Construction potential safety hazard analysis system and method based on image recognition

Abstract

The invention provides a construction potential safety hazard analysis system and method based on image recognition, and relates to the technical field of potential safety hazard analysis, and the method comprises the steps: obtaining an excavation video of excavated earthwork and a three-dimensional point cloud model of unexcavated earthwork; determining a plurality of initial safety excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of preliminary excavation lines based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, wherein each preliminary excavation line passes through each initial safe excavation point; generating a simulation excavation video of each preliminary excavation line based on the preliminary excavation lines, the three-dimensional point cloud model of the unexcavated earthwork and the excavation video of the excavated earthwork; and determining a target excavation line based on the simulated excavation video of each preliminary excavation line. The method can accurately determine the target excavation line of the earthwork excavation operation.

Description

Technical Field

[0001] The present invention relates to the technical field of safety hazard analysis, and in particular to a construction safety hazard analysis system and method based on image recognition. Background Art

[0002] In the field of construction, with the increasing complexity of projects, the safety hazard analysis and construction efficiency of earthwork excavation operations face severe challenges. Traditional construction safety management mainly relies on manual on-site surveys and empirical judgments. It is difficult to accurately quantify the geological structural stability of unexcavated earthwork, and there is a lack of real-time monitoring methods for the dynamic changes of the soil in the excavated area. At the same time, when manually planning excavation routes, it is impossible to efficiently use three-dimensional geological data for global safety assessments, resulting in long excavation scheme design cycles and high risks. Especially in scenarios such as deep foundation pits and complex rock formations, it is easy to cause collapses, landslides and other accidents due to delayed safety hazard inspections. In existing technologies, it is difficult to achieve full-process digital modeling of construction scenarios through single sensor monitoring or simple algorithm analysis, and there is a lack of intelligent closed-loop management capabilities from data collection, risk analysis to solution optimization, resulting in low efficiency of construction safety control and an inability to meet the needs of modern engineering for refined and intelligent management.

[0003] Therefore, how to accurately determine the target excavation route for earthwork excavation operations is a problem that needs to be solved urgently. Summary of the Invention

[0004] The main technical problem solved by the present invention is how to accurately determine the target excavation route for earthwork excavation operations.

[0005] According to a first aspect, the present invention provides a construction safety hazard analysis method based on image recognition, comprising: obtaining an excavation video of excavated earth and a three-dimensional point cloud model of unexcavated earth; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth; determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0006] In one possible implementation, the determining of the target excavation route based on the simulated excavation video of each preliminary excavation route includes: constructing a knowledge graph, the knowledge graph including multiple preliminary excavation route nodes and multiple edges between the multiple preliminary excavation route nodes, the node feature of each preliminary excavation route node including the simulated excavation video of the preliminary excavation route, and the edges between the preliminary excavation route nodes being the positional relationship between different routes; processing the knowledge graph based on a graph neural network to determine multiple target safe excavation points; determining multiple preferred excavation routes based on the multiple target safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preferred excavation route passing through each target safe excavation point; generating a simulated excavation video of each preferred excavation route based on the multiple preferred excavation routes, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; determining the excavation information of each preferred excavation route using a mining processing model based on the simulated excavation video of each preferred excavation route; and determining the target excavation route based on the excavation information of each preferred excavation route.

[0007] In one possible implementation, the generating of a simulated excavation video for each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork includes: generating a simulated excavation video for each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork using a generative adversarial network.

[0008] In a possible implementation, the mining processing model is a Transformer model.

