Image acquisition method and system for electric power inspection robot

Abstract

The present invention discloses an image acquisition method and system for an electric power inspection robot, and the present invention relates to the technical field of image acquisition of electric power inspection robots, including a dust visual acquisition module, a dust identification and processing module, a dust monitoring and analysis module, and a dust removal execution module. The dust visual acquisition module is used to obtain dust data and inspection image data; the dust identification and processing module is divided into a dust three-dimensional model construction unit that integrates discrete dust Z-axis slice images and their distribution positions into a dust three-dimensional model, a dust feature extraction unit that extracts dust thickness from the dust three-dimensional model, and a dust model construction unit that obtains the inspection image impact results corresponding to different dust thicknesses and constructs a dust identification scheduling model; the dust monitoring and analysis module analyzes dust thickness data, evaluates the scheduling index of the inspection robot image acquisition, and transmits the analysis results to the dust removal execution module; the dust removal execution module receives the inspection robot scheduling index results and performs corresponding operations to respond.

Description

Technical Field

[0001] The present invention relates to the technical field of image acquisition by electric power inspection robots, and in particular to an image acquisition method and system for electric power inspection robots. Background Art

[0002] The image acquisition system of the power inspection robot is an important technical support for the development of the modern power industry. With the continuous expansion of the scale of the power grid and the increasing number of power equipment, the traditional manual inspection method has been unable to meet the needs of efficient and accurate inspections. Therefore, the power inspection robot came into being and was equipped with an advanced image acquisition system. The system uses high-definition sensors and advanced camera technology to capture high-definition images of power equipment in real time, providing reliable data support for subsequent image recognition and analysis. At the same time, combined with image recognition and analysis technology, the power inspection robot can automatically identify key information of power equipment, such as equipment type, operating status, etc., and realize automatic judgment and early warning of the status of power equipment. This automated and intelligent inspection method not only improves the efficiency and accuracy of inspections, but also greatly reduces the safety risks of inspectors. Therefore, the image acquisition system of the power inspection robot has become one of the indispensable important technologies in the modern power industry.

[0003] Although the existing power inspection robot image acquisition system has made great progress, there are still some problems that need to be optimized. When the power inspection robot image acquisition system is collecting information about the structure and layout of the mine and whether the appearance of the power equipment in the mine is intact, due to the high dust concentration in the mine, dust particles are easily attached to the camera lens of the power inspection robot, resulting in reduced clarity of the image collected by the camera, and even blurred and incomplete images. This will not only affect the operation and maintenance personnel's accurate judgment of the equipment status, but may also delay the discovery and handling of faults. Summary of the invention

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: an image acquisition system for a power inspection robot, comprising a dust visual acquisition module, a dust identification and processing module, a dust monitoring and analysis module, and a dust removal execution module;

[0005] The dust visual acquisition module is used to obtain dust data and inspection image data;

[0006] The dust identification and processing module is divided into a dust three-dimensional model construction unit, a dust feature extraction unit and a dust model construction unit, wherein the dust three-dimensional model construction unit is used to integrate discrete dust Z-axis slice images and their distribution position data into a dust three-dimensional model; the dust feature extraction unit is used to extract dust thickness data from the dust three-dimensional model; the dust model construction unit is used to obtain the effects of different dust thicknesses on the inspection image and construct a dust identification scheduling model.

[0007] The dust monitoring and analysis module analyzes dust thickness data, evaluates the scheduling index of the inspection robot image acquisition, and transmits the analysis results to the dust removal execution module;

[0008] The dust removal execution module receives the inspection robot scheduling index result and executes corresponding operations to respond.

[0009] Furthermore, the dust data is a dust Z-axis slice image and its distribution position data, and the process of obtaining the dust data includes:

[0010] A1. According to the size range of dust particles, select the laser wavelength and magnification, adjust the scanning speed to be consistent with the inspection image acquisition frame rate, and use the laser scanning confocal microscope LSCM to perform a low-resolution scan on the lens to obtain the distribution position of dust particles;

[0011] A2. Plan the scanning area according to the distribution of dust particles, then start the high-resolution scanning mode, move the laser beam, scan the lens surface point by point, and capture the dust depth information by automatically adjusting the focal length to obtain discrete dust Z-axis slice images.

