Multi-modal bird identification method and system

Abstract

The invention relates to the technical field of bird recognition, and particularly discloses a multi-modal bird recognition method and system.The multi-modal bird recognition method comprises the steps that a multi-modal recognition system is constructed by fusing three types of data of biological traces, environmental disturbance and a biological magnetic field, firstly, regional positions and bird biological carrier data are extracted, initial bird characteristics are generated through screening in combination with a migration path library, and then the bird biological carrier data are extracted; secondly, analyzing atmospheric ion distribution and a landing shock wave spectrum, indirectly calculating body weight and landing speed through a shock wave instantaneous peak value and ion parameters, deriving bird tendon compressive strength as a mechanical characteristic, then converting a wing moving magnetic field signal into a frequency domain eddy current spectrogram, extracting a main frequency band energy proportion coefficient to form a biological magnetic fingerprint characteristic, and calculating the bird tendon compressive strength according to the biological magnetic fingerprint characteristic; the three types of features are standardized and coded into a bird incidence matrix, species identification is carried out on the bird incidence matrix based on a distributed bird feature library, the method breaks through traditional single-mode limitation, high robustness is still kept in complex scenes such as image shielding and audio noise, and birds are identified.

Description

Technical Field

[0001] This invention relates to the field of bird identification technology, and in particular to a multimodal bird identification method and system. Background Technology

[0002] Bird species identification refers to the process of classifying a target bird species into a specific species by observing and analyzing its morphological characteristics, behavioral habits, ecological environment, and other information. This process involves knowledge from multiple disciplines such as biology, ecology, and ornithology, and is the foundation of bird research, conservation, and birdwatching activities. Existing bird identification methods typically rely on extracting bird audio features from Mel spectrograms and acquiring images for identification. However, environmental noise interference causes the recognition rate to plummet when the signal-to-noise ratio drops below 10dB, and it is unable to distinguish between songbirds of the same frequency distributed in the same region (such as great tits and marsh tits). Furthermore, the acquired images are easily obscured, making them impossible to acquire. Consequently, the combination of audio and images cannot adequately address the problems in bird identification. Therefore, a multimodal bird identification method and system are needed to solve these problems. Summary of the Invention

[0003] The purpose of this invention is to provide a multimodal bird identification method and system to solve the technical problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A multimodal bird identification method includes: Acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data; Based on the biological trace data, regional location data and bird biological carrier data are obtained, and bird generalization data is obtained based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird characteristic data. Atmospheric ion distribution information and bird landing shock wave spectrum are obtained based on the environmental disturbance data, and bird tendon compressive strength is obtained based on the atmospheric ion distribution information and bird landing shock wave spectrum. The bird wing kinetic magnetic field signal is obtained based on the biomagnetic field data, and the bird wing kinetic magnetic field signal is converted into a frequency domain eddy current spectrum. The main frequency band energy proportion coefficient is obtained based on the frequency domain eddy current spectrum. A bird correlation matrix is ​​constructed based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy proportion coefficient. The bird association matrix is ​​identified based on a distributed bird feature database to obtain bird credentials.

[0005] Preferably, the step of obtaining bird generalization data based on the bird biological carrier data, and filtering the bird generalization data based on regional location data to obtain initial bird characteristic data includes: Based on the bird biological carrier data, information on bird fecal contaminants and the appearance of scattered feathers is obtained; based on the bird fecal contaminant information, information on hormone levels and digestive metabolism is obtained; and based on the hormone levels and digestive metabolism information, information on physiological characteristics is obtained. Based on the scattered feather appearance information, feather color information and feather shape information are obtained, and based on the feather color information and feather shape information, appearance feature information is obtained, and based on the physiological feature information and appearance feature information, bird generalization data is obtained; Information on migratory birds is obtained by comparing a pre-set database of bird migration routes with regional location data. Based on the migratory bird information, multiple bird population information is obtained, and based on the multiple bird population information, bird generalization data is filtered to obtain initial bird characteristic data.

[0006] Preferably, the step of obtaining the compressive strength of bird tendons based on the atmospheric ion distribution information and the bird landing shock wave spectrum includes: The atmospheric ion distribution information is used to obtain the ion charge gradient and ion mobility, and the atmospheric ion distribution coefficient is obtained based on the ion charge gradient and the ion mobility. The instantaneous peak value and attenuation coefficient of the shock wave are obtained from the spectrum of the bird landing shock wave, and the bird's weight is obtained from the instantaneous peak value and attenuation coefficient of the shock wave. The bird's landing speed is obtained based on the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, and the bird's tendon compressive strength is obtained based on the bird's landing speed and the bird's weight.

