A multi-scale feature quantization insulator automatic extraction method and system

Abstract

The application relates to the technical field of electric power inspection, and discloses a multi-scale feature quantization insulator automatic extraction method and system, which comprises the following steps: step 1: based on basic point cloud data, feature extraction is carried out, and an electric tower head point cloud is extracted; step 2: based on the electric tower head point cloud, different geometric features of the electric tower head point cloud under different scales are calculated; the scale comprises an optimal neighborhood of each point and a self-defined neighborhood; the geometric features comprise feature value features, density features and projection features of the insulator; the contribution degrees of different geometric features to the extraction of the insulator are quantized, the weights of the geometric features are obtained, the geometric features are weighted and fused, and then a comprehensive evaluation score of the insulator is obtained; and step 3: based on the comprehensive evaluation score of the insulator, a complete insulator string point cloud is extracted. The scheme provided by the application has wide applicability, can accurately extract insulator point clouds of different types from a complex scene, and has high extraction accuracy.

Description

Technical Field

[0001] This invention relates to the field of power line inspection technology, specifically to an automatic extraction method and system for insulators using multi-scale feature quantification. Background Technology

[0002] Insulators, as crucial components supporting overhead lines and ensuring insulation from towers in high-voltage power corridors, are vital to the safe and stable operation of the power system. During daily power grid operation, insulators exposed to dust for extended periods can easily become contaminated, reducing their insulating effectiveness. In more severe cases, they may overheat and spontaneously combust after being subjected to strong power surges from switches or lightning, causing short circuits and widespread power outages. Therefore, insulator inspection is a critical part of routine power system maintenance, requiring regular checks on insulator strings for dirt and integrity, and timely repair or replacement of insulators.

[0003] In recent years, with the rapid development of lightweight drones, drones equipped with optical cameras, thermal infrared cameras, and LiDAR (light detection and ranging) systems have been widely used in power grid inspection. Compared with traditional methods, drone-based inspection methods have advantages such as convenience, economy, speed, and safety. However, when applied to insulator inspection, these methods are affected by factors such as the inaccessibility of the area, limitations of drone signals, the complex distribution of insulators, and the large number of inspection points, making it difficult for manual drone operation to perform precise inspections. Generally, it is necessary to pre-plan the flight path and arrange the drone for automatic inspection. The LiDAR system on the drone can acquire high-precision terrain information and 3D maps of the power corridor, extract the 3D point cloud of the insulators and extract the inspection points, and then perform safe path planning to achieve automatic inspection, greatly improving the efficiency of drone inspection.

[0004] Among these, the accurate extraction of insulator point clouds and inspection points is a key step in achieving refined automatic inspection. Previously, extracting insulator point clouds relied on tedious manual labor. However, automatic insulator point cloud extraction algorithms face the following problems: (1) The large amount of data and the surrounding environment seriously affect the extraction efficiency and accuracy; (2) Insulators are relatively small and have a complex distribution. The linear structure similarity and complex topological relationship formed with power lines and towers make it difficult to accurately extract insulators; (3) The size, shape, and number of layers of power towers are diverse, and the length and type of insulators vary with the size and function of the power towers, resulting in low applicability of the algorithm. Existing image-based insulator recognition methods cannot obtain the spatial location of insulators, while point cloud-based insulator extraction methods have low extraction accuracy and low applicability, and cannot achieve accurate extraction in complex scenarios. Summary of the Invention

[0005] The present invention aims to provide an automatic insulator extraction method and system with multi-scale feature quantization, which has wide applicability and can accurately extract point clouds of different types of insulators from complex scenes with high extraction accuracy.

[0006] To achieve the above objectives, the basic solution provided by this invention is as follows:

[0007] Option 1

[0008] An automatic insulator extraction method based on multi-scale feature quantization includes the following steps:

[0009] Step 1: Based on the basic point cloud data, perform feature extraction and obtain the point cloud of the power tower head;

[0010] Step 2: Based on the point cloud of the power tower head, calculate its different geometric features at different scales; the scale includes the best neighborhood and custom neighborhood of each point; the geometric features include the insulator's eigenvalue features, density features and projection features; quantify the contribution of different geometric features to the extraction of the insulator, obtain the weight of each geometric feature, perform weighted fusion of each geometric feature, and then obtain the comprehensive evaluation score of the insulator.

[0011] Step 3: Extract the complete point cloud of the insulator string based on the comprehensive evaluation score of the insulator.

[0012] Furthermore, in step 1, when performing feature extraction, a vertical slicing method is used to extract the power tower point cloud from the basic point cloud data.

