Unmanned aerial vehicle intelligent inspection path generation method for interior of building based on BIM model feature extraction

Abstract

The invention discloses a BIM model feature extraction-based unmanned aerial vehicle intelligent inspection path generation method for the interior of a building. The method comprises the steps of extracting a spatial geometric feature set, a spatial topological feature set and a semantic feature set from original BIM model data; constructing a basic space graph representing a space structure and a semantic enhancement graph representing a function relationship based on the feature set, and generating a multi-level semantic graph; task targets and dependency relationships are extracted from the inspection task description and the multi-level semantic graph, and a task weight matrix is generated; generating an optimized path set based on the multi-level semantic graph and the task weight matrix; performing multi-dimensional scoring and real-time optimization on the optimized path set to generate an emergency adjustment strategy; and finally verifying the inspection path. According to the method, through multi-level feature extraction and multi-level semantic graph construction, accurate understanding of the building space is realized, and the intelligent level and reliability of routing inspection path planning are improved.

Description

Technical Field

[0001] The present invention relates to a method for generating a drone path, in particular to a method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction. Background Art

[0002] With the growing demand for smart buildings and facility management, drone-based indoor inspection systems have gradually become an important technical development direction. Such systems can autonomously complete tasks such as status monitoring, equipment inspection, and safety inspections inside buildings, significantly reducing labor costs and improving inspection efficiency and safety. Especially in complex environments such as large-scale complex buildings, data centers, and industrial plants, drone inspections can overcome the problems of high work intensity, high risk factors, and many inspection blind spots in traditional manual inspections. At the same time, with the widespread application of BIM technology in the full life cycle management of buildings, using the rich spatial information provided by BIM models to guide drone indoor inspections has become a key technical approach to improving the level of intelligent inspections.

[0003] The current indoor drone inspection system mainly adopts positioning and navigation methods based on laser SLAM and path planning strategies based on preset waypoints. In terms of spatial perception, it mainly relies on laser radar and visual sensors to obtain real-time environmental information, and realizes autonomous positioning through feature point matching and pose estimation; in terms of path planning, it mainly adopts improved A*, RRT and other classic algorithms, combined with basic collision detection to generate flight trajectories; in terms of task execution, it usually adopts pre-set fixed inspection routes and simple obstacle avoidance strategies. These methods can realize basic inspection functions in simple indoor environments, but still face many challenges in complex architectural spaces.

[0004] The existing technology has the following specific problems: First, in terms of spatial feature extraction, the existing methods often simplify the building space into a two-dimensional plane or a rough three-dimensional grid, and fail to make full use of the component attributes, spatial relationships and functional information contained in the BIM model, resulting in an insufficient understanding of the complex spatial structure; second, in terms of spatial expression, there is a lack of effective multi-level semantic modeling methods, which cannot accurately describe the mapping relationship between the physical characteristics and functional attributes of the space, affecting the rationality of path planning; third, in terms of path generation, the existing algorithms mainly consider geometric constraints and distance optimization, and do not consider multi-dimensional goals such as task requirements, energy consumption, and execution efficiency, making it difficult to generate the optimal path that meets the actual inspection needs; finally, in terms of path optimization, there is a lack of dynamic adjustment mechanism based on real-time environmental perception, which cannot effectively respond to emergencies and environmental changes inside the building, affecting the reliability and safety of the inspection task. These technical problems seriously restrict the actual application effect of the indoor drone inspection system. Summary of the invention

[0005] The purpose of the invention is to provide a method for generating an intelligent drone inspection path for the interior of a building based on BIM model feature extraction, in order to solve at least one technical problem existing in the prior art.

[0006] Technical solution: A method for generating intelligent inspection paths for drones inside buildings based on BIM model feature extraction, comprising the following steps:

[0007] Step S1, extracting spatial boundary points, voxel states, and local descriptors from pre-stored original BIM model data to generate a spatial geometric feature set; constructing spatial adjacency relationships and key node sets based on the spatial geometric feature set to generate a spatial topological feature set; combining the original BIM model data and the spatial topological feature set to extract component attributes, spatial functions, and regional importance information to generate a semantic feature set;

[0008] Step S2: construct a basic spatial graph representing the spatial structure based on the spatial geometric feature set and the spatial topological feature set; construct a semantic enhancement graph representing the functional relationship using the semantic feature set and the basic spatial graph; perform multi-level mapping association between the basic spatial graph and the semantic enhancement graph to generate a multi-level semantic graph;

[0009] Step S3, extracting task objectives and dependencies from pre-stored inspection task descriptions and multi-level semantic graphs to generate task subsets; mapping the task subsets to the spatial structure of the multi-level semantic graph to generate a task space mapping matrix; calculating multi-dimensional weight values ​​based on the task subsets and the task space mapping matrix to generate a task weight matrix;

[0010] Step S4, generating an initial candidate path set based on the multi-level semantic graph and the task weight matrix; generating a feasible path set by combining the initial candidate path set and the pre-stored spatial constraint condition set; performing multi-objective optimization using the feasible path set and the task weight matrix to generate an optimized path set;

[0011] Step S5: Calculate multi-dimensional scores based on the optimized path set and the pre-stored evaluation index set to generate a path scoring matrix; select the optimal path using the path scoring matrix and the pre-stored path constraints to generate an optimal inspection path; design a dynamic adjustment mechanism based on the optimal inspection path and the real-time environmental data collected in real time to generate an emergency adjustment strategy;

[0012] Step S6: Perform a physical feasibility test based on the optimal inspection path and the pre-stored set of spatial constraints to generate a feasibility assessment result; perform a safety assessment based on the optimal inspection path and the emergency adjustment strategy to generate a safety assessment report; perform a final verification based on the feasibility assessment result and the safety assessment report to generate a verified inspection path.

[0013] Beneficial effects: Through multi-level feature extraction and multi-level semantic graph construction, accurate understanding of building space is achieved, and the intelligence level and reliability of inspection path planning are improved. The relevant technical effects will be described in detail in conjunction with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention.

[0015] Figure 2 It is a flow chart of step S1 of the present invention.

[0016] Figure 3 It is a flow chart of step S2 of the present invention.

[0017] Figure 4 It is a flow chart of step S3 of the present invention.

[0018] Figure 5 It is a flow chart of step S4 of the present invention.

[0019] Figure 6 It is a flow chart of step S5 of the present invention.

[0020] Figure 7 It is a flow chart of step S6 of the present invention. DETAILED DESCRIPTION

[0021] like Figure 1 As shown, a method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction is provided, comprising the following steps:

[0022] Step S1, extracting spatial boundary points, voxel states, and local descriptors from pre-stored original BIM model data to generate a spatial geometric feature set; constructing spatial adjacency relationships and key node sets based on the spatial geometric feature set to generate a spatial topological feature set; combining the original BIM model data and the spatial topological feature set to extract component attributes, spatial functions, and regional importance information to generate a semantic feature set;

[0023] Step S2: construct a basic spatial graph representing the spatial structure based on the spatial geometric feature set and the spatial topological feature set; construct a semantic enhancement graph representing the functional relationship using the semantic feature set and the basic spatial graph; perform multi-level mapping association between the basic spatial graph and the semantic enhancement graph to generate a multi-level semantic graph;

[0024] Step S3, extracting task objectives and dependencies from pre-stored inspection task descriptions and multi-level semantic graphs to generate task subsets; mapping the task subsets to the spatial structure of the multi-level semantic graph to generate a task space mapping matrix; calculating multi-dimensional weight values ​​based on the task subsets and the task space mapping matrix to generate a task weight matrix;

[0025] Step S4, generating an initial candidate path set based on the multi-level semantic graph and the task weight matrix; generating a feasible path set by combining the initial candidate path set and the pre-stored spatial constraint condition set; performing multi-objective optimization using the feasible path set and the task weight matrix to generate an optimized path set;

[0026] Step S5: Calculate multi-dimensional scores based on the optimized path set and the pre-stored evaluation index set to generate a path scoring matrix; select the optimal path using the path scoring matrix and the pre-stored path constraints to generate an optimal inspection path; design a dynamic adjustment mechanism based on the optimal inspection path and the real-time environmental data collected in real time to generate an emergency adjustment strategy;

[0027] Step S6: Perform a physical feasibility test based on the optimal inspection path and the pre-stored set of spatial constraints to generate a feasibility assessment result; perform a safety assessment based on the optimal inspection path and the emergency adjustment strategy to generate a safety assessment report; perform a final verification based on the feasibility assessment result and the safety assessment report to generate a verified inspection path.