[0009] According to a second aspect, the present invention provides a construction safety hazard analysis system based on image recognition, comprising: an acquisition module for acquiring an excavation video of excavated earth and a three-dimensional point cloud model of unexcavated earth; a safety point determination module for determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth; a route generation module for determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preliminary excavation route passing through each initial safe excavation point; a simulation generation module for generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; and a target determination module for determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0010] In one possible implementation, the target determination module is also used to: construct a knowledge graph, the knowledge graph includes multiple preliminary excavation route nodes and multiple edges between the multiple preliminary excavation route nodes, the node feature of each preliminary excavation route node includes a simulated excavation video of the preliminary excavation route, and the edges between the preliminary excavation route nodes are the positional relationships between different routes; process the knowledge graph based on a graph neural network to determine multiple target safe excavation points; determine multiple preferred excavation routes based on the multiple target safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preferred excavation route passes through each target safe excavation point; generate a simulated excavation video of each preferred excavation route based on the multiple preferred excavation routes, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; determine the excavation information of each preferred excavation route using a mining processing model based on the simulated excavation video of each preferred excavation route; and determine the target excavation route based on the excavation information of each preferred excavation route.

[0011] In one possible implementation, the simulation generation module is specifically used to generate a simulated excavation video of each preliminary excavation route using a generative adversarial network based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

[0012] In a possible implementation, the mining processing model is a Transformer model.

[0013] According to a third aspect, an embodiment of the present invention provides an electronic device, comprising: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: obtaining an excavation video of excavated earth and a three-dimensional point cloud model of unexcavated earth; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth; determining a plurality of preliminary excavation routes based on the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned construction safety hazard analysis method based on image recognition, the method comprising: obtaining an excavation video of the excavated earth and a three-dimensional point cloud model of the unexcavated earth; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth; determining a plurality of preliminary excavation routes based on the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0015] The present invention provides a construction safety hazard analysis system and method based on image recognition. The method includes obtaining an excavation video of excavated earth and a three-dimensional point cloud model of unexcavated earth; determining multiple initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth; determining multiple preliminary excavation routes based on the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earth, and the excavation video of the excavated earth; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route. This method can accurately determine the target excavation route for earthwork excavation operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a flow chart of a construction safety hazard analysis method based on image recognition provided by an embodiment of the present invention; Figure 2 A schematic diagram of a process for determining a target excavation route provided by an embodiment of the present invention; Figure 3 A schematic diagram of a construction safety hazard analysis system based on image recognition provided by an embodiment of the present invention; Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0017] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present invention to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted under different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present invention are not shown or described in the specification. This is to avoid the core of the present invention being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0018] In an embodiment of the present invention, there is provided Figure 1 The construction safety hazard analysis method based on image recognition shown in FIG. 1 includes steps S1 to S5: Step S1: Obtain an excavation video of the excavated earthwork and a three-dimensional point cloud model of the unexcavated earthwork.

[0019] Excavated earthwork is the earthwork area where excavation work has been completed during the construction process.

[0020] Unexcavated earthwork refers to the original earthwork area within the construction scope that has not yet been excavated.

[0021] The excavation video of the excavated earth is a video shot by a camera to record the construction process of the excavated earth area.

[0022] A 3D point cloud model of an unexcavated area is a collection of discrete points captured using laser scanning technology. Each point contains precise X, Y, and Z coordinates and reflectivity properties. This 3D point cloud model fully reconstructs the surface morphology and spatial structure, accurately restoring the original topography and geological structure of the unexcavated area.

[0023] Step S2: determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork.

[0024] In some embodiments, a safety point determination model can be used to determine multiple initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earth. The safety point determination model is a convolutional neural network model. The input of the safety point determination model is the three-dimensional point cloud model of the unexcavated earth, and the output of the safety point determination model is multiple initial safe excavation points.

[0025] Convolutional neural network models include convolutional neural networks (CNNs), a model architecture designed for grid-like data in the field of deep learning. The core of a convolutional neural network consists of convolutional layers, pooling layers, activation functions, and fully connected layers. Convolutional layers use learnable convolution kernels to slide over the data to extract local features, while also sharing weights to reduce the number of parameters. Pooling layers reduce data dimensionality to enhance the model's robustness to data deformations. Activation functions impart nonlinear expressiveness to the model. Fully connected layers perform final classification or regression tasks based on the extracted features. Convolutional neural networks can be used to process complex data types such as 3D point cloud models. Their advantage lies in their ability to automatically learn characteristic patterns in the data and reduce the cost and error of manually designing features.

