Unmanned aerial vehicle detection method and device based on multiple sensors and confidence evaluation

Abstract

The invention discloses an unmanned aerial vehicle detection method and device based on multiple sensors and confidence evaluation. The method comprises the following steps: S1, data acquisition; s2, data preprocessing; s3, target type probability calculation; s4, confidence coefficient calculation; s5, carrying out weight calculation; s6, target type probability fusion; and S7, target type judgment. In order to solve the detection problem of the unmanned aerial vehicle in a complex environment (such as rainstorm, high temperature and night), a solution integrating multi-modal sensor fusion (millimeter wave radar, visible light camera and infrared thermal imaging) and confidence coefficient dynamic evaluation is provided, and the reliability and accuracy of multi-sensor fusion are ensured in combination with real-time environmental factors.

Description

Technical Field

[0001] This invention relates to drone detection, and in particular to a drone detection method and apparatus based on multi-sensor and confidence assessment. Background Technology

[0002] With the increasingly widespread application of drones, drone activity in the low-altitude airspace is becoming more frequent. At the same time, the number of public safety incidents caused by drones is also on the rise. Therefore, real-time and accurate detection of drones has become an urgent need to ensure public safety and maintain order in specific areas.

[0003] Drones are typical "low, slow, and small" targets, flying at low altitudes, at slow speeds, and relatively small in size. In low-altitude environments, many other objects of similar size and speed to drones exist, such as birds and balloons, significantly increasing the difficulty of drone detection. Currently, multi-sensor technology is increasingly being applied in drone detection. Millimeter-wave radar can provide information such as target distance, speed, and angle, offering high accuracy in spatial positioning. Visible light cameras can acquire visual features such as target texture and shape, aiding in intuitive target identification. Infrared thermal imagers utilize the difference in thermal radiation between the target and the background for detection, enabling operation at night and in adverse weather conditions. Theoretically, fully integrating data from these multiple sensors could significantly improve the accuracy and reliability of drone detection. However, achieving efficient multi-sensor data fusion presents numerous technical challenges that need to be overcome.

[0004] The existing technology has the following problems: (1) Insufficient sensor reliability assessment: Existing technologies often lack quantitative assessment of the reliability of sensor data based on the current environment. It is difficult to adjust sensor weights or fusion strategies based on real-time environmental factors (such as light, rainfall, temperature, and wind speed), which can easily lead to misjudgment or omission in complex environments.

[0005] (2) Poor adaptability of multi-sensor fusion: Existing methods are difficult to cope with complex and ever-changing environments, especially under extreme conditions such as night, rain, fog, and strong winds, the system performance is unstable and the detection accuracy will drop significantly. Summary of the Invention

[0006] To address the aforementioned problems, this invention provides a drone detection method based on multi-sensor and confidence assessment.

[0007] This invention provides the following technical solution: A drone detection method based on multi-sensor and confidence assessment includes the following steps: S1. Collect sensor data of the target and meteorological data of the target location; S2. Preprocess the sensor data and meteorological data to obtain the preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. S3. Input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector; S4. Calculate the confidence level based on the meteorological data level and the target characteristic temperature level; S5. Determine if the confidence level is greater than the confidence level threshold. If so, calculate the weight based on the confidence level; otherwise, issue an alarm. S6. Multiply the target type probability distribution vector by the weight to obtain the target type fusion probability distribution vector; S7. Determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

[0008] Further, step S1 specifically involves: collecting sensor data of the target and meteorological data of the target location. The sensor data includes radar echo signals, images, and thermal images, and the meteorological data includes rainfall, light intensity, ambient temperature, and wind speed.

