Method and system for identifying power poles and towers applicable to drones

Abstract

This invention provides a method and system for identifying power poles using unmanned aerial vehicles (UAVs), belonging to the field of power pole inspection technology. The identification method includes: acquiring an image of the power pole to be identified; preprocessing the power pole image to obtain a standard image; and inputting the standard image into a preset image recognition network to determine the location of the power pole. This identification method and system identify power poles by combining images captured by a UAV, enabling the UAV to independently complete power pole inspection work. Compared to existing technologies, the identification method and system provided by this invention have faster identification efficiency.

Description

Technical Field

[0001] This invention relates to the field of power pole inspection technology, and more specifically to a power pole identification method and system applicable to unmanned aerial vehicles (UAVs). Background Technology

[0002] Power poles, as an indispensable component of power transmission and distribution systems, bear the crucial responsibility of transmitting and distributing electrical energy to end users. These towering structures are frequently exposed to harsh natural environments, including strong winds, snow, earthquakes, and other potentially destructive conditions. Therefore, ensuring the stability and reliability of power poles is of paramount importance to safeguarding the safety of residents' electricity use.

[0003] Traditional power pole inspection work mainly relies on manual labor. Inspectors need to regularly check the physical condition of the poles, including but not limited to checking the structural integrity of the pole itself, the condition of the insulators, the tension of the conductors, and whether there are any potential hazards in the surrounding environment that could threaten the safety of the pole.

[0004] With the development of drone technology, the technology of using drones for facility inspection has also emerged. However, since power poles are mostly located in complex environments, conventional visual recognition technology is difficult to meet the inspection needs of drones. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying power poles suitable for drones, which can control and ensure that drones can independently complete inspection tasks.

[0006] To achieve the above objectives, embodiments of the present invention provide a method for identifying power poles suitable for unmanned aerial vehicles (UAVs), comprising:

[0007] Acquire images of the power poles to be identified;

[0008] The images of the power poles are preprocessed to obtain standard images;

[0009] The standard image is input into a preset image recognition network to determine the location of the power pole.

[0010] Optionally, the power pole images are preprocessed to obtain standard images, including:

[0011] A high-pass filter is used to perform spatial filtering on the power pole image;

[0012] Perform histogram equalization on the power pole image after spatial filtering;

[0013] The power pole image after histogram equalization is then denoised.

[0014] Optionally, the image recognition network includes:

[0015] A residual convolutional unit is used to read the image of the power pole to be identified;

[0016] The first convolutional unit, with its first end connected to the residual convolutional unit;

[0017] The second convolutional unit, wherein the first end of the second convolutional unit is connected to the second end of the first convolutional unit;

[0018] The third convolutional unit, wherein the first end of the third convolutional unit is connected to the second end of the second convolutional unit;

[0019] The fourth convolutional unit, wherein the first end of the fourth convolutional unit is connected to the second end of the third convolutional unit;

[0020] The fifth convolutional unit, wherein the first end of the fifth convolutional unit is connected to the second end of the fourth convolutional unit;

[0021] The sixth convolutional unit, wherein the first end of the sixth convolutional unit is connected to the second end of the fifth convolutional unit;

[0022] The first integration unit, wherein the first end of the first integration unit is connected to the third end of the first convolution unit;

[0023] A first classification unit, wherein a first end of the first classification unit is connected to a second end of the first integration unit;

[0024] A first upsampling and convolution unit, one end of which is connected to the third end of the third convolution unit, and the other end of which is connected to the third end of the first integration unit;

[0025] The second integration unit, the first end of which is connected to the third end of the fourth convolution unit;

[0026] The second classification unit, wherein the first end of the second classification unit is connected to the second end of the second integration unit;

[0027] The second upsampling and convolution unit has one end connected to the second end of the sixth convolution unit, and the second end of the second upsampling and convolution unit is connected to the third end of the second integration unit.

[0028] The third classification unit is connected to the third end of the sixth convolutional unit;

[0029] The third upsampling and convolution unit, wherein the first end of the third upsampling and convolution unit is connected to the third end of the fifth convolution unit;

[0030] The third integration unit has a first end connected to the second convolution unit and a second end connected to the other end of the third upsampling and convolution unit.