[0013] Furthermore, when extracting the point cloud of the power tower head, the main direction of the power tower is redirected to face the X-axis; then the power tower point cloud is horizontally sliced ​​and the power tower head point cloud is extracted.

[0014] Furthermore, the eigenvalue features include minimum eigenvalue, flatness, linearity, surface variation, PCA1, PCA2, and perpendicularity; the density features include point density; and the projection features include width and the sum of length and width.

[0015] Furthermore, in step 2, when calculating geometric features, a point cloud kd-tree is first constructed, and the feature entropy within a scale range of [0.8-2.0m] with an interval of 0.1m is calculated. The neighborhood radius corresponding to the minimum feature entropy is obtained as the optimal neighborhood radius. The geometric features are then calculated by combining the optimal neighborhood radius with multiple custom neighborhood point clouds.

[0016] Furthermore, in step 2, the contribution of different geometric features to the extraction of the insulator is quantified based on the feature information entropy, including the following sub-steps:

[0017] S1: Normalize each geometric feature; construct the eigenvalue matrix;

[0018] S2: Perform weight calculation; and the calculation formula is as follows:

[0019]

[0020] Among them, w i The weight represents the i-th geometric feature; The information entropy represents the i-th geometric feature; Represents the feature set; α i Let represent the normalized feature value of point i; m represent the number of geometric features in the feature set; n represent the number of feature values; and information entropy. a i This represents the normalized feature value at point i.

[0021] Further, during the normalization process, the geometric features are first divided into positive and negative features based on the insulator characteristic distribution; then the positive features are further divided according to α... j =v max -v j / v max -v min Perform normalization; divide negative features according to α j =v j -v min / v max -v min Normalization is performed; where, (v min v max ) represent the minimum and maximum values ​​of the eigenvalue, respectively; α j This represents the normalized feature value at point j.

[0022] Furthermore, in step 3, a regional erosion and growth strategy is used to extract the complete insulator string point cloud; the regional erosion and growth strategy includes: eroding the noise point cloud based on the fixed geometric features of the insulator; and growing the point cloud along the main direction of the insulator until the end of the complete insulator is reached.

[0023] Furthermore, when eroding the noise point cloud, the noise points are eroded based on the standardized point density threshold range [0.6-1] and the width range [0.6-1.4m].

[0024] Option 2

[0025] An automatic insulator extraction system with multi-scale feature quantization is provided for performing an automatic insulator extraction method with multi-scale feature quantization as described in Scheme 1, comprising a feature extraction module, a feature quantization module, and an insulator extraction module.

[0026] The feature extraction module is used to extract features based on the basic point cloud data and extract the point cloud of the tower head.

[0027] The feature quantization module is used to calculate different geometric features of the tower head point cloud at different scales. The scale includes the best neighborhood and custom neighborhood of each point. The geometric features include feature value features, density features and projection features. It is also used to quantify the contribution of different geometric features to the extraction of insulators, obtain the weight of each geometric feature, perform weighted fusion of each geometric feature, and then obtain the comprehensive evaluation score of the insulator.

[0028] The insulator extraction module is used to extract a complete point cloud of insulator strings based on the comprehensive evaluation score of the insulators.

[0029] The working principle and advantages of this invention are as follows: First, the tower head point cloud is extracted from the basic point cloud data. Then, through multi-scale geometric feature analysis and comprehensive evaluation scores of different geometric features, candidate insulator point clouds can be confirmed from the tower head point cloud. Finally, the complete insulator string point cloud is obtained by further segmenting and extracting the data using regional erosion and growth strategies.

[0030] This invention presents an automatic insulator extraction method and system based on multi-scale feature quantization, applicable to the extraction of insulators with various complex tower shapes and distributions, exhibiting wide applicability and high extraction accuracy. The key advantages are: firstly, this scheme selects a neighborhood point cloud computing geometric feature model; it does not need to consider the influence of the overall insulator distribution and tower shape; and it requires a smaller feature set for analysis. While meeting the insulator extraction requirements, this approach helps reduce computational load and improve extraction efficiency and accuracy.

[0031] Second, this scheme operates based on multi-scale feature quantization. By analyzing different types of geometric features at different scales, the applicability of this scheme can be effectively expanded to meet the analysis needs of different tower structures and different insulator structures. Furthermore, multi-feature quantization is performed, and the comprehensive evaluation score obtained by quantization weighting is used as the basis for insulator point cloud extraction. This effectively avoids the impact of different discriminative values ​​and thresholds of multiple features under multi-scale conditions on the extraction results, ensuring the stability and accuracy of extraction under multi-scale and multi-feature conditions.

