Detection and early warning system of optical cable cross-connecting box

Abstract

The invention discloses a detection and early warning system for an optical cable cross-connecting box, and particularly relates to the field of automation device monitoring, and the system comprises a laser scanning module, a point cloud processing module, a point cloud detection module and a visual monitoring module. And the laser scanning module is used for installing a laser scanner into the optical cable cross-connecting box, recording coordinates and reflection intensity information of discrete points where laser beams intersect with the target surface in the optical cable cross-connecting box, and generating point cloud data by receiving reflection signals. Through the laser scanning and point cloud processing technology, the monitoring precision and efficiency can be significantly improved, the three-dimensional structure information in the optical cable cross-connecting box is comprehensively captured, the limitation of human factors is overcome, the risks of missing detection and error detection are reduced, the potential fault area is more accurately identified through the convolutional neural network, and the fault detection accuracy is improved. And early warning information is generated in time, so that maintenance personnel are helped to quickly identify areas needing to be concerned, the risk of wrong identification is reduced, and the operation safety is improved.

Description

Technical Field

[0001] The invention relates to the field of automation device monitoring, and in particular to a detection and early warning system for an optical cable junction box. Background Art

[0002] The optical cable junction box is a key device in the optical fiber communication network. It is usually set up at the node position of the communication network to play the role of signal transfer and distribution, ensuring the smooth transmission of data between different networks. In modern communication networks, the importance of the optical cable junction box is self-evident. With the continuous increase in data traffic, optical fiber communication has become the mainstream transmission method. As the core component of the optical fiber network, the optical cable junction box undertakes the task of signal transfer and distribution, ensuring the efficient operation of the network. Its reliability directly affects the stability and service quality of the entire communication network. With the rapid development of information technology, the optical fiber communication network has become one of the basic infrastructures of modern communications and is widely used in telecommunications, the Internet and various data transmission scenarios. The health status and stability of the optical cable connection, node joints, optical fiber distribution and other components inside the optical cable junction box are directly related to the performance and reliability of the entire communication network. Therefore, real-time monitoring, maintenance and fault warning of the optical cable junction box are of great practical significance.

[0003] Traditional optical cable junction box detection methods mainly rely on manual inspections and regular maintenance, which is not only inefficient, but also often difficult to detect hidden dangers or faults due to human factors. With the continuous growth of data traffic, the stability and security requirements of optical fiber communication networks are becoming higher and higher. Therefore, introducing advanced technical means to dynamically monitor and intelligently manage optical cable junction boxes has become an important research direction.

[0004] The existing technology has the following deficiencies: the existing optical cable junction box detection method is not only time-consuming and labor-intensive, but also easily affected by human factors, such as the experience, attention and working status of the inspectors. Manual inspections often fail to fully cover all areas, resulting in some potential problems being ignored, which may lead to more serious failures or accidents. In addition, the frequency of manual inspections is usually low, and abnormal conditions cannot be discovered and handled in a timely manner, increasing the risk of equipment failure. The existing monitoring technology of optical cable junction boxes relies on static sensors and simple threshold judgments, and cannot obtain the detailed status inside the optical cable junction box in real time. When facing complex environments, it often cannot provide sufficient information, resulting in wrong judgments and decisions. Data analysis is performed through simple fixed algorithms, lacks intelligent processing capabilities, and is difficult to quickly and accurately identify abnormal conditions, resulting in extended response time, affecting the timeliness and effectiveness of maintenance work. In addition, the existing technology is insufficient in its ability to handle data noise and redundant information, which may lead to incorrect abnormal identification and increase maintenance costs. The information transmission is too low, and there is a lack of intuitive visualization inside the optical cable junction box, making it difficult for maintenance personnel to quickly understand the abnormal location, affecting the timeliness and accuracy of on-site maintenance.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a detection and early warning system for an optical cable junction box to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above-mentioned object, the present invention provides the following technical solutions: a detection and early warning system for an optical cable junction box, comprising a laser scanning module, a point cloud processing module, a point cloud detection module and a visual monitoring module; Laser scanning module: Install the laser scanner in the optical cable junction box, record the coordinates and reflection intensity information of the discrete points where the laser beam intersects with the target surface inside the optical cable junction box, and generate point cloud data by receiving the reflection signal; Point cloud processing module: receives point cloud data, removes invalid points through random forest and extracts neighboring point density features, adaptively and dynamically deletes noise points in point cloud data based on neighboring point density features, and implements voxel grid filtering for multi-scale filtering processing; Point cloud detection module: The point cloud data processed by multi-scale filtering is converted into a three-dimensional mesh model inside the optical cable junction box through Poisson reconstruction. The convolutional neural network is used to apply the joint position features, cable direction features, geometric state features and structural information features, and the probability distribution of abnormal monitoring of the optical cable junction box is calculated to generate early warning information; Visual monitoring module: Convert the coordinates of the abnormal position into the abnormal position of the three-dimensional grid model according to the early warning information, create a three-dimensional visualization interface, add support for user interactive operations, and dispatch professional maintenance personnel to conduct a comprehensive inspection.

