Power grid line ice melting effect detection method based on deep learning

Abstract

The invention belongs to the technical field of power grid lines, and particularly relates to a power grid line ice melting effect detection method based on deep learning, and the method comprises the following specific steps: S1, constructing a power grid line ice melting effect detection data set; s2, building a power grid line ice melting effect detection network GIMEDNet; s3, training a power grid line ice melting effect detection network GIMEDNet; and S4, detecting the ice melting effect of the power grid line by using the GIMEDNet. According to the invention, the GIMEDNet can greatly improve the quality and efficiency of ice melting effect detection through a bidirectional attention mechanism, and guarantees the high-quality detection of the ice melting effect of a power grid line.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid lines, and in particular to a method for detecting ice melting effects of power grid lines based on deep learning. Background Art

[0002] Power grid lines are prone to snow accumulation and ice formation in cold seasons, especially after icing weather. The melting of ice has a significant impact on the safety and stability of power equipment. Detecting the ice melting effect after icing is key to ensuring the normal operation of power grid lines and avoiding accidents. Traditional ice melting detection methods, such as manual inspections or automated detection based on simple image processing techniques (such as edge detection, threshold segmentation, etc.), usually have many limitations. Manual detection is not only inefficient and labor-intensive, but also easily affected by weather, viewing angles, and human factors, and it is difficult to achieve large-scale and real-time monitoring. Traditional image processing methods have poor adaptability to complex environments (such as different ice thicknesses, lighting changes, and background interference), making it difficult to achieve accurate and stable detection.

[0003] In summary, the traditional ice melting detection method has the following disadvantages:

[0004] (1) Traditional ice melting effect detection methods rely on manual observation and analysis, which has the disadvantages of high labor costs, slow detection speed, strong subjectivity of results, and poor adaptability.

[0005] (2) Solutions that have adopted deep learning technology often have problems such as too many deep learning network parameters, time-consuming training, and low operating efficiency. In addition, they have not been optimized for the detection of ice melting effects on power grid lines, resulting in poor detection results. Summary of the Invention

[0006] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0007] A method for detecting ice melting effects of power grid lines based on deep learning includes the following specific steps:

[0008] S1: Construct a dataset for detecting the de-icing effect of power grid lines;

[0009] S2: Build the power grid line ice melting effect detection network GIMEDNet;

[0010] S3: Training the power line ice melting effect detection network GIMEDNet;

[0011] S4: Apply GIMEDNet to detect the de-icing effect of power grid lines.

[0012] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the specific steps of S1 are as follows:

[0013] S11: collecting image data of power grid line ice melting results;

[0014] S12: The collected image data is annotated using annotation tool software to annotate the ice-covered areas in the image pixel by pixel. After the annotation is completed, a de-icing result mask map is obtained. The power grid line de-icing effect detection dataset consists of the original power grid line de-icing result image and its corresponding mask map.

[0015] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the specific steps of S2 are as follows:

[0016] S21: The power line ice melting result image F1 is input into GIMEDNet and first processed by the Conv(3×3)_BN_ReLU module to obtain the ice melting effect feature map F2. The Conv(3×3)_BN_ReLU module is composed of a Conv layer with a convolution kernel size of 3×3 and a stride of 1, a BN layer, and a ReLU layer in series.

[0017] S22: After F2 is downsampled by Conv(7×7), it is input into the RMBAMB module to obtain the ice melting effect feature map F3, and the resolution of F3 is half of F2; the RMBAMB module includes the DCMRB module, the MFEDSB module and the BAMB module; the DCMRB module uses deformable convolution for feature extraction and spatial adaptive transformation to adapt to the ice melting effect features of different deformations; the MFEDSB module uses depthwise separable convolution, which can improve the ice melting effect feature extraction capability while reducing computational complexity, reduce redundant calculations, improve the efficiency and generalization ability of the model, and enable the network to more fully capture and integrate key features of different scales; the BAMB module is used to implement a bidirectional attention mechanism, and the BAMB module combines the void convolution and HCA activation function, effectively improving the ice melting effect feature extraction capability;

[0018] S23: After the downsampling operation of Conv(3×3), F3 is input into the RMBAMB module, and the ice melting effect features of F3 are extracted and refined to obtain the ice melting effect feature map F4, whose resolution is half of F3; after the downsampling operation of Conv(3×3), F4 is input into the DCMRB module, and the ice melting effect features of F4 are extracted and refined to obtain the ice melting effect feature map F5, whose resolution is half of F4;

[0019] S24: After upsampling F5 through the EEAUP module, a concat operation is performed with F3. The concat feature map is input into the RBAMB module to obtain the ice melting effect feature map F6. The resolution of F6 is twice that of F5. The RBAMB module is composed of a DCMRB module and a BAMB module in series. After upsampling F6 through the EEAUP module, a concat operation is performed with F2. The concat feature map is input into the BAMB module. The result processed by the BAMB module is subjected to a Conv(3×3) convolution operation to obtain the ice melting effect feature map F7. The resolution of F7 is twice that of F6.

