An adaptive highway low-altitude patrol method and system for extreme environment

Abstract

This invention relates to the field of highway inspection technology, specifically to an adaptive low-altitude highway inspection method and system for extreme environments. The method includes: collecting heterogeneous data from UAV airborne sensors, roadside weather stations, and geographic information systems; constructing an environmental feature vector including a sensor reliability index; retrieving an extreme scenario database based on this vector; running a hybrid planning model to generate a flexible task package including a main execution strategy and a scenario-response strategy; scheduling a UAV swarm to execute the flexible task package; calculating an environmental comprehensive score and dynamic link weights to dynamically reconstruct the network topology of the UAV swarm; adaptively enhancing the collected data to remove interference; and using a nonlinear confidence model to fuse the judgment results of multiple algorithms to generate a highway inspection judgment report. This invention ensures flight safety and communication stability in extreme environments through environment-driven flexible task planning and dynamic topology reconstruction.

Description

Technical Field

[0001] This invention relates to the field of highway inspection technology, and more specifically, to an adaptive low-altitude highway inspection method and system for extreme environments. Background Technology

[0002] As the arteries of modern economic development, highway networks often traverse complex geographical environments such as high-altitude mountains, deep canyons, and coastal areas. Ensuring the safe operation of these facilities requires regular, meticulous inspections to identify potential hazards such as road surface cracks, unstable slopes, and damage to ancillary facilities. With the rapid development of low-altitude economy and aviation technology, unmanned aerial vehicles (UAVs), with their high mobility, wide coverage, and ability to hover and conduct detailed inspections, have gradually become the main force in highway inspections. In practical applications, using UAV swarms equipped with high-definition cameras, LiDAR, and thermal imagers enables large-scale data collection and status monitoring of the road network. Especially in complex scenarios with rapidly changing weather conditions and harsh terrain, UAV systems are required to possess higher autonomy and environmental adaptability to ensure the continuity of inspection tasks and the reliability of data. Developing intelligent inspection systems capable of sensing environmental changes and adjusting operational strategies in real time has become a key direction in the evolution of smart transportation maintenance technology. However, existing UAV inspection technologies have significant limitations when facing complex and ever-changing extreme environments. On the one hand, most path planning algorithms rely on static pre-planning, lacking flexibility to cope with sudden weather changes. When encountering sudden strong crosswinds or heavy rainfall, UAVs often cannot adjust their flight strategies in time, easily leading to mission interruption or even crashes. On the other hand, the communication network of a cluster typically uses a fixed topology, which is prone to link instability or coordination failure in environments with electromagnetic interference or terrain obstruction. Furthermore, traditional data processing methods not only ignore sensor performance degradation caused by environmental factors but also lack an effective fusion mechanism for the results of multiple algorithms. This results in high false alarm and false negative rates for defect identification when processing blurred images acquired in harsh environments, making it difficult to meet the high-precision requirements of highway maintenance. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive low-altitude highway inspection method and system for extreme environments. Through environment-driven flexible task planning and dynamic topology reconstruction, it ensures flight safety and communication stability in extreme environments. By utilizing adaptive data augmentation and nonlinear multi-algorithm fusion, it significantly improves the inspection and judgment accuracy under harsh conditions and realizes all-weather intelligent maintenance.

[0004] This invention is achieved through the following technical solution:

[0005] An adaptive low-altitude highway inspection method for extreme environments, comprising the following steps: Heterogeneous data from UAV airborne sensors, roadside weather stations, and geographic information systems are collected through a protocol conversion gateway and then standardized to construct an environmental feature vector that includes a sensor reliability index. Based on the environmental feature vectors, an extreme scenario library is retrieved, and a hybrid programming model that integrates stochastic programming and robust optimization is run to generate a resilient task package that includes a main execution strategy and a scenario-response strategy. The system schedules a cluster of drones to execute the elastic task package and calculates the comprehensive environmental score and dynamic link weight in real time based on environmental feature vectors during execution, so as to dynamically reconstruct the network topology of the drone cluster. The collected data is adaptively enhanced to remove interference using an environment-degradation mapping knowledge base, and the judgment results of multiple algorithms are fused using a nonlinear confidence model to generate a highway inspection and judgment report.

[0006] Optionally, the construction includes an environmental feature vector containing a sensor reliability index. Its calculation formula is a 7-tuple:

[0007] in, For visibility, For precipitation intensity, The average wind speed, For wind shear intensity, Due to terrain complexity, Electromagnetic interference intensity, This is the sensor reliability index.

