Transmission line sag detection method and system based on lightweight neural network model

Abstract

The present invention relates to the field of sag detection technology and provides a transmission line sag detection method and system based on a lightweight neural network model. The method comprises acquiring a transmission line image through aerial photography by an unmanned aerial vehicle and performing preprocessing to obtain a sag image; extracting multi-scale features of the sag image through an improved lightweight neural network model, constructing a multi-scale feature pyramid based on the multi-scale feature map of the sag image to output a multi-scale heat map; performing feature fusion on the multi-scale heat map through a dynamic weight fusion module to obtain a fused feature heat map; locating sag key points of the fused feature heat map through a heat map regression method, calculating the sag physical distance based on the sag key points, and correcting the sag physical distance in real time through a temperature compensation algorithm to obtain the actual sag distance. The present invention systematically solves the core bottleneck of sag measurement in power inspections and achieves fully automatic detection with high precision, high efficiency, and high robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of sag detection, and in particular to a transmission line sag detection method and system based on a lightweight neural network model. Background Art

[0002] Sag detection directly links the safety margin and load capacity of transmission line conductors by monitoring the sag of transmission line conductors. Accurate sag detection can effectively prevent excessive sag, which may cause discharge to the ground or across objects, leading to short circuits or fire risks. Excessive sag may also cause line disconnection risks under extreme conditions such as strong winds and icing. Real-time sag data can guide the power grid to dynamically adjust transmission capacity and avoid power loss caused by conductor overheating. Combined with meteorological data, abnormal changes in sag can provide early warning of the impact of environmental disasters such as icing and high temperatures on the line.

[0003] Transmission lines are often deployed in complex environments such as mountains, forests, and rain and fog. Low light, rain, snow, and vegetation obstruction can degrade image quality, making traditional algorithms susceptible to failure. When captured by high-altitude drones, the conductor width only occupies a few pixels in the image, making it difficult for conventional models to capture high-frequency details and resulting in a high rate of missed detections. Conductor sag fluctuates with temperature, affecting measurement accuracy.

[0004] Existing sag detection methods often rely on manual measurement or traditional image processing, which is inefficient and susceptible to environmental interference. Deep learning object detection models (such as Faster R-CNN) are computationally intensive and difficult to deploy on drone edge devices. Summary of the Invention

[0005] The present invention aims to address at least one of the technical problems existing in the related art. To this end, the present invention provides a transmission line sag detection method and system based on a lightweight neural network model. Through lightweight model optimization, multi-scale feature fusion, temperature compensation algorithm and other technologies, the present invention systematically solves the core bottleneck of sag measurement in power inspections, significantly reduces the model computing power, and achieves high-precision, high-efficiency, and high-robustness fully automatic detection.

[0006] The present invention provides a transmission line sag detection method based on a lightweight neural network model, comprising:

[0007] S1: Acquire transmission line images through drone aerial photography and preprocess the transmission line images to obtain sag images;

[0008] S2: Extract multi-scale features of the sag image through the improved MobileNetV3 lightweight neural network model to obtain the multi-scale feature map of the sag image;

[0009] S3: Construct a multi-scale feature pyramid based on the multi-scale feature map of the sag image, and predict the multi-scale feature map of the sag image through the multi-scale feature pyramid to obtain a multi-scale heat map;

[0010] S4: Perform feature fusion on the multi-scale heat map through the dynamic weight fusion module to obtain the fused feature heat map;

[0011] S5: The sag key points of the fused feature heat map are located by the heat map regression method, the sag physical distance is calculated based on the sag key points, and the sag physical distance is corrected in real time by the temperature compensation algorithm to obtain the actual sag distance.

[0012] According to the present invention, a transmission line sag detection method based on a lightweight neural network model further includes the preprocessing including insulator detection, segmented transformation and conductor characteristic enhancement.

[0013] The insulator detection method is to locate the coordinates of the top and bottom of the drone aerial image through the lightweight HRNet model to obtain the insulator detection image.

[0014] The segmented transformation is to divide the insulator detection image into N segments, independently calculate the transformation matrix for each segment, and correct each segment of the conductor area into a horizontal straight line using the transformation matrix;

[0015] The wire characteristic enhancement is to enhance the wire features in a low-contrast environment by using a CLAHE algorithm.