[0028] By innovatively integrating BIM model feature extraction, multi-level semantic graph construction, multi-dimensional path planning and real-time optimization and adjustment technologies, a complete indoor drone intelligent inspection system has been established. The solution overcomes the problems of incomplete feature extraction, insufficient spatial expression, unreasonable path planning, and poor real-time adaptability in traditional inspection methods. Through multi-level feature extraction and fusion, a comprehensive understanding of the building space is achieved; through multi-level construction of semantic graphs, the physical and functional relationships of the space are accurately expressed; through multi-objective optimization and dynamic adjustment of the path, the efficient completion of the inspection task is ensured; through multiple verifications and safety assessments, the reliable operation of the system is guaranteed. This solution significantly improves the intelligence level, task completion efficiency and system reliability of indoor drone inspections, and has important theoretical value and practical application value.

[0029] According to one aspect of the present application, step S1 specifically comprises:

[0030] Step S11, reading three-dimensional coordinate information from pre-stored original BIM model data, extracting a set of space boundary point coordinates; constructing a three-dimensional grid space based on the boundary point coordinate set, calculating the occupancy state value of each voxel; calculating a local space description value for each boundary point; combining the boundary point set, voxel state value and local space description value to generate a space geometric feature set;

[0031] Step S12, calculating the connectivity values ​​between spatial regions based on the spatial geometric feature set to generate a spatial adjacency matrix; identifying the three-dimensional coordinates of feature positions such as doors and corners from the spatial geometric feature set to generate a key node set; constructing a regional connectivity graph based on the spatial adjacency matrix and the key node set; combining the spatial adjacency matrix, the key node set and the regional connectivity graph to generate a spatial topological feature set;

[0032] Step S13, extracting attribute information such as component type and material from the pre-stored original BIM model data to generate a component attribute set; determining the function type label of each spatial region based on the regional connectivity relationship of the spatial topological feature set; calculating the importance value of each region; and combining the component attribute set, function type label and importance value to generate a semantic feature set.

[0033] By innovatively deconstructing BIM model data into a three-layer structure of spatial geometric feature set, spatial topological feature set and semantic feature set, all-round feature extraction of the building's internal space is achieved. Among them, the spatial geometric feature set not only retains the geometric information of the space through the combination of boundary points, voxel states and local descriptors, but also captures the microscopic features of the space through local descriptors; the spatial topological feature set accurately describes the connectivity and key position information of the space through the spatial adjacency matrix and key nodes; the semantic feature set integrates component attributes, functional labels and importance information, so that feature extraction includes both physical attributes and functional attributes. This multi-level and multi-dimensional feature extraction method overcomes the limitations of traditional methods that only focus on geometric features or only focus on semantic features, and provides rich decision-making basis for subsequent path planning. At the same time, through the hierarchical organization of features, the efficiency and accuracy of feature extraction are significantly improved, and the computational complexity of feature extraction is reduced.

[0034] According to one aspect of the present application, step S2 is specifically:

[0035] Step S21, constructing a graph node set based on the spatial boundary information in the spatial geometric feature set and the key node set in the spatial topological feature set; establishing an edge set between nodes using the spatial adjacency matrix in the spatial topological feature set; assigning a weight value to each node according to the importance of the spatial feature; and combining the graph node set, edge set and node weight value to generate a basic spatial graph;

[0036] Step S22, constructing a semantic node set based on the function label information in the semantic feature set; establishing an edge set between the semantic nodes according to the degree of association of the spatial functions; calculating the weight value of the semantic node using the regional importance information in the semantic feature set; combining the semantic node set, the semantic edge set and the semantic weight value to generate a semantic enhancement graph;

[0037] Step S23, establish a node correspondence matrix based on the basic spatial graph and the semantic enhancement graph; construct a cross-layer edge set that represents the association relationship between nodes at different levels; calculate the association strength weight of the cross-layer edges; combine the basic spatial graph, the semantic enhancement graph, the node correspondence matrix, the cross-layer edge set and the association strength weight to generate a multi-level semantic graph.

[0038] By adopting an innovative multi-level semantic graph construction method, the organic unity of spatial structure and functional relationship is achieved through the double-layer structure of basic spatial graph and semantic enhanced graph. In the basic spatial graph, the physical layout and connectivity of the space are accurately expressed through the construction of graph node sets and edge sets, combined with the allocation of node weight values; in the semantic enhanced graph, the functional association of the space is effectively described through the construction of semantic node sets and semantic edge sets, combined with the calculation of semantic weight values; the mapping relationship between physical space and functional space is established through the design of node correspondence matrix and cross-layer edge sets. This multi-level graph structure not only overcomes the problem that traditional single graph structure is difficult to express physical and functional relationships at the same time, but also realizes the complementarity and enhancement of information through the association between levels, providing a decision-making basis for subsequent path planning that takes into account both physical constraints and functional requirements. At the same time, through the hierarchical organization of the graph structure, the integrity and accuracy of spatial expression are greatly improved.

[0039] According to one aspect of the present application, step S3 is specifically:

[0040] Step S31, parsing a task target list from a pre-stored inspection task description; establishing dependencies between tasks based on the semantic relationship of a multi-level semantic graph; extracting timing constraints from the inspection task description; combining the task target list, task dependencies and timing constraints to generate a task subset;

[0041] Step S32, establishing a correspondence between tasks and space nodes based on task subsets and multi-level semantic graphs; calculating the access frequency value of each space node according to task requirements; determining the access order constraint of the space node; combining the task space correspondence, access frequency value and access order constraint to generate a task space mapping matrix;

[0042] Step S33, calculate the task priority weight value based on the task subset; calculate the time urgency weight value according to the timing constraints in the task subset; calculate the resource consumption weight value based on the task space mapping matrix; combine the task priority weight value, time urgency weight value and resource consumption weight value to generate a task weight matrix.

[0043] The innovative use of task decomposition and multi-dimensional weight calculation methods has achieved accurate modeling and optimized configuration of inspection tasks. Through task goal analysis, dependency establishment and timing constraint extraction, complex inspection tasks are decomposed into manageable task subsets; through the construction of the task space mapping matrix, the correspondence between task requirements and space resources is established; through the calculation of multi-dimensional weight values, factors such as task priority, time urgency and resource consumption are comprehensively considered. This task modeling method overcomes the problem of oversimplification of task description in traditional methods. Through the refined decomposition of tasks and multi-dimensional weight calculation, task planning is more reasonable and efficient. At the same time, the task space mapping mechanism ensures the feasibility of task execution and significantly improves the completion quality and efficiency of inspection tasks.

[0044] According to one aspect of the present application, step S4 is specifically:

[0045] Step S41, constructing a basic path based on a multi-level semantic graph; performing path expansion in combination with a task weight matrix; increasing path diversity through random sampling; and combining the basic path, the expanded path, and the randomly sampled path to generate an initial candidate path set;

[0046] Step S42: Perform geometric feasibility check based on the pre-stored spatial constraint condition set and the initial candidate path set; perform dynamic constraint verification on the paths that pass the geometric check; perform energy constraint evaluation on the paths that pass the dynamic verification; and combine the paths that pass all constraint verifications to generate a feasible path set;

[0047] Step S43, construct a multi-objective optimization function based on the feasible path set; use the task weight matrix to perform path optimization calculation; perform local search improvement on the optimized path; combine the optimized and improved paths to generate an optimized path set.

[0048] By innovatively combining multi-level semantic graphs and task weight matrices, and adopting a three-stage path generation strategy, efficient and reliable inspection path planning is achieved. First, through the combination of basic path generation, path expansion, and random sampling, a diverse set of initial candidate paths is generated; then, through multiple verifications of geometric feasibility, dynamic constraints, and energy constraints, the actual feasibility of the path is ensured; finally, through multi-objective optimization and local search, the global optimization of the path is achieved. This path planning method overcomes the problem that traditional methods are prone to fall into local optimality or ignore actual constraints. Through diversified path generation and multi-level path optimization, it not only ensures the quality of the planning results, but also improves the efficiency of planning. At the same time, the verification of multiple constraints ensures the actual executability of the path.

