Electrical equipment installation inclination degree identification method and system, medium and equipment

Abstract

The invention relates to the technical field of image processing, and provides a power equipment installation inclination degree identification method and system, a medium and equipment, and the method comprises the steps: obtaining a power equipment inspection image, and carrying out the segmentation of a power equipment box through a deep learning instance segmentation method, and obtaining a box mask image; based on the box body mask image, carrying out box body minimum enclosing rectangle fitting by adopting an equal-interval rotation search method to obtain an inclination angle of the box body; wherein the equal-interval rotation searching method comprises the following steps: continuously rotating a box mask image for multiple times in the same direction, increasing the rotation angle at equal intervals, recording the minimum enclosing rectangle of the box mask image in each rotation, and calculating the area of the minimum enclosing rectangle obtained in each rotation; and finally, the minimum enclosing rectangle corresponding to the minimum area is selected as the minimum enclosing rectangle obtained by fitting, and the rotation angle of the minimum enclosing rectangle obtained by fitting is taken as the inclination angle of the box body, so that the calculation amount and the calculation complexity are effectively reduced.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method, system, medium and device for recognizing the tilt degree of power equipment installation. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The installation quality of power equipment is directly related to the safe and stable operation of the power grid. However, traditional methods for inspecting the installation quality of power equipment usually rely on manual inspections. This method is not only inefficient, but also easily affected by human factors, resulting in low accuracy of the inspection results.

[0004] Currently, a line detection method based on Hough transform exists for identifying the tilt of installed power equipment. This method acquires an image of the power equipment and detects straight lines along its edges to determine if it is tilted. However, this method has the following drawbacks: (1) Complexity and computational cost: The Hough transform requires traversing every pixel in the image and voting in the parameter space, which is computationally expensive, especially in high-resolution images; in addition, in order to accurately detect lines, multiple parameters may need to be adjusted, which increases the complexity of the line detection method. (2) Limited adaptability: Hough transform is mainly suitable for detecting objects with obvious straight line features. However, the shape and installation of power equipment may vary depending on the actual environment. If the straight line features are not obvious or there is interference, the Hough transform method may not work effectively, leading to misjudgment. (3) Environmental factors have a great influence: The Hough transform method relies on the detection of straight line features in the image. However, these features are easily affected by environmental factors such as changes in light, shadow interference, and dirt occlusion, which leads to inaccurate detection of straight lines and thus affects the judgment of the degree of tilt. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a method, system, medium, and device for identifying the tilt degree of power equipment installation. After segmenting the enclosure mask, the minimum bounding rectangle of the enclosure is fitted using an equally spaced rotation search method. The tilt angle is then automatically calculated by fitting the minimum bounding rectangle of the power equipment, eliminating the need to traverse every pixel in the image or adjust multiple parameters, thus effectively reducing the computational load and complexity.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for identifying the tilt degree of power equipment installation, comprising: Acquire images of power equipment during inspection, and use a deep learning instance segmentation method to segment the power equipment enclosure to obtain the enclosure mask image; Based on the box mask image, an equal-interval rotation search method is used to fit the minimum bounding rectangle of the box to obtain the tilt angle of the box. The equal-interval rotation search method refers to rotating the box mask image continuously multiple times in the same direction, with the rotation angle increasing at equal intervals. Each rotation records the minimum bounding rectangle of the box mask image, and the area of ​​the minimum bounding rectangle obtained in each rotation is calculated. Finally, the minimum bounding rectangle corresponding to the smallest area is selected as the fitted minimum bounding rectangle, and the rotation angle of the fitted minimum bounding rectangle is used as the tilt angle of the box.

[0007] Furthermore, a method for identifying the tilt degree of power equipment installation also includes: calculating the area ratio of the box mask image and the fitted minimum bounding rectangle, and using the area ratio as the confidence level of the tilt angle; if the confidence level is less than a threshold, it is determined that the power equipment inspection image acquisition is non-compliant.

[0008] Furthermore, a method for identifying the tilt degree of power equipment installation also includes: if the confidence level is greater than a threshold, comparing the tilt angle of the enclosure with the angle threshold to determine whether the power equipment installation is up to standard.

