A multi-sensor fusion drone detection method and system

Abstract

The present invention provides a multi-sensor fusion unmanned aerial vehicle detection method and system, wherein the present invention receives data streams from various sensors to obtain a comprehensive perception information set, uses a deep learning model to analyze the comprehensive perception information set, generates dynamic target behavior pattern data, applies a method combining adaptive noise cancellation technology with a machine learning optimized filter to optimize signal quality, generates an optimized detection signal, predicts a flight path through a predictive analysis algorithm, generates predicted flight path data, processes the comprehensive perception information set through a distributed edge computing framework, optimizes the deep learning model through a federated learning mechanism, generates an optimized detection strategy, and generates an updated detection process based on the predicted flight path data and the optimized detection strategy; the technical solution provided by the present invention enhances the accuracy of target recognition, ensures the high quality of the signal through the detection capability under low visibility conditions, and achieves rapid response.

Description

Technical Field

[0001] The embodiments of the present invention relate to the technical field of drone detection, and in particular to a drone detection method using multiple sensor fusion. Background Art

[0002] In the field of drone detection, with the development of technology and the diversification of application scenarios, the traditional single sensor detection method can no longer meet the high-precision detection needs in complex environments. The existing drone detection methods lack effective fusion of multi-source sensor data and fail to fully utilize the advantages of different types of sensors to improve detection performance; secondly, the traditional methods fail to fully consider the influence of noise interference and background complexity in the signal processing process, resulting in a decrease in the accuracy of target recognition; thirdly, the existing flight path prediction algorithms are mostly based on static environment assumptions, ignoring the impact of real-time meteorological conditions and terrain changes on drone flight paths; finally, the centralized cloud computing architecture not only increases the burden of data transmission, but also limits the system's response speed and privacy protection capabilities. Summary of the invention

[0003] The embodiments of the present invention provide a multi-sensor fusion drone detection method and system to solve the problems of insufficient multi-source sensor data fusion, lack of noise suppression and background complexity consideration during signal processing, neglect of dynamic environmental factors in flight path prediction, and real-time and privacy protection caused by centralized cloud computing architecture in existing drone detection technologies.

[0004] In a first aspect, an embodiment of the present invention provides a method for detecting a drone using multiple sensor fusions, including:

[0005] Receiving data streams from various sensors, synchronizing and correlating the data streams from various sensors in time domain and space domain, and obtaining a comprehensive perception information set;

[0006] Analyzing the comprehensive perception information set using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions;

[0007] Based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with a machine learning optimized filter is used to optimize signal quality, and spectrum analysis technology is used to assist in identification to generate an optimized detection signal;

[0008] In combination with geographic information systems, real-time meteorological data, three-dimensional terrain models, and urban structure databases, a predictive analysis algorithm is used to predict the flight path of the optimized detection signal to generate predicted flight path data;

[0009] A distributed edge computing framework is used to process the comprehensive perception information set, and a deep learning model is optimized through a federated learning mechanism to generate an optimized detection strategy. Based on the predicted flight path data and the optimized detection strategy, an updated detection process is generated.

[0010] Optionally, based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with a machine learning optimized filter is used to optimize signal quality, and spectrum analysis technology is used to assist in identification to generate an optimized detection signal, including:

[0011] Adaptive noise cancellation technology is used to pre-process the original sensor signal in combination with real-time environmental parameters to obtain a preliminary purified signal;

[0012] According to the preliminary purified signal, filtering is performed through a multi-layer filter optimized by machine learning, and an optimal filtering path is determined through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter, and a neural network filter;

[0013] Utilize spectrum analysis technology to assist in identifying the high-quality detection signal, analyze the specific electromagnetic spectrum characteristics emitted by the drone, and combine historical flight records and dynamic target behavior pattern data for cross-verification to generate a detection signal with material characteristics and behavior features;

[0014] Based on the detection signal with material properties and behavioral characteristics, the spatiotemporal correlation analysis algorithm is used to integrate the preliminary purification signal and the high-quality detection signal, evaluate the continuity and consistency of the UAV in different time and space dimensions, and obtain the optimized detection signal.

[0015] Optionally, according to the preliminary purified signal, filtering processing is performed through a multi-layer filter optimized by machine learning, and an optimal filtering path is determined through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter and a neural network filter, including:

[0016] The Kalman filter is combined with an adaptive gain adjustment mechanism to perform the initial dynamic state estimation processing on the preliminary purified signal, obtain the initial filtering result, and record the state parameters and uncertainty measurement during the filtering process;

[0017] Based on the initial filtering results, state parameters and uncertainty measures, a particle filter is applied in combination with a Bayesian update rule and a resampling strategy to perform nonlinear secondary filtering on the initial filtering results to obtain secondary filtering results, and particle distribution, weights and prediction error covariance are simultaneously evaluated;

[0018] Using a deep neural network filter, based on the secondary filtering results and particle distribution, using transfer learning technology to initialize network parameters, and introducing an adversarial training mechanism to perform deep learning optimization processing to generate a filtering signal optimized by the neural network;

[0019] Combining the initial filtering results, the secondary filtering results and the filtering signal after neural network optimization, the performance of different filters is evaluated by the model selection algorithm under the integrated learning framework, the confidence, relevance and diversity of the output of each layer of filters are evaluated, and the optimal filtering path is determined. At the same time, the weight distribution of each layer of filters is dynamically adjusted by the reinforcement learning algorithm, and the output of each layer of filters is integrated to obtain the comprehensive filtering signal after the optimized path;

[0020] According to the comprehensive filtered signal after the optimization path, a context-aware mechanism is introduced, and real-time environmental parameters, historical flight data and dynamic target behavior patterns are used as auxiliary inputs. The relationship between sensors is modeled using graph neural networks to capture spatiotemporal dependencies and generate high-quality detection signals.

[0021] Optionally, combining the primary filtering result, the secondary filtering result and the filtering signal after the neural network optimization, the performance of different filters is evaluated by the model selection algorithm under the integrated learning framework, the confidence, relevance and diversity of the output of each layer of filters are evaluated, and the optimal filtering path is determined. At the same time, the weight distribution of each layer of filters is dynamically adjusted by using the reinforcement learning algorithm, and the output of each layer of filters is integrated to obtain the comprehensive filtering signal after the optimized path, including:

[0022] Using the model selection algorithm under the ensemble learning framework, based on the primary filtering results, secondary filtering results and the filtered signals after neural network optimization, the confidence, relevance and diversity of the output of each layer of filters are evaluated and processed to obtain a preliminary performance evaluation report;

[0023] According to the preliminary performance evaluation report, the Bayesian optimization algorithm is used to dynamically adjust the weight distribution of filters in each layer to generate a set of filtered signals after weight adjustment;

[0024] Based on the weighted filtered signal set, an adaptive feedback mechanism is constructed through a reinforcement learning algorithm to monitor and adjust the filtering path in real time to obtain an optimized path solution;

[0025] The outputs of each layer of filters in the optimized path scheme are integrated to form a comprehensive filtering signal, and a context perception module is introduced to take environmental parameters, historical flight data and dynamic target behavior patterns as auxiliary inputs to obtain a comprehensive filtering signal after the optimized path.

[0026] Optionally, the comprehensive perception information set is analyzed using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions, including:

[0027] Based on the comprehensive perception information set, a pre-trained deep learning model is used to perform preliminary feature extraction processing on the sensor data to obtain a multi-dimensional feature representation;

[0028] According to the multi-dimensional feature representation, a time series analysis algorithm is applied to identify and extract the historical flight path, speed change, and altitude adjustment of the UAV to obtain flight characteristic data;

[0029] In combination with the flight characteristic data, an attitude estimation algorithm is used to analyze and process the acceleration parameters and flight attitude of the UAV to generate a preliminary draft of the behavior pattern;

[0030] By using the environment interaction analysis module, the draft of the behavior pattern is combined with the real-time environment interaction information, and an environment adaptability adjustment process is performed to generate behavior pattern data after environment adaptation;

[0031] The behavioral pattern data after environmental adaptation, the draft behavioral pattern, the flight characteristic data and the multi-dimensional feature representation are integrated, and a comprehensive evaluation is performed through the high-level abstract layer of the deep learning model to obtain the dynamic target behavior pattern data.

[0032] Optionally, combining a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, a predictive analysis algorithm is used to predict the flight path of the optimized detection signal to generate predicted flight path data, including:

[0033] Using geographic information system and real-time meteorological data, the optimized detection signal is subjected to environmental factor fusion processing to obtain an environmentally fused detection signal;

[0034] Based on the detection signal after the environment fusion, combined with the three-dimensional terrain model, spatial feature matching processing is performed through a terrain adaptability adjustment algorithm to generate a detection signal after terrain matching;

[0035] According to the signal after terrain matching, the building and infrastructure information in the urban structure database is applied to perform obstacle avoidance and path optimization processing to obtain a preliminary flight path prediction;

[0036] Combined with the preliminary flight path prediction, using a predictive analysis algorithm, integrating historical flight data and the trend of the current detection signal, performing dynamic prediction processing of the flight path, and generating a dynamic flight path prediction result;

[0037] The dynamic flight path prediction results, the detection signals after environmental fusion and the preliminary flight path prediction are integrated, and the predicted flight path data is generated through multi-source data fusion technology.

