An ecological protection area-oriented intelligent identification method for invading drones

Abstract

The present application relates to the technical field of image recognition, and particularly relates to an intelligent identification method for invasion of unmanned aerial vehicles in ecological protection zones, which determines whether to trigger an acoustic detection device at each moment according to the static and dynamic characteristics of heat source targets in infrared images of the ecological protection zone at each moment, processes time-domain sound signals into a mel-frequency spectrum graph after triggering the acoustic detection device, determines the energy characteristics of the mel-frequency spectrum graph according to the uniformity of energy distribution in each mel-frequency band, and determines the unmanned aerial vehicle invasion identification result according to the correlation between the energy characteristics of the heat source target and the change of the running speed. The present application triggers control through preliminary screening of infrared images, and then performs acoustic fine analysis, finally obtains the unmanned aerial vehicle invasion identification result according to cross-modal physical correlation fusion, improves the identification accuracy, reduces the false alarm rate, improves the energy efficiency, and effectively deals with the complex acoustic and thermal background interference in the ecological protection zone, and improves the environmental adaptability.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and more specifically to an intelligent identification method for intrusive drones in ecological protection areas. Background Technology

[0002] Current methods for identifying drones that have intruded into ecological protection zones typically rely on acoustic detection. This involves setting up microphone arrays at various locations within the ecological protection zone to collect sound signals generated by the drone's propellers to detect intrusions. However, acoustic detection is susceptible to interference from environmental noise, such as wind and animal calls, resulting in low accuracy.

[0003] To prevent drones from interfering with ecological protection areas, effective identification of intruding drones is necessary. Existing identification technologies mainly include:

[0004] Infrared image-based identification: This method captures the thermal radiation from drone engines or propellers using thermal imaging cameras. However, it has the drawback that there are many mobile heat sources such as wild animals in ecological reserves, and relying solely on visual features can easily lead to misjudgments. Furthermore, it has limited ability to distinguish heat sources from distant or small drones.

[0005] Acoustic detection-based identification: This method collects the characteristic acoustic patterns of drone propellers using a microphone array. However, the drawback is that background noise in the natural environment, such as wind, animal calls, and water flow, can severely interfere with or even drown out the target signal, leading to a decrease in the recognition rate.

[0006] None of the aforementioned single technologies can achieve stable and accurate identification in complex scenarios such as ecological protection areas. Although there are solutions that simply use multiple sensors in parallel, this leads to continuous high power consumption, heavy data processing burden, and fails to solve the problem of effective coordination and decision fusion between different modal information. Summary of the Invention

[0007] The technical problem that this invention aims to solve is: how to provide an intelligent identification method for intrusion drones that can adapt to the complex environment of ecological protection zones, effectively integrate infrared and acoustic information, and reduce system energy consumption and false alarm rate while ensuring high recognition accuracy.

[0008] The purpose of this invention is to provide an intelligent identification method for intrusive drones in ecological protection areas, and the specific technical solution adopted is as follows:

[0009] This invention provides an intelligent identification method for intrusive drones in ecological protection areas, comprising:

[0010] Based on the static and dynamic characteristics of the heat source target in the infrared images of the ecological reserve at various times, it is determined whether the acoustic detection device is triggered at each time; the static characteristics include shape and temperature distribution, and the dynamic characteristics are the motion trajectory.

[0011] After the acoustic detection device is triggered, the time-domain sound signals collected by the acoustic detection device at each moment are processed into Mel spectrograms.

[0012] The energy characteristics of the Mel spectrum are determined based on the uniformity of energy distribution within each Mel band in the Mel spectrum.

[0013] Based on the correlation between the energy characteristics and operating speed of the heat source target, the probability of drone intrusion into the ecological protection zone is obtained.

[0014] The drone intrusion identification result is determined based on the likelihood of drone intrusion.

[0015] In an exemplary embodiment, determining whether an acoustic detection device is triggered at each moment based on the static and dynamic characteristics of the heat source target in the infrared images of the ecological protection area at each moment includes:

[0016] Based on the shape of the heat source target in the infrared image at each time, the shape regularity of the heat source target at each time is obtained;

[0017] The disorder of temperature distribution in the infrared image of the heat source target at each time moment is determined, and combined with the shape regularity, the degree of UAV suspicion of the heat source target at each time moment is obtained; the degree of UAV suspicion is positively correlated with both the disorder of temperature distribution and the shape regularity.

[0018] Based on the difference in motion curvature between two adjacent moments in the motion trajectory of the heat source target at each moment in the local time window, the degree of chaos in the motion trajectory of the heat source target at each moment is obtained.

[0019] Based on the suspected drone level and the degree of motion trajectory disorder, the acoustic detection trigger probability of the heat source target at each moment is obtained; the acoustic detection trigger probability is positively correlated with both the suspected drone level and the degree of motion trajectory disorder.

[0020] Based on the probability of acoustic detection triggering at each moment of the heat source target, determine whether the acoustic detection device is triggered at each moment.

[0021] In an exemplary embodiment, the process of obtaining the shape regularity includes:

[0022] Determine the minimum bounding rectangle of the heat source target in the infrared image;

[0023] Obtain the aspect ratio of the minimum bounding rectangle as the shape regularity of the heat source target.

[0024] In one exemplary embodiment, the process of obtaining the temperature distribution disorder includes:

[0025] Determine the temperature gradient characteristics of each pixel in the infrared image of the heat source target;

[0026] The influence weight of the temperature gradient feature of each pixel is obtained from the temperature value of each pixel in the infrared image of the heat source target; the influence weight is positively correlated with the temperature value.

[0027] Based on the influence weight of the temperature gradient features of each pixel, the temperature gradient features of each pixel are weighted and summed to obtain the disorder of temperature distribution of the heat source target in the infrared image at each time.

[0028] In an exemplary embodiment, the process of obtaining the degree of motion trajectory disorder includes:

[0029] Based on the difference in the curvature of movement between two adjacent moments within the local time window of the heat source target at each moment, the characteristics of the difference in curvature of movement between two adjacent moments are obtained.

[0030] Based on the frame difference method, the number of moving pixels in the infrared image of the heat source target at each time moment is determined;

[0031] Based on the difference in the number of moving pixels between two adjacent moments in the local time window of the heat source target at each moment, a reference weight for the difference in the curvature of movement between two adjacent moments in the local time window of the heat source target at each moment is obtained; the reference weight is inversely correlated with the difference in the number of moving pixels.

[0032] Based on the reference weights of the differences in motion curvature between two adjacent moments within the local time window of each moment of the heat source target, the differences in motion curvature between two adjacent moments within the local time window of each moment of the heat source target are weighted and summed to obtain the degree of chaos in the motion trajectory of the heat source target at each moment.