[0009] Furthermore, the deep learning instance segmentation method includes: Brightness is adjusted for power equipment inspection images, and features are extracted and recalibrated through a backbone network and a channel attention mechanism. Multi-scale context features are extracted in parallel using dilated convolutions with different dilation rates. The multi-scale context features are adaptively fused using a dynamically generated gating weight matrix to obtain multi-level fused features. For multi-level fusion features, global context enhancement features at each level are obtained through multi-level feature fusion. Based on the global context enhancement features at each level, the box mask image is obtained through a background-aware gating mechanism.

[0010] This invention employs dilated convolutions with different dilation rates to extract multi-scale features in parallel. By adaptively fusing dynamically generated gating weight matrices, the importance of features at different scales can be balanced, thereby improving segmentation accuracy.

[0011] Furthermore, the specific steps of the brightness adjustment include: after converting the power equipment inspection image from RGB space to HSV space, dividing it into three regions according to brightness: highlights, midtones, and shadows; performing nonlinear compression on the highlight region, applying S-curve transformation to the midtone region, and performing adaptive Gamma correction on the shadow region before converting it back to RGB space.

[0012] This invention can effectively solve the problem of blurred equipment edges caused by uneven lighting (such as strong light reflection and shadow occlusion) in inspection images by adjusting the brightness of different regions, thereby improving the accuracy of subsequent power equipment inspection image segmentation.

[0013] Furthermore, the multi-level feature fusion includes: aligning the multi-level fused features and then concatenating them to obtain concatenated features; enhancing the aligned multi-level fused features with local window self-attention and then fusing them with the concatenated features to obtain global context-enhanced features at each level.

[0014] This invention enhances multi-level features through a local window self-attention mechanism, which can capture the contextual relationship between the box and the background, reduce background noise interference, and make the segmentation mask more closely match the actual contour.

[0015] Furthermore, the background-aware gating mechanism includes: performing bilinear upsampling on the global context enhancement features at each level, followed by splicing and compression operations; generating a mask by applying spatial attention to the compressed features; and generating a box mask image by hard thresholding binarization of the mask.

[0016] This invention uses upsampling and splicing of multi-level features to generate a spatial attention mask, which can focus on the box area and suppress background interference. Hard threshold binarization further ensures the clarity of the mask edges.

[0017] A second aspect of the present invention provides a system for identifying the tilt level of power equipment installation, comprising: The image segmentation module is configured to: acquire power equipment inspection images and use a deep learning instance segmentation method to segment the power equipment enclosure to obtain an enclosure mask image; The tilt angle calculation module is configured to: based on the box mask image, use an equal-interval rotation search method to fit the minimum bounding rectangle of the box to obtain the tilt angle of the box; wherein, the equal-interval rotation search method refers to performing multiple consecutive rotations of the box mask image in the same direction, with the rotation angle increasing at equal intervals, recording the minimum bounding rectangle of the box mask image in each rotation, calculating the area of ​​the minimum bounding rectangle obtained in each rotation, and finally selecting the minimum bounding rectangle corresponding to the minimum area as the fitted minimum bounding rectangle, and using the rotation angle of the fitted minimum bounding rectangle as the tilt angle of the box.

[0018] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for identifying the tilt level of an electrical equipment installation.

[0019] A fourth aspect of the present invention provides a computer device including a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for identifying the tilt of an electrical equipment installation.

[0020] Compared with the prior art, the beneficial effects of the present invention are: This invention innovatively proposes an automated tilt angle calculation method for power equipment. After segmenting the box mask, the method uses an equal-interval rotation search to fit the minimum bounding rectangle of the box. The tilt angle is then automatically calculated by fitting the minimum bounding rectangle of the power equipment. This eliminates the need to traverse every pixel in the image or perform multi-parameter adjustments, effectively reducing the amount of computation and computational complexity.

[0021] This invention innovatively proposes an image acquisition compliance judgment method. It determines the confidence level of the tilt angle of power equipment by calculating the area ratio of the enclosure in the smallest bounding rectangle, and judges whether the image acquisition is compliant by setting a confidence level threshold. This avoids misjudgment due to image quality issues and improves adaptability.