[0016] According to a transmission line sag detection method based on a lightweight neural network model provided by the present invention, the improved MobileNetV3 lightweight neural network model includes a shallow feature reuse module, a dynamic depth-separable convolution and an adaptive spatial attention module;

[0017] The shallow feature reuse module performs synchronous downsampling through the convolution path and the mean pooling path;

[0018] Dynamic depth-wise separable convolution automatically adjusts the depth of the convolution kernel according to the complexity of the input image;

[0019] An adaptive spatial attention module is inserted before the residual connection of each neck network of the MobileNetV3 model.

[0020] According to the present invention, a transmission line sag detection method based on a lightweight neural network model further includes obtaining multiple key layer features through an improved MobileNetV3 lightweight neural network model, constructing multiple feature layers based on the multiple key layer features, and obtaining a multi-scale feature pyramid;

[0021] Each key layer feature has a different resolution, and each feature layer detects and locates targets of different scales according to the resolution.

[0022] According to a transmission line sag detection method based on a lightweight neural network model provided by the present invention, the sag key points include the suspension point and the lowest point of the transmission line conductor.

[0023] According to a transmission line sag detection method based on a lightweight neural network model provided by the present invention, the method also includes the dynamic weight fusion module dynamically adjusting the multi-scale thermal map feature weight according to the height of the drone, and performing feature fusion on the multi-scale thermal map according to the multi-scale thermal map feature weight to obtain a fused feature thermal map.

[0024] According to the present invention, a transmission line sag detection method based on a lightweight neural network model further includes locating sag key points of a fused feature heat map by a heat map regression method, including:

[0025] S51: Obtain an optimized heat map by removing adjacent response peaks on the fusion feature heat map;

[0026] S52: Find the position with the largest response value on the optimization heat map, and obtain the coordinates of the sag key point on the heat map;

[0027] S53: Repeat step S52 to obtain the coordinates of multiple sag key points on the thermal map;

[0028] S54: The coordinates of the sag key points on the heat map are mapped back to the coordinate space of the drone aerial image by scaling to obtain the positions of the sag key points.

[0029] According to a transmission line sag detection method based on a lightweight neural network model provided by the present invention, the calculation expression of the sag physical distance is:

[0030]

[0031] in, is the sag physical distance, is the control coefficient, is the vertical pixel distance of the sag key point in the fusion feature heat map, is the horizontal tension of the conductor, is the field of view of the drone camera, is the pixel height of the tower, is the tangent function.

[0032] According to a transmission line sag detection method based on a lightweight neural network model provided by the present invention, the calculation expression of the actual sag distance is:

[0033]

[0034] in, is the actual sag distance, is the sag physical distance, is the conductor temperature expansion coefficient, The difference between the current temperature and the reference temperature.

[0035] The present invention also provides a transmission line sag detection system based on a lightweight neural network model, which is used to execute any of the above-mentioned transmission line sag detection methods based on a lightweight neural network model, comprising:

[0036] A sag image acquisition module, wherein the sag image acquisition module acquires a transmission line image through drone aerial photography and pre-processes the transmission line image to obtain a sag image;

[0037] A feature extraction module, wherein the feature extraction module extracts multi-scale features of the sag image through an improved MobileNetV3 lightweight neural network model to obtain a multi-scale feature map of the sag image;

[0038] A multi-scale heat map acquisition module, which constructs a multi-scale feature pyramid based on the multi-scale feature map of the sag image, predicts the multi-scale feature map of the sag image through the multi-scale feature pyramid, and obtains a multi-scale heat map;

[0039] A fusion module, which performs feature fusion on the multi-scale heat map through a dynamic weight fusion module to obtain a fused feature heat map;

[0040] The sag distance acquisition module locates the sag key points of the fused feature heat map through the heat map regression method, calculates the sag physical distance based on the sag key points, and corrects the sag physical distance in real time through the temperature compensation algorithm to obtain the actual sag distance.

