Intelligent wiring automatic laying method based on image recognition

Abstract

The invention discloses an intelligent wiring automatic laying method based on image recognition, and the method comprises the following steps: S1, collecting a multi-angle image, recognizing an obstacle, a passing region and an interface point, and constructing a wiring environment diagram; s2, constructing graph structure representation based on the environment graph and the wiring parameters; s3, inputting the graph structure into a graph attention gating capsule network, extracting path features, and generating candidate paths; s4, performing multi-objective optimization on the candidate path by adopting a variational singular ant colony optimization algorithm to obtain an optimal wiring path; s5, generating a wiring action instruction sequence containing a node control instruction and pose information; s6, controlling the wiring equipment to execute and detect deviation in real time, and re-planning a path if deviation occurs; and S7, after wiring is completed, image verification and path consistency comparison are carried out, and an evaluation report is output. According to the invention, wiring automation, high-precision path planning and dynamic execution control in a complex environment are realized.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to an intelligent wiring automatic laying method based on image recognition. Background Art

[0002] Against the backdrop of the rapid development of intelligent manufacturing and industrial automation, cabling, a critical infrastructure deployment link in scenarios such as industrial sites, power systems, buildings, and data centers, has a direct impact on system construction efficiency and operational reliability. Traditional cabling methods rely primarily on manual measurement and construction, which often suffers from low efficiency, suboptimal routing, and high error rates when faced with complex environments or large-scale cabling tasks. In recent years, with advances in robotic control and image recognition technologies, some systems have attempted to introduce cabling equipment to perform automated laying operations and combine image navigation for path tracking, but these systems still face key technical bottlenecks.

[0003] In existing technologies, wiring path planning generally relies on manual annotation or pre-built CAD models for path extraction, lacking the ability to perceive complex spatial environments in real time, and is particularly difficult to complete path adaptation and obstacle avoidance in dynamic or unstructured scenarios. At the same time, most path planning methods use traditional heuristic algorithms or search mechanisms based on fixed cost functions, which makes it difficult to achieve multi-objective optimization between path length, obstacle avoidance capability, and traffic stability. As a result, the generated paths are difficult to meet the unified requirements of construction feasibility and cost rationality. Although some studies have introduced graph neural networks or deep learning models for path evaluation, the structural modeling of the paths is insufficient and there is a lack of accurate expression of spatial topological details. In addition, open-loop control modes are generally used during path execution, and no dynamic perception mechanism for environmental changes and path deviations has been established. There is a lack of closed-loop adjustment capabilities, which can easily lead to path failure and construction interruption during execution.

[0004] In response to the above-mentioned shortcomings, the present invention proposes an automatic laying method for intelligent wiring based on image recognition, which constructs a wiring environment graph through multi-angle image acquisition and semantic segmentation, combines graph structure representation with capsule recurrent convolutional network model to extract path features, and introduces graph attention gating mechanism to enhance the modeling ability of spatial structure. At the same time, a variational singular ant colony optimization algorithm is used to perform multi-objective optimization on candidate paths to achieve a balance between path connectivity, obstacle avoidance integrity and length cost. During the execution process, a path feedback closed-loop system is constructed, path deviation detection is performed through real-time image recognition, and the action instruction sequence is dynamically updated based on the local subgraph, thereby improving wiring accuracy and environmental adaptability. The present invention effectively solves the defects of the existing technology in perception accuracy, path optimization capability and execution robustness, and has a stronger level of automation and breadth of applicability.

[0005] Therefore, how to provide an intelligent wiring automatic laying method based on image recognition is a problem that those skilled in the art urgently need to solve. Summary of the Invention

[0006] One purpose of the present invention is to propose an intelligent wiring automatic laying method based on image recognition. The present invention fully combines multi-angle image acquisition, semantic segmentation modeling, graph structure path expression, graph attention gated capsule network and variational singular ant colony optimization algorithm, and describes in detail the full process control mechanism of wiring environment perception, path planning and automatic execution. It has the advantages of strong environmental adaptability, high path planning accuracy and high degree of intelligent wiring execution.

[0007] According to an embodiment of the present invention, a method for automatic intelligent wiring laying based on image recognition includes the following steps:

[0008] S1. Collect image data from several angles of the wiring area, use a semantic segmentation model to identify obstacles, traffic areas, and cable interface points in the image, and construct a wiring environment map containing spatial connectivity information and interface coordinates;

[0009] S2. constructing a graph structure representation for path planning based on the wiring environment graph and the starting position, end position, and wiring preference parameters set by the user;

[0010] S3. Input the graph structure representation into a capsule recurrent convolutional network model containing a graph attention gated capsule layer to extract routing path features and generate a set of candidate paths.

