Unmanned aerial vehicle detection method and system based on infrared image analysis

Abstract

The application relates to the technical field of unmanned aerial vehicle detection, and discloses an unmanned aerial vehicle detection method and system based on infrared image analysis. The method collects infrared image sequence data within a preset time window; based on the sequence data, environment stability analysis is performed to determine an image heat distribution stability factor, an environment temperature stability factor and a motion characteristic stability factor; fuzzy logic inference rules are applied to infer an infrared response interval in combination with the three factors, and based on the interval, target analysis is performed on the infrared image sequence data according to a preset detection item set to obtain a periodic detection data set; and the set is identified by using an infrared target identifier to obtain an unmanned aerial vehicle detection result. The method can fully consider dynamic changes in the environment by analyzing the environment stability factors, combining fuzzy logic inference to determine the infrared response interval, and then performing target analysis and identification, thereby improving the adaptability and effectiveness of unmanned aerial vehicle detection, and being suitable for unmanned aerial vehicle detection scenes in complex environments.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle detection, in particular to an unmanned aerial vehicle detection method and system based on infrared image analysis. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, its application in civilian, commercial and military fields is increasingly widespread. At the same time, the safety hazards caused by illegal flight of unmanned aerial vehicles are becoming increasingly prominent. For example, the intrusion of unmanned aerial vehicles into the airport clearance area may cause flight delays or even accidents, and the illegal shooting of unmanned aerial vehicles at important places may leak sensitive information. Therefore, accurate and efficient detection of unmanned aerial vehicles has become a problem to be solved.

[0003] Existing unmanned aerial vehicle detection methods are diverse, including radar detection, radio detection, visual detection, etc. Although radar detection has a long detection distance, it is easily affected by weather and has low identification accuracy for small unmanned aerial vehicles, and has a high false alarm rate. Radio detection achieves detection by capturing the communication signals between the unmanned aerial vehicle and the remote controller. However, with the development of unmanned aerial vehicle communication technology, the application of encryption communication and frequency hopping technology has greatly reduced the detection effect of this method, especially in complex electromagnetic environments, its reliability is further reduced.

[0004] Visual detection is concerned due to its low cost and easy deployment, but it is severely dependent on lighting conditions and its detection performance decreases dramatically at night or in bad weather. Infrared images can be imaged without being affected by light, and can be imaged by the temperature difference between the target and the environment, providing a possibility for unmanned aerial vehicle detection at night and in complex environments. However, existing unmanned aerial vehicle detection methods based on infrared images do not fully consider the influence of dynamic changes in the environment on the detection results, such as fluctuations in environmental temperature and changes in background heat distribution, resulting in low detection accuracy and insufficient stability in complex environments, which makes it difficult to meet the actual application requirements. Therefore, how to effectively utilize infrared image information, combined with dynamic characteristics of the environment, to improve the accuracy and stability of unmanned aerial vehicle detection, has become a research hotspot and difficulty in this field. SUMMARY

[0005] The present application aims to provide an unmanned aerial vehicle detection method and system based on infrared image analysis to solve the problems raised in the background.

[0006] To achieve the above-mentioned purpose, the present application provides an unmanned aerial vehicle detection method based on infrared image analysis, which comprises:

[0007] Collecting infrared image sequence data within a preset time window;

[0008] performing environment stability analysis based on the infrared image sequence data to determine an image heat distribution stability factor, an environment temperature stability factor, and a motion feature stability factor;

[0009] applying fuzzy logic inference rules based on the image heat distribution stability factor, the environment temperature stability factor, and the motion feature stability factor to infer an infrared response interval; performing target analysis on the infrared image sequence data according to a preset detection item set based on the infrared response interval to obtain a periodic detection data set;

[0010] performing identification on the periodic detection data set by using an infrared target identifier to obtain a UAV detection result.

[0011] Preferably, the collecting infrared image sequence data within a preset time window comprises: acquiring monitoring data of a plurality of infrared sensor arrays; performing outlier rejection and standardization processing on the monitoring data to generate a cleaned infrared image data set; and extracting a heat distribution parameter set, a temperature parameter set, and a motion feature parameter set based on the cleaned infrared image data set.

[0012] Preferably, the performing environment stability analysis based on the infrared image sequence data comprises: traversing the heat distribution parameter set, the temperature parameter set, and the motion feature parameter set; calculating a variance value of the heat distribution parameter set as an image heat distribution stability factor; analyzing a historical change trend of the temperature parameter set to determine an environment temperature stability factor; and counting a fluctuation range of the motion feature parameter set to generate a motion feature stability factor.

[0013] Preferably, the applying fuzzy logic inference rules to infer an infrared response interval comprises: defining fuzzy input variables, wherein the fuzzy input variables comprise image heat distribution stability, environment temperature stability, and motion feature stability; setting a fuzzy output variable as the infrared response interval; establishing a fuzzy rule base, wherein the fuzzy rule base contains rules for mapping values of the image heat distribution stability factor, the environment temperature stability factor, and the motion feature stability factor to the infrared response interval; inputting the image heat distribution stability factor, the environment temperature stability factor, and the motion feature stability factor into a fuzzy inference system; and generating the infrared response interval through a process of fuzzification, rule evaluation, and defuzzification.