[0049] According to one aspect of the present application, step S5 is specifically:

[0050] Step S51, calculating a path length score based on the optimized path set; calculating an energy consumption score based on a pre-stored evaluation index set; evaluating a task completion efficiency score; and combining the path length score, energy consumption score, and task completion efficiency score to generate a path scoring matrix;

[0051] Step S52: perform fuzzy comprehensive evaluation based on the path scoring matrix; perform multi-objective decision-making in combination with pre-stored path constraints; perform sensitivity analysis on the evaluation results; select and generate the optimal inspection path according to the analysis results;

[0052] Step S53, establishing status monitoring indicators based on real-time environmental data collected in real time; setting adjustment trigger conditions according to the optimal inspection path; building an emergency plan database; combining the status monitoring indicators, adjustment trigger conditions and emergency plan data to generate an emergency adjustment strategy.

[0053] The innovative method of combining multi-dimensional scoring and dynamic adjustment has achieved accurate evaluation and real-time optimization of inspection paths. A comprehensive path evaluation system has been established through comprehensive scoring of path length, energy consumption and task completion efficiency; scientific selection of the optimal path has been achieved through fuzzy comprehensive evaluation and multi-objective decision-making methods; and a complete dynamic adjustment mechanism has been established through the design of status monitoring indicators, adjustment trigger conditions and emergency plans. This evaluation and optimization method overcomes the problems of single evaluation dimension and lack of adaptability of traditional methods, ensures the comprehensiveness and accuracy of the evaluation through multi-dimensional scoring, improves the robustness and adaptability of the system through the dynamic adjustment mechanism, and significantly improves the reliability and adaptability of the inspection system.

[0054] According to one aspect of the present application, step S6 is specifically:

[0055] Step S61, verifying kinematic constraints based on the optimal inspection path; checking collision risks according to a pre-stored set of spatial constraints; evaluating passing performance indicators; combining verification and evaluation results to generate a feasibility evaluation result;

[0056] Step S62: Analyze potential risk points based on the optimal inspection path; evaluate emergency response capabilities in combination with emergency adjustment strategies; verify safety margin parameters; and generate a safety assessment report by combining analysis and assessment results;

[0057] Step S63: Conduct a comprehensive assessment based on the feasibility assessment results and the safety assessment report; make necessary optimization adjustments to the problems found in the assessment; confirm the final inspection path plan; and generate a verified inspection path.

[0058] The innovative method of combining physical feasibility verification and safety assessment is used to achieve the final confirmation and safety assurance of the inspection path. The feasibility of the path at the physical level is ensured through kinematic constraint verification, collision risk inspection and performance evaluation; the safety of path execution is guaranteed through potential risk analysis, emergency response capability assessment and safety margin verification; and the final optimization of the path plan is achieved through comprehensive evaluation and necessary adjustments. This verification and evaluation method overcomes the problem of traditional methods ignoring actual execution risks. Through multi-level feasibility verification and safety assessment, it not only ensures the actual executability of the path, but also improves the safety of execution. At the same time, through the final comprehensive optimization, the integrity and reliability of the path plan are further improved.

[0059] According to one aspect of the present application, step S11 is specifically:

[0060] Step S111, reading component geometry information from pre-stored original BIM model data; performing boundary tracking processing on the geometry information to extract boundary vertices; performing spatial continuity analysis on the boundary vertices; organizing the processed boundary vertices to generate a spatial boundary point coordinate set;

[0061] Among them, boundary tracking is specifically: boundary point tracking value BTV(p) = μ1·G(p) + μ2·E(p)·exp(-D(p) / σ) + μ3·C(p); where: G(p) is the gradient strength; E(p) is the edge continuity; C(p) is the corner point response; D(p) is the distance to the known boundary; μ1, μ2, μ3 are weight coefficients and Σμi=1; σ is the distance attenuation parameter.

[0062] In another embodiment of the present application, BIM data standardization is also included, specifically: standardized BIM eigenvalue NBF(x) = β1·F(x) + β2·S(x)·(1+α·Q(x)) + β3·R(x); wherein: F(x) is the geometric eigenvalue; S(x) is the semantic eigenvalue; Q(x) is the data quality score; R(x) is the relationship completeness; β1, β2, β3 are weight coefficients and Σβi=1; α is the quality gain parameter.

[0063] Step S112, determining the boundary of the space range based on the coordinate set of the space boundary points; constructing a three-dimensional grid space using an adaptive grid division method; uniformly sampling the three-dimensional grid space to generate a sampling point cloud; and calculating the occupancy state value of each voxel based on the sampling point cloud;

[0064] Step S113, determining a calculation area based on a set of spatial boundary point coordinates; constructing a local coordinate system for each boundary point; calculating geometric features in the local coordinate system; and converting the geometric features into local space description values;

[0065] The calculation of the local space description value is as follows: local space description value LSV(p) = α1·V(p) + α2·C(p) + α3·A(p) + α4·E(p); where: V(p) = Σ(vi·exp(-||p-pi||² / σ²)), V(p) is the voxel occupancy density feature; C(p) = |Σ((pi-p)×(pi+1-p))| / ||pi-p||·||pi+1-p||, C(p) is the curvature feature; A(p) = arccos(n(p)·nref) / π, A(p) is the normal angle feature; E(p) = -Σ(P(θi)·log(P(θi))), E(p) is the local entropy feature; p is the current boundary point; pi is the neighborhood point; vi is the voxel state value; n(p) is the normal vector; nref is the reference direction; P(θi) is the direction histogram probability; α1, α2, α3, α4 are feature weight coefficients, and Σαi=1; σ is the Gaussian kernel parameter.

[0066] Step S114, combining the spatial boundary point coordinate set, the occupancy state value and the local space description value; normalizing the combined features; and combining the normalized features to generate a spatial geometric feature set.

[0067] Among them, the normalization processing is specifically: normalized eigenvalue NF(x) = γ1·tanh(β1(x-μx) / σx) + γ2·(1-exp(-β2·||x-xref||² / δ²)); where: x is the original eigenvalue; μx is the feature mean; σx is the feature standard deviation; xref is the reference eigenvalue; γ1,γ2 are weight coefficients, and γ1+γ2=1; β1,β2 are scaling parameters; δ is the feature distance scale; tanh is the hyperbolic tangent function; this method combines statistical normalization and relative distance normalization to improve the feature discrimination.

[0068] The innovative four-stage spatial geometric feature extraction method is used to achieve accurate geometric description of the interior space of the building. Through boundary tracking processing and spatial continuity analysis, the coordinate set of spatial boundary points is accurately extracted; through adaptive grid division and uniform sampling strategy, an accurate three-dimensional grid space model is established and the voxel occupancy state is calculated; through local coordinate system construction and geometric feature calculation, the local spatial description value of the boundary point is obtained; through feature combination and normalization processing, a standardized set of spatial geometric features is generated. This feature extraction method overcomes the problems of insufficient accuracy or high computational complexity of traditional methods. Through multi-stage feature extraction and processing, high-precision acquisition of spatial geometric features is achieved. At the same time, the standardized processing of features ensures the reliability of subsequent processing, significantly improving the accuracy and computational efficiency of spatial description.

[0069] According to one aspect of the present application, step S13 is specifically:

[0070] Step S131, read component basic information from pre-stored original BIM model data; convert the component basic information into standardized attributes using an attribute mapping table; organize the standardized attributes using a hierarchical classification method; establish an attribute dependency graph according to the association relationship between the attributes; perform integrity verification based on the attribute dependency graph; combine the verified attributes to generate a component attribute set;

[0071] In another embodiment of the present application, a spatial rule base construction process is also included, specifically: rule weight value RW(r) = ζ1·H(r) + ζ2·U(r)·exp(-T(r) / φ) + ζ3·V(r); wherein: H(r) is the historical usage frequency; U(r) is the rule universality; V(r) is the verification pass rate; T(r) is the time decay factor; ζ1, ζ2, ζ3 are weight coefficients and Σζi=1; φ is the time decay parameter.

