CN-121980408-A - Abnormal behavior detection method and system based on flight task context awareness

CN121980408ACN 121980408 ACN121980408 ACN 121980408ACN-121980408-A

Abstract

The application provides a flight task context awareness-based abnormal behavior detection method and system, which are applied to the technical field of data processing. The unmanned aerial vehicle abnormal behavior judgment method comprises the steps of processing unmanned aerial vehicle flight dynamic data and task protocol data, obtaining task types through analyzing MASSION _ITEM fields in a task protocol to generate dynamic labels, combining GPS positioning data with normalized mutual information and random forest importance scores to screen task sensitive features to generate task related feature subsets, processing the task related feature subsets based on a random forest model, training exclusive classification models for each task type to generate various task exclusive classification models, predicting unmanned aerial vehicle real-time flight state data based on various task exclusive classification models to obtain probability distribution of each task type, judging according to set thresholds, and generating an unmanned aerial vehicle abnormal behavior judgment result.

Inventors

FANG WEI
SUN BINFANG
QUAN XIAOWEN
HAN WEIDONG

Assignees

远江盛邦安全科技集团股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (8)

1. The abnormal behavior detection method based on the context awareness of the flight mission is characterized by comprising the following steps of: Acquiring real-time flight state data of an unmanned aerial vehicle, flight dynamic data of the unmanned aerial vehicle, GPS positioning data and task protocol data; processing unmanned aerial vehicle flight dynamic data and task protocol data, and obtaining a task type through analyzing MASSION _ITEM fields in a task protocol to generate a dynamic tag; The GPS positioning data is combined with normalized mutual information and random forest importance scores to screen task sensitive features, and a task related feature subset is generated; Processing the task related feature subsets based on the random forest model, training an exclusive classification model for each task type, and generating various task exclusive classification models; And predicting the real-time flight state data of the unmanned aerial vehicle based on various task-specific classification models to obtain probability distribution of each task type, and judging according to a set threshold value to generate an abnormal behavior judgment result of the unmanned aerial vehicle.
2. The method of claim 1, wherein processing the unmanned aerial vehicle flight dynamic data, the mission protocol data, and obtaining the mission type by parsing a MASSION _item field in the mission protocol, generating the dynamic tag, comprises: Analyzing MASSION _ITEM field in the task protocol data to obtain task type and generating primarily determined task type information; Checking and confirming the primarily determined task type information to generate a precise task type result; carrying out segment association on unmanned aerial vehicle flight dynamic data according to accurate task type results to generate a preliminary marked data segment; and performing format specification and information perfection on the preliminarily marked data segment to generate a dynamic label.
3. The method of claim 1, wherein screening task sensitive features for GPS positioning data in combination with normalized mutual information and random forest importance scores to generate a subset of task related features, comprising: The GPS positioning data is subjected to format unification and missing value filling processing, and the primarily processed GPS positioning data is generated; combining the primarily processed GPS positioning data with task type labels, calculating normalized mutual information, and generating normalized mutual information associated data; carrying out random forest importance scoring calculation on the GPS positioning data subjected to preliminary processing to generate random forest scoring data; the normalized mutual information associated data and random forest scoring data are based on a formula Performing fusion calculation to generate a feature weight calculation result, wherein, Represent the first The combined weight of the individual features is used, In order to be able to adapt the coefficients, For normalizing mutual information, for measuring Features of With task type tags The correlation between the two is that, Represents the first The individual features are a specific feature selected from a plurality of data features collected by the unmanned aerial vehicle, A task type tag for identifying the type of task currently being performed by the drone, Is calculated by a random forest algorithm Features of Importance score of (2); And sorting the feature weight calculation results according to the descending order of the weights, reserving the features of which the weight sum is 80% before, and generating a task related feature subset.
4. The method of claim 1, wherein processing the task-related feature subsets based on a random forest model trains a proprietary classification model for each task type, generating various types of task-proprietary classification models, comprising: Acquiring a task related feature subset and a preset random forest initial model, wherein the task related feature subset is used for representing the screened key data features under different task types; Carrying out data denoising, feature screening and standardization processing on the task related feature subsets to generate preprocessed feature subset data; Grouping and sorting the preprocessed feature subset data according to task types to generate a task type grouping data set; performing training parameter setting processing on the task type grouping data set to generate random forest model training parameters; Performing iterative training treatment on a preset random forest initial model based on random forest model training parameters to generate a random forest classification model for preliminary training; And performing performance evaluation and super-parameter adjustment processing on the preliminarily trained random forest classification model to generate various task-specific classification models.
5. The method of claim 4, wherein predicting the real-time flight state data of the unmanned aerial vehicle based on each type of task-specific classification model to obtain probability distribution of each task type, and determining according to a set threshold value to generate an abnormal behavior determination result of the unmanned aerial vehicle comprises: Processing the real-time flight state data of the unmanned aerial vehicle based on various task-specific classification models to generate initial probability distribution information of each task type; classifying and sorting the initial probability distribution information of each task type to generate probability distribution grouping information under different task types; Carrying out key probability value extraction processing on probability distribution grouping information to generate probability characteristic data sets of different groups; Judging and processing the probability characteristic data sets of different groups by combining with a target threshold value to obtain an abnormal behavior judgment primary result of the unmanned aerial vehicle; performing complementary analysis processing on probability characteristic data sets of different groups based on the preliminary result of abnormal behavior judgment of the unmanned aerial vehicle, and generating detailed attribute information of abnormal behavior judgment corresponding to the different groups; And generating comprehensive and accurate unmanned aerial vehicle abnormal behavior judgment results based on detailed attribute information of abnormal behavior judgment corresponding to different groups.
6. An abnormal behavior detection device based on flight mission context awareness, the device comprising: the data collection module is used for acquiring real-time flight state data of the unmanned aerial vehicle, flight dynamic data of the unmanned aerial vehicle, GPS positioning data and task protocol data; The system comprises a data processing module, a task type generation module, a task related feature subset generation module, a task model generation module and a task model generation module, wherein the data processing module is used for processing unmanned aerial vehicle flight dynamic data and task protocol data, acquiring task types through a MASSION _ITEM field in a task protocol to generate dynamic labels, combining GPS positioning data with normalized mutual information and random forest importance scores to screen task sensitive features to generate task related feature subsets, processing the task related feature subsets based on the random forest models, training exclusive classification models for each task type to generate various task exclusive classification models, predicting unmanned aerial vehicle real-time flight state data based on various task exclusive classification models to obtain probability distribution of each task type, judging according to set thresholds, and generating an unmanned aerial vehicle abnormal behavior judgment result.
7. An electronic device, comprising: and a memory for storing executable instructions of the first processor; Wherein the first processor is configured to perform the method of any one of claims 1-5 based on flight mission context awareness abnormal behavior detection via execution of the executable instructions.
8. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a second processor implements the method for detecting abnormal behavior based on flight mission context awareness according to any one of claims 1 to 5.