[0031] The fourth classification unit, one end of which is connected to the third end of the third integration unit;

[0032] The DS unit, connected to the first, second, third, and fourth classification units, is used to determine the final classification result based on the classification results of the first, second, third, and fourth classification units using evidence theory.

[0033] Optionally, at least one of the first convolutional unit, the second convolutional unit, the third convolutional unit, the fourth convolutional unit, the fifth convolutional unit, and the sixth convolutional unit includes:

[0034] The first residual convolutional layer, one end of which is used to receive the input of the previous network layer;

[0035] A two-dimensional convolutional layer, one end of which is connected to the other end of the first residual convolutional layer;

[0036] The second residual convolutional layer has one end connected to the other end of the two-dimensional convolutional layer and one end of the first residual convolutional layer.

[0037] Optionally, at least one of the first classification unit, the second classification unit, the third classification unit, and the fourth classification unit includes:

[0038] First convolutional layer;

[0039] A first pooling layer, one end of which is connected to the other end of the first convolutional layer;

[0040] The first multiplication module has a first end connected to one end of the first convolutional layer and a second end connected to the other end of the first pooling layer.

[0041] A global average pooling layer, one end of which is connected to the third end of the first multiplication module;

[0042] The second convolutional layer, one end of which is connected to the other end of the global average pooling layer;

[0043] A mapping layer, one end of which is connected to the other end of the second convolutional layer;

[0044] The second multiplication module has a first end connected to the other end of the mapping layer, and a second end connected to the fourth end of the first multiplication module.

[0045] The third convolutional layer, one end of which is connected to the third end of the second multiplication module;

[0046] A fully connected layer, one end of which is connected to the other end of the third convolutional layer, and the other end of which is used to output the classification result.

[0047] Optionally, the loss function for training the image recognition network is Equation (1) and Equation (2).

[0048]

[0049] VFL=-β(1-p i ) 2 log(p j )-(1-β)p i 2 log(1-p j (2)

[0050] Where Y is the loss function, λ1, λ2, and λ3 are the adaptive weights, n is the number of training samples, and y i For the numerical expression of the standard value, y i Numerical representation of the predicted value. Here, b is the graphical representation of the predicted value, α is the graphical representation of the standard value, β is the hyperparameter, and p is the adjustment factor. i p j These are the probabilities of the i-th and j-th categories, respectively.

[0051] On the other hand, the present invention also provides a power pole identification system suitable for drones, the identification system including a processor configured to perform the identification method as described above.

[0052] In another aspect, the present invention also provides a computer-readable storage medium storing instructions for being read by a machine to cause the machine to perform any of the identification methods described above.

[0053] Through the above technical solution, the embodiments of the present invention provide a method and system for identifying power poles using drones. This method and system identify power poles by combining images captured by the drone, enabling the drone to independently complete the inspection of power poles. Compared with the prior art, the identification method and system provided by the present invention have faster identification efficiency.

[0054] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a flowchart of a method for identifying power poles applicable to unmanned aerial vehicles (UAVs) according to an embodiment of the present invention;

[0057] Figure 2 This is an example image of a power pole according to one embodiment of the present invention;

[0058] Figure 3 This is a flowchart of a preprocessing operation according to an embodiment of the present invention;

[0059] Figure 4 This is a block diagram of an image recognition network according to an embodiment of the present invention;

[0060] Figure 5 This is a block diagram of a first convolutional unit, a second convolutional unit, a third convolutional unit, a fourth convolutional unit, a fifth convolutional unit, and a sixth convolutional unit according to an embodiment of the present invention;

[0061] Figure 6 This is a block diagram of a first classification unit, a second classification unit, a third classification unit, and a fourth classification unit according to an embodiment of the present invention. Detailed Implementation

[0062] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0063] like Figure 1 The diagram shows a flowchart of a method for identifying power poles for unmanned aerial vehicles (UAVs) according to an embodiment of the present invention. Figure 1 In this context, the identification method may include the following steps:

[0064] In step S10, an image of the power pole to be identified is acquired;

[0065] In step S11, the power pole images are preprocessed to obtain standard images;

[0066] In step S12, a standard image is input into a preset image recognition network to determine the location of the power pole.

[0067] In such Figure 1 In the identification method shown, step S10 can be used to acquire an image of the power pole to be identified. This power pole image can be, for example... Figure 2 As shown, this was captured in real-time by a drone.