[0007] Preferably, the step of converting the bird wing-kinematic magnetic field signal into a frequency domain eddy current spectrum and obtaining the main frequency band energy proportion coefficient based on the frequency domain eddy current spectrum includes: The wing-induced geomagnetic field and muscle biomagnetic field were obtained based on the bird wing-induced magnetic field signal. Based on time-frequency transformation technology, an energy distribution characterization map is generated for the wing-moving geomagnetic field and the muscle biomagnetic field. The response frequency for a preset time is obtained based on the bird wing kinetic magnetic field signal, and a frequency-energy distribution map is obtained based on the energy distribution characterization map and the response frequency. The frequency-energy distribution map is defined as a frequency domain eddy current spectrum. The total energy at each frequency point is obtained from the frequency domain eddy current spectrum. Based on the frequency domain eddy current spectrum, frequency band energies greater than a preset threshold are selected and used as the main frequency band energies. The main frequency band energy proportion coefficient is calculated based on the ratio of the main frequency band energy to the total energy.

[0008] Preferably, the step of constructing a bird correlation matrix based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy proportion coefficient includes: Based on the initial bird feature data, the initial bird feature data is mapped to the coding space according to a preset reception time sequence for encoding to obtain the initial bird feature code, wherein the initial bird feature code includes body shape code, biomarker code and spatial behavior feature code; The body shape encoding, the biomarker encoding, and the spatial behavior feature encoding are vector-mapped to obtain the body shape encoding vector, the biomarker encoding vector, and the spatial behavior feature encoding vector; By performing vector mapping between the compressive strength of the bird tendon and the energy proportion coefficient of the main frequency band, we can obtain the bird tendon compressive strength vector and the main frequency band energy proportion coefficient vector. An initial coordination matrix is ​​formed by combining the body shape encoding vector, the biomarker encoding vector, the spatial behavior feature encoding vector, the bird tendon compressive strength vector, and the main frequency band energy proportion coefficient vector in a column-wise combination. The initial coordination matrix is ​​standardized to eliminate the dimensional differences of different parameters, thus obtaining the bird correlation matrix.

[0009] Preferably, the step of identifying the bird association matrix based on a distributed bird feature database to obtain bird credentials includes: The covariance matrix is ​​obtained by identifying the bird correlation matrix. The covariance matrix is ​​decomposed into eigenvalues, and multiple eigenvectors corresponding to the top K largest eigenvalues ​​are extracted. Based on the feature space formed by the multiple feature vectors, a core factor matrix is ​​generated, wherein each column of the core factor matrix corresponds to a bird feature; Based on a distributed bird feature database, multiple bird features are identified to obtain bird credentials.

[0010] This application also provides a multimodal bird identification system, including: The first acquisition module is used to acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data. The second acquisition module is used to acquire regional location data and bird biological carrier data based on the biological trace data, and to acquire bird generalization data based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird feature data. The third acquisition module is used to acquire atmospheric ion distribution information and bird landing shock wave spectrum based on the environmental disturbance data, and to acquire bird tendon compressive strength based on the atmospheric ion distribution information and bird landing shock wave spectrum. The fourth acquisition module is used to acquire bird wing-moving magnetic field signals based on the biomagnetic field data, convert the bird wing-moving magnetic field signals into frequency domain eddy current spectra, and acquire the main frequency band energy proportion coefficient based on the frequency domain eddy current spectra. The construction module is used to construct a bird correlation matrix based on the initial bird characteristic data, the compressive strength of the bird tendons, and the energy proportion coefficient of the main frequency band; The identification module is used to identify the bird association matrix based on the distributed bird feature database to obtain bird credentials.

[0011] Preferably, the second acquisition module includes: The first acquisition unit is used to acquire bird fecal contaminant information and scattered feather appearance information based on the bird biological carrier data, acquire hormone level information and digestive metabolism information based on the bird fecal contaminant information, and acquire physiological characteristic information based on the hormone level information and digestive metabolism information. The second acquisition unit is used to acquire feather color information and feather shape information based on the scattered feather appearance information, acquire appearance feature information based on the feather color information and feather shape information, and acquire bird generalization data based on the physiological feature information and the appearance feature information. The comparison unit is used to compare the data with a pre-set bird migration route database and regional location data to obtain information on migratory birds. The third acquisition unit is used to acquire multiple bird population information based on the migratory bird information, and to filter the bird generalization data based on the multiple bird population information to obtain initial bird feature data.