[0072] Step S132: establish a spatial function association diagram based on the regional connectivity relationship in the spatial topological feature set; use the pre-stored spatial function rule library to perform preliminary function identification; perform function verification based on the function constraint relationship of adjacent spaces; use the spatial syntax analysis method to calculate the function propagation probability; optimize the function type based on the function propagation probability; generate a function type label for each spatial area;

[0073] Among them, the preliminary function recognition is specifically: function recognition score FRS(r) = λ1·T(r) + λ2·S(r) + λ3·M(r)·exp(-d(r) / ρ); where: T(r)= Σ(wt·ft(r)) is the topological feature score; S(r)= Σ(ws·fs(r)) is the semantic feature score; M(r)= Π(P(r|ri)) is the neighborhood propagation score; d(r) is the topological distance; ft(r) is the topological feature function; fs(r) is the semantic feature function; P(r|ri) is the conditional probability; wt, ws are feature weights; λ1, λ2, λ3 are combination weights; ρ is the distance decay parameter.

[0074] Step S133, read the functional type label of each area; calculate the basic importance based on the pre-stored regional level rules; calculate the regional accessibility index in combination with the spatial topological feature set; calculate the regional influence using the spatial centrality analysis method; calculate the usage intensity based on the regional functional density distribution; weightedly combine the basic importance, accessibility index, regional influence and usage intensity to generate an importance value;

[0075] The calculation process of importance is as follows: regional importance value RIV(z) = η1·B(z) + η2·R(z)·(1+κ·C(z)) + η3·D(z); where: B(z)= Σ(wb·fb(z)) is the basic importance; R(z)= min(1,Σ(wr·N(z,r) / Nmax)) is the accessibility index; C(z)= |E(z)| / (|V(z)|·(|V(z)|-1)) is the centrality; D(z)= Σ(wd·fd(z)) is the density index; fb(z),fd(z) are evaluation functions; N(z,r) is the number of reachable nodes within the range of r; Nmax is the maximum number of nodes; wb,wr,wd are weight coefficients; η1,η2,η3 are combined weights, and Σηi=1; κ is the centrality gain coefficient.

[0076] Step S134, organize the component attribute set in a structured manner; map the function type labels according to the regional codes; normalize the importance values; integrate the above data using a feature fusion algorithm; perform consistency check on the integrated data; and generate a semantic feature set.

[0077] Among them, the feature fusion algorithm is specifically as follows: fusion feature value FF(f) = π1·P(f) + π2·L(f)·(1+γ·M(f)) + π3·N(f); where: P(f) is the physical feature; L(f) is the logical feature; M(f) is the mapping completeness; N(f) is the association degree; π1, π2, π3 are weight coefficients and Σπi=1; γ is the mapping gain parameter.

[0078] The innovative semantic feature extraction method of multi-source information fusion is used to achieve a comprehensive semantic description of the architectural space. Through the construction of hierarchical classification and attribute dependency graph, the standardized organization of component attributes is achieved; through the spatial function association graph and function propagation probability analysis, the functional type of the spatial area is accurately identified; through the fusion and weighted combination of multi-dimensional indicators, the importance of the area is scientifically calculated; through feature fusion and consistency test, a complete set of semantic features is generated. This semantic feature extraction method overcomes the problem of incomplete semantic information or weak correlation in traditional methods. Through the systematic fusion of multi-source information, the comprehensive acquisition of spatial semantic features is achieved. At the same time, the consistency test of features ensures the accuracy of semantic description, which greatly improves the effect and reliability of spatial function analysis.

[0079] According to one aspect of the present application, step S22 is specifically:

[0080] Step S221, extracting function label information from the semantic feature set; performing semantic layering processing on the function label information; performing function normalization processing based on the semantic ontology library; hierarchically organizing the normalized function nodes; performing correlation analysis on the organized nodes; organizing the processed nodes to generate a semantic node set;

[0081] Step S222, read node pair information from the semantic node set; calculate the semantic similarity between nodes based on the pre-stored functional association rule library; calculate the indirect association strength using the semantic transfer probability model; calculate the comprehensive association degree by combining the direct association and the indirect association; filter the effective connection according to the association degree threshold; generate a semantic edge set from the filtered node connection relationship;

[0082] The semantic similarity calculation process is as follows:

[0083] Semantic similarity SS(i,j) = π1·H(i,j) + π2·O(i,j)·exp(-L(i,j) / τ) + π3·F(i,j); where: H(i,j) = 1-|hi-hj| / hmax is the hierarchical similarity; O(i,j) = |Oi∩Oj| / |Oi∪Oj| is the ontology overlap; F(i,j) = Σ(wf·cos(fi,fj)) is the functional similarity; L(i,j) is the shortest path length; hi,hj are the hierarchical depths; Oi,Oj are the ontology sets; fi,fj are the function vectors; π1,π2,π3 are the weight coefficients; τ is the path attenuation parameter.

[0084] Step S223, extracting regional importance information based on the semantic feature set; calculating basic weights based on the hierarchical levels of semantic nodes; calculating structural weights based on node connectivity; calculating position weights using spatial position relationships; fusing various weights using a multi-criteria weighting method; and generating weight values ​​for semantic nodes;

[0085] Among them, the fusion weights are as follows: fusion weight value FW(n) = ϕ1·B(n)·(1+μ·log(1+|En|)) + ϕ2·H(n)·exp(-D(n) / ξ) + ϕ3·P(n); among them: B(n) is the basic weight; H(n) is the hierarchical weight; P(n) is the position weight; |En| is the number of connected edges; D(n) is the distance to important nodes; ϕ1, ϕ2, ϕ3 are fusion coefficients and Σϕi=1; μ is the connectivity gain parameter; ξ is the distance attenuation coefficient.

[0086] Step S224, encode and identify the semantic node set; establish an adjacency list representation for the semantic edge set; map the semantic weight value to the corresponding node; use a graph structure organization method to integrate the node, edge and weight information; verify the consistency of the integrated graph structure; and generate a semantically enhanced graph.

[0087] The innovative use of semantic enhanced graph construction method realizes the deep expression of spatial functional relations. Through semantic stratification and semantic ontology mapping, a standardized set of semantic nodes is established; through semantic similarity calculation and transfer probability model, a complete set of semantic edges is constructed; through multi-criteria weight fusion, the importance of semantic nodes is determined; through graph structure organization and consistency verification, a reliable semantic enhanced graph is generated. This graph construction method overcomes the problem of insufficient functional relationship expression of traditional methods. Through deep analysis and relationship construction at the semantic level, it realizes the accurate description of spatial functional relations. At the same time, the standardized organization of the graph structure ensures the integrity of semantic expression, which significantly improves the accuracy and practicality of spatial functional analysis.

[0088] According to one aspect of the present application, step S31 is specifically:

[0089] Step S311: read the original task text from the pre-stored inspection task description; perform semantic word segmentation processing using task parsing rules; identify key task elements based on a preset task template; classify and organize the recognition results; organize the organized task elements to generate a task target list.

[0090] Step S312: Extract semantic relationship information from the multi-level semantic graph output by step S2; establish a task node graph in combination with the task target list; analyze the sequence between nodes based on the graph traversal algorithm; verify the rationality of the sequence using task logic rules; generate dependency relationships between tasks from the verified sequence relationship.

[0091] Step S313: extracting time requirement information from the pre-stored inspection task description; parsing the time window constraint based on the timing template; analyzing the execution timing requirement in combination with the task dependency relationship; performing conflict detection on the timing requirement; generating timing constraint conditions from the processed timing requirement.

[0092] Step S314: organize the task target list outputted in step S311 into a structured manner; convert the task dependency relationship outputted in step S312 into an adjacency matrix representation; normalize the timing constraint conditions outputted in step S313; combine the above data using a task integration algorithm; perform integrity check on the combined result; and generate a task subset.

[0093] The innovative use of task decomposition and dependency building methods has achieved refined modeling of inspection tasks. Through semantic word segmentation and template matching, the task target elements are accurately identified; through graph traversal and logic verification, reliable task dependencies are established; through timing analysis and conflict detection, reasonable execution constraints are determined; through task integration and integrity verification, standardized task subsets are generated. This task modeling method overcomes the problems of rough task description or unclear constraints in traditional methods. Through refined task decomposition and systematic organization, the accurate definition of inspection tasks is achieved. At the same time, strict verification of the relationship between tasks ensures the feasibility of execution, greatly improving the scientificity and operability of task planning.

[0094] According to one aspect of the present application, step S41 is specifically:

[0095] Step S411: extract node and edge information from the multi-level semantic graph output from step S2; use the improved Dijkstra algorithm to calculate the shortest path between nodes; analyze the reachable path based on node connectivity; generate a path skeleton based on the topological structure of the graph; organize the generated path skeleton into a basic path.

