A bird-detecting radar target recognition system and method based on multi-sensor data fusion

Abstract

The present invention relates to the technical field of bird-detecting radar target recognition, and specifically to a bird-detecting radar target recognition system and method based on multi-sensor data fusion. In the present invention, a data collection module collects radar sensor data, temperature sensor data and image sensor data, and performs adaptive filtering and graph neural network processing to obtain corresponding feature vectors. A high-level target evaluation module then uses these feature vectors as high-level data and original data as low-level data. The posterior probability of the target bird species is obtained by calculating the high-level data to determine the high-level target species. A fusion processing module uses a Bayesian formula to calculate the probability of the low-level data under the high-level target species to obtain a fused probability distribution, and then calculates and compares the low-level target species. If they are consistent, they are used as the final target species. If they are inconsistent, the final target species is selected according to the peak value of the probability distribution. The scheme solves the problem of insufficient multi-sensor data fusion and improves the accuracy and robustness of target recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of bird-detecting radar target recognition, and in particular to a bird-detecting radar target recognition system and method based on multi-sensor data fusion. Background Art

[0002] Traditional bird-detecting radar target recognition systems face many technical challenges and limitations.

[0003] First, such systems usually rely on a single sensor or a limited number of data sources, making it difficult to fully acquire and integrate diverse environmental information, resulting in low accuracy and robustness in bird target recognition. Moreover, when faced with complex and changing natural environments and different types of birds, data from a single sensor is often insufficient to provide a comprehensive and reliable feature description. Therefore, such systems are susceptible to noise and external interference when identifying target species, and have a high probability of misjudgment.

[0004] In addition, traditional systems usually lack in-depth analysis and utilization of probability distribution, and fail to fully tap the potential features in sensor data. Traditional methods rely more on fixed algorithms and pattern matching, lacking adaptability and dynamic adjustment capabilities, which significantly reduces the recognition efficiency and accuracy of the system in practical applications, especially when facing multiple types of birds and complex environments. At the same time, traditional systems are relatively rigid in processing and fusing multi-level data, and usually cannot play the complementary role of data at each level, thus limiting the effect of overall data fusion.

[0005] In summary, the traditional bird-detecting radar target recognition system has significant limitations in terms of the diversity of data sources, the depth and accuracy of data fusion, and dynamic adaptability, and cannot meet the actual needs of high precision and high robustness. Summary of the invention

[0006] The purpose of the present invention is to provide a bird detection radar target recognition system based on multi-sensor data fusion and a method thereof to solve the problems raised in the above background technology.

[0007] To achieve the above object, one of the objects of the present invention is to provide a bird detection radar target recognition system based on multi-sensor data fusion, comprising a data collection module, a high-level target evaluation module and a fusion processing module, wherein:

[0008] The data collection module collects radar sensor data acquired by the radar sensor, temperature sensor data acquired by the infrared sensor, and image sensor data acquired by the optical sensor; performs adaptive filtering and graph neural network processing on the radar sensor data, temperature sensor data, and image sensor data to obtain radar feature vectors, infrared temperature feature vectors, and optical image feature vectors; and obtains target features from the feature database, where the target features include multiple bird species;

[0009] The high-level target assessment module uses the radar feature vector, the infrared temperature feature vector and the optical image feature vector as high-level data, and uses the radar data, the temperature data and the image data as low-level data; based on the high-level data, calculates the posterior probabilities of multiple bird species in the target feature, and selects the bird species in the target feature with the largest posterior probability as the high-level target species;

[0010] The fusion processing module uses the Bayesian formula to calculate the probability of each sensor data in the low-level data under the high-level target type, and obtains the fusion probability distribution of the low-level data under the high-level target type; according to the fusion probability distribution, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the low-level target type; the high-level target type is compared with the low-level target type, wherein:

[0011] If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category;

[0012] If the comparison is inconsistent, the larger one is selected as the final target category based on the peak value of the fusion probability distribution of the low-level data under the high-level target category and the peak value of the fusion probability distribution of the high-level data under the high-level target category.