[0008] Optionally, the elastic task package is specifically implemented by solving the following objective function:

[0009] in, The main decision-making variable for the execution strategy. For extreme scenarios The following are the decision variables for coping strategies; The cost coefficient vector of the main execution strategy. For the scene The cost coefficient vector of the response strategy; For the scene The estimated probability of occurrence, In order to match the scene Weights associated with highway safety events; For robust risk aversion coefficient, This represents the total number of scenes in the scene library. This is the set of currently active scenes.

[0010] Optionally, the dynamic reconstruction of the network topology of the drone swarm includes constructing a dynamic neighbor set. Specifically: Define the node environment fitness scoring function ,in, to All are weighting coefficients; A candidate neighbor set is determined based on physical communication constraints, and the best neighbor set that satisfies these constraints is selected from the candidate neighbor set. The node is taken as the final neighbor, where, This is an adaptive threshold.

[0011] Optionally, the dynamic link weights are calculated. Specifically, it is based on the following calculation formula:

[0012] in, This is the distance after terrain correction. For signal-to-noise ratio, Due to the difference in wind speed between the two machines, For neighboring nodes Sensor reliability; These are the weighting coefficients; When updating the network topology, the dynamic link weights are normalized to execute an environment-aware distributed consensus protocol.

[0013] Optionally, the scheduling of the drone cluster to execute the elastic task package specifically includes: Continuous monitoring of environmental feature vectors The rate of change of each component; If the instantaneous value of any critical environmental component changes more than a preset threshold relative to the time of task startup, and this change triggers a scenario in the extreme scenario library... The drone swarm will then automatically switch to execute the corresponding scenario-response strategy. .

[0014] Optionally, the specific process of the adaptive interference reduction enhancement is as follows: Query Environment - Degradation Mapping Knowledge Base Based on the current environmental feature vector Predicted Degradation Type Set ; according to Dynamically assemble processing algorithm sequences from a predefined algorithm pool. It processes the raw collected data to output enhanced data.

[0015] Optionally, the judgment results obtained by fusing multiple algorithms using a nonlinear confidence model specifically include calculating the first... Confidence function of the algorithm :

[0016] in, It is the Sigmoid activation function. It is a linear weight vector. It is a symmetric matrix used to characterize the interactions between different environmental factors. For bias terms; Final assessment results Obtained through weighted calculation:

[0017] in, For the Kronecker delta function, For candidate results, For the algorithm The results of the analysis are represented as the original analysis results.

[0018] Optionally, it also includes the logic for determining the final assessment result, which is as follows: If the confidence level of the final assessment result is lower than the threshold, the environmental vector will be... Original assessment results and human feedback information Constructed as training samples; The parameters of the confidence function are updated using an online incremental learning algorithm. , , To minimize prediction error.

[0019] An adaptive low-altitude highway inspection system for extreme environments includes: The multi-source sensing and fusion module is configured to access heterogeneous data and extract environmental feature vectors containing sensor reliability indices. The predictive task planning module is configured to manage a library of extreme scenarios and run a hybrid planning model that combines stochastic programming and robust optimization to generate flexible task packages. The elastic collaborative scheduling module is configured to calculate the comprehensive environmental score of its candidate neighbor nodes for any node, and determine the candidate nodes whose comprehensive environmental score is higher than the preset adaptive threshold as neighbor nodes; the preset adaptive threshold is dynamically adjusted based on the current environmental feature vector. The intelligent judgment and self-evolution module is configured to drive data augmentation using an environment-degradation knowledge base and to fuse the judgment results of multiple algorithms using a nonlinear confidence model containing an interaction term matrix of environmental factors.

[0020] The technical solution of the present invention has at least the following advantages and beneficial effects: This invention, on the one hand, generates flexible task packages with main execution strategies and scenario response strategies by constructing multi-dimensional environmental feature vectors and employing a hybrid programming model that integrates stochastic programming and robust optimization. This enables the system to autonomously switch strategies when facing sudden risks, significantly improving flight safety and mission completion rate. On the other hand, the invention's dynamic network topology reconstruction mechanism based on environmental scoring ensures the reliability of communication links and the consistency of coordination within the cluster under strong interference conditions. Furthermore, this method not only utilizes an environment-degradation knowledge base to adaptively enhance the quality of perceived data but also employs a nonlinear confidence model to fuse results from multiple algorithms, effectively eliminating the impact of environmental noise on judgment accuracy. This ensures that the system can still output high-precision inspection reports under extreme conditions, achieving a closed-loop evolution from environmental perception to intelligent decision-making. Attached Figure Description