[0033] In an exemplary embodiment, determining whether an acoustic detection device is triggered at each moment based on the acoustic detection trigger probability of the heat source target at each moment includes: determining that the acoustic detection device is triggered when the acoustic detection trigger probability is greater than a preset trigger probability threshold.

[0034] In an exemplary embodiment, the step of processing the time-domain sound signals collected by the acoustic detection device at each moment into a Mel spectrogram includes: processing the time-domain sound signals of local time windows at each moment into Mel spectrograms at each moment.

[0035] In an exemplary embodiment, the process of obtaining the energy characteristics of the Mel spectrum includes:

[0036] The Mel frequency bands in the Mel spectrogram are converted to grayscale to obtain the regional grayscale map of each Mel frequency band.

[0037] The uniformity of energy distribution in each Mel frequency band is obtained from the standard deviation of gray values ​​in the regional grayscale images of each Mel frequency band; the uniformity of energy distribution is inversely correlated with the standard deviation of gray values.

[0038] When the maximum energy distribution uniformity meets the preset condition, the energy of the Mel frequency band corresponding to the maximum energy distribution uniformity is obtained, and the energy characteristics of the Mel spectrum are obtained.

[0039] In an exemplary embodiment, determining the likelihood of a drone intrusion into an ecological protection zone based on the correlation between the energy characteristics of the heat source target and changes in its operating speed includes:

[0040] Determine the differences in energy characteristics of the Mel spectrograms at two adjacent time points, as well as the differences in the running speed of the heat source target at two adjacent time points;

[0041] Based on the difference between the energy characteristics and the operating speed, the probability of drone intrusion into the ecological protection zone at two adjacent moments is obtained; the probability of drone intrusion is inversely correlated with the difference between the energy characteristics and the operating speed.

[0042] In an exemplary embodiment, determining the drone intrusion identification result based on the drone intrusion probability includes: when a preset number of consecutive drone intrusion probabilities are all greater than a preset intrusion probability threshold, it is determined that there is a drone intrusion in the ecological protection area.

[0043] This invention has the following beneficial effects: By determining whether to trigger acoustic detection equipment based on the static and dynamic characteristics of heat source targets in infrared images of ecological protection areas, this invention achieves initial screening and trigger control of infrared images. After triggering the acoustic detection equipment, the time-domain sound signal is processed into a Mel spectrogram. Based on the energy characteristics shown in the Mel spectrogram, combined with the correlation between the energy characteristics of the heat source target and the change in running speed, the probability of UAV intrusion into the ecological protection area is obtained, achieving fine acoustic analysis. Based on cross-modal physical correlation fusion, the final UAV intrusion identification result is obtained. This invention creatively solves the limitations of single detection methods in ecological protection area scenarios. By utilizing dual-modal information complementarity and physical correlation verification, the identification accuracy can be improved and the false alarm rate can be reduced. Moreover, the acoustic detection equipment is not always in operation, but works on demand, which can improve energy efficiency and effectively cope with complex acoustic and thermal background interference in ecological protection areas, thus improving environmental adaptability. Attached Figure Description

[0044] Figure 1 This is a flowchart of an intelligent identification method for intrusive drones in ecological protection areas provided by an embodiment of the present invention;

[0045] Figure 2 This is a flowchart illustrating the specific implementation of step S1 provided in one embodiment of the present invention;

[0046] Figure 3 This is a flowchart illustrating the process of obtaining temperature distribution disorder according to an embodiment of the present invention;

[0047] Figure 4 This is a flowchart illustrating the process of obtaining the degree of motion trajectory disorder according to an embodiment of the present invention;

[0048] Figure 5 This is a flowchart illustrating the acquisition of energy characteristics provided in one embodiment of the present invention. Detailed Implementation

[0049] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the specific implementation methods, structures, features, and effects of the present invention are described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0050] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. All data and information collected in this application have been obtained with full consent.

[0051] This embodiment provides an intelligent identification method for intruding drones in ecological protection zones, applied to the detection and identification of drone intrusions in ecological protection zones. In a specific application, when deploying an infrared detection system in an ecological protection zone, an infrared thermal imager with an uncooled vanadium oxide (VOx) microbolometer as its core is selected. The infrared thermal imager is deployed at high points such as existing communication towers in the ecological protection zone, or at the top of specially established monitoring pillars. Based on the area of ​​the ecological protection zone and the monitoring range of the infrared thermal imager, the distance interval between adjacent infrared thermal imagers and the number of infrared thermal imagers are determined to achieve comprehensive monitoring of the ecological protection zone or monitoring of key areas. In an exemplary embodiment, the infrared thermal imager can be integrated into a heavy-duty gimbal, possessing 360° continuous horizontal rotation and a wide range of pitch capabilities, enabling all-weather three-dimensional monitoring, and equipped with a 4G / 5G or microwave wireless communication module to achieve real-time data transmission.

[0052] In terms of acoustic detection, multiple acoustic detection devices are deployed at pre-set monitoring locations within the ecological protection zone, or at key locations such as valleys, dense forest edges, and core area boundaries. Each acoustic detection device must be equipped with a microphone array. It should be understood that infrared thermal imagers and acoustic detection devices can be set up in pairs, with the acoustic detection devices positioned around their corresponding infrared thermal imagers. Once activated, the acoustic detection devices can collect sound signals from the vicinity of the corresponding infrared thermal imagers. Each infrared thermal imager and each acoustic detection device is connected to the data processing equipment, forming a hardware system. During operation, the system first relies on the infrared thermal imagers to conduct real-time monitoring of the ecological protection zone, identifying abnormal targets through thermal imaging data.

[0053] It should be understood that the infrared thermal imager acquires infrared images according to a preset sampling frequency, which can be determined empirically. The acoustic detection device, after startup, also acquires time-domain sound signals according to a preset sampling frequency. When the infrared thermal imager and acoustic detection device operate simultaneously, they can sample synchronously at the same frequency.

[0054] like Figure 1 As shown in the figure, the intelligent identification method for intruding drones in ecological protection areas provided in this embodiment includes the following steps:

[0055] Step S1: Based on the static and dynamic characteristics of the heat source target in the infrared images of the ecological reserve at each time, determine whether the acoustic detection equipment is triggered at each time.

[0056] Step S2: After triggering the acoustic detection device, process the time-domain sound signals collected by the acoustic detection device at each moment into a Mel spectrogram;

[0057] Step S3: Determine the energy characteristics of the Mel spectrum based on the uniformity of energy distribution within each Mel band in the Mel spectrum.