[0012] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0013] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0014] The beneficial effects of this application are as follows: This invention constructs a multimodal recognition system by integrating three types of data: biological traces, environmental disturbances, and biomagnetic fields. First, regional location and bird biological carrier information are extracted from the biological trace data, and initial bird feature data is generated by filtering using a pre-set migration path database. Second, the atmospheric ion distribution and landing shock wave spectrum in the environmental disturbance data are converted into the compressive strength of bird tendons. Body weight and landing velocity are indirectly calculated using the instantaneous peak value of the shock wave and atmospheric ion parameters, establishing mechanical behavior characteristics. Next, the wing kinetic magnetic field signal in the biomagnetic field data is converted into a frequency domain eddy current spectrum, and the energy proportion coefficient of the main frequency band is extracted as a biomagnetic fingerprint feature. Finally, the above three types of features are standardized and encoded into a bird association matrix. Parallel feature matching and covariance matrix decomposition are performed using a distributed feature database, achieving highly robust bird recognition in complex environments. This method overcomes the limitations of traditional single-modal methods and can still work stably in occluded or noisy scenarios. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0017] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] like Figures 1-3 As shown, this application provides a multimodal bird identification method and system, including: S1. Acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data; S2. Obtain regional location data and bird biological carrier data based on the biological trace data, and obtain bird generalization data based on the bird biological carrier data. Filter the bird generalization data based on the regional location data to obtain initial bird characteristic data. S3. Obtain atmospheric ion distribution information and bird landing shock wave spectrum based on the environmental disturbance data, and obtain bird tendon compressive strength based on the atmospheric ion distribution information and bird landing shock wave spectrum. S4. Obtain bird wing-moving magnetic field signals based on the biomagnetic field data, and convert the bird wing-moving magnetic field signals into frequency domain eddy current spectra. Obtain the main frequency band energy proportion coefficient based on the frequency domain eddy current spectra. S5. Construct a bird correlation matrix based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy ratio coefficient; S6. Based on the distributed bird feature database, identify the bird association matrix to obtain bird credentials.