[0032] Third, this scheme's multi-scale approach encompasses both the optimal scale based on feature entropy minimization and fixed scales corresponding to multiple features. By combining the accurate description of the target using the optimal neighborhood based on feature entropy with the expression of neighborhoods at multiple scales, it addresses the problem of poor size extraction accuracy in existing single-scale analysis methods applied to various tower-shaped scenarios. Fourth, this scheme also employs a region erosion and growth strategy, enabling refined segmentation and extraction of insulators based on the candidate insulator point cloud. By eroding the noise point cloud and growing the point cloud along the main direction, noise can be meticulously eliminated, and complete insulator strings can be obtained, resulting in complete and accurate insulator segmentation. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the method flow of an embodiment of the automatic insulator extraction method and system for multi-scale feature quantization according to the present invention;

[0034] Figure 2 This is a schematic diagram of the tower head plane position acquisition process in an embodiment of the multi-scale feature quantization insulator automatic extraction method and system of the present invention;

[0035] Figure 3 This is a schematic diagram of the tower head segmentation process in an embodiment of the multi-scale feature quantization automatic insulator extraction method and system of the present invention;

[0036] Figure 4 This is a schematic diagram of the feature and frequency distribution of negative features in an embodiment of the automatic extraction method and system for multi-scale feature quantization of insulators according to the present invention.

[0037] Figure 5 This is a schematic diagram of the feature and frequency distribution of the positive features in an embodiment of the automatic insulator extraction method and system for multi-scale feature quantization according to the present invention.

[0038] Figure 6 This is a schematic diagram of the insulator extraction process in an embodiment of the multi-scale feature quantization automatic insulator extraction method and system of the present invention;

[0039] Figure 7 This is a schematic diagram of the insulator extraction results in an embodiment of the multi-scale feature quantization automatic insulator extraction method and system of the present invention. Detailed Implementation

[0040] The following detailed explanation illustrates the specific implementation methods:

[0041] The basic implementation examples are as follows: Figure 1 As shown: An automatic insulator extraction method based on multi-scale feature quantization includes the following steps:

[0042] Step 1: Based on the basic point cloud data, perform feature extraction and obtain the point cloud of the power tower head;

[0043] The basic point cloud data is point cloud data of the power transmission corridor scene.

[0044] As attached Figure 2 As shown, vertical slicing is used to extract the power tower point cloud from the base point cloud data during feature extraction. Specifically, in actual power transmission scenarios, power towers are vertically continuous and have relatively large height differences under the sparse grid of transmission lines. Based on this, vertical slicing can quickly identify the center of the power tower, and then extract the overall point cloud of the power tower, including components such as the tower body and tower head, from the base point cloud data.

[0045] When extracting the point cloud of the power tower head, the tower's main direction is redirected to face the X-axis of the coordinate system; this avoids the incorrect removal of insulators. The power tower point cloud is then horizontally sliced, and the point cloud of the tower head is extracted. Specifically, after horizontal slicing, each layer of the tower body's point cloud presents a regular equilateral rectangle. Taking advantage of the rectangle's characteristic that its aspect ratio is always 1 at any angle, the tower body is removed, thus obtaining the power tower head point cloud, as shown in the attached figure. Figure 3 As shown.

[0046] After feature extraction, voxel sampling with a size of 0.2 μm was performed on the point cloud at the top of the power tower. This step yields a more uniform point density and further reduces the amount of point cloud data, making subsequent analysis easier.

[0047] Step 2: Based on the point cloud of the power tower head, calculate its different geometric features at different scales; the scale includes the best neighborhood and custom neighborhood of each point; the geometric features include the eigenvalue features, density features and projection features of the insulator; specifically, the insulator includes tension insulators and suspension insulators; the eigenvalue features include minimum eigenvalue (ME), flatness (PL), linearity (LI), surface variation (SV), PCA1, PCA2 and verticality (VE); the density features include point density (PD); the projection features include width (WI) and the sum of length and width (LS).