[0020] S25: After the ice melting effect feature maps F5, F6, and F7 are restored to the resolution of F1 through upsampling operations, they are spliced ​​in the channel dimension. The spliced ​​result is converted into a single channel after the Conv(3×3) convolution operation, and then processed through a linear layer to obtain the power grid line ice melting effect detection result.

[0021] As a preferred solution of the power line ice melting effect detection method based on deep learning of the present invention, in which: in the DCMRB module, the input feature map Input DCMRB The calculation is performed on two branches respectively. The specific steps are as follows:

[0022] In the first branch, the input feature map Input DCMRB After the Conv(1x1) convolution operation and GN normalization, the ice melting effect feature map D1 is obtained; after D1 is activated by LeakyReLU, the ice melting effect feature map D2 is obtained; D2 enters the first sub-branch and the second sub-branch for processing respectively:

[0023] In the first sub-branch, D2 is subjected to a deformable convolution DConv(3x3) operation and then normalized by GN to obtain the ice-melting effect feature map D3. D3 is activated by LeakyReLU and then subjected to a Conv(1x1) operation to obtain the ice-melting effect feature map D4.

[0024] In the second sub-branch, D2 is subjected to a Conv(1x1) convolution operation and then the channel attention is enhanced through SE Block to obtain the ice melting effect feature map D5;

[0025] In the second branch, the input feature map Input DCMRB After performing the Conv(1x1) convolution operation, the channel attention is enhanced through the SE Block to obtain the ice melting effect feature map D6;

[0026] After element-wise addition of D4, D5, and D6, activation is performed through LeakyReLU to obtain the output feature map of the DCMRB module. DCMRB Output DCMRB The calculation process can be described by formula (1)-(7);

[0027] D1=GN(Conv 1×1 (Input DCMRB )) (1)

[0028] D2=LeakyReLU(D1) (2)

[0029] D3=GN(Conv 3×3 (D2)) (3)

[0030] D4=Conv 1×1 (LeakyReLU(D3)) (4)

[0031] D5=SE(Conv 1×1 (D2)) (5)

[0032] D6=SE(Conv 1×1 (Input DCMRB )) (6)

[0033] Output DCMRB =LeakyReLU(Add(D4,D5,D6)) (7).

[0034] As a preferred solution of the power line ice melting effect detection method based on deep learning of the present invention, in which: in the MFEDSB module, the input feature map Input MFEDSB Calculations are performed on four branches respectively. The specific steps are as follows:

[0035] In the first branch, for Input MFEDSB No treatment is done;

[0036] In the second branch, set Input MFEDSB The Conv(1×1) convolution operation is used to reduce the number of channels and computational complexity, and the ice melting effect feature map M1 is obtained.

[0037] In the third branch, set Input MFEDSB The number of channels is reduced through the Conv(1×1) convolution operation, and then the depthwise separable convolution DSConv(3×3) operation is performed to extract the ice melting effect features and obtain the ice melting effect feature map M2;

[0038] In the fourth branch, set Input MFEDSBThe number of channels is reduced through the Conv(1×1) convolution operation, and then two consecutive DSConv(3×3) convolution operations are performed to extract deeper ice melting effect features, obtaining the ice melting effect feature map M3. The two DSConv(3×3) convolution operations have the same receptive field as the DSConv(5×5) convolution operation, but with fewer calculation parameters, which can reduce the number of network parameters.

[0039] Combine M1, M2, M3 and Input MFEDSB After the Concat operation, the DSConv (3×3) operation is used to extract the ice melting effect features to obtain the output feature map of the MFEDSB module. MFEDSB Output MFEDSB The calculation process can be described by equations (8)-(11);

[0040] M1=conv 1×1 (Input MFEDSB ) (8)

[0041] M2=DSConv 3×3 (conv 1×1 (M1)) (9)

[0042] M3=DSConv 3×3 (DSConv 3×3 (conv 1×1 (Input MFEDSB ))) (10)

[0043] Output MFESB =DSConv 3×3 Concat(M1,M2,M3,Input MFEDSB )) (11).

[0044] As a preferred solution of the power line ice melting effect detection method based on deep learning described in the present invention, in which: in the BAMB module, the input feature map Input BAMB Calculations are performed on four branches respectively. The specific steps are as follows:

[0045] In the first branch, for Input BAMB After performing DConv (3x3) operation and activating it through HCA activation function, we get ice melting effect feature map B1;

[0046] In the second branch, for Input BAMB No treatment is done;

[0047] In the third branch, the Input BAMB No processing is performed;

[0048] In the fourth branch, the Input BAMB After performing the GAP operation, the ice-melting effect feature map B3 is obtained; after B3 undergoes two Conv(3x3)(2dilated) operations and is activated by the HCA activation function, the ice-melting effect feature map B4 is obtained;

[0049] B1 and Input BAMB After matrix multiplication, we get the ice melting effect feature map B2; B4 and Input BAMB After matrix multiplication, the ice melting effect characteristic map B5 is obtained;

[0050] Perform element-wise addition of B4 and B5 to obtain the output feature map of the BAMB module Output BAMB Output BAMB The calculation process can be described by equations (12)-(17);

[0051] B1=HCA(DConv 3×3 (Input BAMB )) (12)

[0052]

[0053] B3=GAP(Input BAMB ) (14)

[0054] B4=HCA(DConv 3×3 (DConv 3×3 (B3))) (15)

[0055]

[0056] Output BAMB =Add(B4,B5) (17).