[0068] Step S11 can be used to preprocess the images of power poles to obtain standard images. The specific method for this preprocessing operation can be of various forms known to those skilled in the art. In one example of the present invention, considering that drone inspections are generally conducted on sunny days with high light intensity, resulting in brighter images, and given the specific outdoor environment, the depth of field in the images is relatively complex, the preprocessing operation can include, for example... Figure 3 The method shown. In this Figure 3 In this process, the preprocessing operation may include the following steps:

[0069] In step S20, a high-pass filter is used to perform spatial filtering on the power pole image, thereby improving the edge features of the power pole image;

[0070] In step S21, histogram equalization is performed on the power pole image after spatial filtering to reduce the grayscale range of the power pole image.

[0071] In step S22, a denoising operation is performed on the power pole image after histogram equalization to reduce the noise effect of the power pole image.

[0072] Step S12 can be used to input a standard image into a preset image recognition network to determine the location of the power pole. While the specific structure of this image recognition network can take many forms known to those skilled in the art, considering the characteristics of power pole images described above, in one example of the present invention, the image recognition network can be as follows: Figure 4 As shown. In this Figure 4The image recognition network may include residual convolution unit 01, first convolution unit 02, second convolution unit 03, third convolution unit 04, fourth convolution unit 05, fifth convolution unit 06, sixth convolution unit 07, first upsampling and convolution unit 08, first integration unit 09, first classification unit 10, second upsampling and convolution unit 11, second integration unit 12, second classification unit 13, third classification unit 14, third upsampling and convolution unit 15, third integration unit 16, fourth classification unit 17, and a judgment unit (not shown in the figure).

[0073] The residual convolutional unit 01 can be used to read images of power poles to be identified. The first end of the first convolutional unit 02 can be connected to the residual convolutional unit 01 to perform feature extraction on the feature map output by the residual convolutional unit 01. The first end of the second convolutional unit 03 can be connected to the second end of the first convolutional unit 02 to perform feature extraction on the feature map output by the first convolutional unit 02. The first end of the third convolutional unit 04 can be connected to the second end of the second convolutional unit 03 to perform feature extraction on the feature map output by the second convolutional unit 03. The first end of the fourth convolutional unit 05 can be connected to the second end of the third convolutional unit 04 to perform feature extraction on the feature map output by the third convolutional unit 04. The first end of the fifth convolutional unit 06 can be connected to the second end of the fourth convolutional unit 05 to perform feature extraction on the feature map output by the fourth convolutional unit 05. The first end of the sixth convolutional unit 07 can be connected to the second end of the fifth convolutional unit 06 to perform feature extraction on the feature map output by the fifth convolutional unit 06. The first end of the first integration unit 09 can be connected to the third end of the first convolutional unit 02. The first end of the first classification unit 10 can be connected to the second end of the first integration unit 09. One end of the first upsampling and convolutional unit 08 can be connected to the third end of the third convolutional unit 04, and the other end of the first upsampling and convolutional unit 08 can be connected to the third end of the first integration unit 09. The first upsampling and convolutional unit 08 can be used to sequentially perform upsampling and convolution operations on the feature map output by the third convolutional unit 04, thereby ensuring that the output feature map maintains the same dimension as the feature map output by the first convolutional unit 02. The first integration unit 09 (concat unit) can be used to add the feature maps output by the first convolutional unit 02 and the first upsampling and convolutional unit 08, and output the result to the first classification unit 10 for classification. The first end of the second integration unit 12 can be connected to the third end of the fourth convolutional unit 05. The first end of the second classification unit 13 can be connected to the second end of the second integration unit 12. One end of the second upsampling and convolution unit 11 can be connected to the second end of the sixth convolution unit 07, and the second end of the second upsampling and convolution unit 11 is connected to the third end of the second integration unit 12. The second upsampling and convolution unit 11 can be used to sequentially upsample and convolve the feature map output by the sixth convolution unit 07, thereby ensuring that the output feature map maintains the same dimension as the feature map output by the fourth convolution unit 05. The second integration unit 12 (concat unit) can be used to add the feature maps output by the fourth convolution unit 05 and the second upsampling and convolution unit 11, and output the result to the second classification unit 13 for classification. The third classification unit 14 can be connected to the third end of the sixth convolution unit 07 and is used to perform classification on the feature map output by the sixth convolution unit 07.The first end of the third upsampling and convolution unit 15 can be connected to the third end of the fifth convolution unit 06. The first end of the third integration unit 16 can be connected to the third end of the second convolution unit 03, and the second end of the third integration unit 16 can be connected to the other end of the third upsampling and convolution unit 15. One end of the fourth classification unit 17 can be connected to the third end of the third integration unit 16. The DS unit can be connected to the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17, and is used to determine the final classification result based on the classification results of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 using evidence theory.

