enhance the detection, and design the amplification factor to amplify the regression loss of the prior frame, as shown in the following formula: Loss Enl arg eIoU =μLoss CIoU Where: w g Indicates the width of the minimum prior box; H g represents the height of the minimum prior box; μ represents the magnification factor; Introduce the dynamic proportional factor and combine it with the above formula to form the final loss function Loss of the insulator detector. DynamicIoU , as shown in the following formula: Loss DynamicIoU =DLoss Enl arg eIoU , Where: D represents the gradient gain; θ represents the dynamic scaling factor; δ and γ represent hyperparameters.

6. The high-altitude insulator detection method based on laser radar point cloud data decoding according to claim 1 is characterized in that: The distance detection steps of the insulator are as follows: Each point in the point cloud data represents a reflection point on the surface of the insulator, and the reflection point contains three-dimensional coordinates; The Euclidean distance from the sensor position to the insulator point cloud is calculated to obtain the depth information of the insulator. For an insulator, there are multiple reflection points. The depth information of all points is calculated, and the shortest distance in the depth information is selected as the final detection result.

7. A high altitude insulator detection system based on laser radar point cloud data decoding, characterized in that: It includes a pseudo-image conversion module, an insulator detection module, a distance detection module and a warning module; The pseudo image conversion module is used to convert the laser radar point cloud data into a pseudo image; The insulator detection module is used to construct an insulator detector based on a convolutional backbone network fused with a dynamic deformation convolutional layer, and input the pseudo image into the insulator detector for insulator detection; The distance detection module is used to perform distance detection on each insulator by calculating the depth information of the point cloud; The warning module is used to issue a warning when the detection distance of the insulator is less than a preset threshold.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

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

Images