[0008] Preferably, install a laser scanner in the optical cable junction box, set the scanning resolution to 2000000PPS and the accuracy to ±1mm, start the laser scanner and set the scanning time to 5min, adjust the position of the laser scanner for multi-angle scanning, and make the laser beam fully cover the optical cable junction box through 5 different scanning angles, record the discrete point coordinates and reflection intensity information of the intersection of the laser beam and the target surface inside the optical cable junction box, and generate point cloud data by receiving the reflected signal.

[0009] Preferably, a tree structure is constructed, the branch nodes inside the tree structure represent the invalid point features of the point cloud data, the leaf nodes inside the tree structure represent the invalid removal results, and multiple decision trees are constructed by randomly selecting invalid point features and point cloud data. The invalid point features of the branch nodes are randomly selected according to the leaf nodes for optimal splitting. An independent classification prediction is performed on each decision tree, and the classification prediction that appears most frequently in all decision trees is selected as the invalid point classification, and the invalid points of the point cloud data are removed by judgment.

[0010] Preferably, the neighboring point density feature of the point cloud data is extracted, the neighboring point radius is dynamically adjusted according to the local change of the neighboring point density feature, the minimum number of neighboring points is selected dynamically through the local point cloud data, the distance from each point to other points is calculated by the Euclidean distance metric algorithm, and the number of points in the neighborhood with the neighboring point radius for each point is checked through the KD tree. When the number of points in the neighborhood is ≥ the minimum number of neighboring points, the point is judged to be a core point and marked, and an empty cluster list is initialized. For the current core point that has not been marked and does not belong to any cluster, a new cluster is created and the current core point and the points in the neighborhood with the neighboring point radius are added to the new cluster, and the new cluster is added recursively. The current non-core point and the point in the neighborhood with the neighboring point radius of a cluster are marked as boundary points, and all points not assigned to any cluster are marked as noise points and deleted.

[0011] Preferably, voxel grid filtering is implemented, and the voxel size is dynamically adjusted according to the neighboring point density characteristics of the point cloud data. When the neighboring point density characteristics are ≥10 points / cm³, it is defined as a dense area, and the voxel size is set to 0.1 times the average distance between the points of the point cloud data. When the neighboring point density characteristics are <10 points / cm³, it is defined as a sparse area, and the voxel size is set to 0.5 times the average distance between the points of the point cloud data. The largest voxel size is applied for multi-scale filtering processing, and the voxel size is gradually reduced for multi-scale filtering processing.

[0012] Preferably, the point cloud data is converted into a three-dimensional mesh model of the interior of the optical cable junction box through Poisson reconstruction, the mesh optimization tool is used to remove redundant vertices and faces of the three-dimensional mesh model, the sharp edges and unnecessary details of the three-dimensional mesh model are removed through Laplace smoothing, the three-dimensional mesh model is detailed and modified with engraving tools, and bumps and subtle depictions are added, and the texture coordinates are added and mapped through UV unfolding according to the color information of the point cloud data.