[0014] Preferably, the performing target analysis on the infrared image sequence data according to a preset detection item set based on the infrared response interval comprises: setting the preset detection item set to include heat feature detection, motion trajectory detection, and abnormal behavior detection; dividing time intervals of the infrared image sequence data based on the infrared response interval; and applying the heat feature detection, the motion trajectory detection, and the abnormal behavior detection in each time interval to generate intermediate detection data.

[0015] Preferably, the obtaining the periodic detection data set comprises: integrating the heat distribution value, the temperature change value and the motion vector value of the intermediate detection data; arranging the heat distribution value, the temperature change value and the motion vector value in time sequence; and generating a periodic detection data array.

[0016] Preferably, the identifying the periodic detection data set by using the infrared target recognizer comprises: extracting a potential target feature in the periodic detection data array; matching the potential target feature with a preset unmanned aerial vehicle feature library; calculating a matching degree score; and classifying a target type based on the matching degree score.

[0017] Preferably, the obtaining the unmanned aerial vehicle detection result comprises: outputting a classification result of the target type; generating detection report data; and storing the detection report data into a historical database.

[0018] Preferably, the method further comprises: updating the infrared image sequence data in real time; based on the updated infrared image sequence data, recalculating the image heat distribution stability factor, the environmental temperature stability factor and the motion feature stability factor; and dynamically adjusting the infrared response interval according to the recalculated result.

[0019] Preferably, the present application further comprises an unmanned aerial vehicle detection system based on infrared image analysis, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the unmanned aerial vehicle detection method based on infrared image analysis as described above.

[0020] Compared with the prior art, the present application has the following beneficial effects:

[0021] By collecting the infrared image sequence data within a preset time window, continuous information of the target within a certain time can be obtained, which provides a rich data source for subsequent analysis. Based on the infrared image sequence data, the environmental stability analysis is performed to determine the image heat distribution stability factor, the environmental temperature stability factor and the motion feature stability factor, which can comprehensively consider the stability degree of the environment in terms of heat distribution, temperature and motion feature, so as to more accurately grasp the dynamic change characteristics of the environment.

[0022] The fuzzy logic inference rule is applied to deduce the infrared response interval in combination with the three stability factors, the advantages of the fuzzy logic in processing uncertain and fuzzy information are fully utilized, the appropriate infrared response interval is adaptively determined according to the complex and changeable environmental characteristics, so that the interval is more suitable for the actual environmental conditions. The periodic detection data set is obtained by performing target analysis on the infrared image sequence data according to the preset detection item set based on the infrared response interval, the target analysis process is more targeted and timely, the invalid analysis caused by the unstable environment is reduced, and the effectiveness of target information extraction is improved.

[0023] The periodic detection data set is identified by using the infrared target recognizer to obtain a UAV detection result, since the environmental analysis, response interval deduction and target analysis in the early stage all fully consider the dynamic characteristics of the environment, the data input to the recognizer is more accurate and reliable, which helps to improve the recognition effect of the recognizer on the UAV. The method is closely related and cooperates in each link from data acquisition to output of the final recognition result, can better adapt to the complex and changeable environment, and can maintain good detection performance in different scenes, thereby providing a more effective technical approach for UAV detection. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 A working principle diagram of the UAV detection method based on infrared image analysis is described.

[0025] Figure 2 A flowchart for collecting infrared image sequence data in a preset time window is described.

[0026] Figure 3 A flowchart for deducing the infrared response interval by using the fuzzy logic inference rule is described.

[0027] Figure 4 A flowchart for target analysis based on the infrared response interval is described.

[0028] Figure 5 A flowchart for identification by using the infrared target recognizer is described. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0030] Please refer to Figure 1 The present application provides a UAV detection method based on infrared image analysis, which comprises:

[0031] The infrared image sequence data within a preset time window is collected, and environmental stability analysis is performed based on the infrared image sequence data to determine an image thermal distribution stability factor, an environmental temperature stability factor, and a motion feature stability factor. Fuzzy logic inference rules are applied to infer an infrared response interval based on the aforementioned three stability factors. Based on the infrared response interval, target analysis is performed on the infrared image sequence data according to a preset detection item set, thereby obtaining a periodic detection data set. The periodic detection data set is identified by using an infrared target identifier, and finally, a UAV detection result is obtained. This method optimizes the identification process of the UAV in a complex infrared background by dynamically evaluating the environmental stability and adjusting the analysis frequency accordingly.