[0022] This invention innovatively proposes a deep learning instance segmentation method, which performs refined positioning and identification of power equipment enclosures through deep learning instance segmentation. This effectively removes interference from background information, improves the accuracy of power equipment identification, and can adapt to power equipment inspection images in different scenarios. It can effectively identify and judge regardless of changes in lighting, shadows, or the angle and size of the enclosure. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart of a method for identifying the tilt level of power equipment installation according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of the precision positioning of the metering box according to Embodiment 1 of the present invention; Figure 3 This is a diagram showing the refined positioning result of the metering box in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the minimum circumscribed rectangle fitting of the metering box body according to Embodiment 1 of the present invention; Figure 5 This is a schematic diagram showing the proportion of the box body in the smallest bounding rectangle according to Embodiment 1 of the present invention; Figure 6 This is a schematic diagram illustrating the estimation of the tilt degree of the metering box according to Embodiment 1 of the present invention; Figure 7 This is a schematic diagram of the structure of a computer device according to Embodiment 4 of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0026] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] Example 1 This embodiment provides a method for identifying the tilt level of power equipment installation.

[0028] This embodiment provides a method for identifying the tilt level of power equipment installation, which aims to improve detection efficiency and accuracy, while enabling remote monitoring and real-time response.

[0029] In this embodiment, a metering box, an electrical device, is used as an example for explanation.

[0030] This embodiment provides a method for identifying the tilt level of installed power equipment, such as... Figure 1 As shown, it includes the following steps: Step 1: Obtain the inspection image of the metering box (also known as the original metering box image).

[0031] Step 2: Precise positioning of the metering box.

[0032] For an inspection image of a metering box, fine-grained positioning and identification of the metering box in the image can effectively remove interference from background information.

[0033] In this embodiment, the refined positioning and recognition uses a deep learning instance segmentation method to segment the meter box, which can obtain a mask image of the box.

[0034] In this embodiment, the metering box is precisely positioned, such as... Figure 2 As shown, it includes the following steps: Step 201: Inspection images of the metering box Apply dynamic range brightness adjustment. Specifically, this includes: (1) Input image Image obtained by converting from RGB space to HSV space. ; (2) For the image The system is divided into three regions based on brightness: highlights, midtones, and shadows. Nonlinear mapping is applied to each region: nonlinear compression is performed on highlight regions that exceed the threshold to preserve metallic reflection details; S-curve transformation is applied to midtone light regions to equalize brightness and improve contrast; and adaptive Gamma correction is performed on shadow regions to enhance shadow details. It can be represented as: ; in, , , , , , , The average value of the shaded area. The standard deviation is the shaded area. Representing an image One pixel in the image. Representing an image The maximum brightness value in; (3) Then, the adjusted image Convert back to RGB space to obtain the brightness-adjusted image (i.e., the brightness-adjusted image of the metering box). .

[0035] Step 202: The backbone network extracts image features and enhances image features through multi-scale contextual interaction.

[0036] Specifically, brightness adjustment image Multi-level feature maps were extracted from the backbone network. ( After that, the features are first recalibrated using a channel attention mechanism to suppress channel responses irrelevant to the target, thus obtaining the features. Subsequently, dilated convolutions with dilation rates of 1, 3, and 5 were used in parallel to extract multi-scale contextual features. Furthermore, adaptive fusion is achieved through dynamically generated gating weight matrices to obtain multi-level fused features. .in, This represents the number of channels in the feature map, corresponding to the dimension of the extracted semantic features; and These represent the height and width of the feature map, respectively, corresponding to the resolution of the image in the spatial dimension.

[0037] This can be expressed as: ;in, Indicates global max pooling. , It is a fully connected layer. The compression ratio is... For the sigmoid function, Initialize to 1 and optimize during training; ;in, The tensor is represented by the tensor rate. 3×3 convolution; ;in, For learnable parameters, Calculated along the channel dimension, This indicates multiplication by channel.

[0038] Step 203: Multi-level feature fusion to improve the accuracy of metering box segmentation task.

[0039] In this embodiment, multi-level fusion features are fused with progressive features through a hybrid attention mechanism. This retains the advantages of the U-shaped structure while introducing global-local collaborative modeling, improving the edge accuracy of the bin segmentation task. Specifically, this includes: (1) First, multi-level features Feature alignment is performed to obtain Then spliced ​​together : ; ; in, This represents a convolution operation with a 1×1 kernel. This indicates a splicing operation along the channel dimension.

[0040] (2) Then, local window self-attention is used to enhance multi-level features: ; in, , , These correspond to the linear projections of the attention mechanism query, key, and value, respectively. For channel dimensions.

[0041] (3) Finally, compensation vectors are generated through global average pooling to enhance the global context features at each level: .

[0042] in, This indicates that global average pooling is performed on the input feature map along the spatial dimension; This represents a multilayer perceptron.

[0043] Step 204: Background-aware gating mechanism, which generates a spatial attention mask through a lightweight MLP to dynamically distinguish between foreground and background regions.