[0057] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the HCA activation function is defined as follows:

[0058] HCA(x)=α·Cosh(β·x+γ)+δ·log(1+∈·|x|) (18)

[0059] Among them, α is the amplitude of the activation function, which controls the maximum value of the output; β is the degree of influence of the input on the activation function, which controls the steepness of the function; γ is a translation parameter, which controls the translation effect of the input; δ adjusts the amplitude of the second term, which affects the response to the absolute value of the input; ∈ adjusts the weighting degree of the absolute value of the input, which adds a certain degree of smoothness;

[0060] The Cosh function is defined as follows:

[0061]

[0062] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the upsampling process of the EEAUP module is as follows:

[0063] First, given the input feature map X and the target upsampling scale s, bilinear interpolation or deconvolution is used to generate the initial upsampling result:

[0064] X up =U(X,s) (20)

[0065] Among them, U represents the conventional upsampling operation;

[0066] Then, for any pixel position (i, j) in the feature map, the gradient magnitude is calculated using the Sobel filter or gradient operator:

[0067]

[0068] After normalization, we get the edge attention weight:

[0069]

[0070] Here, ∈ is a small constant to prevent numerical instability;

[0071] Finally, the edge information is fused and the preliminary upsampling result is adaptively adjusted to obtain the final upsampling result:

[0072] X EEAUP (i,j)=X up (i,j)+αW edge (i,j)·X up (i,j) (23)

[0073] Among them, Δ is a hyperparameter used to control the strength of edge enhancement.

[0074] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the specific steps of S3 are as follows:

[0075] S31: Use the power line ice melting effect detection dataset in S1 to train the power line ice melting effect detection network GIMEDNet built in S2;

[0076] S32: During the training process, GIMEDNet optimizes the loss function by adjusting the weight values ​​in the network using the Adam algorithm, so that the loss function converges and the network reaches a balanced state.

[0077] As a preferred solution of the method for detecting ice melting effect of power grid lines based on deep learning according to the present invention, the specific steps of S4 are as follows:

[0078] S41: After GIMEDNet training is completed, GIMEDNet is used to detect the de-icing effect of power grid lines;

[0079] S42: In actual applications, the collected power grid line ice melting result image is input, and GIMEDNet processes the ice melting result image and outputs the ice melting effect detection result.

[0080] Compared with existing technologies:

[0081] The GIMEDNet of the present invention can significantly improve the quality and efficiency of ice-melting effect detection through a bidirectional attention mechanism, ensuring high-quality detection of the ice-melting effect of power grid lines. GIMEDNet also has the ability to continuously learn and self-optimize, and can continuously improve the detection accuracy of the ice-melting effect of power grid lines through continuous training and iteration. Based on this, the present invention has a faster detection speed than complex methods and a higher detection quality than traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0083] Figure 2 This is a schematic diagram of the GIMEDNet network structure of the present invention;

[0084] Figure 3 This is a schematic diagram of the RMBAMB module structure of the present invention;

[0085] Figure 4 This is a schematic diagram of the DCMRB module structure of the present invention;

[0086] Figure 5 This is a schematic diagram of the MFEDSB module structure of the present invention;

[0087] Figure 6 Schematic diagram of the BAMB module structure of the present invention. DETAILED DESCRIPTION

[0088] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0089] The present invention provides a method for detecting the ice melting effect of power grid lines based on deep learning. Figures 1-6 , including the following specific steps:

[0090] S1: Construct a dataset for detecting the de-icing effect of power grid lines.

[0091] The specific steps of S1 are as follows:

[0092] S11: Collect image data of power grid line ice melting results (taken on site or using existing public datasets);

[0093] S12: The collected image data is annotated using annotation tool software to annotate the ice-covered areas in the image pixel by pixel. After the annotation is completed, a de-icing result mask map is obtained. The power grid line de-icing effect detection dataset consists of the original power grid line de-icing result image and its corresponding mask map.

[0094] There are many labeling tool softwares. In the present invention, LabelMe is used as the labeling tool software.

[0095] S2: Build a power grid line ice melting effect detection network GIMEDNet.

[0096] Build a grid line ice melting effect detection network GIMEDNet (GridIce Melting EffectDetection Net) for power line ice melting effect detection. The GIMEDNet network structure is as follows Figure 2 As shown, the ⊕ in the figure is the Concat operation.