[0021] As described in steps S1-S6 above, existing bird identification methods in outdoor bird detection typically rely on extracting bird audio features from mellitus spectrograms and image acquisition for bird identification. However, environmental noise interference causes a sharp drop in recognition rate when the signal-to-noise ratio falls below 10dB, and it cannot distinguish between songbirds of the same frequency distributed in the same region (such as great tits and marsh tits). Furthermore, the acquired images are easily obstructed, making acquisition impossible. Therefore, audio and image methods are not suitable for bird identification. This invention first acquires multimodal bird identification data, including biological trace data, environmental disturbance data, and biomagnetic field data. This multidimensional data approach can overcome the limitations of traditional single-modality methods (such as audio or image-only methods). By overcoming the limitations of traditional methods, this system integrates three types of data—biological traces, environmental disturbances, and biomagnetic fields—to form a multi-dimensional identification system. This system can comprehensively capture bird characteristics, reduce identification errors caused by environmental interference or feature loss in single modalities, and ensure that biomagnetic field data can be stably acquired in complex environments such as image occlusion and high audio noise. This compensates for the shortcomings of traditional methods in terms of sharp drop in recognition rate under harsh conditions, improving the robustness of the system. Furthermore, different modalities of data reflect bird characteristics from multiple aspects such as physiology, behavior, and environmental interaction. For example, biological trace data reflects physiological characteristics, environmental disturbance data reflects the impact of behavior on the environment, and biomagnetic field data reveals movement characteristics. These elements complement each other and improve the accuracy of identification. The specific implementation process is as follows: Biological trace data is collected from bird activity areas by deploying a sensor network to collect biological carriers such as feces and feathers. Spectroscopic analysis equipment is used to detect hormone levels and digestive metabolite components in the feces. Image acquisition equipment is used to obtain appearance information such as feather color and shape. Environmental disturbance data is recorded in real time using an atmospheric ion monitor to record ion distribution information in the monitored area, including parameters such as ion charge gradient and mobility. Shock wave sensors are deployed at potential bird landing points to collect the spectral signals of shock waves generated when birds land. Biomagnetic field data is collected using high-sensitivity magnetic sensors (e.g., integrated tunneling magnetoresistive (TMR) sensors and CMOS technology) to monitor magnetic field signals generated during bird wing movement, including changes in the wing-movement geomagnetic field and muscle biomagnetic field, ensuring effective acquisition during bird flight or roosting. All of the above sensors are installed according to a pre-defined data collection site. Next, regional location data and bird biological carrier data are obtained based on the biological trace data, and bird generalization data is obtained based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird feature data. In this way, physiological features (hormones, metabolism) and appearance features (color, shape) are extracted from feces and feathers to form comprehensive bird generalization data, avoiding the one-sidedness of relying solely on appearance or physiological features. Furthermore, by comparing the data with the regional location data based on the preset migration route database, the range of bird populations is narrowed, the workload of feature filtering is reduced, and the identification efficiency is improved. At the same time, interference from birds in non-migratory areas is eliminated, improving the targeting of identification. Secondly, the generalization data is filtered based on migratory bird information to make the initial features more consistent with the actual situation of birds in the target area, reduce interference from irrelevant features, and improve the accuracy of subsequent identification. The process of acquiring biological trace data involves laboratory analysis of bird fecal contaminant information, detection of hormone types and concentrations using chromatography-mass spectrometry, acquisition of the composition and content of digestive metabolites through biochemical analysis, and inference of physiological characteristics of birds, such as metabolic rate and health status. Furthermore, image recognition technology is used to process the appearance information of scattered feathers, analyzing feather color parameters (such as RGB values ​​and spectral reflectance) and shape characteristics (such as length, width, and barb distribution) to generate appearance feature information. Simultaneously, the acquired regional location data (such as latitude, longitude, and altitude) is spatially matched with a pre-set bird migration route database (containing information on the migration routes and stopover areas of different birds) to determine the species of migratory birds that may appear in the current area, thus obtaining migratory bird information. Then, atmospheric ion distribution information and bird landing shock wave spectrum are obtained based on the environmental disturbance data. Furthermore, the tendon compressive strength of the birds is obtained based on these atmospheric ion distribution information and the bird landing shock wave spectrum. This transforms the bird landing behavior into a quantifiable tendon compressive strength index, characterizing bird behavior from a mechanical perspective and providing a unique physical dimension for identification. Moreover, through multi-parameter calculations of atmospheric ion distribution and landing shock wave spectrum, a correlation model between the environment and bird behavior is established, making feature extraction more scientific and logical. This method avoids directly capturing birds; by indirectly obtaining weight and landing speed through shock wave spectrum and atmospheric ion data, it avoids disturbing the birds and meets ecological protection requirements. The process of analyzing atmospheric ion distribution information and the bird landing shock wave spectrum involves: analyzing atmospheric ion distribution information to extract ion charge gradient (the change in ion charge per unit distance) and ion mobility (the speed at which ions move in an electric field). An atmospheric ion distribution coefficient, reflecting the influence of the atmospheric environment on bird landing, is calculated using a formula. The bird landing shock wave spectrum is then processed to identify the instantaneous peak value (the maximum pressure generated at the moment of landing) and the attenuation coefficient (the rate at which pressure decays over time). Using a shock wave mechanics model, the bird's weight is estimated based on these two parameters. Furthermore, combining the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, the bird's landing velocity is calculated using fluid dynamics and kinetic models. Finally, based on the landing velocity and the estimated weight, the compressive strength that the bird's tendons can withstand during landing is calculated using materials mechanics formulas. Next, bird wing-movement magnetic field signals are obtained based on the biomagnetic field data, and these signals are converted into frequency domain eddy current spectra. The main frequency band energy ratio coefficient is then obtained from the frequency domain eddy current spectra. In this way, the wing-movement magnetic field signals reflect the muscle activity and wing movement patterns of birds during flight. The frequency domain eddy current spectra and main frequency band energy ratios of different birds are unique and can be used as a "biomagnetic fingerprint" for identification. Furthermore, the magnetic field signals are converted into energy distribution characterization maps and frequency-energy distribution maps through time-frequency transformation technology, which finely characterizes the magnetic field signal features from both time and frequency dimensions, improving feature resolution. Moreover, since biomagnetic field signals are physical field signals, they are less affected by environmental noise (such as sound waves and optical interference) and can still be stably acquired in complex environments, making up for the shortcomings of audio and image recognition. The specific implementation process is as follows: Spectral analysis is performed on the bird wing-movement magnetic field signal to separate the wing-movement geomagnetic field (the disturbance of the geomagnetic field by bird wing movement) and the muscle biomagnetic field (the magnetic field generated by muscle electrical activity during wing movement). These are processed separately, and time-frequency transformation techniques such as short-time Fourier transform or wavelet transform are used to convert the time-domain signals of the wing-movement geomagnetic field and the muscle biomagnetic field into frequency-domain signals, generating an energy distribution characterization map to show the energy distribution of different frequency components. Based on the response frequency of the bird wing-movement magnetic field signal within a preset time period (e.g., the dominant frequency within one wing movement cycle), combined with the energy distribution characterization map, a frequency-energy distribution map, i.e., a frequency-domain eddy current spectrum, is plotted to visually present the frequency characteristics and energy distribution of the magnetic field signal. Finally, the energy at all frequency points in the frequency-domain eddy current spectrum is summed to obtain the total energy. The energy in the frequency band with a duration greater than a preset threshold (e.g., 50ms) is selected as the dominant frequency band energy, and the ratio of the dominant frequency band energy to the total energy is calculated to obtain the dominant frequency band energy proportion coefficient. Next, a bird association matrix is ​​constructed based on the initial bird feature data, the bird tendon compressive strength, and the main frequency band energy proportion coefficient. This encodes and standardizes multi-source features to form a unified-dimensional association matrix, providing structured data for subsequent species identification. It also eliminates the dimensional differences of different features (such as physiological indicators and physical parameters), facilitating feature fusion and pattern recognition by machine learning models. Furthermore, features are coded into body shape, biomarkers, spatial behavior, etc., making the matrix structure clearer, which is convenient for subsequent feature analysis and recognition model processing, improving feature expression capabilities. Through encoding and vector mapping, high-dimensional features are transformed into vectors of a unified dimension, reducing data dimensionality while retaining key feature information, improving computational efficiency and model training speed. Finally, the bird association matrix is ​​identified based on the distributed bird feature database to obtain bird credentials. This method, through covariance matrix eigenvalue decomposition, enables feature dimensionality reduction, retains the most discriminative key features, and reduces redundant information interference. Furthermore, by storing a large amount of bird feature data in the distributed bird feature database and performing parallel computing to match and identify features in the core factor matrix, the identification speed and scale are improved, making it suitable for large-scale bird population identification. The distributed feature database can be continuously updated and expanded to adapt to changes in the characteristics of newly discovered birds or species. At the same time, the eigenvalue decomposition method has a certain suppression effect on data noise, improving the robustness and scalability of the identification system. Thus, it is possible to identify birds even when bird audio features and image acquisition are affected.

