An air-ground collaborative multi-source perception mode recognition method and system for intelligent logistics

Abstract

The application relates to the technical field of intelligent logistics, and discloses an air-ground collaborative multi-source perception mode identification method and system for intelligent logistics, which comprises the following steps: constructing a logistics context information set; acquiring multi-source perception data from air-ground collaborative perception devices, performing time synchronization and format conversion processing, and obtaining processed multi-source perception data; when it is monitored that a transport vehicle has a behavior anomaly, generating a context hypothesis describing the state of the transport vehicle; calling multi-source perception data related to the context hypothesis for verification scoring; iteratively performing hypothesis generation and verification scoring until a preset iteration termination condition is met, and determining an effective context hypothesis; based on the effective context hypothesis, reconstructing corresponding scene semantics; when the scene semantics are identified as a dangerous context and the score of the effective context hypothesis reaches a preset warning threshold, issuing a warning information and providing decision support. The application solves the problem that complex abnormal modes cannot be accurately identified and timely warning cannot be performed in the prior art.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics technology, and more specifically, to an air-ground collaborative multi-source sensing pattern recognition method and system for intelligent logistics. Background Technology

[0002] In modern intelligent logistics systems, air-ground collaborative sensing and control systems are widely deployed to achieve precise management and automated scheduling of cargo transportation. This system relies on the collaborative work of various devices, such as drones, ground sensor networks, and automated guided vehicles (AGVs), to collect real-time environmental and cargo status information. However, in actual operation, complex environmental factors, such as localized electromagnetic interference, may lead to a decline in data quality and inconsistencies in some key sensing devices, making existing data processing and anomaly detection methods inadequate.

[0003] However, when localized, intermittent electromagnetic interference occurs, RFID readers may experience tag reading failures, delays, or errors. The wireless data link between the AGV and the central control system may also experience packet loss and transmission delays, leading to slight jumps or lags in the real-time position and status data reported by the AGV. In this situation, the system's original mechanism of relying on high-precision numerical positioning data for cross-validation is challenged. Inconsistencies in the quality of multi-source heterogeneous sensing data make it difficult for the system to accurately determine the source of data conflicts, resulting in uncertainty in the integrated AGV's precise position information.

[0004] To compensate for the lack of positioning data, the system attempted to allocate more weight to optical vision information. However, the accuracy of drone optical cameras in determining the precise position of AGVs at long distances is limited, and the coverage of ground-based fixed cameras is also limited. More importantly, the system's original recognition patterns mainly rely on high-precision numerical positioning data, which is insufficient for accurately extracting sub-meter-level positional deviations of AGVs from continuous video streams. This results in inconsistent effective redundancy across different data sources for specific tasks.

[0005] This fluctuation in the quality of multi-source data and insufficient recognition capabilities make it difficult for the system to identify complex, progressively evolving anomaly patterns caused by multiple factors. For example, when an AGV deviates slightly and continuously from its preset path due to electromagnetic interference and heads towards an unreported open maintenance area, the fragmented information received by the system (AGV reports an object ahead, drones and ground cameras show the AGV moving but fail to identify the slight deviation, and the system lacks maintenance area status updates) cannot be effectively integrated to form meaningful scene semantics. This prevents the system from accurately identifying potential dangerous situations in the early stages, ultimately forcing it to urgently stop the AGV and issue the highest-level alarm, causing task delays and exposing the system's deep-seated deficiencies in dealing with dynamic environmental changes, fluctuations in the quality of multi-source data, and the recognition of complex, progressive anomaly patterns.

[0006] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this application provides a method and system for air-ground collaborative multi-source sensing pattern recognition for intelligent logistics. This addresses the deep-seated deficiencies of existing intelligent logistics systems in recognizing complex environmental factors, fluctuations in the quality of multi-source data, and complex, progressive anomaly patterns, making it difficult to accurately identify potential hazardous situations in the early stages.

[0008] In a first aspect, this application discloses an air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics, comprising the following steps: Acquire and dynamically update status information describing the logistics environment to construct a set of logistics contextual information; The system acquires multi-source sensing data from air-ground collaborative sensing devices in real time, performs time synchronization and format conversion on the multi-source sensing data, and obtains processed multi-source sensing data. When abnormal behavior of a transport vehicle is detected, at least one scenario hypothesis describing the current state of the transport vehicle is generated based on the logistics scenario information set and the preset scenario hypothesis template library. Retrieve multi-source sensory data related to the contextual assumptions for verification and scoring; Iteratively perform hypothesis generation and validation scoring until the score of the scenario hypothesis meets the preset iteration termination condition, and the scenario hypothesis that meets the preset iteration termination condition is determined as a valid scenario hypothesis. Based on the effective context hypothesis, the corresponding scenario semantics are reconstructed. When the scenario semantics are identified as a dangerous situation and the score of the effective context hypothesis reaches the preset warning threshold, a warning message is issued and decision support is provided.

[0009] Through this technical solution, this application can effectively integrate multi-source sensing data from air and ground collaboration, and through the generation, verification and iteration of scenario hypotheses, it can identify complex abnormal patterns, issue early warnings and provide decision support before dangerous situations occur, thereby solving the problem of difficulty in accurately identifying complex abnormal patterns and providing timely early warnings in existing technologies.

[0010] Furthermore, in some implementation schemes, the logistics context information set includes environmental facility information, operation planning information, equipment operation information, and environmental status information.

[0011] Through this technical solution, this application constructs a logistics context information set containing multi-dimensional information, providing more comprehensive and accurate background knowledge for the generation of context hypotheses, thereby improving the accuracy and recognition efficiency of context hypotheses.

[0012] Furthermore, in some implementation schemes, the multi-source sensing data includes real-time video streams, vehicle sensor data, and RFID tag data; detecting abnormal behavior of the transport vehicle includes detecting that the lateral position deviation between the positioning sensor data in the vehicle sensor data and the preset path continuously exceeds a first preset threshold within a preset time window, or that the obstacle avoidance sensor data in the vehicle sensor data indicates the presence of environmental facilities not registered in the environmental facility information.

[0013] By introducing various types of perception data and combining them with specific abnormal behavior judgment criteria, this application can more sensitively and accurately capture abnormal behavior of transportation vehicles, providing timely and reliable triggering conditions for subsequent scenario hypothesis generation.

[0014] Based on the above, this application further proposes that the steps for retrieving multi-source sensory data related to the contextual hypothesis for verification scoring include: Based on the semantic content of the context hypothesis, one or more target perception data related to the verification of the context hypothesis are selected from the multi-source perception data; Analyze and process target perception data to obtain verification evidence that supports or refutes situational hypotheses; The effectiveness contribution of the verification evidence is calculated by combining the reliability assessment weights of the target perception data and the semantic matching degree between the verification evidence and the contextual hypothesis. Based on the validity contribution of all valid evidence, a score for the situational hypothesis is generated through normalization.