[0058] Step S4: Based on the correlation between the energy characteristics of the heat source target and the changes in its operating speed, the probability of drone intrusion into the ecological protection zone is obtained;

[0059] Step S5: Determine the drone intrusion identification result based on the likelihood of drone intrusion.

[0060] The following detailed explanation of each step, in conjunction with the accompanying drawings, will elaborate on each step.

[0061] Step S1: Based on the static and dynamic characteristics of the heat source target in the infrared images of the ecological reserve at each time, determine whether the acoustic detection equipment is triggered at each time.

[0062] Taking any infrared thermal imager as an example, the imager acquires infrared images at various times according to a preset sampling frequency, resulting in consecutive frames of infrared images. The frame difference method, by subtracting consecutive frames, can efficiently separate moving pixel regions. Therefore, the frame difference method is used to obtain the moving pixel regions in the infrared images. Each moving pixel region obtained from the infrared images is defined as a heat source target, serving as the object of subsequent data analysis. This yields the heat source targets in the infrared images at each time point. It should be understood that there can be only one heat source target or multiple heat source targets in the infrared images; when there are multiple targets, each heat source target is processed separately.

[0063] In infrared images, heat sources are essentially regions composed of multiple pixels, and each pixel represents a temperature value. This embodiment converts the infrared image to grayscale; therefore, the higher the grayscale value of a pixel, the higher its temperature. Furthermore, this embodiment normalizes the grayscale values ​​of the converted infrared image, for example, by dividing the grayscale value of each pixel by 255, ensuring that the grayscale value of each pixel falls within the range of 0-1. This eliminates the influence of dimensions, making the grayscale values ​​of each pixel dimensionless data, which facilitates subsequent processing. The grayscale values ​​of each pixel in the infrared images mentioned below are all normalized dimensionless data.

[0064] In embodiments of the present invention, whenever "normalization" or "normalization" is mentioned, it is considered a standard term. Unless otherwise specified, functions can use the maximum and minimum value normalization method. The maximum and minimum values ​​on which the calculation depends are dynamically determined based on the dataset within the current local time window or a sliding time window of a preset length.

[0065] It should be understood that the identified heat source target may be interfering factors such as birds, other animals, or falling leaves. However, the static and dynamic characteristics of the drone differ significantly from these interfering factors. Therefore, this embodiment determines whether the acoustic detection device is triggered at each moment based on the static and dynamic characteristics of the heat source target in the infrared image at each time. Static characteristics include shape and temperature distribution, while dynamic characteristics are the motion trajectory. In an exemplary embodiment, such as... Figure 2 As shown, the following is a specific implementation process of step S1:

[0066] Step S11: Based on the shape of the heat source target in the infrared image at each time moment, obtain the shape regularity of the heat source target at each time moment.

[0067] For drones, especially multi-rotor models, thermal imaging characteristics are very prominent: the main heat sources are the motors and electronic speed controllers, thus appearing as multiple (e.g., four) bright spots in the image. These spots exhibit a high degree of spatial regularity, such as symmetrical square or cross-shaped geometric configurations, with relatively uniform distance and brightness between them. Correspondingly, if the minimum bounding rectangle of a drone in an infrared image is obtained, it is close to a square. Conversely, moving birds, other animals, and other biological heat sources typically appear as a single, irregularly shaped block of bright light. Correspondingly, if the minimum bounding rectangle of moving birds, hot leaves, or other objects in an infrared image is obtained, it deviates significantly from a square.

[0068] In an exemplary embodiment, for an infrared image at any given time, the minimum bounding rectangle of the heat source target region in the infrared image is obtained, and the length and width of the minimum bounding rectangle are also obtained. The ratio of the width to the length is calculated as the width-to-length ratio. The value range of the width-to-length ratio is (0,1]. The larger the value, the closer the shape of the minimum bounding rectangle is to a square, that is, the more regular the shape of the minimum bounding rectangle. Therefore, the width-to-length ratio of the minimum bounding rectangle is used as the shape regularity of the heat source target. Thus, the shape regularity of the heat source target at each time is obtained. The higher the shape regularity of the heat source target at each time, the greater the probability that the heat source target belongs to a drone in the infrared image at the corresponding time. Therefore, the higher the suspicion degree of the heat source target as a drone at the corresponding time, and the suspicion degree of drone is positively correlated with the shape regularity.

[0069] Step S12: Determine the disorder of temperature distribution of the heat source target in the infrared image at each time moment, and combine it with the regularity of shape to obtain the degree of suspicion of the UAV of the heat source target at each time moment.

[0070] It should be understood that relying solely on the regularity of the shape of the heat source target may not be sufficient for accurate drone identification, as some drones may not have particularly regular shapes. To further improve the accuracy of drone identification, it is also necessary to consider the temperature distribution of the heat source target in the infrared image.

[0071] For birds, their bodies are unified organisms, and their body temperature is maintained by metabolism. Therefore, the temperature distribution throughout their bodies (except for the beak, feet, and other extremities) is relatively uniform and continuous. In infrared images, birds typically appear as a single, irregularly shaped heat patch with a gentle temperature gradient. Drones, however, are completely different. Their heat distribution is more chaotic: the motors and electronic speed controllers are high-temperature points, while the arms, fuselage, and payload (especially plastic and carbon fiber components) are much cooler, close to ambient temperature. Therefore, in infrared images, drones appear as a combination of bright, high-temperature hotspots (hotspots with high temperatures) and low-temperature areas, resulting in a highly uneven overall temperature distribution and a strong contrast. Therefore, if the internal temperature distribution of a heat source target is uneven, it is far more likely to be a drone than a bird with a uniform temperature distribution. Correspondingly, when acquiring the temperature distribution disorder of heat source targets in infrared images at various times, the temperature distribution disorder of drones is higher than that of organisms such as birds. In an exemplary embodiment, such as Figure 3 As shown, the following is a process for obtaining the disordered temperature distribution of a heat source target, including:

[0072] Step S121: Determine the temperature gradient characteristics of each pixel in the infrared image of the heat source target.

[0073] For any given moment in the infrared image, the temperature gradient value of each pixel representing the heat source target is obtained. Since temperature is represented by grayscale values, the gradient value of each pixel is used as its temperature gradient value. The larger the gradient value, the more drastic the grayscale change, i.e., the stronger the temperature change, indicating a more uneven temperature distribution, and the more likely the heat source target is a drone.

[0074] The temperature gradient values ​​of each pixel in the infrared image of the heat source target are normalized, and the normalized results are used as the temperature gradient features of each pixel. Since the grayscale values ​​of each pixel in the infrared image of the heat source target are normalized dimensionless data, verification shows that the numerical range of the temperature gradient values ​​of each pixel is 0- Therefore, the normalization method for the temperature gradient values ​​of each pixel can be: .

