Power transmission line intelligent ice melting device and ice melting method
By combining the IPC de-icing computing chip with thermal imaging and optical image sensors, high-precision identification and safety control of de-icing areas on transmission lines have been achieved, solving the problems of low efficiency and low accuracy in traditional de-icing methods and improving the safety and reliability of de-icing on transmission lines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG UNIV OF TECH
- Filing Date
- 2025-09-23
- Publication Date
- 2026-05-15
AI Technical Summary
Traditional methods for de-icing power transmission lines are inefficient and unsafe, especially in complex environments where the accuracy of detection is low, and the deployment and implementation of existing equipment are difficult.
The system employs an IPC ice-melting computing chip, combined with thermal imaging and optical image sensors. It identifies the ice-melting area through a positioning and scanning unit, controls the mobile device with an intelligent movement and operation unit, performs ice-melting with a heating device, and ensures device safety through a safety protection and monitoring unit, thus constructing a high-precision ice-melting area identification model (IMDM).
It improves the accuracy and response capability of de-icing detection, ensures the safe operation of the device, solves the problems of low efficiency and low accuracy in traditional methods, and adapts to complex and ever-changing de-icing environments.
Smart Images
Figure CN121216336B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power transmission line technology, specifically to an intelligent de-icing device and method for power transmission lines. Background Technology
[0002] Icing on transmission lines is a major hidden danger threatening the safe and stable operation of the power grid. When winter temperatures drop sharply and humidity is high, ice easily accumulates on the surface of transmission lines. As the ice thickness increases, the load on the lines gradually rises, leading to increased mechanical loads on the towers. In mild cases, this can cause slight deformation of the towers, affecting the normal sag and electrical performance of the line; in severe cases, it can lead to serious accidents such as tower collapse and line breakage, causing large-scale power outages.
[0003] Traditional methods for de-icing power transmission lines include manual knocking, short-circuit de-icing, and DC de-icing. Manual knocking is extremely inefficient and poses significant personal safety risks when working at heights, making it unsuitable for large-scale transmission line de-icing needs. Short-circuit de-icing requires power outages, impacting power supply and dispatch, especially during peak hours, severely limiting its application. While DC de-icing improves efficiency and reduces outage time to some extent, it involves large equipment investments, complex operation, and requires specialized DC de-icing equipment and skilled technicians. Furthermore, for transmission lines in remote areas, the deployment and implementation of DC de-icing equipment is challenging due to inconvenient transportation and limited power supply.
[0004] Based on the above, traditional methods for de-icing power transmission lines have the following problems:
[0005] 1. Traditional manual de-icing detection methods often rely on manual inspections and experience-based judgment, resulting in low inspection efficiency and a high risk of errors.
[0006] 2. Existing image-based de-icing detection methods are susceptible to the complex and ever-changing de-icing inspection environment, resulting in low stability and accuracy of the detection results.
[0007] Therefore, an intelligent de-icing device and de-icing method for power transmission lines were invented. Summary of the Invention
[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution:
[0009] A smart de-icing device for power transmission lines includes an IPC de-icing computing chip. The input terminal of the IPC de-icing computing chip is connected to a thermal imaging sensor for detecting the area of the power transmission line requiring de-icing and obtaining a thermal image of the power transmission line, and an image sensor for obtaining an optical image of the power transmission line. The output terminal of the IPC de-icing computing chip is connected to a moving device for moving the de-icing device to a designated area to be de-iced, and a heating device for heating and de-icing the area to be de-iced.
[0010] The IPC ice-melting computing chip includes:
[0011] The positioning and scanning unit is used to identify and locate the area to be melted in the on-site ice melting and fusion image;
[0012] Intelligent movement and operation unit, used to control the mobile device to move the ice-melting device to the designated area to be melted;
[0013] The thermal de-icing unit is used to control the heating device to heat and melt the ice in the area to be melted;
[0014] The safety protection and monitoring unit is used to monitor the heating device controlled by the thermal de-icing unit in real time to ensure the safe operation of the device.
[0015] In a preferred embodiment of the intelligent de-icing device for power transmission lines described in this invention, the heating device comprises:
[0016] Coil housing;
[0017] A heating roller, which is fixedly installed on the middle part of the coil housing;
[0018] A focusing magnetic core is provided on the outer side of the heating roller;
[0019] A high-frequency induction coil array is provided on the outer side of several of the focusing magnetic cores;
[0020] A temperature probe is provided in the middle of the heating roller.
[0021] A smart de-icing method for power transmission lines includes the following specific steps:
[0022] S1: Collect on-site data and perform data preprocessing using thermal imaging sensors and image sensors;
[0023] S2: Design an IPC ice-melting computing chip to fuse on-site ice-melting images. Inputting data into the positioning and scanning unit of the IPC ice-melting computing chip identifies and locates the on-site ice-melting and fusion images. The chip locates the ice-melting area; then, the intelligent movement and operation unit in the chip controls the moving device to move the ice-melting device to the designated ice-melting area; then, the chip's thermal ice-melting unit controls the heating device to heat and melt the ice in the ice-melting area; while performing the above operations, the chip's safety protection and monitoring unit monitors the heating device controlled by the thermal ice-melting unit in real time to ensure the safe operation of the device.
[0024] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of step S1 are as follows:
[0025] S11, the thermal imaging sensor is used to detect the area of the transmission line that needs de-icing, and a thermal imaging image D of the transmission line is obtained. thermal The optical image sensor is used to detect the area of the power transmission line that needs de-icing, and an optical image D of the power transmission line is obtained. image ;
[0026] S12, for D thermal and D image Perform data preprocessing: for D thermal and D image Median filtering is performed to remove isolated points in the image. Then, different weights are assigned based on the importance of the thermal imaging data and the optical image data. Finally, the data at corresponding locations are weighted and summed to obtain the de-icing fusion image of the transmission line. ;
[0027]
[0028] in, Assigning importance weights to on-site thermal imaging data. Assign importance weights to on-site image data. .