[0041] The above one or more technical solutions in the embodiments of the present invention have at least one of the following technical effects:

[0042] Through technologies such as lightweight model optimization, multi-scale feature fusion, and temperature compensation algorithms, the core bottleneck of sag measurement in power inspections has been systematically solved, the model computing power has been significantly reduced, and fully automatic detection with high precision, high efficiency, and high robustness has been achieved.

[0043] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 It is a flow chart of a transmission line sag detection method based on a lightweight neural network model provided by the present invention.

[0046] Figure 2 It is a structural schematic diagram of the transmission line sag detection system based on the lightweight neural network model provided by the present invention.

[0047] Reference numerals:

[0048] 101. Sag image acquisition module; 102. Feature extraction module; 103. Multi-scale thermal map acquisition module; 104. Fusion module; 105. Sag distance acquisition module. DETAILED DESCRIPTION

[0049] To make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below. Obviously, the embodiments described are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0050] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0051] The following combination Figures 1 to 2 The present invention describes a transmission line sag detection method and system based on a lightweight neural network model.

[0052] like Figure 1As shown, a transmission line sag detection method based on a lightweight neural network model includes:

[0053] S1: Acquire transmission line images through drone aerial photography and preprocess the transmission line images to obtain sag images;

[0054] In some specific embodiments of the present invention, transmission line images are acquired by using a visible light camera mounted on a drone.

[0055] Transmission lines include high-voltage cables, substation transmission lines, photovoltaic power station transmission lines, and railway catenary transmission lines.

[0056] The drone's GPS synchronously records the latitude, longitude, and altitude of each frame, and corrects the pitch / yaw angles through a combined inertial navigation system. For example, an alarm is triggered when the pitch angle is >30°, and the shutter speed is dynamically adjusted according to the light intensity to ensure that the wire area is not overexposed.

[0057] Pre-processing includes insulator detection, segmentation transformation and conductor characteristic enhancement.

[0058] Insulator detection uses a lightweight HRNet model to locate the coordinates of the top and bottom of the drone aerial image to obtain the insulator detection image.

[0059] The segmented transformation is to divide the insulator detection image into N segments, independently calculate the transformation matrix for each segment, and correct each segment of the conductor area into a horizontal straight line using the transformation matrix;

[0060] Wire feature enhancement is to enhance the wire features in low-contrast environments through the Contrast Limited Adaptive Histogram Equalization algorithm (CLAHE algorithm).

[0061] Low contrast includes rain, fog, and light and shadow interference.

[0062] In some specific embodiments of the present invention, N=3, corresponding to the wire suspension section, the left tower section, and the right tower section.

[0063] S2: Extract multi-scale features of the sag image through the improved MobileNetV3 lightweight neural network model to obtain the multi-scale feature map of the sag image;

[0064] The improved MobileNetV3 lightweight neural network model includes a shallow feature reuse module, dynamic depthwise separable convolution, and an adaptive spatial attention module;

[0065] The shallow feature reuse module performs synchronous downsampling through the convolution path and the mean pooling path;

[0066] The convolution path extracts low-level semantic features of the sag image through 3×3 convolution and then activates it through the Hardswish activation function. The low-level semantic features include the edge of the wire and the texture of the insulator.

[0067] The calculation process of the Hardswish activation function is relatively simple, which can speed up the training and reasoning of the model and automatically adjust the activation level according to the input data.

[0068] The pooling path downsamples the sag image through 3×3 mean pooling with the same parameters as the convolution path;

[0069] Mean pooling retains more overall illumination information than maximum pooling, avoids size mismatch caused by skipping pooling directly, reduces the size of the feature map, reduces the amount of calculation, and retains the overall features in the area. Synchronous downsampling of the convolution path and the mean pooling path can comprehensively utilize the abstract features extracted by the convolution operation and the original information retained by the pooling operation, thereby improving the performance and feature expression ability of the model.

[0070] Dynamic depth-wise separable convolution automatically adjusts the depth of the convolution kernel according to the complexity of the input image;

[0071] Dynamic depth-wise separable convolution can automatically increase the convolution depth in areas with dense wires; Dynamic depth-wise separable convolution includes an inverted residual structure, which consists of three convolutional layers: the first ( ) convolutional layer is used for dimension increase, the middle depth convolutional layer is used for feature extraction, and the last one ( ) Convolutional layers are used for dimensionality reduction. Deep convolution is a core component of this architecture. In sag detection, deep convolution can extract local features from images of power transmission lines, such as line edges and textures.