[0074] In this embodiment, the specific structures of the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 can be various forms known to those skilled in the art. In one example of the present invention, the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 can include, for example: Figure 5 The structure shown. Specifically, in this Figure 5 In this configuration, at least one of the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 may include a first residual convolutional layer 20, a two-dimensional convolutional layer 21, and a second residual convolutional layer 22. One end of the first residual convolutional layer 20 can be used to receive the input from the previous network layer. One end of the two-dimensional convolutional layer 21 can be connected to the other end of the first residual convolutional layer 20. One end of the second residual convolutional layer 22 can be connected to both the other end of the two-dimensional convolutional layer 21 and one end of the first residual convolutional layer 20.

[0075] In this embodiment, the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can all be used for classification operations. The specific structure of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can be of various forms known to those skilled in the art. In one example of the present invention, the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can include, for example: Figure 6 The structure shown. In this Figure 6In this configuration, at least one of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 may include a first convolutional layer 30, a first pooling layer 31, a first multiplication module 32, a global average pooling layer 33, a second convolutional layer 34, a mapping layer 35, a second multiplication module 36, a third convolutional layer 37, and a fully connected layer 38. Specifically, the first convolutional layer 30 can perform convolution operations, and the first pooling layer 31 can perform pooling operations on the result of the convolution operations. The first multiplication module 32 can perform matrix multiplication operations on the input of the first convolutional layer 30 and the output of the first pooling layer 31. One end of the global average pooling layer 33 can be connected to the third end of the first multiplication module 32 for performing global average pooling operations. One end of the second convolutional layer 34 can be connected to the other end of the global average pooling layer 33 for performing convolution operations. One end of the mapping layer 35 can be connected to the other end of the second convolutional layer 34. The first end of the second multiplication module 36 can be connected to the other end of the mapping layer 35, and the second end of the second multiplication module 36 can be connected to the fourth end of the first multiplication module 32, for performing matrix multiplication operations on the output of the mapping layer 35 and the output of the first multiplication module 32. One end of the third convolutional layer 37 can be connected to the third end of the second multiplication module 36. One end of the fully connected layer 38 can be connected to the other end of the third convolutional layer 37, and the other end of the fully connected layer 38 is used to output the classification result.

[0076] In this embodiment, the training steps of the image recognition network can take various forms known to those skilled in the art. In one example of the present invention, considering that various indicators of the predicted value need to be taken into account during the training process, the loss function for training the image recognition network can be the following formulas (1) and (2).

[0077]

[0078] VFL=-β(1-p i ) 2 log(p j )-(1-β)p i 2 log(1-p j (2)

[0079] Where Y is the loss function, λ1, λ2, and λ3 are the adaptive weights, n is the number of training samples, and y i For the numerical expression of the standard value, y i Numerical representation of the predicted value. Here, b is the graphical representation of the predicted value, α is the graphical representation of the standard value, β is the hyperparameter, and p is the adjustment factor. i p j These are the probabilities of the i-th and j-th categories, respectively.

[0080] On the other hand, the present invention also provides a power pole identification system suitable for drones, the identification system including a processor configured to perform the identification method as described above. Specifically, the identification method may include the following steps:

[0081] In step S10, an image of the power pole to be identified is acquired;

[0082] In step S11, the power pole images are preprocessed to obtain standard images;

[0083] In step S12, a standard image is input into a preset image recognition network to determine the location of the power pole.

[0084] In such Figure 1 In the identification method shown, step S10 can be used to acquire an image of the power pole to be identified. This power pole image can be, for example... Figure 2 As shown, this was captured in real-time by a drone.