[0029] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of S2 are as follows:
[0030] S21, Design an ice-melting area identification model, After being input into the positioning and scanning unit, it is automatically identified and located. Location of the melting ice area;
[0031] S22 uses an intelligent motion and control unit to receive the target regression box coordinates of the ice-covered area output by the positioning and scanning unit, generate motion commands for the mobile device, and control the ice melting device to reach the target position.
[0032] S23, the thermal energy de-icing unit controls the heating device to heat and melt the ice in the area to be melted, and sets the de-icing confidence threshold. The positioning and scanning unit is periodically invoked to update the icing area. until At that time, the heating and melting process ends;
[0033] S24, the safety protection and monitoring unit monitors the temperature and current of the heating device in real time when controlling the heating device of the ice melting thermal energy unit to prevent overheating, short circuit or incomplete ice melting, thereby ensuring the safe operation of the device, and records all abnormal events and timestamps for subsequent accident analysis.
[0034] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of S21 are as follows:
[0035] S211, Design a DFEM variable feature extraction module to perform feature extraction operations on the feature map of the melting ice region to obtain a feature map of the melting ice region that incorporates more information about the melting ice region;
[0036] S212, Design a D-Det deformable detector, input the feature map of the melting area to detect the melting area, and thus obtain the regression box and confidence of the melting area;
[0037] S213, based on the DFEM module, DCPF module, and D-Det deformable detector, proposes an IMDM model for identifying melting ice regions. This model accepts melting ice fusion images of transmission lines. As input, it automatically identifies and locates icy areas in the image.
[0038] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of S211 are as follows:
[0039] S2111, Input a feature map of the melting region with dimensions H×W×C. The feature map of the melting region with size H×W×C is obtained by passing the convolution kernel C with a kernel size of 3×3 and a padding value of 2. ;
[0040] S2112, A feature map of the melting region with dimensions H×W×C. The ice-melting region feature maps, each with a size of H×W×C / 2, were obtained by splitting along the channel. and ;
[0041] S2113, will After average pooling, the data is input into a deformable convolution, followed by batch normalization (BN) and a non-linear transformation using the Swish activation function. Finally, average pooling is performed to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ;
[0042] S2114, will After max pooling, the data is fed into a deformable convolution, followed by batch normalization (BN), a non-linear transformation using the Swish activation function, and then max pooling to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ;
[0043] S2115, and Perform a Concat operation along the channel to obtain a feature map of the melting region with dimensions H×W×C. Next The input is fed into a CBS module with C convolutional kernels of size 3×3 and padding of 2, which produces and outputs a feature map of the melting ice region with dimensions H×W×C. ;
[0044] S2116, Input a feature map of the melting region with dimensions H×W×C. In the three different branches, in the first branch, The ice-melting region feature map is calculated using deformable convolutions with C kernels, a size of 3×3, and padding of 2. After batch normalization (BN), a ReLU activation function is applied for non-linear transformation to obtain a feature map of size H×W×C. In the second branch, The ice-melting region feature map of size H×W×C is obtained by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2, followed by BN layers and ReLU activation function. In the third branch, The ice-melting region feature map of size H×W×C is obtained by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2, followed by BN layers and ReLU activation function. ;
[0045] S2117, , , Perform a Concat operation along the channel to obtain a feature map of the melting region with dimensions H×W×3C. Then The input is fed into a CBS module with C kernels, a size of 3×3, and a padding value of 2, to obtain a feature map of the melting region with a size of H×W×C. ;
[0046] S2118, The input is fed into three consecutive max-pooling layers, which sequentially produce feature maps of the melting region with dimensions H×W×C. , , Next , , , Performing a Concat operation along the channel yields a feature map of the melting region with dimensions H×W×4C. Then The input is fed into a CBS module with C convolutional kernels of size 3×3 and padding of 2, resulting in and outputting a feature map of the melting ice region with dimensions H×W×C. .
[0047] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of S212 are as follows:
[0048] S2121, Input a feature map of the melting region with dimensions H×W×C. The CBS module with C kernels, a size of 3×3, and a padding value of 2 is used to obtain a feature map of the melting region with a size of H×W×C. ;
[0049] S2122, will The input is fed into two branches. In the first branch, The input is fed into a deformable convolution with C kernels and a size of 3×3, and then subjected to batch normalization (BN) and ReLU function operations to obtain a feature map of the melting region with a size of H×W×C. In the second branch, The input is fed into a deformable convolution with C kernels and a size of 3×3, and then subjected to batch normalization (BN) and ReLU function operations to obtain a feature map of the melting region with a size of H×W×C. ;Will and By summing the features, a feature map of the melting region with dimensions H×W×C is obtained. Next, Input is routed to two branches, in the upper branch The input is fed into a convolutional kernel with 4 kernels and a size of 1×1 to obtain a feature map of the melting ice region with a size of H×W×4. , The ice-covered target regression box is obtained from the feature image of the melting area; in the next branch The input is fed into a 2D convolutional layer with one kernel and a size of 1×1 to obtain a feature map of the melting region with a size of H×W×1. , The regression bounding boxes and confidence scores of the ice-covered targets in the feature image of the melting ice area were obtained through calculation.