[0096] Among them, the improved Dijkstra algorithm is specifically as follows: path cost PC(u,v) = θ1·L(u,v) + θ2·R(u,v)·(1+ε·S(u,v)) + θ3·W(u,v)·exp(-K(u,v) / ψ); where: L(u,v) is the path length; R(u,v) is the risk; S(u,v) is the smoothness; W(u,v) is the task weight; K(u,v) is the number of key points; θ1, θ2, θ3 are weight coefficients; ε is the smoothness gain parameter; ψ is the key point attenuation coefficient.

[0097] Step S412: read the task weight matrix outputted in step S3; perform path scoring on the basic path based on the weight value; generate a local variation path using a path variation algorithm; generate a combined path using a path recombination method; generate an extended path using the variation path and the combined path.

[0098] The path mutation algorithm is specifically as follows: mutation path value MP(p) = χ1·O(p) + χ2·V(p)·exp(-C(p) / φ) + χ3·Q(p); where: O(p) is the original path score; V(p) is the mutation amplitude; C(p) is the constraint violation degree; Q(p) is the quality score; χ1, χ2, χ3 are weight coefficients and Σχi=1; φ is the constraint attenuation parameter.

[0099] Step S413: Establish a sampling space based on the multi-level semantic graph output from step S2; generate sampling points using a layered random sampling strategy; connect the sampling points using a probabilistic roadmap method; verify the feasibility of the connection path; and generate a random sampling path for the verified path.

[0100] Step S414: encode and identify the basic path outputted from step S411; structure the extended path outputted from step S412; normalize the randomly sampled path outputted from step S413; combine the above paths using a path fusion algorithm; deduplicate the combined path set; and generate an initial candidate path set.

[0101] The innovative use of diversified path generation strategies enables efficient construction of initial candidate paths. The basic path skeleton is generated through the improved Dijkstra algorithm and connectivity analysis; diversified extended paths are generated through path scoring and mutation recombination; supplementary random paths are obtained through stratified random sampling and feasibility verification; and a complete set of candidate paths is generated through path fusion and deduplication processing. This path generation method overcomes the problem of insufficient path diversity or low quality in traditional methods. Through the combined application of multiple strategies, batch generation of high-quality candidate paths is achieved. At the same time, the normalization of paths ensures the efficiency of subsequent optimization, significantly improving the initial quality and convergence speed of path planning.

[0102] According to one aspect of the present application, step S42 is specifically:

[0103] Step S421: Read geometric constraint parameters from a pre-stored set of spatial constraint conditions; extract path geometric features based on the initial candidate path set output in step S41; use a collision detection algorithm to verify clearance; use a curvature analysis method to check the turning radius; verify vertical spacing based on a spatial partitioning algorithm; generate a geometrically feasible path set for the verified paths.

[0104] Step S422: read path information from the geometrically feasible path set; extract dynamic parameters based on the pre-stored spatial constraint condition set; calculate speed constraints using the kinematic model; verify maneuverability using acceleration analysis method; check balance stability based on attitude constraints; generate a dynamically feasible path set from the verified paths.

[0105] Among them, the dynamic constraint value DC(v) = θ1·K(v) + θ2·A(v)·(1+ε·J(v)) + θ3·E(v); where: K(v) is the kinematic constraint; A(v) is the acceleration constraint; J(v) is the jerk constraint; E(v) is the energy constraint; θ1, θ2, θ3 are weight coefficients and Σθi=1; ε is the smoothness parameter.

[0106] Step S423: extract path segments from the power feasible path set; read energy consumption parameters based on the pre-stored space constraint condition set; calculate basic power consumption using the energy consumption model; estimate additional energy consumption based on path characteristics; verify endurance using energy balance analysis; generate an energy feasible path set for the verified paths.

[0107] Step S424: classify and organize the various feasible paths output from steps S421 to S423; evaluate the feasibility level of the paths; screen the paths based on the comprehensive scores; fine-tune the screened paths using the path optimization algorithm; perform final verification on the adjusted paths; and generate a set of feasible paths.

[0108] The innovative use of multi-level constraint verification methods has achieved a comprehensive feasibility assessment of the inspection path. The geometric feasibility of the path is verified through collision detection and curvature analysis; the dynamic feasibility of the path is verified through kinematic models and acceleration analysis; the energy feasibility of the path is verified through energy consumption models and energy balance analysis; and the final feasible path set is generated through feasibility level evaluation and path fine-tuning. This constraint verification method overcomes the problem of incomplete constraint consideration or lax verification in traditional methods. Through strict verification and system optimization in multiple dimensions, it achieves comprehensive guarantee of path feasibility. At the same time, the path fine-tuning mechanism ensures the practicality of the verification results and significantly improves the success rate and reliability of path execution.

[0109] According to one aspect of the present application, step S43 is specifically:

[0110] Step S431: extract path feature data from the feasible path set outputted from step S42; construct an objective function based on path length, smoothness and complexity; generate constraint conditions in combination with the task weight matrix outputted from step S3; establish sub-objective functions using the objective decomposition method; and combine the objective function and constraint conditions to generate a multi-objective optimization function.

[0111] Step S432: Calculate the initial path score based on the multi-objective optimization function; generate a path optimization plan using a genetic algorithm; use the Pareto sorting method to screen the dominant solution; update the optimization plan in combination with the elite retention strategy; generate an optimized path set from the optimized plan.

[0112] Step S433: extracting a path node sequence from the optimized path set; selecting an improved area based on a variable neighborhood search strategy; optimizing the local trajectory using a path smoothing algorithm; adjusting the node distribution using a dynamic programming method; and generating an improved path set from the improved path.

[0113] Step S434: sort and classify the optimized path set outputted from step S432; sort and score the improved path set outputted from step S433; integrate the optimized paths using a path merging algorithm; eliminate redundancy in the integrated paths; and combine the processed paths to generate an optimized path set.

[0114] The innovative method of combining multi-objective optimization and local search is used to achieve precise optimization of the inspection path. Through target decomposition and constraint construction, a complete optimization function system is established; through genetic algorithms and Pareto sorting, global optimization of the path is achieved; through variable neighborhood search and path smoothing, fine adjustment of local trajectories is completed; through path merging and redundancy elimination, the final optimized path set is generated. This path optimization method overcomes the problem that traditional methods are prone to fall into local optimality or insufficient optimization. Through the organic combination of global optimization and local adjustment, a significant improvement in path quality is achieved. At the same time, the redundancy elimination mechanism ensures the simplicity of the optimization results, greatly improving the efficiency and effectiveness of path planning.

[0115] According to one aspect of the present application, step S51 is specifically:

[0116] Step S511: extracting path geometry data from the optimized path set outputted from step S4; calculating the length of each path segment using a path segmentation algorithm; calculating the path smoothness value based on curvature analysis; calculating the complexity value based on the number of turns; and generating a path length score by weighted combination of the path length, smoothness value, and complexity value.

[0117] The path length score is specifically: path score value PS(r) = ω1·D(r)·(1-σ·T(r)) + ω2·S(r)·exp(-C(r) / ζ) + ω3·E(r); where: D(r) is the distance score; T(r) is the number of turns; S(r) is the smoothness score; C(r) is the complexity; E(r) is the efficiency score; ω1, ω2, ω3 are weight coefficients and Σωi=1; σ is the turn penalty coefficient; ζ is the complexity attenuation parameter.

[0118] Step S512: extracting energy consumption evaluation standards based on a pre-stored evaluation index set; extracting motion parameters from an optimized path set; calculating basic energy consumption values ​​using an energy consumption model; estimating dynamic energy consumption values ​​based on path characteristics; and comprehensively calculating basic energy consumption values ​​and dynamic energy consumption values ​​to generate an energy consumption score.

[0119] Step S513: extract the task execution sequence from the optimized path set; obtain the time evaluation parameter based on the pre-stored evaluation index set; calculate the waiting time value between tasks; analyze the impact of path overlap on efficiency; and generate a task completion efficiency score based on the task completion time and path utilization.

[0120] Step S514: normalize the path length score outputted from step S511; perform standardized calculation on the energy consumption score outputted from step S512; perform quantitative mapping on the task completion efficiency score outputted from step S513; integrate the scores using a scoring combination algorithm; and generate a path scoring matrix using the integrated scoring data.