[0097] Concat represents the channel-wise concatenation operation in deep learning. When concatenating in the channel dimension, the feature maps must have the same length and width, but the number of channels can be different. For example, if there are two feature maps A and B with shapes of H×W×16 and H×W×32, respectively, they can be concatenated in the channel dimension to obtain a new feature map with a shape of H×W×48.

[0098] The specific steps of S2 are as follows:

[0099] S21: The power line ice melting result image F1 is input into GIMEDNet and first processed by the Conv(3×3)_BN_ReLU module to obtain the ice melting effect feature map F2. The Conv(3×3)_BN_ReLU module is composed of a Conv layer with a convolution kernel size of 3×3 and a stride of 1, a BN layer, and a ReLU layer in series.

[0100] The convolution kernel size of 3×3 mentioned above means that the length and width of the convolution kernel are both 3.

[0101] S22: After the downsampling operation of Conv (7×7), F2 is input into the RMBAMB (Residual Multi-scale Bilateral Attention Mechanism Block) module to obtain the ice melting effect feature map F3. The resolution of F3 is half of F2. The RMBAMB module structure is as follows: Figure 3 As shown;

[0102] The present invention designs a DCMRB (Deformable Conv Multi-dimension Residual Block) module, which uses deformable convolution (DConv) for feature extraction and spatial adaptive transformation to adapt to the ice melting effect characteristics of different deformations.

[0103] The advantage of this module is that it can not only significantly expand the receptive field without reducing the resolution, but also adapt to the complex ice-melting effect structure through deformable convolution. Combined with the global attention mechanism of SEBlock, it can enhance the network's attention to key features, thereby improving the recognition ability of ice-melting effect features. The structure of the DCMRB module is as follows: Figure 4 shown.

[0104] In the DCMRB module, the input feature map Input DCMRB Calculations are performed on two branches, such as Figure 4 The specific steps are as follows:

[0105] In the first branch, the input feature map Input DCMRB After the Conv(1x1) convolution operation, it is normalized by GN (GroupNormalization) to obtain the ice-melting effect feature map D1; after D1 is activated by LeakyReLU, the ice-melting effect feature map D2 is obtained; D2 enters the first sub-branch and the second sub-branch for processing respectively:

[0106] In the first sub-branch, D2 is subjected to a deformable convolution DConv(3x3) operation and then normalized by GN to obtain the ice-melting effect feature map D3. D3 is activated by LeakyReLU and then subjected to a Conv(1x1) operation to obtain the ice-melting effect feature map D4.

[0107] In the second sub-branch, D2 is subjected to a Conv(1x1) convolution operation and then the channel attention is enhanced through SE Block to obtain the ice melting effect feature map D5;

[0108] In the second branch, the input feature map Input DCMRBAfter performing the Conv(1x1) convolution operation, the channel attention is enhanced through the SE Block to obtain the ice melting effect feature map D6;

[0109] After adding D4, D5 and D6 element by element, we activate them through LeakyReLU to get the output feature map of DCMRB module. DCMRB Output DCMRB The calculation process can be described by formula (1)-(7);

[0110] D1=GN(Conv 1×1 (Input DCMRB )) (1)

[0111] D2=LeakyReLU(D1) (2)

[0112] D3=GN(Conv 3×3 (D2)) (3)

[0113] D4=Conv 1×1 (LeakyReLU(D3)) (4)

[0114] D5=SE(Conv 1×1 (D2)) (5)

[0115] D6=SE(Conv 1×1 (Input DCMRB )) (6)

[0116] Output DCMRB =LeakyReLU(Add(D4,D5,D6)) (7).

[0117] This paper designs a MFEDSB (Multi-scale Feature Extraction-DepthSeparation Block) module to improve the performance of GIMEDNet. The module uses depthwise separable convolution (DSConv), which can improve the ice melting effect feature extraction capability while reducing computational complexity, reduce redundant computations, improve the efficiency and generalization ability of the model, and enable the network to more fully capture and integrate key features of different scales. The MFEDSB module structure is as follows: Figure 5 shown.

[0118] In the MFEDSB module, the input feature map Input MFEDSB Calculations are performed on four branches, such as Figure 5 The specific steps are as follows:

[0119] In the first branch, for Input MFEDSB No treatment is done;

[0120] In the second branch, set Input MFEDSB The Conv(1×1) convolution operation is used to reduce the number of channels and computational complexity, and the ice melting effect feature map M1 is obtained.

[0121] In the third branch, set Input MFEDSB The number of channels is reduced through the Conv(1×1) convolution operation, and then the depthwise separable convolution DSConv(3×3) operation is performed to extract the ice melting effect features and obtain the ice melting effect feature map M2;

[0122] In the fourth branch, set Input MFEDSB The number of channels is reduced through the Conv(1×1) convolution operation, and then two consecutive DSConv(3×3) convolution operations are performed to extract deeper ice melting effect features, obtaining the ice melting effect feature map M3. The two DSConv(3×3) convolution operations have the same receptive field as the DSConv(5×5) convolution operation, but with fewer calculation parameters, which can reduce the number of network parameters.