[0011] S4. Using a variational singular ant colony optimization algorithm to perform multi-objective optimization on the candidate path set, and screen out the optimal wiring path based on indicators such as path connectivity, obstacle avoidance integrity, and path length cost;

[0012] S5. Generate a corresponding wiring action instruction sequence according to the optimal wiring path, wherein the instruction sequence includes an execution order, node control instructions, and target posture information;

[0013] S6. Controlling the wiring device to execute the wiring operation according to the wiring action instruction sequence, and collecting images in real time during the execution process, comparing the current execution status with the optimal wiring path, and triggering local path replanning if path deviation or environmental changes are detected, generating and updating a new action instruction sequence;

[0014] S7. After the wiring is completed, the final path is verified by image recognition and compared with the path consistency, the wiring completeness rate, execution accuracy and connection efficiency are calculated, and an evaluation report is output.

[0015] Optionally, the S1 specifically includes: collecting image data from several angles of the wiring area, unifying the image resolution and acquisition time; performing grayscale standardization, contrast enhancement, illumination balance and distortion correction processing on the image, inputting the preprocessed image into the semantic segmentation neural network model, identifying obstacles, passage areas and cable interface points, and generating a semantic label map, performing spatial fusion based on the multi-perspective semantic label map, constructing a three-dimensional spatial semantic structure, marking the spatial position and boundary information of each element, extracting the position of the connected passage areas and cable interface points, and establishing a wiring environment map containing obstacle distribution, passage area structure and interface point coordinates.

[0016] Optionally, the S2 specifically includes:

[0017] S21. Receive a routing start point and an end point input by a user, wherein the start point and the end point are both represented by three-dimensional spatial coordinates, and are matched to spatial nodes with connectivity properties in a routing environment graph, respectively, as the start node and the target node of the path planning;

[0018] S22. Obtaining routing preference parameters set by the user, wherein the preference parameters include a path length priority weight, an obstacle avoidance priority weight, and a traffic reliability constraint coefficient, which are used to perform weighted control on the importance of path length, obstacle avoidance capability, and stable traffic capability in subsequent path evaluation;

[0019] S23. Based on the wiring environment diagram, and according to the identified traffic area boundaries and obstacle distribution, a graph structure representation model is constructed. The model includes a set of spatial nodes and a set of edges. Nodes correspond to location units or interface points that meet traffic conditions, and edges are used to describe feasible path segments between any two connected nodes.

[0020] S24. Assign attribute indicators such as path length, obstacle proximity, and traffic stability to each edge in the graph structure to form an edge attribute set with multi-dimensional attributes to support subsequent path evaluation and optimization processing;

[0021] S25. Combine the start node and the target node, the routing preference parameters, and the edge attribute set to establish a path cost function model and generate a path planning graph structure model.

[0022] Optionally, the step S25 of establishing a path cost function model specifically includes:

[0023] S251. For each graph structure edge in the wiring environment graph, extract three attribute indicators: path length, obstacle proximity, and traffic stability. These indicators correspond to the path length factor, obstacle avoidance factor, and stability factor, respectively, and serve as basic attribute data for path cost evaluation.

[0024] S252: Define the path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient set by the user as weighted coefficients in the path cost function, and correspond them one-to-one with the above three attributes to control the weight ratio of each attribute in the path score;

[0025] S253. Linearly combine the three attribute factors and the corresponding weighting coefficients to construct a path cost function. The path cost function is used to calculate the comprehensive travel cost of any path segment in the graph structure as a basis for judging the quality of the path.

[0026] S254. Apply the path cost function to all edges in the graph structure one by one to generate a complete set of edge cost values, and use this to construct a path cost matrix, which serves as the basic input for path scoring in the path feature extraction process.

[0027] Optionally, the S3 specifically includes:

[0028] S31. Based on the graph structure model constructed by the wiring environment graph, each spatial node is used as an initial input unit, and a structural connection between nodes is established through the graph adjacency relationship, and an adjacency tensor structure between nodes is constructed to represent spatial connectivity;

[0029] S32. In the model input stage, a graph embedding method is used to vector encode each spatial node in the graph structure. The encoding result is used as the input of the graph attention gated capsule layer. The graph attention gated capsule layer is composed of multiple sublayers, each of which includes a graph attention module, a capsule transformation module and a gated activation unit;

[0030] S33, the graph attention module calculates the attention weight distribution based on the spatial adjacency weight and node feature similarity, realizes the dynamic adjustment of the local spatial path correlation, and outputs the weighted aggregated node representation;

[0031] S34, the capsule transformation module performs a posture matrix transformation on the aggregated node representation to generate a capsule feature vector containing path directionality and node status;

[0032] S35, the gate activation unit dynamically adjusts the information flow weight of each feature channel according to the distribution state of the capsule feature vector, thereby achieving adaptive control of the significance of different path features;

[0033] S36. Input the output representation after processing by the graph attention gated capsule layer into the recurrent convolution layer for temporal feature fusion, extract the path temporal dependency in the wiring graph structure, output the wiring path features, and generate a set of candidate paths.