[0121] The innovative use of multi-dimensional path scoring method realizes the scientific evaluation of inspection paths. Through path segmentation and curvature analysis, the accurate path length score is calculated; through energy consumption model and dynamic estimation, a reliable energy consumption score is obtained; through task analysis and efficiency evaluation, a reasonable completion efficiency score is determined; through scoring combination and normalization processing, a complete path scoring matrix is ​​generated. This scoring method overcomes the problem of single evaluation dimension or unreasonable indicator setting in traditional methods. Through comprehensive evaluation of multiple dimensions, accurate measurement of path quality is achieved. At the same time, the standardized processing of scoring ensures the comparability of evaluation results, which significantly improves the scientificity and accuracy of path selection.

[0122] According to one aspect of the present application, step S53 is specifically:

[0123] Step S531: extract environmental parameters from real-time environmental data collected in real time; process environmental parameters using a multi-sensor data fusion algorithm; calculate parameter change trends based on a sliding window method; identify environmental mutations using anomaly detection methods; generate status monitoring indicators from processed environmental indicators.

[0124] The multi-sensor data fusion includes: fusion data value FD(t) = υ1·M(t)·exp(-ΔT / τ) + υ2·N(t)·(1+ρ·V(t)) + υ3·A(t); where: M(t) is the main sensor data; N(t) is the auxiliary sensor data; A(t) is the environmental data; ΔT is the time difference; V(t) is the data validity; υ1, υ2, υ3 are weight coefficients and Συi=1; τ is the time decay parameter; ρ is the validity gain coefficient.

[0125] Step S532: extract path features from the optimal inspection path output by step S52; determine key monitoring points based on path segmentation analysis; set parameter warning values ​​using threshold calculation methods; establish trigger rules based on environmental change characteristics; generate and adjust trigger conditions based on the monitoring points, warning values ​​and trigger rules.

[0126] In another embodiment of the present application, the data is also updated, and the update frequency is specifically: update frequency value UF(t) = η1·C(t) + η2·V(t)·exp(-D(t) / ρ) + η3·P(t); wherein: C(t) is the computational complexity; V(t) is the data change rate; P(t) is the priority; D(t) is the delay time; η1, η2, η3 are weight coefficients and Σηi=1; ρ is the delay attenuation parameter.

[0127] Step S533: Establish a plan framework based on the pre-stored emergency plan template; build a set of alternative plans in combination with historical emergency data; establish emergency response rules using a scenario matching method; use a plan evaluation model to screen effective plans; and generate emergency plan data from the screened plan data.

[0128] Step S534: classify the status monitoring indicators output by step S531; regularize the adjustment trigger conditions output by step S532; perform scenario mapping on the emergency plan data output by step S533; use a strategy integration algorithm to combine monitoring, triggering and emergency data; and generate an emergency adjustment strategy from the integrated data.

[0129] The innovative use of adaptive adjustment strategies has achieved real-time optimization of inspection paths. Reliable status monitoring indicators have been established through multi-sensor fusion and anomaly detection; reasonable adjustment trigger conditions have been determined through path analysis and early warning settings; an effective emergency plan library has been built through template matching and evaluation screening; and a complete emergency adjustment strategy has been generated through strategy integration and scenario mapping. This adjustment method overcomes the problems of poor adaptability or untimely response of traditional methods. Through the mechanism of real-time monitoring and rapid response, dynamic optimization of paths has been achieved. At the same time, the system design of the emergency plan ensures the reliability of the adjustment and greatly improves the robustness and adaptability of the system.

[0130] According to one aspect of the present application, step S62 is specifically:

[0131] Step S621: extract trajectory parameters from the optimal inspection path output in step S5; use a risk assessment model to identify spatial risk areas; analyze collision risk probability based on path characteristics; calculate traffic risk level in combination with environmental constraints; generate potential risk points based on the risk area, risk probability and risk level.

[0132] Step S622: read emergency rules from the emergency adjustment strategy output in step S5; calculate response time parameters based on rule matching; analyze disposal efficiency using a capability assessment model; verify solution reliability in combination with historical data; generate an emergency response capability value using the response time, disposal efficiency and reliability data.

[0133] The specific emergency response capability value is: response capability value RC(e) = ψ1·T(e)·exp(-D(e) / δ) + ψ2·P(e)·(1+κ·R(e)) + ψ3·H(e); where: T(e) is the response time score; P(e) is the disposal efficiency; H(e) is the historical success rate; D(e) is the distance factor; R(e) is the resource adequacy; ψ1, ψ2, ψ3 are weight coefficients and Σψi=1; δ is the distance attenuation parameter; κ is the resource gain coefficient.

[0134] Step S623: extract key safety parameters based on potential risk points; analyze the space margin using a safety distance calculation model; calculate the motion margin in combination with dynamic constraints; evaluate the residual margin using energy status; generate safety margin parameters from the space margin, motion margin and residual margin.

[0135] Step S624: Classify the potential risk point data output from step S621; quantitatively evaluate the emergency response capability value output from step S622; perform threshold analysis on the safety margin parameter output from step S623; integrate the assessment data using a report generation template; and generate a safety assessment report using the integrated data.

[0136] The innovative use of a comprehensive safety assessment method has achieved risk control on the inspection route. Potential danger points were identified through risk assessment and probability analysis; emergency response capabilities were verified through response time and efficiency assessment; the safety boundary of execution was ensured through multi-dimensional margin analysis; and a systematic safety assessment was completed through report generation and data integration. This assessment method overcomes the problems of inaccurate risk prediction or inadequate prevention and control measures in traditional methods. Through comprehensive risk assessment and systematic safety verification, the safety of route execution is guaranteed. At the same time, the standardization of assessment reports ensures the traceability of safety management, significantly improving the safety and reliability of the inspection system.

[0137] In summary, in order to address the problem of insufficient spatial feature extraction, a three-layer feature extraction architecture (spatial geometric feature set, spatial topological feature set, and semantic feature set) is used to achieve in-depth analysis of BIM model data. Specifically, accurate geometric feature extraction is achieved by combining boundary tracking processing and local spatial description value calculation; a complete topological relationship is established through spatial adjacency matrix and key node identification; and a hierarchical classification and attribute dependency graph are used to achieve systematic extraction of component attributes, spatial functions, and regional importance. This multi-level feature extraction method ensures comprehensive acquisition and accurate understanding of architectural spatial features.

[0138] To address the problem of inaccurate spatial expression: a unified expression of physical space and functional space is achieved through a multi-level semantic graph construction method. A two-layer structure of a basic spatial graph and a semantic enhancement graph is designed, and an accurate mapping of physical features and functional attributes is established through a node correspondence matrix and cross-layer edge sets. This multi-level graph structure not only accurately expresses the physical layout and functional associations of the space, but also improves the integrity of the spatial expression through information complementarity between levels.

[0139] In response to the problem of unreasonable path generation: a complete path generation system has been established. Through task decomposition and multi-dimensional weight calculation, accurate modeling of inspection needs has been achieved; the three-stage strategy of "basic path generation-multi-constraint verification-multi-objective optimization" has been adopted, and multiple dimensions such as geometric constraints, dynamic constraints, and energy constraints have been comprehensively considered to ensure the practicality and superiority of the generated path.

[0140] In order to solve the problem of poor dynamic adaptability, we have built a complete real-time optimization mechanism. We have designed a state monitoring system based on multi-sensor data fusion, combined with adjustment trigger conditions and emergency plan database, to achieve intelligent dynamic adjustment of the path; through comprehensive safety assessment and emergency response mechanism, we have ensured the reliable operation of the system in the face of environmental changes.

[0141] Through the three-layer feature system, the complete extraction from geometric features to functional semantics is achieved. Through the multi-level semantic graph, a unified model of physical space and functional space is established. Through the multi-stage strategy, the rationality and feasibility of path planning are ensured. Through real-time monitoring and dynamic adjustment, the robustness and reliability of the system are improved.

[0142] It not only overcomes the limitations of existing technologies, but also significantly improves the overall performance and practical value of indoor drone inspection systems. Through the organic combination of feature extraction, spatial expression, path generation and dynamic optimization, the intelligent and efficient drone inspection of buildings is achieved.