[0013] Preferably, the joint position features of the three-dimensional grid model inside the optical cable junction box are extracted by color marking, and the three-dimensional coordinate information of the joint position is recorded, the cable direction features of the three-dimensional grid model inside the optical cable junction box are extracted by path lines, and the three-dimensional coordinate information, length and curvature of the cable direction are recorded, the geometric state features and structural information features of the optical cable junction box are extracted, a convolutional neural network model is established, a deformable convolution kernel is added to slide on the three-dimensional grid model, and the joint position features, cable direction features, geometric state features and structural information features are applied, and the specific formula of the convolution operation is:

[0014] in, represents an element in each feature, Represents the three-dimensional coordinate information of the three-dimensional mesh model, represents the weight of the convolution kernel, Respectively represent the position index of the convolution kernel on the three-dimensional grid model, Respectively represent the index of each feature in height, depth and width on the three-dimensional grid model, introduce the Swish activation function, perform spatial dimension average pooling on each feature, flatten each feature output by the convolution layer and the pooling layer into a one-dimensional vector, and connect it to the fully connected layer for classification, and use the softmax function to calculate the probability distribution of abnormal monitoring of the optical cable junction box. The specific formula is:

[0015] in, It represents the probability that the output category O is g given the input Q. represents the Swish activation function value, Indicates the information level value of the current abnormal category, It indicates the total number of abnormal categories. The probability distribution threshold is set to 0.5. When the probability distribution is higher than the probability distribution threshold, it is marked as an abnormality inside the optical cable junction box and an early warning message is generated.

[0016] Preferably, a three-dimensional grid model and a three-dimensional scene are constructed using the Three.js command, a three-dimensional grid model is imported and a three-dimensional scene is added, the coordinates of the abnormal position are converted into the abnormal position of the three-dimensional grid model according to the early warning information, the abnormal position area of ​​the optical cable junction box is marked in red, and the warning position area of ​​the optical cable junction box is marked in yellow, a three-dimensional visualization interface is created, support for user interactive operations is added, free rotation, zooming in and out of the optical cable junction box is performed, an on-site inspection plan for the optical cable junction box is formulated according to the early warning information and the three-dimensional scene, and professional maintenance personnel are dispatched for a comprehensive inspection, the abnormal position of the three-dimensional grid model is used as on-site evidence to implement maintenance of the optical cable junction box, and on-site feedback information is recorded.

[0017] In the above technical solution, the technical effects and advantages provided by the present invention are: Laser scanning and point cloud processing technology can significantly improve the accuracy and efficiency of monitoring, fully capture the three-dimensional structural information inside the optical cable junction box, overcome the limitations of human factors, and reduce the risk of missed detection and false detection, thereby greatly improving the comprehensiveness and accuracy of monitoring, and better reflecting the actual situation inside the junction box. The random forest algorithm is used to effectively identify and eliminate invalid points in the point cloud data, and accurately extract key feature information. The processed point cloud data is converted into a high-quality three-dimensional mesh model using Poisson reconstruction technology, and the abnormal situation of the optical cable junction box is monitored and analyzed in real time through a convolutional neural network. Using multi-dimensional feature input, the model can more accurately identify potential fault areas and generate timely It generates early warning information, enabling the maintenance team to take prompt action to avoid greater losses. It extracts the density features of neighboring points and dynamically adjusts the radius of neighboring points to adapt to changes in areas of different densities, ensuring that useful information can still be effectively extracted in complex environments. It more accurately identifies structural features in point cloud data, uses KD trees to accelerate the search of neighborhood points, and significantly improves computing efficiency when processing large-scale point cloud data. It builds three-dimensional grid models and scenes to more intuitively display the structure and status of the optical cable junction box, helping maintenance personnel to quickly identify areas that need attention, reduce the risk of misidentification and improve operational safety. Interactive operations allow maintenance personnel to observe the status of the optical cable junction box from different angles, which helps to understand the overall picture of the abnormal location. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0019] Figure 1The present invention is a method flow chart of a detection and early warning system for an optical cable junction box.