[0021] Figure 1 A flowchart illustrating the adaptive low-altitude highway inspection method for extreme environments provided by this invention. Figure 2 This is a schematic diagram illustrating the principle of the adaptive low-altitude highway inspection system for extreme environments provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0023] Summary of the Invention Concept: The core inventive concept of this invention lies in the following: First, constructing a cognitive foundation that comprehensively and dynamically reflects the inspection environment. This is achieved by integrating multi-source heterogeneous data, including data from the UAV's own sensors, ground meteorological facilities, and geospatial information, and innovatively introducing a reliability index of the sensors in the current environment, collectively forming a multi-dimensional environmental feature vector. Second, based on this environmental cognition, the system no longer uses rigid preset paths. Instead, it runs a hybrid programming model that integrates stochastic programming and robust optimization to proactively generate flexible task packages containing a main strategy and multiple scenario response strategies, thereby endowing the UAV swarm with the ability to anticipate and respond to sudden extreme environments. During the task execution phase, swarm collaboration no longer relies on a fixed communication topology. Instead, it achieves adaptive reconstruction of the network topology through real-time calculation of the comprehensive environmental score and dynamic link weights, ensuring collaborative robustness in harsh communication environments. Finally, for the collected data affected by environmental interference, the system uses an environment-degradation mapping knowledge base for adaptive interference removal and enhancement, and adopts a nonlinear confidence model to intelligently integrate the judgment results of multiple algorithms. This model can capture the complex interactions between environmental factors and finally output a high-precision judgment report. It also continuously optimizes its own model through business feedback, forming a complete technical closed loop from environmental perception to intelligent decision-making and then to self-evolution.

[0024] Based on the above-mentioned inventive concept, the present invention proposes an adaptive low-altitude highway inspection scheme for extreme environments, as described in the following embodiments.

[0025] Example 1 like Figure 1 As shown, Figure 1 This is a flowchart illustrating the adaptive low-altitude highway inspection method for extreme environments provided by the present invention. The method can be executed on a centralized ground control station, a cloud server, or a distributed computing system. Specifically, the method includes: S110: Collect heterogeneous data from UAV airborne sensors, roadside weather stations, and geographic information systems through a protocol conversion gateway, and construct an environmental feature vector containing sensor reliability indices after standardization processing.

[0026] It's important to explain that step S110 forms the perception and cognition foundation of the entire adaptive inspection process. Its application scenario involves deploying drone swarms for automated inspections on a vast and environmentally diverse highway network. These drones not only carry various sensors but also communicate with roadside weather stations along the route and with a backend geographic information system. The core task of S110 is to uniformly collect and process this data from different sources and in various formats, transforming it into standardized, machine-understandable quantitative indicators that comprehensively characterize the current operational environment—i.e., environmental feature vectors. The quality of this step directly determines the accuracy and effectiveness of all subsequent decisions.

[0027] S120. Based on the environmental feature vector, retrieve the extreme scenario library and run a hybrid programming model that integrates stochastic programming and robust optimization to generate an elastic task package containing the main execution strategy and the scenario-response strategy.

[0028] Step S120 is the system's "brain," responsible for proactive decision-making and planning. It receives the environmental feature vectors constructed by S110 and first matches them against a predefined library of extreme scenarios to determine what extreme weather or environmental challenges might be encountered in the current or future period. Subsequently, instead of generating a single, fixed flight plan, it runs an advanced hybrid planning model. This model simultaneously considers the probabilistic statistics of historical data and the risks under worst-case scenarios, thereby calculating a "resilient mission package." This mission package not only includes a primary execution strategy under normal circumstances but also pre-prepared response strategies for various possible extreme scenarios. This approach transforms the drone swarm from passive response to proactive defense, significantly improving mission success rate and safety.

[0029] S130. Schedule the UAV cluster to execute the elastic task package, and calculate the comprehensive environmental score and dynamic link weight in real time based on the environmental feature vector during the execution process, so as to dynamically reconstruct the network topology of the cluster.

[0030] Step S130 is the task execution and collaborative control phase. After receiving the elastic task package generated in S120, the UAV swarm begins to execute the main strategy. Unlike traditional fixed formation flight, the UAV swarm in this method is a flexible self-organizing network. During flight, each UAV continuously senses its surrounding environment and calculates a "comprehensive environmental score" in real time based on the latest environmental feature vectors to assess the quality of its own and its neighboring nodes' environments. Simultaneously, the "weights" of the communication links between UAVs are no longer fixed values ​​but dynamically change based on various environmental factors such as distance, signal quality, and wind speed differences. In this way, the swarm can intelligently choose to communicate with neighbors in better "situations," dynamically adjusting and optimizing the communication network topology of the entire swarm, thus maintaining efficient and reliable collaborative operation even under adverse conditions such as signal interference or terrain obstruction.

[0031] S140. Adaptive interference removal and enhancement of the collected data are performed using the environment-degradation mapping knowledge base, and the judgment results of multiple algorithms are fused using a nonlinear confidence model to generate a highway inspection and judgment report.