[0012] Furthermore, the inspection image data is the inspection image data corresponding to each time when the dust data is collected, and the acquisition process includes:

[0013] The inspection camera is physically connected to the LSCM, sharing a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time. The inspection camera is used to collect images in real time to obtain the inspection image data corresponding to each moment of collecting dust data.

[0014] Furthermore, the dust three-dimensional model building unit integrates the discrete dust Z-axis slice images and their distribution position data into the dust three-dimensional model, including:

[0015] The collected discrete dust Z-axis slice images and their distribution position data are imported into ZEN Software. The overlapping area size and tolerance range parameters are adjusted. The images are spliced ​​by detecting and matching the feature points of each layer to obtain the spliced ​​dust image. The maximum intensity projection (MIP) is selected to convert the spliced ​​image into a three-dimensional model to obtain the dust three-dimensional model.

[0016] Furthermore, the dust feature extraction unit extracts dust thickness data from the dust three-dimensional model in a process including:

[0017] Initialize the level set function Φ=(x, y, z), where the zero level set Φ=(x, y, z)=0 is the boundary of the dust layer. Define the evolution normal velocity of the level set function, which controls the direction and rate of interface movement. Based on the mean curvature flow equation, capture the changes of dust boundaries in the three-dimensional dust model. Use the finite difference method to discretize the level set equation, and update the level set function through an iterative process until a stable state is reached and the update stops. After the evolution is completed, extract the zero level set to obtain the upper and lower boundaries of the dust layer. For each position, calculate the shortest distance between the upper and lower boundaries, and then smooth the shortest distance data to obtain the dust thickness data.

[0018] Furthermore, the dust model building unit obtains the effects of different dust thicknesses on the inspection image in the following process:

[0019] B1. Performing size adjustment and normalization processing on the image data corresponding to each moment of collecting dust data, presetting dust thickness thresholds, namely, a first dust thickness threshold and a second dust thickness threshold, according to which dust thickness below the first dust thickness threshold is set as slight pollution; dust thickness between the first dust thickness threshold and the second dust thickness threshold is set as moderate pollution; and dust thickness above the second dust thickness threshold is set as severe pollution;

[0020] B2. The standardized image data and its corresponding dust thickness level are used as labels to form the training set, validation set and test set of the convolutional neural network (CNN) supervised learning. The training set is used to train the CNN model. During the training process, the inspection images corresponding to different dust thickness levels are divided into high-quality images, qualified images and unqualified images. The cross entropy loss function is used to evaluate the model performance, and the overfitting phenomenon of the model is checked through the validation set.

[0021] B3. After the model training is completed, the generated CNN model is evaluated using the test set, the parameters are adjusted to optimize the model, and the CNN model is deployed to the inspection robot image acquisition system. The dust thickness data is input, and the model output corresponds to the inspection image impact result.

[0022] Furthermore, the dust model building unit builds a dust identification scheduling model, including:

[0023] C1. Input the results of the impact of different dust thicknesses on the inspection images into the random forest model and combine them into a data set. The data set is divided into a training set and a test set in a ratio of 7:3. The number of decision trees is selected as 1 and the maximum depth is 3.

[0024] C2. Use the training set to train the random forest model. During the training process, the inspection image influence results of high-quality images, qualified images, and unqualified images are assigned to dispatch indices of 0%, 50%, and 100%, respectively;

[0025] C3. After the model training is completed, the generated random forest model is evaluated using the test set, the parameters of the model are adjusted to optimize the model, and the random forest model is deployed to the inspection robot image acquisition system. The data on the impact of different dust thicknesses on the inspection images is input, and the corresponding scheduling index is output to obtain the dust recognition scheduling model.