[0022] In one embodiment, step S2, which involves obtaining bird generalization data based on the bird biological carrier data and filtering the bird generalization data based on regional location data to obtain initial bird feature data, includes: S201. Obtain bird fecal contaminant information and scattered feather appearance information based on the bird biological carrier data, obtain hormone level information and digestive metabolism information based on the bird fecal contaminant information, and obtain physiological characteristic information based on the hormone level information and digestive metabolism information. S202. Obtain feather color information and feather shape information based on the scattered feather appearance information, obtain appearance feature information based on the feather color information and feather shape information, and obtain bird generalization data based on the physiological feature information and the appearance feature information. S203. Based on the preset bird migration route database and regional location data, obtain migratory bird information; S204. Obtain multiple bird population information based on the migratory bird information, and filter the bird generalization data based on the multiple bird population information to obtain initial bird characteristic data.

[0023] As described in steps S201-S204 above, this invention first obtains information on bird fecal contaminants and the appearance of scattered feathers based on the bird biological carrier data, and then obtains hormone level information and digestive metabolism information based on the bird fecal contaminant information, and finally obtains physiological characteristic information based on the hormone level information and digestive metabolism information. This breaks through the limitations of traditional visual recognition, distinguishes similar-looking species through physiological indicators, and provides molecular-level data support for ecological research. Moreover, closely related species distributed in the same domain (such as great tits and marsh tits) have subtle differences in appearance, while the hormone levels and metabolites in feces are species-specific and can be used as key distinguishing features. For example, when collecting blood pheasant feces in the Qinling Nature Reserve, the cortisol concentration (a characteristic of the breeding season) was detected by mass spectrometry analysis. Combined with digestive metabolism information (protein absorption rate), it was confirmed that the individual was in a breeding active state, thus distinguishing it from individuals in the non-breeding season. Next, feather color and shape information are obtained based on the scattered feather appearance information, and appearance feature information is obtained based on the feather color and shape information. Bird generalization data is obtained based on the physiological feature information and the appearance feature information. In this way, through cross-analysis of fecal contaminants and feather appearance information, physiological indicators such as hormone levels and digestive metabolism are combined with appearance features such as feather color and shape to form a "physiological-appearance" complementary generalization data system, avoiding the one-sidedness of single-dimensional features and improving the completeness of feature description. Secondly, by comparing the preset bird migration route database with regional location data, information on migratory birds can be obtained. This spatial matching based on the preset migration route database and regional location data can dynamically narrow the range of target bird populations, making feature selection have clear geographical orientation and reducing interference from irrelevant species. Finally, based on the migratory bird information, multiple bird population information is obtained, and the bird generalization data is filtered based on these multiple bird population information to obtain initial bird feature data. This avoids interference from non-target population features and improves recognition accuracy. For example, it can accurately locate the target species among multiple populations distributed in the same region. For instance, traditional bird identification directly uses all feature data; when analyzing the features of 200 bird species, the computational load increases exponentially with the dimensionality. This step reduces the target population categories through migration path matching, thereby improving recognition accuracy.

[0024] In one embodiment, step S3, which involves obtaining the compressive strength of a bird's tendons based on the atmospheric ion distribution information and the spectrum of the bird's landing shock wave, includes: S301. Obtain the ion charge gradient and ion mobility based on the atmospheric ion distribution information, and obtain the atmospheric ion distribution coefficient based on the ion charge gradient and the ion mobility. S302. Obtain the instantaneous peak value and shock wave attenuation coefficient of the shock wave based on the spectrum of the bird landing shock wave, and obtain the bird's weight based on the instantaneous peak value and the shock wave attenuation coefficient. S303. Obtain the bird's landing speed based on the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, and obtain the bird's tendon compressive strength based on the bird's landing speed and the bird's weight.