[0034] Optionally, the S4 specifically includes:

[0035] S41, using the candidate path set as the initial solution space for optimization, initializing the ant colony individuals, each ant colony individual representing a specific candidate path, setting the maximum number of iterations and the initial pheromone value;

[0036] S42. Calculate a path connectivity score, an obstacle avoidance integrity score, and a path length for each candidate path to form three factors for path cost evaluation, which are used to evaluate whether the path is fully accessible, whether it effectively avoids obstacles, and whether it is economical to lay.

[0037] S43. Calculate the comprehensive cost value for each path based on the set path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient, which are used as weight parameters of the path comprehensive scoring function. The path cost value function is defined as:

[0038] F(p)=w L L(p)+w O O(p)+w S S(p);

[0039] Among them, p represents the path individual, L(p) represents the path length, O(p) represents the obstacle avoidance completeness, S(p) represents the path connectivity, and w L 、w O 、w S are the weight coefficients of path length, obstacle avoidance and connectivity respectively;

[0040] S44. In each round of path search, a singular potential guidance mechanism is introduced to apply a guiding potential field to the node area near the obstacle boundary in the path to prevent the path from passing through high-risk areas, thereby improving the safety and practical feasibility of the path.

[0041] S45. Based on the results of the path cost function and the singular guidance mechanism, the ant colony is guided to perform path search and pheromone update, and the variational strategy is used to adaptively adjust the pheromone evaporation rate and the probability distribution of individual path selection;

[0042] S46. Update the optimal path record after each round of search, and output the path with the minimum path cost as the final optimal routing path after the termination condition is met.

[0043] Optionally, the variational strategy described in S45 adaptively adjusts the pheromone evaporation rate and the individual path selection probability distribution, specifically including: dynamically adjusting the pheromone evaporation rate according to the variance of the path cost value in the current ant colony, reducing the evaporation rate if the variance is higher than a preset first threshold, and increasing the evaporation rate if the variance is lower than a preset second threshold; and adjusting the path selection probability according to the path cost value ranking result.

[0044] Optionally, the S6 specifically includes:

[0045] S61, inputting the generated wiring action instruction sequence into the wiring device control module, driving the device to execute path segment laying along the optimal wiring path, and collecting wiring area images in real time;

[0046] S62, inputting the wiring area image into a lightweight recognition module with the same structure as the semantic segmentation model, extracting the current wiring device spatial position, and comparing it with the coordinates of the corresponding path segment of the optimal wiring path;

[0047] S63: Calculate the spatial offset distance between the current wiring device spatial position and the target path segment. If it exceeds a set threshold, determine it as a path offset or environment change event.

[0048] S64. Based on the graph structure representation and the path cost function model, a local subgraph is constructed with the current coordinate as the starting point, and a variational singular ant colony optimization algorithm is called to generate a local optimal path segment;

[0049] S65. Convert the local optimal path segment into a wiring action instruction subsequence and dynamically insert it into the control sequence to achieve real-time correction of the original wiring process and generate and update a new action instruction sequence.

[0050] The beneficial effects of the present invention are:

[0051] First, the present invention combines multi-angle image acquisition with a semantic segmentation model to achieve accurate identification of obstacles, passages, and cable interface points in the wiring area. It can automatically construct a wiring environment map containing spatial connectivity information, avoiding the inflexibility of traditional reliance on manual modeling or CAD drawings, and significantly improving environmental adaptability and automation levels.

[0052] Secondly, the present invention introduces a graph attention gated capsule recurrent convolutional network model to extract structured features of the wiring path, and combines it with the variational singular ant colony optimization algorithm to perform multi-objective joint optimization of the candidate path set. It can achieve a dynamic trade-off between path connectivity, obstacle avoidance integrity and path length cost, thereby generating a better, safer and more constructible wiring path solution.

[0053] In addition, the present invention constructs an execution closed-loop feedback mechanism, which collects images in real time during the execution of the wiring equipment, judges the position deviation against the optimal path, and generates a new wiring action instruction subsequence through local replanning, ensuring that the wiring task can still be completed stably when the environment changes or the path shifts, significantly improving the robustness and intelligent response capabilities of the wiring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0055] Figure 1 This is an overall flow chart of an intelligent wiring automatic laying method based on image recognition proposed by the present invention;

[0056] Figure 2 This is a graph attention gated capsule recurrent convolutional network structure for an intelligent wiring automatic laying method based on image recognition proposed by the present invention;

[0057] Figure 3 This is a path optimization flowchart of the intelligent wiring automatic laying method based on image recognition proposed by the present invention. DETAILED DESCRIPTION

[0058] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0059] refer to Figure 1-3 , an intelligent wiring automatic laying method based on image recognition, comprising the following steps:

[0060] S1. Collect image data from several angles of the wiring area, use a semantic segmentation model to identify obstacles, traffic areas, and cable interface points in the image, and construct a wiring environment map containing spatial connectivity information and interface coordinates;

[0061] S2. constructing a graph structure representation for path planning based on the wiring environment graph and the starting position, end position, and wiring preference parameters set by the user;

[0062] S3. Input the graph structure representation into a capsule recurrent convolutional network model containing a graph attention gated capsule layer to extract routing path features and generate a set of candidate paths.