[0085] Step S11 can be used to preprocess the images of power poles to obtain standard images. The specific method for this preprocessing operation can be of various forms known to those skilled in the art. In one example of the present invention, considering that drone inspections are generally conducted on sunny days with high light intensity, resulting in brighter images, and given the specific outdoor environment, the depth of field in the images is relatively complex, the preprocessing operation can include, for example... Figure 3 The method shown. In this Figure 3 In this process, the preprocessing operation may include the following steps:

[0086] In step S20, a high-pass filter is used to perform spatial filtering on the power pole image, thereby improving the edge features of the power pole image;

[0087] In step S21, histogram equalization is performed on the power pole image after spatial filtering to reduce the grayscale range of the power pole image.

[0088] In step S22, a denoising operation is performed on the power pole image after histogram equalization to reduce the noise effect of the power pole image.

[0089] Step S12 can be used to input a standard image into a preset image recognition network to determine the location of the power pole. While the specific structure of this image recognition network can take many forms known to those skilled in the art, considering the characteristics of power pole images described above, in one example of the present invention, the image recognition network can be as follows: Figure 4 As shown. In this Figure 4The image recognition network may include residual convolution unit 01, first convolution unit 02, second convolution unit 03, third convolution unit 04, fourth convolution unit 05, fifth convolution unit 06, sixth convolution unit 07, first upsampling and convolution unit 08, first integration unit 09, first classification unit 10, second upsampling and convolution unit 11, second integration unit 12, second classification unit 13, third classification unit 14, third upsampling and convolution unit 15, third integration unit 16, fourth classification unit 17, and a judgment unit (not shown in the figure).

[0090] The residual convolutional unit 01 can be used to read images of power poles to be identified. The first end of the first convolutional unit 02 can be connected to the residual convolutional unit 01 to perform feature extraction on the feature map output by the residual convolutional unit 01. The first end of the second convolutional unit 03 can be connected to the second end of the first convolutional unit 02 to perform feature extraction on the feature map output by the first convolutional unit 02. The first end of the third convolutional unit 04 can be connected to the second end of the second convolutional unit 03 to perform feature extraction on the feature map output by the second convolutional unit 03. The first end of the fourth convolutional unit 05 can be connected to the second end of the third convolutional unit 04 to perform feature extraction on the feature map output by the third convolutional unit 04. The first end of the fifth convolutional unit 06 can be connected to the second end of the fourth convolutional unit 05 to perform feature extraction on the feature map output by the fourth convolutional unit 05. The first end of the sixth convolutional unit 07 can be connected to the second end of the fifth convolutional unit 06 to perform feature extraction on the feature map output by the fifth convolutional unit 06. The first end of the first integration unit 09 can be connected to the third end of the first convolutional unit 02. The first end of the first classification unit 10 can be connected to the second end of the first integration unit 09. One end of the first upsampling and convolutional unit 08 can be connected to the third end of the third convolutional unit 04, and the other end of the first upsampling and convolutional unit 08 can be connected to the third end of the first integration unit 09. The first upsampling and convolutional unit 08 can be used to sequentially perform upsampling and convolution operations on the feature map output by the third convolutional unit 04, thereby ensuring that the output feature map maintains the same dimension as the feature map output by the first convolutional unit 02. The first integration unit 09 (concat unit) can be used to add the feature maps output by the first convolutional unit 02 and the first upsampling and convolutional unit 08, and output the result to the first classification unit 10 for classification. The first end of the second integration unit 12 can be connected to the third end of the fourth convolutional unit 05. The first end of the second classification unit 13 can be connected to the second end of the second integration unit 12. One end of the second upsampling and convolution unit 11 can be connected to the second end of the sixth convolution unit 07, and the second end of the second upsampling and convolution unit 11 is connected to the third end of the second integration unit 12. The second upsampling and convolution unit 11 can be used to sequentially upsample and convolve the feature map output by the sixth convolution unit 07, thereby ensuring that the output feature map maintains the same dimension as the feature map output by the fourth convolution unit 05. The second integration unit 12 (concat unit) can be used to add the feature maps output by the fourth convolution unit 05 and the second upsampling and convolution unit 11, and output the result to the second classification unit 13 for classification. The third classification unit 14 can be connected to the third end of the sixth convolution unit 07 and is used to perform classification on the feature map output by the sixth convolution unit 07.The first end of the third upsampling and convolution unit 15 can be connected to the third end of the fifth convolution unit 06. The first end of the third integration unit 16 can be connected to the third end of the second convolution unit 03, and the second end of the third integration unit 16 can be connected to the other end of the third upsampling and convolution unit 15. One end of the fourth classification unit 17 can be connected to the third end of the third integration unit 16. The DS unit can be connected to the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17, and is used to determine the final classification result based on the classification results of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 using evidence theory.