[0123] Combine M1, M2, M3 and Input MFEDSB After the Concat operation, the DSConv (3×3) operation is used to extract the ice melting effect features to obtain the output feature map of the MFEDSB module. MFEDSB Output MFEDSB The calculation process can be described by equations (8)-(11);

[0124] M1=conv 1×1 (Input MFEDSB ) (8)

[0125] M2=DSConv 3×3 (conv 1×1 (M1)) (9)

[0126] M3=DSConv 3×3 (DSConv 3×3 (conv 1×1 (Input MFEDSB ))) (10)

[0127] Output MFEDSB =DSConv 3×3 Concat(M1,M2,M3,Input MFEDSB )) (11).

[0128] The present invention designs a BAMB (Bilateral Attention Mechanism Block) module to implement a bidirectional attention mechanism. The module combines dilated convolution and HCA (HyperLog-Cosh Activation) activation function to effectively improve the ability to extract ice-melting effect features. Dilated convolution expands the receptive field and captures multi-scale features while maintaining computational efficiency. The HCA activation function optimizes training stability and accelerates convergence. The bidirectional attention mechanism enables the network to focus more on key features, further improving the detection accuracy of ice-melting effects. The structure of the BAMB module is as follows: Figure 6 shown. Figure 6 middle is matrix multiplication.

[0129] Among them, dilated convolution, also known as expanded convolution, is a variant that introduces dilation into the standard convolution operation. It inserts holes between convolution kernel elements to increase the receptive field without increasing computational complexity or reducing resolution.

[0130] For dilated convolution, the distance between convolution kernels is determined by a dilation rate.

[0131] Assume that the input feature map is I, the convolution kernel is K, the convolution kernel size is k×k, and the input feature map size is H in ×W in , the hole rate is d, and the output feature map size is H out ×W out , then the calculation formula of the dilated convolution is as follows:

[0132]

[0133] In the BAMB module, the input feature map Input BAMB Calculations are performed on four branches, such as Figure 6 The specific steps are as follows:

[0134] In the first branch, for Input BAMB After performing DConv (3x3) operation and activating it through HCA activation function, we get ice melting effect feature map B1;

[0135] In the second branch, for Input BAMB No treatment is done;

[0136] In the third branch, the Input BAMB No processing is performed;

[0137] In the fourth branch, the Input BAMB After performing the GAP (global average pooling) operation, the ice-melting effect feature map B3 is obtained. After B3 undergoes two Conv(3x3)(2dilated) operations and is activated by the HCA activation function, the ice-melting effect feature map B4 is obtained.

[0138] B1 and Input BAMB After matrix multiplication, we get the ice melting effect feature map B2; B4 and Input BAMB After matrix multiplication, the ice melting effect characteristic map B5 is obtained;

[0139] Perform element-wise addition of B4 and B5 to obtain the output feature map of the BAMB module. BAMB Output BAMB The calculation process can be described by equations (12)-(17);

[0140] B1=GCA(DConv 3×3 (Input BAMB )) (12)

[0141]

[0142] B3=GAP(Input BAMB ) (14)

[0143] B4=HCA(DConv 3×3 (DConv 3×3 (B3))) (15)

[0144]

[0145] Output BAMB =Add(B4,B5) (17).

[0146] The present invention designs a hyperlogarithmic hyperbolic cosine activation function (HyperLog-Cosh Activation, HCA), which combines an exponential function and a parameterized adjustment factor to make the activation function more flexible and complex.

[0147] The HCA activation function is defined as follows:

[0148] HCA(x)=α·Cosh(β·x+γ)+δ·log(1+∈·|x|) (18)

[0149] Among them, α is the amplitude of the activation function, which controls the maximum value of the output; β is the degree of influence of the input on the activation function, which controls the steepness of the function; γ is a translation parameter, which controls the translation effect of the input; δ adjusts the amplitude of the second term, which affects the response to the absolute value of the input; ∈ adjusts the weighting degree of the absolute value of the input, which adds a certain degree of smoothness;

[0150] The Cosh function is defined as follows:

[0151]

[0152] S23: After the downsampling operation of Conv(3×3), F3 is input into the RMBAMB module, and the ice melting effect features of F3 are extracted and refined to obtain the ice melting effect feature map F4, whose resolution is half of F3; after the downsampling operation of Conv(3×3), F4 is input into the DCMRB module, and the ice melting effect features of F4 are extracted and refined to obtain the ice melting effect feature map F5, whose resolution is half of F4;

[0153] S24: After upsampling F5 through the EEAUP module, a concat operation is performed with F3. The concat feature map is input into the RBAMB module to obtain the ice melting effect feature map F6. The resolution of F6 is twice that of F5. The RBAMB module is composed of a DCMRB module and a BAMB module in series. After upsampling F6 through the EEAUP module, a concat operation is performed with F2. The concat feature map is input into the BAMB module. The result processed by the BAMB module is subjected to a Conv(3×3) convolution operation to obtain the ice melting effect feature map F7. The resolution of F7 is twice that of F6.