[0032] Step S140 is the data processing and intelligent analysis stage. Images or data collected by drones in extreme environments are often of poor quality, such as blurred images due to strong winds or images covered in stripes due to rainfall. This step first queries an "environment-degradation mapping knowledge base" based on the current environmental feature vector to intelligently determine what type of interference the data may have suffered, and automatically calls the most suitable algorithm combination for interference removal and enhancement. When identifying defects in the enhanced data, the system runs multiple different identification algorithms simultaneously and uses a non-linear confidence model to decide which algorithm's result to accept. This model can understand which algorithm's historical performance is more reliable under specific environments. By weighted fusion of the results from multiple algorithms, a highway inspection and assessment report with high confidence and accuracy is finally generated.

[0033] As described above, the method provided by the embodiments of the present invention, by constructing a comprehensive environmental understanding, performing forward-looking and flexible task planning, realizing adaptive cluster collaborative control, and intelligently enhancing and analyzing the collected data, forms a complete closed-loop solution that systematically solves the problem of UAV inspection operations in extreme environments.

[0034] Example 2 Based on the above embodiment 1, this embodiment will provide a more detailed explanation of the key technical steps involved.

[0035] First, the process of constructing the environmental feature vector in step S110 will be explained in detail.

[0036] In a specific example of the present invention, the environmental feature vector is defined as a seven-tuple. These seven components together constitute a comprehensive and detailed digital description of the inspection environment.

[0037] Quantity Representing horizontal visibility, its unit is typically meters. It is directly related to the effective working distance and image quality of optical sensors on the drone, such as high-definition cameras. When planning a mission, high visibility means that a higher flight altitude can be used, improving efficiency; while during mission execution, a sudden drop in visibility is an important basis for triggering risk warnings and adjusting strategies. This data can be estimated in real time using an atmospheric transmittance model through the forward-facing visual sensor onboard the drone, or it can be obtained directly from nearby roadside weather stations via a wireless link.

[0038] Quantity Representing the type and intensity of precipitation, it is a discrete enumeration value. For example, its value set can be defined as {no precipitation, light rain, moderate rain, heavy rain, torrential rain, snow, hail}. Different types and intensities of precipitation have drastically different effects on UAV flight: rain can affect the stability of flight control and propulsion systems and obstruct optical lenses; while snow or hail may cause physical damage to the airframe structure. This information is mainly derived through comprehensive analysis of reports from roadside weather stations and data from the UAV's onboard raindrop sensor, after classification and quantification.

[0039] Quantity This represents the average wind speed, measured in meters per second. This is one of the most critical factors affecting the flight stability and energy consumption of drones. Sustained strong winds significantly increase a drone's energy consumption, shorten its flight range, and may even prevent it from maintaining its planned flight path. This data can be calculated precisely using the drone's built-in inertial measurement unit (IMU) and GPS data, combined with airspeed tube readings, to accurately determine the current wind speed and direction.

[0040] Quantity Crosswind shear intensity is represented by meters per second (m / s). Compared to average wind speed, crosswind shear describes drastic changes in wind speed across space, especially when crossing canyons, buildings, or at the boundaries of different terrains. Strong wind shear poses a significant challenge to the attitude control of drones and is one of the main causes of flight instability and even crashes. Its calculation requires higher-frequency flight control data and complex aerodynamic models.

[0041] Quantity Representing terrain ruggedness, this is a dimensionless numerical value. It describes the degree of surface undulation in the inspection area. In plains, this value is close to 0; while in mountainous or hilly areas, it increases significantly. Terrain ruggedness not only affects the flight path planning of UAVs, requiring frequent altitude adjustments, but also directly relates to the quality of communication links between UAVs and between UAVs and ground stations, as mountain obstruction is a major cause of signal interruption. This data is typically calculated based on a high-precision digital elevation model (DEM) provided by a Geographic Information System (GIS) by running terrain relief or surface roughness algorithms within a specific area.

[0042] Quantity Electromagnetic interference (EMI) intensity, typically measured in dBm, represents the background noise level within the frequency band used for UAV communication. In areas near strong electromagnetic radiation sources such as high-voltage power lines, radar stations, and mobile communication base stations, EMI intensity increases significantly, directly leading to a decrease in the signal-to-noise ratio of the UAV's control and data transmission links, shortened communication distance, or even complete interruption. This data is collected in real-time by a dedicated spectrum monitoring module mounted on the UAV.