[0038] Optionally, a distributed edge computing framework is used to process the comprehensive perception information set, a deep learning model is optimized through a federated learning mechanism, an optimized detection strategy is generated, and an updated detection process is generated based on the predicted flight path data and the optimized detection strategy, including:

[0039] Using a distributed edge computing framework, the comprehensive perception information set is locally preprocessed and preliminarily analyzed to obtain an edge processing result;

[0040] Based on the edge processing results, combined with the federated learning mechanism, the deep learning model is independently trained on each edge node to obtain a trained deep training model, and the trained deep training model is updated and uploaded to the central server for aggregation processing to generate a globally optimized deep learning model;

[0041] According to the globally optimized deep learning model, it is applied to each edge node to perform intelligent analysis and processing on the received sensor data to generate an optimized detection strategy;

[0042] Combining the predicted flight path data with the optimized detection strategy, using a decision fusion algorithm to integrate multi-source information, formulate a specific action plan, and obtain an initial detection process;

[0043] Based on the initial detection process, a feedback adjustment mechanism is introduced to dynamically modify the detection strategy and the detection process according to the actual execution situation to obtain an updated detection process.

[0044] In a second aspect, an embodiment of the present invention provides a multi-sensor fusion drone detection system, including:

[0045] A receiving module, used to receive the data streams of each sensor, synchronize and associate the data streams of each sensor in the time domain and the spatial domain, and obtain a comprehensive perception information set;

[0046] An analysis module for analyzing the comprehensive perception information set using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions;

[0047] An optimization module, for optimizing signal quality based on the dynamic target behavior pattern data by combining adaptive noise cancellation technology with a machine learning optimized filter, and using spectrum analysis technology to assist in identification, thereby generating an optimized detection signal;

[0048] A prediction module, for combining a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, and using a predictive analysis algorithm to predict the flight path of the optimized detection signal to generate predicted flight path data;

[0049] A generation module is used to process the comprehensive perception information set using a distributed edge computing framework, optimize the deep learning model through a federated learning mechanism, generate an optimized detection strategy, and generate an updated detection process based on the predicted flight path data and the optimized detection strategy.

[0050] In a third aspect, an embodiment of the present invention provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a drone detection method using multiple sensor fusion as described in any one of the first aspects.

[0051] In a fourth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a drone detection method using multiple sensor fusion as described in any one of the first aspects.

[0052] In an embodiment of the present invention, data streams of various sensors are received, and the data streams of the various sensors are synchronized and associated in the time domain and the spatial domain to obtain a comprehensive perception information set; the comprehensive perception information set is analyzed using a deep learning model, and dynamic target behavior pattern data is generated by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions; based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with a machine learning optimized filter is used to optimize signal quality, and spectral analysis technology is used to assist in identification to generate an optimized detection signal; in combination with a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, an optimized detection signal is generated by using a method combining adaptive noise cancellation technology with a machine learning optimized filter. The predictive analysis algorithm predicts the flight path of the optimized detection signal and generates predicted flight path data; the distributed edge computing framework is used to process the comprehensive perception information set, the deep learning model is optimized through the federated learning mechanism, and the optimized detection strategy is generated; based on the predicted flight path data and the optimized detection strategy, an updated detection process is generated; the technical solution provided by the present invention improves the flexibility and adaptability of the detection system, enhances the accuracy of target recognition, effectively reduces background noise interference, enhances the detection capability under low visibility conditions, ensures the high quality of the signal, reduces the cloud computing burden, achieves rapid response, and protects user privacy, ensuring the efficiency and security of the system;

[0053] Furthermore, in-depth improvements have been made to the performance evaluation and optimization path selection in the filtering process. First, the confidence, relevance and diversity of the output of each layer of filters are evaluated based on the initial filtering results, the secondary filtering results and the filtered signals after neural network optimization using the model selection algorithm under the integrated learning framework, and a preliminary performance evaluation report is obtained to ensure a comprehensive evaluation of the performance of different filters and provide a scientific basis for subsequent optimization; based on the preliminary performance evaluation report, the Bayesian optimization algorithm is used to dynamically adjust the weight distribution of each layer of filters to generate a set of filtered signals after adjusting the weights. In this way, the best performance of each layer of filters can be brought into play during the fusion process, the limitations that may be brought by a single filter are avoided, and the overall filtering effect is improved; next, based on the set of filtered signals after the weights are adjusted, an adaptive feedback mechanism is constructed through a reinforcement learning algorithm, the filtering path is monitored and adjusted in real time, and an optimized path plan is obtained, thereby realizing dynamic adjustment of the filtering path, and being able to flexibly respond to changes in the detection environment and maintain efficient filtering performance; finally, the outputs of each layer of filters in the optimized path plan are integrated to form a comprehensive filtering signal, and a context perception module is introduced, and environmental parameters, historical flight data, and dynamic target behavior patterns are used as auxiliary inputs to obtain a comprehensive filtering signal after the optimized path, which not only enhances the adaptability and accuracy of the comprehensive filtering signal, but also ensures the coherence and consistency in the process from preliminary performance evaluation to high-quality detection signal generation; in summary, the quality and stability of the filtered signal are significantly improved, thereby providing a more reliable foundation for subsequent drone detection.

[0054] These and other aspects of the present invention will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flowchart of a drone detection method using multiple sensor fusion provided by an embodiment of the present invention;

[0057] Figure 2 A schematic diagram of the structure of a drone detection system with multiple sensor fusion provided by an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the structure of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0060] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0062] Figure 1 A flowchart of a method for detecting a drone using multiple sensor fusion is provided for an embodiment of the present invention. Figure 1 As shown, the method includes:

[0063] Step 101: receiving data streams from various sensors, synchronizing and associating the data streams from various sensors in the time domain and the spatial domain, and obtaining a comprehensive sensing information set;

[0064] In this step, the comprehensive perception information set refers to the data set collected by various sensors (such as radar, camera, infrared imager, acoustic sensor, etc.). These sensor data contain various information about the target, such as position, speed, shape, temperature distribution, etc. In order to ensure that data from different sources can accurately correspond to the same time point and spatial position, synchronization and association in the time domain and space domain must be performed. It is the basis for all subsequent analysis and ensures the consistency and integrity of the data;

[0065] First, it receives data streams from multiple different types of sensors, including but not limited to radar reflection signals, high-resolution optical images, infrared thermal imaging data, and acoustic wave signals. Then, it uses timestamp alignment technology and spatial coordinate conversion algorithms to accurately match these heterogeneous data in time and space dimensions, thereby forming a unified comprehensive perception information set, ensuring that all processing in subsequent steps is based on high-quality data under the same time and space benchmark;

[0066] In a practical application, suppose that drone activities are being monitored in an urban area. The system collects real-time data from radar stations, camera networks, and infrared sensors installed at different locations simultaneously. The time synchronization protocol (NTP) ensures that the timestamps of all devices are consistent, and GPS positioning technology is used to calibrate the spatial coordinates of each sensor. Subsequently, these synchronized data are transmitted to the central processing unit, where they are integrated into a comprehensive set of perception information for subsequent deep learning analysis.

[0067] Step 102: Analyze the comprehensive perception information set using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions;

[0068] In this step, dynamic target behavior pattern data refers to the behavior characteristics extracted by analyzing the historical flight records and other relevant characteristics of the drone (such as speed, altitude, acceleration, flight attitude, etc.), as well as its interaction with the surrounding environment. This data can reveal the movement patterns of the drone and its possible destination or intention, which is crucial for predicting its future behavior;

[0069] Using pre-trained deep learning models, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), we conduct in-depth analysis of multimodal data such as images, video clips, radar echoes, etc. in the integrated perception information set. The model not only identifies static features (such as object outlines), but also captures dynamic characteristics (such as speed changes and flight attitudes). In addition, it combines environmental factors (such as wind speed and temperature gradients) to further understand the interactive relationship between drones and the environment, and ultimately generates a data set that describes the drone's behavior patterns.

[0070] Continuing with the above urban monitoring scenario, once the comprehensive perception information set is ready, the system will call the pre-trained deep learning model, which will analyze the flight trajectory of the drone over the past period of time and identify specific speed change patterns, altitude adjustment behaviors, and interactions with buildings or other obstacles. For example, if a drone frequently flies around high-rise buildings, the model may mark this as abnormal behavior. In this way, a detailed database of dynamic target behavior patterns is established, providing an important reference for the next step of signal optimization.

[0071] Step 103: Based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with a machine learning optimized filter is used to optimize signal quality, and spectrum analysis technology is used to assist identification to generate an optimized detection signal;

[0072] In this step, adaptive noise cancellation technology is a technology that can automatically adjust parameters according to the characteristics of background noise to eliminate or reduce interference. The machine learning optimized filter refers to a filter trained by machine learning methods, which can more effectively remove noise and retain useful signals. Spectral analysis technology is used to analyze the characteristics of different frequency bands of electromagnetic waves to help distinguish targets from backgrounds.

[0073] Based on the dynamic target behavior pattern data, adaptive noise cancellation technology is first applied to reduce the impact of background noise. Then, machine learning optimized filters are used to further improve signal quality to ensure that the presence of drones can be clearly detected even in complex backgrounds. Finally, spectrum analysis technology is used to supplement the optimized signal, especially for the unique electromagnetic spectrum characteristics reflected or emitted by drones, to enhance recognition accuracy.

[0074] In the aforementioned urban monitoring environment, when the system has built dynamic target behavior pattern data, the next step is to optimize the quality of the detection signal. Specifically, the system will first use adaptive noise cancellation technology to filter out urban traffic noise and radio interference. After that, through advanced filtering technologies such as Kalman filters or particle filters optimized by machine learning, the signal is further purified to make the characteristics of the drone more prominent. Finally, with the help of spectral analysis technology, the system can confirm its type and status based on the specific electromagnetic spectrum reflected by the drone material, such as distinguishing civilian drones from military drones, or identifying drones disguised as birds.