[0032] Embodiment 1: refer to Figure 2 The collection of infrared image sequence data is the foundation of the entire method, involving the complete process from physical sensors to data preprocessing. Multiple infrared sensor arrays are deployed in the monitoring area, and these arrays continuously capture the infrared radiation information of the scene at a fixed frame rate, generating raw monitoring data streams. In the initial state, this data stream contains various interference factors, such as random fluctuations caused by sensor thermal noise, local outliers caused by instantaneous strong heat sources, or error pixels generated during data transmission. To eliminate these interferences, systematic data cleaning operations are performed. Outlier rejection uses a statistical distribution-based method to calculate the temperature value distribution of each pixel point or region in a continuous frame sequence, sets a standard deviation threshold range, and identifies and removes data points that exceed this threshold range; at the same time, it is supplemented by a neighboring data point comparison strategy to detect isolated abnormal points that have significant temperature differences with spatially adjacent points or temporally preceding and following frames. After completing outlier rejection, data standardization processing is performed. This processing aims to solve the problem of inconsistent responses between different infrared sensor units due to manufacturing tolerances, aging levels, or environmental differences. The standardization process maps the data of each sensor unit to a unified numerical reference, such as converting the original radiation intensity values of all pixels to relative temperature values, or adjusting the data to a preset standard numerical interval through linear transformation. After the above cleaning and standardization steps, a high-quality, comparable cleaned infrared image data set is generated.

[0033] Based on the cleaned infrared image dataset, key parameter extraction is performed. This process aims to extract three types of structured information from the image sequence that characterize the environmental state and target cues. The thermal distribution parameter set focuses on the overall and local thermal patterns of the images. The calculation includes the average temperature value of the entire image frame, reflecting the overall thermal level of the scene; analyzing the temperature gradient distribution, identifying the boundary between high-temperature core areas and low-temperature background areas; counting the area proportion of pixels above a certain temperature threshold, quantifying the range of significant heat sources; and possibly including the number, size, and spatial distribution statistics of thermal spots. The temperature parameter set is more focused on the quantitative description of the environmental background temperature. Extract the average temperature value of the environmental background area (usually excluding potential moving target areas) as the baseline of the environmental temperature; record the highest temperature and the lowest temperature of the background area within the time window; calculate the change amount or rate of the background temperature between consecutive frames to capture the dynamic characteristics of the environmental temperature. The motion feature parameter set aims to capture dynamic change information in the scene. By calculating the displacement vector of corresponding pixels or regions between consecutive image frames, a motion vector field is generated; based on the motion vector field, the pixel area of the motion region and its change rate over time are counted; the direction distribution consistency of the motion vector is analyzed.

[0034] Environmental stability analysis is a key step in evaluating whether the monitoring scene is suitable for effective unmanned aerial vehicle detection. The core is to calculate three factors that quantify the stability of the environment. The analysis process traverses the previously extracted thermal distribution parameter set, temperature parameter set, and motion feature parameter set. The calculation of the image thermal distribution stability factor focuses on the fluctuation degree of the thermal pattern over time. For each parameter in the thermal distribution parameter set, calculate the statistical variance value of all frames within the preset time window. For example, calculate the variance of the average temperature sequence of the last N frames. This variance value is directly used as a measure of the image thermal distribution stability factor. The smaller the variance value, the more consistent the thermal distribution pattern of the scene changes over time, indicating that the thermal background is relatively stable, which is conducive to distinguishing the thermal characteristics of small targets such as unmanned aerial vehicles. A larger variance indicates that the thermal background has significant fluctuations, which may be caused by large-scale heat source changes or strong interference.

[0035] The determination of the environmental temperature stability factor relies on the analysis of the historical trend of the temperature parameter set. The parameter in the temperature parameter set that represents the ambient background temperature is of primary interest. The pattern of change of this parameter over a time window is analyzed, and a sequence of its moving average is computed, and the trend of this sequence is evaluated. Specifically, the slope of the linear fit of the temperature trend can be computed, or the difference between the maximum and minimum values of the temperature over the entire time window (range) can be computed. Based on the result of this trend analysis, the environmental temperature stability factor is determined. For example, if the absolute value of the slope of the temperature trend is below a certain pre-set threshold, or the temperature range is below a certain range, then the environmental temperature stability is deemed to be high, and a high stability factor value is assigned. Conversely, if the temperature exhibits a clear upward or downward trend, or a dramatic fluctuation, then a low stability factor value is assigned. Stable environmental temperature helps to establish a reliable thermal contrast baseline.

[0036] The generation of the motion feature stability factor is based on the fluctuation range statistics of the motion feature parameter set. This factor aims to quantify the severity of non-target motion interference in the scene. For key indicators in the motion feature parameter set, such as the area change rate of the motion region, the average motion speed amplitude, or the consistency indicator of the main motion direction, the fluctuation range of these indicators over a pre-set time window is computed. The fluctuation range can be obtained by computing the difference between the maximum and minimum values of the parameter sequence, or by computing its standard deviation. This fluctuation range value is quantified as the motion feature stability factor. A small fluctuation range means that the background motion in the scene is relatively smooth and regular, and the motion feature stability factor value is high. A large fluctuation range indicates the presence of significant and irregular background motion interference, and the motion feature stability factor value is low. A lower level of motion interference helps to clearly identify and track the potential UAV target motion trajectory. By integrating these three stability factors, the system can quantitatively evaluate the degree of favorability or unfavorability of the current environmental conditions for infrared UAV detection.