[0015] Preferably, in some implementations, the reliability assessment weights of the target perception data are obtained through the following steps: At least one data quality indicator of the target perception data is monitored in real time. When the real-time monitoring value of the data quality indicator deviates from the preset normal range threshold within a preset time window, the target perception data is determined to be in a state of reduced reliability. Based on the degree of deviation between the real-time monitoring value of the data quality indicator and the preset normal range threshold, the reliability assessment weight of the target perception data is calculated.

[0016] In one implementation, the degree of semantic matching between the verification evidence and the contextual hypothesis is obtained through the following steps: Extract semantic or quantitative features from the verification evidence to support or refute the situational hypothesis, evaluate the semantic or logical consistency and support between the extracted semantic or quantitative features and the hypothesis features corresponding to the situational hypothesis, and calculate the semantic matching degree.

[0017] Based on the above, this application further proposes that the steps of determining a scenario hypothesis as a valid scenario hypothesis include: the scoring of the scenario hypothesis satisfies a preset iteration termination condition, and the determination of the scenario hypothesis that satisfies the preset iteration termination condition as a valid scenario hypothesis. If there exists a first scenario hypothesis, which is the only scenario hypothesis with the highest score among all scenario hypotheses and the difference between its score and the score of the second highest-scoring scenario hypothesis satisfies a preset score difference condition, then the first scenario hypothesis is determined to satisfy the preset iteration termination condition and is identified as a valid scenario hypothesis. If there are multiple identical scenario hypotheses with the highest scores, a valid scenario hypothesis is determined from the multiple scenario hypotheses according to a predetermined priority rule.

[0018] As a technical improvement, this application also proposes that, based on the effective context assumption, the steps for reconstructing the corresponding scene semantics include: By semantically associating and integrating the current state of the transport vehicle described by the effective scenario hypothesis, the multi-source perception data related to the scenario hypothesis, and the set of logistics scenario information, a scenario semantic description with complete causal logic is reconstructed.

[0019] As an enhancement to the functionality, this application also proposes steps for issuing early warning information and providing decision support, including: Generate early warning messages that include descriptions of hazardous situations, relevant multi-source sensing data, and suggested response measures; The warning message is displayed to the operator through a visual interface, and an alarm is issued through an audible and visual alarm.

[0020] Secondly, this application also discloses an air-ground collaborative multi-source sensing pattern recognition system for intelligent logistics, used to perform the aforementioned method, the system comprising: The logistics context information set construction module is used to acquire and dynamically update the status information describing the logistics environment in order to construct the logistics context information set. The multi-source sensing data acquisition module is used to acquire multi-source sensing data from air-ground collaborative sensing devices in real time, and to perform time synchronization and format conversion on the multi-source sensing data to obtain processed multi-source sensing data. The scenario hypothesis generation module is used to generate at least one scenario hypothesis describing the current state of the transport vehicle based on the logistics scenario information set and the preset scenario hypothesis template library when abnormal behavior of the transport vehicle is detected. The verification and scoring module is used to retrieve multi-source perception data related to the contextual assumptions for verification and scoring. The iteration module is used to iteratively perform hypothesis generation and validation scoring until the score of the scenario hypothesis meets the preset iteration termination condition, and the scenario hypothesis that meets the preset iteration termination condition is determined as a valid scenario hypothesis. The early warning module is used to reconstruct the corresponding scene semantics based on the effective scenario hypothesis. When the scene semantics are identified as a dangerous situation and the score of the effective scenario hypothesis reaches the preset early warning threshold, an early warning message is issued and decision support is provided.

[0021] This application provides a system implementation method that modularizes the various functions in the aforementioned method, thereby improving the system's feasibility, maintainability, and scalability, and providing a complete solution for air-ground collaborative multi-source perception pattern recognition in intelligent logistics.

[0022] In summary, this application provides a method and system for air-ground collaborative multi-source sensing pattern recognition for intelligent logistics. The method acquires and dynamically updates a logistics context information set, and acquires and processes multi-source sensing data in real time. When abnormal behavior of a transport vehicle is detected, a context hypothesis is generated based on the logistics context information set and a pre-set context hypothesis template library. Subsequently, relevant multi-source sensing data is retrieved for verification and scoring, and hypothesis generation and verification scoring are iteratively executed until the score of the context hypothesis meets a pre-set iteration termination condition, thus determining a valid context hypothesis. Finally, the scene semantics are reconstructed based on the valid context hypothesis, and when a dangerous situation is identified and the score reaches a warning threshold, a warning message is issued and decision support is provided. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating a multi-source sensing pattern recognition method for air-ground collaborative sensing in intelligent logistics, provided as an embodiment of this application.

[0024] Figure 2 This is a schematic diagram of the structure of an air-ground collaborative multi-source sensing pattern recognition system for intelligent logistics, provided in an embodiment of this application.

[0025] Labeling Explanation: 210, Logistics Context Information Set Construction Module; 220, Multi-Source Sensing Data Acquisition Module; 230, Context Hypothesis Generation Module; 240, Verification and Scoring Module; 250, Iteration Module; 260, Early Warning Module. Detailed Implementation

[0026] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0027] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0028] In modern intelligent logistics systems, while air-ground collaborative sensing and control systems are widely deployed, complex environmental factors, such as localized electromagnetic interference, can lead to data quality degradation and inconsistencies in some key sensing devices during actual operation. This makes it difficult for existing data processing and anomaly identification methods to cope with the inconsistencies in data quality among multi-source heterogeneous sensing data, thus hindering the accurate determination of the source of data conflicts and resulting in uncertainty in the integrated precise location information. This fluctuation in multi-source data quality and insufficient identification capabilities make it difficult for the system to identify complex anomaly patterns that develop gradually due to multiple factors, ultimately potentially causing the system to fail to accurately identify potential dangerous situations in the early stages, resulting in task delays.

[0029] In this regard, firstly, referring to Figure 1 This application proposes a multi-source sensing pattern recognition method for air-ground collaborative sensing in intelligent logistics, including: S1. Obtain and dynamically update the status information describing the logistics environment to construct a set of logistics context information; S2. Acquire multi-source sensing data from air-ground cooperative sensing devices in real time, and perform time synchronization and format conversion processing on the multi-source sensing data to obtain processed multi-source sensing data. S3. When abnormal behavior of the transport vehicle is detected, at least one scenario hypothesis describing the current state of the transport vehicle is generated based on the logistics scenario information set and the preset scenario hypothesis template library. S4. Retrieve multi-source sensory data related to the contextual assumptions for verification and scoring; S5. Iteratively perform hypothesis generation and verification scoring until the score of the scenario hypothesis meets the preset iteration termination condition, and determine the scenario hypothesis that meets the preset iteration termination condition as a valid scenario hypothesis. S6. Based on the effective scenario hypothesis, reconstruct the corresponding scenario semantics. When the scenario semantics are identified as a dangerous scenario and the score of the effective scenario hypothesis reaches the preset warning threshold, issue a warning message and provide decision support.

[0030] The logistics contextual information set refers to comprehensive state information describing the logistics environment, which may include environmental facility information, operation planning information, equipment operation information, and environmental status information, etc., to provide comprehensive background knowledge for the generation of contextual hypotheses.