[0020] Figure 2 The present invention is a module schematic diagram of a detection and early warning system for an optical cable junction box. DETAILED DESCRIPTION

[0021] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0022] The present invention provides Figure 1 A detection and early warning system for an optical cable junction box is shown, comprising a laser scanning module, a point cloud processing module, a point cloud detection module and a visual monitoring module; Laser scanning module: Install the laser scanner in the optical cable junction box, record the coordinates and reflection intensity information of the discrete points where the laser beam intersects with the target surface inside the optical cable junction box, and generate point cloud data by receiving the reflection signal; Install the laser scanner in the optical cable junction box, set the scanning resolution to 2000000PPS and the accuracy to ±1mm, start the laser scanner and set the scanning time to 5min, adjust the position of the laser scanner for multi-angle scanning, and make the laser beam fully cover the optical cable junction box through 5 different scanning angles. Record the discrete point coordinates and reflection intensity information where the laser beam intersects with the target surface inside the optical cable junction box, and generate point cloud data by receiving the reflected signal.

[0023] Point cloud processing module: receives point cloud data, removes invalid points through random forest and extracts neighboring point density features, adaptively and dynamically deletes noise points in point cloud data based on neighboring point density features, and implements voxel grid filtering for multi-scale filtering processing; A tree structure is constructed, and the branch nodes inside the tree structure represent the invalid point features of the point cloud data, and the leaf nodes inside the tree structure represent the invalid removal results. Multiple decision trees are constructed by randomly selecting invalid point features and point cloud data. The invalid point features of the branch nodes are randomly selected according to the leaf nodes for optimal splitting. Independent classification prediction is performed on each decision tree, and the classification prediction that appears most frequently in all decision trees is selected as the invalid point classification to determine and remove the invalid points of the point cloud data.

[0024] The neighboring point density features of the point cloud data are extracted, where the neighboring point density features are used to identify the sparse areas of the points inside the optical cable junction box. The neighboring point radius is dynamically adjusted according to the local changes of the neighboring point density features. The minimum number of neighboring points is selected dynamically through the local point cloud data. The distance from each point to other points is calculated by the Euclidean distance measurement algorithm. The number of points in the neighborhood of each point with the neighboring point radius is checked by the KD tree. When the number of points in the neighborhood is ≥ the minimum number of neighboring points, the point is judged to be a core point and marked. The empty cluster list is initialized. For the current core point that has not been marked and does not belong to any cluster, a new cluster is created and the current core point and the points in the neighborhood with the neighboring point radius are added to the new cluster. The new cluster is added recursively. The current non-core point and the point in the neighborhood with the neighboring point radius of a cluster are marked as boundary points. All points that are not assigned to any cluster are marked as noise points and deleted.

[0025] Voxel grid filtering is implemented, and the voxel size is dynamically adjusted according to the neighboring point density characteristics of the point cloud data. When the neighboring point density characteristics are ≥10 points / cm³, it is defined as a dense area, and the voxel size is set to 0.1 times the average distance between the points of the point cloud data. When the neighboring point density characteristics are <10 points / cm³, it is defined as a sparse area, and the voxel size is set to 0.5 times the average distance between the points of the point cloud data. The largest voxel size is applied for multi-scale filtering, and the voxel size is gradually reduced for multi-scale filtering to retain important information while reducing redundant points.

[0026] Point cloud detection module: The point cloud data processed by multi-scale filtering is converted into a three-dimensional mesh model inside the optical cable junction box through Poisson reconstruction. The convolutional neural network is used to apply the joint position features, cable direction features, geometric state features and structural information features, and the probability distribution of abnormal monitoring of the optical cable junction box is calculated to generate early warning information; The point cloud data is converted into a 3D mesh model of the interior of the optical cable junction box through Poisson reconstruction. The mesh optimization tool is used to remove redundant vertices and faces of the 3D mesh model to ensure that the 3D mesh model is more efficient during visualization and rendering. The sharp edges and unnecessary details of the 3D mesh model are removed through Laplace smoothing to ensure the natural appearance of the 3D mesh model. The 3D mesh model is modified with details using engraving tools, and concave and convex and subtle depictions are added to ensure that the 3D mesh model is accurate and reflects the actual structure inside the optical cable junction box. According to the color information of the point cloud data, the texture coordinate setting is added through UV unfolding and the texture of the 3D mesh model is mapped to achieve more delicate detail presentation of the 3D mesh model.