[0050] As a preferred embodiment of the intelligent de-icing method for transmission lines according to the present invention, the specific steps of S213 are as follows:
[0051] S2131, Input a transmission line de-icing fusion image with dimensions H×W×3. The feature map of the melting ice region with a size of H / 2×W / 2×64 is obtained by using a CBS module with 64 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then The input is fed into a CBS module with 128 kernels, a size of 3×3, a stride of 2, and a padding value of 1, resulting in a feature map of the melting ice region with a size of H / 4×W / 4×128. ;
[0052] S2132, will The input is fed into a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with a size of H / 8×W / 8×256. Feature map of the melting area with dimensions H / 8×W / 8×256 ;Will The input is fed into a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. This is then fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will The input is processed by a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is processed by a DFEM module to obtain a feature map of the melting ice region with a size of H / 32×W / 32×512. ;Will The input is fed into DCPF to obtain a feature map of the melting ice region with dimensions of H / 32×W / 32×512. ;
[0053] S2133, will An upsample operation was performed to obtain a feature map of the melting region with dimensions H / 16×W / 16×512. ;Will and After performing a Concat operation along the channel and inputting it into DFEM, a feature map of the melting region with dimensions of H / 16×W / 16×1024 is obtained. ;
[0054] S2134, will An upsample operation was performed to obtain a feature map of the melting region with dimensions H / 8×W / 8×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting area with dimensions H / 8×W / 8×768. ;
[0055] S2135, Inputting the CBS module with 256 kernels, a size of 3×3, a stride of 2, and padding of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ;Will and After performing a Concat operation along the channel, inputting the data into DFEM yields a feature map of the melting area with dimensions H / 8×W / 8×1280. ;
[0056] S2136, feature map of melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting ice region with dimensions of H / 16×W / 16×1024. ;
[0057] S2137, feature map of the melting ice area Inputting a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ; Map the characteristics of the melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 32×W / 32×512. ; Map the characteristics of the melting ice area Inputting a CBS module with 1024 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 64×W / 64×1024. ;
[0058] S2138, , , The input is fed into the D-Det deformable detector to obtain and output the icing identification result.
[0059] Compared with existing technologies:
[0060] This invention designs an IPC (Internet Protocol Cell) de-icing calculation chip. The chip inputs a de-icing fusion image of a transmission line into its positioning and scanning unit. This unit automatically identifies and locates the area to be de-iced in the image. Then, the chip's intelligent movement and operation unit controls a mobile device to move the de-icing device to the designated area. The chip's thermal de-icing unit then controls a heating device to melt the ice in the area. Simultaneously, the chip's safety protection and monitoring unit monitors the heating device in real time to ensure safe operation. Furthermore, this invention constructs a high-precision de-icing area recognition model architecture (IMDM) within the positioning and scanning unit of the IPC de-icing calculation chip. This significantly improves the model's ability to extract features from the de-icing fusion image, thereby greatly enhancing the model's detection accuracy and real-time response capability for de-icing areas. This solves the problem of traditional image-based de-icing detection struggling to correctly detect ice in complex and changing de-icing inspection environments. Attached Figure Description
[0061] Figure 1 This is a schematic diagram of the framework of the present invention;
[0062] Figure 2 This is a structural diagram of the DFEM of the present invention;
[0063] Figure 3 This is a structural diagram of the DCPF of the present invention;
[0064] Figure 4 This is a structural diagram of the D-Det of the present invention;
[0065] Figure 5 This is a structural diagram of the IMDM model of the present invention;
[0066] Figure 6 This is a schematic diagram of the heating device structure of the present invention.
[0067] In the figure: coil housing 101, high-frequency induction coil array 201, heating roller 202, focusing magnetic core 203, temperature probe 204. Detailed Implementation
[0068] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.
[0069] This invention provides an intelligent de-icing device for power transmission lines. Please refer to [link / reference]. Figures 1-6The device includes an IPC de-icing computing chip. The input terminal of the IPC de-icing computing chip is connected to a thermal imaging sensor for detecting the field area of the transmission line that needs de-icing and obtaining a thermal image of the transmission line, and an image sensor for obtaining an optical image of the transmission line. The output terminal of the IPC de-icing computing chip is connected to a moving device for moving the de-icing device to a designated area to be de-iced, and a heating device for heating and de-icing the area to be de-iced.
[0070] The IPC ice-melting computing chip includes:
[0071] The positioning and scanning unit is used to identify and locate the area to be melted in the on-site ice melting and fusion image;
[0072] Intelligent movement and operation unit, used to control the mobile device to move the ice-melting device to the designated area to be melted;
[0073] The thermal de-icing unit is used to control the heating device to heat and melt the ice in the area to be melted;
[0074] The safety protection and monitoring unit is used to monitor the heating device controlled by the thermal de-icing unit in real time to ensure the safe operation of the device.
[0075] The heating device includes: a coil housing 101, a high-frequency induction coil array 201, a heating roller 202, a focusing magnetic core 203, and a temperature measuring probe 204;
[0076] The heating roller 202 is fixedly installed on the middle part of the coil housing 101. A plurality of focusing magnetic cores 203 are provided on the outer side of the heating roller 202. A high-frequency induction coil array 201 is provided on the outer side of the plurality of focusing magnetic cores 203. A temperature measuring probe 204 is provided in the middle part of the heating roller 202.
[0077] When the device is in operation, the control system drives the high-frequency induction coil array 201 to generate a high-frequency alternating magnetic field. The focusing magnetic core 203 concentrates the magnetic field energy and precisely guides it to the heating roller 202. A strong eddy current is induced in the metal body of the heating roller 202, thereby rapidly heating it to generate heat for ice melting. During this process, the temperature probe 204 monitors the temperature of the heating roller 202 in real time and feeds the temperature data back to the control system to achieve closed-loop power regulation and ensure that the ice melting process is stable and efficient.