[0031] Multi-source sensing data refers to data from different types of sensing devices, such as video streams, vehicle sensor data, and RFID tag data. After time synchronization and format conversion, this data is used to support the verification of scenario hypotheses.

[0032] Transportation vehicles refer to equipment that performs transportation tasks in a logistics environment, such as automated guided vehicles (AGVs) and drones.

[0033] Contextual assumptions are speculative descriptions of the current state of a transport vehicle. They are generated based on a set of logistics contextual information and a contextual assumption template library, and need to be verified through multi-source sensing data.

[0034] A valid scenario hypothesis is a scenario hypothesis that, after iterative verification, satisfies the preset iteration termination condition and is considered to be the hypothesis that most accurately describes the state of the transport vehicle.

[0035] Scene semantics refers to a scene description reconstructed based on valid scenario assumptions, possessing a complete causal logical relationship, used to identify dangerous situations.

[0036] The warning threshold refers to the scoring criteria used to determine whether to issue a warning message.

[0037] The method proposed in this application first acquires and dynamically updates state information describing the logistics environment to construct a logistics context information set. This logistics context information set can be viewed as a dynamic knowledge base, whose content is updated in real time as the logistics environment changes. For example, environmental facility information (such as warehouse layout and shelf location), operational planning information (such as transportation routes and scheduling plans), equipment operation information (such as AGV power and operating status), and environmental state information (such as temperature, humidity, and traffic conditions) can be acquired through various means such as sensor networks, database queries, and manual input. This information is integrated and structured to form a comprehensive logistics context information set, providing a foundation for subsequent context hypothesis generation.

[0038] Simultaneously, multi-source sensing data from air-to-ground cooperative sensing devices is acquired in real time, and time synchronization and format conversion processing are performed on the multi-source sensing data to obtain processed multi-source sensing data. Air-to-ground cooperative sensing devices can include, but are not limited to, aerial drones, fixed ground cameras, vehicle-mounted sensors (such as positioning sensors and obstacle avoidance sensors), and RFID readers. These devices generate heterogeneous data, such as video streams, numerical sensor data, and discrete RFID tag data. To ensure that this data can be effectively utilized, time synchronization processing is required, for example, by using a unified timestamp or timestamp alignment algorithm to ensure that data from different data sources remain consistent in the time dimension. Furthermore, format conversion processing is also necessary to unify data of different formats into a standard format that can be processed by the system, such as converting video streams into image sequences and converting raw sensor data into structured data.

[0039] When an abnormal behavior of a transport vehicle is detected, at least one scenario hypothesis describing the current state of the transport vehicle is generated based on the logistics context information set and a pre-set scenario hypothesis template library. Detection of abnormal behavior can be achieved in various ways. For example, real-time analysis of onboard sensor data can be performed; if the lateral position deviation between the positioning sensor data and the preset path continuously exceeds a first preset threshold within a preset time window, it can be determined that the transport vehicle is exhibiting abnormal behavior. Alternatively, if obstacle avoidance sensor data indicates the presence of environmental facilities not registered in the environmental facility information, it can also be considered an abnormal behavior. Once an abnormal behavior is detected, the system uses the constructed logistics context information set and the pre-set scenario hypothesis template library to generate a scenario hypothesis. The scenario hypothesis template library pre-stores various possible abnormal situation patterns, such as AGV deviation from the path, AGV entering a restricted area, and AGV collision. The system selects or combines from the template library to generate one or more specific scenario hypotheses based on the currently detected abnormality type and logistics context information.

[0040] Subsequently, multi-source perception data related to the scenario hypothesis is retrieved for verification scoring. For each generated scenario hypothesis, the system filters one or more target perception data points associated with verifying the hypothesis from the processed multi-source perception data, based on their semantic content. For example, if the scenario hypothesis is that the AGV deviates from a preset path, it may be necessary to retrieve the AGV's positioning sensor data, AGV video streams captured by drones, and video streams of the AGV area captured by ground cameras. This target perception data is analyzed to obtain verification evidence supporting or refuting the scenario hypothesis. For example, by analyzing positioning sensor data, the deviation of the AGV's actual position from the preset path can be calculated; by analyzing video streams, the AGV's actual travel trajectory can be identified. The validity contribution of the verification evidence is calculated by combining the reliability assessment weights of the target perception data and the semantic matching degree between the verification evidence and the scenario hypothesis. For example, if the quality of a sensor data point is low, its reliability assessment weight will be reduced accordingly. Based on the validity contributions of all verification evidence, a score for the scenario hypothesis is generated through normalization processing.

[0041] This application iteratively performs hypothesis generation and validation scoring until the scenario hypothesis's score meets a preset iteration termination condition. Scenario hypotheses meeting the preset iteration termination condition are then identified as valid scenario hypotheses. The iterative process means that the system may adjust or generate new scenario hypotheses based on the current validation scoring results, and then perform validation scoring again to gradually converge to the most accurate scenario description. The iteration termination condition can include various cases. For example, if there exists a first scenario hypothesis that is the only scenario hypothesis with the highest score among all scenario hypotheses, and the difference between its score and the score of the second-highest-scoring scenario hypothesis meets a preset score difference condition, then this first scenario hypothesis is determined to meet the preset iteration termination condition and is identified as a valid scenario hypothesis. This indicates that the system has found an explanation that is significantly better than the other hypotheses. If there are multiple scenario hypotheses with the same highest score, a valid scenario hypothesis can be determined from the multiple scenario hypotheses according to predetermined priority rules (e.g., based on the severity of the hypothesis, historical frequency, etc.).

[0042] Finally, based on the valid scenario hypothesis, the corresponding scenario semantics are reconstructed. When the scenario semantics are identified as a dangerous situation and the score of the valid scenario hypothesis reaches a preset warning threshold, a warning message is issued and decision support is provided. The steps for reconstructing the scenario semantics include semantically associating and integrating the current state of the transport vehicle described by the valid scenario hypothesis, multi-source perception data related to the scenario hypothesis, and a set of logistics scenario information to reconstruct a scenario semantic description with complete causal logic. For example, if the valid scenario hypothesis is that the AGV deviates from the preset path due to a navigation system malfunction and heads towards an unregistered maintenance area, the system will combine the AGV's real-time location data, the environmental facility information of the maintenance area, and the operating status of the navigation system to reconstruct the scenario semantics that the AGV is entering the maintenance area abnormally and there is a risk of collision. When this scenario semantics are identified as a dangerous situation (e.g., by matching with a preset dangerous situation pattern library) and the score of the valid scenario hypothesis reaches a preset warning threshold, the system will immediately issue a warning message. The warning message may include a description of the dangerous situation, relevant multi-source perception data (such as video screenshots, sensor curves), and suggested handling measures (such as immediately stopping the AGV and dispatching personnel for inspection). These early warning messages can be displayed to operators through a visual interface and alerted by sound and light alarms, thus providing timely decision support.