[0044] Specifically, the input consists of global context enhancement features at each level. Bilinear upsampling is used to align the resolutions of global context enhancement features across all levels, followed by concatenation and compression. Spatial attention is then used to generate a mask. ,Right now: ; in, Indicates that the core is Convolution operation, For upsampling operation, This represents the Sigmoid activation function.

[0045] Then, a gating matrix is ​​generated by hard thresholding binarization to suppress the background region and generate a mask image of the metering box: .

[0046] The final generated metering box mask image is as follows: Figure 3 As shown.

[0047] Step 3: Fitting the minimum circumscribed rectangle structure of the metering box.

[0048] Based on the binary image of the box mask identified in step 2, a method of equally spaced rotation search is used to fit the minimum bounding rectangle of the box. The binary image of the box mask is rotated at equal intervals within a 90° range, and the bounding rectangle parameters of its contour in the coordinate system direction are recorded each time. The minimum bounding rectangle is obtained by calculating the area of ​​the bounding rectangle. Figure 4 As shown.

[0049] The specific steps are as follows: Step 301: By calculating the maximum and minimum values ​​of the box distribution coordinates in the binary image of the measuring box mask, obtain the circumscribed rectangle of the box in the current image. Record the length, width, and area of ​​the circumscribed rectangle. Obtain the minimum circumscribed rectangle RectMin and assign its area value to the variable AreaMin. At this time, the rotation angle is... ; Step 302: Rotate the binary mask image by an angle △x, and according to step 301, find the minimum bounding rectangle RectTmp of the rotated box region (binary mask image), and calculate the area of ​​the minimum bounding rectangle, denoted as AreaTmp. Step 303: Rotate the angle at this point. Compare the sizes of AreaTmp and AreaMin, assign the smaller area value to AreaMin, and assign the rotation angle at this point to AreaMin. x The rectangle information is assigned to RectMin; Step 304: Increment j by 1, and repeat steps 302 and 303. Finally, we can obtain a minimum bounding rectangle RectMin and its corresponding rotation angle. x ; Step 305: Rotate RectMin in the opposite direction x The angle is the smallest bounding rectangle of the box in the binary image of the metering box mask. x This refers to the tilt angle of the box.

[0050] Step 4: Estimate the tilt of the measuring box.

[0051] Step 401: Calculate the confidence level of the box tilt angle.

[0052] Based on the minimum bounding rectangle of the box obtained in step 3, the proportion of the box's area within the minimum bounding rectangle can be calculated. The larger the proportion of the box's area within the minimum bounding rectangle, the smaller the deformation of the box in the image, and the higher the confidence level of the calculated box tilt angle. The specific calculation steps are as follows: (1) Based on the minimum bounding rectangle obtained in step 3, a binary image of the minimum bounding rectangle mask of the box can be generated; (2) Given the binary mask image of the box and the binary mask image of the fitted minimum bounding rectangle, the area of ​​the box and the area of ​​the minimum bounding rectangle can be calculated. Then, the proportion of the box in the minimum bounding rectangle can be determined. ; (3) If R=1, it means that the box has no deformation in the image and the box is tilted at an angle. x The confidence level is conf=1; R<1 indicates that the box is deformed in the image. The R value represents the degree of deformation. The smaller the R value, the greater the deformation and the box tilt angle. x The lower the confidence level, such as Figure 5 As shown.

[0053] Step 402: Estimation of the tilt of the metering box.

[0054] Based on the tilt angle of the metering box and its confidence level, the degree of tilt of the metering box is estimated as follows: (1) Set a confidence threshold thre, compare the confidence with the threshold, and if the confidence is less than the threshold thre, it is considered that the box has a large deformation in the image, and the image acquisition is deemed non-compliant, and an image non-compliant error is returned. (2) If the confidence level is not less than the set threshold, then the tilt angle is determined, and the angle threshold is set according to the actual business needs. and , ; (3) Compare the tilt angles x With threshold and The size between, if x < Then determine the meter box installation specifications; if Then it is determined that the metering box is installed at a slight tilt; if Then it is determined that the metering box is installed at a serious angle, such as Figure 6 As shown.