[0078] A smart de-icing method for power transmission lines includes the following specific steps:
[0079] S1: Collect on-site data and perform data preprocessing using thermal imaging sensors and image sensors.
[0080] The specific steps of S1 are as follows:
[0081] S11, the thermal imaging sensor is used to detect the area of the transmission line that needs de-icing, and a thermal imaging image D of the transmission line is obtained. thermal The optical image sensor is used to detect the area of the power transmission line that needs de-icing, and an optical image D of the power transmission line is obtained. image ;
[0082] S12, for D thermal and D image Perform data preprocessing: for D thermal and D image Median filtering is performed to remove isolated points in the image. Then, different weights are assigned based on the importance of the thermal imaging data and the optical image data. Finally, the data at corresponding locations are weighted and summed to obtain the de-icing fusion image of the transmission line. ;
[0083]
[0084] in, Assigning importance weights to on-site thermal imaging data. Assign importance weights to on-site image data. .
[0085] S2: Design an IPC ice-melting computing chip (IceMelt Pro-Chip) to fuse on-site ice-melting images. Inputting data into the positioning and scanning unit of the IPC ice-melting computing chip identifies and locates the on-site ice-melting and fusion images. The chip locates the ice-melting area; then, the intelligent movement and operation unit in the chip controls the moving device to move the ice-melting device to the designated ice-melting area; then, the chip's thermal ice-melting unit controls the heating device to heat and melt the ice in the ice-melting area; while performing the above operations, the chip's safety protection and monitoring unit monitors the heating device controlled by the thermal ice-melting unit in real time to ensure the safe operation of the device.
[0086] The specific steps of S2 are as follows:
[0087] S21, Design an ice-melting area identification model, After being input into the positioning and scanning unit, it is automatically identified and located. The model is run in the positioning and scanning unit of the IPC ice melting computing chip, which determines the location of the ice melting area.
[0088] The specific steps of S21 are as follows:
[0089] S211, Design a DFEM (Deformable Feature Extraction Module) to perform feature extraction operations on the feature map of the melting ice region, and obtain a feature map of the melting ice region that incorporates more information about the melting ice region;
[0090] This module dynamically adjusts the sampling space distribution of the convolution kernels through two parallel sets of deformable convolution branches, expanding the receptive field of detailed features in the image of the melting ice region. This allows for better capture of information about the melting ice region in the image, enhancing the model's ability to recognize the melting ice region. The structure of the DFEM module is as follows: Figure 2 As shown.
[0091] The specific steps of S211 are as follows:
[0092] S2111, Input a feature map of the melting region with dimensions H×W×C. The feature map of the melting region with size H×W×C is obtained by passing the convolution kernel C with a kernel size of 3×3 and a padding value of 2. ;
[0093] in:
[0094] The CBS module is a commonly used deep learning module consisting of Convolutional layers (Conv), Batch Normalization (BN) layers, and SiLu (activation function). It is primarily used for feature extraction and transformation. It extracts features from the input data through convolutional operations, performs normalization using BN layers, and provides non-linearity through the SiLu activation function. The feature map obtained after computation by the CBS module for any image has the same number of channels as the number of convolutional kernels in the CBS module.
[0095] An image with dimensions H×W×C refers to an image with a length of H pixels, a width of W pixels, and a number of channels of C.
[0096] S2112, A feature map of the melting region with dimensions H×W×C. The ice-melting region feature maps, each with a size of H×W×C / 2, were obtained by splitting along the channel. and ;
[0097] S2113, will After average pooling, the data is input into a deformable convolution, followed by batch normalization (BN) and a non-linear transformation using the Swish activation function. Finally, average pooling is performed to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ;
[0098]
[0099] in, This indicates the average pooling operation. Represents deformable convolution. This indicates batch normalization.
[0100] S2114, will After max pooling, the data is fed into a deformable convolution, followed by batch normalization (BN), a non-linear transformation using the Swish activation function, and then max pooling to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ;
[0101]
[0102] in, This indicates a max pooling operation.
[0103] S2115, and Perform a Concat operation (i.e., feature stitching) along the channel to obtain a feature map of the melting area with dimensions H×W×C. Next The input is fed into a CBS module with C convolutional kernels of size 3×3 and padding of 2, which produces and outputs a feature map of the melting ice region with dimensions H×W×C. ;
[0104] A DCPF (Deformable Convolution Fusion Spatial Pooling Factorization) pooling module was designed to extract features from the feature map of the melting ice region, resulting in a feature map that incorporates more information from melting ice regions at different scales. In the DCPF pooling module, parallel deformable convolutional branches are used to extract features from the melting ice region image, and pooling operations are used to fuse the features of melting ice region images at different scales, improving the model's ability to identify icy areas in melting ice region images at different scales. The structure of the DCPF module is as follows: Figure 3 As shown.
[0105] S2116, Input a feature map of the melting region with dimensions H×W×C. In the three different branches, in the first branch, The ice-melting region feature map is calculated using deformable convolutions with C kernels, a size of 3×3, and padding of 2. After batch normalization (BN), a ReLU activation function is applied for non-linear transformation to obtain a feature map of size H×W×C. In the second branch, The ice-melting region feature map of size H×W×C is obtained by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2, followed by BN layers and ReLU activation function. In the third branch, The ice-melting region feature map of size H×W×C is obtained by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2, followed by BN layers and ReLU activation function. ;
[0106] S2117, , , Perform a Concat operation along the channel to obtain a feature map of the melting region with dimensions H×W×3C. Then The input is fed into a CBS module with C kernels, a size of 3×3, and a padding value of 2, to obtain a feature map of the melting region with a size of H×W×C. ;
[0107]
[0108] in, This is for overlay operations along the channel.