[0013] As a further improvement of the technical solution, the data collection module uses a radar sensor to detect the distance and speed of the target by sending and receiving electromagnetic wave pulses, as radar sensor data ; The infrared sensor detects the infrared radiation of the target object and obtains the temperature information of the target as temperature sensing data ; Optical sensors capture the image information of the target through cameras or other optical devices as image sensing data ; Perform adaptive filtering and graph neural network processing on radar sensor data, temperature sensor data and image sensor data to obtain radar feature vectors , infrared temperature feature vector and the optical image feature vector ; Get target features from the feature database , target features Including multiple bird species, indicating the Bird species target.

[0014] As a further improvement of the technical solution, the process of generating high-level target types by the high-level target evaluation module is as follows:

[0015] The radar feature vector , infrared temperature feature vector and the optical image feature vector As high-level data ,in ;

[0016] For the target feature For each bird species in The conditional probability , calculated according to Gaussian distribution, the calculation formula is as follows:

[0017] ,in and Respectively represent Mean and covariance matrix of bird species characteristics;

[0018] Assume that all types of birds have the same probability of appearing, that is, ,in is the total number of bird species;

[0019] Use the Bayesian formula to calculate the posterior probability, ,in An index variable used to iterate over all bird species;

[0020] Find the bird species with the highest posterior probability, the formula is as follows:

[0021] ,in It means finding the variable that makes a function reach its maximum value;

[0022] The bird species corresponding to the calculated maximum posterior probability As a high-level target category.

[0023] As a further improvement of the technical solution, the fusion processing module includes a low-level target evaluation unit, and the process of generating the low-level target type by the low-level target evaluation unit is as follows:

[0024] For the identified high-level target categories , calculate the high-level target category of each sensor data in the low-level data The probability of , where:

[0025] For radar sensor data , calculate its high-level target type The conditional probability under ;

[0026] For temperature sensing data , calculate its high-level target type The conditional probability under ;

[0027] For image sensing data , calculate its high-level target type The conditional probability under ;

[0028] Using the above conditional probability, we can get the target type at the high level. The fusion probability distribution of the low-level data under is as follows:

[0029] ,in , represents the set of all low-level data;

[0030] Using the above fused probability distribution, calculate the posterior probability of each bird species. For each bird species in , calculate the conditional probability of the lower-level data under it, the formula is as follows:

[0031] ;

[0032] For all birds Compute the sum of the denominators: ;

[0033] The posterior probability of each bird is calculated using the Bayesian formula:

[0034] ;

[0035] After calculating the posterior probabilities of multiple bird species, the bird species with the largest posterior probability is selected as the low-level target species: .

[0036] As a further improvement of the technical solution, the fusion processing module includes a comparison processing unit, and the process of generating the final target type by the comparison processing unit is as follows:

[0037] The high-level target category and low-level target types For comparison, where:

[0038] If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category. ;

[0039] If the comparison is inconsistent, the larger one is selected as the final target type according to the peak value of the fusion probability distribution of the low-level data under the high-level target type and the peak value of the fusion probability distribution of the high-level data under the high-level target type, where the peak value of the fusion probability distribution of the low-level data under the high-level target type is , the peak value of the fusion probability distribution of high-level data under high-level target categories is , then the final target type The calculation formula is as follows:

[0040] .

[0041] The second object of the present invention is to provide a method for using a bird detection radar target recognition system based on multi-sensor data fusion, comprising the following method steps:

[0042] S1. Collect radar sensor data acquired by radar sensors, temperature sensor data acquired by infrared sensors, and image sensor data acquired by optical sensors; perform adaptive filtering and graph neural network processing on the radar sensor data, temperature sensor data, and image sensor data to obtain radar feature vectors, infrared temperature feature vectors, and optical image feature vectors; obtain target features from a feature database, where the target features include multiple bird species;

[0043] S2, taking the radar feature vector, infrared temperature feature vector and optical image feature vector as high-level data, and taking the radar data, temperature data and image data as low-level data; calculating the posterior probabilities of multiple bird species in the target feature according to the high-level data, and selecting the bird species in the target feature with the largest posterior probability as the high-level target species;