[0048] Among them, the minimum eigenvalue (ME) varies significantly at different scales, exhibiting good characteristics for multi-scale fusion and is suitable for evaluating tension insulators; PCA1 and PCA2 are significant features that help distinguish insulators, and based on these eigenvalues, insulator types can be accurately identified, and linear structures of this type of insulator can be extracted; surface variation (SV) shows prominent numerical performance at the connection points between the insulator ends and power lines and towers, and based on these eigenvalues, components at the connection points can be accurately removed; point density (PD) and width (WI) features are used to optimize the insulator extraction results during the extraction optimization stage. Additionally, a specially designed length-to-width sum (LS) feature is used to reduce the similarity between the point cloud of suspended insulators and the vertical structure of the tower, helping to improve the accuracy of suspended insulator point cloud extraction.

[0049] Furthermore, different geometric features are selected and calculated for different types of insulators. The specific selected geometric features and calculation methods are shown in Appendix Table 1.

[0050] Table 1 Calculation characteristics of insulators

[0051]

[0052] Furthermore, when calculating geometric features, a point cloud kd-tree is first constructed, and the feature entropy within a scale range of [0.8-2.0m] with intervals of 0.1m is calculated. The neighborhood radius corresponding to the minimum feature entropy is obtained as the optimal neighborhood radius. The geometric features are then calculated by combining the optimal neighborhood radius with multiple custom neighborhood point clouds. The formula for calculating the feature entropy is as follows:

[0053]

[0054] In the formula, e1, e2 and e3 (e i =λ i / ∑λ) represents the result of the normalization of the corresponding eigenvalues. Here, the three eigenvalues ​​are exactly the three eigenvalues ​​of the principal component analysis of the neighborhood point cloud.

[0055] In practical applications, the tower structure and insulator exhibit different characteristics at different scales. At smaller scales, both show similar linear structures, while at larger scales, the tower frame structure exhibits better planarity. The linear structure of the tower tends to be a linear expression of the optimal neighborhood at a small scale. To better represent the planarity of the tower, a scale range of [0.8-2.0m] with 0.1m intervals is selected, and the feature entropy within this range is calculated. Based on the optimal scale for the minimum feature entropy (corresponding to the aforementioned optimal neighborhood radius) and the corresponding PCA2, surface variation, PCA1, and verticality features, for tension insulators, when calculating geometric features, the minimum feature value at a radius of 0.8m, the linearity at a radius of 1.3m, and the planarity feature at a radius of 1.8m are added. For suspension insulators, when calculating geometric features, the sum of length and width at a radius of 1.5m is added. Among them, the 0.8m radius, 1.3m radius, 1.5m radius and 1.8m radius all refer to the custom neighborhood radius of the custom neighborhood point cloud.

[0056] Using the above method, feature recognition and calculation are performed based on the local neighborhood features (neighborhood point cloud) within this area. It does not require consideration of the overall distribution of insulators and the shape of power towers. The resulting feature data is less and more representative, which can meet the feature extraction requirements of insulators while helping to further reduce the amount of calculation and improve the extraction accuracy.

[0057] Furthermore, the contribution of different geometric features to the extraction of insulators is quantified, and the weight of each geometric feature is obtained. Weighted fusion of these features yields a comprehensive evaluation score for the insulator. Based on this comprehensive evaluation score, candidate insulator point clouds are extracted for further refined point cloud segmentation in step 3. Specifically, when extracting candidate insulator point clouds, the comprehensive extraction score [0.85-0.95] for tension insulators is used as a benchmark, and point clouds within this benchmark range are extracted as candidate tension insulator point clouds; the comprehensive extraction score [0.9-1] for suspension insulators is used as a benchmark, and point clouds within this benchmark range are extracted as candidate suspension insulator point clouds. Here, this scheme sets different extraction standards for candidate point clouds for different types of insulators, demonstrating strong targeting.

[0058] Here, this scheme specifically adopts the entropy weight method to fuse multiple geometric features, and only assigns a threshold to the comprehensive evaluation score of the multi-feature weighting. This can avoid the influence of different neighborhood scales and different feature thresholds on the insulator extraction results, and ensure that the extraction results are accurate and reasonable.

[0059] Specifically, in this embodiment, the contribution of different geometric features to the extraction of insulators is quantified based on feature information entropy, including the following sub-steps:

[0060] S1: Normalize each geometric feature; construct the eigenvalue matrix.

[0061] During normalization, the geometric features are first divided into positive and negative features according to the insulator type, as shown in the attached figure. Figure 4 and Figure 5 As shown; then, the positive features are arranged according to α j =v max -v j / v max -v min Perform normalization; divide negative features according to α j =v j -v min / v max -v min Normalization is performed; where, (v min v max ) represent the minimum and maximum values ​​of the eigenvalue, respectively; α j This represents the normalized feature value at point j.