[0027] The joint position features of the three-dimensional grid model inside the optical cable junction box are extracted by color marking, and the three-dimensional coordinate information of the joint position is recorded. The cable direction features of the three-dimensional grid model inside the optical cable junction box are extracted by path lines, and the three-dimensional coordinate information, length and curvature of the cable direction are recorded. The geometric state features and structural information features of the optical cable junction box are extracted, and a convolutional neural network model is established. A deformable convolution kernel is added to slide on the three-dimensional grid model, and the joint position features, cable direction features, geometric state features and structural information features are applied. The specific formula of the convolution operation is:

[0028] in, represents an element in each feature, Represents the three-dimensional coordinate information of the three-dimensional mesh model, represents the weight of the convolution kernel, Respectively represent the position index of the convolution kernel on the three-dimensional grid model, Respectively represent the index of each feature in height, depth and width on the three-dimensional grid model, introduce the Swish activation function, perform spatial dimension average pooling on each feature, flatten each feature output by the convolution layer and the pooling layer into a one-dimensional vector, and connect it to the fully connected layer for classification, and use the softmax function to calculate the probability distribution of abnormal monitoring of the optical cable junction box. The specific formula is:

[0029] in, It represents the probability that the output category O is g given the input Q. represents the Swish activation function value, Indicates the information level value of the current abnormal category, It indicates the total number of abnormal categories. The probability distribution threshold is set to 0.5. When the probability distribution is higher than the probability distribution threshold, it is marked as an abnormality inside the optical cable junction box and an early warning message is generated.

[0030] Visual monitoring module: Converts the coordinates of abnormal locations into abnormal locations of a three-dimensional grid model based on early warning information, creates a three-dimensional visualization interface, adds support for user interactive operations, and dispatches professional maintenance personnel for comprehensive inspections; Use the Three.js command to build a 3D grid model and 3D scene, import the 3D grid model and add the 3D scene, convert the coordinates of the abnormal position into the abnormal position of the 3D grid model according to the early warning information, mark the abnormal position area of ​​the optical cable junction box in red, and mark the warning position area of ​​the optical cable junction box in yellow, create a 3D visualization interface, add support for user interactive operations, and freely rotate, zoom in and out the inside of the optical cable junction box to view the abnormal positions at different angles. According to the early warning information and the 3D scene, formulate an on-site inspection plan for the optical cable junction box, dispatch professional maintenance personnel for a comprehensive inspection, use the abnormal position of the 3D grid model as on-site evidence to implement maintenance of the optical cable junction box, and record on-site feedback information.

[0031] The present invention introduces efficient and significant point cloud processing. By removing invalid points through the random forest algorithm, the quality of point cloud data can be significantly improved. By extracting neighboring point density features and dynamically adjusting the neighboring point radius, the system can more flexibly identify sparse areas and dense areas. After implementing voxel grid filtering, it can effectively retain important information while reducing redundant points. By dynamically adjusting the voxel size, the system can effectively balance between dense areas and sparse areas to ensure the integrity and accuracy of point cloud data. For example, the original point cloud data contains 100,000 points, of which 15,000 are identified as invalid points. After processing with the random forest algorithm, these invalid points are removed, and the number of remaining valid points is 85,000. In the extraction of neighboring point density features, the point density in one area is 12 points / cm³, and the point density in another area is 5 points / cm³. The dynamic adjustment of the neighboring point radius is: for the area with a point density of 12 points / cm³, the neighboring point radius is set to 0.2cm, and the core is identified. The number of points is 1,000, forming 5 clusters. For the area with a point density of 5 points / cm³, the radius of neighboring points is set to 0.5 cm. The number of core points identified is 300, forming 2 clusters. In the voxel grid filtering stage, the voxel size is set to 0.1 times the average distance between points in dense areas. Assuming the average distance is 0.5 cm, the voxel size is 0.05 cm. In sparse areas, the voxel size is set to 0.5 times the average distance between points, and the voxel size is 0.25 cm.