[0043] The proposed air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics effectively addresses the challenge of traditional methods in accurately identifying complex anomaly patterns when faced with fluctuations in the quality and inconsistencies of multi-source heterogeneous sensing data. This is achieved by introducing a contextual hypothesis generation, iterative verification, and scoring mechanism. Traditional systems often rely on a single data source or simple multi-source data fusion. When some data sources malfunction, their recognition capabilities are significantly reduced, leading to false alarms or missed alarms. For example, when local electromagnetic interference causes RFID readers to fail to read tags, experience delays, or make errors, or when there are packet loss and transmission delays in the wireless data link between the AGV and the central control system, traditional systems struggle to accurately determine the source of data conflicts, resulting in uncertainty in the integrated AGV's precise location information.

[0044] For example, when an AGV deviates slightly and continuously from its preset path due to electromagnetic interference and heads towards an unregistered open maintenance area, the fragmented information received by traditional systems (AGV reports an object ahead, drones and ground cameras show the AGV moving but fail to recognize the slight deviation, and the system lacks maintenance area status updates) cannot be effectively integrated to form meaningful scene semantics. The method in this application, however, can generate scenario hypotheses such as the AGV possibly deviating from its preset path due to navigation system malfunction or the AGV possibly entering an unregistered area. By retrieving and scoring various data sources, including AGV positioning data, drone video, and ground camera video, and combining this with data reliability assessment weights, the system can more accurately determine which scenario hypothesis best reflects the actual situation. Through iterative verification, a valid scenario hypothesis is finally determined, and based on this, scene semantics with complete causal logic is reconstructed, such as the AGV entering the maintenance area abnormally, posing a collision risk. When this scene semantics is identified as a dangerous situation and the score reaches the warning threshold, the system can promptly issue a warning and provide decision support, thus avoiding the shortcomings of traditional systems that can only urgently stop the AGV and issue the highest-level alarm, causing task delays. Therefore, this application significantly improves the ability of intelligent logistics systems to identify abnormal patterns and improves their early warning efficiency in complex and dynamic environments.

[0045] Specifically, the aforementioned logistics context information set may include environmental facility information, operation planning information, equipment operation information, and environmental status information.

[0046] The logistics contextual information set aims to comprehensively and dynamically describe the current state and potential risks of the intelligent logistics environment. Specifically, environmental facility information refers to data on fixed infrastructure within the logistics park, such as warehouse layout, shelf locations, charging pile distribution, road networks, and restricted areas—both geospatial and attribute information. This information provides the basic physical environment background for the operation of transport vehicles. Operational planning information refers to the planning and scheduling data of logistics tasks, such as the preset routes, task lists, timetables, cargo types, and priorities of transport vehicles. This information defines the expected behavior patterns of transport vehicles. Equipment operation information refers to the real-time status data of transport vehicles and their onboard equipment, such as the speed, location, battery level, load, sensor operating status, and fault codes of the transport vehicles. This information reflects the actual operating status of the transport vehicles. Environmental status information refers to dynamic or external factor data within the logistics environment, such as weather conditions, traffic flow, personnel activity, and emergencies (such as road construction and temporary obstacles). This information provides the external environmental context affecting the operation of transport vehicles. By integrating the above types of information, a multi-dimensional and high-precision set of logistics context information can be constructed, providing comprehensive data support for the subsequent generation and verification of context hypotheses.

[0047] The aforementioned technical solution clarifies the specific components of the logistics context information set, making the constructed set more comprehensive, detailed, and structured. This effectively improves the accuracy and completeness of the description of the logistics environment, providing richer and more reliable contextual information for subsequent contextual hypothesis generation. This not only helps improve the quality of generated contextual hypotheses but also provides stronger data support for subsequent verification and scoring, thereby enhancing the robustness and accuracy of the entire pattern recognition method and reducing the risk of false positives or false negatives.

[0048] Specifically, in some of the above embodiments, in order to perceive the logistics environment more comprehensively and accurately and identify abnormal behavior of transport vehicles, this application further defines the content of multi-source perception data and the monitoring method of abnormal behavior.

[0049] Multi-source sensing data includes real-time video streams, vehicle sensor data, and RFID tag data; Detecting abnormal behavior of a transport vehicle includes detecting that the lateral position deviation between the positioning sensor data in the vehicle sensor data and the preset path continuously exceeds a first preset threshold within a preset time window, or that the obstacle avoidance sensor data in the vehicle sensor data indicates the presence of environmental facilities that are not registered in the environmental facility information.

[0050] Multi-source sensing data refers to data acquired from various types of sensing devices, aiming to provide richer and more comprehensive environmental information. Specifically, real-time video streams can provide visual information for identifying objects, human behavior, or environmental changes; vehicle-mounted sensor data, including but not limited to positioning sensor data, obstacle avoidance sensor data, and speed sensor data, is used to obtain local information about the operating status of the transport vehicle and its surrounding environment; RFID tag data can be used to identify the identity and location of specific items or vehicles. These different types of data, after time synchronization and format conversion, can be integrated and utilized to form a comprehensive view of the logistics environment and the status of transport vehicles.

[0051] Furthermore, the specific methods for detecting abnormal behavior in transport vehicles have been refined. One anomaly is that, by monitoring positioning sensor data from the vehicle's onboard sensors, it is found that the lateral position deviation of the transport vehicle continuously exceeds a first preset threshold within a preset time window. This indicates that the transport vehicle may have deviated from the preset path, posing a risk of abnormal driving. Another anomaly is that obstacle avoidance sensor data from the vehicle's onboard sensors indicates the presence of environmental facilities not registered in the aforementioned environmental facility information. This means that the transport vehicle may have encountered unknown obstacles or entered unplanned areas, which also constitutes potentially dangerous behavior. Through these two specific monitoring methods, potential risky behaviors of transport vehicles can be identified more accurately.

[0052] Furthermore, by specifically defining the monitoring mechanism for abnormal vehicle behavior—namely, judging path deviation based on positioning sensor data and judging unknown obstacles based on obstacle avoidance sensor data—the system can directly and quantitatively identify potential dangerous behaviors at the data level. When positioning sensor data indicates that the vehicle is continuously deviating from the preset path, the system can promptly detect the abnormality in its trajectory; when obstacle avoidance sensor data detects unregistered environmental facilities, the system can immediately identify potential collision risks. This clear anomaly judgment logic provides clear triggering conditions and data basis for subsequent scenario hypothesis generation and verification.

[0053] Through the aforementioned technical solutions, this application enables a more comprehensive and refined perception of the logistics environment and the status of transport vehicles, thereby improving the accuracy and timeliness of abnormal behavior identification. Specifically, the introduction of multi-source perception data allows the system to acquire information from multiple dimensions such as vision, motion, and recognition, effectively compensating for the limitations of a single data source and enhancing the understanding of complex logistics scenarios. Simultaneously, by quantitatively defining and monitoring abnormal transport vehicle behavior, the system can detect potential risks earlier and more accurately, such as path deviations or unknown obstacles, thus providing a solid data foundation and judgment basis for subsequent early warning and decision support, significantly improving the safety and reliability of the intelligent logistics system.