[0154] The present invention designs an EEAUP upsampling module that combines edge perception and adaptive adjustment (see Figure 2 ), which aims to improve the ability to retain feature details during the upsampling process. This method first performs conventional upsampling on the input feature map (such as bilinear interpolation or deconvolution), then calculates the edge response using gradient information to generate edge enhancement weights. This weight is used to adaptively adjust the upsampled features, resulting in additional enhancement of edge regions, effectively reducing blur and improving detail clarity. Compared to traditional upsampling methods, EEAUP more accurately restores edge details while maintaining overall structural information, making it better suited for ice melting effect detection tasks.

[0155] The upsampling process of the EEAUP module is as follows:

[0156] First, given the input feature map X and the target upsampling scale s, bilinear interpolation or deconvolution is used to generate the initial upsampling result:

[0157] X up =U(X,s) (20)

[0158] Among them, U represents the conventional upsampling operation;

[0159] Then, for any pixel position (i, j) in the feature map, the gradient magnitude is calculated using the Sobel filter or gradient operator:

[0160]

[0161] After normalization, we get the edge attention weight:

[0162]

[0163] Here, ∈ is a small constant to prevent numerical instability;

[0164] Finally, the edge information is fused and the preliminary upsampling result is adaptively adjusted to obtain the final upsampling result:

[0165] X EEAUP (i,j)=X up (i,j)+αW edge (i,j)·X up (i,j) (23)

[0166] Among them, α is a hyperparameter used to control the strength of edge enhancement.

[0167] S25: After the de-icing effect feature maps F5, F6, and F7 are restored to the resolution of F1 through upsampling operations, they are concatenated (Concat) in the channel dimension. The concatenated result is converted into a single channel after a Conv(3×3) convolution operation, and then processed through a linear layer to obtain the power grid line de-icing effect detection result.

[0168] The detection result of the ice melting effect of the power grid line is a numerical value. The larger the output value is, the better the ice melting effect is.

[0169] S3: Training the power grid line ice melting effect detection network GIMEDNet.

[0170] The specific steps of S3 are as follows:

[0171] S31: Use the power line de-icing effect detection dataset in S1 to train the power line de-icing effect detection network GIMEDNet built in S2; initialize the hyperparameters required for training the GIMEDNet network, such as the batch size, training rounds, and initial learning rate;

[0172] S32: During the training process, GIMEDNet optimizes the loss function by adjusting the weight values ​​in the network using the Adam algorithm, so that the loss function converges and the network reaches a balanced state.

[0173] Among them, the Adam optimization algorithm is a common optimization algorithm in deep learning.

[0174] S4: Apply GIMEDNet to detect the de-icing effect of power grid lines.

[0175] The specific steps of S4 are as follows:

[0176] S41: After GIMEDNet training is completed, GIMEDNet is used to detect the de-icing effect of power grid lines;

[0177] S42: In actual applications, the collected power grid line ice melting result image is input, and GIMEDNet processes the ice melting result image and outputs the ice melting effect detection result.

[0178] Example: Apply GIMEDNet to detect the de-icing effect of power grid lines.

[0179] Assume that the input power line de-icing result image F1 has a dimension of 1×512×512 (channels × width × height). F1 is input into GIMEDNet and first processed by the Conv(3×3)_BN_ReLU module to obtain a de-icing effect feature map F2 with a dimension of 16×512×512. F2 is downsampled by the Conv(7×7) operation and input into the RMBAMB module to obtain a de-icing effect feature map F3 with a dimension of 32×256×256. F3 is downsampled by the Conv(3×3) operation and input into the RMBAMB module to extract and refine the de-icing effect features of F3, resulting in a de-icing effect feature map F4 with a dimension of 48×128×128. After downsampling F4 using the Conv(3×3) operation, it is fed into the DCMRB module, where its ice-melting effect features are extracted and refined, resulting in a 64×64×64 ice-melting effect feature map F5. F5 is upsampled using the EEAUP module and concatenated with F3. This concatenated feature map is fed into the RBAMB module, resulting in a 48×256×256 ice-melting effect feature map F6. F6 is upsampled using the EEAUP module and concatenated with F2. This concatenated feature map is fed into the BAMB module, where the result of the BAMB processing is subjected to a Conv(3×3) convolution operation, resulting in a 48×512×512 ice-melting effect feature map F7. After upsampling the ice-melting effect feature maps F5, F6, and F7 to a resolution of 512×512, they are concatenated in the channel dimension. The resulting feature maps, with dimensions of (64+48+48)×512×512, undergo a Conv(3×3) convolution operation, converting them to a single channel (1×512×512). They are then processed through a linear layer with a weight matrix of 512×512 to produce the final power line ice-melting effect detection output. The power line ice-melting effect detection result is a numerical value ranging from 0 to 100, with larger values ​​indicating better ice-melting effect.