[0091] In this embodiment, the specific structures of the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 can be various forms known to those skilled in the art. In one example of the present invention, the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 can include, for example: Figure 5 The structure shown. Specifically, in this Figure 5 In this configuration, at least one of the first convolutional unit 02, the second convolutional unit 03, the third convolutional unit 04, the fourth convolutional unit 05, the fifth convolutional unit 06, and the sixth convolutional unit 07 may include a first residual convolutional layer 20, a two-dimensional convolutional layer 21, and a second residual convolutional layer 22. One end of the first residual convolutional layer 20 can be used to receive the input from the previous network layer. One end of the two-dimensional convolutional layer 21 can be connected to the other end of the first residual convolutional layer 20. One end of the second residual convolutional layer 22 can be connected to both the other end of the two-dimensional convolutional layer 21 and one end of the first residual convolutional layer 20.

[0092] In this embodiment, the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can all be used for classification operations. The specific structure of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can be of various forms known to those skilled in the art. In one example of the present invention, the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 can include, for example: Figure 6 The structure shown. In this Figure 6In this configuration, at least one of the first classification unit 10, the second classification unit 13, the third classification unit 14, and the fourth classification unit 17 may include a first convolutional layer 30, a first pooling layer 31, a first multiplication module 32, a global average pooling layer 33, a second convolutional layer 34, a mapping layer 35, a second multiplication module 36, a third convolutional layer 37, and a fully connected layer 38. Specifically, the first convolutional layer 30 can perform convolution operations, and the first pooling layer 31 can perform pooling operations on the result of the convolution operations. The first multiplication module 32 can perform matrix multiplication operations on the input of the first convolutional layer 30 and the output of the first pooling layer 31. One end of the global average pooling layer 33 can be connected to the third end of the first multiplication module 32 for performing global average pooling operations. One end of the second convolutional layer 34 can be connected to the other end of the global average pooling layer 33 for performing convolution operations. One end of the mapping layer 35 can be connected to the other end of the second convolutional layer 34. The first end of the second multiplication module 36 can be connected to the other end of the mapping layer 35, and the second end of the second multiplication module 36 can be connected to the fourth end of the first multiplication module 32, for performing matrix multiplication operations on the output of the mapping layer 35 and the output of the first multiplication module 32. One end of the third convolutional layer 37 can be connected to the third end of the second multiplication module 36. One end of the fully connected layer 38 can be connected to the other end of the third convolutional layer 37, and the other end of the fully connected layer 38 is used to output the classification result.

[0093] In this embodiment, the training steps of the image recognition network can take various forms known to those skilled in the art. In one example of the present invention, considering that various indicators of the predicted value need to be taken into account during the training process, the loss function for training the image recognition network can be the following formulas (1) and (2).

[0094]

[0095] VFL=-β(1-p i ) 2 log(p j )-(1-β)p i 2 log(1-p j (2)

[0096] Where Y is the loss function, λ1, λ2, and λ3 are the adaptive weights, n is the number of training samples, and y i For the numerical expression of the standard value, y i Numerical representation of the predicted value. Here, b is the graphical representation of the predicted value, α is the graphical representation of the standard value, β is the hyperparameter, and p is the adjustment factor. i p j These are the probabilities of the i-th and j-th categories, respectively.

[0097] In another aspect, the present invention also provides a computer-readable storage medium storing instructions for being read by a machine to cause the machine to perform any of the identification methods described above.

[0098] Through the above technical solution, the embodiments of the present invention provide a method and system for identifying power poles using drones. This method and system identify power poles by combining images captured by the drone, enabling the drone to independently complete the inspection of power poles. Compared with the prior art, the identification method and system provided by the present invention have faster identification efficiency.