[0054] Specifically, the steps for retrieving and validating multi-source perception data related to the contextual assumptions include the following:

[0055] Based on the semantic content of the context hypothesis, one or more target perception data related to the verification of the context hypothesis are selected from the multi-source perception data; Analyze and process target perception data to obtain verification evidence that supports or refutes situational hypotheses; The effectiveness contribution of the verification evidence is calculated by combining the reliability assessment weights of the target perception data and the semantic matching degree between the verification evidence and the contextual hypothesis. Based on the validity contribution of all valid evidence, a score for the situational hypothesis is generated through normalization.

[0056] Specifically, selecting one or more target perception data related to verifying the scenario hypothesis from multi-source perception data based on the semantic content of the scenario hypothesis means identifying and extracting data segments or data streams directly related to the scenario hypothesis from real-time acquired air-ground coordinated multi-source perception data, based on the specific meaning of the current state of the transport vehicle described by the scenario hypothesis. For example, if the scenario hypothesis describes the transport vehicle deviating from a preset path, it is necessary to select vehicle-mounted sensor data related to the transport vehicle's position and trajectory, as well as video stream data that may contain path information.

[0057] Furthermore, analyzing and processing the target perception data to obtain verification evidence that supports or refutes the situational hypothesis refers to conducting in-depth analysis of the selected target perception data to extract specific information that can directly prove or refute the situational hypothesis. For example, for selected positioning sensor data, its actual deviation from the preset path can be calculated; for video stream data, image recognition technology can be used to detect the actual driving status of the transport vehicle or the surrounding environment. These analytical results constitute verification evidence.

[0058] Furthermore, by combining the reliability assessment weights of the target perception data with the semantic matching degree between the verification evidence and the contextual hypothesis, the effectiveness contribution of the verification evidence is calculated. This means that when evaluating verification evidence, not only the content of the evidence itself must be considered, but also the reliability of its source and its fit with the contextual hypothesis. The reliability assessment weights of the target perception data reflect the quality and credibility of the data source; for example, sensor malfunctions or signal interference may lead to a decrease in weight. The semantic matching degree between the verification evidence and the contextual hypothesis measures the semantic or logical consistency of the evidence with the contextual hypothesis. By combining these two factors, the actual contribution of each piece of verification evidence to the verification of the contextual hypothesis can be quantified.

[0059] Therefore, generating a score for the situational hypothesis based on the validity contributions of all validating evidence involves summarizing the validity contributions of all calculated validating evidence and converting them into a unified score using a normalization method. This score can intuitively reflect the degree to which the situational hypothesis is supported by multi-source perceptual data, thus providing a quantitative basis for subsequent iterative decisions. Normalization ensures a fair comparison of the contributions of different pieces of evidence and keeps the final score within a pre-defined effective range.

[0060] The above technical solution enables refined management of the verification and scoring process for scenario hypotheses, significantly improving the accuracy and reliability of the scoring. This method not only effectively utilizes multi-source sensory data but also reduces the risk of misjudgment due to data quality issues or insufficient evidence relevance by considering data reliability and semantic matching. Therefore, the pattern recognition method of this application can more accurately determine the current state of transport vehicles when identifying abnormal behavior, thus providing more reliable decision support and early warning information for intelligent logistics systems.

[0061] Specifically, the reliability assessment weights of the aforementioned target perception data are obtained through the following steps.

[0062] The reliability assessment weight of the target perception data is obtained through the following steps: real-time monitoring of at least one data quality indicator of the target perception data; when the real-time monitoring value of the data quality indicator continuously deviates from the preset normal range threshold within a preset time window, the target perception data is determined to be in a state of reduced reliability; and the reliability assessment weight of the target perception data is calculated based on the degree of deviation between the real-time monitoring value of the data quality indicator and the preset normal range threshold.

[0063] Data quality indicators can include, but are not limited to, data integrity, timeliness, accuracy, consistency, or availability. For example, for real-time video streams, data quality indicators may include frame rate, resolution, sharpness, or the presence of occlusion; for vehicle sensor data, data quality indicators may include the stability of sensor readings, noise level, or data transmission latency. A preset time window defines the duration of continuous deviation to avoid misjudgments caused by instantaneous fluctuations. A preset normal range threshold defines the expected normal fluctuation range of the data quality indicators. When the real-time monitoring value of a data quality indicator continuously exceeds the preset normal range threshold within the preset time window—for example, if the frame rate of the video stream is lower than normal for an extended period, or if the positioning sensor data exhibits continuous drift—the target perception data is considered to be in a state of reduced reliability. Furthermore, the degree of deviation can be quantified based on the absolute difference, relative percentage deviation, or a specific functional mapping relationship between the real-time monitoring value and the normal range threshold. For example, a larger degree of deviation indicates lower data reliability. Therefore, the reliability assessment weight can be calculated as a value inversely proportional to the degree of deviation, or mapped through a predefined weighting function, such that the lower the data quality, the lower the weight.

[0064] Through the above technical solution, this application can effectively solve the problem that inaccurate verification results may occur due to the deterioration of the quality of some sensing data during the fusion and verification of multi-source sensing data. By introducing dynamic reliability assessment weights, the sensitivity of the verification scoring process to data quality is ensured, enabling the system to more robustly handle various complex and uncertain logistics environments, improving the accuracy and reliability of situational hypothesis verification, and thus enhancing the accuracy of the entire pattern recognition method and the effectiveness of decision support.

[0065] Specifically, in the above-mentioned air-ground collaborative multi-source perception pattern recognition method for intelligent logistics, the method for obtaining the semantic matching degree between verification evidence and contextual hypothesis can be further refined.

[0066] The semantic match between the above verification evidence and the contextual hypothesis is obtained through the following steps: Extract semantic or quantitative features from the verification evidence to support or refute the situational hypothesis, evaluate the semantic or logical consistency and support between the extracted semantic or quantitative features and the hypothesis features corresponding to the situational hypothesis, and calculate the semantic matching degree.

[0067] Specifically, verification evidence refers to information obtained through analysis and processing of target perception data that supports or refutes the contextual hypothesis. The contextual hypothesis describes the current state of the transport vehicle. To assess the correlation between verification evidence and the contextual hypothesis, key information needs to be extracted from the verification evidence. This key information can be semantic features or quantitative features. Semantic features refer to words, phrases, or concepts with specific meanings extracted from unstructured data (such as video descriptions, voice recordings, and text reports) using techniques such as natural language processing and text analysis. For example, if the contextual hypothesis is that the transport vehicle deviates from the preset path, then descriptions in the verification evidence such as the vehicle making a sharp left turn or deviating from the lane line would be considered semantic features. Quantitative features refer to quantifiable values ​​or indicators extracted from structured data (such as sensor data and positioning data). For example, if the contextual hypothesis is that the transport vehicle is speeding, then the current speed of 80 km / h in the verification evidence would be considered a quantitative feature.