[0026] Furthermore, the dust monitoring and analysis module analyzes dust thickness data in the following process:

[0027] The constructed CNN model and dust recognition scheduling model are used to analyze the dust thickness data. When the dust thickness is lower than the first dust thickness threshold, the inspection robot image acquisition device is slightly contaminated, and the corresponding image impact result is a high-quality image with high image quality, and the corresponding scheduling index is 0%; when the dust thickness is between the first dust thickness threshold and the second dust thickness threshold, the inspection robot image acquisition device is moderately contaminated, and the corresponding image impact result is a qualified image with average image quality, and the corresponding scheduling index is 50%; when the dust thickness is higher than the second dust thickness threshold, the inspection robot image acquisition device is severely contaminated, and the corresponding image impact result is an unqualified image with low image quality, and the corresponding scheduling index is 100%, and the generated scheduling index result is transmitted to the dust removal execution module.

[0028] Furthermore, the dust removal execution module receives the inspection robot scheduling index result and executes corresponding operations to respond, including:

[0029] When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the automatic cleaning device of the inspection robot cannot clean it thoroughly, so the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are required to respond.

[0030] A method for collecting images of a power inspection robot is implemented based on the image collection system of the power inspection robot described above, and comprises the following steps:

[0031] Step 1: Use a laser scanning confocal microscope (LSCM) to obtain a Z-axis slice image of discrete dust and its distribution position. The inspection camera is physically connected to the LSCM to share a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time, and the inspection camera is used to collect the inspection image data corresponding to the dust data.

[0032] Step 2: Integrate the discrete dust Z-axis slice images and their distribution position data into a dust 3D model through ZEN Software;

[0033] Step 3: Using the level set function to extract dust thickness data from the dust three-dimensional model;

[0034] Step 4: Standardize the image data corresponding to each moment of collecting dust data, preset the dust thickness threshold, and divide the dust thickness pollution level. By building a CNN model, obtain the inspection image impact results corresponding to the dust thickness pollution level;

[0035] Step 5: By constructing a random forest model, different scheduling indexes for different dust thicknesses corresponding to the inspection image impact result data are obtained;

[0036] Step 6: Analyze the dust thickness data, evaluate the scheduling index of the inspection robot image acquisition, and transmit the analysis results to the dust removal execution module;

[0037] Step 7. When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the inspection robot's automatic cleaning device cannot clean it thoroughly, and the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are performed to respond.

[0038] Beneficial effects of the present invention: Compared with the traditional image acquisition method and system of electric power inspection robot, the discrete dust Z-axis slicing technology, three-dimensional modeling technology, dust feature extraction technology, CNN model construction technology, and random forest model construction technology in the system of the present invention are closely combined with modern information technology, and the laser scanning confocal microscope LSCM technology, image acquisition technology and TTL pulse synchronization timing technology are integrated to accurately capture the dust data and inspection image data required for monitoring by the inspection robot to collect images, thereby achieving real-time and comprehensive monitoring of the centralized monitoring system of the inspection robot, and through ZEN The software integrates the discrete dust Z-axis slice image and its distribution position data into a dust 3D model, and then uses the level set function to extract the dust thickness data from the dust 3D model. Through the dust thickness data and inspection image data obtained after data processing, a CNN model is constructed to obtain the inspection image impact result data of different dust thicknesses, and a random forest model is constructed to obtain the corresponding scheduling index of the inspection image impact result data of different inspection images. The software analyzes the impact of dust on image acquisition when the inspection robot is working in the mining area in real time, monitors anomalies in real time, and responds to anomalies encountered during system operation in a timely manner, solving the problem of the power inspection robot image acquisition system in collecting the structure and layout of the mine cave. , during the working process of checking whether the appearance of power equipment in mines is intact, due to the high dust concentration in mines, dust particles are easily attached to the camera lens of the power inspection robot, resulting in reduced clarity of the image collected by the camera, and even blurred and incomplete images. This will not only affect the operation and maintenance personnel's accurate judgment of the equipment status, but may also delay the discovery and handling of faults. It is ensured that the method in the present invention can refine the dynamic monitoring standard of the image acquisition system of the power inspection robot within a more precise range, so that the monitored data becomes a more accurate indicator under the same conditions. The development and application of this method have significantly enhanced the intelligence of the inspection robot in the image acquisition process. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a block diagram of an image acquisition system for a power inspection robot according to the present invention;