[0099] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0104] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0105] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0106] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0107] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for identifying power poles suitable for unmanned aerial vehicles (UAVs), characterized in that, The identification method includes: Acquire images of the power poles to be identified; The images of the power poles are preprocessed to obtain standard images; The standard image is input into a preset image recognition network to determine the location of the power pole; The image recognition network includes: A residual convolutional unit is used to read the image of the power pole to be identified; The first convolutional unit, with its first end connected to the residual convolutional unit; The second convolutional unit, wherein the first end of the second convolutional unit is connected to the second end of the first convolutional unit; The third convolutional unit, wherein the first end of the third convolutional unit is connected to the second end of the second convolutional unit; The fourth convolutional unit, wherein the first end of the fourth convolutional unit is connected to the second end of the third convolutional unit; The fifth convolutional unit, wherein the first end of the fifth convolutional unit is connected to the second end of the fourth convolutional unit; The sixth convolutional unit, wherein the first end of the sixth convolutional unit is connected to the second end of the fifth convolutional unit; The first integration unit, wherein the first end of the first integration unit is connected to the third end of the first convolution unit; A first classification unit, wherein a first end of the first classification unit is connected to a second end of the first integration unit; A first upsampling and convolution unit, one end of which is connected to the third end of the third convolution unit, and the other end of which is connected to the third end of the first integration unit; The second integration unit, the first end of which is connected to the third end of the fourth convolution unit; The second classification unit, wherein the first end of the second classification unit is connected to the second end of the second integration unit; The second upsampling and convolution unit has one end connected to the second end of the sixth convolution unit, and the second end of the second upsampling and convolution unit is connected to the third end of the second integration unit. The third classification unit is connected to the third end of the sixth convolutional unit; The third upsampling and convolution unit, wherein the first end of the third upsampling and convolution unit is connected to the third end of the fifth convolution unit; The third integration unit has a first end connected to the second convolution unit and a second end connected to the other end of the third upsampling and convolution unit. The fourth classification unit, one end of which is connected to the third end of the third integration unit; The DS unit, connected to the first, second, third, and fourth classification units, is used to determine the final classification result based on the classification results of the first, second, third, and fourth classification units using evidence theory.

2. The identification method according to claim 1, characterized in that, The images of the power poles are preprocessed to obtain standard images, including: A high-pass filter is used to perform spatial filtering on the power pole image; Perform histogram equalization on the power pole image after spatial filtering; The power pole image after histogram equalization is then denoised.

3. The identification method according to claim 2, characterized in that, At least one of the first convolutional unit, the second convolutional unit, the third convolutional unit, the fourth convolutional unit, the fifth convolutional unit, and the sixth convolutional unit includes: The first residual convolutional layer, one end of which is used to receive the input of the previous network layer; A two-dimensional convolutional layer, one end of which is connected to the other end of the first residual convolutional layer; The second residual convolutional layer has one end connected to the other end of the two-dimensional convolutional layer and one end of the first residual convolutional layer.

4. The identification method according to claim 1, characterized in that, At least one of the first classification unit, the second classification unit, the third classification unit, and the fourth classification unit includes: First convolutional layer; A first pooling layer, one end of which is connected to the other end of the first convolutional layer; The first multiplication module has a first end connected to one end of the first convolutional layer and a second end connected to the other end of the first pooling layer. A global average pooling layer, one end of which is connected to the third end of the first multiplication module; The second convolutional layer, one end of which is connected to the other end of the global average pooling layer; A mapping layer, one end of which is connected to the other end of the second convolutional layer; The second multiplication module has a first end connected to the other end of the mapping layer, and a second end connected to the fourth end of the first multiplication module. The third convolutional layer, one end of which is connected to the third end of the second multiplication module; A fully connected layer, one end of which is connected to the other end of the third convolutional layer, and the other end of which is used to output the classification result.

5. The identification method according to claim 1, characterized in that, The loss functions for training the image recognition network are Equations (1) and (2). ，（1） ，（2） in, For loss function, , and For adaptive weights, The number of training samples, The numerical expression of the standard value. Numerical representation of the predicted value, A graphical representation of the predicted values. A graphical representation of the standard value. For hyperparameters, As a regulating factor, , The first The and the first The probability of each category.

6. A power pole identification system suitable for unmanned aerial vehicles (UAVs), characterized in that, The identification system includes a processor configured to perform the identification method as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that are read by a machine to cause the machine to perform the identification method as described in any one of claims 1 to 5.

Images