[0063] S4. Using a variational singular ant colony optimization algorithm to perform multi-objective optimization on the candidate path set, and screen out the optimal wiring path based on indicators such as path connectivity, obstacle avoidance integrity, and path length cost;

[0064] S5. Generate a corresponding wiring action instruction sequence according to the optimal wiring path, wherein the instruction sequence includes an execution order, node control instructions, and target posture information;

[0065] S6. Controlling the wiring device to execute the wiring operation according to the wiring action instruction sequence, and collecting images in real time during the execution process, comparing the current execution status with the optimal wiring path, and triggering local path replanning if path deviation or environmental changes are detected, generating and updating a new action instruction sequence;

[0066] S7. After the wiring is completed, the final path is verified by image recognition and compared with the path consistency, the wiring completeness rate, execution accuracy and connection efficiency are calculated, and an evaluation report is output.

[0067] By building a complete closed-loop wiring system from image recognition, path modeling, feature extraction, path optimization to automatic execution, this invention has end-to-end automation and dynamic adaptability compared to existing technologies that rely on manual design or static path planning, significantly improving wiring efficiency and execution accuracy, and is suitable for intelligent wiring scenarios in a variety of complex environments.

[0068] In this embodiment, S1 specifically includes: collecting image data from several angles of the wiring area, unifying the image resolution and acquisition time; performing grayscale standardization, contrast enhancement, illumination balance and distortion correction on the image, inputting the preprocessed image into the semantic segmentation neural network model, identifying obstacles, passage areas and cable interface points, and generating a semantic label map, performing spatial fusion based on the multi-perspective semantic label map, constructing a three-dimensional spatial semantic structure, marking the spatial position and boundary information of each element, extracting the position of the connected passage areas and cable interface points, and establishing a wiring environment map containing obstacle distribution, passage area structure and interface point coordinates.

[0069] The present invention utilizes multi-angle image acquisition combined with a semantic segmentation model to achieve accurate identification of obstacles, passage areas, and interface points, and construct a three-dimensional semantic environment map, avoiding traditional CAD dependence and manual labeling. It has the advantages of automated environment construction, wide adaptation scenarios, and accurate semantic understanding, thereby improving the perception quality before wiring.

[0070] In this embodiment, S2 specifically includes:

[0071] S21. Receive a routing start point and an end point input by a user, wherein the start point and the end point are both represented by three-dimensional spatial coordinates, and are matched to spatial nodes with connectivity properties in a routing environment graph, respectively, as the start node and the target node of the path planning;

[0072] S22. Obtaining routing preference parameters set by the user, wherein the preference parameters include a path length priority weight, an obstacle avoidance priority weight, and a traffic reliability constraint coefficient, which are used to perform weighted control on the importance of path length, obstacle avoidance capability, and stable traffic capability in subsequent path evaluation;

[0073] S23. Based on the wiring environment diagram, and according to the identified traffic area boundaries and obstacle distribution, a graph structure representation model is constructed. The model includes a set of spatial nodes and a set of edges. Nodes correspond to location units or interface points that meet traffic conditions, and edges are used to describe feasible path segments between any two connected nodes.

[0074] S24. Assign attribute indicators such as path length, obstacle proximity, and traffic stability to each edge in the graph structure to form an edge attribute set with multi-dimensional attributes to support subsequent path evaluation and optimization processing;

[0075] S25. Combine the start node and the target node, the routing preference parameters, and the edge attribute set to establish a path cost function model and generate a path planning graph structure model.

[0076] The present invention constructs a graph structure model based on the wiring environment graph, expresses spatial connectivity and path attributes through nodes and edges, and generates a path planning graph in combination with user preference parameters, providing structured input for subsequent neural networks and optimization algorithms. It has the advantages of reasonable modeling form, rich graph expression information, and flexible path target trade-offs.

[0077] In this embodiment, the specific contents of establishing the path cost function model in S25 include:

[0078] S251. For each graph structure edge in the wiring environment graph, extract three attribute indicators: path length, obstacle proximity, and traffic stability. These indicators correspond to the path length factor, obstacle avoidance factor, and stability factor, respectively, and serve as basic attribute data for path cost evaluation.

[0079] S252: Define the path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient set by the user as weighted coefficients in the path cost function, and correspond them one-to-one with the above three attributes to control the weight ratio of each attribute in the path score;

[0080] S253. Linearly combine the three attribute factors and the corresponding weighting coefficients to construct a path cost function. The path cost function is used to calculate the comprehensive travel cost of any path segment in the graph structure as a basis for judging the quality of the path.