[0025] As described in steps S301-S303 above, this invention first obtains the ion charge gradient and ion mobility based on the atmospheric ion distribution information. Specifically, an atmospheric ion monitoring instrument (such as the AIM-2000 model) is deployed in the bird activity area to monitor the concentration of positive and negative ions in real time, obtaining the ion charge gradient (unit: nC / m) and ion mobility (unit: cm² / (V·s)). For example, at a wetland monitoring point, the positive ion charge gradient is measured to be 2.3 nC / m, and the negative ion mobility is 1.8 cm² / (V·s). The atmospheric ion distribution coefficient is then obtained based on the ion charge gradient and the ion mobility, using the Nernst-Einstein equation and combining the ion charge gradient (G) and mobility (K). c is the atmospheric ion distribution coefficient, and G is the ion charge gradient. Bird weight and landing velocity mobility are indirectly obtained through atmospheric ion distribution and landing shock wave spectrum, avoiding interference with birds by traditional capture and measurement, meeting ecological protection requirements, and realizing quantitative analysis of dynamic behavior. Furthermore, atmospheric physical parameters (ion charge gradient, mobility) and biomechanical parameters (shock wave peak value, attenuation coefficient) are coupled and weighted for calculation. Standardization processing is required before calculation to eliminate the dimensional differences of different parameters, establish environmental-biological interactive data, make feature extraction more physically logical, and improve the interpretability of features. Then, the instantaneous peak value and attenuation coefficient of the shock wave are obtained from the spectrum of the bird landing shock wave, and the bird's weight is obtained from the instantaneous peak value and attenuation coefficient. In this way, the instantaneous peak value and attenuation coefficient are extracted from the landing shock wave spectrum to deduce the bird's weight, realizing body size estimation under non-visual conditions. Secondly, the shock wave parameters are positively correlated with the bird's mass, which can overcome the limitations of visual obstruction and provide physical characteristics for bird identification in concealed environments (such as dense forests and night). Furthermore, the mechanical waves generated when birds land contain species-specific information. The instantaneous peak value reflects the magnitude of the impact force, and the attenuation coefficient reflects the body size and ground buffering capacity, which is an important supplement to traditional audio / image recognition. For example, piezoresistive pressure sensors can be buried at the possible landing points of birds (such as mudflats and treetops) to collect the time-domain signal of the shock wave when birds land. For example, the peak value of the shock wave waveform measured when a white spoonbill lands is 75 Pa, and the attenuation time constant is 0.15 s. Wavelet transform is used to denoise the shock wave spectrum to extract the instantaneous peak value (Ppeak, in Pa) and the attenuation coefficient (α, in 1 / s). The center frequency of the shock wave was calculated using Hilbert transform, and its correlation with the bird's body weight was verified. Finally, the bird's landing speed is obtained based on the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, and the bird's tendon compressive strength is obtained based on the bird's landing speed and the bird's weight. In this way, the landing speed is derived by combining the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, and then the tendon compressive strength is calculated by combining it with the weight. This reflects the bird's physiological function and behavioral adaptability. At the same time, tendon compressive strength is a key indicator of the bird's flight ability and landing stability, and can be used to distinguish closely related species (such as the significant difference in tendon strength between birds of prey and songbirds). Specific process: Bird landing speed Where v is the landing speed of birds, The instantaneous peak value of the shock wave (extracted from the shock wave spectrum of bird landing). Let be the air density (taken as 1.2 kg / m³ under standard conditions), and c be the atmospheric ion distribution coefficient. For ease of calculation, the above process can be standardized first to eliminate dimensional differences between different parameters. In one embodiment, step S4, which involves converting the bird wing-kinematic magnetic field signal into a frequency domain eddy current spectrum and obtaining the main frequency band energy proportion coefficient based on the frequency domain eddy current spectrum, includes: S401. Obtain the wing-moving geomagnetic field and muscle biomagnetic field based on the bird wing-moving magnetic field signal; S402. Based on time-frequency transformation technology, an energy distribution characterization map is generated for the wing-moving geomagnetic field and the muscle biomagnetic field. S403. Obtain the response frequency for a preset time based on the bird wing kinetic magnetic field signal, obtain the frequency-energy distribution map based on the energy distribution characterization map and the response frequency, and define the frequency-energy distribution map as a frequency domain eddy current spectrum. S404. Obtain the total energy at each frequency point based on the frequency domain eddy current spectrum. S405. Select the frequency band energy that is greater than a preset threshold according to the frequency domain eddy current spectrum and take it as the main frequency band energy. Calculate the main frequency band energy ratio coefficient according to the ratio of the main frequency band energy to the total energy.