[0075] Step S122: Obtain the influence weight of the temperature gradient feature of each pixel point from the temperature value of each pixel point in the infrared image of the heat source target.

[0076] The temperature value (grayscale value) of each pixel in the infrared image of the heat source target is obtained. A larger grayscale value has a greater impact on the temperature gradient characteristics and carries more weight when determining the disorder of temperature distribution. Therefore, the influence weight of the temperature gradient characteristics of each pixel is obtained from the temperature value of the heat source target in the infrared image, and the influence weight is positively correlated with the temperature value. Based on the above logic, a specific calculation method for the influence weight is given below: Calculate the sum of the grayscale values ​​of all pixels of the heat source target in the infrared image, and then calculate the ratio of the grayscale value of each pixel of the heat source target in the infrared image to this sum. This ratio is used as the influence weight of each pixel of the heat source target in the infrared image. This method not only limits the numerical range of the influence weight of each pixel of the heat source target in the infrared image to 0-1, but also ensures that the sum of the influence weights of each pixel of the heat source target in the infrared image is 1.

[0077] It should be understood that if the sum of the gray values ​​of all pixels in the infrared image of the heat source target is 0, it means that the identified heat source target does not emit heat, and it is directly determined that there is no drone in the infrared image, and no further data processing is performed.

[0078] Step S123: Based on the influence weight of the temperature gradient features of each pixel, perform a weighted summation of the temperature gradient features of each pixel to obtain the disorder of temperature distribution of the heat source target in the infrared image at each time.

[0079] Based on the influence weight of the temperature gradient features of each pixel, the temperature gradient features of each pixel are weighted and summed to obtain the result of the disordered temperature distribution of the heat source target in the infrared image. The calculation formula is as follows:

[0080] ;

[0081] in, This indicates the disorder of temperature distribution in the infrared image of the heat source target at time t. This represents the influence weight of the temperature gradient feature of the q-th pixel in the infrared image at time t. Let Q represent the temperature gradient feature of the q-th pixel in the infrared image of the heat source target at time t, where Q represents the number of pixels of the heat source target in the infrared image at time t.

[0082] Using the above process, the temperature distribution disorder of the heat source target in the infrared image at each time moment is obtained. The larger the value of the temperature distribution disorder, the more uneven the temperature distribution of the heat source target in the infrared image, and the greater the probability that the heat source target belongs to a drone in the infrared image at the corresponding time moment. Therefore, the higher the suspicion degree of the heat source target as a drone at the corresponding time moment, and the suspicion degree of drone is positively correlated with the temperature distribution disorder.

[0083] Based on the disordered temperature distribution and regularity of the shape of the heat source target in the infrared images at various times, the drone suspicion level of the heat source target at each time is obtained. Based on the above logical analysis, a specific calculation method for the drone suspicion level is given below: Taking time t as an example, calculate the average of the disordered temperature distribution and the regularity of the shape of the heat source target in the infrared image at time t. The result is used as the drone suspicion level of the heat source target at time t.

[0084] This allows us to obtain the drone suspicion level of the heat source target at various times. By analyzing the static characteristics of the heat source target at various times, we obtain the drone suspicion level based on the static characteristics.

[0085] Step S13: Based on the difference in motion curvature between two adjacent moments in the motion trajectory of the heat source target at each moment in the local time window, the degree of chaos in the motion trajectory of the heat source target at each moment is obtained.

[0086] Taking time t as an example, a local time window for time t is obtained. In an exemplary embodiment, time t is taken as the end time, a time before time t is taken as the start time, and the local time interval between the start time and the end time is taken as the local time window for time t. The length of the local time window, i.e., the number of times it contains, is determined by the implementer based on experience.

[0087] Based on the dynamic characteristics reflected by the heat source target within a local time window at time t, specifically the difference in curvature between adjacent moments within the trajectory at time t, the degree of disorder in the heat source target's trajectory at time t is obtained. In an exemplary embodiment, such as... Figure 4 As shown, the following is a specific process for obtaining the degree of chaos in a motion trajectory:

[0088] Step S131: Based on the difference in the curvature of the movement between two adjacent moments in the local time window of the heat source target at each moment, obtain the characteristics of the difference in the curvature of the movement between two adjacent moments.

[0089] The flight control mechanisms of drones and birds are fundamentally different. Bird flight relies on a complex coordination of muscles, aerodynamics, and neural reflexes. Its movement changes are a continuous and smooth adjustment process. Therefore, the curvature changes caused by changes in its heading are gradual, gentle, and continuous. The trajectory of the bird is usually a smooth curve with relatively gentle curvature changes.

[0090] Conversely, the flight of a drone is achieved by the flight controller directly driving the motor through algorithms. Its steering commands are digital, pursuing precision and efficiency. Moreover, after a drone intrudes into an ecological protection zone, it often needs to collect images of different areas within the ecological protection zone. In addition, the drone control is relatively flexible and fast. Therefore, the drone's movement trajectory usually shows a curve with more dramatic changes. The curvature changes more drastically, exhibiting characteristics of non-smooth, step-like, or with obvious inflection points.

[0091] The curvature of the heat source target at each moment within a local time window at time t is obtained and defined as the motion curvature. In an exemplary embodiment, the position of the heat source target in the infrared image at each moment within the local time window at time t is obtained, and the position of the center point of the heat source target is taken as the position of the heat source target. In this embodiment, the center point of the smallest bounding rectangle of the heat source target can be taken as the center point of the heat source target. Thus, based on the position of the heat source target in the infrared image at each moment within the local time window at time t, the motion trajectory of the heat source target in the local time window at time t is obtained.

[0092] Based on the trajectory of the heat source target within its local time window at time t, the curvature of the heat source target at each time point within that local time window is obtained. Specifically, for any three consecutive center points on the trajectory... , as well as The curvature at time i can be calculated as the reciprocal of the radius of the unique circumcircle determined by these three points. If the three points are collinear, the curvature is 0. Curvature represents the degree of bending of the trajectory at a certain point; the greater the curvature, the greater the degree of bending.

[0093] The difference in motion curvature (i.e., the absolute value of the difference in curvature) between any two adjacent moments within the local time window of the heat source target at moment t is calculated. Then, based on the difference in motion curvature between any two adjacent moments within the local time window of the heat source target at moment t, the feature of the difference in motion curvature between any two adjacent moments is obtained. In an exemplary embodiment, the difference in motion curvature is normalized, and the normalized result is used as the feature of the difference in motion curvature. In an exemplary embodiment, a maximum / minimum value normalization method can be used here.