[0055] This embodiment provides a method for identifying the tilt level of power equipment installation, which has the following characteristics: High-precision positioning and identification: The deep learning instance segmentation method is used to perform fine positioning and identification of power equipment enclosures, which effectively removes the interference of background information and improves the accuracy of power equipment identification; High degree of automation: The entire identification process, including box positioning, minimum bounding rectangle fitting, tilt angle calculation and tilt degree estimation, is automated and requires no manual intervention, which greatly improves inspection efficiency. Highly adaptable: It can adapt to power equipment inspection images in different scenarios, and can effectively identify and judge changes in lighting, shadows, or the angle and size of the enclosure. Accurate tilt estimation: By calculating the area ratio of the enclosure in the smallest bounding rectangle, i.e. the confidence level of the enclosure tilt angle, and combining it with the actual value of the tilt angle, the tilt degree of the power equipment can be accurately estimated, providing strong support for subsequent maintenance and management. Self-verification: By setting a confidence threshold, it can intelligently determine whether image acquisition is compliant, avoiding misjudgments caused by image quality issues; Flexible business adaptability: Based on actual business needs, angle thresholds can be flexibly set to classify and manage power equipment with different degrees of tilt, providing clear guidance for inspection personnel; Highly scalable: It is not only applicable to the tilt identification of metering boxes, but can also be extended to the status identification and inspection of other power equipment, showing good application prospects.

[0056] Example 2 This embodiment provides a power equipment installation tilt recognition system, including: The image segmentation module is configured to: acquire power equipment inspection images and use a deep learning instance segmentation method to segment the power equipment enclosure to obtain an enclosure mask image; The tilt angle calculation module is configured to: based on the box mask image, use an equal-interval rotation search method to fit the minimum bounding rectangle of the box to obtain the tilt angle of the box; wherein, the equal-interval rotation search method refers to performing multiple consecutive rotations of the box mask image in the same direction, with the rotation angle increasing at equal intervals, recording the minimum bounding rectangle of the box mask image in each rotation, calculating the area of ​​the minimum bounding rectangle obtained in each rotation, and finally selecting the minimum bounding rectangle corresponding to the minimum area as the fitted minimum bounding rectangle, and using the rotation angle of the fitted minimum bounding rectangle as the tilt angle of the box; The image compliance judgment module is configured to: calculate the area ratio of the box mask image and the minimum bounding rectangle obtained by fitting, and use the area ratio as the confidence level of the tilt angle; if the confidence level is less than the threshold, it is judged that the power equipment inspection image acquisition is non-compliant. The installation compliance judgment module is configured to: if the confidence level is greater than the threshold, compare the tilt angle of the enclosure with the angle threshold to determine whether the installation of the power equipment is in compliance with regulations.

[0057] Furthermore, deep learning instance segmentation methods include: Brightness is adjusted for power equipment inspection images, and features are extracted and recalibrated through a backbone network and a channel attention mechanism. Multi-scale context features are extracted in parallel using dilated convolutions with different dilation rates. The multi-scale context features are adaptively fused using a dynamically generated gating weight matrix to obtain multi-level fused features. For multi-level fusion features, global context enhancement features at each level are obtained through multi-level feature fusion. Based on the global context enhancement features at each level, the box mask image is obtained through a background-aware gating mechanism.

[0058] Furthermore, the specific steps for brightness adjustment include: after converting the power equipment inspection image from RGB space to HSV space, dividing it into three regions according to brightness: highlights, midtones, and shadows; performing nonlinear compression on the highlight region, applying S-curve transformation to the midtone region, and performing adaptive Gamma correction on the shadow region before converting it back to RGB space.

[0059] Furthermore, the multi-level feature fusion includes: aligning the multi-level fused features and then concatenating them to obtain concatenated features; then, after enhancing the aligned multi-level fused features with local window self-attention, fusing them with the concatenated features to obtain global context-enhanced features at each level.

[0060] Furthermore, the background-aware gating mechanism includes: applying bilinear upsampling to the global context enhancement features at each level, then performing splicing and compression operations, and applying spatial attention to the compressed features to generate a mask; for the mask, generating a box mask image through hard thresholding binarization.

[0061] It should be noted that each module in this embodiment corresponds one-to-one with each step in Embodiment 1, and their specific implementation processes are the same, so they will not be repeated here.

[0062] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the method for identifying the tilt level of power equipment installation as described in Embodiment 1 above.