[0068] Hypothesis features refer to the specific characteristics inherent in the contextual hypothesis itself, used to describe its content. For example, the hypothesis features of the contextual hypothesis "the transport vehicle deviates from the preset path" might include semantic elements such as deviation, path, and transport vehicle, or quantitative conditions such as a lateral positional deviation greater than X meters. After extracting the semantic or quantitative features of the verification evidence, it is necessary to compare and evaluate them with the hypothesis features corresponding to the contextual hypothesis. This evaluation aims to determine the semantic or logical consistency and support level between the two. Consistency refers to the degree of agreement in meaning, while support level refers to the strength of support the verification evidence provides for the contextual hypothesis. For example, it can be evaluated by calculating semantic similarity (e.g., using word vector models, ontology matching), logical reasoning (e.g., judging whether conditions are met), or rule matching. Thus, by comprehensively evaluating the consistency and support level of these features, the semantic matching degree can be calculated. This matching degree is usually a numerical value representing the strength of support the verification evidence provides for the contextual hypothesis; a higher value indicates a higher matching degree and stronger support.

[0069] The above technical solution enables a refined assessment of the correlation between verification evidence and situational hypotheses, improving the accuracy and objectivity of verification scoring. Specifically, by extracting semantic and quantitative features and evaluating their consistency and support for hypothesis features, the system can more accurately determine the supporting role of verification evidence for situational hypotheses, thereby enhancing the reliability of situational hypothesis verification and providing stronger data support for ultimately determining valid situational hypotheses and identifying dangerous situations.

[0070] When performing pattern recognition on abnormal behavior of transport vehicles, the basic approach simply relies on pre-defined iteration termination conditions to determine valid context hypotheses during the iterative execution of hypothesis generation and verification scoring. This can lead to challenges in practical applications. Specifically, when the scores of multiple context hypotheses are very close, or when no single context hypothesis scores significantly higher than the others, the system struggles to explicitly select the most accurate and reliable valid context hypothesis. This uncertainty may affect the accuracy of subsequent scene semantic reconstruction and the timeliness of hazardous situation identification.

[0071] In response, this application further proposes a step for determining the scenario hypothesis whose score satisfies a preset iteration termination condition, and for identifying scenario hypotheses that satisfy the preset iteration termination condition as valid scenario hypotheses, including: If there exists a first scenario hypothesis, which is the only scenario hypothesis with the highest score among all scenario hypotheses and the difference between its score and the score of the second highest-scoring scenario hypothesis satisfies a preset score difference condition, then the first scenario hypothesis is determined to satisfy the preset iteration termination condition and is identified as a valid scenario hypothesis. If there are multiple identical scenario hypotheses with the highest scores, a valid scenario hypothesis is determined from the multiple scenario hypotheses according to a predetermined priority rule.

[0072] Specifically, the first scenario hypothesis refers to the single scenario hypothesis identified as having the highest score after verification and scoring in the current iteration round. The unique highest-scoring scenario hypothesis means that among all generated candidate scenario hypotheses, only one hypothesis has the highest score, and that score is higher than all other hypotheses. The second-highest-scoring scenario hypothesis is the one whose score is second only to the highest score among all scenario hypotheses. The preset score difference condition is a pre-defined numerical threshold used to measure the minimum difference between the highest and second-highest scores. This condition aims to ensure that the selected highest-scoring scenario hypothesis has sufficient confidence advantage, avoiding misjudgments due to small score differences. For example, this score difference condition can be adjusted according to the robustness requirements of the decision-making in the actual application scenario to balance the sensitivity and accuracy of identification.

[0073] In cases where multiple identical scenario hypotheses receive the highest scores, a predefined priority rule needs to be introduced. This priority rule can be set based on various factors. For example, it can prioritize hypotheses with more specific and detailed semantic content; it can prioritize hypotheses related to more types of multi-source sensory data; or it can assign different priority weights to different types of scenario hypotheses based on pre-defined business logic or domain knowledge. For instance, in a logistics scenario, hypotheses involving safety risks may be given higher priority. In this way, even when scores are the same, the system can determine a unique and valid scenario hypothesis, thereby ensuring the certainty and continuity of decision-making.

[0074] Through the above technical solutions, this application can significantly improve the accuracy and reliability of determining valid context hypotheses. By introducing a score difference condition, the system can avoid making hasty judgments when the differences in context hypothesis scores are not significant, thus ensuring that only hypotheses with sufficient advantage are confirmed as valid. This effectively reduces the risk of misjudgment due to similar scores. Simultaneously, for situations where multiple context hypotheses have the same score, a predetermined priority rule is used for decision-making, eliminating ambiguity in selection and ensuring that a clear and valid context hypothesis can be determined even in complex situations. This not only improves the robustness of the pattern recognition process but also provides a more solid and reliable foundation for subsequent scene semantic reconstruction and early warning decisions, thereby improving the efficiency and accuracy of the entire intelligent logistics system in identifying and responding to abnormal behavior.

[0075] In some preferred embodiments, it is assumed that when performing pattern recognition on abnormal behavior of the transport vehicle, the system iteratively generates multiple situational hypotheses and verifies and scores them. For example, in a certain iteration, the system obtains the following situational hypotheses and their scores: Scenario Assumption A (e.g., the vehicle deviates from the preset route, possibly due to a navigation system malfunction), Score: 0.92 Scenario assumption B (e.g., the transport vehicle deviates from the preset route, possibly due to driver fatigue), score: 0.88 Scenario assumption C (e.g., the transport vehicle deviates from the preset route, possibly due to road construction not being updated in a timely manner), score: 0.75 At this point, the first scenario hypothesis is scenario hypothesis A, with a score of 0.92. The second highest-scoring scenario hypothesis is scenario hypothesis B, with a score of 0.88. If the preset score difference condition is 0.03, then 0.92 - 0.88 = 0.04, which is greater than 0.03. Therefore, scenario hypothesis A is determined to meet the preset iteration termination condition and is identified as a valid scenario hypothesis.

[0076] For example, in another scenario, the system obtained the following situational assumptions and their scores: Scenario Assumption D (e.g., the transport vehicle is speeding, potentially indicating an emergency transport need), Score: 0.95 Scenario hypothesis E (e.g., the transport vehicle is speeding, which may cause false alarms in the system), score: 0.95 Scenario assumption F (e.g., the transport vehicle is speeding, which may indicate driver error), score: 0.80 In this scenario, both scenario hypothesis D and scenario hypothesis E have a score of 0.95, making them the highest-scoring scenario hypotheses with the same score. The system will then select the hypothesis based on predefined priority rules. For example, if the predefined priority rules stipulate that hypotheses concerning urgent transportation needs take precedence over hypotheses that result in false alarms, then scenario hypothesis D will be determined as the valid scenario hypothesis. In this way, even when scores are the same, the system can make a clear and evidence-based decision.