[0044] S3. Using the Bayesian formula, the probability of each sensor data in the low-level data under the high-level target category is calculated respectively, and the fusion probability distribution of the low-level data under the high-level target category is obtained; according to the fusion probability distribution, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the low-level target category; the high-level target category and the low-level target category are compared, wherein:

[0045] If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category;

[0046] If the comparison is inconsistent, the larger one is selected as the final target category based on the peak value of the fusion probability distribution of the low-level data under the high-level target category and the peak value of the fusion probability distribution of the high-level data under the high-level target category.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] In the bird-detecting radar target recognition system and method based on multi-sensor data fusion, the Bayesian formula is used to calculate the probability of each sensor data in the low-level data under the high-level target type, which can effectively construct the fusion probability distribution of the low-level data under the high-level target type. This first ensures that the potential characteristics of the sensor data can be fully mined and utilized under the current high-level target type assumption through the comprehensive information of the multi-sensor data; then, according to the fusion probability distribution, Bayesian reasoning is used to further calculate the posterior probabilities of multiple bird types in the target features, and the bird type in the target feature with the largest posterior probability is selected as the low-level target type; this step extracts the most likely target type from the sensor data through the probability theory method, thereby improving the credibility and accuracy of target recognition;

[0049] In addition, the unique role of high-level target categories in this process is that they not only serve as initial assumptions to guide the probability distribution calculation of low-level data, but also play a key role in the overall target recognition decision. By comparing high-level target categories with low-level target categories, errors and deviations that may be caused by single-level data can be identified and corrected. When the high-level and low-level prediction results are consistent, the target recognition results are more stable and reliable; conversely, by comprehensively comparing and weighing the information from different data sources, the final target category selection can be dynamically adjusted and optimized, thereby enhancing the robustness and adaptability of the system;

[0050] This two-level comparison and fusion mechanism makes the high-level target type act as a constraint and verification in the entire target recognition process, improving the overall performance of the system; ultimately, through the comparison and comprehensive determination of high-level and low-level target types, the system can more accurately identify and classify bird targets, significantly improving the effect and efficiency of multi-sensor data fusion in target recognition, thereby achieving higher classification accuracy and reliability in complex environments and diverse bird target scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0052] Figure 2 It is a schematic diagram of the overall module unit of the present invention.

[0053] In the figure: 100, data collection module; 200, high-level target evaluation module; 300, fusion processing module; 301, low-level target evaluation unit; 302, comparison processing unit. DETAILED DESCRIPTION

[0054] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] Unless otherwise specified, the terms (including technical terms) used herein have the commonly understood meanings to those skilled in the art. In addition, it is understood that the terms defined in commonly used dictionaries should be understood to have the same meanings as those in the context of the relevant fields, and should not be understood as idealized or overly formal meanings.

[0056] According to the document, the following are some definitions of terms:

[0057] High-level data: refers to radar feature vectors, infrared temperature feature vectors, and optical image feature vectors obtained through adaptive filtering and graph neural network processing. This feature vector is based on high-level sensor data and is used to calculate the posterior probability of the target type, thereby determining the high-level target type;

[0058] Low-level data: refers to unprocessed radar sensor data, temperature sensor data, and image sensor data, which are used to calculate the fusion probability distribution of low-level data after determining the type of high-level target, so as to evaluate the type of low-level target;

[0059] High-level target type: The initial target type is calculated based on the radar feature vector, infrared temperature feature vector and optical image feature vector. The high-level target type is determined by calculating the posterior probability of each bird species and selecting the bird species with the largest posterior probability.

[0060] Low-level target types: Based on the determined high-level target types, the target types are obtained by calculating the conditional probability of each sensor data in the low-level data and fusing them. The low-level target types are determined by calculating the posterior probabilities of multiple bird species in the low-level data and selecting the bird species with the largest posterior probability.

[0061] Fusion probability distribution: Under the assumption of high-level target type, the fusion probability distribution calculated with low-level data, and the fusion probability distribution calculated with high-level data; when the high-level target type is inconsistent with the low-level target type, the final target type is determined by comparing the peak values ​​of the two fusion probability distributions and selecting the larger one.