[0109] S2118, Input into 3 consecutive max pooling layers (i.e.) Figure 3 Maxpooling (in the process) yields feature maps of melting regions with dimensions H×W×C, respectively. , , Next , , , Performing a Concat operation along the channel yields a feature map of the melting region with dimensions H×W×4C. Then The input is fed into a CBS module with C kernels, a size of 3×3, and a padding value of 2, to obtain and output a feature map of the melting ice region with a size of H×W×C. ;
[0110]
[0111] S212, a D-Det (Deformable Detection) detector is designed. It takes a feature map of the ice-melting region as input and performs ice-melting region detection to obtain the regression box and confidence score of the ice-melting region. The D-Det deformable detector uses a deformable hybrid convolutional structure to significantly expand the effective receptive field of the convolutional kernel while maintaining the feature map resolution. This helps to extract more contextual information from the input feature map and capture target features at different scales. This allows the detection head to extract rich feature information, thereby improving detection accuracy and efficiency. The structure of the D-Det deformable detector is as follows: Figure 4 As shown.
[0112] The specific steps of S212 are as follows:
[0113] S2121, Input a feature map of the melting region with dimensions H×W×C. The CBS module with C kernels, a size of 3×3, and a padding value of 2 is used to obtain a feature map of the melting region with a size of H×W×C. ;
[0114] S2122, will The input is fed into two branches. In the first branch, The input is fed into a deformable convolution with C kernels and a size of 3×3, and then subjected to batch normalization (BN) and ReLU function operations to obtain a feature map of the melting region with a size of H×W×C. In the second branch, The input is fed into a deformable convolution with C kernels and a size of 3×3, and then subjected to batch normalization (BN) and ReLU function operations to obtain a feature map of the melting region with a size of H×W×C. ;Will and Perform feature addition (i.e.) Figure 4 The ADD operation in the middle obtains a feature map of the melting region with dimensions H×W×C. Next, Input is routed to two branches, in the upper branch The input is fed into a convolutional kernel with 4 kernels and a size of 1×1 to obtain a feature map of the melting ice region with a size of H×W×4. , The ice-covered target regression box is obtained from the feature image of the melting area; in the next branch The input is fed into a 2D convolutional layer with one kernel and a size of 1×1 to obtain a feature map of the melting region with a size of H×W×1. , The regression bounding boxes and confidence scores of the ice-covered targets in the feature image of the melting ice area were obtained through calculation.
[0115] S213, based on the DFEM module, DCPF module, and D-Det deformable detector, proposes an Ice-Melt Region Detection Model (IMDM). This model accepts ice-melting fusion images of transmission lines. Using this as input, the system automatically identifies and locates icy areas in the image. The structure of the IMDM model is shown in the figure.
[0116] The specific steps of S213 are as follows:
[0117] S2131, Input a transmission line de-icing fusion image with dimensions H×W×3. The feature map of the melting ice region with a size of H / 2×W / 2×64 is obtained by using a CBS module with 64 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then The input is fed into a CBS module with 128 kernels, a size of 3×3, a stride of 2, and a padding value of 1, resulting in a feature map of the melting ice region with a size of H / 4×W / 4×128. ;
[0118] S2132, will The input is fed into a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with a size of H / 8×W / 8×256. Feature map of the melting area with dimensions H / 8×W / 8×256 ;Will The input is fed into a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. This is then fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will The input is processed by a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is processed by a DFEM module to obtain a feature map of the melting ice region with a size of H / 32×W / 32×512. ;Will The input is fed into DCPF to obtain a feature map of the melting ice region with dimensions of H / 32×W / 32×512. ;
[0119] S2133, Performing an upsample operation yields a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will and After performing a Concat operation along the channel and inputting it into DFEM, a feature map of the melting region with dimensions of H / 16×W / 16×1024 is obtained. ;
[0120] S2134, will An upsample operation was performed to obtain a feature map of the melting region with dimensions H / 8×W / 8×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting area with dimensions H / 8×W / 8×768. ;
[0121] S2135, Inputting the CBS module with 256 kernels, a size of 3×3, a stride of 2, and padding of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ;Will and After performing a Concat operation along the channel, inputting the data into DFEM yields a feature map of the melting area with dimensions H / 8×W / 8×1280. ;
[0122] S2136, feature map of melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting ice region with dimensions of H / 16×W / 16×1024. ;
[0123] S2137, feature map of melting ice area Inputting a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ; Map the characteristics of the melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 32×W / 32×512. ; Map the characteristics of the melting ice area Inputting a CBS module with 1024 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 64×W / 64×1024. ;
[0124] S2138, , , The input is fed into a D-Det deformable detector, which yields and outputs icing identification results, including bounding boxes of iced areas in the image and melting confidence scores. .
[0125] S22 uses an intelligent motion and control unit to receive the target regression box coordinates of the ice-covered area output by the positioning and scanning unit, generate motion commands for the mobile device, and control the ice melting device to reach the target position.
[0126] S23, the thermal energy de-icing unit controls the heating device to heat and melt the ice in the area to be melted, and sets the de-icing confidence threshold. The positioning and scanning unit is periodically invoked to update the icing area. until At that time, the heating and melting process ends;
[0127] Its heating element consists of a coil housing, a high-frequency induction coil array, a heating roller, a focusing magnetic core, and a temperature probe. Compared to other similar devices, this heating element exhibits significant advantages in material selection, structural design, safety protection, and monitoring functions. It employs several advanced technologies; the grounding resistance of the coil housing and the inclusion of a leakage current protector effectively prevent leakage accidents. The spiral cooling water channel design inside the heating roller effectively removes heat, preventing overheating and protecting the normal operation of the device. The safety protection and monitoring unit can monitor the temperature and current of the heating element in real time, preventing overheating, short circuits, or incomplete de-icing. Simultaneously, it records all abnormal events and timestamps for subsequent accident analysis, providing strong evidence for device maintenance and improvement, greatly enhancing the device's safety and reliability. A top-view cross-sectional view of the device is shown below. Figure 6 As shown.