[0072] An adaptive spatial attention module is inserted before the residual connection of each neck network of the MobileNetV3 model.

[0073] By inserting an adaptive spatial attention module before the residual connection of each neck network, the MobileNetV3 model can pay more attention to the key features of the transmission line, such as the edges and contours of the line, thereby improving the performance in the real-time detection of transmission line sag.

[0074] The adaptive spatial attention module is inserted before the residual connection of each neck network to use the attention mechanism to allow the model to focus on the key feature area and suppress useless information; the dynamic depth-wise separable convolution uses dynamic depth-wise separable convolution to dynamically adjust the convolution kernel parameters according to the input to better adapt to different features and enhance the model's feature extraction capability.

[0075] S3: Construct a multi-scale feature pyramid based on the multi-scale feature map of the sag image, and predict the multi-scale feature map of the sag image through the multi-scale feature pyramid to obtain a multi-scale heat map;

[0076] The improved MobileNetV3 lightweight neural network model is used to obtain multiple key layer features. Based on these key layer features, multiple feature layers are constructed to obtain a multi-scale feature pyramid.

[0077] Each key layer feature has a different resolution, and each feature layer detects and locates targets of different scales according to the resolution.

[0078] S4: Perform feature fusion on the multi-scale heat map through the dynamic weight fusion module to obtain the fused feature heat map;

[0079] Features at different levels have different characteristics. Shallow features typically contain more specific details, such as low-level features like the edges and textures of transmission lines. Deep features, on the other hand, focus more on abstract semantic information, capturing the overall structure of the transmission line and high-level features related to sag. The dynamic weight fusion module integrates these different levels of features based on weights, enabling better sag detection.

[0080] The dynamic weight fusion module dynamically adjusts the multi-scale heat map feature weights according to the UAV altitude and mission requirements. The calculation expression of the multi-scale heat map feature weights is:

[0081]

[0082] in, For the The heat map feature weights of each scale, is the slope factor, is the altitude of the drone, For the The preset height reference value of the feature of each scale,

[0083] At the same time, ensure that the sum of all weights is 1. The calculation expression is:

[0084]

[0085] in, is the number of characteristic scales.

[0086] S5: The sag key points of the fused feature heat map are located by the heat map regression method, the sag physical distance is calculated based on the sag key points, and the sag physical distance is corrected in real time by the temperature compensation algorithm to obtain the actual sag distance.

[0087] The key sag points include the suspension point and the lowest point of the transmission line conductor.

[0088] The key points of sag located by heat map regression include:

[0089] S51: Obtain an optimized heat map by removing adjacent response peaks on the fusion feature heat map;

[0090] S52: Find the position with the largest response value on the optimization heat map, and obtain the coordinates of the sag key point on the heat map;

[0091] S53: Repeat step S52 to obtain the coordinates of multiple sag key points on the thermal map;

[0092] S54: The coordinates of the sag key points on the heat map are mapped back to the coordinate space of the drone aerial image by scaling to obtain the positions of the sag key points.

[0093] The calculation expression of sag physical distance is:

[0094]

[0095] in, is the sag physical distance, is the control coefficient, is the vertical pixel distance of the sag key point in the fusion feature heat map, is the horizontal tension of the conductor, is the field of view of the drone camera, is the pixel height of the tower, is the tangent function.

[0096] The selection and innovation of this formula are derived from the catenary sag calculation formula in the "Design Manual of High-Voltage Transmission Lines for Power Engineering". The image information is converted into physical quantities. Considering the coupling relationship between sag length and tension, and that this relationship is related to environmental factors such as temperature, icing, and wind speed, as well as conductor properties such as conductor elastic modulus and cross-sectional area, the control coefficient is introduced after simplification through the mixed operation of various parameters in the line manual. , The unit is , It is not a fixed value. Due to changes in the environment, it can be obtained by calibrating the drone on a line path with known sag.