[0040] Figure 2 The present invention provides a flow chart of an image acquisition method for an electric power inspection robot. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] like Figure 1 and Figure 2 As shown, the present invention provides a technical solution: an image acquisition system for a power inspection robot, including a dust visual acquisition module, a dust identification and processing module, a dust monitoring and analysis module, and a dust removal execution module;

[0043] Dust visual collection module, used to obtain dust data and inspection image data;

[0044] The dust recognition processing module is divided into a dust three-dimensional model construction unit, a dust feature extraction unit and a dust model construction unit. The dust three-dimensional model construction unit is used to integrate discrete dust Z-axis slice images and their distribution position data into a dust three-dimensional model; the dust feature extraction unit is used to extract dust thickness data from the dust three-dimensional model; the dust model construction unit is used to obtain the effects of different dust thicknesses on the inspection image and construct a dust recognition scheduling model.

[0045] The dust monitoring and analysis module analyzes dust thickness data, evaluates the scheduling index of the inspection robot's image acquisition, and transmits the analysis results to the dust removal execution module;

[0046] The dust removal execution module receives the inspection robot scheduling index results and performs corresponding operations to respond.

[0047] The dust data is the dust Z-axis slice image and its distribution position data, and the acquisition process includes:

[0048] A1. According to the size range of dust particles, select the laser wavelength and magnification, adjust the scanning speed to be consistent with the inspection image acquisition frame rate, and use the laser scanning confocal microscope LSCM to perform a low-resolution scan on the lens to obtain the distribution position of dust particles;

[0049] A2. Plan the scanning area according to the distribution of dust particles, then start the high-resolution scanning mode, move the laser beam, scan the lens surface point by point, and capture the dust depth information by automatically adjusting the focal length to obtain discrete dust Z-axis slice images.

[0050] The inspection image data is the inspection image data corresponding to each time when the dust data is collected, and the acquisition process includes:

[0051] The inspection camera is physically connected to the LSCM, sharing a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time. The inspection camera is used to collect images in real time to obtain the inspection image data corresponding to each moment of collecting dust data.

[0052] The dust 3D model building unit integrates the discrete dust Z-axis slice image and its distribution position data into the dust 3D model, including:

[0053] The collected discrete dust Z-axis slice images and their distribution position data are imported into ZEN Software. The overlapping area size and tolerance range parameters are adjusted. The images are spliced ​​by detecting and matching the feature points of each layer to obtain the spliced ​​dust image. The maximum intensity projection (MIP) is selected to convert the spliced ​​image into a three-dimensional model to obtain the dust three-dimensional model.

[0054] The dust feature extraction unit extracts dust thickness data from the dust three-dimensional model in the following process:

[0055] Initialize the level set function Φ=(x, y, z), where the zero level set Φ=(x, y, z)=0 is the boundary of the dust layer. Define the evolution normal velocity of the level set function, which controls the direction and rate of interface movement. Based on the mean curvature flow equation, capture the changes of dust boundaries in the three-dimensional dust model. Use the finite difference method to discretize the level set equation, and update the level set function through an iterative process until a stable state is reached and the update stops. After the evolution is completed, extract the zero level set to obtain the upper and lower boundaries of the dust layer. For each position, calculate the shortest distance between the upper and lower boundaries, and then smooth the shortest distance data to obtain the dust thickness data.

[0056] The dust model building unit obtains the effects of different dust thicknesses on the inspection image, and the process includes:

[0057] B1. Performing size adjustment and normalization processing on the image data corresponding to each moment of collecting dust data, presetting dust thickness thresholds, namely, a first dust thickness threshold and a second dust thickness threshold, according to which dust thickness below the first dust thickness threshold is set as slight pollution; dust thickness between the first dust thickness threshold and the second dust thickness threshold is set as moderate pollution; and dust thickness above the second dust thickness threshold is set as severe pollution;

[0058] B2. The standardized image data and its corresponding dust thickness level are used as labels to form the training set, validation set and test set of the convolutional neural network (CNN) supervised learning. The training set is used to train the CNN model. During the training process, the inspection images corresponding to different dust thickness levels are divided into high-quality images, qualified images and unqualified images. The cross entropy loss function is used to evaluate the model performance, and the overfitting phenomenon of the model is checked through the validation set.