[0037] The environmental stability analysis is based on the aforementioned extracted thermal distribution parameter set, temperature parameter set and motion feature parameter set. The analysis process iterates through these three parameter sets. For the thermal distribution parameter set, the statistical variance value of the multiple parameters it contains within a preset time window is calculated, which is directly defined as the image thermal distribution stability factor. The smaller the variance value, the more gentle the thermal distribution changes over time, and the more stable the environmental thermal background. For the temperature parameter set, its historical trend is analyzed, which includes calculating the moving average of the temperature mean or median, and evaluating the linear fitting slope of the moving average or calculating the temperature variation amplitude within a certain time period. Based on the results of this trend analysis, the environmental temperature stability factor is determined, for example, a trend with a variation amplitude below a preset threshold is determined to be stable. For the motion feature parameter set, the fluctuation range of the motion vector amplitude or motion area area and other parameters it contains within the time window is calculated, such as the difference between the maximum and minimum values or the standard deviation. This fluctuation range is quantified as the motion feature stability factor, and a small fluctuation range indicates less large-scale or intense motion interference in the environment.

[0038] Example 2: see Figure 3 The application of fuzzy logic inference rules is the core process of converting the three quantitative factors generated by the environmental stability analysis into specific infrared response intervals. This process starts with clearly defining the input and output variables of the fuzzy inference system. The input variables correspond to the three factors derived from the aforementioned environmental stability analysis: image thermal distribution stability, environmental temperature stability, and motion feature stability. Each input variable is assigned a set of language values that describe its state. These language values are usually qualitative descriptions in natural language, such as "low", "medium", "high" stability levels. These language values are not precise numerical limits, but represent the fuzzy concept range of stability. At the same time, a unique fuzzy output variable is defined as the infrared response interval, and its language values also use qualitative descriptions, such as "very short", "short", "medium", "long", "very long", to represent different time interval length concepts. Each language value is associated with a membership function that defines the degree to which a specific numerical point in the numerical domain belongs to a certain language value, usually represented by a value between 0 and 1, where 0 represents complete non-membership and 1 represents complete membership. The shape of the membership function can be triangular, trapezoidal or Gaussian, and its parameters are set according to domain knowledge or experience.

[0039] Establishing the fuzzy rule base is the core knowledge base of the fuzzy inference system. The rule base contains multiple fuzzy conditional statements, and each rule expresses the causal relationship or mapping relationship between the combination of the language values of the input variables (three stability factors) and the language value of the output variable (infrared response interval). The expression form of the rule is usually in the structure of "IF... THEN...". For example, a rule can be: "If the image thermal distribution stability is 'high', and the environmental temperature stability is 'high', and the motion feature stability is 'high', then the infrared response interval is 'long'". Another rule can be: "If the image thermal distribution stability is'medium', and the environmental temperature stability is 'low', and the motion feature stability is 'high', then the infrared response interval is'medium'". The design of the rule base aims to cover all reasonable combinations of input variable language values, ensuring that the system can respond to various possible environmental stability states. The construction of the rule base usually relies on the experience knowledge of domain experts, or is induced from the analysis of historical data.

[0040] The inference process begins with the input of the three precise numerical values of the image thermal distribution stability factor, the environmental temperature stability factor, and the motion feature stability factor into the fuzzy inference system. The first step is fuzzy processing. For each input variable, according to its pre-defined membership function, the membership degree of the input precise value to all language values of the variable is calculated. For example, assuming that the input image thermal distribution stability factor value is 0.85, this value may belong to the "high" stability language value with a membership degree of 0.6, belong to the "medium" stability language value with a membership degree of 0.4, and belong to the "low" stability language value with a membership degree of 0.0. Similarly, the precise values of the environmental temperature stability factor and the motion feature stability factor are also converted into membership degree sets corresponding to their language values. This process converts the precise numerical input into a fuzzy set representation.

[0041] The second step is rule evaluation. The fuzzy inference system traverses all rules in the rule base. For each rule, the language values corresponding to each input variable in its premise condition (i.e. the IF part) are checked. According to the membership degrees obtained after fuzzy processing of the input values, the satisfaction degree of the premise condition of the rule is determined. The commonly used method is to take the minimum value or product of the membership degrees of each clause in the premise condition to obtain the activation strength or trigger weight of the rule. Then, according to the output language value specified in the conclusion part (i.e. the THEN part) of the rule and the calculated activation strength, a fuzzy output subset is generated. This fuzzy output subset represents the fuzzy value and its confidence level of the output variable (infrared response interval) suggested by the rule under the current input. For example, a rule with an activation strength of 0.6 whose conclusion is that the infrared response interval is "long" will generate a "long" output fuzzy subset with a membership degree of 0.6.

[0042] The third step is defuzzification. After the rule evaluation, usually multiple rules are activated, each contributing a fuzzy output subset. The first step of defuzzification is to aggregate all these fuzzy output subsets from the activated rules to form a total set representing the final fuzzy output of the system. The aggregation operation usually takes the maximum, i.e. for each point in the output variable's domain, the maximum of the membership degrees of all fuzzy output subsets at that point. This results in an aggregated fuzzy output set. The final step is to calculate a crisp, numerical infrared response interval from this aggregated fuzzy output set. Common defuzzification methods include the center of gravity method, which calculates the x-coordinate of the center of area of the aggregated fuzzy output set's membership function curve, or the maximum average method, which takes the average of all points where the membership degree reaches its maximum. For example, the center of gravity method yields a numerical value of 15 seconds, which is the final output of the fuzzy inference system as the precise value of the infrared response interval. This value will be used as the basis for dividing the time intervals of the infrared image sequence data in the subsequent steps. The entire fuzzy inference process converts the qualitative environmental stability assessment into a quantitative time control parameter, enabling the system to adaptively adjust the analysis pace according to the dynamic characteristics of the environment.

