A robot inspection system for power equipment based on AI vision

Abstract

The present invention discloses an AI vision-based power equipment robot inspection system, which belongs to the field of visual detection technology, and specifically comprises: a route planning module, which is used to plan an inspection route according to a power equipment area to be inspected; an image acquisition module, which is used to acquire the image of the power equipment to be inspected based on an image sensor, and preprocess the image of the power equipment; an image processing module, which is used to perform edge detection on the preprocessed power equipment image, identify the edge line of the power equipment, and segment the power equipment area image from the background area according to the edge line; a feature extraction module, which is used to extract features from the power equipment area image; a defect detection module, which is used to establish a defect detection model based on a database and through a Transformer network, input the power equipment area image into the defect detection model, and output a corresponding defect recognition result; the present invention realizes efficient and low-cost visual detection of power equipment.

Description

Technical Field

[0001] The present invention relates to the field of visual detection technology, and in particular to an AI vision-based power equipment robot inspection system. Background Art

[0002] With the rapid development of the power industry, the number and types of power equipment are increasing, which puts higher requirements on the inspection and maintenance of power equipment. The traditional manual inspection method has problems such as low efficiency, high cost, and great safety hazards. Therefore, the development of a power equipment robot inspection system based on AI vision has important practical significance and broad market prospects.

[0003] At present, some power equipment inspection solutions based on image recognition technology have been proposed. These solutions mainly collect power equipment images through cameras installed on drones or robots, and then use computer vision algorithms to process and analyze the images, thereby realizing automatic edge detection of power equipment. However, existing solutions still have certain limitations in practical applications, such as insufficient processing capabilities for power equipment image segmentation under complex backgrounds, and high computing power costs. Summary of the invention

[0004] The purpose of the present invention is to provide an AI vision-based power equipment robot inspection system to solve the following technical problems:

[0005] The image segmentation and processing capabilities for power equipment in complex backgrounds are insufficient and the required computing power cost is high.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An AI vision-based power equipment robot inspection system, comprising:

[0008] The route planning module is used to plan the inspection route according to the power equipment area to be inspected and send the inspection route to the robot;

[0009] An image acquisition module, used for acquiring images of the electric power equipment to be inspected based on an image sensor mounted on the robot, and preprocessing the images of the electric power equipment;

[0010] An image processing module is used to perform edge detection on the preprocessed power equipment image, identify the edge line of the power equipment, and segment the power equipment area image from the background area according to the edge line;

[0011] A feature extraction module, used for extracting features from the power equipment area image;

[0012] The defect detection module is used to establish a defect detection model based on the defect image database and through the Transformer network, input the power equipment area image into the defect detection model, and output the corresponding defect recognition result.

[0013] As a further solution of the present invention: in the image acquisition module, the preprocessing process is:

[0014] The power equipment image is processed in the frequency domain by vortex filtering, and the contrast of the power equipment image is stretched to a set value, and the geometric distortion in the edge area of ​​the image is corrected. The definition of the edge area of ​​the image is set according to the focal length of the image sensor.

[0015] As a further solution of the present invention: the process of the image processing module identifying the edge line is:

[0016] Convert the power equipment image into a grayscale image, start from the edge of one side of the grayscale image perpendicular to the ground, select a column of pixels in sequence, that is, a column of pixels perpendicular to the ground, calculate the mean grayscale value of the selected column of pixels and compare it with the mean grayscale value of the next column of pixels, when the difference in the mean is greater than a preset threshold a, mark the straight line between the two columns of pixels as a pending edge line, mark the first pending edge line as the start line, continue to detect until the other side of the image edge, and mark the last pending edge line as the end line;

[0017] When the number of pending edge lines is greater than 2, the pending edge lines within a distance of m pixels near the start line are obtained and marked as the start line group, where m is a preset value. The pending edge lines within m pixels near the end line are marked as the end line group. The internal longitudinal pixel value detection is performed on each pending edge line in the start line group and the end line group respectively. The pixel value of the current pixel point in a single pending edge line is compared with the pixel value of the next point in turn. The segmentation point between two pixel points whose difference is greater than a preset threshold a is marked as an edge point. All edge points in the start line group and the end line group are connected respectively, and the connected edge points are fitted with a straight line equation, and the fitted straight line is marked as an edge line.

[0018] As a further solution of the present invention: when the number of the pending edge lines is equal to 2, the two pending edge lines are directly marked as edge lines.

[0019] As a further solution of the present invention: when the number of edge points in a single pending edge line is greater than or equal to 2, the middle point position between the first edge point and the last edge point is marked as an edge point representing the pending edge line.

[0020] As a further solution of the present invention: when the angle difference between any edge line and the vertical line of the ground is greater than or equal to 5 degrees, the electrical equipment is marked as abnormally placed equipment.