[0075] Step 104: combining the geographic information system, real-time meteorological data, three-dimensional terrain model and urban structure database, using a predictive analysis algorithm to predict the flight path of the optimized detection signal to generate predicted flight path data;

[0076] In this step, Geographic Information System (GIS) is an integrated tool for storing, managing, analyzing and displaying all forms of geographic data. Real-time meteorological data provides information on current weather conditions, such as temperature, humidity, wind speed, etc., which is very important for predicting the flight path of drones. The three-dimensional terrain model depicts the ground undulations and building layout in detail, which helps to simulate the physical obstacles that drones may encounter. The "Urban Structure Database" contains specific information about urban infrastructure, such as roads, bridges, high-rise buildings, etc.;

[0077] By analyzing the optimized detection signals and combining them with rich background information provided by GIS, real-time meteorological data, 3D terrain models and urban structure databases, the system uses a predictive analysis algorithm to infer the future flight path of the drone. This method not only takes into account the current position and speed of the drone, but also fully considers the impact of external environmental factors on the flight path, thereby improving the accuracy of the prediction;

[0078] In the urban monitoring scenario, after completing the signal optimization, the system begins to predict the flight path of the drone. It combines the map information provided by GIS, the latest meteorological data (such as wind speed and direction), the three-dimensional terrain model (such as the distribution of hills and tall buildings), and the urban structure database (such as the location of bridges and tunnels), and calculates the most likely flight path through complex predictive analysis algorithms. For example, if the meteorological data shows that strong winds are about to blow from west to east, the system will predict that the drone may adjust its route to avoid flying against the wind, while taking into account the blocking effect of nearby tall buildings, and choose a safer and more efficient flight route.

[0079] Step 105: Processing the comprehensive perception information set using a distributed edge computing framework, optimizing the deep learning model through a federated learning mechanism, generating an optimized detection strategy, and generating an updated detection process based on the predicted flight path data and the optimized detection strategy;

[0080] In this step, the distributed edge computing framework allows data processing to be performed close to the data source (i.e., edge nodes), reducing data transmission latency and bandwidth requirements. The federated learning mechanism is a distributed machine learning method that allows multiple participants to jointly train models without sharing original data, which protects privacy and improves model performance. The optimized detection strategy refers to an action plan adjusted based on the latest detection results, aiming to improve detection efficiency and accuracy;

[0081] The distributed edge computing framework deployed on each edge node is responsible for processing the local comprehensive perception information set and continuously optimizing the deep learning model through the federated learning mechanism. The benefit of this is that all participating nodes can benefit from the global optimal model without sacrificing privacy. Based on the predicted flight path data and the optimized detection strategy, the system will generate a new detection process to ensure that each detection can make the best decision based on the latest intelligence;

[0082] In urban monitoring scenarios, once the flight path prediction is completed, the system will use a distributed edge computing framework to process local integrated perception information sets at various monitoring sites (such as the top of a tall building or a key intersection). These sites not only process data independently, but also regularly upload model updates to the central server through a federated learning mechanism to achieve continuous optimization of the global model. Ultimately, based on the predicted flight path data and the optimized detection strategy, the system will generate an updated detection process to guide each site on how to better allocate resources and adjust the monitoring angle to ensure effective tracking and response to drone activities. For example, if the forecast shows that a drone will fly to the city center, the system will notify monitoring sites near the city center in advance to strengthen monitoring efforts and prepare emergency measures.

[0083] Based on this, the present invention provides a specific embodiment, wherein step 103, based on the dynamic target behavior pattern data, applies a method combining adaptive noise cancellation technology with a machine learning optimized filter to optimize signal quality, and uses spectrum analysis technology to assist in identification, to generate an optimized detection signal, specifically comprising the following steps:

[0084] Step 201: using adaptive noise cancellation technology and combining real-time environmental parameters to pre-process the original sensor signal to obtain a preliminary purified signal;

[0085] In this step, adaptive noise cancellation technology can automatically adjust parameters according to the background noise characteristics to eliminate or reduce interference. Real-time environmental parameters include temperature, humidity, wind speed and other data, which can affect the quality of the detection signal;

[0086] After receiving the original sensor signal, an adaptive noise cancellation algorithm is applied to dynamically adjust internal parameters to adapt to the changing background noise environment. At the same time, real-time environmental parameters such as temperature, humidity, wind speed, etc. from the weather station and the sensor itself are collected and integrated to further optimize the noise cancellation effect. After this series of processing, non-correlated noise is effectively removed, generating a preliminary purified signal, providing a high-quality data basis for subsequent analysis;

[0087] When monitoring drone activity in an urban environment, multiple types of sensors (radar, camera, infrared imager) capture a large amount of raw signal containing background noise. To ensure signal quality, adaptive noise cancellation technology is used to eliminate interference caused by traffic noise, construction site machinery, etc. Real-time environmental parameters are obtained from nearby weather stations and this information is fed back to the noise cancellation algorithm so that it can better adapt to current conditions. For example, in windy weather, the parameters are adjusted appropriately to compensate for the impact of wind noise, thereby obtaining a clearer initial clean signal.

[0088] Step 202: According to the preliminary purified signal, filtering is performed through a multi-layer filter optimized by machine learning, and an optimal filtering path is determined through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter, and a neural network filter;

[0089] In this step, multi-layer filters refer to multi-level filter structures trained by machine learning methods, including Kalman filters, particle filters, and neural network filters. Each filter is good at processing different types of data features, and their combination can significantly improve signal processing effects; the model selection algorithm is used to evaluate the performance of different filters and select the best combination to ensure the highest quality of the output signal;

[0090] Based on the preliminary purified signal, the system sequentially performs initial state estimation through the Kalman filter, nonlinear secondary filtering through the particle filter, and finally advanced optimization processing through the deep neural network filter. At each stage, the system records key performance indicators and finally comprehensively evaluates the effects of each layer of filters through the model selection algorithm. The model selection algorithm not only considers the independent performance of each layer of filters, but also evaluates the synergy between them, so as to determine the optimal filtering path and generate high-quality detection signals;

[0091] Continuing with the above urban monitoring scenario, once the preliminary purification signal is obtained, the system will start multi-layer filtering processing. First, the Kalman filter will perform an initial state estimation on the signal to eliminate short-term fluctuations; then the particle filter will process complex nonlinear changes and capture possible irregular movements of the drone; finally, the deep neural network filter will further optimize the signal through the deep learning model to identify weak but important features. During the whole process, the system continuously monitors the filtering effect and determines the optimal filtering path through the model selection algorithm. For example, if there is a large amount of random noise in a certain signal segment, the system may rely more on the powerful denoising ability of the particle filter. The high-quality detection signal finally generated can accurately reflect the actual flight status of the drone.

[0092] Step 203: Utilize spectrum analysis technology to assist in identifying the high-quality detection signal, analyze the specific electromagnetic spectrum characteristics emitted by the UAV, and perform cross-validation in combination with historical flight records and dynamic target behavior pattern data to generate a detection signal with material characteristics and behavior features;

[0093] In this step, spectrum analysis technology is used to analyze the characteristics of different frequency bands of electromagnetic waves to help distinguish targets from backgrounds. Specific electromagnetic spectrum characteristics refer to the unique electromagnetic wave frequency range reflected or emitted by drones. These characteristics can help accurately identify the type of drone and its material. Historical flight records and dynamic target behavior pattern data provide information about the drone's past behavior, which helps to enhance identification accuracy;

[0094] Spectral analysis technology is used to conduct in-depth analysis of high-quality detection signals, especially focusing on the specific electromagnetic spectrum characteristics reflected or emitted by drones. It not only enhances the ability to identify the type of drone, but also reveals its surface material properties. Subsequently, the analysis results are cross-validated with historical flight records and dynamic target behavior pattern data to ensure the consistency and reliability of the identification results. Ultimately, a detection signal with material properties and behavioral characteristics is generated, providing an important basis for subsequent decision-making;

[0095] After completing the multi-layer filtering process, spectral analysis technology is further used to confirm the identity of the drone. Some drones may emit radio signals of specific frequencies, or their surface materials may reflect light of specific wavelengths. By analyzing these signal characteristics, the specific model and purpose of the drone can be determined. In addition, the historical flight records and known behavior patterns of the drone are reviewed to check whether the new findings are consistent with previous records. For example, if a commercial drone often performs missions in the same area and the flight path is similar each time, its identity can be confirmed by comparison, generating a detection signal with detailed material characteristics and behavioral characteristics.

[0096] Step 204: Based on the detection signal with material characteristics and behavior characteristics, a spatiotemporal correlation analysis algorithm is used to integrate the preliminary purification signal and the high-quality detection signal, and the continuity and consistency of the UAV in different time and space dimensions are evaluated to obtain an optimized detection signal;

[0097] In this step, the spatiotemporal correlation analysis algorithm refers to an algorithm that can capture the relationship between time series and spatial distribution, and is suitable for analyzing the behavior patterns of drones at different times and locations. Continuity and consistency refer to the degree to which the behavior patterns of drones remain coherent and logical when they move between multiple observation points;

[0098] Using the spatiotemporal correlation analysis algorithm, the detection signals with material characteristics and behavioral features are integrated with the preliminary purification signals and high-quality detection signals. The algorithm evaluates the continuity and consistency of the drone's behavior pattern by comparing data at different time periods and locations. This not only verifies the accuracy of the detection signal, but also detects potential anomalies. Ultimately, an optimized detection signal is generated, providing a solid foundation for subsequent decision-making;

[0099] Once a detection signal with material properties and behavioral characteristics has been generated, the next step is to ensure that this information is consistent in time and space. For example, if a drone appears frequently in a certain area over a period of time, its flight trajectory will be checked to see if it is coherent and whether there are any sudden deviations from the normal path. Through the spatiotemporal correlation analysis algorithm, the behavior pattern of the drone throughout the flight is evaluated to ensure that all observations support each other. If there are any inconsistencies, they are marked for further investigation. The resulting optimized detection signal not only contains rich details, but also ensures its high consistency in the spatiotemporal dimensions.