[0026] The initial safe excavation point is a basic safe point in the unexcavated earthwork suitable for excavation, as output by the safety point determination model. The initial safe excavation point satisfies basic safety constraints such as geological stability and accessibility.

[0027] The 3D point cloud model of the unexcavated earthwork is a digital representation of the construction site terrain. It contains a large number of discrete points, each of which contains precise 3D coordinate information. These points can fully present details such as the slope, concave and convex shape, and geological faults of the earthwork, thereby providing intuitive and accurate basic data for construction planning. The 3D point cloud model data can be converted into structured data that can be processed by a convolutional neural network through multi-view projection processing. The convolutional neural network can use its local feature extraction capabilities to accurately capture geometric features related to excavation safety, such as the slope of the earthwork and loose areas. It then analyzes the local geometric information of the point cloud and outputs the safety probability of each point. Points above the threshold will be marked, and finally the convolutional neural network can be combined with the operating requirements of the construction machinery and geological conditions to screen and optimize the initial safe excavation points.

[0028] In some embodiments, the safety point determination model includes a terrain analysis layer, a regional assessment layer, and a safety point determination layer. The terrain analysis layer, regional assessment layer, and safety point determination layer all include convolutional neural network structures. The terrain analysis layer inputs a three-dimensional point cloud model of the unexcavated earthwork, and the terrain analysis layer outputs a terrain slope map, fault location markers, and surface concave-convex area distribution. The regional assessment layer inputs a terrain slope map, fault location markers, and surface concave-convex area distribution, and the regional assessment layer outputs multiple segmented areas, a slope safety score for each area, a fault risk level for each area, and a mechanical access score for each area. The safety point determination layer inputs multiple segmented areas, a slope safety score for each area, a fault risk level for each area, and a mechanical access score for each area, and the safety point determination layer outputs multiple initial safe excavation points.

[0029] The terrain slope map is a visualization of the unexcavated earthwork terrain slope distribution calculated using 3D point cloud model data. The terrain slope map marks the slope size of different areas (such as flat areas and steep slope areas) and can intuitively reflect the degree of undulation of the unexcavated earthwork terrain.

[0030] The fault location marker is a location identifier of the geological fault area identified in the 3D point cloud model data. It marks the specific coordinate range of the fault zone and can indicate high-risk geological areas for construction.

[0031] The surface concave-convex area distribution is a distribution annotation of the raised and concave areas on the surface of the unexcavated earth terrain. This annotation can be used to distinguish the spatial positions of different terrain forms, and then assist in judging the difficulty of mechanical passage.

[0032] Multiple segmentation regions are used to refine terrain safety analysis. Multiple segmentation regions divide the overall 3D point cloud model into multiple irregularly shaped regions based on terrain characteristics (such as slope changes, fault distribution, and surface convexity). Each region has independent terrain attributes, facilitating the calculation of targeted safety indicators.

[0033] The slope safety score of each area is a quantitative score of the maximum slope value of each divided area. The higher the score, the flatter the slope and the higher the construction safety.

[0034] The fault risk level of each area is assessed based on the distance between the divided area and the fault zone. The higher the level, the greater the risk is closer to the fault zone.

[0035] The mechanical accessibility score for each area is based on the flatness of the terrain within the divided area, such as whether there are bumps and depressions, to determine the difficulty of smooth mechanical passage. The mechanical accessibility score is used to evaluate construction accessibility.

[0036] Different layers are responsible for different levels of information processing. The terrain analysis layer extracts intuitive terrain features from the 3D point cloud model. The regional assessment layer converts terrain features into segmented regions for detailed analysis and quantifiable regional safety indicators. The safety point determination layer selects initial safe excavation points based on regional safety indicators. This layered approach breaks down complex terrain safety analysis into a modular process of feature extraction, regional scoring, and point determination. Each layer can focus on a single task, improving processing efficiency and facilitating targeted accuracy optimization of each step. Ultimately, the reliability and rationality of the initial safe excavation points can be ensured.