[0094] Step S132: Based on the frame difference method, determine the number of moving pixels in the infrared image of the heat source target at each time moment.

[0095] This embodiment also needs to consider the reliability of the curvature difference characteristics between every two adjacent moments within the local time window of the heat source target at moment t, that is, the reliability of the curvature change between every two adjacent moments. This is because if the drone's own trajectory at two adjacent moments is superimposed with the additional motion exerted by factors such as wind, it becomes impossible to separate the "controllable flight characteristics" that reflect its true intention. Specifically, curvature analysis can distinguish between drones (mechanical) and birds (fluid) because it captures the target's own steering model. However, in strong winds, the actual movement of a drone is the result of its own power and the combined effect of wind force. Its trajectory may be blown away by the wind, resulting in non-autonomous and sudden drift. The "curvature change" calculated at this time contains a lot of noise, reflecting the disturbance of the wind field rather than the inherent control mode of the drone. This data obtained under interference is "contaminated" and not representative. Using it for pattern recognition will seriously mislead the algorithm.

[0096] Therefore, the number of moving pixels in the infrared image of the heat source target at each time step is first determined based on the frame difference method. Specifically, for any given time step, the absolute value of the difference between the grayscale values ​​of the same pixel in the infrared image at that time and its adjacent time steps is obtained using the frame difference method, resulting in a difference image. Then, the difference image is converted into a binary image according to a preset threshold to distinguish the moving region from the background. The preset threshold can be a fixed threshold or an adaptive threshold obtained through the Otsu's method. The number of pixels with a value of 1 in the binary image is counted as the number of moving pixels in the infrared image at that time step. This yields the number of moving pixels at each time step within the local time window of the heat source target at time step t.

[0097] Step S133: Based on the difference in the number of moving pixels between two adjacent moments in the local time window of the heat source target at each moment, obtain the reference weight of the movement curvature difference feature between two adjacent moments in the local time window of the heat source target at each moment.

[0098] For any two adjacent moments within the local time window of the heat source target at moment t, calculate the absolute value of the difference in the number of moving pixels between those two moments, which is taken as the difference in the number of moving pixels between those two moments. The larger the difference in the number of moving pixels, the more noise is included in the curvature change of the heat source target between those two moments, the less referential the curvature difference feature is, and the lower the reference value of the curvature change between those two moments. Therefore, the reference value of the curvature change between those two moments is inversely correlated with the difference in the number of moving pixels between those two moments. Based on this logic, a specific calculation method for the reference value of curvature change is given below:

[0099] ;

[0100] in, This represents the reference degree of curvature change between two adjacent time points (i.e., the i-th time point and the (i-1)-th time point). This represents a normalization function, such as a maximum / minimum value normalization function. This represents the number of pixels that moved at the (i-1)th time step of the heat source target. This represents the number of moving pixels of the heat source target at time i.

[0101] Then, based on the curvature change reference degree of each two adjacent moments within the local time window of the heat source target at time t, the reference weight of the moving curvature difference feature of each two adjacent moments within the local time window of the heat source target at time t is obtained. The higher the curvature change reference degree, the higher the desirability of the moving curvature difference feature of the corresponding two adjacent moments, that is, the larger the reference weight of the moving curvature difference feature, and the two are positively correlated. Based on the above logic, the following is a specific calculation method for the reference weight of the moving curvature difference feature: calculate the sum of the curvature change reference degrees of all two adjacent moments within the local time window of the heat source target at time t, and then calculate the ratio of the curvature change reference degree of each two adjacent moments within the local time window of the heat source target at time t to the sum value, which is used as the reference weight of the moving curvature difference feature of each two adjacent moments within the local time window of the heat source target at time t. In this way, not only can the reference weight of the movement curvature difference feature of the heat source target at time t be limited to the range of 0-1 between each two adjacent times in the local time window, but the sum of the reference weights of the movement curvature difference feature of the heat source target at time t in the local time window is also made to be 1.

[0102] Step S134: Based on the reference weight of the difference in motion curvature between two adjacent moments in the local time window of each moment of the heat source target, perform a weighted summation of the difference in motion curvature between two adjacent moments in the local time window of each moment of the heat source target to obtain the degree of chaos of the motion trajectory of the heat source target at each moment.

[0103] Based on the reference weights of the differences in motion curvature between two adjacent moments within the local time window of the heat source target at moment t, the differences in motion curvature between two adjacent moments within the local time window of the heat source target at moment t are weighted and summed to obtain the degree of motion trajectory disorder of the heat source target at moment t. The calculation formula is as follows:

[0104] ;

[0105] in, This indicates the degree of disorder in the trajectory of the heat source target at time t. This represents the difference in the curvature of movement of the heat source target at time t within a local time window, specifically between time i and time i-1. The reference weight represents the difference in the movement curvature characteristics between the i-th and (i-1)-th times within the local time window of the t-th time of the heat source target, where I represents the number of times within the local time window of the t-th time.

[0106] The higher the degree of disorder in the trajectory of the heat source target at time t, and the more irregular the movement of the heat source target at time t, the higher the probability that the heat source target belongs to a drone, which can suggest the unique mechanical control logic of the drone.

[0107] Step S14: Based on the degree of suspicion of the drone and the degree of disorder in its movement trajectory, obtain the acoustic detection trigger probability of the heat source target at each moment.

[0108] Through the above steps, the drone suspicion level and trajectory disorder level of the heat source target at each time point are obtained, thereby determining the acoustic detection trigger probability of the heat source target at each time point. The higher the probability that the heat source target is a drone, the higher the acoustic detection trigger probability. Therefore, the acoustic detection trigger probability is positively correlated with both the drone suspicion level and the trajectory disorder level. In an exemplary embodiment, the average of the drone suspicion level and trajectory disorder level of the heat source target at time t is calculated, and the result is used as the acoustic detection trigger probability of the heat source target at time t. By averaging, the static and dynamic characteristics of the heat source target are fused to obtain the acoustic detection trigger probability of the heat source target at each time point.

[0109] Step S15: Based on the acoustic detection trigger probability of the heat source target at each moment, determine whether the acoustic detection device is triggered at each moment.