[0043] Example 3, see Figure 4 Based on the numerical value of the infrared response interval derived from the fuzzy logic inference, the target analysis of the infrared image sequence data is performed. The preset detection item set contains three core analysis dimensions: thermal feature detection, motion trajectory detection, and abnormal behavior detection. Thermal feature detection focuses on identifying and quantifying the thermal radiation characteristics of potential targets, analyzing their temperature distribution, thermal spot morphology, and thermal contrast differences with the background. Motion trajectory detection aims to track the spatial position changes of targets in consecutive image frames, depicting their motion paths, speed changes, and direction trends. Abnormal behavior detection focuses on whether the target's motion pattern deviates from the norm, identifying such as sudden acceleration, deceleration, irregular circling, long-time hovering, or trajectories that do not conform to the common motion patterns of the environment. The infrared response interval defines the length of the time period for performing these analysis tasks. Using this interval value, the continuous infrared image sequence data is divided into a series of consecutive time intervals on the time axis. The length of each time interval is equal to the current infrared response interval value. For example, if the infrared response interval is 15 seconds and the image frame rate is 10 frames per second, each time interval contains 150 frames of image data. The division process ensures that the time intervals continuously cover the entire monitoring period without overlap or gaps.

[0044] Within each predefined time interval, three detection tasks from a pre-defined set of detection items are executed sequentially. Thermal feature detection is performed first. The thermal distribution data of all image frames within that time interval are analyzed. Identifying potential target regions is typically based on temperature threshold segmentation or connected component analysis to find pixel regions with temperatures significantly higher than the ambient background temperature. For each identified potential target region, its thermal distribution feature values ​​are calculated. These values ​​include, but are not limited to: the region's average temperature, reflecting the overall thermal radiation intensity of the target; the standard deviation of the temperature within the region, characterizing the uniformity of the internal thermal distribution of the target; the pixel area of ​​the target region, indicating the size of the target in the image; the compactness or morphology factor of the target region, describing its shape characteristics; and the average temperature difference between the target region and the local background, quantifying its thermal contrast. These calculated thermal feature values ​​are recorded as the thermal feature detection results for that time interval.

[0045] Motion trajectory detection follows. This detection analyzes the target displacement between consecutive image frames within a time interval. The core is calculating the motion vector of the potential target. For targets identified in thermal feature detection, tracking is performed in subsequent frames. Target tracking can employ feature matching or motion estimation methods. Once the target is associated with consecutive frames, its motion vector can be calculated. This vector represents the displacement direction and distance of the target on the image plane. The calculation of the motion vector can be expressed as:

[0046]

[0047] in: Indicates the target is in the first place. Frame and the Motion vectors between frames (pixels / second). The target center point is at the 1st Frame and the Horizontal displacement (pixels) between frames. The target center point is at the 1st Frame and the Vertical displacement (pixels) between frames. It is the first Frame and the The time interval (in seconds) between frames. By connecting the position points of the target in all consecutive frames within this time interval, a preliminary motion trajectory of the target is formed. Based on this trajectory, further trajectory parameters are calculated, such as the target's average velocity, total displacement, direction of motion, and velocity variation within the interval. These trajectory parameters are recorded as the motion trajectory detection result for this time interval.

[0048] Finally, abnormal behavior detection is performed. This detection analyzes the trajectory parameters obtained from motion trajectory detection and the target's continuous motion pattern. A predefined set of rules or pattern recognition algorithms are applied to determine whether the target's motion is abnormal. The rule set may include: detecting whether the target's speed changes drastically in a short period of time; determining whether the target moves repeatedly or circles around a certain position (hovering pattern); identifying whether the target maintains a fixed position for a long time in a non-predetermined area (hovering); checking whether the target's motion direction changes frequently and irregularly; or evaluating whether the target's overall motion path conforms to common motion patterns in the environment. The detection results are usually output as one or more flags or quantitative indicators, such as "sudden acceleration exists," "hovering behavior detected," "hovering time exceeds the limit," etc., or a comprehensive abnormal behavior score is calculated. These abnormal behavior flags or indicators are recorded as the abnormal behavior detection results for that time interval. After completing the analysis of all three detection items within a time interval, intermediate detection data for that time interval is generated. This data is a structured record containing summary information of the aforementioned thermal feature detection results, motion trajectory detection results, and abnormal behavior detection results.