[0009] Further, step S2 specifically involves: preprocessing the radar echo signal to obtain an RDM image and a micro-Doppler time-frequency image; preprocessing the image to obtain a standardized RGB image; preprocessing the thermal image to obtain a thermal grayscale image; preprocessing the meteorological data to obtain rainfall level, illumination level, temperature level, and wind speed level; obtaining the target characteristic temperature based on the thermal image; and obtaining the target characteristic temperature level based on the target characteristic temperature. Further, step S3 specifically involves: inputting the RDM image and the micro-Doppler time-frequency image into the first target type probability distribution model to obtain the first target type probability distribution vector; inputting the standardized RGB image into the second target type probability distribution model to obtain the second target type probability distribution vector; and inputting the thermal imaging grayscale image into the third target type probability distribution model to obtain the third target type probability distribution vector. Further, step S4 specifically involves: calculating the first confidence level based on the rainfall level and wind speed level; calculating the second confidence level based on the rainfall level and light intensity level; and calculating the third confidence level based on the target characteristic temperature level, temperature level, and wind speed level. Further, step S5 specifically involves: determining whether the first confidence level, the second confidence level, and the third confidence level are all greater than the confidence threshold; if so, calculating the first weight, the second weight, and the third weight based on the first confidence level, the second confidence level, and the third confidence level, respectively; otherwise, stopping the operation and issuing an alarm. Further, step S6 specifically involves multiplying the first target type probability distribution vector by the first weight, the second target type probability distribution vector by the second weight, and the third target type probability distribution vector by the third weight, and then adding the three together to obtain the target type fusion probability distribution vector.

[0010] Furthermore, in step S2, the preprocessing includes noise reduction, calibration, format conversion, and time synchronization.

[0011] Furthermore, in step S4, ; in, The first rainfall impact coefficient represents the effect of energy attenuation caused by rainfall on the first confidence level. The first wind speed influence coefficient represents the impact of wind-induced turbulent disturbances on the first confidence level.

[0012] Furthermore, in step S4, ; in, The illumination effect coefficient represents the degree to which the illumination deviation from the optimal state affects the second confidence level. The second rainfall influence coefficient represents the degree of influence of rainfall intensity on the second confidence level; the image signal-to-noise ratio is optimal under the best illumination level, and the reliability of feature extraction is the highest.

[0013] Furthermore, in step S4, ; in, The target temperature difference influence coefficient represents the degree of influence of the temperature difference between the environment and the target characteristics on the third confidence level. The second wind speed influence coefficient represents the degree of influence of wind speed on the third confidence level. The temperature influence coefficient represents the degree to which temperature deviation from the optimal state affects the third confidence level; thermal images at the optimal temperature level have the strongest stability and the highest clarity and detail reproduction.

[0014] Furthermore, in step S5, First weight = (first confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ); Second weight = (Second confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ); Third weight = (Third confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ).

[0015] A drone detection device based on multi-sensor and confidence assessment includes: The data acquisition module is used to collect sensor data of the target and meteorological data of the target location; The processing module is used to preprocess sensor data and meteorological data to obtain preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. The first calculation module is used to input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector. The second calculation module is used to calculate the confidence level based on the meteorological data level and the target characteristic temperature level. The first judgment module is used to determine whether the confidence level is greater than the confidence threshold. If it is, the weight is calculated based on the confidence level; otherwise, an alarm is triggered. The third calculation module multiplies the target type probability distribution vector by the weights to obtain the target type fusion probability distribution vector; The second judgment module is used to determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

[0016] An electronic device, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0017] A computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the method described above.

[0018] The beneficial effects of the present invention are as follows: Multi-sensor reliability quantitative assessment: To quantify the impact of the environment on sensors, multiple factors are integrated through physical mechanisms. Based on real-time meteorological environmental data (such as rainfall, light, temperature, and wind speed), and following the principle of prioritizing meteorological dominant factors, confidence models for three types of sensors are constructed to dynamically assess the reliability (confidence) of each sensor under current conditions, providing a scientific and accurate data basis for dynamic fusion.

[0019] Dynamic fusion strategy design: Based on a confidence-weighted multi-sensor dynamic fusion algorithm, a squared term is introduced to strengthen the weight of high-confidence sensors, realizing data fusion of multiple sensors (millimeter-wave radar, visible light camera, infrared thermal imager), adapting to both scenarios where all sensors are effective and single sensor fails, achieving complementary advantages of multi-source data, improving the ability to identify low, slow, and small targets such as drones, birds, and balloons in complex environments, significantly reducing misjudgment and missed detection problems in the detection process, and providing high-precision and robust data support for drone detection missions. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a schematic diagram of the modules of the device of the present invention; The module includes a data acquisition module 1, a processing module 2, a first calculation module 3, a second calculation module 4, a first judgment module 5, a third calculation module 6, and a second judgment module 7. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] To address the challenges of UAV detection in complex environments (such as heavy rain, high temperatures, and nighttime), a solution integrating multimodal sensor fusion (millimeter-wave radar, visible light camera, and infrared thermal imaging) with dynamic confidence assessment is proposed. This solution combines real-time environmental factors to ensure the reliability and accuracy of multi-sensor fusion.