[0128] in:
[0129] Coil housing: Injection molded from high-strength glass fiber reinforced polyamide with 60% glass fiber content, tensile strength ≥260 MPa, flexural modulus ≥21 GPa, grounding resistance ≤0.1Ω, equipped with a leakage current protector with an operating current of 30mA and an operating time of 0.1s.
[0130] High-frequency induction coil array: Employs a double-helix parallel winding structure, using Φ8mm×1mm T2 copper tubing with a 10mm spacing between each turn, totaling 24 turns, covering an effective heating zone of 1100mm on the roller. The coil is 3mm away from the surface of the heating roller, secured by 99.7% alumina, with a gap error ≤±0.1mm.
[0131] Heating roller: The outer layer is a seamless stainless steel tube with a wall thickness of 12mm~18mm and a hard chrome plating thickness of 15μm~25μm. The inner layer adopts a spiral cooling water channel with a pitch of 40mm~60mm and a channel diameter of Φ10mm~Φ14mm.
[0132] Focusing magnetic core: Made of manganese-zinc ferrite (PC40 material), with an initial permeability ≥2300, saturation magnetic flux density Bs ≥0.39T, and an operating frequency range of 50kHz~200kHz. The magnetic core is divided into 8 arc-shaped blocks (each covering an angle of 45°), which are fixed to the outside of the heating roller by spring clips.
[0133] Temperature probe: A Pt100 thin-film platinum resistance probe is used as the temperature probe. The probe shell is a 3 mm × 15 mm 316L stainless steel sheath, and the end is filled with high thermal conductivity magnesium oxide insulating powder. It is fixed to the inner wall of the heating roller by M8 × 1 thread and polytetrafluoroethylene gasket. The wire is a 3-core silver-plated FEP shielded wire, and the shielding layer is grounded at one end to suppress common mode interference.
[0134] S24, the safety protection and monitoring unit monitors the temperature and current of the heating device in real time when controlling the heating device of the ice melting thermal energy unit to prevent overheating, short circuit or incomplete ice melting, thereby ensuring the safe operation of the device, and records all abnormal events and timestamps for subsequent accident analysis.
[0135] Based on the above, the present invention includes, but is not limited to, the following embodiments:
[0136] A 512×512×3 thermal image of the ice melting site was obtained using a thermal imaging sensor. thermal An image sensor was used to obtain a 512×512×3 image of the area requiring ice melting. image .
[0137] Thermal imaging image of transmission line D thermal Image D of transmission lines image Perform data preprocessing. For D thermal and D image Assign different weights, among which The weight is 0.3. With a weight of 0.7, the data at corresponding locations are weighted and summed to obtain a 512×512×3 image of the power transmission line de-icing process. .
[0138]
[0139] Will Inputting the data into the positioning and scanning unit of the IPC ice-melting computing chip, the positioning and scanning unit automatically identifies and locates the data. The chip identifies the area to be melted, and then the chip's intelligent movement and operation unit controls the moving device to move the ice-melting device to the designated area. The chip's thermal ice-melting unit then controls the heating device to melt the ice in the area. Simultaneously, the chip's safety protection and monitoring unit monitors the heating device controlled by the thermal ice-melting unit in real time to ensure the safe operation of the device. The specific process is as follows:
[0140] The size is 512×512×3 The feature map of the melting ice region, with a size of 256×256×64, is obtained by using a CBS module with 64 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then The input is fed into a CBS module with 128 kernels, a size of 3×3, a stride of 2, and a padding value of 1, resulting in a feature map of the melting ice region with a size of 128×128×128. .
[0141] Will The input is fed into a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is fed into a DFEM for feature extraction, resulting in a feature map of the melting ice region with a size of 64×64×256. and a feature map of the melting ice area with dimensions of 64×64×256. .Will The input is fed into a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. This is then fed into a DFEM module for feature extraction, resulting in a 32×32×512 feature map of the melting ice region. .Will Inputting the data into a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, followed by inputting DFEM, yields a feature map of the melting ice region with a size of 16×16×512. .Will The input is fed into DCPF to obtain a feature map of the melting ice region with a size of 16×16×512. .
[0142] Will An upsample operation was performed to obtain a feature map of the melting ice region with a size of 32×32×512. .Will and After performing a Concat operation along the channel, the DFEM image is input to obtain a feature map of the melting area with a size of 32×32×1024. .
[0143] Will An upsample operation was performed to obtain a feature map of the melting ice region with a size of 64×64×512. .Will and After performing a Concat operation along the channel, the DFEM image was input to obtain a feature map of the melting area with dimensions of 64×64×768. .
[0144] Will Inputting the CBS module with 256 kernels, a size of 3×3, a stride of 2, and padding of 1, yields a feature map of the melting ice region with a size of 32×32×256. .Will and After performing a concat operation along the channel, the DFEM image is input to obtain a feature map of the melting area with dimensions of 64×64×1280. .
[0145] Feature map of the melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with a size of 32×32×512. .Will and After performing a Concat operation along the channel, the DFEM image is input to obtain a feature map of the melting area with a size of 32×32×1024. .
[0146] Feature map of the melting ice area Inputting a CBS module with 256 kernels, a size of 3×3, a stride of 2, and padding of 1, yields a feature map of the melting ice region with a size of 32×32×256. Feature map of the melting ice area. Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with a size of 16×16×512. Feature map of the melting ice area. Inputting the CBS module with 1024 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with a size of 8×8×1024. .