[0059] B3. After the model training is completed, the generated CNN model is evaluated using the test set, the parameters are adjusted to optimize the model, and the CNN model is deployed to the inspection robot image acquisition system. The dust thickness data is input, and the model output corresponds to the inspection image impact result.

[0060] Dust model building unit,The process of building a dust identification and scheduling model includes:

[0061] C1. Input the results of the impact of different dust thicknesses on the inspection images into the random forest model and combine them into a data set. The data set is divided into a training set and a test set in a ratio of 7:3. The number of decision trees is selected as 1 and the maximum depth is 3.

[0062] C2. Use the training set to train the random forest model. During the training process, the inspection image influence results of high-quality images, qualified images, and unqualified images are assigned to dispatch indices of 0%, 50%, and 100%, respectively;

[0063] C3. After the model training is completed, the generated random forest model is evaluated using the test set, the parameters of the model are adjusted to optimize the model, and the random forest model is deployed to the inspection robot image acquisition system. The data on the impact of different dust thicknesses on the inspection images is input, and the corresponding scheduling index is output to obtain the dust recognition scheduling model.

[0064] Dust monitoring and analysis module, the process of analyzing dust thickness data includes:

[0065] The constructed CNN model and dust recognition scheduling model are used to analyze the dust thickness data. When the dust thickness is lower than the first dust thickness threshold, the inspection robot image acquisition device is slightly contaminated, and the corresponding image impact result is a high-quality image with high image quality, and the corresponding scheduling index is 0%; when the dust thickness is between the first dust thickness threshold and the second dust thickness threshold, the inspection robot image acquisition device is moderately contaminated, and the corresponding image impact result is a qualified image with average image quality, and the corresponding scheduling index is 50%; when the dust thickness is higher than the second dust thickness threshold, the inspection robot image acquisition device is severely contaminated, and the corresponding image impact result is an unqualified image with low image quality, and the corresponding scheduling index is 100%, and the generated scheduling index result is transmitted to the dust removal execution module.

[0066] The dust removal execution module receives the inspection robot scheduling index result and performs corresponding operations to respond, including:

[0067] When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the automatic cleaning device of the inspection robot cannot clean it thoroughly, so the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are required to respond.

[0068] A method for collecting images of a power inspection robot is implemented based on the image collection system of the power inspection robot, and comprises the following steps:

[0069] Step 1: Use a laser scanning confocal microscope (LSCM) to obtain a Z-axis slice image of discrete dust and its distribution position. The inspection camera is physically connected to the LSCM to share a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time, and the inspection camera is used to collect the inspection image data corresponding to the dust data.

[0070] Step 2: Integrate the discrete dust Z-axis slice images and their distribution position data into a dust 3D model through ZEN Software;

[0071] Step 3: Using the level set function to extract dust thickness data from the dust three-dimensional model;

[0072] Step 4: Standardize the image data corresponding to each moment of collecting dust data, preset the dust thickness threshold, and divide the dust thickness pollution level. By building a CNN model, obtain the inspection image impact results corresponding to the dust thickness pollution level;

[0073] Step 5: By constructing a random forest model, different scheduling indexes for different dust thicknesses corresponding to the inspection image impact result data are obtained;

[0074] Step 6: Analyze the dust thickness data, evaluate the scheduling index of the inspection robot image acquisition, and transmit the analysis results to the dust removal execution module;

[0075] Step 7. When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the inspection robot's automatic cleaning device cannot clean it thoroughly, and the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are performed to respond.

[0076] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions. The sentence "includes an element defined by ... does not exclude the existence of other identical elements in the process, method, article or device including the element".