[0043] Quantity This is an innovative metric in this solution: the sensor reliability index. It's a dimensionless value between 0 and 1, used to comprehensively characterize the reliability of various sensors mounted on the UAV, such as cameras and radar, under specific environmental conditions. For example, in dense fog, the reliability index of the visual camera... It will significantly reduce; in environments with strong electromagnetic interference, the GPS positioning module's... It may also decrease. The index is calculated using a specialized reliability assessment model. This model considers three aspects: first, the sensor's built-in self-test (BIST) status flags; second, the sensor's performance statistics in similar environments based on historical data; and finally, the current environmental feature vector. This includes knowledge about the correlation between other components and the sensor's performance. For example, the knowledge base might store information such as "when rainfall intensity..." The model follows a rule that "during periods of heavy rain, the noise rate of the lidar point cloud data increases by 30%." By combining this information, the model outputs a probability value representing the likelihood that the sensor's current reading is accurate. The index enables the system to treat data from different sensors differently in subsequent decisions, and to place greater trust in information sources with high reliability.

[0044] Next, the process of generating the elastic task package in step S120 will be explained in detail.

[0045] The core of this process is to solve a hybrid programming model that combines stochastic programming and robust optimization. Its objective function takes the following form:

[0046] This objective function consists of three parts, which respectively reflect a comprehensive consideration of conventional costs, expected risks, and worst-case risks.

[0047] Part One This represents the expected cost of the main execution strategy. Among them, These are the decision variables for the first stage, representing the basic inspection plan that needs to be pre-defined before the uncertain environment is specifically observed. Examples include the takeoff point of the drone swarm, the approximate inspection route sequence, and the initial formation configuration. (Vector) This is the corresponding cost coefficient vector, whose elements quantify the cost of executing these basic decisions, such as the electricity consumed to fly one kilometer or the time cost of flying one minute. One of the optimization goals is to minimize this part of the conventional costs.

[0048] Part Two This represents the expected response cost under all possible extreme scenarios. Here, These are the decision variables for the second stage, representing the situation under a specific extreme scenario. The emergency adjustment plan adopted by the system when events such as "winds of level 5 or above" or "visibility less than 500 meters" occur. For example, the adjustment plan... This could include specific maneuvers such as reducing flight speed, choosing an alternative safe route, or returning to base. Vector This is the cost factor for these countermeasures, which includes not only additional energy consumption, but more importantly, the risk cost equivalent to the decrease in inspection coverage that may result from strategy adjustments. It is a scene The estimated probability of occurrence is based on historical meteorological and operational data. It is related to the scene The associated highway safety event weight is a scaling factor greater than 1, used to amplify the impact of high-risk scenarios in the objective function. For example, a scenario that might lead to the failure of slope collapse monitoring... The value will be much higher than in a scenario that only affects the clarity of road markings. This represents the total number of predefined scenarios in the scenario library. The goal of this part is to minimize the weighted average response cost across all possible scenarios.

[0049] Part Three This embodies the idea of ​​robust optimization, which is to avoid the risk of the worst case. It is based on the current environmental feature vector The set of activated scenarios identified. This item means that among all currently possible scenarios, the "worst-case" scenario that would result in the highest response cost is identified, and its cost is directly incorporated into the optimization objective. Parameters It is a risk aversion coefficient set by the user; it is a non-negative scalar. When A larger value indicates that the decision-maker is extremely risk-averse, and the optimization model will tend to choose a strategy with relatively small losses even in the worst-case scenario, even if the cost of this strategy is slightly higher under normal circumstances. Conversely, when... When the size is small, the model will focus more on optimizing average performance.

[0050] By solving this three-in-one optimization objective function, the system can obtain an optimal elastic task package that is both economical and robust, achieving a delicate balance between efficiency and security.

[0051] Then, the process of dynamically reconstructing the cluster network topology in step S130 will be explained in detail.

[0052] This process involves two core steps: constructing a dynamic neighbor set and calculating dynamic link weights.

[0053] Dynamic Neighbor Sets The construction of drone nodes The intelligent selection of communication partners involves two steps. The first step is preliminary screening based on physical communication constraints to determine a candidate neighbor set. This ensures that all candidate nodes are physically reachable. The second step is optimal selection based on environmental awareness within the candidate set. To this end, the system first defines an environmental comprehensive scoring function, specifically in the form of… This function represents the complex, multidimensional environment vector in which a node resides. This is mapped to a single scalar score. Visibility is one of these scores. and sensor reliability The higher the value, the higher the score; while the electromagnetic interference intensity... and average wind speed The larger the value, the lower the score. to These are weighting coefficients used to adjust the importance of each factor. With this scoring function, the nodes... They will then prioritize those with high environmental ratings from among their candidate neighbors. Above an adaptive threshold nodes As its ultimate communication neighbor. The logic behind this strategy is that cooperating with a node that has a good environment, reliable data, and stable communication is far more efficient and secure than communicating with a node that is in deep trouble.