[0032] The present invention introduces high-precision point cloud detection and converts point cloud data into a high-quality three-dimensional mesh model through Poisson reconstruction, so that the internal structure of the optical cable junction box can be accurately represented. The convolutional neural network is used to extract the joint position features, cable direction features, geometric state features and structural information features, which improves the monitoring ability of the internal state of the optical cable junction box and helps to discover potential abnormal situations at an early stage. The convolutional neural network model is used to perform abnormality detection on the optical cable junction box, and the generated probability distribution can quickly identify abnormal situations. For example, among the street location features, there are (5.2, 3.3, 4. 1), (6.8, 3.2, 4.5), (5.7, 2.9, 4.0), then three joint positions are identified, and the cable direction features are (4.5, 2.9, 3.1), (5.5, 3.1, 4.0), then the recorded cable direction length is 4.9 meters, and the curvature is 15 degrees. After constructing the convolutional neural network, the probability of abnormal type A is 0.65, and the probability of abnormal type B is 0.30. Since both abnormal types A and B are higher than the set threshold, the abnormal type A is marked inside the optical cable junction box, and an early warning message is generated to notify the maintenance team in time for inspection.

[0033] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A detection and early warning system for an optical cable junction box, characterized in that: It includes laser scanning module, point cloud processing module, point cloud detection module and visual monitoring module; Laser scanning module: Install the laser scanner in the optical cable junction box, record the coordinates and reflection intensity information of the discrete points where the laser beam intersects with the target surface inside the optical cable junction box, and generate point cloud data by receiving the reflection signal; Point cloud processing module: receives point cloud data, removes invalid points through random forest and extracts neighboring point density features, adaptively and dynamically deletes noise points in point cloud data based on neighboring point density features, and implements voxel grid filtering for multi-scale filtering processing; Point cloud detection module: The point cloud data processed by multi-scale filtering is converted into a three-dimensional mesh model inside the optical cable junction box through Poisson reconstruction. The convolutional neural network is used to apply the joint position features, cable direction features, geometric state features and structural information features, and the probability distribution of abnormal monitoring of the optical cable junction box is calculated to generate early warning information; Visual monitoring module: Convert the coordinates of the abnormal position into the abnormal position of the three-dimensional grid model according to the early warning information, create a three-dimensional visualization interface, add support for user interactive operations, and dispatch professional maintenance personnel to conduct a comprehensive inspection.

2. The detection and early warning system for an optical cable junction box according to claim 1, characterized in that: In the laser scanning module, the laser scanner is installed in the optical cable junction box, the scanning resolution is set to 2000000 PPS and the accuracy is ±1 mm, the laser scanner is started and the scanning time is set to 5 minutes, the position of the laser scanner is adjusted to perform multi-angle scanning, and the laser beam is fully covered by 5 different scanning angles. The discrete point coordinates and reflection intensity information of the intersection of the laser beam and the target surface inside the optical cable junction box are recorded, and point cloud data is generated by receiving the reflection signal.

3. The detection and early warning system for an optical cable junction box according to claim 1, characterized in that: In the point cloud processing module, a tree structure is constructed, and the branch nodes inside the tree structure represent the invalid point features of the point cloud data, and the leaf nodes inside the tree structure represent the invalid removal results. Multiple decision trees are constructed by randomly selecting invalid point features and point cloud data, and the invalid point features of the branch nodes are randomly selected according to the leaf nodes for optimal splitting. Independent classification prediction is performed on each decision tree, and the classification prediction that appears most in all decision trees is selected as the invalid point classification. The invalid points of the point cloud data are judged and removed, and the neighboring point density features of the point cloud data are extracted. The neighboring point radius is dynamically adjusted according to the local changes of the neighboring point density features. The minimum number of neighboring points is selected through the local dynamic selection of the point cloud data, and the distance from each point to other points is calculated through the Euclidean distance metric algorithm. The KD tree is used to check the neighboring points of each point. The number of points in the neighborhood of the nearest point radius. When the number of points in the neighborhood is ≥ the minimum number of neighboring points, the point is judged as a core point and marked. The empty cluster list is initialized. If the current core point has not been marked and does not belong to any cluster, a new cluster is created and the current core point and the points in the neighborhood with the neighboring point radius are added to the new cluster. The new cluster is added recursively. The current non-core point and the point in the neighborhood with the neighboring point radius of a cluster are marked as boundary points. All points that are not assigned to any cluster are marked as noise points and deleted. Voxel grid filtering is implemented. The voxel size is dynamically adjusted according to the density characteristics of the neighboring points of the point cloud data. The voxel size is set to 0.5 times the average distance between the points of the point cloud data. The largest voxel size is applied for multi-scale filtering, and the voxel size is gradually reduced for multi-scale filtering.