[0026] As described in steps S401-S405 above, the present invention first obtains the wing-movement geomagnetic field and muscle biomagnetic field based on the bird wing-movement magnetic field signal. In this way, the geomagnetic field component (environmental background magnetic field) and the muscle biomagnetic field (bioelectric signal generated by muscle contraction) are separated from the bird wing-movement magnetic field signal, providing a pure biological signal source for subsequent time-frequency analysis. The geomagnetic field component can be used to calibrate environmental interference. The muscle biomagnetic field directly reflects the muscle activity pattern during wing movement and is a key feature for distinguishing the flight behavior of different birds. At the same time, when birds move their wings, the muscle electrical activity will generate a weak magnetic field. This signal carries species-specific motion characteristics (such as wing movement frequency and muscle force exertion timing), while traditional audio / image recognition cannot capture such bioelectromagnetic signals. Then, based on time-frequency transformation technology, an energy distribution characterization map is generated from the wing-moving geomagnetic field and the muscle biomagnetic field. In this way, time-frequency transformation technology (such as short-time Fourier transform) is used to convert the time-domain magnetic field signal into a three-dimensional "time-frequency-energy" distribution characterization map, which intuitively shows the distribution characteristics of energy in different frequency bands during wing movement. Secondly, time-frequency analysis can capture the transient changes of the wing-moving magnetic field (such as frequency shifts during acceleration / deceleration). Differences in energy distribution can reflect physical attributes such as bird size and wingspan. Moreover, time-domain signals are difficult to directly reflect frequency characteristics. Time-frequency transformation converts the magnetic field signal into a visualized energy spectrum, which is convenient for extracting key parameters such as the main frequency band, thus solving the limitations of traditional time-domain analysis. Next, the response frequency for a preset time is obtained based on the bird's wing-movement magnetic field signal. A frequency-energy distribution map is obtained based on the energy distribution characterization map and the response frequency. The frequency-energy distribution map is defined as a frequency domain eddy current spectrum. In this way, by combining the response frequency (such as the average frequency within a preset 1 second) and the energy distribution characterization map, a two-dimensional "frequency-energy" distribution map (frequency domain eddy current spectrum) is generated, which focuses on the main energy frequency bands of the wing-movement magnetic field. At the same time, the frequency domain eddy current spectrum can quantify the energy proportion of different frequency bands, reflecting the muscle force intensity and frequency stability during bird wing movement, and providing electromagnetic characteristic basis for species classification. Secondly, the total energy of frequency points is obtained from the frequency domain eddy current spectrum. By accumulating the energy values ​​of all frequency points in the frequency domain eddy current spectrum, the total energy value is obtained, which reflects the overall intensity of muscle activity during wing movement. The total energy is positively correlated with the bird's body size and wing movement amplitude (e.g., the total energy of wing movement of birds of prey is higher than that of songbirds). It can be used as an auxiliary feature to distinguish different groups of birds. Furthermore, the total energy value can quantify the "energy consumption level" of wing movement. Combined with the energy proportion of the main frequency band, it can more comprehensively describe the biomechanical characteristics of bird flight behavior. For example, the total energy of the frequency points of the wing movement magnetic field of a sparrowhawk is calculated to be -15dB·Hz, while the total energy of a common kestrel distributed in the same domain is -20dB·Hz. The former has a significantly higher total energy due to its larger body size and wider wing movement amplitude. Finally, based on the frequency domain eddy current spectrum, frequency band energies exceeding a preset threshold are selected and designated as the dominant frequency band energy. The dominant frequency band energy proportion coefficient is calculated based on the ratio of the dominant frequency band energy to the total energy. This process selects frequency band energies with durations exceeding a preset threshold (e.g., 500ms) as dominant frequency band energies, and calculates their ratio to the total energy to obtain the dominant frequency band energy proportion coefficient. Furthermore, through the technical path of "magnetic field signal decomposition - time-frequency transformation - spectrum construction - energy statistics," the shortcomings of traditional methods—lack of dynamic features, sensitivity to environmental interference, and insufficient time-frequency analysis—are effectively addressed. Its core advantage lies in utilizing the physical stability of biological magnetic field signals and the species-specific nature of time-frequency domain features to transform wing movement—a dynamic behavior—into a quantifiable dominant frequency band energy proportion index. Simultaneously, it provides a dynamic feature dimension unaffected by environmental interference for multimodal recognition, complementing modes such as biological traces and environmental disturbances, significantly improving the system's robustness in complex scenarios.

[0027] In one embodiment, step S5, which involves constructing a bird correlation matrix based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy proportion coefficient, includes: S501. Based on the initial bird feature data, the initial bird feature data is mapped to the coding space according to a preset reception time sequence for encoding to obtain the initial bird feature code, wherein the initial bird feature code includes body shape code, biomarker code and spatial behavior feature code; S502. Perform vector mapping on the body shape encoding, the biomarker encoding, and the spatial behavior feature encoding to obtain the body shape encoding vector, the biomarker encoding vector, and the spatial behavior feature encoding vector; S503. Vector mapping is performed between the compressive strength of the bird tendon and the energy proportion coefficient of the main frequency band to obtain the bird tendon compressive strength vector and the energy proportion coefficient vector of the main frequency band. S504. An initial cooperative matrix is ​​formed by combining the body shape encoding vector, the biomarker encoding vector, the spatial behavior feature encoding vector, the bird tendon compressive strength vector, and the main frequency band energy proportion coefficient vector in column order. S505. The initial coordination matrix is ​​standardized to eliminate the dimensional differences of different parameters, and the bird correlation matrix is ​​obtained.