[0143] In another embodiment of the present application, the data processing flow is as follows:

[0144] S1. Multi-dimensional feature extraction of BIM model

[0145] S11: Geometric feature extraction

[0146] Obtain the original BIM model data and extract the spatial boundary point set P={p1,p2,...,pn}, where pi=(xi,yi,zi) represents the three-dimensional coordinates of the boundary point; construct the spatial voxel matrix V, V(i,j,k)∈{0,1} represents the voxel occupancy state; calculate the local spatial descriptor L(p)={l1,l2,...,lm}, where li represents the local geometric features at point p.

[0147] Generate spatial geometric feature set G={P,V,L}

[0148] S12: Topological feature extraction

[0149] Read the spatial geometric feature set G and construct the spatial adjacency matrix A, where Aij represents the connectivity between spatial regions i and j;

[0150] Extract the key node set K={k1,k2,...,kr}, including feature points such as doors and corners;

[0151] Calculate the regional connectivity graph C, C=(V,E), V is the set of spatial regions, E is the set of connectivity relations;

[0152] Generate spatial topological feature set T={A,K,C}

[0153] S13: Semantic feature extraction

[0154] Obtain original BIM model data and spatial topology feature set T;

[0155] Extract component attribute set M={m1,m2,...,mt}, including component type, material, etc.;

[0156] Construct a spatial function label set F={f1,f2,...,fu};

[0157] Establish regional importance score R={r1,r2,...,rv};

[0158] Generate a semantic feature set S={M,F,R}.

[0159] S2: Multi-level semantic graph construction

[0160] S21: Basic layer construction

[0161] Read the spatial geometric feature set G and the spatial topological feature set T;

[0162] Construct graph node set Vb based on key node set K;

[0163] Establish edge set Eb based on spatial adjacency matrix A;

[0164] Assign node weights wv based on spatial feature importance;

[0165] Generate the basic space graph Gb=(Vb,Eb,wv);

[0166] S22: Semantic layer construction

[0167] Obtain the semantic feature set S and the basic spatial graph Gb;

[0168] Construct semantic node set Vs and integrate spatial function labels;

[0169] Establish semantic edge set Es to represent functional association relationship;

[0170] Calculate the semantic weight ws based on the region importance score;

[0171] Generate semantic enhancement graph Gs=(Vs,Es,ws);

[0172] S23: Inter-layer relationship mapping

[0173] Obtain the basic spatial graph Gb and the semantic enhancement graph Gs;

[0174] Establish an inter-layer mapping matrix H, where Hij represents the correspondence between nodes in two layers;

[0175] Construct a cross-layer edge set Ec to represent the associations between different levels;

[0176] Calculate the cross-layer weight wc to reflect the strength of the relationship between layers;

[0177] Generate a multi-level semantic graph G={Gb,Gs,H,Ec,wc};

[0178] S3: Inspection task requirements analysis

[0179] S31: Task Decomposition

[0180] Obtain inspection task description and multi-level semantic graph G;

[0181] Parsing task target set O={o1,o2,...,ox};

[0182] Establish task dependency D={d1,d2,...,dy};

[0183] Determine the task timing constraints Ct={t1,t2,...,tz};

[0184] Generate task subset T={O,D,Ct};

[0185] S32: Spatial Mapping

[0186] Obtain task subset T and multi-level semantic graph G;

[0187] Establish task-space correspondence;

[0188] Calculate the frequency of spatial access;

[0189] Determine spatial access order constraints;

[0190] Generate task space mapping matrix M;

[0191] S33: Weight calculation

[0192] Get the task subset T and the task space mapping matrix M;

[0193] Calculate the task priority weight wp;

[0194] Evaluation time urgency weight wt;

[0195] Estimate resource consumption weight wr;

[0196] Generate task weight matrix W={wp,wt,wr};

[0197] S4: Adaptive path planning model construction

[0198] S41: Candidate Path Generation

[0199] Obtain multi-level semantic graph G and task weight matrix W;

[0200] Apply the improved A* algorithm to generate the basic path;

[0201] Expanding the set of paths using probabilistic roadmap methods;

[0202] Incorporate random sampling to increase diversity;

[0203] Generate an initial candidate path set P0;

[0204] S42: Dynamic Constraint Processing

[0205] Obtain the initial candidate path set P0 and the spatial constraint condition set C;

[0206] Check geometric feasibility constraints; evaluate dynamic constraints; consider energy constraints; generate feasible path set P1;

[0207] S43: Adaptive Optimization

[0208] Obtain the feasible path set P1 and the task weight matrix W;

[0209] Construct multi-objective optimization function; apply improved genetic algorithm optimization; integrate local search strategy; generate optimization path set P2;

[0210] S5: Path optimization and dynamic adjustment

[0211] S51: Path evaluation

[0212] Obtain the optimized path set P2 and the evaluation index set I; calculate the path length score sl; evaluate the energy consumption to get se; estimate the task completion efficiency score st; generate the path scoring matrix S={sl,se,st};

[0213] S52: Optimal Path Selection

[0214] Obtain the path scoring matrix S and path constraint conditions C; apply the fuzzy comprehensive evaluation method; consider the weights of multi-objective decision-making; conduct sensitivity analysis; determine the optimal inspection path P*;

[0215] S53: Dynamic adjustment mechanism

[0216] Input: optimal inspection path P*, real-time environmental data E Output: emergency adjustment strategy D Processing: establish a state monitoring mechanism; design trigger condition thresholds; build an alternative solution library; generate emergency adjustment strategy D;

[0217] S6: Path feasibility verification

[0218] S61: Physical feasibility test

[0219] Obtain the optimal inspection path P* and the set of spatial constraints C; verify kinematic constraints; check collision risks; evaluate passing performance; generate feasibility assessment results R1;

[0220] S62: Safety Assessment

[0221] Obtain the optimal inspection path P* and emergency adjustment strategy D; analyze potential risk points; evaluate emergency response capabilities; verify safety margins; generate safety assessment report R2;

[0222] S63: Final Verification

[0223] Obtain feasibility assessment results R1 and safety assessment report R2; comprehensive assessment results; necessary adjustments and optimizations; confirm the final plan; output the verified inspection path Pf.

[0224] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for generating intelligent inspection paths for drones inside buildings based on BIM model feature extraction, characterized in that: The steps include: Step S1, extracting spatial boundary points, voxel states, and local descriptors from pre-stored original BIM model data to generate a spatial geometric feature set; constructing spatial adjacency relationships and key node sets based on the spatial geometric feature set to generate a spatial topological feature set; combining the original BIM model data and the spatial topological feature set to extract component attributes, spatial functions, and regional importance information to generate a semantic feature set; Step S2, constructing a basic spatial graph representing the spatial structure based on the spatial geometric feature set and the spatial topological feature set; Constructing a semantic enhancement graph representing functional relationships using a semantic feature set and the basic spatial graph; performing multi-level mapping association between the basic spatial graph and the semantic enhancement graph to generate a multi-level semantic graph; Step S3, extracting task objectives and dependencies from pre-stored inspection task descriptions and multi-level semantic graphs to generate task subsets; Mapping the task subsets onto the spatial structure of a multi-level semantic graph to generate a task space mapping matrix; calculating multi-dimensional weight values ​​based on the task subsets and the task space mapping matrix to generate a task weight matrix; Step S4, generating an initial candidate path set based on the multi-level semantic graph and the task weight matrix; generating a feasible path set by combining the initial candidate path set and the pre-stored spatial constraint condition set; performing multi-objective optimization using the feasible path set and the task weight matrix to generate an optimized path set; Step S5: Calculate multi-dimensional scores based on the optimized path set and the pre-stored evaluation index set to generate a path scoring matrix; select the optimal path using the path scoring matrix and the pre-stored path constraints to generate an optimal inspection path; design a dynamic adjustment mechanism based on the optimal inspection path and the real-time environmental data collected in real time to generate an emergency adjustment strategy; Step S6: Perform a physical feasibility test based on the optimal inspection path and the pre-stored spatial constraint condition set to generate a feasibility assessment result; Perform safety assessment in combination with the optimal inspection path and emergency adjustment strategy, and generate a safety assessment report; The feasibility assessment results and safety assessment report are combined to perform final verification and generate a verified inspection route.

2. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11, reading three-dimensional coordinate information from pre-stored original BIM model data, extracting a set of space boundary point coordinates; constructing a three-dimensional grid space based on the boundary point coordinate set, calculating the occupancy state value of each voxel; calculating a local space description value for each boundary point; combining the boundary point set, voxel state value and local space description value to generate a space geometric feature set; Step S12, calculating the connectivity values ​​between spatial regions based on the spatial geometric feature set to generate a spatial adjacency matrix; identifying the three-dimensional coordinates of feature positions such as doors and corners from the spatial geometric feature set to generate a key node set; constructing a regional connectivity graph based on the spatial adjacency matrix and the key node set; combining the spatial adjacency matrix, the key node set and the regional connectivity graph to generate a spatial topological feature set; Step S13: extracting attribute information such as component type and material from the pre-stored original BIM model data to generate a component attribute set; Based on the regional connectivity relationship of the spatial topological feature set, the functional type label of each spatial region is determined; the importance value of each region is calculated; and the component attribute set, the functional type label and the importance value are combined to generate a semantic feature set.

3. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21, constructing a graph node set based on the spatial boundary information in the spatial geometric feature set and the key node set in the spatial topological feature set; establishing an edge set between nodes using the spatial adjacency matrix in the spatial topological feature set; assigning a weight value to each node according to the importance of the spatial feature; and combining the graph node set, edge set and node weight value to generate a basic spatial graph; Step S22, constructing a semantic node set based on the function label information in the semantic feature set; establishing an edge set between the semantic nodes according to the degree of association of the spatial functions; and calculating the weight value of the semantic node using the regional importance information in the semantic feature set; Combining the semantic node set, the semantic edge set and the semantic weight value to generate a semantic enhancement graph; Step S23: establishing a node correspondence matrix based on the basic spatial graph and the semantic enhancement graph; constructing a cross-layer edge set representing the association relationship between nodes at different levels; and calculating the association strength weight of the cross-layer edges; The basic spatial graph, the semantic enhancement graph, the node correspondence matrix, the cross-layer edge set and the association strength weight are combined to generate a multi-level semantic graph.

4. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S3 is specifically as follows: Step S31, parsing a task target list from a pre-stored inspection task description; establishing dependencies between tasks based on the semantic relationship of a multi-level semantic graph; extracting timing constraints from the inspection task description; Combining the task target list, task dependencies and timing constraints to generate a task subset; Step S32: establishing a correspondence between tasks and spatial nodes based on the task subset and the multi-level semantic graph; Calculate the access frequency value of each spatial node according to the task requirements; determine the access order constraint of the spatial node; combine the task space correspondence, the access frequency value and the access order constraint to generate a task space mapping matrix; Step S33, calculate the task priority weight value based on the task subset; calculate the time urgency weight value according to the timing constraints in the task subset; calculate the resource consumption weight value based on the task space mapping matrix; combine the task priority weight value, time urgency weight value and resource consumption weight value to generate a task weight matrix.

5. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S4 is specifically as follows: Step S41, constructing a basic path based on a multi-level semantic graph; performing path expansion in combination with a task weight matrix; increasing path diversity through random sampling; and combining the basic path, the expanded path, and the randomly sampled path to generate an initial candidate path set; Step S42: performing a geometric feasibility check based on the pre-stored spatial constraint condition set and the initial candidate path set; Perform dynamic constraint verification on paths that pass geometric verification; Energy constraint assessment of the pathways validated by kinetics; The paths that pass all constraint verifications are combined to generate a set of feasible paths; Step S43, constructing a multi-objective optimization function based on the feasible path set; The task weight matrix is ​​used to perform path optimization calculations; local search and improvement are performed on the optimized path; and the optimized and improved paths are combined to generate an optimized path set.

6. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S5 is specifically as follows: Step S51, calculating a path length score based on the optimized path set; calculating an energy consumption score based on a pre-stored evaluation index set; evaluating a task completion efficiency score; and combining the path length score, energy consumption score, and task completion efficiency score to generate a path scoring matrix; Step S52: perform fuzzy comprehensive evaluation based on the path scoring matrix; perform multi-objective decision-making in combination with pre-stored path constraints; perform sensitivity analysis on the evaluation results; select and generate the optimal inspection path based on the analysis results; Step S53, establishing status monitoring indicators based on real-time environmental data collected in real time; setting adjustment trigger conditions according to the optimal inspection path; building an emergency plan database; combining the status monitoring indicators, adjustment trigger conditions and emergency plan data to generate an emergency adjustment strategy.

7. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S6 is specifically as follows: Step S61, verifying kinematic constraints based on the optimal inspection path; checking collision risks according to a pre-stored set of spatial constraints; evaluating passing performance indicators; combining verification and evaluation results to generate a feasibility evaluation result; Step S62: analyzing potential risk points based on the optimal inspection path; Evaluate emergency response capabilities in conjunction with emergency adjustment strategies; Verify safety margin parameters; Combine the analysis and assessment results to generate a safety assessment report; Step S63: Perform a comprehensive assessment based on the feasibility assessment results and the safety assessment report; and make necessary optimization adjustments to the problems found in the assessment; Confirm the final inspection path plan; generate the verified inspection path.

8. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S11 is specifically as follows: Step S111, reading component geometry information from pre-stored original BIM model data; performing boundary tracking processing on the geometry information to extract boundary vertices; performing spatial continuity analysis on the boundary vertices; organizing the processed boundary vertices to generate a spatial boundary point coordinate set; Step S112: determining the boundary of the space range based on the coordinate set of the space boundary points; and constructing a three-dimensional grid space using an adaptive grid division method; Uniformly sampling the three-dimensional grid space to generate a sampling point cloud; Computing an occupancy state value of each voxel based on the sampling point cloud; Step S113, determining a calculation area based on a set of spatial boundary point coordinates; constructing a local coordinate system for each boundary point; calculating geometric features in the local coordinate system; and converting the geometric features into local space description values; Step S114, combining the spatial boundary point coordinate set, the occupancy state value and the local space description value; normalizing the combined features; and combining the normalized features to generate a spatial geometric feature set.

9. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1 is characterized in that: Step S13 is specifically as follows: Step S131, reading component basic information from pre-stored original BIM model data; converting the component basic information into standardized attributes using an attribute mapping table; A hierarchical classification approach is used to organize standardized attributes; Establish an attribute dependency graph according to the association relationship between attributes; perform integrity verification based on the attribute dependency graph; combine the verified attributes to generate a component attribute set; Step S132: establishing a spatial function association diagram based on the regional connectivity relationship in the spatial topological feature set; performing preliminary function identification using a pre-stored spatial function rule library; and performing function verification based on the functional constraint relationship of adjacent spaces; The function propagation probability is calculated by using the space syntax analysis method; the function type is optimized based on the function propagation probability; and the function type label of each spatial area is generated; Step S133, read the functional type label of each area; calculate the basic importance based on the pre-stored regional level rules; calculate the regional accessibility index in combination with the spatial topological feature set; calculate the regional influence using the spatial centrality analysis method; calculate the usage intensity based on the regional functional density distribution; weightedly combine the basic importance, accessibility index, regional influence and usage intensity to generate an importance value; Step S134, structuring the component attribute set; mapping the function type label according to the regional code; Normalize the importance values; The above data are integrated using feature fusion algorithm; the integrated data are checked for consistency; and a semantic feature set is generated.

10. The method for generating a drone intelligent inspection path for the interior of a building based on BIM model feature extraction according to claim 1, characterized in that: Step S22 is specifically as follows: Step S221, extracting function label information from the semantic feature set; performing semantic layering processing on the function label information; performing function normalization processing based on the semantic ontology library; hierarchically organizing the normalized function nodes; performing correlation analysis on the organized nodes; organizing the processed nodes to generate a semantic node set; Step S222, reading node pair information from the semantic node set; calculating the semantic similarity between nodes based on the pre-stored functional association rule library; The indirect association strength is calculated using the semantic transfer probability model; Calculate the comprehensive association degree by combining direct association and indirect association; filter the effective connection according to the association degree threshold; generate the semantic edge set from the filtered node connection relationship; Step S223, extracting region importance information based on the semantic feature set; Calculate the basic weights based on the hierarchical levels of the semantic nodes; Calculate structural weights based on node connectivity; Calculate position weights using spatial position relationships; A multi-criteria weighted method is used to fuse various weights; Generate weight values ​​of semantic nodes; Step S224, encoding and marking the semantic node set; Create an adjacency list to represent the semantic edge set; Map the semantic weight values ​​to the corresponding nodes; Use graph structure organization method to integrate node, edge and weight information; verify the consistency of the integrated graph structure; Generate semantically enhanced graphs.

Images