[0023] The embodiments of the present invention will be further described below with reference to several examples.

[0024] Example 1 like Figure 1 A method for detecting unmanned aerial vehicles (UAVs) based on multi-sensor and confidence assessment includes the following steps: S1. Collect sensor data of the target and meteorological data of the target location; Collect sensor data of the target and meteorological data of the target location. The sensor data includes radar echo signals, images and thermal images. The meteorological data includes rainfall, light intensity, ambient temperature and wind speed. The targets include five types: drones, birds, balloons, kites, and others. The weather station acquires multi-dimensional meteorological data in real time via the meteorological software's API, including rainfall, light intensity, ambient temperature, and wind speed. Simultaneously, it connects to an NTP server and synchronizes to standard UTC time via a GPS timing module, ensuring consistency with the GPS clock references of millimeter-wave radar, visible light cameras, and infrared thermal imagers. The weather station collects environmental data at a dynamic minimum cycle, performing NTP dynamic calibration and timestamp binding on each set of raw data to ensure temporal alignment between environmental parameters and sensor data.

[0025] S2. Preprocess the sensor data and meteorological data to obtain the preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. The radar echo signal is preprocessed to obtain the RDM image and the micro-Doppler time-frequency image; the image is preprocessed to obtain the standardized RGB image; the thermal image is preprocessed to obtain the thermal grayscale image; the meteorological data is preprocessed to obtain the rainfall level, illumination level, temperature level, and wind speed level. The specific conversion methods are shown in Table 1; the target characteristic temperature is obtained from the thermal image, and the target characteristic temperature level is obtained from the target characteristic temperature. The conversion method for the target characteristic temperature level can also be adopted using the temperature level conversion method in Table 1. Specifically, for radar echo signals, the processing procedures in steps S1 and S2 are as follows: The radar echo signal of frequency-modulated continuous wave is received by millimeter-wave radar. Background noise is eliminated by a static clutter MTI filter. Two-dimensional data matrices of fast time (range dimension) and slow time (Doppler dimension) are extracted to preserve the time-domain signal characteristics of the target reflection.

[0026] Fast Fourier Transform (FFT) is performed on the fast time dimension to obtain the range spectrum. The target range gate is located by constant false alarm rate detection (CFAR) to filter out the range cells containing the target. FFT is performed on the slow time series within the target range gate to obtain the Doppler spectrum. The spectrum is adjusted to a 128×128 range-Doppler map (RDM) by bilinear interpolation and the power spectral density value is normalized to [0,1].

[0027] The slow time series of the target range gate is used to generate 16 frames of Doppler time-frequency map (time × frequency × energy) through continuous wavelet transform (CWT) to capture dynamic features such as periodic modulation of UAV rotor and non-periodic fluctuations of bird flapping wings. The time series length is matched with the radar pulse repetition frequency (PRF).

[0028] Specifically, for the image, the processing steps S1 and S2 are as follows: Acquire RGB three-channel color images, including the texture and shape features of the target (such as drone rotors and kite tail fins).

[0029] Adaptive histogram equalization is performed on the acquired low-light images, and gamma correction is applied to overexposed images to enhance the contrast between the target and the background. For images that have undergone illumination enhancement, bilinear interpolation is used to adjust the images to a fixed size of 640×640, and the pixel values ​​are normalized to [0,1].

[0030] A standard RGB image is a standardized RGB image with fixed dimensions and pixel normalization.

[0031] Specifically, for thermal imaging images, the processing steps S1 and S2 are as follows: A 16-bit thermal image was acquired, with pixel values ​​reflecting the infrared radiation intensity of the target. An adaptive histogram equalization and linear mapping strategy was used to convert the 16-bit thermal image into an 8-bit thermal image, balancing the global contrast. The image size was then adjusted to 320×256 using an interpolation algorithm. A Gaussian filter kernel was applied to filter the thermal image, suppressing thermal noise and highlighting key features such as the high-temperature area of ​​the UAV motor and the body temperature area of ​​birds, thus obtaining a thermal grayscale image. Dynamic thresholding was used to distinguish the target area from the background on the 16-bit thermal image, and Gaussian mixture clustering (GMM) was used to extract the temperature values ​​of the cluster centers in the target area as the target feature temperature.