[0081] S254. Apply the path cost function to all edges in the graph structure one by one to generate a complete set of edge cost values, and use this to construct a path cost matrix, which serves as the basic input for path scoring in the path feature extraction process.

[0082] The present invention constructs a cost function model that integrates path length, obstacle avoidance integrity and traffic connectivity, and implements multi-objective constraints through user-set weight parameters, replacing the traditional single cost optimization method. It has the significant advantages of more practical path generation, strong trade-off ability, and adaptability to different wiring requirements.

[0083] In this embodiment, S3 specifically includes:

[0084] S31. Based on the graph structure model constructed by the wiring environment graph, each spatial node is used as an initial input unit, and a structural connection between nodes is established through the graph adjacency relationship, and an adjacency tensor structure between nodes is constructed to represent spatial connectivity;

[0085] S32. In the model input stage, a graph embedding method is used to vector encode each spatial node in the graph structure. The encoding result is used as the input of the graph attention gated capsule layer. The graph attention gated capsule layer is composed of multiple sublayers, each of which includes a graph attention module, a capsule transformation module and a gated activation unit;

[0086] S33, the graph attention module calculates the attention weight distribution based on the spatial adjacency weight and node feature similarity, realizes the dynamic adjustment of the local spatial path correlation, and outputs the weighted aggregated node representation;

[0087] S34, the capsule transformation module performs a posture matrix transformation on the aggregated node representation to generate a capsule feature vector containing path directionality and node status;

[0088] S35, the gate activation unit dynamically adjusts the information flow weight of each feature channel according to the distribution state of the capsule feature vector, thereby achieving adaptive control of the significance of different path features;

[0089] S36. Input the output representation after processing by the graph attention gated capsule layer into the recurrent convolution layer for temporal feature fusion, extract the path temporal dependency in the wiring graph structure, output the wiring path features, and generate a set of candidate paths.

[0090] This paper introduces the graph attention gated capsule network for the first time to extract features from wiring paths, integrating the graph attention mechanism with capsule dynamic routing to capture path directionality and local correlation while maintaining spatial topology. It has the innovative advantages of strong expressiveness, high feature abstraction ability, and accurate modeling of path structure details.

[0091] In this embodiment, the S4 specifically includes:

[0092] S41, using the candidate path set as the initial solution space for optimization, initializing the ant colony individuals, each ant colony individual representing a specific candidate path, setting the maximum number of iterations and the initial pheromone value;

[0093] S42. Calculate a path connectivity score, an obstacle avoidance integrity score, and a path length for each candidate path to form three factors for path cost evaluation, which are used to evaluate whether the path is fully accessible, whether it effectively avoids obstacles, and whether it is economical to lay.

[0094] S43. Calculate the comprehensive cost value for each path based on the set path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient, which are used as weight parameters of the path comprehensive scoring function. The path cost value function is defined as:

[0095] F(p)=w L L(p)+w O O(p)+w S S(p);

[0096] Among them, p represents the path individual, L(p) represents the path length, O(p) represents the obstacle avoidance completeness, S(p) represents the path connectivity, and w L 、w O 、w S are the weight coefficients of path length, obstacle avoidance and connectivity respectively;

[0097] S44. In each round of path search, a singular potential guidance mechanism is introduced to apply a guiding potential field to the node area near the obstacle boundary in the path to prevent the path from passing through high-risk areas, thereby improving the safety and practical feasibility of the path.

[0098] S45. Based on the results of the path cost function and the singular guidance mechanism, the ant colony is guided to perform path search and pheromone update, and the variational strategy is used to adaptively adjust the pheromone evaporation rate and the probability distribution of individual path selection;

[0099] S46. Update the optimal path record after each round of search, and output the path with the minimum path cost as the final optimal routing path after the termination condition is met.

[0100] The present invention designs an ant colony optimization algorithm that integrates singular potential guidance and variational scheduling mechanism. It can balance connectivity, obstacle avoidance and path cost in multi-objective path optimization, avoid local optimality, and has the advantages of high search efficiency, strong global optimization capability and good path practicality, breaking through the limitations of traditional heuristic optimization.

[0101] In this embodiment, the variational strategy described in S45 adaptively adjusts the pheromone evaporation rate and the individual path selection probability distribution, specifically including: dynamically adjusting the pheromone evaporation rate according to the variance of the path cost value in the current ant colony, reducing the evaporation rate when the variance is higher than a preset first threshold, and increasing the evaporation rate when the variance is lower than a preset second threshold; and adjusting the path selection probability according to the path cost value ranking result.

[0102] The present invention realizes dynamic scheduling capability in the optimization process by introducing a pheromone evaporation rate and path selection probability adjustment mechanism driven by the variance of path cost value. It has the innovative advantages of responding quickly to changes in path distribution, improving convergence stability, and maintaining path diversity, thus solving the problem that ant colonies are prone to falling into local optimality.