[0097] According to the flight attitude sensor data of the UAV or the geometric relationship between the wire and the tower in the image, the vertical angle at the time of shooting is determined, and then the ;

[0098] Consult the design documents, operation and maintenance records of the transmission line, or obtain the horizontal tension of the conductor through real-time measurement by a tension sensor;

[0099] Get the image resolution and other information set when the drone camera takes the image, and directly get the image height;

[0100] Obtain the current temperature from the meteorological monitoring equipment in the transmission line environment and determine the conductor temperature expansion coefficient based on the conductor's relevant material properties;

[0101] In a specific embodiment of the present invention, the conductor model is LGJ-185 / 35, the span is 150 meters, and the vertical pixel distance of the sag key point in the fusion feature heat map is , the drone's camera field of view is known through the attitude sensor carried by the drone during shooting ,but .

[0102] Consult the transmission line design documents to find out the horizontal tension of the conductor The vertical pixel distance of the sag key points in the fusion feature heat map generated by the lightweight model of the drone camera image , , tower pixel height , tower height , , , the current temperature is 30℃ and the reference temperature is 20℃, and the temperature change is is 10℃.

[0103] The physical distance of the sag is obtained from this .

[0104] Eliminate seasonal errors through real-time temperature correction.

[0105] The calculation expression of actual sag distance is:

[0106]

[0107] in, is the actual sag distance, is the sag physical distance, is the conductor temperature expansion coefficient, The difference between the current temperature and the reference temperature.

[0108] The actual sag distance is thus:

[0109]

[0110] Through real-time temperature correction, the sag distance accuracy error is less than 5%.

[0111] The present invention uses the UAV visual geometry and catenary mechanics theory to consider not only the influence of temperature change on the wire, but also the change of thermal expansion coefficient with the change of environment, wire aging, corrosion and other factors. In order to cope with the impact of these changes, the present invention is a reasonable application and optimization of existing standards in specific scenarios; the vertical pixel distance of the sag key point in the fusion feature heat map The original theory of overhead transmission line tension is coupled with sag and image pixels. By continuously fitting and adjusting relevant parameters for adverse conditions such as corrosive chemical areas and excessive water mist, the accuracy of conductor sag measurement is greatly improved.

[0112] And the control coefficient It is calibrable and scene-adaptable and is not a fixed value.

[0113] like Figure 2 As shown, a transmission line sag detection system based on a lightweight neural network model is used to detect a transmission line sag based on a lightweight neural network model, including:

[0114] The sag image acquisition module 101 acquires a transmission line image through drone aerial photography and pre-processes the transmission line image to obtain a sag image;

[0115] The feature extraction module 102 extracts multi-scale features of the sag image through the improved MobileNetV3 lightweight neural network model to obtain a multi-scale feature map of the sag image;

[0116] The multi-scale heat map acquisition module 103 constructs a multi-scale feature pyramid based on the multi-scale feature map of the sag image, and predicts the multi-scale feature map of the sag image through the multi-scale feature pyramid to obtain a multi-scale heat map;

[0117] The fusion module 104 performs feature fusion on the multi-scale heat map through the dynamic weight fusion module to obtain a fused feature heat map;

[0118] The sag distance acquisition module 105 locates the sag key points of the fused feature heat map through the heat map regression method, calculates the sag physical distance based on the sag key points, and corrects the sag physical distance in real time through the temperature compensation algorithm to obtain the actual sag distance.

[0119] Through the collaborative work of the above modules, and through technologies such as lightweight model optimization, multi-scale feature fusion, and temperature compensation algorithm, the core bottleneck of sag measurement in power inspections is systematically solved, the model computing power is significantly reduced, and high-precision, high-efficiency, and high-robust fully automatic detection is achieved.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A transmission line sag detection method based on a lightweight neural network model, characterized in that: include: S1: Acquire transmission line images through drone aerial photography and preprocess the transmission line images to obtain sag images; S2: Extract multi-scale features of the sag image through the improved MobileNetV3 lightweight neural network model to obtain the multi-scale feature map of the sag image; The improved MobileNetV3 lightweight neural network model includes a shallow feature reuse module, dynamic depth-separable convolution, and an adaptive spatial attention module; The shallow feature reuse module performs synchronous downsampling through the convolution path and the mean pooling path; Dynamic depth-wise separable convolution automatically adjusts the depth of the convolution kernel according to the complexity of the input image; Insert an adaptive spatial attention module before the residual connection of each neck network of the MobileNetV3 model; S3: Construct a multi-scale feature pyramid based on the multi-scale feature map of the sag image, and predict the multi-scale feature map of the sag image through the multi-scale feature pyramid to obtain a multi-scale heat map; S4: Perform feature fusion on the multi-scale heat map through the dynamic weight fusion module to obtain the fused feature heat map; S5: The sag key points of the fused feature heat map are located by the heat map regression method, the sag physical distance is calculated based on the sag key points, and the sag physical distance is corrected in real time by the temperature compensation algorithm to obtain the actual sag distance.