[0100] Based on this, the present invention provides a specific embodiment, in which step 202 performs filtering processing on the preliminary purified signal through a multi-layer filter optimized by machine learning, and determines the optimal filtering path through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter, and a neural network filter, and specifically includes the following steps:

[0101] Step 301: using a Kalman filter combined with an adaptive gain adjustment mechanism to perform initial dynamic state estimation processing on the preliminary purified signal, obtain an initial filtering result, and record the state parameters and uncertainty measurement during the filtering process;

[0102] In this step, the Kalman filter is a recursive solution that is widely used to estimate the state variables of the system, especially suitable for dealing with linear and Gaussian noise environments. The adaptive gain adjustment mechanism allows the Kalman filter to dynamically adjust its gain value according to the current signal characteristics to optimize the estimation effect. The state parameters and uncertainty measures are used to describe the system state and its estimation error, which helps the subsequent steps to better understand the signal characteristics;

[0103] The Kalman filter is applied to perform the initial dynamic state estimation of the preliminary purified signal. In this process, the gain value is dynamically adjusted according to the characteristics of the input signal to ensure the best estimation effect. At the same time, the state parameters and uncertainty measures of each filtering operation are recorded so that the subsequent steps can make full use of this information. The initial filtering result not only reduces the noise component in the signal, but also lays the foundation for the subsequent nonlinear processing;

[0104] When the preliminary cleansed signal is ready, the Kalman filter is called for the initial dynamic state estimation. Assuming that a certain drone is flying steadily, the Kalman filter dynamically adjusts the gain value to optimize the estimation effect based on its speed, direction and other information. The state parameters and uncertainty measures of each filtering operation are recorded, such as the position error range of the drone. These records are crucial for subsequent nonlinear processing because they provide important clues about the state of the drone. The initial filtering results not only make the signal clearer, but also provide a reliable reference for subsequent steps.

[0105] Step 302: Based on the initial filtering result, state parameters and uncertainty measurement, a particle filter is applied in combination with a Bayesian update rule and a resampling strategy to perform nonlinear secondary filtering on the initial filtering result to obtain a secondary filtering result, and the particle distribution, weight and prediction error covariance are simultaneously evaluated;

[0106] In this step, the particle filter is a probabilistic filtering method for nonlinear and non-Gaussian problems, which is particularly suitable for processing signals in complex environments. The Bayesian update rule is used to update the probability distribution of particles based on new observation data. The resampling strategy aims to avoid particle impoverishment and ensure particle diversity. Particle distribution, weights, and prediction error covariance are key indicators for measuring the performance of particle filters, reflecting the reliability and stability of the filtering results;

[0107] Based on the initial filtering results, state parameters and uncertainty measures, a particle filter is applied for nonlinear secondary filtering. The particle filter simulates a large number of possible states (i.e., particles) and continuously adjusts the probability distribution of these particles according to the Bayesian update rule. The resampling strategy ensures that the particle distribution always maintains diversity and prevents particle impoverishment. Finally, the particle distribution, weights, and prediction error covariance are simultaneously evaluated to generate secondary filtering results, which further improves the purity and reliability of the signal;

[0108] After the initial filtering is completed, the particle filter is called for nonlinear secondary filtering. For example, if a drone suddenly changes its flight direction, the Kalman filter may not be able to fully capture this nonlinear change. At this time, the particle filter simulates a large number of possible flight paths (particles) and updates the probability distribution of these paths based on actual observation data to more accurately track the new movements of the drone. The resampling strategy ensures the diversity of particle distribution and maintains good tracking effects even in extreme cases. The final secondary filtering result not only reflects the flight status of the drone more accurately, but also provides high-quality input for subsequent deep learning optimization processing.

[0109] Step 303: using a deep neural network filter, based on the secondary filtering result and particle distribution, using transfer learning technology to initialize network parameters, and introducing an adversarial training mechanism to perform deep learning optimization processing to generate a filtering signal after neural network optimization;

[0110] In this step, the deep neural network filter is an advanced filtering tool implemented by a deep learning model, which is particularly suitable for processing complex signals. Transfer learning technology allows the use of pre-trained models as a starting point to accelerate the training process of new models. The adversarial training mechanism is a reinforcement learning method that improves the robustness and generalization ability of the model through adversarial generative networks (GANs). The neural network optimized filtered signal refers to a more pure and information-rich signal after being processed by a deep learning model;

[0111] Based on the secondary filtering results and particle distribution, deep neural network filters are used for deep learning optimization processing. In order to speed up the training and improve the model performance, the transfer learning technology is used to initialize the network parameters, borrowing the successful experience of previous similar tasks. At the same time, the adversarial training mechanism is introduced to enhance the robustness and generalization ability of the model through the adversarial generative network (GAN). Finally, the filter signal optimized by the generated neural network is significantly improved, which significantly improves the signal quality and detection accuracy;

[0112] After completing the nonlinear secondary filtering, the deep neural network filter is called for deep learning optimization processing. Suppose a drone uses a new stealth technology, which is difficult to effectively identify with traditional filtering methods. Using the pre-trained deep neural network model, it can quickly adapt to new tasks through transfer learning technology. The adversarial training mechanism ensures that the model can cope with various complex situations, such as drones disguised as birds or other objects. The resulting neural network optimized filter signal is not only purer, but also can reveal the subtle features of the drone, such as special flight posture or surface material, providing strong support for subsequent decision-making.

[0113] Step 304: combining the initial filtering result, the secondary filtering result and the filtering signal after the neural network optimization, evaluating the performance of different filters through the model selection algorithm under the integrated learning framework, evaluating the confidence, relevance and diversity of the output of each layer of filters, determining the optimal filtering path, and dynamically adjusting the weight distribution of each layer of filters by using the reinforcement learning algorithm, integrating the output of each layer of filters, and obtaining the comprehensive filtering signal after the optimized path;

[0114] In this step, the ensemble learning framework refers to a method that combines multiple different types of models or algorithms to improve the overall performance. The model selection algorithm is used to evaluate the performance of different filters and select the best combination. Confidence, relevance, and diversity are key indicators for evaluating the output of each layer of filters, reflecting the reliability and complementarity of the filtering results. The reinforcement learning algorithm is a method of learning through trial and error, which is used to dynamically adjust the weight distribution of each layer of filters to achieve optimal performance;

[0115] Combining the results of the initial filtering, the secondary filtering, and the filtered signal after neural network optimization, the performance of different filters is evaluated through the model selection algorithm under the integrated learning framework. The model selection algorithm not only considers the independent performance of each layer of filters, but also evaluates the synergy between them to determine the optimal filtering path. At the same time, the reinforcement learning algorithm is used to dynamically adjust the weight distribution of each layer of filters to ensure that each layer can achieve maximum efficiency. Finally, the output of each layer of filters is integrated to generate a comprehensive filtered signal after the optimized path, which significantly improves the overall effect of signal processing;

[0116] After completing three different filtering processes, the performance of each layer of filters is evaluated through an integrated learning framework. For example, the Kalman filter performs well in processing linear changes, the particle filter is good at capturing nonlinear dynamics, and the deep neural network filter is able to identify complex features. The model selection algorithm comprehensively evaluates the performance of these filters and determines the optimal filtering path. At the same time, the weight distribution of each layer of filters is dynamically adjusted through the reinforcement learning algorithm to ensure the best effect in different environments. The final generated integrated filtered signal after the optimized path not only combines the advantages of each layer of filters, but also achieves higher signal quality and detection accuracy.

[0117] Step 305: According to the comprehensive filtered signal after the optimized path, a context-aware mechanism is introduced, and real-time environmental parameters, historical flight data, and dynamic target behavior patterns are used as auxiliary inputs. The relationship between sensors is modeled using a graph neural network to capture spatiotemporal dependencies and generate high-quality detection signals.

[0118] In this step, the context-aware mechanism refers to the ability to make intelligent decisions based on the current environment and other relevant information. Real-time environmental parameters, historical flight data, and dynamic target behavior patterns provide comprehensive information about the drone and its surroundings, which helps improve detection accuracy. Graph neural networks are a type of neural network specifically designed to process graph-structured data, and are particularly suitable for modeling complex relationships between sensors. Spatiotemporal dependency refers to the interrelationship of events in time and space, which is very important for understanding drone behavior;

[0119] Based on the comprehensive filtered signal after the optimized path, the context perception mechanism is introduced, and the real-time environmental parameters, historical flight data and dynamic target behavior patterns are used as auxiliary inputs to further enrich the signal information. At the same time, the relationship between sensors is modeled using graph neural networks to capture spatiotemporal dependencies, ensuring that the generated high-quality detection signals are not only accurate, but also highly coherent and consistent. This method not only improves detection accuracy, but also enhances adaptability and flexibility;

[0120] When the integrated filtered signal after the optimized path is ready, the context-aware mechanism is introduced to further improve the detection signal by combining real-time environmental parameters, historical flight data, and dynamic target behavior patterns. For example, if a drone is approaching a tall building, the detection angle and sensitivity are adjusted considering the height and shape of the building, as well as the wind speed and direction at the time. At the same time, the graph neural network analyzes the relationship between each sensor to ensure that all information is fully utilized. The high-quality detection signal finally generated not only accurately reflects the current position and status of the drone, but also foresees its future movements, providing a solid guarantee for urban safety.