4. The detection and early warning system for an optical cable junction box according to claim 1, characterized in that: In the point cloud detection module, the point cloud data is converted into a three-dimensional mesh model inside the optical cable junction box through Poisson reconstruction, the mesh optimization tool is used to remove redundant vertices and faces of the three-dimensional mesh model, the sharp edges and unnecessary details of the three-dimensional mesh model are removed through Laplace smoothing, the three-dimensional mesh model is modified with carving tools, and bumps and subtle depictions are added, and the texture coordinates are added and mapped through UV unfolding according to the color information of the point cloud data.

5. The detection and early warning system for an optical cable junction box according to claim 1, characterized in that: In the point cloud detection module, the joint position features of the three-dimensional grid model inside the optical cable junction box are extracted by color marking, and the three-dimensional coordinate information of the joint position is recorded. The cable direction features of the three-dimensional grid model inside the optical cable junction box are extracted by path lines, and the three-dimensional coordinate information, length and curvature of the cable direction are recorded. The geometric state features and structural information features of the optical cable junction box are extracted, and a convolutional neural network model is established. A deformable convolution kernel is added to slide on the three-dimensional grid model, and the joint position features, cable direction features, geometric state features and structural information features are applied. The Swish activation function is introduced, and each feature is averagely pooled in the spatial dimension. Each feature output by the convolution layer and the pooling layer is flattened into a one-dimensional vector and connected to the fully connected layer for classification. The probability distribution of abnormal monitoring of the optical cable junction box is calculated using the softmax function, and the probability distribution threshold is set to 0.

5. When the probability distribution is higher than the probability distribution threshold, it is marked as an abnormality inside the optical cable junction box, and an early warning information is generated.

6. The detection and early warning system for an optical cable junction box according to claim 5, characterized in that: The specific formula of the convolution operation is: ； in, represents an element in each feature, Represents the three-dimensional coordinate information of the three-dimensional mesh model, represents the weight of the convolution kernel, Respectively represent the position index of the convolution kernel on the three-dimensional grid model, Represents the index of each feature in height, depth and width on the 3D mesh model respectively.

7. The detection and early warning system for an optical cable junction box according to claim 5, characterized in that: The specific formula for calculating the probability distribution of abnormal monitoring of the optical cable junction box using the softmax function is: ； in, It represents the probability that the output category O is g given the input Q. represents the Swish activation function value, Indicates the information level value of the current abnormal category, Represents the total number of abnormal categories.

8. The detection and early warning system for an optical cable junction box according to claim 1, characterized in that: In the visual monitoring module, a three-dimensional grid model is imported and a three-dimensional scene is added. The coordinates of the abnormal position are converted into the abnormal position of the three-dimensional grid model according to the early warning information, a three-dimensional visualization interface is created, and support for user interactive operations is added. The inside of the optical cable junction box is freely rotated, enlarged and reduced. According to the early warning information and the three-dimensional scene, an on-site inspection plan for the optical cable junction box is formulated, and professional maintenance personnel are dispatched to conduct a comprehensive inspection. The abnormal position of the three-dimensional grid model is used as on-site evidence to implement maintenance of the optical cable junction box, and on-site feedback information is recorded.

Images