[0077] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image acquisition system for an electric power inspection robot, comprising a dust visual acquisition module, a dust identification and processing module, a dust monitoring and analysis module, and a dust removal execution module, characterized in that: The dust visual acquisition module is used to obtain dust data and inspection image data; The dust identification and processing module is divided into a dust three-dimensional model construction unit, a dust feature extraction unit and a dust model construction unit, wherein the dust three-dimensional model construction unit is used to integrate discrete dust Z-axis slice images and their distribution position data into a dust three-dimensional model; the dust feature extraction unit is used to extract dust thickness data from the dust three-dimensional model; the dust model construction unit is used to obtain the effects of different dust thicknesses on the inspection image and construct a dust identification scheduling model. The dust monitoring and analysis module analyzes dust thickness data, evaluates the scheduling index of the inspection robot image acquisition, and transmits the analysis results to the dust removal execution module; The dust removal execution module receives the inspection robot scheduling index result and executes corresponding operations to respond.

2. The image acquisition system of a power inspection robot according to claim 1, characterized in that: The dust data is a dust Z-axis slice image and its distribution position data, and the process of obtaining the dust data includes: A1. According to the size range of dust particles, select the laser wavelength and magnification, adjust the scanning speed to be consistent with the inspection image acquisition frame rate, and use the laser scanning confocal microscope LSCM to perform a low-resolution scan on the lens to obtain the distribution position of dust particles; A2. Plan the scanning area according to the distribution of dust particles, then start the high-resolution scanning mode, move the laser beam, scan the lens surface point by point, and capture the dust depth information by automatically adjusting the focal length to obtain discrete dust Z-axis slice images.

3. The image acquisition system of a power inspection robot according to claim 2 is characterized in that: The inspection image data is the inspection image data corresponding to each time when the dust data is collected, and the acquisition process includes: The inspection camera is physically connected to the LSCM, sharing a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time. The inspection camera is used to collect images in real time to obtain the inspection image data corresponding to each moment of collecting dust data.

4. The image acquisition system of a power inspection robot according to claim 3 is characterized in that: The dust 3D model building unit integrates the discrete dust Z-axis slice images and their distribution position data into the dust 3D model, including: The collected discrete dust Z-axis slice images and their distribution position data are imported into ZEN Software. The overlapping area size and tolerance range parameters are adjusted. The images are spliced ​​by detecting and matching the feature points of each layer to obtain the spliced ​​dust image. The maximum intensity projection (MIP) is selected to convert the spliced ​​image into a three-dimensional model to obtain the dust three-dimensional model.

5. The image acquisition system of a power inspection robot according to claim 4 is characterized in that: The process of extracting dust thickness data from the dust three-dimensional model by the dust feature extraction unit includes: Initialize the level set function Φ=(x, y, z), where the zero level set Φ=(x, y, z)=0 is the boundary of the dust layer. Define the evolution normal velocity of the level set function, which controls the direction and rate of interface movement. Based on the mean curvature flow equation, capture the changes of dust boundaries in the three-dimensional dust model. Use the finite difference method to discretize the level set equation, and update the level set function through an iterative process until a stable state is reached and the update stops. After the evolution is completed, extract the zero level set to obtain the upper and lower boundaries of the dust layer. For each position, calculate the shortest distance between the upper and lower boundaries, and then smooth the shortest distance data to obtain the dust thickness data.

6. The image acquisition system of a power inspection robot according to claim 5, characterized in that: The process of obtaining the effects of different dust thicknesses on the inspection image by the dust model building unit includes: B1. Performing size adjustment and normalization processing on the image data corresponding to each moment of collecting dust data, presetting dust thickness thresholds, namely, a first dust thickness threshold and a second dust thickness threshold, according to which dust thickness below the first dust thickness threshold is set as slight pollution; dust thickness between the first dust thickness threshold and the second dust thickness threshold is set as moderate pollution; and dust thickness above the second dust thickness threshold is set as severe pollution; B2. The standardized image data and its corresponding dust thickness level are used as labels to form the training set, validation set and test set of the convolutional neural network (CNN) supervised learning. The training set is used to train the CNN model. During the training process, the inspection images corresponding to different dust thickness levels are divided into high-quality images, qualified images and unqualified images. The cross entropy loss function is used to evaluate the model performance, and the overfitting phenomenon of the model is checked through the validation set. B3. After the model training is completed, the generated CNN model is evaluated using the test set, the parameters are adjusted to optimize the model, and the CNN model is deployed to the inspection robot image acquisition system. The dust thickness data is input, and the model output corresponds to the inspection image impact result.