[0147] Feature map of the melting ice area , , The input is fed into the D-Det deformable detector to obtain and output the icing recognition results, which include the bounding boxes and confidence scores of the icing areas in the image. .
[0148] The intelligent motion and control unit generates motion commands for the mobile device by receiving the bounding box coordinates of the icing area output by the positioning and scanning unit, thereby controlling the ice-melting device to reach the target position. An ice-melting confidence threshold K=0.5 is set, and then the thermal ice-melting unit controls the heating device to heat and melt the ice in the area to be melted. The positioning and scanning unit is periodically invoked to update the icing area. until At that time, the heating and de-icing process ends. Simultaneously, the safety protection and monitoring unit continues to operate, monitoring the temperature and current of the heating device in real time to prevent overheating, short circuits, or incomplete de-icing, thereby ensuring the safe operation of the device. It also records all abnormal events and timestamps for subsequent analysis.
[0149] Although the present invention has been described above with reference to embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of the invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments can be combined with each other in any manner. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.
Claims
1. A smart de-icing method for power transmission lines, characterized in that, The specific steps are as follows: S1: Collect on-site data and perform data preprocessing using thermal imaging sensors and image sensors; S2: Design an IPC ice-melting computing chip to fuse on-site ice-melting images. Inputting data into the positioning and scanning unit of the IPC ice-melting computing chip identifies and locates the on-site ice-melting and fusion images. The chip locates the ice-melting area; then, the chip's intelligent movement and operation unit controls the movement device to move the ice-melting device to the designated ice-melting area; the chip's thermal ice-melting unit controls the heating device to heat and melt the ice in the ice-melting area; while performing the above operations, the chip's safety protection and monitoring unit monitors the heating device controlled by the thermal ice-melting unit in real time to ensure the safe operation of the device. The specific steps of S1 are as follows: S11, the thermal imaging sensor is used to detect the area of the transmission line that needs de-icing, and a thermal imaging image D of the transmission line is obtained. thermal The optical image sensor is used to detect the area of the power transmission line that needs de-icing, and an optical image D of the power transmission line is obtained. image ; S12, for D thermal and D image Perform data preprocessing: For D thermal and D image Median filtering is performed to remove isolated points in the image. Then, different weights are assigned based on the importance of the thermal imaging data and the optical image data. Finally, the data at corresponding locations are weighted and summed to obtain the de-icing fusion image of the transmission line. ; in, Assigning importance weights to on-site thermal imaging data. Assign importance weights to on-site image data. ; The specific steps of S2 are as follows: S21, Design an ice-melting area identification model, After being input into the positioning and scanning unit, it is automatically identified and located. Location of the melting ice area; S22 uses an intelligent motion and control unit to receive the target regression box coordinates of the ice-covered area output by the positioning and scanning unit, generate motion commands for the mobile device, and control the ice melting device to reach the target position. S23, the thermal energy de-icing unit controls the heating device to heat and melt the ice in the area to be melted, and sets the de-icing confidence threshold. The positioning and scanning unit is periodically invoked to update the icing area. until At that time, the heating and melting process ends; S24, the safety protection and monitoring unit monitors the temperature and current of the heating device in real time when controlling the heating device of the ice melting thermal energy unit to prevent overheating, short circuit or incomplete ice melting, thereby ensuring the safe operation of the device, and records all abnormal events and timestamps for subsequent accident analysis. The specific steps of S21 are as follows: S211, a DFEM variable feature extraction module is designed to perform feature extraction operations on the feature map of the melting ice region, resulting in a feature map of the melting ice region that incorporates more information about the melting ice region. This module dynamically adjusts the sampling space distribution of the convolution kernel through two sets of parallel deformable convolution branches, expanding the receptive field of detailed features in the melting ice region image, thereby better capturing the information of the melting ice region in the image and enhancing the model's ability to recognize the melting ice region. A DCPF pooling module is also designed to perform feature extraction operations on the feature map of the melting ice region, resulting in a feature map of the melting ice region that incorporates more information about the melting ice region at different scales. In the DCPF pooling module, parallel deformable convolution branches are used to extract features from the image of the melting ice region, and pooling operations are used to fuse the features of the image of the melting ice region at different scales, improving the model's ability to recognize the icy areas in the melting ice region images at different scales. S212, Design a D-Det deformable detector, input the feature map of the melting ice region to detect the melting ice region, and thus obtain the regression box and confidence of the melting ice region; The D-Det deformable detector significantly expands the effective receptive field of the convolution kernel while maintaining the resolution of the feature map through a deformable hybrid convolution structure. S213, based on the DFEM module, DCPF module, and D-Det deformable detector, proposes an IMDM model for identifying melting ice regions. This model accepts melting ice fusion images of transmission lines. As input, it automatically identifies and locates icy areas in the image.
2. The intelligent de-icing method for transmission lines according to claim 1, characterized in that, The specific steps of the DFEM variable feature extraction module in S211 are as follows: S2111, Input a feature map of the melting region with dimensions H×W×C. The feature map of the melting region with size H×W×C is obtained by passing the convolution kernel C with a kernel size of 3×3 and a padding value of 2. ; S2112, A feature map of the melting region with dimensions H×W×C. The ice-melting region feature maps, each with a size of H×W×C / 2, were obtained by splitting along the channel. and ; S2113, will After average pooling, the data is input into a deformable convolution, followed by batch normalization (BN) and a non-linear transformation using the Swish activation function. Finally, average pooling is performed to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ; S2114, will After max pooling, the data is fed into a deformable convolution, followed by batch normalization (BN), a non-linear transformation using the Swish activation function, and then max pooling to obtain a feature map of the melting ice region with dimensions H×W×C / 2. ; S2115, and Perform a Concat operation along the channel to obtain a feature map of the melting region with dimensions H×W×C. Next The input is fed into a CBS module with C convolutional kernels of size 3×3 and padding of 2, which produces and outputs a feature map of the melting ice region with dimensions H×W×C. .