[0121] Based on this, the present invention provides a specific embodiment, wherein step 304 combines the initial filtering result, the secondary filtering result and the filtering signal after the neural network optimization, evaluates the performance of different filters through the model selection algorithm under the integrated learning framework, evaluates the confidence, relevance and diversity of the output of each layer of filters, determines the optimal filtering path, and dynamically adjusts the weight distribution of each layer of filters using the reinforcement learning algorithm, integrates the output of each layer of filters, and obtains the comprehensive filtering signal after the optimized path, specifically including the following steps:

[0122] Step 401: using a model selection algorithm under an integrated learning framework, based on the primary filtering result, the secondary filtering result, and the filtering signal after neural network optimization, the confidence, relevance, and diversity of the output of each layer of filters are evaluated and processed to obtain a preliminary performance evaluation report;

[0123] In this step, the ensemble learning framework is a method that combines multiple different types of models or algorithms to improve the overall performance. The model selection algorithm is used to evaluate the performance of different filters and select the best combination. Confidence, relevance, and diversity are key indicators for evaluating the output of each layer of filters, reflecting the reliability and complementarity of the filtering results;

[0124] Through the model selection algorithm under the integrated learning framework, the confidence, relevance and diversity of the output of each layer of filters are evaluated based on the initial filtering results, the secondary filtering results and the filtered signals after neural network optimization. This algorithm not only considers the independent performance of each layer of filters, but also evaluates the synergy between them, and finally generates a preliminary performance evaluation report to provide a scientific basis for subsequent steps;

[0125] In specific applications, after completing three different filtering processes, the model selection algorithm is called to evaluate the performance of each layer of filters. For example, the Kalman filter performs well in processing linear changes, the particle filter is good at capturing nonlinear dynamics, and the deep neural network filter can recognize complex features. The algorithm comprehensively evaluates the performance of these filters, determines the advantages and disadvantages of each filter under specific conditions, and generates a detailed preliminary performance evaluation report as the basis for the next adjustment.

[0126] Step 402: According to the preliminary performance evaluation report, the Bayesian optimization algorithm is used to dynamically adjust the weight distribution of each layer of filters to generate a set of filtered signals after weight adjustment;

[0127] In this step, the Bayesian optimization algorithm is a method of learning through trial and error, which is used to dynamically adjust the weight distribution of filters in each layer to achieve optimal performance. The set of filtered signals after weight adjustment refers to the new set of filtered signals after weight adjustment, ensuring that each layer can achieve maximum performance;

[0128] According to the preliminary performance evaluation report, the Bayesian optimization algorithm is used to dynamically adjust the weight distribution of each layer of filters. In this way, it can ensure that each layer of filters can play the best performance in the fusion process, avoid the limitations that may be brought by a single filter, and improve the overall filtering effect. Finally, a set of filtered signals with adjusted weights is generated, providing a high-quality data foundation for subsequent steps;

[0129] Based on the preliminary performance evaluation report, the Bayesian optimization algorithm is applied to adjust the weight distribution of each layer of filters. For example, if the Kalman filter performs better than other filters in some cases, its weight is appropriately increased; conversely, for filters that do not perform as expected, their weight is reduced. Through this dynamic adjustment, it is ensured that all filters can perform optimally in different environments, thereby generating a set of filtered signals with adjusted weights, further improving the quality and reliability of the filtered signals.

[0130] Step 403: Based on the weighted filtered signal set, an adaptive feedback mechanism is constructed through a reinforcement learning algorithm to monitor and adjust the filtering path in real time to obtain an optimized path solution;

[0131] In this step, the reinforcement learning algorithm is a method of learning by trial and error, which is used to dynamically adjust the weight distribution of each layer of filters to achieve optimal performance. The adaptive feedback mechanism allows the system to make intelligent decisions based on the current environment and other relevant information. The optimization path solution refers to the solution that adjusts the filter path to achieve the best filtering effect;

[0132] Based on the set of filtered signals after adjusting the weights, an adaptive feedback mechanism is constructed through a reinforcement learning algorithm to monitor and adjust the filtering path in real time. This method realizes the dynamic adjustment of the filtering path, enabling the system to flexibly respond to changes in the detection environment and maintain efficient filtering performance. Ultimately, an optimized path solution is obtained to ensure that each filtering operation can make the best decision based on the latest intelligence;

[0133] When the weighted filtered signal set is ready, the adaptive feedback mechanism is activated to monitor and adjust the filtering path in real time through the reinforcement learning algorithm. For example, if there is a lot of random noise in a certain signal segment, the system will automatically adjust the filtering path, relying more on the powerful denoising ability of the particle filter. At the same time, based on the real-time monitoring results, the filtering path is continuously optimized to ensure that each filtering operation can make the best decision based on the latest intelligence, and finally generate an optimized path plan, which significantly improves the filtering effect.

[0134] Step 404: Integrate the outputs of the filters at each layer in the optimized path solution to form a comprehensive filtering signal, and introduce a context perception module to take environmental parameters, historical flight data and dynamic target behavior patterns as auxiliary inputs to obtain a comprehensive filtering signal after the optimized path;

[0135] In this step, the integrated filter signal refers to the high-quality filter signal formed by integrating the outputs of filters at each layer. The context-aware module refers to the ability to make intelligent decisions based on the current environment and other relevant information. The integrated filter signal after the optimized path is the final filter signal generated based on the optimized path plan, combined with environmental parameters, historical flight data and dynamic target behavior patterns;

[0136] The outputs of each layer of filters in the optimized path plan are integrated to form a comprehensive filter signal, and a context-aware module is introduced to take environmental parameters, historical flight data, and dynamic target behavior patterns as auxiliary inputs. This approach not only enhances the adaptability and accuracy of the comprehensive filter signal, but also ensures the coherence and consistency from preliminary performance evaluation to high-quality detection signal generation;

[0137] On the basis of the optimized path plan, the outputs of each layer of filters are integrated to form a comprehensive filtered signal. In order to further enhance the accuracy and adaptability of the signal, a context-aware module is introduced to take environmental parameters (such as temperature, humidity, wind speed), historical flight data (such as the historical flight trajectory of the drone) and dynamic target behavior patterns (such as the flight attitude and speed changes of the drone) as auxiliary inputs. For example, if a drone is approaching a tall building, the detection angle and sensitivity are adjusted considering the height and shape of the building, as well as the wind speed and direction at the time. The final generated comprehensive filtered signal after the optimized path not only accurately reflects the current position and status of the drone, but also foresees its future movements, providing a solid guarantee for subsequent decision-making.

[0138] Based on this, the present invention provides a specific embodiment, wherein the step 102 uses a deep learning model to analyze the comprehensive perception information set, and generates dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions, and specifically includes the following steps:

[0139] Step 501: Based on the comprehensive perception information set, use a pre-trained deep learning model to perform preliminary feature extraction processing on the sensor data to obtain a multi-dimensional feature representation;

[0140] In this step, the comprehensive perception information set refers to the data set collected by multiple sensors, which contains various information about the target. The pre-trained deep learning model refers to a deep learning model that has been trained with a large amount of data and has a strong feature extraction capability. The multi-dimensional feature representation refers to the data features extracted by the model, which contains information in multiple dimensions such as time and space;

[0141] Based on the comprehensive perception information set, the pre-trained deep learning model is used to perform preliminary feature extraction processing on the sensor data. Through multi-level abstraction and transformation, the model extracts multi-dimensional features such as time and space of the data to form a multi-dimensional feature representation. This not only simplifies the original data, but also provides a richer information basis for subsequent analysis;

[0142] When monitoring drone activities in an urban environment, it is assumed that a comprehensive set of perception information from multiple sensors such as radar, camera, infrared imager, etc. has been collected. Using the pre-trained deep learning model, preliminary feature extraction processing is performed on this data. For example, the model can extract the speed and altitude change of the drone from the radar reflection signal, the shape and color features of the drone from the camera image, and the temperature distribution of the drone from the infrared imager data. These multi-dimensional feature representations provide a solid foundation for subsequent in-depth analysis.

[0143] Step 502: Based on the multi-dimensional feature representation, a time series analysis algorithm is applied to identify and extract the historical flight path, speed change, and altitude adjustment of the UAV to obtain flight characteristic data;

[0144] In this step, the time series analysis algorithm is an algorithm specifically used to analyze time series data, which is suitable for processing data characteristics that change over time. Flight characteristic data refers to the behavior pattern of the drone revealed by analyzing the drone's historical flight path, speed changes, and altitude adjustments;

[0145] Based on the multi-dimensional feature representation, the time series analysis algorithm is applied to identify and extract the historical flight path, speed changes, and altitude adjustments of the drone. The algorithm analyzes the time series data to reveal the behavior patterns of the drone and generate flight characteristic data that describes its movement laws, which not only enhances the understanding of the drone's movement, but also provides important reference information for subsequent analysis;

[0146] After completing the multi-dimensional feature representation, the time series analysis algorithm is applied to identify and extract the historical flight path, speed change, and altitude adjustment of the drone. For example, the algorithm can identify whether the drone's flight path is stable from radar data, extract the drone's speed change trend from camera images, and analyze the drone's altitude adjustment from infrared imager data. These flight characteristic data provide rich information for subsequent behavior pattern analysis, which helps to more accurately understand the movement patterns of drones.

[0147] Step 503: combining the flight characteristic data, using an attitude estimation algorithm, analyzing and processing the acceleration parameters and flight attitude of the UAV, and generating a preliminary draft of the behavior pattern;

[0148] In this step, the attitude estimation algorithm is an algorithm used to estimate the flight attitude of the drone (such as pitch angle, roll angle, yaw angle). The preliminary behavior pattern refers to the behavior pattern description initially generated by analyzing the acceleration parameters and flight attitude of the drone;

[0149] Combined with flight characteristic data, the attitude estimation algorithm is used to analyze and process the acceleration parameters and flight attitude of the drone. The algorithm analyzes the changes in the acceleration of the drone, reveals the changing rules of its flight attitude, and generates a preliminary draft of the behavior pattern. This not only enhances the understanding of the drone's attitude, but also provides important clues for subsequent environmental adaptability adjustments;

[0150] Based on the flight characteristic data, the attitude estimation algorithm is applied to analyze and process the acceleration parameters and flight attitude of the drone. For example, the algorithm can estimate the pitch angle, roll angle and yaw angle of the drone based on the acceleration changes in the radar data. This information not only reveals the law of attitude changes of the drone, but also provides valuable reference for subsequent behavior pattern analysis. The generated draft of the behavior pattern lays a solid foundation for subsequent environmental adaptability adjustments.