[0037] Step S3 : determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, wherein each preliminary excavation route passes through each initial safe excavation point.

[0038] In some embodiments, a route planning model can be used to determine multiple preliminary excavation routes based on the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth. The route planning model is a deep neural network model. The input of the route planning model is the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earth, and the output of the route planning model is multiple preliminary excavation routes.

[0039] Deep neural network models include deep neural networks (DNNs). DNNs are machine learning models with multi-layered neural structures that mimic the workings of neurons in the brain to learn and process complex data. DNNs consist of an input layer, multiple hidden layers, and an output layer, with neurons connected by weights. Data enters the input layer, undergoes neuron activation operations and feature extraction in the hidden layers, and ultimately produces results in the output layer. As the number of network layers increases, DNNs are able to automatically learn deep, abstract feature representations of the data.

[0040] The multiple preliminary excavation routes are multiple possible excavation paths determined by the route planning model, and each preliminary excavation route passes through each initial safe excavation point.

[0041] Deep neural networks can extract features from complex spatial data in three-dimensional point cloud models, identifying key information such as slope and obstacle distribution in earthwork terrain. Deep neural networks can also learn the positional relationships of initial safe excavation points and, combined with terrain characteristics, simulate a variety of route combinations that meet construction requirements. Furthermore, by incorporating constraints such as construction safety and efficiency as optimization objectives, deep neural networks can screen multiple reasonable preliminary excavation routes from a vast number of potential paths, providing a rich set of candidate paths for construction plan development.

[0042] Step S4: generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

[0043] In some embodiments, a simulated excavation video of each preliminary excavation route may be generated using a generative adversarial network based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

[0044] A generative adversarial network (GAN) is a deep learning architecture consisting of two neural networks: a generator and a discriminator. The generator generates data, while the discriminator attempts to distinguish between generated data and real data. Through adversarial training, the generator and discriminator are continuously optimized, ultimately enabling the GAN to generate highly realistic data that conforms to specific patterns.

[0045] The simulated excavation video is a virtual video output by the generative adversarial network to simulate the earth excavation process along the preliminary excavation line.

[0046] Generative adversarial networks (GANs) possess powerful feature learning and adversarial optimization capabilities. Their generator extracts terrain features from a 3D point cloud model of unexcavated earthwork and converts preliminary excavation routes into spatiotemporal trajectories. It then learns the dynamic patterns of mechanical motion from excavation videos of excavated earthwork, integrating this information into a video frame sequence. The discriminator assesses the authenticity of the generated results based on visual features from real excavation videos and then guides the generator to refine details through gradient feedback. This collaborative training process between the generator and discriminator ensures that the resulting simulated excavation videos conform to construction logic and approximate real-world visual quality and temporal coherence.

[0047] Step S5: determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0048] In some embodiments, Figure 2 A schematic diagram of a process for determining a target excavation route is provided in an embodiment of the present invention. The process for determining a target excavation route includes steps S21 to S26: Step S21, constructing a knowledge graph, wherein the knowledge graph includes multiple preliminary mining route nodes and multiple edges between the multiple preliminary mining route nodes, the node feature of each preliminary mining route node includes a simulated mining video of the preliminary mining route, and the edges between the preliminary mining route nodes are the positional relationships between different routes.

[0049] A knowledge graph is a data structure composed of nodes and edges, where nodes are used to represent various entities, and edges are used to represent the relationships between entities. In some embodiments, each node can represent each preliminary excavation route, and edges are used to describe the positional relationship, sequence, etc. between different preliminary excavation routes. Each node feature is also accompanied by attribute information, such as a simulated excavation video of the preliminary excavation route. Through this structured representation method, the knowledge graph can integrate complex and scattered information into an interconnected knowledge network system, thereby clearly presenting the attributes of each route and its spatial relationship.