[0049] The process of obtaining a periodic detection dataset involves systematically integrating and structuring intermediate detection data generated across all time intervals. The integration operation extracts key numerical information from the intermediate detection data of each time interval. Three main types of values ​​are extracted: thermal distribution values, temperature change values, and motion vector values. Thermal distribution values ​​typically include the average temperature of the main (or most prominent) target detected within the interval, the temperature difference between the target and the background, and the area of ​​the target region. Temperature change values ​​may include the amount or rate of change of the target's average temperature within the time interval. Motion vector values ​​include the target's average velocity vector (containing X and Y components), the target's total displacement vector within the interval, and the average motion direction angle. If multiple targets exist, the feature values ​​of each target may be recorded, or the feature values ​​of the most prominent target may be selected. After extraction, these values ​​are arranged in a time series. The arrangement is based on the start or center timestamp of the time interval, ensuring that data from the earliest time interval is placed first, with data from subsequent time intervals arranged sequentially. After arrangement, the thermal distribution values, temperature change values, and motion vector values, organized in time series, are combined to form a structured two-dimensional data array. The rows of this array correspond to different time intervals (arranged in chronological order), while the columns correspond to different dimensions of the detected data. For example, columns might include: timestamp, target average temperature, target background temperature difference, target area, velocity X component, velocity Y component, displacement, orientation angle, anomaly flag 1, anomaly flag 2, and anomaly comprehensive score. Each row of data represents a summary of target analysis for a specific time interval within the monitored scene. This final two-dimensional data array is the periodic detection data set, organized in chronological order, containing information on the evolution of target thermal characteristics, motion states, and abnormal behaviors over time, providing structured input data for the subsequent target identification stage.

[0050] Example 4: See Figure 5The infrared target identifier processes periodic detection datasets to classify and identify potential targets. This process begins with feature extraction. The periodic detection dataset is a two-dimensional data array arranged chronologically, with each row corresponding to a time interval and each column representing a dimension of the detection data. The feature extraction operation selects or calculates feature values ​​from this array that effectively distinguish the drone from other objects. These feature values ​​can be directly selected from the column data in the array, such as the target's average temperature, average velocity amplitude (calculated from the X and Y components of velocity), azimuth angle change (the average difference between azimuth angles in consecutive time intervals), and anomaly score, etc. Derived features can also be calculated based on the array data, such as calculating the autocorrelation coefficient of the target velocity vector (reflecting the stability of velocity changes) over several consecutive time intervals, calculating the slope of the target temperature trend over multiple time intervals, or statistically analyzing the frequency of specific anomalous behavior indicators in the most recent N time intervals. The goal of feature extraction is to generate a feature vector that comprehensively characterizes the key patterns of the potential target's thermal properties and motion behavior.

[0051] Next, the extracted feature vectors are matched against a pre-defined UAV feature library. The UAV feature library stores feature templates or model parameters for various known UAV types and common interfering objects under typical operating conditions. These templates define the typical performance range or pattern of various targets in the infrared detection dimension. For example, a feature template for a quadcopter UAV might include: cruising speed range, hovering capability indicators, altitude change patterns, infrared radiation intensity range (related to ambient temperature), and size range (represented as hotspot area range in images). The matching process involves calculating the similarity or distance between the currently extracted feature vector and each template in the feature library. Depending on the feature type, the matching algorithm may employ different metrics. For numerical features, the Euclidean distance or cosine similarity between the feature vector and the template vector may be calculated; for categorical features or behavioral patterns, likelihood may be calculated using rule-based matching or probabilistic models. The matching results generate a matching score for each target type in the feature library, which quantifies the degree of conformity between the currently observed feature and the target type. An example of a matching score, based on the matching results of partial features across three time intervals with three target types in the feature library, is shown in Table 1.

[0052] Table 1: Example of target feature matching score.

[0053] Time interval Target type Speed matching degree Temperature matching degree Behavior matching degree Comprehensive matching degree score T1 Quadcopter unmanned aerial vehicle 0.92 0.88 0.78 0.86 T1 Fixed-wing unmanned aerial vehicle 0.75 0.82 0.65 0.74 T1 Bird 0.45 0.60 0.30 0.45 T2 Quadcopter unmanned aerial vehicle 0.85 0.90 0.85 0.87 T2 Fixed-wing unmanned aerial vehicle 0.70 0.85 0.55 0.70 T2 Bird 0.50 0.65 0.40 0.52 T3 Quadcopter unmanned aerial vehicle 0.88 0.85 0.90 0.88 T3 Fixed-wing unmanned aerial vehicle 0.72 0.80 0.70 0.74 T3 Bird 0.42 0.58 0.35 0.45

[0054] Among them, speed matching degree, temperature matching degree, and behavior matching degree represent the degree of matching between the corresponding dimensions of the feature vector and the target type template. The comprehensive matching degree score is a weighted average of the matching degrees of each dimension or a fusion result of a specific algorithm.