[0077] Specifically, the steps for reconstructing the corresponding scene semantics based on the effective context assumption include: By semantically associating and integrating the current state of the transport vehicle described by the effective scenario hypothesis, the multi-source perception data related to the scenario hypothesis, and the set of logistics scenario information, a scenario semantic description with complete causal logic is reconstructed.

[0078] Semantic association and integration refers to a deep understanding and connection of information from different sources and types to reveal their inherent connections and logical relationships. Specifically, this includes unified semantic analysis and information fusion of the current state of the transport vehicle described in the effective scenario hypothesis (e.g., its position, speed, direction, load, etc.), multi-source sensor data related to this scenario hypothesis (e.g., object recognition results in video images, abnormal readings in sensor data, etc.), and a set of logistics contextual information (e.g., environmental facility information, operational planning information, equipment operation information, environmental status information, etc.). The aim is to integrate these discrete pieces of information into a coherent and meaningful whole. Furthermore, reconstructing a scene semantic description with complete causal logical relationships means not merely aggregating information, but constructing a logical model that can explain the causes, development process, and possible outcomes of an event. For example, if the effective scenario hypothesis indicates that the transport vehicle has deviated from its preset path, the scene semantic description will not only include the fact of the deviation but will also combine multi-source sensor data (such as road obstacles, weather conditions) and logistics contextual information (such as the area's construction plan) to explain why the deviation occurred and predict its potential impact.

[0079] The aforementioned technical solutions enable a deep understanding and accurate modeling of logistics scenarios. Traditional scenario understanding may only remain at the data surface level, failing to capture the underlying causes and potential risks of events. However, this application, by semantically associating and integrating the current state of transport vehicles, multi-source sensing data, and logistics contextual information, can reconstruct a semantic description of the scenario with complete causal logic, thereby significantly improving the ability to identify complex logistics scenarios and the accuracy of early warnings. This method not only identifies anomalies but also reveals the underlying causes, providing more insightful decision support for intelligent logistics systems and effectively avoiding misjudgments or delayed responses caused by information fragmentation.

[0080] Specifically, the steps for issuing early warning information and providing decision support include: Generate early warning messages that include a description of the hazardous situation, relevant multi-source sensing data, and suggested handling measures; display the early warning messages to the operator through a visual interface and issue an alarm via an audible and visual alarm.

[0081] The warning messages aim to provide comprehensive and easily understandable information about hazardous situations. Specifically, the description of the hazardous situation is a detailed textual explanation of the currently identified dangerous event, such as a vehicle deviating from its preset path and posing a collision risk. Related multi-source perception data refers to raw or processed perception data directly associated with the hazardous situation, such as positioning sensor data, obstacle avoidance sensor data, or video stream clips that led to the identification of the hazardous situation. Its purpose is to provide operators with the original basis for situational judgment. Recommended actions are responses automatically generated by the system based on the nature and severity of the hazardous situation, or matched from a preset rule base, such as immediately stopping the vehicle and checking the path.

[0082] Furthermore, the display of warning messages is designed to be multimodal to ensure that operators can receive and understand the warning information in a timely and effective manner. The visual interface can be a monitor, tablet, or console screen, used to intuitively display the full content of the warning message in graphical and textual forms, including a description of the hazardous situation, relevant multi-source sensing data, and suggested handling measures. The operator refers to the person responsible for monitoring and managing the logistics and transportation process. Audible and visual alarms serve as an auxiliary alarm mechanism, attracting the operator's attention by emitting sound and / or flashing lights, especially when the operator may not be continuously monitoring the visual interface, ensuring the immediate delivery of warning information.

[0083] Through the aforementioned technical solutions, this application can significantly improve the timeliness, accuracy, and operability of hazardous situation early warnings in intelligent logistics systems. Specifically, detailed hazardous situation descriptions and related multi-source sensing data provide operators with comprehensive situational awareness capabilities, enabling them to quickly grasp the nature and evidence of events. Simultaneously, the suggested remedial measures provided by the system greatly simplify the operator's decision-making process and reduce the possibility of human error. Furthermore, the combination of a visual interface and audible and visual alarms ensures that early warning information reaches operators in the most effective way, effectively attracting their attention and prompting them to take action even in complex or high-pressure working environments, thereby effectively avoiding or mitigating potential logistics accidents and losses.

[0084] Traditional intelligent logistics pattern recognition methods often face the challenge of effectively transforming abstract recognition logic into an operational system architecture during actual deployment and operation. While simple method descriptions provide theoretical guidance, the lack of a concrete system to efficiently and stably execute these methods in the face of complex logistics environments and diverse sensor data may lead to difficulties in implementation or low execution efficiency.

[0085] Regarding this, secondly, refer to Figure 2 This application proposes an air-ground cooperative multi-source sensing pattern recognition system for intelligent logistics, used to execute the aforementioned air-ground cooperative multi-source sensing pattern recognition method for intelligent logistics. The system includes: The logistics context information set construction module 210 is used to acquire and dynamically update the status information describing the logistics environment in order to construct the logistics context information set; The multi-source sensing data acquisition module 220 is used to acquire multi-source sensing data from air-ground cooperative sensing devices in real time, and to perform time synchronization and format conversion on the multi-source sensing data to obtain processed multi-source sensing data. The scenario hypothesis generation module 230 is used to generate at least one scenario hypothesis describing the current state of the transport vehicle based on the logistics scenario information set and the preset scenario hypothesis template library when an abnormal behavior of the transport vehicle is detected. The verification scoring module 240 is used to retrieve multi-source perception data related to the context hypothesis for verification scoring. The iteration module 250 is used to iteratively perform hypothesis generation and verification scoring until the score of the scenario hypothesis meets the preset iteration termination condition, and the scenario hypothesis that meets the preset iteration termination condition is determined as a valid scenario hypothesis. The early warning module 260 is used to reconstruct the corresponding scene semantics based on the effective scenario hypothesis. When the scene semantics are identified as a dangerous situation and the score of the effective scenario hypothesis reaches the preset early warning threshold, an early warning message is issued and decision support is provided.

[0086] Specifically, the logistics context information set construction module 210 is responsible for continuously collecting and organizing various static and dynamic information within the logistics station, such as infrastructure layout, preset routes of transport vehicles, equipment operating status, and real-time weather or traffic conditions. This information is integrated into a comprehensive logistics context information set, providing basic data support for subsequent context analysis. The multi-source sensing data acquisition module 220 ensures that sensing data from different sources and in different formats can be processed uniformly, eliminating obstacles caused by data heterogeneity and providing consistent input for subsequent pattern recognition. Air-ground collaborative sensing devices can include drones, ground mobile robots, fixed cameras, vehicle-mounted sensors, and RFID readers. Upon receiving an abnormal behavior signal, the context hypothesis generation module 230 combines the current logistics context information set with a selection or combination from a predefined context hypothesis template library to form multiple possible explanations for the current abnormal state, such as vehicle deviation from the predetermined path or the presence of unknown obstacles. The verification and scoring module 240, based on the semantic content of each context hypothesis, selects the most relevant data fragments from the processed multi-source sensing data and analyzes them to obtain verification evidence supporting or refuting the context hypothesis. The iterative module 250 continuously adjusts and evaluates scenario hypotheses through iterative optimization until it finds the most effective scenario hypothesis that explains the current anomaly. The early warning module 260 is the final output of the entire system, responsible for presenting the identified hazardous situations to the operator in an intuitive way and providing corresponding handling suggestions.