[0054] Dynamic Link Weight The calculation determines the neighboring nodes. Information in the node The proportion of decision-making power. The formula for calculation is:

[0055] This formula takes into account four factors. The first is the distance factor, specifically the distance after terrain correction. The closer the node, the higher its weight. The second factor is communication quality; the node... Receive from node Signal-to-noise ratio The higher the value, the higher the weight. The third factor is the environmental factors of cooperative flight, specifically the wind speed difference between the two drones. The smaller the value, the more similar the flow field environment they are in, and the more stable their cooperative flight, thus the higher the weight. The fourth factor is data reliability, specifically neighboring nodes. Sensor reliability index The higher the value, the more credible the information it provides, and the higher its weight should be. These are the weighting coefficients used to balance these four factors. The calculated weights... Before being used in distributed consensus protocols, normalization is performed to ensure the convergence of the protocol.

[0056] In addition, the system continuously monitors environmental feature vectors during task execution. The change is detected. Once any critical environmental component, such as wind speed or electromagnetic interference, is detected whose change exceeds a preset safety threshold, and this change happens to activate a scene in the scene library... The drone swarm will then automatically execute the main strategy without human intervention. Switch to a pre-planned scenario and coping strategy This enables a rapid, closed-loop response to environmental changes.

[0057] Finally, the process of intelligent judgment and self-evolution in step S140 is explained in detail.

[0058] The first step is adaptive interference removal enhancement. This occurs when the drone transmits back data containing environmental feature vectors. When processing raw inspection data, the system will query an environment-degradation mapping knowledge base. This knowledge base is essentially a pre-trained classifier that can adapt to the current environment. Predict the set of most likely degradation types in the data. For example, when the input is "heavy rain, moderate wind speed", the knowledge base might output "image stripes, motion blur". Then, the system will determine the appropriate degradation type based on this set. It dynamically assembles the most suitable sequence of processing algorithms from a predefined pool of algorithms, like building blocks. It processes the original data and outputs enhanced data.

[0059] Secondly, a nonlinear confidence model is used to fuse the results of multiple algorithms. The enhanced data is then fed into various disease identification algorithms (such as CNN-based and traditional image processing-based algorithms) for parallel analysis to obtain a preliminary set of judgment results. To determine which result to ultimately accept, the system performs a process for each algorithm. They all maintain a confidence function. Its specific form is as follows:

[0060] in, It's the Sigmoid activation function, guaranteeing the output is between 0 and 1. The core of this function is that it considers not only the linear impact of a single environmental factor on the algorithm's performance, but also the influence of the weight vector... It is manifested; more importantly, it is achieved through a symmetric matrix. The quadratic term was introduced. This is used to explicitly model the nonlinear interactions between different environmental factors. For example, The off-diagonal elements in the model can capture the combined effect on the performance of a particular algorithm when "low visibility" and "strong crosswinds" occur simultaneously, an effect far exceeding the sum of the individual effects of the two. The confidence level of each algorithm under the current environment is calculated. Finally, the final assessment results Produced through a weighted voting process: The essence of this process is to give greater influence to algorithms that have historically performed better and have higher confidence levels in the current environment.

[0061] Finally, there's the model's self-evolution. If the overall confidence level of the final judgment falls below a certain threshold, it indicates the system lacks confidence in its assessment. In this case, the case, along with all relevant data, is submitted to human experts for a final decision. Human feedback information... Will be related to this environmental vector Original results of each algorithm Together, these form a valuable training sample. The system will use this new sample, through an online incremental learning algorithm, to refine the parameters in the confidence function. , , Fine-tuning and updates are performed to minimize future prediction errors. In this way, the system learns continuously through interaction with humans, and its analytical capabilities are constantly evolving and improving.

[0062] Example 3 like Figure 2 As shown, Figure 2 This is a schematic diagram illustrating the principle of the adaptive low-altitude highway inspection system for extreme environments provided by the present invention. The system 100 includes: a multi-source sensing and fusion module 110, a predictive task planning module 120, a flexible collaborative scheduling module 130, and an intelligent judgment and self-evolution module 140.

[0063] The multi-source sensing and fusion module 110 is configured to perform the data aggregation and feature extraction functions described in the aforementioned method. Specifically, this module integrates a protocol conversion gateway, enabling seamless access to heterogeneous data streams from UAV-borne sensor arrays (such as GPS, IMU, pitot tube, visual camera, and spectrum analyzer), roadside weather stations deployed along the route, and a backend geographic information system (GIS). It performs real-time parsing, format unification, and time synchronization of this data, and then extracts and calculates data including horizontal visibility using a series of predefined algorithms and models. Precipitation type and intensity Average wind speed Crosswind shear intensity Ruggedness of terrain Electromagnetic interference intensity And the crucial sensor reliability index Seven-dimensional environmental feature vectors This module acts as the "sensory organs" of the entire system, providing real-time, accurate, and comprehensive environmental situation information for all higher-level decisions.