[0055] Based on the calculated matching scores, target types are classified. Various classification strategies can be employed. A common approach is to set a confidence threshold. For example, a threshold is set; if the overall matching score for a target type exceeds this threshold, the target is classified as that type. If the highest score does not reach the threshold but is significantly higher than other types, it may be classified as "suspected [target type]". Another approach is to use the nearest neighbor principle, directly classifying the target as the type with the highest matching score. More complex classifiers can comprehensively consider all matching scores and the feature vectors themselves in their decision-making. The classification results distinguish targets into clear category labels, such as "confirmed drone - quadcopter", "confirmed drone - fixed-wing", "suspected drone", "non-drone - bird", "non-drone - unknown object", or "background interference". This classification result represents the system's preliminary determination of the detected target's identity. Obtaining the final drone detection results involves output and storage operations. The output operation organizes the target classification results and related information into structured detection report data. The report data typically includes the following key information items: Timestamp: the time point when the target was detected or the time interval of the target's activity; Target Location Information: if the system has positioning capabilities, the estimated location of the target in the scene; Target Classification Type: the target category label obtained from the previous classification steps; Confidence / Match Score: the confidence level of the classification result or the highest match score, reflecting the reliability of the judgment; Summary of Main Features: a brief description of the main features observed, such as average speed, average temperature, and significant behavioral patterns; Description of Abnormal Behavior: if abnormal behavior is identified during the detection process, it is described here. The format of the report data can be structured data records or more readable natural language summary text, depending on the application requirements. The generated detection report data is then stored in a historical database. This historical database is designed for persistent storage and management of detection reports. The database structure typically contains data tables, whose fields correspond to the information in the detection report data. The database management system provides data indexing functions, such as quick queries based on timestamp range, target type, or location area. It supports retrospective analysis of historical detection report data, such as querying activity records of a specific type of drone within a specific time period, or analyzing target activity patterns in a specific area. The establishment of historical databases provides a data foundation for long-term situational awareness, threat assessment, and system performance analysis. For example, security personnel can query the database to understand all drone activity records in the monitored area over the past 24 hours; system maintenance personnel can analyze historical data to evaluate the effectiveness of detection algorithms or identify potential false alarm patterns.

[0056] Example 5: This method employs a dynamic mechanism that continuously adapts to environmental changes. After initial execution and outputting detection results, the system does not remain static but continuously receives and processes new infrared image data. This real-time updating is achieved by the infrared sensor array continuously acquiring new image frames at a fixed or configurable sampling rate. Newly acquired image frames are added to the currently processed data stream in real time. To maintain a constant preset time window length, this mechanism typically follows a first-in, first-out (FIFO) principle; that is, when a new frame is added, the oldest frame in the window is automatically discarded. Therefore, the infrared image sequence data maintained in the system always contains the latest acquired frame and covers a fixed time span, which is the preset time window. The continuously updated infrared image sequence data forms the dynamic basis for environmental condition monitoring.

[0057] Based on the updated infrared image sequence data—a dataset containing the latest observation information but with the overall time window length unchanged—the system re-executes the environmental stability analysis steps. This analysis process completely reuses the procedures and methods used in the initial analysis, but the input is data reflecting the latest environmental conditions. Specifically, for the updated data sequence, the system re-traverses the thermal distribution parameter set, temperature parameter set, and motion feature parameter set to calculate real-time stability factors. For the calculation of the image thermal distribution stability factor, the system re-analyzes the updated thermal distribution parameter set, calculating the statistical variance of these parameters within the updated time window, thereby obtaining new factor values ​​reflecting the current degree of thermal background fluctuation. For the environmental temperature stability factor, the system re-analyzes its changing trends based on the updated temperature parameter set (especially the average temperature sequence of the background region). This includes recalculating the moving average sequence or re-evaluating the difference between the maximum and minimum temperatures and the slope of change within the latest window to determine the current stability of the environmental temperature and generate new factor values. For the motion feature stability factor, the system statistically analyzes the fluctuation range of the updated motion feature parameter set within the latest time window, thereby quantifying the severity of background motion interference in the current environment and generating new factor values. These three recalculated stability factors represent the latest environmental conditions faced by the current system operation.

[0058] Based on the recalculated image thermal distribution stability factor, ambient temperature stability factor, and motion feature stability factor, the system dynamically adjusts the key parameter used for target analysis: the infrared response interval. This adjustment process is not arbitrary but fully reuses the complete workflow of the aforementioned fuzzy logic inference system. The system uses the newly calculated precise values ​​of the three stability factors as new inputs to the fuzzy inference system. During fuzzification, the latest precise values ​​are converted into corresponding fuzzy input membership degrees according to a preset input variable membership function. After completion, the system proceeds to the rule evaluation stage. Based on a pre-established fuzzy rule base, all rules matching the current fuzzy input are identified, and corresponding fuzzy output subsets are generated based on the rule activation strength. Then, a defuzzification operation is performed, aggregating all fuzzy output subsets generated by the activation rules, and calculating a new, precise infrared response interval value using the selected defuzzification method. This final new value represents the target analysis cycle length that the system considers most suitable under the current environmental conditions. This newly calculated infrared response interval value immediately replaces the previously used old interval value.