[0103] In this embodiment, S6 specifically includes:

[0104] S61, inputting the generated wiring action instruction sequence into the wiring device control module, driving the device to execute path segment laying along the optimal wiring path, and collecting wiring area images in real time;

[0105] S62, inputting the wiring area image into a lightweight recognition module with the same structure as the semantic segmentation model, extracting the current wiring device spatial position, and comparing it with the coordinates of the corresponding path segment of the optimal wiring path;

[0106] S63: Calculate the spatial offset distance between the current wiring device spatial position and the target path segment. If it exceeds a set threshold, determine it as a path offset or environment change event.

[0107] S64. Based on the graph structure representation and the path cost function model, a local subgraph is constructed with the current coordinate as the starting point, and a variational singular ant colony optimization algorithm is called to generate a local optimal path segment;

[0108] S65. Convert the local optimal path segment into a wiring action instruction subsequence and dynamically insert it into the control sequence to achieve real-time correction of the original wiring process and generate and update a new action instruction sequence.

[0109] The present invention establishes a feedback closed-loop system during the wiring execution process, identifies the device status in real time and triggers local path reconstruction, realizes dynamic adjustment of action instructions, and has the advantages of strong fault tolerance, high environmental adaptability, and good task continuity, breaking through the technical difficulty of existing open-loop control path failure that cannot be corrected.

[0110] Example 1:

[0111] In order to verify the feasibility of the present invention in practice, the present invention was applied to the automatic cable laying operation in a certain industrial equipment control center. The scene includes multiple power distribution control cabinets, communication equipment racks, and obstacles such as pipes, columns, and ventilation shafts in the environment. The spatial structure is complex and the wiring paths are irregular. Traditional manual mapping and wiring methods are not only time-consuming, but also have problems such as misconnection, redundant winding, and insufficient safety clearance. For the above typical wiring environment, the image recognition-based intelligent wiring automatic laying method provided by the present invention uses a multi-angle camera array to capture images of the wiring area, and uses a semantic segmentation model to complete the intelligent identification of obstacles, passage areas, and cable interface points, thereby automatically generating a visual wiring environment map.

[0112] The system uses graph modeling to transform the environment graph into a node-edge structure. Nodes represent key locations or device interfaces with spatial accessibility, while edges represent possible routing path segments. The path cost function incorporates three metrics: path length, obstacle avoidance integrity, and access reliability. Furthermore, the graph-attention gated capsule network designed by the present invention extracts path features and outputs a set of candidate paths. A variational singular ant colony optimization algorithm then performs multi-objective screening on these candidate paths, ultimately selecting the optimal routing path and generating the required routing action instruction sequence for device execution.

[0113] In actual deployment, wiring equipment performs wiring operations according to generated action instructions. Simultaneously, a lightweight image recognition module acquires wiring position data in real time. If any deviation is detected after comparison with the preset path, the local graph structure is updated and the path replanning mechanism is immediately initiated. The entire wiring process is completed smoothly without human intervention, and the system demonstrates a high degree of automation and stability in all four stages: path identification, generation, execution, and correction.

[0114] In actual comparative tests, the performance of the proposed method was compared with traditional manual wiring and a conventional static path algorithm. Manual wiring took an average of approximately 82 minutes, with an average of four misconnected nodes and a total path length of approximately 42.7 meters. The static algorithm took 56 minutes, with two misconnected nodes and a path length of 41.1 meters. In contrast, the proposed system completed the task in just 37 minutes under the same conditions, with zero misconnected nodes and a total path length of 39.4 meters. Furthermore, its obstacle avoidance success rate reached 98.6%, and the routing trajectory duplication rate was only 3.2%, significantly outperforming other solutions.

[0115] The following table summarizes key data collected during the on-site wiring process, based on the average of three consecutive rounds of repeated experiments. This table demonstrates the comprehensive improvements achieved by the present invention in terms of construction efficiency, execution accuracy, and path optimization quality.

[0116] Table 1: Performance comparison test results of three wiring schemes

[0117]

[0118]

[0119] Regarding “Table 1: Performance comparison experimental results of three wiring schemes”, a comprehensive analysis can be conducted from four aspects: wiring efficiency, accuracy, path optimization quality and overall control capability to verify the significant advantages of the present invention in practical applications.

[0120] In terms of wiring efficiency, the average wiring time of the method of the present invention is 37 minutes, which is much lower than the 56 minutes of the static path algorithm and the 82 minutes of the manual wiring solution. It reflects the automation capability of the end-to-end path planning and execution process, and effectively reduces the time consumed by manual operation and path determination.

[0121] In terms of routing accuracy, the present invention achieved zero misconnected nodes in three rounds of experiments, outperforming both the static path algorithm (two misconnections) and manual routing (four misconnections). This demonstrates that the path features extracted by the graph attention gated capsule network have high practical feasibility, and combined with the variational singular ant colony optimization mechanism, they can ensure a high degree of consistency in path design during construction.