[0110] The higher the probability of acoustic detection triggering at each moment, the more likely the heat source target is to be a drone at that moment, and the more necessary it is to trigger the acoustic detection device to start sound detection. In an exemplary embodiment, this embodiment presets a trigger probability threshold, which is used to determine whether the probability of acoustic detection triggering at each moment is high. The value of the preset trigger probability threshold ranges from 0 to 1, and the specific value is set according to the actual judgment needs. This embodiment uses 0.7 as an example for explanation. When the probability of acoustic detection triggering at a certain moment is greater than the preset trigger probability threshold, it is determined that the acoustic detection device will be triggered at that moment, and an acoustic detection device trigger command is output to control the acoustic detection device to start. It should be understood that this embodiment can only control the acoustic detection device matched with the infrared thermal imager to start. As another implementation, in addition to starting the matched acoustic detection device, other acoustic detection devices within a preset distance can also be controlled to start.

[0111] If the probability of acoustic detection triggering at a certain moment is not greater than a preset triggering probability threshold, the system determines that the acoustic detection device will not be triggered at that moment. In this case, the system maintains infrared tracking of the heat source target and continuously updates its acoustic detection triggering probability in subsequent moments; if the updated probability value exceeds the preset triggering probability threshold, the acoustic detection device is triggered. If the target is lost or its probability value remains below a set termination threshold for multiple consecutive cycles, the system terminates tracking of the target.

[0112] Step S2: After triggering the acoustic detection device, process the time-domain sound signals collected by the acoustic detection device at each moment into a Mel spectrogram.

[0113] After the acoustic detection device is triggered, it starts operating and begins collecting sound signals from the surrounding environment. This embodiment acquires the time-domain sound signals within local time windows at each moment, which are then used as the time-domain sound signals for that moment. Specifically, after the acoustic detection device is triggered at time t, it collects an audio segment centered at time t with a preset duration (e.g., 2 seconds). This audio segment serves as the time-domain sound signal at time t for subsequent processing. Then, the time-domain sound signals within the local time windows at each moment are processed into Mel spectrograms for that moment. Processing time-domain sound signals into Mel spectrograms is a prior art technique, and a simplified description of the process follows.

[0114] First, the time-domain audio signal is processed into a digital signal. To balance the spectrum and enhance high-frequency components, a first-order high-pass filter can be used for pre-emphasis, thereby enhancing high-frequency resonants and sound clarity. Audio signals can be considered approximately stationary over short periods (e.g., 10-30 milliseconds). To analyze how its frequency components change over time, the long signal is divided into multiple short segments. During Mel spectrogram processing, the frame length and frame shift (i.e., step size) need to be determined. Both can be obtained empirically. The frame length can be 20-40 milliseconds; for example, at a 16kHz sampling rate, 25 milliseconds corresponds to 16000 * 0.025 = 400 sampling points. To maintain continuity between frames, there will be overlap. Typically, the frame shift is half the frame length (e.g., 10 milliseconds). This overlap makes the result smoother.

[0115] Perform a Fourier transform on each frame of the audio signal to convert the time-domain audio signal into a frequency-domain representation, obtaining the energy distribution of the audio signal in different frequency components of that frame. The length of the FFT is usually equal to or greater than the frame length (achieved through zero padding). This yields a spectrogram, whose two dimensions are time (frame index) and linear frequency (Hz), and intensity is the amplitude or energy.

[0116] Constructing a Mel filter bank: In the frequency domain (Hz), define a set of triangular bandpass filters, typically 40-80 (e.g., 40). Smooth the frequency domain signal using the Mel filter bank, highlighting key frequencies. Then, perform logarithmic and discrete cosine transform operations to convert the linear frequencies (Hz) to Mel frequencies (Mel), uniformly dividing the frequency range on the Mel scale (e.g., Mel values ​​corresponding to 0 Hz to the Nyquist frequency). This yields the Mel spectrogram. In the Mel spectrogram, arrange the Mel energy calculated for each frame of the audio signal from left to right in chronological order, forming a two-dimensional matrix. The horizontal axis represents time (each unit is one frame), and the vertical axis represents the Mel frequency band index (from low to high frequencies). Each pixel value in the matrix represents Mel energy; higher pixel values, i.e., brighter areas, represent stronger Mel energy. This yields the Mel spectrogram for each time point.

[0117] Step S3: Determine the energy characteristics of the Mel spectrum based on the uniformity of energy distribution within each Mel band in the Mel spectrum.

[0118] For drones, the high-speed rotation of their rotors generates a fundamental frequency and a series of harmonics in their sound signal. These frequency components are stable and continuous, appearing as horizontal stripes with relatively uniform energy distribution and high brightness parallel to the time axis on a Mel spectrogram. In contrast, bird calls or flapping sounds are brief, variable, and have unstable frequency components, appearing as discontinuous, tilted, or curved patches or short lines on the spectrogram. Environmental noise (such as wind noise) is typically an unstructured diffuse background. Therefore, determining the existence of horizontal stripes in each Mel band of the Mel spectrogram is necessary; searching for horizontal stripes is essentially a form of efficient pattern filtering.

[0119] For any given time, the energy characteristics of the Mel spectrum are determined based on the uniformity of energy distribution within each Mel band in the Mel spectrum at that time. In an exemplary embodiment, such as... Figure 5 As shown, the following is a specific process for obtaining the energy characteristics of a Mel spectrogram:

[0120] Step S31: Convert each Mel frequency band in the Mel spectrogram to grayscale to obtain the regional grayscale map of each Mel frequency band.

[0121] It should be understood that the color intensity of pixels in each Mel frequency band of the Mel spectrogram represents the Mel energy of the pixel. Therefore, in this embodiment, each Mel frequency band in the Mel spectrogram is grayscaled to obtain a regional grayscale image of each Mel frequency band. The higher the grayscale value of a pixel, the higher its Mel energy. The grayscale value of each pixel in the regional grayscale image of each Mel frequency band is divided by 255 to normalize the grayscale value, making the grayscale value of each pixel in the regional grayscale image of each Mel frequency band dimensionless data. The grayscale values ​​of each pixel in the regional grayscale images of each Mel frequency band mentioned below are all normalized dimensionless data.

[0122] Step S32: Obtain the uniformity of energy distribution in each Mel band from the standard deviation of gray values ​​in the regional grayscale images of each Mel band.

[0123] For any Mel band, calculate the standard deviation of the gray values ​​in the regional grayscale image of that Mel band. The standard deviation of the gray values ​​characterizes the degree of disorder in the Mel energy distribution of the Mel band. Therefore, based on the standard deviation of the gray values ​​in the regional grayscale image of that Mel band, the uniformity of the energy distribution of that Mel band can be obtained. The uniformity of energy distribution is inversely correlated with the standard deviation of the gray values. In an example, the following is a specific method for calculating the uniformity of the energy distribution of a Mel band:

[0124] ;

[0125] in, This indicates the uniformity of energy distribution in the R-th Mel band. This represents the standard deviation of the grayscale values ​​in the R-th Mel band region. It should be understood that for a dimensionless sequence with a numerical range of 0-1, its standard deviation has a dimensionless range of 0-0.5. Alternatively, the standard deviation of the Mel energy of each pixel in the Mel band can be calculated, and then inverse correlation normalization can be performed to obtain the uniformity of the energy distribution in the Mel band.