[0151] Step 504: using the environment interaction analysis module, combining the draft behavior pattern with the real-time environment interaction information, performing environment adaptability adjustment processing, and generating behavior pattern data after environment adaptation;

[0152] In this step, the environment interaction analysis module is a tool specifically used to analyze the interaction between the drone and the surrounding environment. The behavior pattern data after environmental adaptation refers to the data generated by adjusting the behavior pattern by analyzing the interaction information between the drone and the environment;

[0153] The environment interaction analysis module is used to combine the draft behavior pattern with the real-time environment interaction information to adjust the environment adaptability. The module adjusts the behavior pattern by analyzing the interaction information between the drone and the environment, such as wind speed, temperature gradient, etc., and generates the behavior pattern data after environmental adaptation, which not only enhances the accuracy of the behavior pattern, but also improves the adaptability and flexibility of the system;

[0154] After the first draft of the behavior pattern is completed, the environmental interaction analysis module is applied to combine the first draft with the real-time environmental interaction information to perform environmental adaptability adjustment processing. For example, if a drone is flying in windy weather, the module will adjust the drone's attitude and speed prediction according to the wind speed and direction. Through this environmental adaptability adjustment, the generated environmentally adapted behavior pattern data not only more accurately reflects the actual flight status of the drone, but also foresees its future movements, providing strong support for subsequent decision-making.

[0155] Step 505: Integrate the behavior pattern data after the environmental adaptation, the draft of the behavior pattern, the flight characteristic data and the multi-dimensional feature representation, and perform comprehensive evaluation through the high-level abstract layer of the deep learning model to obtain dynamic target behavior pattern data;

[0156] In this step, dynamic target behavior pattern data refers to the integration of data from multiple sources to reveal the behavior patterns of drones in time and space. The high-level abstraction layer of the deep learning model is used to perform high-level abstraction and transformation on the data to reveal deeper features and relationships;

[0157] Integrate the behavioral pattern data after environmental adaptation, the draft of the behavioral pattern, the flight characteristics data and the multi-dimensional feature representation, and conduct a comprehensive evaluation through the high-level abstraction layer of the deep learning model. The model reveals the behavior patterns of the drone in time and space through high-level abstraction and transformation of these data, and generates dynamic target behavior pattern data. This not only enhances the understanding of drone behavior, but also provides more detailed information for subsequent analysis;

[0158] The behavioral pattern data after environmental adaptation, the draft of the behavioral pattern, the flight characteristics data and the multi-dimensional feature representation are integrated and comprehensively evaluated through the high-level abstraction layer of the deep learning model. For example, the model can combine the historical flight path, speed change, altitude adjustment, attitude change and environmental interaction information of the UAV to generate a comprehensive description of the dynamic target behavior pattern. This description not only reveals the behavior rules of the UAV in different time periods, but also foresees its possible future movements, providing strong support for subsequent decision-making.

[0159] Based on this, the present invention provides a specific embodiment, wherein step 104 combines a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, uses a predictive analysis algorithm to predict the flight path of the optimized detection signal, and generates predicted flight path data, specifically comprising the following steps:

[0160] Step 601: using the geographic information system and real-time meteorological data, the optimized detection signal is subjected to environmental factor fusion processing to obtain an environmentally fused detection signal;

[0161] In this step, the Geographic Information System (GIS) is an integrated tool for storing, managing, analyzing and displaying all forms of geographic data. Real-time meteorological data provides information on current weather conditions, such as temperature, humidity, wind speed, etc., which is very important for predicting the flight path of drones. The detection signal after environmental fusion refers to the signal generated by processing the optimized detection signal by fusing the geographic information system and real-time meteorological data;

[0162] Using geographic information systems and real-time meteorological data, the optimized detection signal is processed by environmental factors. Not only the current position and speed of the drone is taken into account, but also the impact of external environmental factors on the flight path is fully considered to generate environmental fused detection signals. This not only improves the accuracy of the detection signal, but also enhances the adaptability and flexibility of the system.

[0163] In an urban monitoring environment, when the optimized detection signal is ready, the geographic information system and real-time meteorological data are called to perform environmental factor fusion processing. For example, the system will adjust the detection signal by combining map information (such as roads, bridges, and high-rise building locations) with the latest acquired meteorological data (such as wind speed and direction). Through this environmental fusion processing, the generated environmental fusion detection signal not only reflects the current position and status of the drone more clearly, but also foresees its future movements, providing a solid guarantee for subsequent decision-making.

[0164] Step 602: Based on the detection signal after the environment fusion, combined with the three-dimensional terrain model, the spatial feature matching processing is performed through the terrain adaptability adjustment algorithm to generate a detection signal after terrain matching; In this step, the three-dimensional terrain model depicts the ground undulations and building layout in detail, which helps to simulate the physical obstacles that the drone may encounter. The terrain adaptability adjustment algorithm is used to adjust the spatial characteristics of the detection signal according to the three-dimensional terrain model to ensure the accuracy and consistency of the detection signal. The detection signal after terrain matching refers to the signal generated after processing by the terrain adaptability adjustment algorithm;

[0165] Based on the detection signal after environmental fusion, combined with the three-dimensional terrain model, spatial feature matching processing is performed through the terrain adaptability adjustment algorithm. The algorithm compares the data at different time periods and locations to ensure that the spatial characteristics of the detection signal match the actual terrain and generate a detection signal after terrain matching. It not only verifies the accuracy of the detection signal, but also discovers potential abnormalities, improving the reliability and adaptability of the system;

[0166] When the detection signal after environmental fusion is ready, it is combined with the three-dimensional terrain model and processed by the terrain adaptive adjustment algorithm for spatial feature matching. For example, if a drone is approaching a tall building, the algorithm will adjust the spatial features of the detection signal according to the three-dimensional terrain model to ensure that the detection signal matches the actual terrain. Through this terrain matching process, the generated terrain-matched detection signal not only more accurately reflects the current position and status of the drone, but also foresees its future movements, providing a solid guarantee for subsequent decision-making.

[0167] Step 603: According to the signal after terrain matching, using the building and infrastructure information in the urban structure database, obstacle avoidance and path optimization processing are performed to obtain a preliminary flight path prediction;

[0168] In this step, the urban structure database contains specific information about urban infrastructure, such as roads, bridges, high-rise buildings, etc. Obstacle avoidance and path optimization processing refers to adjusting the flight path to avoid obstacles and optimize the path by analyzing the relationship between the drone and the urban structure. The preliminary flight path prediction refers to the drone flight path prediction generated by the above processing;

[0169] According to the signal after terrain matching, the building and infrastructure information in the urban structure database is applied to perform obstacle avoidance and path optimization. Not only the current position and speed of the drone is taken into account, but also the impact of the urban structure on the flight path is fully considered to generate a preliminary flight path prediction. This not only improves the accuracy of the flight path prediction, but also enhances the adaptability and flexibility of the system;

[0170] When the terrain-matched signal is ready, the building and infrastructure information in the urban structure database is applied to perform obstacle avoidance and path optimization. For example, if a drone is approaching a tall building, the system will adjust the flight path based on the urban structure database to ensure that the drone avoids the tall building and other obstacles. Through this obstacle avoidance and path optimization processing, the generated preliminary flight path prediction not only more accurately reflects the current position and status of the drone, but also foresees its future movements, providing a solid guarantee for subsequent decision-making.

[0171] Step 604: Combine the preliminary flight path prediction, use the predictive analysis algorithm, integrate the historical flight data and the trend of the current detection signal, perform dynamic prediction processing of the flight path, and generate a dynamic flight path prediction result;

[0172] In this step, the predictive analysis algorithm is an algorithm used to predict future events and is suitable for processing data features that change over time. The dynamic flight path prediction result refers to the UAV flight path prediction generated by the predictive analysis algorithm, combining historical flight data and the trend of the current detection signal;

[0173] Combined with the preliminary flight path prediction, the predictive analysis algorithm is used to integrate the historical flight data and the trend of the current detection signal to perform dynamic prediction processing of the flight path. The algorithm analyzes the time series data to predict the possible future flight path of the UAV and generate dynamic flight path prediction results. This not only improves the accuracy of flight path prediction, but also enhances the adaptability and flexibility of the system;

[0174] When the preliminary flight path prediction is ready, the predictive analysis algorithm is used to combine historical flight data and the trend of the current detection signal to perform dynamic prediction processing of the flight path. For example, if a drone often performs missions in the same area and the flight path is similar each time, the algorithm will predict the possible future flight path of the drone based on the historical flight data and the trend of the current detection signal. Through this dynamic prediction processing, the generated dynamic flight path prediction results not only more accurately reflect the current position and status of the drone, but also foresee its future movements, providing a solid guarantee for subsequent decision-making.

[0175] Step 605: Integrate the dynamic flight path prediction result, the detection signal after environment fusion and the preliminary flight path prediction, and generate predicted flight path data through multi-source data fusion technology;

[0176] In this step, multi-source data fusion technology refers to the integration of data from multiple different sources to improve overall performance. Predicted flight path data refers to the data generated after the dynamic flight path prediction results, the detection signals after environmental fusion, and the preliminary flight path prediction through multi-source data fusion technology;

[0177] Integrate dynamic flight path prediction results, detection signals after environmental fusion, and preliminary flight path prediction, and generate predicted flight path data through multi-source data fusion technology. This not only improves the accuracy of flight path prediction, but also enhances the adaptability and flexibility of the system. The predicted flight path data finally generated provides a solid foundation for subsequent decision-making;

[0178] When the dynamic flight path prediction results, the detection signals after environmental fusion, and the preliminary flight path prediction are ready, these data are integrated through multi-source data fusion technology to generate predicted flight path data. For example, if a drone is approaching a tall building, the system will adjust the flight path based on the dynamic flight path prediction results, the detection signals after environmental fusion, and the preliminary flight path prediction to ensure that the drone avoids tall buildings and other obstacles. Through this multi-source data fusion, the generated predicted flight path data not only more accurately reflects the current position and status of the drone, but also foresees its future movements, providing a solid guarantee for subsequent decision-making.