[0062] Next, see Figure 1-Figure 2One of the purposes of this embodiment is to provide a bird-detecting radar target recognition system based on multi-sensor data fusion, including a data collection module 100, a high-level target evaluation module 200 and a fusion processing module 300.

[0063] The data collection module 100 collects radar sensor data acquired by the radar sensor, temperature sensor data acquired by the infrared sensor, and image sensor data acquired by the optical sensor; wherein the radar sensor detects the distance and speed of the target by sending and receiving electromagnetic wave pulses, and the radar sensor data is used as , including distance and speed; infrared sensors detect the infrared radiation of target objects and obtain the temperature information of the target as temperature sensing data , including temperature values; optical sensors capture image information of the target through cameras or other optical devices as image sensing data , including image frames.

[0064] The data collection module 100 performs adaptive filtering and graph neural network processing on the radar sensor data, temperature sensor data and image sensor data to obtain the radar feature vector , infrared temperature feature vector and the optical image feature vector ; Get target features from the feature database , target features Including multiple bird species, indicating the Bird species targets, such as represents sparrows, where each bird species target includes corresponding bird features.

[0065] The high-level target assessment module 200 uses the radar feature vector, the infrared temperature feature vector and the optical image feature vector as high-level data, and uses the radar data, the temperature data and the image data as low-level data; based on the high-level data, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the high-level target species, specifically including:

[0066] The radar feature vector , infrared temperature feature vector and the optical image feature vector As high-level data ,in ;

[0067] For the target feature For each bird species in The conditional probability , calculated according to Gaussian distribution, the calculation formula is as follows:

[0068] ,in and Respectively represent Mean and covariance matrix of bird species characteristics;

[0069] Assume that all types of birds have the same probability of appearing, that is, ,in is the total number of bird species;

[0070] The posterior probability is calculated using the Bayesian formula, ,in An index variable used to iterate over all bird species;

[0071] Find the bird species with the highest posterior probability, the formula is as follows:

[0072] ,in It means finding the variable that makes a function reach its maximum value;

[0073] The bird species corresponding to the calculated maximum posterior probability As a high-level target category.

[0074] The low-level target evaluation unit 301 in the fusion processing module 300 uses the Bayesian formula to calculate the probability of each sensor data in the low-level data under the high-level target category, and obtains the fusion probability distribution of the low-level data under the high-level target category; according to the fusion probability distribution, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the low-level target category, which specifically includes:

[0075] For the identified high-level target categories , calculate the high-level target category of each sensor data in the low-level data The probability of , where:

[0076] For radar sensor data , calculate its high-level target type The conditional probability under ;

[0077] For temperature sensing data , calculate its high-level target type The conditional probability under ;

[0078] For image sensing data , calculate its high-level target type The conditional probability under ;

[0079] Using the above conditional probability, we can get the target type at the high level. The fusion probability distribution of the low-level data under (assuming that each low-level data is conditionally independent) is as follows:

[0080] ,in , represents the set of all low-level data;

[0081] Using the above fused probability distribution, calculate the posterior probability of each bird species (i.e., multiple target features). For each bird species in , calculate the conditional probability of the lower-level data under it, the formula is as follows:

[0082] ;

[0083] For all birds Compute the sum of the denominators: ;

[0084] The posterior probability of each bird is calculated using the Bayesian formula:

[0085] ;

[0086] After calculating the posterior probabilities of multiple bird species, the bird species with the largest posterior probability is selected as the low-level target species: .

[0087] The comparison processing unit 302 in the fusion processing module 300 compares the high-level object categories and low-level target types For comparison, where:

[0088] If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category. ; At this time, the consistent result is directly used as the final target category, which simplifies the decision-making process. This method ensures the direct application of consistent prediction results, reduces processing time, and improves the efficiency and accuracy of decision-making;

[0089] If the comparison is inconsistent, the larger one is selected as the final target type according to the peak value of the fusion probability distribution of the low-level data under the high-level target type and the peak value of the fusion probability distribution of the high-level data under the high-level target type, where the peak value of the fusion probability distribution of the low-level data under the high-level target type is , the peak value of the fusion probability distribution of high-level data under high-level target categories is , then the final target type The calculation formula is as follows:

[0090] ;

[0091] Through this trade-off and comparison, we ensure that the most representative and credible target types are selected from the different information provided by multi-source data. This method improves the robustness and flexibility of the system when facing data inconsistencies, ensures that the final decision fully considers the comprehensive information of multi-level data, and is conducive to accurate target identification under complex conditions.