[0050] Step S22: Process the knowledge graph based on the graph neural network to determine multiple target security mining points.

[0051] A graph neural network (GNN) is a deep learning model that can operate directly on knowledge graphs. It establishes an information transfer mechanism between nodes, enabling each node to aggregate feature information from adjacent nodes and its own. Through multiple layers of iterative updates, the GNN learns the feature representation of the entire knowledge graph and the complex relationship patterns between nodes, enabling tasks such as feature extraction and classification prediction from the knowledge graph. The GNN takes the knowledge graph as input and outputs multiple target security mining points.

[0052] Multiple target safe excavation points are optimized for safety and construction feasibility, identified through knowledge graph analysis using a graph neural network. The selection of target safe excavation points integrates dynamic construction performance with global coordination, eliminating conflicting points and prioritizing highly connected points to better meet actual construction needs.

[0053] The knowledge graph provides a rich information foundation for graph neural network processing. Nodes in the knowledge graph represent preliminary excavation routes and constitute the basic unit of analysis. Node features include simulated excavation videos that visually display the dynamic details of each route's construction process, helping the graph neural network learn safety factors and potential risks in excavation operations. Edges represent the positional relationships between routes, allowing the graph neural network to perceive spatial connections between routes, such as whether there are risks of cross-operation and whether efficient connecting paths can be formed.

[0054] The graph neural network learns node features and edge relationships, propagates and aggregates information on the knowledge graph, and then comprehensively evaluates the safety and feasibility of each preliminary excavation line node, thereby screening out target safe excavation points with high safety and easy construction.

[0055] Step S23 : determining a plurality of preferred excavation routes based on the plurality of target safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, wherein each preferred excavation route passes through each target safe excavation point.

[0056] In some embodiments, a preferred route planning model can be used to determine multiple preferred excavation routes based on the multiple target safe excavation points and the three-dimensional point cloud model of the unexcavated earth. The preferred route planning model is a deep neural network model. The input of the preferred route planning model is the multiple target safe excavation points and the three-dimensional point cloud model of the unexcavated earth. The output of the preferred route planning model is multiple preferred excavation routes.

[0057] Multiple preferred excavation routes are determined by the optimal route planning model based on a 3D point cloud analysis of multiple target safe excavation points and unexcavated earthwork, taking into account factors such as construction safety, machine efficiency, and construction time and cost. These multiple preferred excavation routes provide multiple reliable candidate options for construction planning, facilitating comparative evaluation based on multiple dimensions, including safety (avoiding areas prone to collapse), feasibility (suitability for machine operation), and economy (reducing route detours). This helps determine the excavation route that best suits actual construction needs.

[0058] With its powerful end-to-end learning capabilities, deep neural networks can directly map the location information of target safe excavation points and the terrain data of the three-dimensional point cloud model into excavation routes. By automatically learning spatial topological relationships and construction constraints through multi-layer neurons, it can quickly determine multiple optimal excavation routes that take into account both safety and construction efficiency.

[0059] Step S24 : generating a simulated excavation video of each preferred excavation route based on the multiple preferred excavation routes, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

[0060] In some embodiments, a mining deduction model can be used to generate a simulated excavation video of each preferred excavation route based on the multiple preferred excavation routes, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork. The mining deduction model is a generative adversarial network, and the input of the mining deduction model is the multiple preferred excavation routes, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork. The output of the mining deduction model is a simulated excavation video of each preferred excavation route.

[0061] The simulated excavation video of the preferred excavation route is a virtual construction process video simulated by the excavation deduction model. The video can intuitively present dynamic details such as earth deformation, mechanical movement and environmental changes when the machinery operates along the preferred route.

[0062] Step S25 : determining excavation information of each preferred excavation route using an excavation processing model based on the simulated excavation video of each preferred excavation route.

[0063] The mining processing model is a Transformer model, the input of the mining processing is the simulated mining video of each preferred mining route, and the output of the mining processing is the mining information of each preferred mining route.