[0126] Step S33: When the maximum energy distribution uniformity meets the preset condition, obtain the energy of the Mel frequency band corresponding to the maximum energy distribution uniformity, and obtain the energy characteristics of the Mel spectrum.

[0127] In step S32, the energy distribution uniformity of each Mel band in the Mel spectrum is obtained, and the highest energy distribution uniformity is taken as the maximum energy distribution uniformity of the Mel spectrum. The maximum energy distribution uniformity indicates that the energy distribution of the corresponding Mel band is the most uniform among all Mel bands. Since a Mel band with a more uniform energy distribution is better able to display horizontal stripes, this embodiment presets an energy distribution uniformity threshold. This preset threshold is used to compare with the maximum energy distribution uniformity of the Mel spectrum to determine whether the maximum energy distribution uniformity of the Mel spectrum is high. The value range of this preset threshold is 0-1, and the specific value is set according to actual needs. This embodiment uses 0.8 as an example. The maximum energy distribution uniformity of the Mel spectrum satisfies the preset condition: the maximum energy distribution uniformity of the Mel spectrum is greater than or equal to the preset energy distribution uniformity threshold.

[0128] When the maximum energy distribution uniformity of the Mel spectrum meets a preset condition, the energy of the Mel band corresponding to the maximum energy distribution uniformity is obtained. Specifically, the energy value of each pixel in the Mel band corresponding to the maximum energy distribution uniformity is obtained, and then the average value of the energy values ​​is calculated as the energy of the Mel band corresponding to the maximum energy distribution uniformity. Since the energy value can be represented by the gray value of the pixel, the average gray value of each pixel in the Mel band corresponding to the maximum energy distribution uniformity is calculated as the energy feature of the Mel spectrum, i.e., the energy feature of the heat source target. Thus, the energy features of the heat source target at various times are obtained.

[0129] It should be understood that if the maximum uniformity of energy distribution does not meet the preset conditions, it can be directly determined that the possibility of drone intrusion does not meet the requirements, that is, there is no drone intrusion.

[0130] Step S4: Based on the correlation between the energy characteristics of the heat source target and the changes in its operating speed, the probability of drone intrusion into the ecological protection zone is obtained.

[0131] When a drone accelerates, its motors rotate at higher speeds to generate greater lift. This physical process leads to two observable simultaneous changes: first, the increased motor speed raises the fundamental and harmonic frequencies of its noise, increasing sound intensity, which is directly reflected in the increased overall brightness (i.e., energy intensity) of horizontal stripes on a Mel spectrogram; second, the increased power directly leads to a faster drone operating speed (i.e., flight speed). Therefore, based on the correlation between the energy characteristics of the heat source target and changes in operating speed, the likelihood of drone intrusion into the ecological protection zone can be determined.

[0132] In an exemplary embodiment, for any two adjacent time points, the energy characteristic difference of the Mel spectrum at those two time points is calculated, i.e., the energy characteristic difference of the heat source target at two adjacent time points. Specifically, the energy characteristic difference is the difference obtained by subtracting the energy characteristic of the next time point from the energy characteristic of the previous time point. The running speed difference of the heat source target at two adjacent time points is also calculated, specifically the difference obtained by subtracting the running speed of the next time point from the running speed of the previous time point. The running speed at a given time point can be the average of the running speeds at all times within a local time window of that time point. The running speed can be obtained using the frame difference method, where the position difference of the center point of the heat source target in two adjacent infrared images (i.e., the distance between the center points) is obtained using the frame difference method, and then divided by the time interval between the two adjacent time points to obtain the running speed. It should be understood that the running speed is a non-negative scalar value and has no direction.

[0133] The difference between energy characteristics and operating speed at two adjacent moments is determined. The smaller the difference, the more synchronized the changes in energy characteristics and operating speed, and the higher the probability of drone intrusion at adjacent moments. Therefore, the probability of drone intrusion is inversely correlated with the difference between energy characteristics and operating speed. Based on the above logical analysis, a specific method for calculating the probability of drone intrusion is given below:

[0134] ;

[0135] in, This represents the probability of drone intrusion at time j-1 and time j. This represents the energy characteristics of the heat source target at time j-1. This represents the energy characteristics of the heat source target at time j. This represents the difference in energy characteristics between time point (j-1) and time point (j). This represents the numerical value of the running speed of the heat source target at time j-1. This represents the numerical value of the running speed of the heat source target at time j. This represents the difference in running speed between time step (j-1) and time step (j). This represents a normalization function, such as a maximum / minimum value normalization function, which incorporates data from previous time points into the normalization calculation.

[0136] The smaller the value, the higher the synchronization rate, and the closer the calculated probability of drone intrusion is to 1.

[0137] Step S5: Determine the drone intrusion identification result based on the likelihood of drone intrusion.

[0138] Step S4 obtains the probability of drone intrusion into the ecological protection zone at two adjacent moments, achieving real-time detection of the drone intrusion probability. Based on the drone intrusion probability, the drone intrusion identification result is determined. Specifically, when a preset number of consecutive drone intrusion probabilities are all greater than a preset intrusion probability threshold, it is determined that a drone intrusion has occurred in the ecological protection zone. The preset number of consecutive intrusions and the preset intrusion probability threshold can both be set according to actual judgment needs. As an example, the preset number of consecutive intrusions can be 10, and the preset intrusion probability threshold can be 0.7. Therefore, when 10 consecutive drone intrusion probabilities are all greater than 0.7, it is determined that a drone intrusion has occurred in the ecological protection zone. It should be understood that the method of determining the drone intrusion identification result based on the drone intrusion probability is flexibly set by the implementer. For example, as another implementation method, the following judgment method can also be used: when the drone intrusion probability exceeds the preset intrusion probability threshold, it is determined that a drone intrusion has occurred in the ecological protection zone.