[0021] As a further solution of the present invention: when the angle difference between the two edge lines is greater than or equal to 5 degrees, the electric power equipment is marked as an abnormal appearance equipment.

[0022] As a further solution of the present invention: in the defect detection module, feature extraction includes color features, texture features, shape features, spatial features, frequency features, statistical features and transform domain features;

[0023] Among them, color features include color distribution and grayscale value, texture features include texture distribution and texture contrast, shape features include edge contour and geometric shape, spatial features include position relationship and area segmentation, frequency features include high-frequency features and low-frequency features, statistical features include mean, variance and histogram features, and transform domain features include Fourier transform and Laplace transform.

[0024] Beneficial effects of the present invention:

[0025] The present invention provides an AI vision-based power equipment robot inspection system, which realizes low-cost and high-efficiency recognition of the edge of power equipment through image acquisition module, preprocessing, edge detection and defect detection module, and grayscale detection of image edge line, and improves image quality by using vortex filtering and contrast stretching technology for image preprocessing; edge line recognition is performed by grayscale conversion and pixel mean comparison method, which simplifies the calculation process; a defect detection model is established by using Transformer network, which improves the accuracy and robustness of defect detection; at the same time, by judging the slope of a straight line, automatic recognition of abnormal placement of power equipment is realized. The present invention not only reduces the cost and risk of manual inspection, but also greatly improves the inspection efficiency and accuracy, and provides strong support for the maintenance and management of power equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The present invention will be further described below in conjunction with the accompanying drawings.

[0027] Figure 1 It is a module schematic diagram of the present invention. DETAILED DESCRIPTION

[0028] 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.

[0029] See also Figure 1 As shown, the present invention is a power equipment robot inspection system based on AI vision, comprising:

[0030] Route planning module: This module is responsible for developing inspection routes to ensure that the robot can cover all power equipment areas that need to be inspected. It calculates the optimal inspection path based on the distribution of power equipment and terrain information, and sends these paths to the robot performing the inspection task. This step is the starting point of the entire inspection process and provides guidance for subsequent image acquisition and processing.

[0031] Image acquisition module: This module uses the image sensor (HD camera) installed on the robot to capture real-time images of power equipment as the robot moves along the planned route. The collected images may be affected by environmental factors such as light changes, weather conditions, etc., so preprocessing is required to improve image quality.

[0032] Image preprocessing module: In order to improve image quality and reduce the complexity of subsequent processing, this module performs a series of preprocessing operations on the acquired images. This may include denoising, contrast enhancement, color balance adjustment, etc. to ensure that the image is clear and suitable for further analysis.

[0033] Image processing module: At this stage, the system uses edge detection algorithms to identify the edge lines of the power equipment. By converting the image to grayscale and analyzing the changes in pixel values, the outline of the power equipment can be accurately determined. Once the edge lines are identified, the system can segment the power equipment area from the background based on these lines, thereby focusing on the object of interest.

[0034] Feature extraction module: For the segmented power equipment area, the feature extraction module extracts key feature information from it. These features may be shape, size, texture or other attributes that help distinguish different types of power equipment. Feature extraction is an important prerequisite for establishing a defect detection model.

[0035] Defect Detection Module: Finally, the defect detection module uses the information obtained from the feature extraction stage combined with a pre-trained Transformer network model to perform defect detection. This model is trained based on a large number of known defect samples and can identify anomalies or potential problems in the image. When a defect is detected, the system will generate a corresponding report indicating the location and nature of the problem.

[0036] In a preferred embodiment of the present invention, in the image acquisition module, the preprocessing process is:

[0037] 1. Vortex filter processing:

[0038] Firstly, the vortex filter technology is used to process the power equipment image in the frequency domain. The vortex filter is an efficient image denoising method, which is particularly suitable for removing high-frequency noise in images while retaining important details and edge information of the image.

[0039] By converting the image into the frequency domain, the frequency components in the image can be analyzed and processed more intuitively, thereby suppressing noise in a targeted manner and improving the signal-to-noise ratio of the image.

[0040] 2. Contrast stretching:

[0041] After denoising, the power equipment image is subjected to contrast stretching. Contrast stretching is an image enhancement technology that adjusts the grayscale distribution of the image to make the details of the image clearer and the contrast more distinct.

[0042] Specifically, the contrast of the image is stretched to a set value, which is determined based on actual application requirements and image characteristics to ensure that the image has good visual recognizability and layering.

[0043] 3. Geometric distortion correction:

[0044] Special correction processing is performed for possible geometric distortion in power equipment images, especially distortion in the edge area of ​​the image. These distortions may be caused by factors such as the installation angle of the image sensor, lens distortion or shooting environment.