[0179] Based on this, the present invention provides a specific embodiment, in which step 105 uses a distributed edge computing framework to process a comprehensive perception information set, optimizes a deep learning model through a federated learning mechanism, generates an optimized detection strategy, and generates an updated detection process based on the predicted flight path data and the optimized detection strategy, specifically including the following steps:

[0180] Step 701: using a distributed edge computing framework, locally preprocessing and preliminarily analyzing the comprehensive sensing information set to obtain edge processing results;

[0181] In this step, the distributed edge computing framework is a computing architecture that allows data processing to be performed close to the data source (i.e., edge nodes), reducing data transmission latency and bandwidth requirements. The comprehensive perception information set refers to a set of data collected by multiple sensors, which contains various information about the target. The edge processing result refers to the data generated after preprocessing and preliminary analysis are completed on each edge node;

[0182] Using the distributed edge computing framework, local preprocessing and preliminary analysis tasks are deployed on each edge node. These nodes independently preprocess data from multiple sensors such as radar, camera, infrared imager, etc., such as denoising, feature extraction, etc., and perform preliminary analysis, such as identifying potential targets or behavior patterns. Finally, edge processing results are generated to provide a high-quality data foundation for subsequent steps;

[0183] When monitoring drone activity in an urban environment, it is assumed that a comprehensive set of perception information from multiple locations has been collected. Each edge node performs local preprocessing tasks such as filtering to remove noise, compressing data volume, and preliminary analysis tasks such as identifying the presence of drones and their approximate flight direction. After this series of processing, each node generates edge processing results, ensuring the quality and efficiency of subsequent deep learning model training.

[0184] Step 702: Based on the edge processing result, combined with the federated learning mechanism, the deep learning model is independently trained on each edge node to obtain a trained deep training model, and the trained deep training model is updated and uploaded to the central server for aggregation processing to generate a globally optimized deep learning model;

[0185] In this step, the federated learning mechanism is a distributed machine learning method that allows multiple participants to jointly train models without sharing original data, which protects privacy and improves model performance. The globally optimized deep learning model refers to the optimal model generated by aggregating the model updates uploaded by each edge node on the central server through the federated learning mechanism;

[0186] Based on the edge processing results, combined with the federated learning mechanism, the deep learning model is trained independently on each edge node. Each node trains the model based on local data and uploads the trained model updates to the central server. The central server receives updates from all nodes and aggregates them to generate a globally optimized deep learning model. The model is then redeployed to each edge node for real-time processing of newly received sensor data;

[0187] When the edge processing results are ready, the federated learning mechanism is started to train the deep learning model independently on each edge node. For example, each node will train a small convolutional neural network (CNN) based on local sensor data to identify the characteristics of drones. After training, the node will upload the model parameter updates to the central server. The central server summarizes the updates of all nodes and generates a globally optimized deep learning model through an aggregation algorithm. This model not only integrates the data characteristics of each location, but also maintains data privacy, and is eventually redeployed to each edge node for more accurate drone detection.

[0188] Step 703: applying the globally optimized deep learning model to each edge node, performing intelligent analysis and processing on the received sensor data, and generating an optimized detection strategy;

[0189] In this step, the optimized detection strategy refers to the decision rules or action guidelines generated by intelligently analyzing and processing sensor data through the application of a global optimized deep learning model. This includes specific measures such as how to adjust the monitoring angle, increase the monitoring frequency, or trigger an alarm;

[0190] Based on the globally optimized deep learning model, it is applied to each edge node to perform intelligent analysis and processing on the received sensor data. Through multi-level abstraction and transformation, the model extracts the multi-dimensional features of the data such as time and space, and generates data describing the behavior patterns of drones. Based on this data, the system formulates an optimized detection strategy to ensure that each detection can make the best decision based on the latest intelligence;

[0191] When the globally optimized deep learning model is redeployed to each edge node, these nodes begin to use the new model to intelligently analyze and process the received sensor data. For example, if a drone is approaching a tall building, the model can predict the drone's behavior pattern based on real-time data and recommend appropriate detection strategies, such as adjusting the monitoring angle or increasing the monitoring frequency. In this way, the system can more accurately capture the dynamics of the drone and improve the accuracy and response speed of detection.

[0192] Step 704: combining the predicted flight path data with the optimized detection strategy, using a decision fusion algorithm to integrate multi-source information, formulating a specific action plan, and obtaining an initial detection process;

[0193] In this step, the decision fusion algorithm is a method to integrate information from different sources to improve overall performance. The initial detection process refers to the specific action plan generated by integrating the predicted flight path data with the optimized detection strategy through the decision fusion algorithm to guide the next operation of the system;

[0194] Combine the predicted flight path data with the optimized detection strategy, and use the decision fusion algorithm to integrate multi-source information. The algorithm comprehensively considers the historical flight path of the drone, the trend of the current detection signal, and environmental factors, formulates a specific action plan, and obtains the initial detection process. This not only improves the accuracy of detection, but also enhances the adaptability and flexibility of the system;

[0195] When the predicted flight path data and optimized detection strategy are ready, the decision fusion algorithm is started to integrate this information. For example, if the prediction shows that a drone will fly to the city center, the algorithm will recommend a specific action plan based on historical flight data and the trend of the current detection signal, such as notifying monitoring stations near the city center in advance to strengthen monitoring efforts and prepare emergency measures. In this way, the system can respond to drone activities more effectively and ensure that each detection can make the best decision based on the latest intelligence, thereby improving the response speed and accuracy of the entire system.

[0196] Step 705: Based on the initial detection process, a feedback adjustment mechanism is introduced to dynamically modify the detection strategy and the detection process according to the actual execution situation to obtain an updated detection process;

[0197] In this step, the feedback adjustment mechanism refers to a method by which the system continuously adjusts its own strategies and processes according to the actual execution situation to achieve optimal performance. The updated detection process refers to the latest version generated by dynamically modifying the detection strategy and detection process through the feedback adjustment mechanism;

[0198] Based on the initial detection process, a feedback adjustment mechanism is introduced to dynamically modify the detection strategy and detection process according to the actual execution situation. The system continuously monitors the detection effect and adjusts the strategy and process according to the actual situation to ensure that each detection can make the best decision based on the latest intelligence. Finally, an updated detection process is generated, which significantly improves the response speed and accuracy of the system;

[0199] When the initial detection process begins, the system introduces a feedback adjustment mechanism to make dynamic corrections based on the actual execution situation. For example, if a detection fails to successfully capture the expected drone, the system will analyze the cause of the failure, which may be due to changes in wind speed that affect the flight path prediction. Based on this, the system will adjust the detection strategy, such as increasing the monitoring frequency of a specific area or improving the flight path prediction algorithm. Through this dynamic correction, the system continuously optimizes its own detection strategy and process to ensure that each detection can make the best decision based on the latest intelligence, thereby improving the response speed and accuracy of the entire system.

[0200] Figure 2 A schematic diagram of the structure of a drone detection system with multiple sensor fusion is provided in an embodiment of the present invention, such as Figure 2 As shown, the system includes:

[0201] The receiving module 21 is used to receive the data streams of each sensor, synchronize and associate the data streams of each sensor in the time domain and the spatial domain, and obtain a comprehensive perception information set;

[0202] An analysis module 22 is used to analyze the comprehensive perception information set using a deep learning model, and generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions;

[0203] An optimization module 23 is used to optimize the signal quality based on the dynamic target behavior pattern data by combining adaptive noise cancellation technology with a machine learning optimized filter, and to use spectrum analysis technology to assist in identification, so as to generate an optimized detection signal;

[0204] A prediction module 24 is used to combine a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, and use a predictive analysis algorithm to predict the flight path of the optimized detection signal to generate predicted flight path data;

[0205] The generation module 25 is used to process the comprehensive perception information set using the distributed edge computing framework, optimize the deep learning model through the federated learning mechanism, generate an optimized detection strategy, and generate an updated detection process based on the predicted flight path data and the optimized detection strategy.

[0206] Figure 2 The multi-sensor fusion drone detection system can perform Figure 1 The implementation principle and technical effect of the drone detection method with multiple sensor fusion described in the illustrated embodiment will not be described in detail. The specific manner in which each module and unit performs operations in the drone detection system with multiple sensor fusion in the above embodiment has been described in detail in the embodiment of the method, and will not be elaborated here.

[0207] In one possible design, Figure 2 A drone detection system with multiple sensor fusion in the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32;

[0208] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0209] The processing component 32 is used to receive the data streams of each sensor, synchronize and associate the data streams of each sensor in the time domain and the spatial domain, and obtain a comprehensive perception information set; use a deep learning model to analyze the comprehensive perception information set, and generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes and environmental interactions; based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with machine learning optimized filters is used to optimize signal quality, and spectral analysis technology is used to assist in identification to generate an optimized detection signal; in combination with a geographic information system, real-time meteorological data, a three-dimensional terrain model and an urban structure database, a predictive analysis algorithm is used to predict the flight path of the optimized detection signal to generate predicted flight path data; a distributed edge computing framework is used to process the comprehensive perception information set, and a deep learning model is optimized through a federated learning mechanism to generate an optimized detection strategy, and an updated detection process is generated based on the predicted flight path data and the optimized detection strategy.