[0064] The Transformer model is a deep learning model based on self-attention. It uses a multi-head attention mechanism to capture long-range dependencies in sequence data. The core of the Transformer model consists of an encoder and a decoder. The encoder uses self-attention to learn the contextual representation of the input sequence, while the decoder generates the target sequence based on the encoder output.

[0065] The excavation information for a preferred excavation route is a set of structured construction indicators extracted by analyzing simulated excavation videos of the preferred excavation route using a mining processing model. The core of this information mining is to transform the dynamic visual information in the simulated excavation videos into quantifiable and comparable key indicators. These include terrain adaptation data, time-series operation data, safety risk indicators, and construction efficiency parameters. For example, from a simulated excavation video of a preferred excavation route, data such as "a certain route needs to cross an area with a slope greater than 25°, and the earthwork slip warning value reaches 75%," and "the fuel consumption rate of the machine on this route is 15% higher than the baseline value" can be extracted. This excavation information for each preferred excavation route allows the safety and efficiency of different routes to be quantified and ranked, providing a basis for selecting the target excavation route.

[0066] Terrain adaptation data includes three-dimensional space-related parameters such as the extreme terrain slope of the excavation line coverage area, obstacle detour distance, and mechanical operation space margin.

[0067] Time series operation data includes time dimension information such as the total excavation time of a single section of line, the proportion of time spent on process connection, and the proportion of machine idle time.

[0068] Safety risk indicators include safety-related quantitative results such as earthwork slip warning value, machinery collision warning probability, and environmental compliance score (such as dust and noise levels).

[0069] Construction efficiency parameters include efficiency evaluation indicators such as excavation volume per unit length, equipment fuel consumption rate, and capacity utilization rate (actual output / theoretical capacity).

[0070] Simulated excavation videos contain complete spatiotemporal information about the construction process, and the rich visual information they contain can be parsed into structured excavation information. Three-dimensional terrain changes within video frames reflect the progress and shape of earthwork excavation, while the machine's motion trajectory reflects the work path and efficiency. Time-series changes record construction time and process connections. By analyzing pixel-level features (such as machine position and soil color changes) and motion characteristics (such as robotic arm movement speed and material transport frequency) in the video, the model can quantify construction parameters such as excavation depth, excavation volume, and equipment load. Environmental interaction information (such as dust dispersion range and machine collision risk) in the video can also be used to assess safety risks.

[0071] The Transformer model possesses a self-attention mechanism and the ability to efficiently model the spatiotemporal features of simulated mining videos. The Transformer model can convert a sequence of video frames into a computable sequence of image blocks, and through the self-attention mechanism, it captures the temporal dependencies between frames and the spatial relationships within frames. It can process long video sequences without convolutional or recursive structures, thus avoiding the vanishing gradient problem. The Transformer model's encoder also supports multimodal input, fusing video visual features with three-dimensional point cloud spatial data to enhance feature complementarity. For simulated mining videos, the Transformer model can accurately identify key information such as process start and end points, machine trajectories, and earthwork deformation. It can also parallelize the calculation of risk and efficiency indicators for different regions, thereby rapidly outputting structured mining information.

[0072] Step S26: determining a target excavation route based on the excavation information of each preferred excavation route.

[0073] In some embodiments, a target mining route can be determined based on the mining information of each preferred mining route using a target route decision model, wherein the target route decision model is a support vector machine, the input of the target route decision model is the mining information of each preferred mining route, and the output of the target route decision model is the target mining route.

[0074] A support vector machine (SVM) is a supervised learning model based on statistical learning theory. Its core concept is to achieve data classification and regression by constructing an optimal hyperplane. The SVM maps linearly inseparable data in a low-dimensional space to a high-dimensional space using a kernel function by maximizing the margin between samples, thereby finding the linear classification boundary. In some embodiments, the SVM can take mining information of a preferred mining route as input and then, through training, learn decision rules based on dimensions such as safety and efficiency. It then outputs a classification result (e.g., feasible / infeasible) and a ranking score for the route, enabling automated decision-making from mining information of the preferred mining route to the target mining route.