[0062] Specifically, when the insulator is a tension insulator, its positive characteristics include PCA1 and linearity; its negative characteristics include minimum characteristic value, flatness, PCA2, and surface variation. When the insulator is a suspension insulator, its positive characteristics include verticality; its negative characteristics include PCA2 and the sum of length and width.

[0063] S2: Perform weight calculation; and the calculation formula is as follows:

[0064]

[0065] Among them, w i The weight represents the i-th geometric feature; The information entropy represents the i-th geometric feature; Represents the feature set; α i Let represent the normalized feature value of point i; m represent the number of geometric features in the feature set; n represent the number of feature values; and information entropy. a i This represents the normalized feature value at point i.

[0066] Furthermore, if the normalization sign for negative features is defined as "-" and the normalization sign for positive features is defined as "+", then the comprehensive evaluation score of the insulator is calculated according to the following formula:

[0067]

[0068] Among them, S TI S represents the overall evaluation score of tension insulators. SI represents the comprehensive evaluation score of the suspension insulator; w represents the weight of the geometric features, corresponding to equation (1); represents the feature set; r ~ minE represents the optimal neighborhood scale corresponding to the minimum feature entropy.

[0069] Step 3: Extract the complete point cloud of the insulator string based on the comprehensive evaluation score of the insulator.

[0070] Specifically, as shown in the attached document Figure 6 As shown, a regional erosion and growth strategy is used to extract the complete insulator string point cloud. The regional erosion and growth strategy includes: eroding the noise point cloud based on the fixed geometric features of the insulator; and growing the point cloud along the main direction of the insulator until the end of the complete insulator is reached. Furthermore, when eroding the noise point cloud, a standardized point density threshold range [0.6-1] and a width range [0.6-1.4m] are used as benchmarks; points exceeding these benchmark ranges are identified as noise points and eroded accordingly.

[0071] This embodiment also provides an automatic insulator extraction system with multi-scale feature quantization, used to perform the automatic insulator extraction method with multi-scale feature quantization as described above, including a feature extraction module, a feature quantization module and an insulator extraction module;

[0072] The feature extraction module is used to extract features based on the basic point cloud data and extract the point cloud of the tower head.

[0073] The feature quantization module is used to calculate different geometric features of the tower head point cloud at different scales. The scale includes the best neighborhood and custom neighborhood of each point. The geometric features include feature value features, density features and projection features. It is also used to quantify the contribution of different geometric features to the extraction of insulators, obtain the weight of each geometric feature, perform weighted fusion of each geometric feature, and then obtain the comprehensive evaluation score of the insulator.

[0074] The insulator extraction module is used to extract a complete point cloud of insulator strings based on the comprehensive evaluation score of the insulators.

[0075] In practical applications, insulator extraction is performed on transmission corridors containing various types of power towers. These towers include various shapes such as cat-head towers, zigzag towers, goblet towers, gate towers, ram's horn towers, and multi-layer zigzag towers. Basic point cloud data is provided by a point cloud library (PCL), which contains data for 82 power towers, including 359 tension insulators and 295 suspension insulators. Based on this basic point cloud data, the method provided in this embodiment is used to extract insulator point clouds from different types of power towers, such as... Figure 7 As shown, this method achieves a recall rate of 99.16% and an accuracy rate of 98.88% for extracting tension insulators. For suspension insulators, the recall rate is 97.29% and the accuracy rate is 97.63%.

[0076] The extraction results show that this method is applicable to extracting different types of insulators from power towers of different types, sizes, and shapes. Furthermore, even when the sizes and insulator lengths of 220kV and 500kV power towers differ significantly, thanks to the representation of multi-scale features, the insulator extraction operation for both types still achieves ideal results.

[0077] This embodiment provides an automatic insulator extraction method and system based on multi-scale feature quantization. First, it coarsely extracts tower point clouds from transmission corridors. Then, it fuses multi-scale feature weights to obtain a comprehensive evaluation score, thereby extracting candidate insulator point clouds. Finally, it employs regional erosion and growth strategies to segment and obtain complete insulator string point clouds. This solution is applicable to insulator extraction from various complex tower shapes and distributions, accurately extracting different types of insulator point clouds from complex scenes with high precision and wide applicability.

[0078] The above descriptions are merely embodiments of the present invention. Commonly known structures and characteristics of the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are aware of all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, based on the guidance provided in this application, improve and implement this solution in conjunction with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent.