[0063] Example 4 This embodiment provides a computer device, such as... Figure 7 As shown, the device includes a computer-readable storage medium 1003, a processor 1001, a communication interface 1002, and a computer program stored on the computer-readable storage medium 1003 and executable on the processor 1001. The processor 1001, communication interface 1002, and computer-readable storage medium 1003 can be connected via a bus or other means. The communication interface 1002 is used to receive and send data. When the processor 1001 executes the program, it implements the steps in the power equipment installation tilt identification method described in Embodiment 1 above.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the tilt degree of power equipment installation, characterized in that, include: Acquire images of power equipment during inspection, and use a deep learning instance segmentation method to segment the power equipment enclosure to obtain the enclosure mask image; Based on the box mask image, an equal-interval rotation search method is used to fit the minimum bounding rectangle of the box to obtain the tilt angle of the box. The equal-interval rotation search method refers to rotating the box mask image continuously multiple times in the same direction, with the rotation angle increasing at equal intervals. Each rotation records the minimum bounding rectangle of the box mask image, and the area of ​​the minimum bounding rectangle obtained in each rotation is calculated. Finally, the minimum bounding rectangle corresponding to the smallest area is selected as the fitted minimum bounding rectangle, and the rotation angle of the fitted minimum bounding rectangle is used as the tilt angle of the box.

2. The method for identifying the tilt level of power equipment installation as described in claim 1, characterized in that, Also includes: Calculate the area ratio of the box mask image and the fitted minimum bounding rectangle, and use the area ratio as the confidence level of the tilt angle; If the confidence level is less than the threshold, the image acquisition for power equipment inspection is deemed non-compliant.

3. The method for identifying the tilt level of power equipment installation as described in claim 2, characterized in that, Also includes: If the confidence level is greater than the threshold, the tilt angle of the enclosure is compared with the angle threshold to determine whether the installation of the power equipment is up to standard.

4. The method for identifying the tilt level of power equipment installation as described in claim 1, characterized in that, The deep learning instance segmentation method includes: Brightness is adjusted for power equipment inspection images, and features are extracted and recalibrated through a backbone network and a channel attention mechanism. Multi-scale context features are extracted in parallel using dilated convolutions with different dilation rates. The multi-scale context features are adaptively fused using a dynamically generated gating weight matrix to obtain multi-level fused features. For multi-level fusion features, global context enhancement features at each level are obtained through multi-level feature fusion. Based on the global context enhancement features at each level, the box mask image is obtained through a background-aware gating mechanism.

5. The method for identifying the tilt degree of power equipment installation as described in claim 4, characterized in that, The specific steps of the brightness adjustment include: after converting the power equipment inspection image from RGB space to HSV space, dividing it into three regions according to brightness: highlights, midtones and shadows; performing nonlinear compression on the highlight region, applying S-curve transformation to the midtone region, and performing adaptive Gamma correction on the shadow region before converting it back to RGB space.

6. The method for identifying the tilt level of power equipment installation as described in claim 4, characterized in that, The multi-level feature fusion includes: aligning the multi-level fused features and then concatenating them to obtain concatenated features; then strengthening the aligned multi-level fused features with local window self-attention and fusing them with the concatenated features to obtain global context-enhanced features at each level.

7. The method for identifying the tilt level of power equipment installation as described in claim 4, characterized in that, The background-aware gating mechanism includes: performing bilinear upsampling on the global context enhancement features at each level, followed by splicing and compression operations; generating a mask by applying spatial attention to the compressed features; and generating a box mask image by hard thresholding binarization of the mask.

8. A system for identifying the tilt level of power equipment installation, characterized in that, include: The image segmentation module is configured to: acquire power equipment inspection images and use a deep learning instance segmentation method to segment the power equipment enclosure to obtain an enclosure mask image; The tilt angle calculation module is configured to: based on the box mask image, use an equal-interval rotation search method to fit the minimum bounding rectangle of the box to obtain the tilt angle of the box; wherein, the equal-interval rotation search method refers to performing multiple consecutive rotations of the box mask image in the same direction, with the rotation angle increasing at equal intervals, recording the minimum bounding rectangle of the box mask image in each rotation, calculating the area of ​​the minimum bounding rectangle obtained in each rotation, and finally selecting the minimum bounding rectangle corresponding to the minimum area as the fitted minimum bounding rectangle, and using the rotation angle of the fitted minimum bounding rectangle as the tilt angle of the box.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the method for identifying the tilt of power equipment installation as described in any one of claims 1-7.

10. A computer device comprising a computer-readable storage medium, a processor, and a computer program stored on the computer-readable storage medium and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the method for identifying the installation tilt of power equipment as described in any one of claims 1-7.

Images