[0032] Table 1. Conversion Table for Rainfall Level, Sunshine Level, Temperature Level, and Wind Speed ​​Level Environmental parameters Threshold division Scene tags Rainfall 0 mm / h → Level 0 (No rain) 0-1 mm / h → Level 1 (Light rain) 1-2.5 mm / h → Level 2 (Light rain) 2.5-10 mm / h → Level 3 (Moderate rain) 10-50 mm / h → Level 4 (Heavy rain) 50-100 mm / h → Level 5 (Heavy rain) >100 mm / h → Level 6 (Extremely heavy rain) Rainfall level (0-6) Light intensity <100 lux → Level 0 (Dim Light) 100-500 lux → Level 1 (Very Low Light) 500-1000 lux → Level 2 (Weak Light) 1000-2000 lux → Level 3 (Suitable Light) 2000-5000 lux → Level 4 (Medium Light) 5000-10000 lux → Level 5 (Relatively Strong Light) >10,000 lux → Level 6 (Strong Light) Light level (0-6) temperature <0℃→Level 0 (Extremely Cold) 0-10℃→Level 1 (Cold) 10-20℃→Level 2 (Cool) 20-28℃→Level 3 (Suitable) 28-35℃→Level 4 (Warm) 35-45℃→Level 5 (High Temperature) >45℃→Level 6 (Extreme High Temperature) Temperature rating (0-6) wind speed 0-0.2m / s → Level 0 (Calm) 0.3-1.5m / s → Level 1 (Light Wind) 1.5-3.4m / s → Level 2 (Light Breeze) 3.4-5.5m / s → Level 3 (Gentle Breeze) 5.5-10.8m / s → Level 4 (Moderate Wind) 10.8-17.2m / s → Level 5 (Strong Wind) >17.2m / s → Level 6 (Gale) Wind speed rating (0-6)

[0033] S3. Input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector; The RDM image and the micro-Doppler time-frequency image are input into the first target type probability distribution model to obtain the first target type probability distribution vector; the normalized RGB image is input into the second target type probability distribution model to obtain the second target type probability distribution vector; the thermal imaging grayscale image is input into the third target type probability distribution model to obtain the third target type probability distribution vector. A training sample library covering multiple scenarios was constructed, encompassing five categories of low-speed, small targets: drones, birds, balloons, kites, and others. Data was collected under various environments, including daytime, nighttime, sunny, foggy, and rainy conditions. Each sample set includes an RDM image, a 16-frame micro-Doppler time-frequency image sequence, and raw slow-time data. Target regions in the preprocessed RDM images were manually labeled. In addition to category labels, the labels included the coordinates of the bounding rectangle of the target region in the RDM image, the rectangle's width and height, and key time frames representing the micro-Doppler temporal features. The RDM images were randomly rotated (±15°) and Gaussian noise was added. The micro-Doppler temporal features were stretched along the time axis and flipped. The training / validation / test sets were divided in a 7:2:1 ratio, ultimately expanding the sample size to over 10,000.

[0034] A dual-branch network structure (spatial feature branch + temporal feature branch) is employed, and a CNN-Transformer model is trained based on a sample database. The spatial feature branch uses a lightweight CNN (MobileNetV3) model to train on the samples, while the temporal feature branch uses a Transformer encoder as its core to extract temporal target characteristics. During training, five categories are set: "drone," "bird," "balloon," "kite," and "other." Hyperparameters such as learning rate, batch size, and number of iterations are adjusted based on the sample database, and training stops after the model converges. Finally, a converged CNN-Transformer model, i.e., the probability distribution model of the first target type, is obtained through training.

[0035] For real-time target detection, the spatial feature branch inputs the pre-processed RDM image from the millimeter-wave radar into a trained CNN model. It extracts range-Doppler energy distribution features through four convolutional layers, outputting a spatial feature vector. The temporal feature branch inputs a 16-frame sequence of micro-Doppler time-frequency images into a Transformer encoder to capture the periodicity of the micro-Doppler data. A self-attention mechanism is used to capture the dynamic correlation between frames, outputting a temporal feature vector. Finally, the spatial and temporal feature vectors are concatenated, fused through a fully connected layer, and the raw scores for five target classes are output. These scores are then converted into probability distributions using a Softmax function.

[0036] The probability distribution vector for the first target type is P1 = [P11,P12,P13,P14,P15], where P11, P12, P13, P14, and P15 correspond to the probabilities of the appearance of drones, birds, balloons, kites, and other targets, respectively, and the sum of P11, P12, P13, P14, and P15 is 1.