[0122] In terms of path optimization quality, the total length of the path generated by the present invention is 39.4 meters, which is shorter than the static path algorithm (41.1 meters) and manual wiring (42.7 meters), and the wiring track repetition rate is only 3.2%, indicating that it can effectively avoid path redundancy and invalid winding, reflecting the advantages of path accuracy and economy.

[0123] Furthermore, in terms of routing control capabilities, the present invention supports a dynamic path adjustment mechanism, enabling two replannings of action instructions during execution, a capability neither static algorithms nor manual methods possess. This ensures path continuity and routing integrity despite environmental changes or execution deviations. Ultimately, the present invention achieved a 100% overall routing integrity rate, significantly more reliable than static algorithms (94.7%) and manual routing (89.2%).

[0124] In summary, the table fully reflects the comprehensive advantages of the present invention in terms of efficiency, accuracy, intelligent path control and environmental adaptation, and has practical engineering application value in replacing traditional wiring methods in complex industrial scenarios.

[0125] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An intelligent wiring automatic laying method based on image recognition, characterized in that: The steps include: S1. Collect image data from several angles of the wiring area, use a semantic segmentation model to identify obstacles, traffic areas, and cable interface points in the image, and construct a wiring environment map containing spatial connectivity information and interface coordinates; S2. constructing a graph structure representation for path planning based on the wiring environment graph and the starting position, end position, and wiring preference parameters set by the user; S3. Input the graph structure representation into a capsule recurrent convolutional network model containing a graph attention gated capsule layer to extract routing path features and generate a set of candidate paths. S4. Using a variational singular ant colony optimization algorithm to perform multi-objective optimization on the candidate path set, and screen out the optimal wiring path based on indicators such as path connectivity, obstacle avoidance integrity, and path length cost; S5. Generate a corresponding wiring action instruction sequence according to the optimal wiring path, wherein the instruction sequence includes an execution order, node control instructions, and target posture information; S6. Controlling the wiring device to execute the wiring operation according to the wiring action instruction sequence, and collecting images in real time during the execution process, comparing the current execution status with the optimal wiring path, and triggering local path replanning if path deviation or environmental changes are detected, generating and updating a new action instruction sequence; S7. After the wiring is completed, the final path is verified by image recognition and compared with the path consistency, the wiring completeness rate, execution accuracy and connection efficiency are calculated, and an evaluation report is output.

2. The method for automatic intelligent wiring laying based on image recognition according to claim 1, characterized in that: The S1 specifically includes: collecting image data from several angles of the wiring area, unifying the image resolution and acquisition time; performing grayscale standardization, contrast enhancement, illumination balance and distortion correction on the image, inputting the preprocessed image into the semantic segmentation neural network model, identifying obstacles, passage areas and cable interface points, and generating a semantic label map, performing spatial fusion based on the multi-view semantic label map, constructing a three-dimensional spatial semantic structure, marking the spatial position and boundary information of each element, extracting the position of the connected passage area and the cable interface point, and establishing a wiring environment map including obstacle distribution, passage area structure and interface point coordinates.

3. The method for automatic intelligent wiring laying based on image recognition according to claim 1, characterized in that: The S2 specifically includes: S21. Receive a routing start point and an end point input by a user, wherein the start point and the end point are both represented by three-dimensional spatial coordinates, and are matched to spatial nodes with connectivity properties in a routing environment graph, respectively, as the start node and the target node of the path planning; S22. Obtaining routing preference parameters set by the user, wherein the preference parameters include a path length priority weight, an obstacle avoidance priority weight, and a traffic reliability constraint coefficient, which are used to perform weighted control on the importance of path length, obstacle avoidance capability, and stable traffic capability in subsequent path evaluation; S23. Based on the wiring environment diagram, and according to the identified traffic area boundaries and obstacle distribution, a graph structure representation model is constructed. The model includes a set of spatial nodes and a set of edges. Nodes correspond to location units or interface points that meet traffic conditions, and edges are used to describe feasible path segments between any two connected nodes. S24. Assign attribute indicators such as path length, obstacle proximity, and traffic stability to each edge in the graph structure to form an edge attribute set with multi-dimensional attributes to support subsequent path evaluation and optimization processing; S25. Combine the start node and the target node, the routing preference parameters, and the edge attribute set to establish a path cost function model and generate a path planning graph structure model.

4. The method for automatic intelligent wiring laying based on image recognition according to claim 3, characterized in that: The specific contents of establishing the path cost function model in S25 include: S251. For each graph structure edge in the wiring environment graph, extract three attribute indicators: path length, obstacle proximity, and traffic stability. These indicators correspond to the path length factor, obstacle avoidance factor, and stability factor, respectively, and serve as basic attribute data for path cost evaluation. S252: Define the path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient set by the user as weighted coefficients in the path cost function, and correspond them one-to-one with the above three attributes to control the weight ratio of each attribute in the path score; S253. Linearly combine the three attribute factors and the corresponding weighting coefficients to construct a path cost function. The path cost function is used to calculate the comprehensive travel cost of any path segment in the graph structure as a basis for judging the quality of the path. S254. Apply the path cost function to all edges in the graph structure one by one to generate a complete set of edge cost values, and use this to construct a path cost matrix, which serves as the basic input for path scoring in the path feature extraction process.