[0075] The target excavation route is the final construction execution path selected by the target route decision model based on excavation information from the preferred excavation route. It is the optimal route that meets high safety requirements, high construction efficiency, controllable costs, and environmental compliance. The target excavation route has clear three-dimensional spatial coordinates, a time-series operation plan, and an equipment configuration plan.

[0076] Support vector machines are suitable for analyzing structured mining information to identify target excavation routes that meet construction requirements. Using its structured risk minimization theory, support vector machines can efficiently classify and sort high-dimensional mining information (such as safety risk values ​​and construction efficiency parameters) using a limited sample size. Kernel functions are then used to map the low-dimensional indicator space to a higher dimension, capturing the nonlinear relationships between different routes in terms of safety, efficiency, and cost, ultimately constructing an optimal decision boundary. For multi-dimensional evaluation data of optimal excavation routes, support vector machines can not only quickly filter out clearly infeasible options through binary classification, but also rank eligible routes using a multi-classification strategy, ultimately outputting the target route with the best overall performance.

[0077] Based on the same inventive concept, Figure 3 A schematic diagram of a construction safety hazard analysis system based on image recognition provided by an embodiment of the present invention, wherein the construction safety hazard analysis system based on image recognition includes: An acquisition module 31 is used to acquire an excavation video of the excavated earthwork and a three-dimensional point cloud model of the unexcavated earthwork; a safety point determination module 32, configured to determine a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; a route generating module 33 for determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preliminary excavation route passing through each initial safe excavation point; a simulation generation module 34 for generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; The target determination module 35 is configured to determine a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0078] Based on the same inventive concept, an embodiment of the present invention provides an electronic device, such as Figure 4As shown, it includes: a processor 41; a memory 42; and a computer program; wherein the computer program is stored in the memory 42 and is configured to be executed by the processor 41 to implement the construction safety hazard analysis method based on image recognition as provided above, the method including: obtaining an excavation video of the excavated earthwork and a three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0079] Based on the same inventive concept, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by the processor 41, implements the aforementioned construction safety hazard analysis method based on image recognition, the method comprising: obtaining an excavation video of the excavated earthwork and a three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of preliminary excavation routes based on the multiple initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; and determining a target excavation route based on the simulated excavation video of each preliminary excavation route.

[0080] The construction safety hazard analysis method based on image recognition provided in the embodiments of the present application can be applied to terminal devices (such as mobile phones), tablet computers, laptops, ultra-mobile personal computers (UMPCs), handheld computers, netbooks, personal digital assistants (PDAs), wearable devices (such as smart watches, smart glasses or smart helmets, etc.), augmented reality (AR) and virtual reality (VR) devices, smart home devices, car computers and other electronic devices. The embodiments of the present application do not impose any restrictions on this.

[0081] While the basic concepts have been described above, it will be apparent to those skilled in the art that the detailed disclosure is merely illustrative and does not limit this specification. Although not explicitly stated herein, various modifications, improvements, and revisions to this specification may be made by those skilled in the art. Such modifications, improvements, and revisions are suggested in this specification and remain within the spirit and scope of the exemplary embodiments of this specification.

[0082] This specification also uses specific terms to describe the embodiments of this specification. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "one embodiment," "an embodiment," or "an alternative embodiment" two or more times in different locations in this specification do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of this specification may be appropriately combined.

[0083] In addition, unless expressly stated in the claims, the order of the processing elements and sequences, the use of alphanumeric characters, or the use of other names described in this specification are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some of the invention embodiments currently considered useful through various examples, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the spirit and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0084] Similarly, it should be noted that, in order to simplify the presentation of this specification and thus facilitate understanding of one or more embodiments of the invention, the foregoing descriptions of the embodiments of this specification sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this disclosure method does not imply that the subject matter of this specification requires more features than those recited in the claims. In fact, an embodiment may have fewer features than all of the features of a single disclosed embodiment.