[0037] A training sample library covering multiple scenarios was constructed, encompassing five categories of low-speed, small targets: drones, birds, balloons, kites, and others. Data was collected under various environments, including daytime, nighttime, sunny, foggy, and rainy conditions. Each sample set includes RGB images of the five target categories. Standardized RGB images of fixed sizes after data preprocessing were manually labeled, including the target's bounding box in the RGB image, its 3D coordinates in the local Cartesian coordinate system of the visible light camera, and its category label. The images were randomly flipped (±15°), brightness / contrast adjusted (±30%), and Gaussian noise was added. The training / validation / test sets were divided in a 7:2:1 ratio, expanding the sample size to approximately 10,000.

[0038] A deep convolutional neural network detection model based on YOLOv3 was adopted, which consists of three parts: feature extraction, multi-scale feature fusion, and a detection head. Five categories were set: "drone," "bird," "balloon," "kite," and "other." The learning rate parameter was adjusted, and the model was trained based on a sample database until it converged.

[0039] A converged YOLOv3 model, specifically the second object type probability distribution model, is trained and used for real-time detection on preprocessed, normalized RGB images. The Darknet-53 network extracts visual and texture features at different levels. Upsampling and skip connection operations are used to fuse features at different scales to adapt to object detection of varying sizes. Finally, the detection head processes the fused features and outputs the object's bounding box and class probability.

[0040] Output the bounding box of the target and the probability distribution vector of the second target type P2 = [P21,P22,P23,P24,P25], where P21, P22, P23, P24, and P25 correspond to the probabilities of the appearance of drones, birds, balloons, kites, and other targets, respectively, and the sum of P21, P22, P23, P24, and P25 is 1.

[0041] A training sample library was constructed for "drones," "birds," "balloons," "kites," and "others." Each sample group contained thermal imaging grayscale images of at least five target categories. The preprocessed, fixed-size thermal imaging grayscale images were manually annotated, including the target category, the bounding box of the high-temperature region in the thermal image, the target's 3D coordinates in the local Cartesian coordinate system of the infrared thermal imager, and the category label. Temperature range scaling and hotspot translation were performed on the thermal images, and the training / validation / test sets were divided in a 7:2:1 ratio, ultimately expanding the sample size to 10,000.

[0042] A "CNN + Temporal Convolution" network model was constructed. The CNN branch adopted a lightweight ResNet-10 architecture, containing 10 convolutional layers, and the temporal convolutional layer consisted of 3 1D convolutional temporal layers. Five categories were set: "drone," "bird," "balloon," "kite," and "other." The learning rate parameter was adjusted, and the model was trained based on a sample library. Training stopped after the model converged.

[0043] A converged "CNN + temporal convolution" network model, namely the third target type probability distribution model, is obtained through training. This model performs real-time detection on thermal imaging grayscale images. A lightweight ResNet-10 architecture is used to extract the thermal distribution features of single-frame thermal images. A sliding window approach is used to select the latest 8 frames of thermal imaging images and input them into the temporal convolutional layer to capture the motion trajectory of the hot areas and analyze the frequency of thermal radiation changes. Finally, the spatial and temporal feature vectors are concatenated, fused through a fully connected layer, and the raw scores for the five target categories are output. These scores are then converted into probability distributions using a Softmax function.

[0044] Output the probability distribution vector of the third target type P3 = [P31,P32,P13,P34,P35], where P31, P32, P33, P34, and P35 correspond to the probabilities of the appearance of drones, birds, balloons, kites, and other targets, respectively, and the sum of P31, P32, P33, P34, and P35 is 1. S4. Calculate the confidence level based on the meteorological data level and the target characteristic temperature level; The first confidence level is calculated based on rainfall level and wind speed level; the second confidence level is calculated based on rainfall level and light intensity level; and the third confidence level is calculated based on target characteristic temperature level, temperature level, and wind speed level. ; in, The first rainfall impact coefficient represents the impact of energy attenuation caused by rainfall on the first confidence level. In this embodiment, it is set to 0.1. The first wind speed influence coefficient represents the impact of wind-induced turbulence disturbance on the first confidence level. In this embodiment, it is set to 0.075.