2. The method for detecting sag of a transmission line based on a lightweight neural network model according to claim 1, characterized in that: The pre-processing includes insulator detection, segmented transformation and conductor characteristic enhancement. The insulator detection method is to locate the coordinates of the top and bottom of the drone aerial image through the lightweight HRNet model to obtain the insulator detection image. The segmented transformation is to divide the insulator detection image into N segments, independently calculate the transformation matrix for each segment, and correct each segment of the conductor area into a horizontal straight line using the transformation matrix; The wire characteristic enhancement is to enhance the wire features in a low-contrast environment by using a CLAHE algorithm.

3. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, wherein: The improved MobileNetV3 lightweight neural network model is used to obtain multiple key layer features. Based on these key layer features, multiple feature layers are constructed to obtain a multi-scale feature pyramid. Each key layer feature has a different resolution, and each feature layer detects and locates targets of different scales according to the resolution.

4. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, wherein: The key sag points include the suspension point and the lowest point of the transmission line conductor.

5. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, characterized in that: The dynamic weight fusion module dynamically adjusts the multi-scale heat map feature weights according to the altitude of the UAV, performs feature fusion on the multi-scale heat map according to the multi-scale heat map feature weights, and obtains a fused feature heat map.

6. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, characterized in that: The key points of sag located by heat map regression include: S51: Obtain an optimized heat map by removing adjacent response peaks on the fusion feature heat map; S52: Find the position with the largest response value on the optimization heat map, and obtain the coordinates of the sag key point on the heat map; S53: Repeat step S52 to obtain the coordinates of multiple sag key points on the thermal map; S54: The coordinates of the sag key points on the heat map are mapped back to the coordinate space of the drone aerial image by scaling to obtain the positions of the sag key points.

7. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, characterized in that: The calculation expression of sag physical distance is: in, is the sag physical distance, is the control coefficient, is the vertical pixel distance of the sag key point in the fusion feature heat map, is the horizontal tension of the conductor, is the field of view of the drone camera, is the pixel height of the tower, is the tangent function.

8. The method for detecting sag of a power transmission line based on a lightweight neural network model according to claim 1, characterized in that: The calculation expression of actual sag distance is: in, is the actual sag distance, is the sag physical distance, is the conductor temperature expansion coefficient, The difference between the current temperature and the reference temperature.

9. A transmission line sag detection system based on a lightweight neural network model, characterized in that: The method for detecting sag of a power transmission line based on a lightweight neural network model according to any one of claims 1 to 8 comprises: A sag image acquisition module, wherein the sag image acquisition module acquires a transmission line image through drone aerial photography and pre-processes the transmission line image to obtain a sag image; A feature extraction module, wherein the feature extraction module extracts multi-scale features of the sag image through an improved MobileNetV3 lightweight neural network model to obtain a multi-scale feature map of the sag image; A multi-scale heat map acquisition module, which constructs a multi-scale feature pyramid based on the multi-scale feature map of the sag image, predicts the multi-scale feature map of the sag image through the multi-scale feature pyramid, and obtains a multi-scale heat map; A fusion module, which performs feature fusion on the multi-scale heat map through a dynamic weight fusion module to obtain a fused feature heat map; The sag distance acquisition module locates the sag key points of the fused feature heat map through the heat map regression method, calculates the sag physical distance based on the sag key points, and corrects the sag physical distance in real time through the temperature compensation algorithm to obtain the actual sag distance.

Images