7. The image acquisition system of a power inspection robot according to claim 6, characterized in that: The dust model building unit builds a dust identification scheduling model, and the process includes: C1. Input the results of the impact of different dust thicknesses on the inspection images into the random forest model and combine them into a data set. The data set is divided into a training set and a test set in a ratio of 7:

3. The number of decision trees is selected as 1 and the maximum depth is 3. C2. Use the training set to train the random forest model. During the training process, the inspection image influence results of high-quality images, qualified images, and unqualified images are assigned to dispatch indices of 0%, 50%, and 100%, respectively; C3. After the model training is completed, the generated random forest model is evaluated using the test set, the parameters of the model are adjusted to optimize the model, and the random forest model is deployed to the inspection robot image acquisition system. The data on the impact of different dust thicknesses on the inspection images is input, and the corresponding scheduling index is output to obtain the dust recognition scheduling model.

8. The image acquisition system of a power inspection robot according to claim 7, characterized in that: The dust monitoring and analysis module analyzes dust thickness data in the following process: The constructed CNN model and dust recognition scheduling model are used to analyze the dust thickness data. When the dust thickness is lower than the first dust thickness threshold, the inspection robot image acquisition device is slightly contaminated, and the corresponding image impact result is a high-quality image with high image quality, and the corresponding scheduling index is 0%; when the dust thickness is between the first dust thickness threshold and the second dust thickness threshold, the inspection robot image acquisition device is moderately contaminated, and the corresponding image impact result is a qualified image with average image quality, and the corresponding scheduling index is 50%; when the dust thickness is higher than the second dust thickness threshold, the inspection robot image acquisition device is severely contaminated, and the corresponding image impact result is an unqualified image with low image quality, and the corresponding scheduling index is 100%, and the generated scheduling index result is transmitted to the dust removal execution module.

9. The image acquisition system of a power inspection robot according to claim 8, characterized in that: The dust removal execution module receives the inspection robot scheduling index result and executes corresponding operations to respond, including: When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the automatic cleaning device of the inspection robot cannot clean it thoroughly, so the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are required to respond.

10. A method for collecting images of a power inspection robot, implemented based on the image collection system of the power inspection robot according to any one of claims 1 to 9, characterized in that: It consists of the following steps: Step 1: Use a laser scanning confocal microscope (LSCM) to obtain a Z-axis slice image of discrete dust and its distribution position. The inspection camera is physically connected to the LSCM to share a control interface and a TTL pulse. The LSCM and camera acquisition work are started at the same time, and the inspection camera is used to collect the inspection image data corresponding to the dust data. Step 2: Integrate the discrete dust Z-axis slice images and their distribution position data into a dust 3D model through ZEN Software; Step 3: Using the level set function to extract dust thickness data from the dust three-dimensional model; Step 4: Standardize the image data corresponding to each moment of collecting dust data, preset the dust thickness threshold, and divide the dust thickness pollution level. By building a CNN model, obtain the inspection image impact results corresponding to the dust thickness pollution level; Step 5: By constructing a random forest model, different scheduling indexes for different dust thicknesses corresponding to the inspection image impact result data are obtained; Step 6: Analyze the dust thickness data, evaluate the scheduling index of the inspection robot image acquisition, and transmit the analysis results to the dust removal execution module; Step 7. When the scheduling index is 0%, the dust thickness has a low impact on the image acquisition of the inspection robot; when the scheduling index is 50%, the dust thickness has a moderate impact on the image acquisition of the inspection robot, and the inspection robot starts the automatic cleaning device to clean the dust; when the scheduling index is 100%, the dust thickness has a high impact on the image acquisition of the inspection robot, and the inspection robot's automatic cleaning device cannot clean it thoroughly, and the inspection robot sends an alarm to the client, and manual deep cleaning and replacement of lens components are performed to respond.

Images