[0045] ; in, The illumination influence coefficient represents the degree to which the illumination deviation from the optimal state affects the second confidence level; in this embodiment, it is set to 0.4. The second rainfall influence coefficient represents the degree of influence of rainfall intensity on the second confidence level, and is set to 0.2 in this embodiment; the image signal-to-noise ratio is optimal and the feature extraction reliability is highest under the best illumination level, and is set to 3 in this embodiment.

[0046] ; in, The target temperature difference influence coefficient represents the degree of influence of the temperature difference between the environment and the target feature on the third confidence level. In this embodiment, it is set to 0.1. The second wind speed influence coefficient represents the degree of influence of wind speed on the third confidence level; in this embodiment, it is set to 0.23. The temperature influence coefficient represents the degree of influence of temperature deviation from the optimal state on the third confidence level. In this embodiment, it is set to 0.54. At the optimal temperature level, the thermal image has the strongest stability and the highest clarity and detail reproduction.

[0047] S5. Determine if the confidence level is greater than the confidence level threshold. If so, calculate the weight based on the confidence level; otherwise, issue an alarm. Determine whether the first confidence level C1, the second confidence level C2, and the third confidence level C3 are all greater than the confidence threshold. If so, calculate the first weight w1, the second weight w2, and the third weight w3 based on the first confidence level C1, the second confidence level C2, and the third confidence level C3, respectively. w1=(C1 2 ) / (C1 2 +C2 2 +C3 2 ); w2=(C2 2 ) / (C1 2 +C2 2 +C3 2 ); w3=(C3 2 ) / (C1 2 +C2 2 +C3 2 ); Otherwise, stop running and issue an alarm; S6. Multiply the target type probability distribution vector by the weight to obtain the target type fusion probability distribution vector; Multiply the probability distribution vector of the first target type by the first weight, the probability distribution vector of the second target type by the second weight, and the probability distribution vector of the third target type by the third weight, and add the three together to obtain the target type fusion probability distribution vector. Pf = [w1×P1 + w2×P2 + w3×P3] =[(w1×P11+w2×P21+w3×P31), (w1×P12+w2×P22+w3×P32), (w1×P13+w2×P23+w3×P33),(w1×P14+w2×P24+w3×P34), (w1×P15+w2×P25+w3×P35)]; The five probabilities in Pf correspond to the probabilities of drones, birds, balloons, kites, and other targets appearing, respectively. S7. Determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

[0048] The target type is determined by the target type corresponding to the largest probability value in the target type fusion probability distribution vector, that is, the target type corresponding to the element with the largest value in the target type fusion probability distribution vector Pf.

[0049] Example 2 like Figure 2 A drone detection device based on multi-sensor and confidence assessment, comprising: The data acquisition module is used to collect sensor data of the target and meteorological data of the target location; The processing module is used to preprocess sensor data and meteorological data to obtain preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. The first calculation module is used to input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector. The second calculation module is used to calculate the confidence level based on the meteorological data level and the target characteristic temperature level. The first judgment module is used to determine whether the confidence level is greater than the confidence threshold. If it is, the weight is calculated based on the confidence level; otherwise, an alarm is triggered. The third calculation module multiplies the target type probability distribution vector by the weights to obtain the target type fusion probability distribution vector; The second judgment module is used to determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

[0050] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0051] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0052] Accordingly, this application also provides an electronic device, including: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors perform the methods described above.

[0053] Accordingly, this application also provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of any of the above methods.

[0054] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0055] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0056] On the other hand, a computer-readable storage medium stores computer instructions thereon, which, when executed by a processor, implement the steps of the above-described method. When the computer program is executed by the processor, it implements the method as described in any of the first aspects above. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for detecting unmanned aerial vehicles (UAVs) based on multi-sensor and confidence assessment, characterized in that, Includes the following steps: S1. Collect sensor data of the target and meteorological data of the target location; S2. Preprocess the sensor data and meteorological data to obtain the preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. S3. Input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector; S4. Calculate the confidence level based on the meteorological data level and the target characteristic temperature level; S5. Determine if the confidence level is greater than the confidence level threshold. If so, calculate the weight based on the confidence level; otherwise, issue an alarm. S6. Multiply the target type probability distribution vector by the weight to obtain the target type fusion probability distribution vector; S7. Determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

2. The method according to claim 1, characterized in that, Step S1 specifically involves: collecting sensor data of the target and meteorological data of the target location. The sensor data includes radar echo signals, images, and thermal images, and the meteorological data includes rainfall, light intensity, ambient temperature, and wind speed.