[0059] This dynamic adjustment mechanism makes the infrared response interval a constantly updated variable driven by environmental changes. The newly calculated infrared response interval value is immediately applied to guide subsequent target analysis. This means that from the next round of analysis, the period length used by the system to divide the time interval of the infrared image sequence data is the newly adjusted value. For example, if the new infrared response interval is calculated to be 20 seconds, while the previous interval was 15 seconds, then the length of the subsequent time interval will change from 15 seconds to 20 seconds. Through this closed-loop feedback mechanism, the system's analysis frequency can proactively adapt to changes in environmental stability. In a stable state with a relatively quiet background, uniform heat distribution, constant temperature, and weak motion interference, the recalculated stability factor is usually at a "high" level. After receiving a high-stability input, the fuzzy inference rule will usually produce a longer infrared response interval output value. A longer analysis interval means that the system reduces the frequency of data processing, alleviates the computational burden, and saves processing resources. Conversely, when environmental disturbances occur, such as the appearance of local heat sources in the monitoring area, increased fluctuations in ambient temperature, or frequent motion interference in the background, the recalculated stability factor value will decrease. Upon receiving input indicating a decrease in stability, fuzzy inference systems typically derive a shorter infrared response interval. This shorter analysis interval forces the system to increase processing frequency and sampling density, potentially enabling faster capture of fleeting UAV target signals in complex and interfering environments. This dynamic adaptation process continues throughout the system's lifecycle. Each new data update triggers a reassessment of the environmental stability factor, potentially leading to adjustments in the infrared response interval. Therefore, the system does not require a static, fixed analysis cycle; instead, it dynamically responds to the environmental state in real time, constantly adapting its detection and analysis rhythm to the complexity of the environment to maintain effective UAV identification capabilities under changing conditions. This periodic update mechanism is a core adaptive feature of the system in dealing with complex, non-steady-state infrared environments.

[0060] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0061] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for detecting unmanned aerial vehicles (UAVs) based on infrared image analysis, characterized in that, include: Collect infrared image sequence data within a preset time window; Based on the infrared image sequence data, environmental stability analysis is performed to determine the image thermal distribution stability factor, environmental temperature stability factor, and motion characteristic stability factor. Based on the image thermal distribution stability factor, ambient temperature stability factor, and motion feature stability factor, fuzzy logic reasoning rules are applied to infer the infrared response interval; based on the infrared response interval, the infrared image sequence data is subjected to target analysis according to a preset detection item set to obtain a periodic detection data set. The periodic detection data set is identified using an infrared target identifier to obtain the UAV detection results.

2. The UAV detection method based on infrared image analysis according to claim 1, characterized in that, The acquisition of infrared image sequence data within a preset time window includes: acquiring monitoring data from multiple infrared sensor arrays; performing outlier removal and standardization on the monitoring data to generate a cleaned infrared image dataset; and extracting a set of thermal distribution parameters, a set of temperature parameters, and a set of motion feature parameters based on the cleaned infrared image dataset.

3. The UAV detection method based on infrared image analysis according to claim 2, characterized in that, The environmental stability analysis based on the infrared image sequence data includes: traversing the thermal distribution parameter set, temperature parameter set, and motion feature parameter set; calculating the variance of the thermal distribution parameter set as the image thermal distribution stability factor; analyzing the historical change trend of the temperature parameter set to determine the environmental temperature stability factor; and statistically analyzing the fluctuation range of the motion feature parameter set to generate the motion feature stability factor.

4. The UAV detection method based on infrared image analysis according to claim 3, characterized in that, The application of fuzzy logic inference rules to infer the infrared response interval includes: defining fuzzy input variables, which include image thermal distribution stability, ambient temperature stability, and motion feature stability; setting the fuzzy output variable as the infrared response interval; establishing a fuzzy rule base, which contains rules mapping the values ​​of the image thermal distribution stability factor, ambient temperature stability factor, and motion feature stability factor to the infrared response interval; inputting the image thermal distribution stability factor, ambient temperature stability factor, and motion feature stability factor into the fuzzy inference system; and generating the infrared response interval through fuzzification, rule evaluation, and defuzzification processes.

5. The UAV detection method based on infrared image analysis according to claim 4, characterized in that, The step of performing target analysis on the infrared image sequence data according to a preset detection item set based on the infrared response interval includes: setting the preset detection item set to include thermal feature detection, motion trajectory detection, and abnormal behavior detection; dividing the time interval of the infrared image sequence data based on the infrared response interval; and applying the thermal feature detection, motion trajectory detection, and abnormal behavior detection in each time interval to generate intermediate detection data.

6. The UAV detection method based on infrared image analysis according to claim 5, characterized in that, The acquisition of the periodic detection data set includes: integrating the heat distribution value, temperature change value, and motion vector value of the intermediate detection data; arranging the heat distribution value, temperature change value, and motion vector value in a time sequence; and generating a periodic detection data array.

7. The UAV detection method based on infrared image analysis according to claim 6, characterized in that, The process of using an infrared target identifier to identify the periodic detection data set includes: extracting potential target features from the periodic detection data array; matching the potential target features with a preset UAV feature library; calculating a matching score; and classifying the target type based on the matching score.

8. The UAV detection method based on infrared image analysis according to claim 7, characterized in that, The process of obtaining UAV detection results includes: outputting the classification results of the target type; generating detection report data; and storing the detection report data in a historical database.

9. The UAV detection method based on infrared image analysis according to claim 8, characterized in that, The method further includes: updating the infrared image sequence data in real time; recalculating the image thermal distribution stability factor, ambient temperature stability factor, and motion characteristic stability factor based on the updated infrared image sequence data; and dynamically adjusting the infrared response interval according to the recalculated results.

10. A drone detection system based on infrared image analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the UAV detection method based on infrared image analysis as described in any one of claims 1 to 9.

Images