Claims

1. A method for automatic extraction of insulators from multi-scale feature quantization, characterized in that, Includes the following steps: Step 1: Based on the basic point cloud data, perform feature extraction and obtain the point cloud of the power tower head; Step 2: Based on the point cloud of the power tower head, calculate its different geometric features at different scales; The scale includes the best neighborhood and custom neighborhood for each point; the geometric features include the eigenvalue features, density features, and projection features of the insulator; the insulators include tension insulators and suspension insulators; Eigenvalue features include minimum eigenvalue ME, flatness PL, linearity LI, surface variation SV, PCA1, PCA2, and perpendicularity VE; density features include point density PD; projection features include width WI and the sum of length and width LS. The contribution of different geometric features to the extraction of insulators is quantified based on feature information entropy, and the weights of each geometric feature are obtained. These geometric features are then weighted and fused to obtain a comprehensive evaluation score for the insulator. This process includes the following sub-steps: S1: Normalize each geometric feature; construct the eigenvalue matrix; when the insulator is a tension insulator, its positive features include PCA1 and linearity; its negative features include minimum eigenvalue, flatness, PCA2 and surface variation; when the insulator is a suspension insulator, its positive features include verticality; its negative features include PCA2 and the sum of length and width. S2: Perform weight calculation; and the calculation formula is as follows: ； in, The weight represents the i-th geometric feature; The information entropy represents the i-th geometric feature; , represents the feature set; Let represent the normalized feature value of point i; m represent the number of geometric features in the feature set; n represent the number of feature values; and information entropy. ; The normalization sign for negative features is defined as -, and the normalization sign for positive features is defined as +. The comprehensive evaluation score of the insulator is then calculated according to the following formula: ； in, This indicates the overall evaluation score of the tension insulator; The score represents the overall evaluation score of the suspension insulator; w represents the weight of the geometric features. This represents the optimal neighborhood scale corresponding to the minimum feature entropy; 0.8m, 1.3m, 1.5m and 1.8m all refer to the custom neighborhood radius of the custom neighborhood point cloud; Step 3: Based on the comprehensive evaluation score of the insulator, a regional erosion and growth strategy is adopted to extract the complete insulator string point cloud; the regional erosion and growth strategy includes: eroding the noise point cloud based on the fixed geometric features of the insulator; growing the point cloud along the main direction of the insulator until the end of the complete insulator is reached.

2. The automatic insulator extraction method with multi-scale feature quantization according to claim 1, characterized in that, In step 1, when performing feature extraction, a vertical slicing method is used to extract the power tower point cloud from the basic point cloud data.

3. The method of claim 2, wherein the method further comprises: When extracting the point cloud of the power tower head, the main direction of the power tower is redirected to face the X-axis; then the power tower point cloud is horizontally sliced ​​and extracted to obtain the power tower head point cloud.

4. The method of claim 1, wherein the method further comprises: In step 2, when calculating geometric features, a point cloud kd-tree is first constructed, and the feature entropy within the scale range of [0.8-2.0m] with an interval of 0.1m is calculated. The neighborhood radius corresponding to the minimum feature entropy is obtained as the optimal neighborhood radius. The geometric features are then calculated by combining the optimal neighborhood radius with multiple custom neighborhood point clouds.

5. The method of claim 1, wherein the method further comprises: During normalization, the geometric features are first divided into positive and negative features based on the insulator characteristic distribution; then the positive features are further classified according to... Perform normalization processing; classify negative features according to Normalization was performed; among them, These represent the minimum and maximum values ​​of the characteristic value, respectively. This represents the normalized feature value at point j.

6. The method of claim 1, wherein the method further comprises: When eroding noise point clouds, the noise points are eroded based on the standardized point density threshold range [0.6-1] and the width range [0.6-1.4m].

7. An automatic insulator extraction system for multi-scale feature quantization, characterized in that, An automatic insulator extraction method for performing multi-scale feature quantization as described in any one of claims 1-6 includes a feature extraction module, a feature quantization module, and an insulator extraction module; The feature extraction module is used to extract features based on the basic point cloud data and extract the point cloud of the tower head. The feature quantization module is used to calculate the different geometric features of the power tower head point cloud at different scales. The scale includes the best neighborhood and a custom neighborhood for each point; the geometric features include eigenvalue features, density features, and projection features. It is also used to quantify the contribution of different geometric features to the extraction of insulators, obtain the weight of each geometric feature, and perform weighted fusion of the weights of each geometric feature to obtain the comprehensive evaluation score of the insulator. The insulator extraction module is used to extract a complete point cloud of insulator strings based on the comprehensive evaluation score of the insulators.

Images