[0180] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space 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 method for detecting ice melting effects of power grid lines based on deep learning, characterized in that: The specific steps are as follows: S1: Construct a dataset for detecting the de-icing effect of power grid lines; S2: Build the power grid line ice melting effect detection network GIMEDNet; S3: Training the power line ice melting effect detection network GIMEDNet; S4: Apply GIMEDNet to detect the de-icing effect of power grid lines.

2. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 1, characterized in that: The specific steps of S1 are as follows: S11: collecting image data of power grid line ice melting results; S12: The collected image data is annotated using annotation tool software to annotate the ice-covered areas in the image pixel by pixel. After the annotation is completed, a de-icing result mask map is obtained. The power grid line de-icing effect detection dataset consists of the original power grid line de-icing result image and its corresponding mask map.

3. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 1, characterized in that: The specific steps of S2 are as follows: S21: The power line ice melting result image F1 is input into GIMEDNet and first processed by the Conv(3×3)_BN_ReLU module to obtain the ice melting effect feature map F2. The Conv(3×3)_BN_ReLU module is composed of a Conv layer with a convolution kernel size of 3×3 and a stride of 1, a BN layer, and a ReLU layer in series. S22: After F2 is downsampled by Conv(7×7), it is input into the RMBAMB module to obtain the ice melting effect feature map F3, and the resolution of F3 is half of F2; the RMBAMB module includes the DCMRB module, the MFEDSB module and the BAMB module; the DCMRB module uses deformable convolution for feature extraction and spatial adaptive transformation to adapt to the ice melting effect features of different deformations; the MFEDSB module uses depthwise separable convolution, which can improve the ice melting effect feature extraction capability while reducing computational complexity, reduce redundant calculations, improve the efficiency and generalization ability of the model, and enable the network to more fully capture and integrate key features of different scales; the BAMB module is used to implement a bidirectional attention mechanism, and the BAMB module combines the void convolution and HCA activation function, effectively improving the ice melting effect feature extraction capability; S23: After the downsampling operation of Conv(3×3), F3 is input into the RMBAMB module, and the ice melting effect features of F3 are extracted and refined to obtain the ice melting effect feature map F4, whose resolution is half of F3; after the downsampling operation of Conv(3×3), F4 is input into the DCMRB module, and the ice melting effect features of F4 are extracted and refined to obtain the ice melting effect feature map F5, whose resolution is half of F4; S24: After upsampling F5 through the EEAUP module, a concat operation is performed with F3. The concat feature map is input into the RBAMB module to obtain the ice melting effect feature map F6. The resolution of F6 is twice that of F5. The RBAMB module is composed of a DCMRB module and a BAMB module in series. After upsampling F6 through the EEAUP module, a concat operation is performed with F2. The concat feature map is input into the BAMB module. The result processed by the BAMB module is subjected to a Conv(3×3) convolution operation to obtain the ice melting effect feature map F7. The resolution of F7 is twice that of F6. S25: After the ice melting effect feature maps F5, F6, and F7 are restored to the resolution of F1 through upsampling operations, they are spliced ​​in the channel dimension. The spliced ​​result is converted into a single channel after the Conv(3×3) convolution operation, and then processed through a linear layer to obtain the power grid line ice melting effect detection result.

4. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 3, characterized in that: In the DCMRB module, the input feature map Input DCMRB The calculation is performed on two branches respectively. The specific steps are as follows: In the first branch, the input feature map Input DCMRB After the Conv(1x1) convolution operation and GN normalization, the ice melting effect feature map D1 is obtained; after D1 is activated by LeakyReLU, the ice melting effect feature map D2 is obtained; D2 enters the first sub-branch and the second sub-branch for processing respectively: In the first sub-branch, D2 is subjected to a deformable convolution DConv(3x3) operation and then normalized by GN to obtain the ice-melting effect feature map D3. D3 is activated by LeakyReLU and then subjected to a Conv(1x1) operation to obtain the ice-melting effect feature map D4. In the second sub-branch, D2 is subjected to a Conv(1x1) convolution operation and then the channel attention is enhanced through SE Block to obtain the ice melting effect feature map D5; In the second branch, the input feature map Input DCMRB After performing the Conv(1x1) convolution operation, the channel attention is enhanced through the SE Block to obtain the ice melting effect feature map D6; After element-wise addition of D4, D5, and D6, activation is performed through LeakyReLU to obtain the output feature map of the DCMRB module. DCMRB Output DCMRB The calculation process can be described by formula (1)-(7); D1=GN(Conv 1×1 (Input DCMRB )) (1) D2=LeakyReLU(D1) (2) D3=GN(Conv 3×3 (D2)) (3) D4=Conv 1×1 (LeakyReLU(D3)) (4) D5=SE(Conv 1×1 (D2)) (5) D6=SE(Conv 1×1 (Input DCMRB )) (6) Output DCMRB =LeakyReLU(Add(D4,D5,D6)) (7)。 5. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 3, characterized in that: In the MFEDSB module, the input feature map Input MFEDSB Calculations are performed on four branches respectively. The specific steps are as follows: In the first branch, for Input MFEDSB No treatment is done; In the second branch, set Input MFEDSB The Conv(1×1) convolution operation is used to reduce the number of channels and computational complexity, and the ice melting effect feature map M1 is obtained. In the third branch, set Input MFEDSB The number of channels is reduced through the Conv(1×1) convolution operation, and then the depthwise separable convolution DSConv(3×3) operation is performed to extract the ice melting effect features and obtain the ice melting effect feature map M2; In the fourth branch, set Input MFEDSB The number of channels is reduced through the Conv(1×1) convolution operation, and then two consecutive DSConv(3×3) convolution operations are performed to extract deeper ice melting effect features, obtaining the ice melting effect feature map M3. The two DSConv(3×3) convolution operations have the same receptive field as the DSConv(5×5) convolution operation, but with fewer calculation parameters, which can reduce the number of network parameters. Combine M1, M2, M3 and Input MFEDSB After the Concat operation, the DSConv (3×3) operation is used to extract the ice melting effect features to obtain the output feature map of the MFEDSB module. MFEDSB Output MFEDSB The calculation process can be described by equations (8)-(11); M1=conv 1×1 (Inpvt MFEDSB ) (8) M2=DSConv 3×3 (conv 1×1 (M1)) (9) M3=DSConv 3×3 (DSConv 3×3 (conv 1×1 (Input MFEDSB ))) (10) Output MFEDSB =DSConv 3×3 Concat(M1,M2,M3,Intput MFEDSB )) (11)。 6. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 3, characterized in that: In the BAMB module, the input feature map Input BAMB Calculations are performed on four branches respectively. The specific steps are as follows: In the first branch, for Input BAMB After performing DConv (3x3) operation and activating it through HCA activation function, we get ice melting effect feature map B1; In the second branch, for Input BAMB No treatment is done; In the third branch, the Input BAMB No processing is performed; In the fourth branch, the Input BAMB After performing the GAP operation, the ice-melting effect feature map B3 is obtained; after B3 undergoes two Conv(3x3)(2dilated) operations and is activated by the HCA activation function, the ice-melting effect feature map B4 is obtained; B1 and Input BAMB After matrix multiplication, we get the ice melting effect feature map B2; B4 and Input BAMB After matrix multiplication, the ice melting effect characteristic map B5 is obtained; Perform element-wise addition of B4 and B5 to obtain the output feature map of the BAMB module Output BAMB Output BAMB The calculation process can be described by equations (12)-(17); B1=HCA(DConv 3×3 (Input BAMB )) (12) B3=GAP(Input BAMB ) (14) B4=HCA(DConv 3×3 (DConv 3×3 (B3))) (15) Output BAMB =Add(B4,B5) (17)。 7. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 6, characterized in that: The HCA activation function is defined as follows: HCA(x)=α·Cosh(β·x+γ)+δ·log(1+∈·|x|) (18) Among them, α is the amplitude of the activation function, which controls the maximum value of the output; β is the degree of influence of the input on the activation function, which controls the steepness of the function; γ is a translation parameter, which controls the translation effect of the input; δ adjusts the amplitude of the second term, which affects the response to the absolute value of the input; ∈ adjusts the weighting degree of the absolute value of the input, which adds a certain degree of smoothness; The Cosh function is defined as follows:

8. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 3, characterized in that: The upsampling process of the EEAUP module is as follows: First, given the input feature map X and the target upsampling scale s, bilinear interpolation or deconvolution is used to generate the initial upsampling result: X up =U(X,s) (20) Among them, U represents the conventional upsampling operation; Then, for any pixel position (i, j) in the feature map, the gradient magnitude is calculated using the Sobel filter or gradient operator: After normalization, we get the edge attention weight: Here, ∈ is a small constant to prevent numerical instability; Finally, the edge information is fused and the preliminary upsampling result is adaptively adjusted to obtain the final upsampling result: X EEAUP (i,j)=X up (i,j)+αW edge (i,j)·X up (i,j) (23) Among them, α is a hyperparameter used to control the strength of edge enhancement.

9. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 1, characterized in that: The specific steps of S3 are as follows: S31: Use the power line ice melting effect detection dataset in S1 to train the power line ice melting effect detection network GIMEDNet built in S2; S32: During the training process, GIMEDNet optimizes the loss function by adjusting the weight values ​​in the network using the Adam algorithm, so that the loss function converges and the network reaches a balanced state.

10. The method for detecting ice melting effect of power grid lines based on deep learning according to claim 1, characterized in that: The specific steps of S4 are as follows: S41: After GIMEDNet training is completed, GIMEDNet is used to detect the de-icing effect of power grid lines; S42: In actual applications, the collected power grid line ice melting result image is input, and GIMEDNet processes the ice melting result image and outputs the ice melting effect detection result.

Images