[0045] The definition of the image edge area is precisely set according to the focal length of the image sensor. By analyzing the relationship between the focal length and the image edge distortion, the area that needs to be corrected can be determined.

[0046] Using a geometric correction algorithm, the distortion of the edge areas of the image is accurately corrected to ensure that all elements in the image maintain their original shape and proportion, thereby improving the overall quality and accuracy of the image.

[0047] In another preferred embodiment of the present invention, the process of the image processing module identifying the edge line is:

[0048] 1. Grayscale conversion:

[0049] First, the original power equipment image is converted into a grayscale image. This step simplifies the subsequent processing flow because the grayscale image only contains brightness information and removes color information, making the subsequent edge detection more direct and efficient.

[0050] 2. Column pixel selection and gray value calculation:

[0051] From the grayscale image, starting from one side edge in the direction perpendicular to the ground, select a column of pixels in sequence. A column is a column of pixels in the direction perpendicular to the ground. For each column of pixels, calculate the mean of its grayscale value.

[0052] The grayscale mean of the current column is compared with the grayscale mean of the next column. If the grayscale mean difference between the two columns is greater than the preset threshold a, it is considered that there is a potential edge change between the two columns of pixels, so the straight line between the two columns of pixels is marked as a "pending edge line".

[0053] 3. Determination of the starting line and the ending line:

[0054] On one side of the image (usually the left or right side), the first marked potential edge line is defined as the "starting line".

[0055] Continue to detect in the above way until the other edge of the image is found. The last marked pending edge line is defined as the "end line".

[0056] 4. The formation of the starting line group and the ending line group:

[0057] When the number of detected pending edge lines is greater than 2, it means that the edge is not obvious due to the shooting background or the device installation position.

[0058] Within a distance of m pixels (m is a preset value) near the “starting line”, all pending edge lines are marked as “starting line group”.

[0059] Similarly, within a distance of m pixels near the “end line”, all pending edge lines are marked as “end line group”.

[0060] 5. Vertical pixel value detection and edge point marking:

[0061] The internal longitudinal pixel value detection is performed on each pending edge line in the “starting line group” and the “ending line group”.

[0062] For a single undetermined edge line, the pixel value of the current pixel point is compared with the pixel value of the next point in turn. If the difference between the two pixel points is greater than the preset threshold a, the segmentation point between the two pixel points is marked as an "edge point".

[0063] 6. Edge point connection and straight line equation fitting:

[0064] Connect all edge points in the "starting line group" and "ending line group" respectively to form continuous edge segments.

[0065] The connected edge points are fitted with a straight line equation to obtain a smoother and more accurate edge line. The fitted straight line is marked as an “edge line”.

[0066] In a preferred case of the present embodiment, in the image, if only two pending edge lines are detected, and the distance and grayscale contrast between the two lines are significant enough, then they have accurately represented the main edge structures in the image. Therefore, the two pending edge lines are directly marked as edge lines.

[0067] In another preferred case of this embodiment, when the number of edge points in a single pending edge line is greater than or equal to 2, the middle point position between the first edge point and the last edge point is directly marked as the edge point representing the pending edge line.

[0068] It is worth noting that when the angle difference between any edge line and the vertical line of the ground is greater than or equal to 5 degrees, it means that there is a large tilt on the side of the equipment, and the electrical equipment is marked as an abnormally placed device.

[0069] When the angle difference between the two edge lines is greater than or equal to 5 degrees, it means that deformation has occurred on both sides of the device, and the electrical equipment is marked as a device with abnormal appearance.

[0070] In another preferred embodiment of the present invention, in the defect detection module, feature extraction includes color features, texture features, shape features, spatial features, frequency features, statistical features and transform domain features;

[0071] Color feature: This is one of the most intuitive and basic features, including color distribution uniformity analysis and grayscale value statistics. Color distribution can help identify abnormal color spots or discolored areas in the image, while grayscale value analysis helps highlight those defects that are inconsistent with the surrounding environment in brightness.

[0072] Texture features: Texture is a visual characteristic of the surface of an object, reflecting the reflection and absorption of light on the surface of the object. By analyzing the regularity of texture distribution and the strength of texture contrast, it is possible to effectively distinguish between the fine structure of a normal device surface and texture mutations caused by defects, such as cracks, rust, etc.

[0073] Shape features: Shape is an important parameter that describes the appearance of an object, including the clarity of the edge contour and the regularity of the geometric shape. By detecting the edge contour, physical damage or deformation of the device can be identified; while geometric shape analysis helps to find structural anomalies that do not meet design specifications.

[0074] Spatial features: These features focus on the relative positional relationship between the components in the image and the rationality of the regional segmentation. For example, by analyzing whether the spacing between components meets the standard, or whether there are foreign objects that should not appear in a certain area, it can be determined whether there are problems such as installation errors or foreign object interference in the equipment.