[0092] The second purpose of this embodiment is to provide a method for using a bird detection radar target recognition system based on multi-sensor data fusion, including the following method steps:

[0093] S1. Collect radar sensor data acquired by radar sensors, temperature sensor data acquired by infrared sensors, and image sensor data acquired by optical sensors; perform adaptive filtering and graph neural network processing on the radar sensor data, temperature sensor data, and image sensor data to obtain radar feature vectors, infrared temperature feature vectors, and optical image feature vectors; obtain target features from a feature database, where the target features include multiple bird species;

[0094] S2, taking the radar feature vector, infrared temperature feature vector and optical image feature vector as high-level data, and taking the radar data, temperature data and image data as low-level data; calculating the posterior probabilities of multiple bird species in the target feature according to the high-level data, and selecting the bird species in the target feature with the largest posterior probability as the high-level target species;

[0095] S3. Using the Bayesian formula, the probability of each sensor data in the low-level data under the high-level target category is calculated respectively, and the fusion probability distribution of the low-level data under the high-level target category is obtained; according to the fusion probability distribution, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the low-level target category; the high-level target category and the low-level target category are compared, wherein:

[0096] If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category;

[0097] If the comparison is inconsistent, the larger one is selected as the final target category based on the peak value of the fusion probability distribution of the low-level data under the high-level target category and the peak value of the fusion probability distribution of the high-level data under the high-level target category.

[0098] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A bird detection radar target recognition system based on multi-sensor data fusion, characterized in that: The system comprises a data collection module (100), a high-level target evaluation module (200) and a fusion processing module (300), wherein: The data collection module (100) collects radar sensing data acquired by a radar sensor, temperature sensing data acquired by an infrared sensor, and image sensing data acquired by an optical sensor; performs adaptive filtering and graph neural network processing on the radar sensing data, temperature sensing data, and image sensing data to obtain a radar feature vector, an infrared temperature feature vector, and an optical image feature vector; and obtains target features from a feature database, wherein the target features include multiple bird species; The high-level target assessment module (200) uses the radar feature vector, the infrared temperature feature vector and the optical image feature vector as high-level data, and uses the radar data, the temperature data and the image data as low-level data; calculates the posterior probabilities of multiple bird species in the target feature based on the high-level data, and selects the bird species in the target feature with the largest posterior probability as the high-level target species; The fusion processing module (300) uses the Bayesian formula to calculate the probability of each sensor data in the low-level data under the high-level target category, and obtains the fusion probability distribution of the low-level data under the high-level target category; according to the fusion probability distribution, calculates the posterior probabilities of multiple bird species in the target feature, and selects the bird species in the target feature with the largest posterior probability as the low-level target category; compares the high-level target category with the low-level target category, wherein: If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category; If the comparison is inconsistent, the larger one is selected as the final target category based on the peak value of the fusion probability distribution of the low-level data under the high-level target category and the peak value of the fusion probability distribution of the high-level data under the high-level target category.

2. The bird detection radar target recognition system based on multi-sensor data fusion according to claim 1 is characterized in that: The data collection module (100) uses a radar sensor to detect the distance and speed of a target by sending and receiving electromagnetic wave pulses as radar sensing data. ; The infrared sensor detects the infrared radiation of the target object and obtains the temperature information of the target as temperature sensing data ; Optical sensors capture the image information of the target through cameras or other optical devices as image sensing data ; The radar sensor data, temperature sensor data and image sensor data are adaptively filtered and processed by graph neural network to obtain the radar feature vector , infrared temperature feature vector and the optical image feature vector ; Get target features from the feature database , target features Including multiple bird species, indicating the Bird species target.