3. The intelligent de-icing method for transmission lines according to claim 2, characterized in that, The specific steps of the DCPF variable convolutional fusion spatial pooling module in S211 are as follows: S2116, Input a feature map of the melting region with dimensions H×W×C. In the three different branches, in the first branch, The ice-melting region feature map is calculated using deformable convolutions with C kernels, a size of 3×3, and padding of 2. After batch normalization (BN), a ReLU activation function is applied for non-linear transformation to obtain a feature map of size H×W×C. In the second branch, The ice-melting region feature map is calculated by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2. After batch normalization (BN) and a ReLU activation function, a non-linear transformation is applied to obtain the feature map. The result is a feature map of size H×W×C representing the melting region. ; In the third branch, The ice-melting region feature map is calculated by performing deformable convolutions with C kernels, a size of 3×3, and padding of 2. After batch normalization (BN) and a ReLU activation function, a non-linear transformation is applied to obtain the feature map. The result is a feature map of size H×W×C representing the melting region. ; S2117, , , Perform a Concat operation along the channel to obtain a feature map of the melting region with dimensions H×W×3C. Then The input is fed into a CBS module with C kernels, a size of 3×3, and a padding value of 2, to obtain a feature map of the melting region with a size of H×W×C. ; S2118, The input is fed into three consecutive max-pooling layers, which sequentially produce feature maps of the melting region with dimensions H×W×C. , , Next , , , Performing a Concat operation along the channel yields a feature map of the melting region with dimensions H×W×4C. Then The input is fed into a CBS module with C kernels, a size of 3×3, and a padding value of 2, to obtain and output a feature map of the melting ice region with a size of H×W×C. .
4. The intelligent de-icing method for transmission lines according to claim 1, characterized in that, The specific steps of the D-Det deformable detector in S212 are as follows: S2121, Input a feature map of the melting region with dimensions H×W×C. The CBS module with C kernels, a size of 3×3, and a padding value of 2 is used to obtain a feature map of the melting region with a size of H×W×C. ; S2122, will The input is fed into two branches. In the first branch, The input is fed into a deformable convolutional layer with C kernels and a size of 3×3. After batch normalization (BN) and ReLU activation functions, a non-linear transformation is performed to obtain a feature map of the melting ice region with a size of H×W×C. In the second branch, The input is fed into a deformable convolutional layer with C kernels and a size of 3×3. After batch normalization (BN), a non-linear transformation is performed using the ReLU activation function to obtain a feature map of the melting ice region with a size of H×W×C. ;Will and By summing the features, a feature map of the melting region with dimensions H×W×C is obtained. Next, Input is routed to two branches, in the upper branch The input is fed into a convolutional kernel with 4 kernels and a size of 1×1 to obtain a feature map of the melting ice region with a size of H×W×4. , The ice-covered target regression box is obtained from the feature image of the melting area; in the next branch The input is fed into a 2D convolutional layer with one kernel and a size of 1×1 to obtain a feature map of the melting region with a size of H×W×1. , The regression bounding boxes and confidence scores of the ice-covered targets in the feature image of the melting ice area were obtained through calculation.
5. The intelligent de-icing method for transmission lines according to claim 1, characterized in that, The specific steps of the ice melt area identification model IMDM in S213 are as follows: S2131, Input a transmission line de-icing fusion image with dimensions H×W×3. The feature map of the melting ice region with a size of H / 2×W / 2×64 is obtained by using a CBS module with 64 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then The input is fed into a CBS module with 128 kernels, a size of 3×3, a stride of 2, and a padding value of 1, resulting in a feature map of the melting ice region with a size of H / 4×W / 4×128. ; S2132, will The input is fed into a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with a size of H / 8×W / 8×256. ;Will The input is fed into a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. This is then fed into a DFEM module for feature extraction, resulting in a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will The input is processed by a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1. Then, it is processed by a DFEM module to obtain a feature map of the melting ice region with a size of H / 32×W / 32×512. ;Will The input is fed into DCPF to obtain a feature map of the melting ice region with dimensions of H / 32×W / 32×512. ; S2133, will An upsample operation was performed to obtain a feature map of the melting region with dimensions H / 16×W / 16×512. ;Will and After performing a Concat operation along the channel and inputting it into DFEM, a feature map of the melting region with dimensions of H / 16×W / 16×1024 is obtained. ; S2134, will An upsample operation was performed to obtain a feature map of the melting region with dimensions H / 8×W / 8×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting area with dimensions H / 8×W / 8×768. ; S2135, Inputting the CBS module with 256 kernels, a size of 3×3, a stride of 2, and padding of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ;Will and After performing a Concat operation along the channel, inputting the data into DFEM yields a feature map of the melting area with dimensions H / 8×W / 8×1280. ; S2136, feature map of melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×512. ;Will and After performing a concat operation along the channel, the data is input into the DFEM to obtain a feature map of the melting ice region with dimensions of H / 16×W / 16×1024. ; S2137, the feature map of the melting ice area D 11 Inputting a CBS module with 256 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 16×W / 16×256. ; Map the characteristics of the melting ice area Inputting a CBS module with 512 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 32×W / 32×512. ; Map the characteristics of the melting ice area Inputting a CBS module with 1024 kernels, a size of 3×3, a stride of 2, and a padding value of 1, yields a feature map of the melting ice region with dimensions H / 64×W / 64×1024. ; S2138, , , The input is fed into the D-Det deformable detector to obtain and output the icing identification result.