[0085] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A construction safety hazard analysis method based on image recognition, characterized in that: include: Obtain excavation videos of excavated earthwork and 3D point cloud models of unexcavated earthwork; determining a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; determining a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preliminary excavation route passing through each initial safe excavation point; generating a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; A target excavation route is determined based on the simulated excavation video of each preliminary excavation route.

2. The construction safety hazard analysis method based on image recognition according to claim 1, characterized in that: The determining of the target excavation route based on the simulated excavation video of each preliminary excavation route comprises: Constructing a knowledge graph, the knowledge graph including a plurality of preliminary mined route nodes and a plurality of edges between the plurality of preliminary mined route nodes, wherein a node feature of each preliminary mined route node includes a simulated mining video of the preliminary mined route, and the edges between the preliminary mined route nodes represent positional relationships between different routes; Process the knowledge graph based on graph neural network to determine multiple target security mining points; determining a plurality of preferred excavation routes based on the plurality of target safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preferred excavation route passing through each target safe excavation point; generating a simulated excavation video of each preferred excavation route based on the plurality of preferred excavation routes, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; Determining excavation information for each preferred excavation route using a mining processing model based on the simulated excavation video of each preferred excavation route; A target excavation route is determined based on the excavation information of each preferred excavation route.

3. The construction safety hazard analysis method based on image recognition according to claim 1, characterized in that: Generating a simulated excavation video for each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork includes: A simulated excavation video of each preliminary excavation route is generated using a generative adversarial network based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

4. The construction safety hazard analysis method based on image recognition according to claim 2, characterized in that: The mining processing model is a Transformer model.

5. The construction safety hazard analysis system based on image recognition is characterized by: include: An acquisition module, for acquiring excavation videos of excavated earthwork and three-dimensional point cloud models of unexcavated earthwork; a safety point determination module, configured to determine a plurality of initial safe excavation points based on the three-dimensional point cloud model of the unexcavated earthwork; a route generating module, configured to determine a plurality of preliminary excavation routes based on the plurality of initial safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preliminary excavation route passing through each initial safe excavation point; a simulation generation module, configured to generate a simulated excavation video of each preliminary excavation route based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; A target determination module is configured to determine a target excavation route based on the simulated excavation video of each preliminary excavation route.

6. The construction safety hazard analysis system based on image recognition according to claim 5, characterized in that: The target determination module is further configured to: Constructing a knowledge graph, the knowledge graph including a plurality of preliminary mined route nodes and a plurality of edges between the plurality of preliminary mined route nodes, wherein a node feature of each preliminary mined route node includes a simulated mining video of the preliminary mined route, and the edges between the preliminary mined route nodes represent positional relationships between different routes; Process the knowledge graph based on graph neural network to determine multiple target security mining points; determining a plurality of preferred excavation routes based on the plurality of target safe excavation points and the three-dimensional point cloud model of the unexcavated earthwork, each preferred excavation route passing through each target safe excavation point; generating a simulated excavation video of each preferred excavation route based on the plurality of preferred excavation routes, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork; Determining excavation information for each preferred excavation route using a mining processing model based on the simulated excavation video of each preferred excavation route; A target excavation route is determined based on the excavation information of each preferred excavation route.

7. The construction safety hazard analysis system based on image recognition according to claim 5, characterized in that: The simulation generation module is specifically used for: A simulated excavation video of each preliminary excavation route is generated using a generative adversarial network based on the preliminary excavation route, the three-dimensional point cloud model of the unexcavated earthwork, and the excavation video of the excavated earthwork.

8. The construction safety hazard analysis system based on image recognition according to claim 6, characterized in that: The mining processing model is a Transformer model.

9. An electronic device, characterized in that: include: processor; Memory; And a computer program; wherein, the computer program is stored in the memory and is configured to be executed by the processor to implement the construction safety hazard analysis method based on image recognition as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the construction safety hazard analysis method based on image recognition as described in any one of claims 1 to 4 is implemented.

Images