[0064] The predictive task planning module 120, configured as the system's "decision-making brain," is responsible for generating flexible task plans. This module receives environmental feature vectors provided by module 110. This module uses this as an index to query an internally maintained extreme scenario library. This library stores a large number of typical extreme environment patterns and their triggering conditions encountered in highway inspections. After matching a possible scenario, the core function of this module is to run a hybrid programming model that combines stochastic programming and robust optimization. This model solves for a comprehensive objective function that minimizes normal cost, expected risk, and worst-case risk, ultimately outputting a structured, resilient task package. This task package not only defines the main execution strategy under normal circumstances... Also, for each extreme scenario that might be activated. A corresponding scenario-response strategy has been prepared for each. .

[0065] The elastic collaborative scheduling module 130 is configured as the "execution and control center" of the task, ensuring efficient collaboration of the UAV swarm in dynamic environments. This module distributes task packages to the UAV swarm and continuously monitors the task execution status and environmental changes. Its core function is to achieve distributed collaboration based on environmental awareness. During task execution, this module guides each UAV to perform tasks based on real-time environmental feature vectors. The module calculates a comprehensive environmental score for itself and its neighbors, and dynamically calculates link weights with other drones based on multi-dimensional information including distance, signal-to-noise ratio, wind speed difference, and neighbor sensor reliability. Based on these dynamically changing scores and weights, the module executes an environmental awareness consistency protocol, continuously optimizes the selection of communication neighbors, and dynamically reconstructs the communication network topology of the entire cluster. This maximizes the cluster's collaborative capabilities and the continuity of task execution under various interferences.

[0066] The intelligent analysis and self-evolution module 140 is configured as the system's "intelligent core," responsible for extracting high-value information from interfered data and enabling continuous improvement of the system's capabilities. This module receives raw inspection data collected by the UAV and corresponding environmental vectors. It first utilizes its internal environment-degradation mapping knowledge base, based on... The system intelligently identifies potential data degradation types and dynamically invokes the most suitable algorithm sequence for adaptive de-interference and image enhancement. Subsequently, it distributes the enhanced data to multiple parallel analysis algorithms and uses a non-linear confidence model incorporating an interaction matrix of environmental factors to calculate the credibility of each algorithm in the current environment. Finally, it generates the final highway inspection and analysis report by performing a confidence-weighted fusion of the algorithm results. When the analysis result is uncertain or human feedback is received, the module also initiates an online incremental learning mechanism to update its internal confidence model using new samples, achieving system self-evolution and performance iteration.

[0067] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive highway low-altitude patrol method for extreme environments, characterized by, The steps of this method include: Heterogeneous data from UAV airborne sensors, roadside weather stations, and geographic information systems are collected through a protocol conversion gateway and then standardized to construct an environmental feature vector that includes a sensor reliability index. Based on the environmental feature vectors, an extreme scenario library is retrieved, and a hybrid programming model that integrates stochastic programming and robust optimization is run to generate a resilient task package that includes a main execution strategy and a scenario-response strategy. The system schedules a cluster of drones to execute the elastic task package and calculates the comprehensive environmental score and dynamic link weight in real time based on environmental feature vectors during execution, so as to dynamically reconstruct the network topology of the drone cluster. The collected data is adaptively enhanced to remove interference using an environment-degradation mapping knowledge base, and the judgment results of multiple algorithms are fused using a nonlinear confidence model to generate a highway inspection and judgment report. The elastic task package is specifically implemented by solving the following objective function: in, The main decision-making variable for the execution strategy. For extreme scenarios The following are the decision variables for coping strategies; The cost coefficient vector of the main execution strategy. For the scene The cost coefficient vector of the response strategy; For the scene The estimated probability of occurrence, In order to match the scene Weights associated with highway safety events; For robust risk aversion coefficient, This represents the total number of scenes in the scene library. The set of currently active scenes; The dynamic reconstruction of the network topology of the drone swarm includes constructing a dynamic neighbor set. Specifically: Define the node environment fitness scoring function ,in, to All are weighting coefficients. For visibility, Electromagnetic interference intensity, The average wind speed, The sensor reliability index; A candidate neighbor set is determined based on physical communication constraints, and a neighbor set that satisfies these constraints is selected. The node is taken as the final neighbor, where, An adaptive threshold; Calculate the dynamic link weights Specifically, it is based on the following calculation formula: in, This is the distance after terrain correction. For signal-to-noise ratio, Due to the difference in wind speed between the two machines, For neighboring nodes Sensor reliability; These are the weighting coefficients; When updating the network topology, the dynamic link weights are normalized to execute an environment-aware distributed consensus protocol. The specific steps for scheduling the drone cluster to execute the elastic task package are as follows: Continuous monitoring of environmental feature vectors The rate of change of each component; If the instantaneous value of any critical environmental component changes more than a preset threshold relative to the time of task startup, and this change triggers a scenario in the extreme scenario library... The drone swarm will then automatically switch to execute the corresponding scenario-response strategy. .