[0210] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0211] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0212] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0213] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0214] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0215] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0216] The embodiment of the present invention further provides a computer storage medium storing a computer program, which can achieve the above-mentioned Figure 1 A method for detecting drones by integrating multiple sensors according to the illustrated embodiment.

[0217] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0218] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0219] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0220] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A drone detection method using multiple sensor fusions, characterized in that: include: Receiving data streams from various sensors, synchronizing and correlating the data streams from various sensors in time domain and space domain, and obtaining a comprehensive perception information set; Analyzing the comprehensive perception information set using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions; Based on the dynamic target behavior pattern data, a method combining adaptive noise cancellation technology with a machine learning optimized filter is used to optimize signal quality, and spectrum analysis technology is used to assist in identification to generate an optimized detection signal; In combination with geographic information systems, real-time meteorological data, three-dimensional terrain models, and urban structure databases, a predictive analysis algorithm is used to predict the flight path of the optimized detection signal to generate predicted flight path data; A distributed edge computing framework is used to process the comprehensive perception information set, and a deep learning model is optimized through a federated learning mechanism to generate an optimized detection strategy. Based on the predicted flight path data and the optimized detection strategy, an updated detection process is generated.

2. The method according to claim 1, characterized in that Based on the dynamic target behavior pattern data, the signal quality is optimized by combining adaptive noise cancellation technology with machine learning optimized filters, and spectral analysis technology is used to assist in identification to generate an optimized detection signal, including: Adaptive noise cancellation technology is used to pre-process the original sensor signal in combination with real-time environmental parameters to obtain a preliminary purified signal; According to the preliminary purified signal, filtering is performed through a multi-layer filter optimized by machine learning, and an optimal filtering path is determined through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter, and a neural network filter; Utilize spectrum analysis technology to assist in identifying the high-quality detection signal, analyze the specific electromagnetic spectrum characteristics emitted by the drone, and combine historical flight records and dynamic target behavior pattern data for cross-verification to generate a detection signal with material characteristics and behavior features; Based on the detection signal with material properties and behavioral characteristics, the spatiotemporal correlation analysis algorithm is used to integrate the preliminary purification signal and the high-quality detection signal, evaluate the continuity and consistency of the UAV in different time and space dimensions, and obtain the optimized detection signal.

3. The method according to claim 2, characterized in that According to the preliminary purified signal, filtering is performed through a multi-layer filter optimized by machine learning, and an optimal filtering path is determined through a model selection algorithm to generate a high-quality detection signal, wherein the multi-layer filter includes a Kalman filter, a particle filter, and a neural network filter, including: The Kalman filter is combined with an adaptive gain adjustment mechanism to perform the initial dynamic state estimation processing on the preliminary purified signal, obtain the initial filtering result, and record the state parameters and uncertainty measurement during the filtering process; Based on the initial filtering results, state parameters and uncertainty measures, a particle filter is applied in combination with a Bayesian update rule and a resampling strategy to perform nonlinear secondary filtering on the initial filtering results to obtain secondary filtering results, and particle distribution, weights and prediction error covariance are simultaneously evaluated; Using a deep neural network filter, based on the secondary filtering results and particle distribution, using transfer learning technology to initialize network parameters, and introducing an adversarial training mechanism to perform deep learning optimization processing to generate a filtering signal optimized by the neural network; Combining the initial filtering results, the secondary filtering results and the filtering signal after neural network optimization, the performance of different filters is evaluated by the model selection algorithm under the integrated learning framework, the confidence, relevance and diversity of the output of each layer of filters are evaluated, and the optimal filtering path is determined. At the same time, the weight distribution of each layer of filters is dynamically adjusted by the reinforcement learning algorithm, and the output of each layer of filters is integrated to obtain the comprehensive filtering signal after the optimized path; According to the comprehensive filtered signal after the optimization path, a context-aware mechanism is introduced, and real-time environmental parameters, historical flight data and dynamic target behavior patterns are used as auxiliary inputs. The relationship between sensors is modeled using graph neural networks to capture spatiotemporal dependencies and generate high-quality detection signals.

4. The method according to claim 3, characterized in that Combining the initial filtering results, the secondary filtering results and the filtering signal after the neural network optimization, the performance of different filters is evaluated by the model selection algorithm under the integrated learning framework, the confidence, relevance and diversity of the output of each layer of filters are evaluated, and the optimal filtering path is determined. At the same time, the weight distribution of each layer of filters is dynamically adjusted by the reinforcement learning algorithm, and the output of each layer of filters is integrated to obtain the comprehensive filtering signal after the optimized path, including: Using the model selection algorithm under the ensemble learning framework, based on the primary filtering results, secondary filtering results and the filtered signals after neural network optimization, the confidence, relevance and diversity of the output of each layer of filters are evaluated and processed to obtain a preliminary performance evaluation report; According to the preliminary performance evaluation report, the Bayesian optimization algorithm is used to dynamically adjust the weight distribution of filters in each layer to generate a set of filtered signals after weight adjustment; Based on the weighted filtered signal set, an adaptive feedback mechanism is constructed through a reinforcement learning algorithm to monitor and adjust the filtering path in real time to obtain an optimized path solution; The outputs of each layer of filters in the optimized path scheme are integrated to form a comprehensive filtering signal, and a context perception module is introduced to take environmental parameters, historical flight data and dynamic target behavior patterns as auxiliary inputs to obtain a comprehensive filtering signal after the optimized path.

5. The method according to claim 1, characterized in that The comprehensive perception information set is analyzed using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions, including: Based on the comprehensive perception information set, a pre-trained deep learning model is used to perform preliminary feature extraction processing on the sensor data to obtain a multi-dimensional feature representation; According to the multi-dimensional feature representation, a time series analysis algorithm is applied to identify and extract the historical flight path, speed change, and altitude adjustment of the UAV to obtain flight characteristic data; In combination with the flight characteristic data, an attitude estimation algorithm is used to analyze and process the acceleration parameters and flight attitude of the UAV to generate a preliminary draft of the behavior pattern; By using the environment interaction analysis module, the draft of the behavior pattern is combined with the real-time environment interaction information, and an environment adaptability adjustment process is performed to generate behavior pattern data after environment adaptation; The behavioral pattern data after environmental adaptation, the draft behavioral pattern, the flight characteristic data and the multi-dimensional feature representation are integrated, and a comprehensive evaluation is performed through the high-level abstract layer of the deep learning model to obtain the dynamic target behavior pattern data.

6. The method according to claim 1, characterized in that Combined with geographic information system, real-time meteorological data, three-dimensional terrain model and urban structure database, a predictive analysis algorithm is used to predict the flight path of the optimized detection signal to generate predicted flight path data, including: Using geographic information system and real-time meteorological data, the optimized detection signal is subjected to environmental factor fusion processing to obtain an environmentally fused detection signal; Based on the detection signal after the environment fusion, combined with the three-dimensional terrain model, spatial feature matching processing is performed through a terrain adaptability adjustment algorithm to generate a detection signal after terrain matching; According to the signal after terrain matching, the building and infrastructure information in the urban structure database is applied to perform obstacle avoidance and path optimization processing to obtain a preliminary flight path prediction; Combined with the preliminary flight path prediction, using a predictive analysis algorithm, integrating historical flight data and the trend of the current detection signal, performing dynamic prediction processing of the flight path, and generating a dynamic flight path prediction result; The dynamic flight path prediction results, the detection signals after environmental fusion and the preliminary flight path prediction are integrated, and the predicted flight path data is generated through multi-source data fusion technology.

7. The method according to claim 1, characterized in that The integrated perception information set is processed by using a distributed edge computing framework, and the deep learning model is optimized by a federated learning mechanism to generate an optimized detection strategy. Based on the predicted flight path data and the optimized detection strategy, an updated detection process is generated, including: Using a distributed edge computing framework, the comprehensive perception information set is locally preprocessed and preliminarily analyzed to obtain an edge processing result; Based on the edge processing results, combined with the federated learning mechanism, the deep learning model is independently trained on each edge node to obtain a trained deep training model, and the trained deep training model is updated and uploaded to the central server for aggregation processing to generate a globally optimized deep learning model; According to the globally optimized deep learning model, it is applied to each edge node to perform intelligent analysis and processing on the received sensor data to generate an optimized detection strategy; Combining the predicted flight path data with the optimized detection strategy, using a decision fusion algorithm to integrate multi-source information, formulate a specific action plan, and obtain an initial detection process; Based on the initial detection process, a feedback adjustment mechanism is introduced to dynamically modify the detection strategy and the detection process according to the actual execution situation to obtain an updated detection process.

8. A multi-sensor fusion drone detection system, characterized in that: include: A receiving module, used to receive the data streams of each sensor, synchronize and associate the data streams of each sensor in the time domain and the spatial domain, and obtain a comprehensive perception information set; An analysis module for analyzing the comprehensive perception information set using a deep learning model to generate dynamic target behavior pattern data by identifying historical flight paths, speed changes, altitude adjustments, acceleration parameters, flight attitudes, and environmental interactions; An optimization module, for optimizing signal quality based on the dynamic target behavior pattern data by combining adaptive noise cancellation technology with a machine learning optimized filter, and using spectrum analysis technology to assist in identification, thereby generating an optimized detection signal; A prediction module, for combining a geographic information system, real-time meteorological data, a three-dimensional terrain model, and an urban structure database, and using a predictive analysis algorithm to predict the flight path of the optimized detection signal to generate predicted flight path data; A generation module is used to process the comprehensive perception information set using a distributed edge computing framework, optimize the deep learning model through a federated learning mechanism, generate an optimized detection strategy, and generate an updated detection process based on the predicted flight path data and the optimized detection strategy.

9. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a drone detection method with multiple sensor fusion as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a drone detection method using multiple sensor fusion as described in any one of claims 1 to 7 is implemented.

Images