[0087] Through the aforementioned system solution, this application transforms the complex air-ground collaborative multi-source sensing pattern recognition method into a deployable and operable physical system. This system, through the close collaboration of its various functional modules, achieves comprehensive perception of the logistics environment, intelligent identification of abnormal behaviors, and timely early warning and decision support for hazardous situations. Compared to solutions that only describe the method, this system provides a stable and efficient operating platform, ensuring the accurate and reliable execution of the method logic in practical applications. It significantly improves situational awareness and risk management capabilities in intelligent logistics scenarios, reduces the frequency and difficulty of manual intervention, and thus enhances the automation and intelligence of overall logistics operations.

[0088] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for air-ground collaborative multi-source sensing pattern recognition for intelligent logistics, characterized in that, include: Acquire and dynamically update status information describing the logistics environment to construct a set of logistics contextual information; Multi-source sensing data from air-ground cooperative sensing devices is acquired in real time, and the multi-source sensing data is processed by time synchronization and format conversion to obtain the processed multi-source sensing data. When abnormal behavior of a transport vehicle is detected, at least one scenario hypothesis describing the current state of the transport vehicle is generated based on the logistics scenario information set and the preset scenario hypothesis template library. Retrieve the multi-source perception data related to the scenario hypothesis for verification and scoring; The hypothesis generation and verification scoring are performed iteratively until the score of the scenario hypothesis meets a preset iteration termination condition, and the scenario hypothesis that meets the preset iteration termination condition is determined as a valid scenario hypothesis. Based on the effective scenario hypothesis, the corresponding scenario semantics are reconstructed. When the scenario semantics are identified as a dangerous scenario and the score of the effective scenario hypothesis reaches a preset warning threshold, a warning message is issued and decision support is provided.

2. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 1, characterized in that, The logistics context information set includes environmental facility information, operation planning information, equipment operation information, and environmental status information.

3. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 2, characterized in that, The multi-source sensing data includes real-time acquired video streams, vehicle sensor data, and RFID tag data; The detected abnormal behavior of the transport vehicle includes the detection that the lateral position deviation between the positioning sensor data in the vehicle sensor data and the preset path continuously exceeds a first preset threshold within a preset time window, or the obstacle avoidance sensor data in the vehicle sensor data indicates the existence of environmental facilities not registered in the environmental facility information.

4. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 1, characterized in that, The step of retrieving and validating the multi-source sensing data related to the scenario hypothesis includes: Based on the semantic content of the context hypothesis, one or more target perception data related to verifying the context hypothesis are selected from the multi-source perception data; The target perception data is analyzed and processed to obtain verification evidence that supports or refutes the situational hypothesis; The validity contribution of the verification evidence is calculated by combining the reliability assessment weight of the target perception data and the semantic matching degree between the verification evidence and the context hypothesis. Based on the validity contribution of all valid evidence, a score for the scenario hypothesis is generated through normalization.

5. A method for air-ground collaborative multi-source sensing pattern recognition for intelligent logistics according to claim 4, characterized in that, The reliability assessment weights of the target sensing data are obtained through the following steps: At least one data quality indicator of the target perception data is monitored in real time. When the real-time monitoring value of the data quality indicator deviates from a preset normal range threshold within a preset time window, the target perception data is determined to be in a state of reduced reliability. Based on the degree of deviation between the real-time monitoring value of the data quality indicator and the preset normal range threshold, the reliability assessment weight of the target perception data is calculated.

6. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 4, characterized in that, The semantic matching degree between the verification evidence and the contextual hypothesis is obtained through the following steps: Semantic or quantitative features used to support or refute the situational hypothesis are extracted from the verification evidence. The semantic or quantitative features are evaluated to assess the semantic or logical consistency and support level between the extracted semantic or quantitative features and the hypothesis features corresponding to the situational hypothesis. The semantic matching degree is then calculated.

7. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 1, characterized in that, The step of determining a scenario hypothesis whose score satisfies a preset iteration termination condition and identifying the scenario hypothesis that satisfies the preset iteration termination condition as a valid scenario hypothesis includes: If there exists a first scenario hypothesis, wherein the first scenario hypothesis is the only scenario hypothesis with the highest score among all scenario hypotheses and the difference between its score and the score of the second highest-scoring scenario hypothesis satisfies a preset score difference condition, then the first scenario hypothesis is determined to satisfy the preset iteration termination condition and is identified as the valid scenario hypothesis. If there are multiple identical scenario hypotheses with the highest scores, then a valid scenario hypothesis is determined from the multiple scenario hypotheses according to a predetermined priority rule.

8. The air-ground collaborative multi-source sensing pattern recognition method for intelligent logistics according to claim 1, characterized in that, The step of reconstructing the corresponding scene semantics based on the effective context assumption includes: The current state of the transport vehicle described by the effective scenario hypothesis, the multi-source perception data related to the scenario hypothesis, and the set of logistics scenario information are semantically correlated and integrated to reconstruct a scenario semantic description with complete causal logic.

9. A method for air-ground collaborative multi-source sensing pattern recognition for intelligent logistics according to claim 1, characterized in that, The steps of issuing early warning information and providing decision support include: Generate the warning message, which includes a description of the dangerous situation, the relevant multi-source sensing data, and suggested handling measures; The warning message is displayed to the operator through a visual interface, and an alarm is triggered by an audible and visual alarm.

10. A multi-source sensing pattern recognition system for air-ground cooperative intelligent logistics, used to execute the multi-source sensing pattern recognition method for air-ground cooperative intelligent logistics as described in any one of claims 1 to 9, characterized in that, The system includes: The logistics context information set construction module is used to acquire and dynamically update the status information describing the logistics environment in order to construct the logistics context information set. The multi-source sensing data acquisition module is used to acquire multi-source sensing data from air-ground cooperative sensing devices in real time, and to perform time synchronization and format conversion on the multi-source sensing data to obtain the processed multi-source sensing data. The scenario hypothesis generation module is used to generate at least one scenario hypothesis describing the current state of the transport vehicle based on the logistics scenario information set and the preset scenario hypothesis template library when abnormal behavior of the transport vehicle is detected. The verification and scoring module is used to retrieve the multi-source perception data related to the scenario hypothesis for verification and scoring. An iterative module is used to iteratively perform hypothesis generation and verification scoring until the score of the scenario hypothesis meets a preset iteration termination condition, and the scenario hypothesis that meets the preset iteration termination condition is determined as a valid scenario hypothesis. The early warning module is used to reconstruct the corresponding scene semantics based on the effective scenario hypothesis. When the scene semantics are identified as a dangerous situation and the score of the effective scenario hypothesis reaches a preset early warning threshold, an early warning message is issued and decision support is provided.

Images