3. The bird detection radar target recognition system based on multi-sensor data fusion according to claim 2 is characterized in that: The process of generating high-level target categories by the high-level target assessment module (200) is as follows: The radar feature vector , infrared temperature feature vector and the optical image feature vector As high-level data ,in ; For the target feature For each bird species in The conditional probability , calculated according to Gaussian distribution, the calculation formula is as follows: ,in and Respectively represent Mean and covariance matrix of bird species characteristics; Assume that all types of birds have the same probability of appearing, that is, ,in is the total number of bird species; Use the Bayesian formula to calculate the posterior probability, ,in An index variable used to iterate over all bird species; Find the bird species with the highest posterior probability, the formula is as follows: ,in It means finding the variable that makes a function reach its maximum value; The bird species corresponding to the calculated maximum posterior probability As a high-level target category.

4. The bird detection radar target recognition system based on multi-sensor data fusion according to claim 3 is characterized in that: The fusion processing module (300) comprises a low-level target evaluation unit (301), and the process of generating a low-level target type by the low-level target evaluation unit (301) is as follows: For the identified high-level target categories , calculate the high-level target category of each sensor data in the low-level data The probability of , where: For radar sensor data , calculate its high-level target type The conditional probability under ; For temperature sensing data , calculate its high-level target type The conditional probability under ; For image sensing data , calculate its high-level target type The conditional probability under ; Using the above conditional probability, we can get the target type at the high level. The fusion probability distribution of the low-level data under is as follows: ,in , represents the set of all low-level data; Using the above fused probability distribution, calculate the posterior probability of each bird species. For each bird species in , calculate the conditional probability of the lower-level data under it, the formula is as follows: ； For all birds Compute the sum of the denominators: ; The posterior probability of each bird is calculated using the Bayesian formula: ； After calculating the posterior probabilities of multiple bird species, the bird species with the largest posterior probability is selected as the low-level target species: .

5. The bird detection radar target recognition system based on multi-sensor data fusion according to claim 4 is characterized in that: The fusion processing module (300) comprises a comparison processing unit (302), and the process of generating the final target category by the comparison processing unit (302) is as follows: The high-level target category and low-level target types For comparison, where: If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category. ; If the comparison is inconsistent, the larger one is selected as the final target type according to the peak value of the fusion probability distribution of the low-level data under the high-level target type and the peak value of the fusion probability distribution of the high-level data under the high-level target type, where the peak value of the fusion probability distribution of the low-level data under the high-level target type is , the peak value of the fusion probability distribution of high-level data under high-level target categories is , then the final target type The calculation formula is as follows: 。 6. A method for using the bird detection radar target recognition system based on multi-sensor data fusion according to any one of claims 1 to 5, characterized in that: The method comprises the following steps: S1. Collect radar sensor data acquired by the radar sensor, temperature sensor data acquired by the infrared sensor, and image sensor data acquired by the optical sensor; perform adaptive filtering and graph neural network processing on the radar sensor data, temperature sensor data, and image sensor data to obtain radar feature vectors, infrared temperature feature vectors, and optical image feature vectors; Acquire target features from a feature database, where the target features include multiple bird species; S2, taking the radar feature vector, infrared temperature feature vector and optical image feature vector as high-level data, and taking the radar data, temperature data and image data as low-level data; calculating the posterior probabilities of multiple bird species in the target feature according to the high-level data, and selecting the bird species in the target feature with the largest posterior probability as the high-level target species; S3. Using the Bayesian formula, the probability of each sensor data in the low-level data under the high-level target category is calculated respectively, and the fusion probability distribution of the low-level data under the high-level target category is obtained; according to the fusion probability distribution, the posterior probabilities of multiple bird species in the target feature are calculated, and the bird species in the target feature with the largest posterior probability is selected as the low-level target category; Compare high-level object categories to low-level object categories, where: If they are consistent, it means that the target category prediction of the high-level data is the same as the target category prediction of the low-level data, and it is taken as the final target category; If the comparison is inconsistent, the larger one is selected as the final target category based on the peak value of the fusion probability distribution of the low-level data under the high-level target category and the peak value of the fusion probability distribution of the high-level data under the high-level target category.

Images