[0139] After determining that a drone has intruded into the ecological protection zone, as a specific application, subsequent drone tracking and locking can also be performed. For example, the system will immediately control a high-resolution optical zoom gimbal to automatically and continuously track the heat source target, ensuring that the drone remains centered in the frame. Subsequently, the system performs the following key operations in sequence: First, it activates a high-speed continuous shooting mode to capture a set of high-definition images from different focal lengths, including both the drone's overall attitude information and key local details such as body markings and modification marks; simultaneously, it starts high-definition video recording, continuously recording several seconds of flight behavior to clearly present its flight characteristics and movement patterns. All image data will be automatically integrated and packaged with metadata such as timestamps, GPS positioning information, and target movement trajectory, and transmitted in real time to the command center data server through the wireless private network deployed in the protection zone (including 4G / 5G mobile networks or microwave links). After the server parses the received data, it will immediately send key intelligence such as the drone's precise location, characteristic photos, and behavioral videos to the patrol team members' mobile terminals through both push notifications and SMS. This detailed on-site information provides patrol personnel with precise navigation and situational awareness for their missions, thereby achieving a fully automated closed loop from target discovery and tracking to intelligence distribution.

[0140] This embodiment also provides an intelligent identification system for intrusive drones in ecological protection areas, including: a memory and a processor; the memory is connected to the processor, and the memory is used to store program instructions; the processor is used to implement the steps in the above-described embodiment of the intelligent identification method for intrusive drones in ecological protection areas when the program instructions are executed.

[0141] In one exemplary embodiment, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the above embodiment of the intelligent identification method for intrusive drones in ecological protected areas.

[0142] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0143] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for intelligent identification of intruding drones in ecological protection zones, characterized in that, include: Based on the static and dynamic characteristics of heat source targets in infrared images of the ecological reserve at various times, it is determined whether acoustic detection equipment is triggered at each time. The static features include shape and temperature distribution, and the dynamic features are motion trajectories; After the acoustic detection device is triggered, the time-domain sound signals collected by the acoustic detection device at each moment are processed into Mel spectrograms. The energy characteristics of the Mel spectrum are determined based on the uniformity of energy distribution within each Mel band in the Mel spectrum. Based on the correlation between the energy characteristics and operating speed of the heat source target, the probability of drone intrusion into the ecological protection zone is obtained. Determine the drone intrusion identification result based on the likelihood of drone intrusion; The method of determining the likelihood of drone intrusion into the ecological protection zone based on the correlation between the energy characteristics of the heat source target and changes in its operating speed includes: Determine the differences in energy characteristics of the Mel spectrograms at two adjacent time points, as well as the differences in the running speed of the heat source target at two adjacent time points; Based on the difference between the energy characteristics and the operating speed, the probability of drone intrusion into the ecological protection zone at two adjacent moments is obtained; the probability of drone intrusion is inversely correlated with the difference between the energy characteristics and the operating speed.

2. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 1, characterized in that, The method of determining whether acoustic detection equipment is triggered at each moment based on the static and dynamic characteristics of heat source targets in infrared images of the ecological protection area at various times includes: Based on the shape of the heat source target in the infrared image at each time, the shape regularity of the heat source target at each time is obtained; The disorder of temperature distribution in the infrared image of the heat source target at each time moment is determined, and combined with the shape regularity, the degree of UAV suspicion of the heat source target at each time moment is obtained; the degree of UAV suspicion is positively correlated with both the disorder of temperature distribution and the shape regularity. Based on the difference in motion curvature between two adjacent moments in the motion trajectory of the heat source target at each moment in the local time window, the degree of chaos in the motion trajectory of the heat source target at each moment is obtained. Based on the suspected drone level and the degree of motion trajectory disorder, the acoustic detection trigger probability of the heat source target at each moment is obtained; the acoustic detection trigger probability is positively correlated with both the suspected drone level and the degree of motion trajectory disorder. Based on the probability of acoustic detection triggering at each moment of the heat source target, determine whether the acoustic detection device is triggered at each moment.

3. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 2, characterized in that, The process of obtaining the shape regularity includes: Determine the minimum bounding rectangle of the heat source target in the infrared image; Obtain the aspect ratio of the minimum bounding rectangle as the shape regularity of the heat source target.

4. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 2, characterized in that, The process of obtaining the temperature distribution disorder includes: Determine the temperature gradient characteristics of each pixel in the infrared image of the heat source target; The influence weight of the temperature gradient feature of each pixel is obtained from the temperature value of each pixel in the infrared image of the heat source target; the influence weight is positively correlated with the temperature value. Based on the influence weight of the temperature gradient features of each pixel, the temperature gradient features of each pixel are weighted and summed to obtain the disorder of temperature distribution of the heat source target in the infrared image at each time.

5. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 2, characterized in that, The process of obtaining the degree of disorder in the motion trajectory includes: Based on the difference in the curvature of movement between two adjacent moments within the local time window of the heat source target at each moment, the characteristics of the difference in curvature of movement between two adjacent moments are obtained. Based on the frame difference method, the number of moving pixels in the infrared image of the heat source target at each time moment is determined; Based on the difference in the number of moving pixels between two adjacent moments in the local time window of the heat source target at each moment, a reference weight for the difference in the curvature of movement between two adjacent moments in the local time window of the heat source target at each moment is obtained; the reference weight is inversely correlated with the difference in the number of moving pixels. Based on the reference weights of the differences in motion curvature between two adjacent moments within the local time window of each moment of the heat source target, the differences in motion curvature between two adjacent moments within the local time window of each moment of the heat source target are weighted and summed to obtain the degree of chaos in the motion trajectory of the heat source target at each moment.

6. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 2, characterized in that, The step of determining whether the acoustic detection device is triggered at each moment based on the acoustic detection trigger probability of the heat source target at each moment includes: determining that the acoustic detection device is triggered when the acoustic detection trigger probability is greater than a preset trigger probability threshold.

7. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 1, characterized in that, The process of processing the time-domain sound signals collected by the acoustic detection device at each moment into Mel spectrograms includes: processing the time-domain sound signals of local time windows at each moment into Mel spectrograms at each moment.

8. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 1, characterized in that, The process of obtaining the energy characteristics of the Mel spectrum includes: The Mel frequency bands in the Mel spectrogram are converted to grayscale to obtain the regional grayscale map of each Mel frequency band. The uniformity of energy distribution in each Mel frequency band is obtained from the standard deviation of gray values ​​in the regional grayscale images of each Mel frequency band; the uniformity of energy distribution is inversely correlated with the standard deviation of gray values. When the maximum energy distribution uniformity meets the preset condition, the energy of the Mel frequency band corresponding to the maximum energy distribution uniformity is obtained, and the energy characteristics of the Mel spectrum are obtained.

9. The intelligent identification method for intrusive drones in ecological protection zones as described in claim 1, characterized in that, The step of determining the drone intrusion identification result based on the drone intrusion probability includes: when a preset number of consecutive drone intrusion probabilities are all greater than a preset intrusion probability threshold, it is determined that there is a drone intrusion in the ecological protection area.

Images