3. The method according to claim 2, characterized in that, Step S2 specifically involves: preprocessing the radar echo signal to obtain the RDM image and the micro-Doppler time-frequency image; preprocessing the image to obtain a standardized RGB image; preprocessing the thermal image to obtain a thermal grayscale image; preprocessing the meteorological data to obtain the rainfall level, illumination level, temperature level, and wind speed level; obtaining the target characteristic temperature based on the thermal image; and obtaining the target characteristic temperature level based on the target characteristic temperature.

4. The method according to claim 3, characterized in that, Step S3 specifically involves: inputting the RDM image and the micro-Doppler time-frequency image into the first target type probability distribution model to obtain the first target type probability distribution vector; inputting the standardized RGB image into the second target type probability distribution model to obtain the second target type probability distribution vector; and inputting the thermal imaging grayscale image into the third target type probability distribution model to obtain the third target type probability distribution vector.

5. The method according to claim 4, characterized in that, Step S4 specifically involves: calculating the first confidence level based on the rainfall level and wind speed level; calculating the second confidence level based on the rainfall level and light intensity level; and calculating the third confidence level based on the target temperature difference level, temperature level, and wind speed level.

6. The method according to claim 5, characterized in that, Step S5 specifically involves determining whether the first confidence level, the second confidence level, and the third confidence level are all greater than the confidence threshold. If so, the first weight, the second weight, and the third weight are calculated based on the first confidence level, the second confidence level, and the third confidence level, respectively. Otherwise, the process is stopped and an alarm is triggered.

7. The method according to claim 6, characterized in that, Step S6 specifically involves multiplying the first target type probability distribution vector by the first weight, the second target type probability distribution vector by the second weight, and the third target type probability distribution vector by the third weight, and then adding the three together to obtain the target type fusion probability distribution vector.

8. The method according to any one of claims 1-7, characterized in that, In step S2, the preprocessing includes noise reduction, calibration, format conversion, and time synchronization.

9. The method according to any one of claims 5-7, characterized in that, In step S4, ; in, The first rainfall impact coefficient represents the effect of energy attenuation caused by rainfall on the first confidence level. The first wind speed influence coefficient represents the impact of wind-induced turbulent disturbances on the first confidence level.

10. The method according to any one of claims 5-7, characterized in that, In step S4, ； in, The illumination effect coefficient represents the degree to which the illumination deviation from the optimal state affects the second confidence level. The second rainfall influence coefficient represents the degree of influence of rainfall intensity on the second confidence level; the image signal-to-noise ratio is optimal under the best illumination level, and the reliability of feature extraction is the highest.

11. The method according to any one of claims 5-7, characterized in that, In step S4, ; in, The target temperature difference influence coefficient represents the degree of influence of the temperature difference between the environment and the target characteristic on the third confidence level. The second wind speed influence coefficient represents the degree of influence of wind speed on the third confidence level. The temperature influence coefficient represents the degree to which temperature deviation from the optimal state affects the third confidence level; thermal images at the optimal temperature level have the strongest stability and the highest clarity and detail reproduction.

12. The method according to claim 6, characterized in that, In step S5, First weight = (first confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ); Second weight = (Second confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ); Third weight = (Third confidence level) 2 ) / (First confidence level) 2 +Second confidence level 2 +Third confidence level 2 ).

13. A drone detection device based on multi-sensor and confidence assessment, characterized in that, include: The data acquisition module is used to collect sensor data of the target and meteorological data of the target location; The processing module is used to preprocess sensor data and meteorological data to obtain preprocessed sensor data, meteorological data level and target characteristic temperature level, respectively. The first calculation module is used to input the preprocessed sensor data into the target type probability distribution model to obtain the target type probability distribution vector. The second calculation module is used to calculate the confidence level based on the meteorological data level and the target characteristic temperature level. The first judgment module is used to determine whether the confidence level is greater than the confidence threshold. If it is, the weight is calculated based on the confidence level; otherwise, an alarm is triggered. The third calculation module multiplies the target type probability distribution vector by the weights to obtain the target type fusion probability distribution vector; The second judgment module is used to determine the target type based on the target type corresponding to the largest probability value in the target type fusion probability distribution vector.

14. An electronic device, characterized in that, include: one or more processors; a memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

15. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

Images