2. The adaptive low-altitude highway inspection method for extreme environments according to claim 1, characterized in that, The construction includes an environmental feature vector containing a sensor reliability index. Its calculation formula is a 7-tuple: in, For visibility, For precipitation intensity, The average wind speed, For wind shear intensity, Due to terrain complexity, Electromagnetic interference intensity, This is the sensor reliability index.

3. The adaptive low-altitude highway inspection method for extreme environments according to claim 1, characterized in that, The specific process of the adaptive interference removal enhancement is as follows: Query Environment - Degradation Mapping Knowledge Base Based on the current environmental feature vector Predicted Degradation Type Set ; according to Dynamically assemble processing algorithm sequences from a predefined algorithm pool. It processes the raw collected data to output enhanced data.

4. The adaptive low-altitude highway inspection method for extreme environments according to claim 1, characterized in that, The judgment results obtained by fusing multiple algorithms using a nonlinear confidence model specifically include calculating the first... Confidence function of the algorithm : in, It is the Sigmoid activation function. It is a linear weight vector. It is a symmetric matrix used to characterize the interactions between different environmental factors. For bias terms; Final assessment results Obtained through weighted calculation: in, For the Kronecker delta function, For candidate results, For the algorithm The results of the analysis are represented as the original analysis results.

5. The adaptive low-altitude highway inspection method for extreme environments according to claim 4, characterized in that, It also includes the logic for determining the final assessment result, which is as follows: If the confidence level of the final assessment result is lower than the threshold, the environmental vector will be... Original assessment results and human feedback information Constructed as training samples; The parameters of the confidence function are updated using an online incremental learning algorithm. , , To minimize prediction error.

6. An adaptive low-altitude highway inspection system for extreme environments, characterized in that, include: The multi-source sensing and fusion module is configured to access heterogeneous data and extract environmental feature vectors containing sensor reliability indices. The predictive task planning module is configured to manage a library of extreme scenarios and run a hybrid planning model that combines stochastic programming and robust optimization to generate flexible task packages. The elastic collaborative scheduling module is configured to calculate the comprehensive environmental score of its candidate neighbor nodes for any node, and determine the candidate nodes whose comprehensive environmental score is higher than the preset adaptive threshold as neighbor nodes. The preset adaptive threshold is dynamically adjusted based on the current environmental feature vector; The intelligent judgment and self-evolution module is configured to drive data augmentation using an environment-degradation knowledge base and to fuse the judgment results of multiple algorithms using a nonlinear confidence model containing an interaction term matrix of environmental factors. The elastic task package is specifically implemented by solving the following objective function: in, The main decision-making variable for the execution strategy. For extreme scenarios The following are the decision variables for coping strategies; The cost coefficient vector of the main execution strategy. For the scene The cost coefficient vector of the response strategy; For the scene The estimated probability of occurrence, In order to match the scene Weights associated with highway safety events; For robust risk aversion coefficient, This represents the total number of scenes in the scene library. The set of currently active scenes; Dynamically reconstructing the network topology of a drone swarm includes building a dynamic neighbor set. Specifically: Define the node environment fitness scoring function ,in, to All are weighting coefficients. For visibility, Electromagnetic interference intensity, The average wind speed, The sensor reliability index; A candidate neighbor set is determined based on physical communication constraints, and a neighbor set that satisfies these constraints is selected. The node is taken as the final neighbor, where, An adaptive threshold; Calculate dynamic link weights Specifically, it is based on the following calculation formula: in, This is the distance after terrain correction. For signal-to-noise ratio, Due to the difference in wind speed between the two machines, For neighboring nodes Sensor reliability; These are the weighting coefficients; When updating the network topology, the dynamic link weights are normalized to execute an environment-aware distributed consensus protocol. The specific steps for scheduling a drone swarm to execute elastic task packages are as follows: Continuous monitoring of environmental feature vectors The rate of change of each component; If the instantaneous value of any critical environmental component changes more than a preset threshold relative to the time of task startup, and this change triggers a scenario in the extreme scenario library... The drone swarm will then automatically switch to execute the corresponding scenario-response strategy. .

Images