Description

Abnormal behavior detection method and system based on flight task context awareness Technical Field The invention relates to the technical field of data processing, in particular to a flight task context awareness-based abnormal behavior detection method and system. Background In the current technological development wave, unmanned aerial vehicles are widely applied in a plurality of fields such as logistics, agriculture, rescue and the like, and low-altitude economy is also brought into the category of the national strategic emerging industry. As a low-altitude economic core carrier, safe, efficient and compliant operation of unmanned aerial vehicles becomes a key place for industrial landing. Under the background, the importance of unmanned aerial vehicle supervision technology is increasingly highlighted, and the abnormal behavior detection technology is a key link for guaranteeing unmanned aerial vehicle operation safety. Currently, existing unmanned aerial vehicle abnormal behavior detection techniques rely mainly on static thresholds (e.g., geofences, speed, altitude overruns) or general sensor data to analyze abnormal behavior. However, these conventional techniques have significant limitations. Firstly, the adaptability is insufficient, and the static threshold is difficult to be matched with the dynamic behavior modes of different task types (such as logistics transportation, inspection and emergency response). Taking logistics transportation tasks as an example, different cargo weights, distribution areas and ageing requirements can enable behavior modes such as unmanned aerial vehicle flight speed, height and the like to be huge, and static thresholds cannot be flexibly adapted to the changes, so that misjudgment on normal flight behaviors or missed judgment on abnormal behaviors can be caused. Secondly, the calculation efficiency is low, and a great deal of feature redundancy can be generated by directly using the original sensor data. The method not only increases the consumption of computing resources and prolongs the data processing time, but also can lead to the increase of the false alarm rate and reduce the accuracy and the instantaneity of detection. Thirdly, semantic association is lost, the prior art does not closely associate the flight task type with the behavior characteristics, and task camouflage attacks are difficult to identify. For example, the act of performing illegal photographing on the behalf of a patrol is difficult to discover and refrain in time due to the lack of comprehensive analysis of task types and behavioral characteristics. It should be noted that the information disclosed in the above background section is only for enhancing understanding of the background of the present disclosure and thus may include information that does not constitute prior art known to those of ordinary skill in the art. Disclosure of Invention The application aims to provide an abnormal behavior detection method and system based on flight task context awareness, which at least overcome the problems existing in the prior art to a certain extent, acquire unmanned aerial vehicle flight dynamics, GPS positioning and task protocol data, analyze the protocol to generate dynamic labels, and then combine NMI and random forest importance scoring and screening characteristics to train a dedicated random forest model of each task. These models are used to predict flight status data, determine anomalies, and analyze attributes. The detection precision and the real-time performance are improved through dynamic feature selection and task-behavior verification, and the safe and compliant operation of the unmanned aerial vehicle is ensured. Other features and advantages of the application will be apparent from the following detailed description, or may be learned by the practice of the application. According to one aspect of the application, the method for detecting the abnormal behavior based on the context awareness of the flight task comprises the steps of obtaining real-time flight state data of an unmanned aerial vehicle, flight dynamic data of the unmanned aerial vehicle, GPS positioning data and task protocol data, processing the unmanned aerial vehicle flight dynamic data and the task protocol data, obtaining task types through analyzing MASSION _ITEM fields in the task protocol to generate dynamic labels, screening task sensitive features by combining normalized mutual information and random forest importance scores to generate task related feature subsets, processing the task related feature subsets based on the random forest models, training exclusive classification models for each task type to generate various task exclusive classification models, predicting the unmanned aerial vehicle real-time flight state data based on the various task exclusive classification models to obtain probability distribution of each task type, judging according to set threshold values, and generating an unmanned