5. The method for automatic intelligent wiring laying based on image recognition according to claim 1, characterized in that: The S3 specifically includes: S31. Based on the graph structure model constructed by the wiring environment graph, each spatial node is used as an initial input unit, and a structural connection between nodes is established through the graph adjacency relationship, and an adjacency tensor structure between nodes is constructed to represent spatial connectivity; S32. In the model input stage, a graph embedding method is used to vector encode each spatial node in the graph structure. The encoding result is used as the input of the graph attention gated capsule layer. The graph attention gated capsule layer is composed of multiple sublayers, each of which includes a graph attention module, a capsule transformation module and a gated activation unit; S33, the graph attention module calculates the attention weight distribution based on the spatial adjacency weight and node feature similarity, realizes the dynamic adjustment of the local spatial path correlation, and outputs the weighted aggregated node representation; S34, the capsule transformation module performs a posture matrix transformation on the aggregated node representation to generate a capsule feature vector containing path directionality and node status; S35, the gate activation unit dynamically adjusts the information flow weight of each feature channel according to the distribution state of the capsule feature vector, thereby achieving adaptive control of the significance of different path features; S36. Input the output representation after processing by the graph attention gated capsule layer into the recurrent convolution layer for temporal feature fusion, extract the path temporal dependency in the wiring graph structure, output the wiring path features, and generate a set of candidate paths.

6. The method for automatic intelligent wiring laying based on image recognition according to claim 1, characterized in that: The S4 specifically includes: S41, using the candidate path set as the initial solution space for optimization, initializing the ant colony individuals, each ant colony individual representing a specific candidate path, setting the maximum number of iterations and the initial pheromone value; S42. Calculate a path connectivity score, an obstacle avoidance integrity score, and a path length for each candidate path to form three factors for path cost evaluation, which are used to evaluate whether the path is fully accessible, whether it effectively avoids obstacles, and whether it is economical to lay. S43. Calculate the comprehensive cost value for each path based on the set path length priority weight, obstacle avoidance priority weight, and traffic reliability constraint coefficient, which are used as weight parameters of the path comprehensive scoring function. The path cost value function is defined as: F(p)=w L ·L(p)+w O ·O(p)+w S ·S(p); Among them, p represents the path individual, L(p) represents the path length, O(p) represents the obstacle avoidance completeness, S(p) represents the path connectivity, and w L 、w O 、w S are the weight coefficients of path length, obstacle avoidance and connectivity respectively; S44. In each round of path search, a singular potential guidance mechanism is introduced to apply a guiding potential field to the node area near the obstacle boundary in the path to prevent the path from passing through high-risk areas, thereby improving the safety and practical feasibility of the path. S45. Based on the results of the path cost function and the singular guidance mechanism, the ant colony is guided to perform path search and pheromone update, and the variational strategy is used to adaptively adjust the pheromone evaporation rate and the probability distribution of individual path selection; S46. Update the optimal path record after each round of search, and output the path with the minimum path cost as the final optimal routing path after the termination condition is met.

7. The method for automatic intelligent wiring laying based on image recognition according to claim 6, characterized in that: The variational strategy described in S45 adaptively adjusts the pheromone evaporation rate and the individual path selection probability distribution, specifically including: dynamically adjusting the pheromone evaporation rate according to the variance of the path cost value in the current ant colony, reducing the evaporation rate when the variance is higher than a preset first threshold, and increasing the evaporation rate when the variance is lower than a preset second threshold; and adjusting the path selection probability according to the path cost value ranking result.

8. The method for automatic intelligent wiring laying based on image recognition according to claim 1, characterized in that: The S6 specifically includes: S61, inputting the generated wiring action instruction sequence into the wiring device control module, driving the device to execute path segment laying along the optimal wiring path, and collecting wiring area images in real time; S62, inputting the wiring area image into a lightweight recognition module with the same structure as the semantic segmentation model, extracting the current wiring device spatial position, and comparing it with the coordinates of the corresponding path segment of the optimal wiring path; S63: Calculate the spatial offset distance between the current wiring device spatial position and the target path segment. If it exceeds a set threshold, determine it as a path offset or environment change event. S64. Based on the graph structure representation and the path cost function model, a local subgraph is constructed with the current coordinate as the starting point, and a variational singular ant colony optimization algorithm is called to generate a local optimal path segment; S65. Convert the local optimal path segment into a wiring action instruction subsequence and dynamically insert it into the control sequence to achieve real-time correction of the original wiring process and generate and update a new action instruction sequence.

Images