[0075] Frequency features: Based on the perspective of signal processing, the image is converted from the spatial domain to the frequency domain for analysis. High-frequency features correspond to the rapidly changing parts of the image, such as edges and detail information, while low-frequency features reflect the overall trend and smooth areas of the image. This analysis helps to reveal subtle flaws that are not easily noticeable in the spatial domain.

[0076] Statistical features: Through statistical analysis of image pixel values, we can extract indicators such as mean (reflecting average brightness), variance (measurement of the discreteness of data distribution) and histogram features (displaying pixel value distribution). These statistics can quantitatively describe the overall characteristics and local differences of the image, providing a numerical basis for defect detection.

[0077] Transform domain features: Mathematical transformation tools such as Fourier transform and Laplace transform are used to transform the image from the original coordinate system to a new coordinate system for analysis. Fourier transform is good at spectrum analysis and can reveal the frequency composition of the image; while Laplace transform is often used for edge detection, strengthening the high-frequency components of the image and making the edges more prominent. These transform domain features provide strong support for in-depth analysis and precise positioning of defects.

[0078] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A robot inspection system for power equipment based on AI vision, characterized in that: include: The route planning module is used to plan the inspection route according to the power equipment area to be inspected and send the inspection route to the robot; An image acquisition module, used for acquiring images of the electric power equipment to be inspected based on an image sensor mounted on the robot, and preprocessing the images of the electric power equipment; An image processing module is used to perform edge detection on the preprocessed power equipment image, identify the edge line of the power equipment, and segment the power equipment area image from the background area according to the edge line; A feature extraction module, used for extracting features from the power equipment area image; A defect detection module, used to establish a defect detection model based on a defect image database and through a Transformer network, input the power equipment area image into the defect detection model, and output a corresponding defect recognition result; The process of the image processing module identifying the edge line is as follows: Convert the power equipment image into a grayscale image, start from the edge of one side of the grayscale image perpendicular to the ground, select a column of pixels in sequence, that is, a column of pixels perpendicular to the ground, calculate the mean grayscale value of the selected column of pixels and compare it with the mean grayscale value of the next column of pixels, when the difference in the mean is greater than a preset threshold a, mark the straight line between the two columns of pixels as a pending edge line, mark the first pending edge line as the start line, continue to detect until the other side of the image edge, and mark the last pending edge line as the end line; When the number of pending edge lines is greater than 2, the pending edge lines within a distance of m pixels near the start line are obtained and marked as the start line group, where m is a preset value. The pending edge lines within m pixels near the end line are marked as the end line group. The internal longitudinal pixel value detection is performed on each pending edge line in the start line group and the end line group respectively. The pixel value of the current pixel point in a single pending edge line is compared with the pixel value of the next point in turn. The segmentation point between two pixel points whose difference is greater than a preset threshold a is marked as an edge point. All edge points in the start line group and the end line group are connected respectively, and the connected edge points are fitted with a straight line equation, and the fitted straight line is marked as an edge line.

2. According to claim 1, the power equipment robot inspection system based on AI vision is characterized in that: In the image acquisition module, the preprocessing process is: The power equipment image is processed in the frequency domain by vortex filtering, and the contrast of the power equipment image is stretched to a set value, and the geometric distortion in the edge area of ​​the image is corrected. The definition of the edge area of ​​the image is set according to the focal length of the image sensor.

3. The AI ​​vision-based power equipment robot inspection system according to claim 1 is characterized in that: When the number of pending edge lines is equal to 2, the two pending edge lines are directly marked as edge lines.

4. The AI ​​vision-based power equipment robot inspection system according to claim 1 is characterized in that: When the number of edge points in a single pending edge line is greater than or equal to 2, the middle point position between the first edge point and the last edge point is marked as an edge point representing the pending edge line.

5. The AI ​​vision-based power equipment robot inspection system according to claim 1 is characterized in that: When the angle difference between any edge line and the vertical line of the ground is greater than or equal to 5 degrees, the electrical equipment will be marked as abnormally placed equipment.

6. The AI ​​vision-based power equipment robot inspection system according to claim 1 is characterized in that: When the angle difference between the two edge lines is greater than or equal to 5 degrees, the electrical equipment is marked as an abnormal appearance device.

7. The AI ​​vision-based power equipment robot inspection system according to claim 1 is characterized in that: In the defect detection module, feature extraction includes color features, texture features, shape features, spatial features, frequency features, statistical features and transform domain features; Among them, color features include color distribution and grayscale value, texture features include texture distribution and texture contrast, shape features include edge contour and geometric shape, spatial features include position relationship and area segmentation, frequency features include high-frequency features and low-frequency features, statistical features include mean, variance and histogram features, and transform domain features include Fourier transform and Laplace transform.

Images