[0028] As described in steps S501-S505 above, the present invention first maps the initial bird feature data to the coding space according to a preset reception time sequence for encoding, thereby obtaining the initial bird feature code. The initial bird feature code includes body shape coding, biomarker coding, and spatial behavior feature coding, thus establishing a three-dimensional coding space (body shape-biomarker-spatial behavior), with different coding rules for each dimension. For example, the body shape dimension includes five parameters such as body size index (body length / wingspan) and feather density, which are processed using one-hot coding and numerical normalization. Secondly, according to the feature acquisition time sequence (e.g., acquiring feather appearance first, then fecal physiological data), the feature parameters are mapped to the corresponding coordinate axes in the coding space to ensure the correlation of features over time. For example, the feather color collected at t=0 is encoded as [0.8,0.2,0], and the hormone level collected at t=10min is encoded as [0.3,0.7,0]. Then, physiological characteristics (such as testosterone concentration of 14ng / g), appearance characteristics (such as flight feather length diameter of 6.5cm), and spatial behavior characteristics (such as flight altitude of 30m) are mapped as encoding vectors and combined to form an initial feature encoding matrix. Secondly, the body shape encoding, the biomarker encoding, and the spatial behavior feature encoding are vector-mapped to obtain body shape encoding vector, biomarker encoding vector, and spatial behavior feature encoding vector. This converts various encodings (body shape, biomarker, and spatial behavior) into standard vector form, which facilitates matrix operations and feature fusion. The fusion process can rely on vector input through machine learning models (such as SVM and neural networks). The mapping process transforms discrete encodings into a continuous vector space, improving feature representation capabilities. Next, the compressive strength of the bird tendon and the energy proportion coefficient of the main frequency band are vector-mapped to obtain the bird tendon compressive strength vector and the energy proportion coefficient vector of the main frequency band. In this way, physical parameters such as bird tendon compressive strength and energy proportion of the main frequency band are converted into vectors, which are unified with the feature encoding vector and realize cross-domain fusion of biological features and physical features. For example, the tendon compressive strength vector can reflect the mechanical characteristics of birds and complement their body features. Next, an initial collaborative matrix is ​​formed by combining the body posture encoding vector, the biomarker encoding vector, the spatial behavior feature encoding vector, the bird tendon compressive strength vector, and the dominant frequency band energy proportion coefficient vector column by column. This column-by-column combination of various feature vectors forms an initial collaborative matrix, realizing the structured integration of multimodal features. The matrix form facilitates subsequent feature dimensionality reduction and pattern recognition. For example, the rows of the matrix represent samples, and the columns represent different feature dimensions. At the same time, multimodal features need to be jointly represented through the matrix form. The collaborative matrix can capture the correlation between features (such as the correlation between body posture and tendon strength), providing multidimensional decision-making basis for species identification. Finally, the initial collaborative matrix is ​​standardized to eliminate dimensional differences in different parameters, resulting in a bird association matrix. Through a technical approach of "feature encoding-vector mapping-matrix standardization," the core problems of dimensional differences, dimensional explosion, and feature redundancy in multimodal recognition are systematically solved. Its core advantage lies in transforming heterogeneous data such as biological traces, environmental disturbances, and biomagnetic fields into a uniform-scale association matrix. This preserves the uniqueness of each modality's features while effectively addressing issues such as poor compatibility of multi-source data, low efficiency of high-dimensional computation, and interference from redundant features. It provides a high-quality matrix representation for subsequent accurate recognition based on a distributed feature library, fully leveraging the complementary advantages of multimodal features and improving recognition accuracy and generalization ability after standardization.

[0029] In one embodiment, step S6, which involves identifying the bird association matrix based on a distributed bird feature database to obtain bird credentials, includes: S601. Identify the bird correlation matrix to obtain the covariance matrix; S602. Perform eigenvalue decomposition on the covariance matrix and extract multiple eigenvectors corresponding to the first K largest eigenvalues; S603. Generate a core factor matrix based on the feature space formed by the multiple feature vectors, wherein each column of the core factor matrix corresponds to a bird feature; S604. Based on the distributed bird feature database, identify multiple bird features to obtain bird credentials.

[0030] As described in steps S601-S604 above, the present invention first identifies the bird correlation matrix to obtain a covariance matrix. Then, statistical analysis is performed on the bird correlation matrix to calculate the covariance between each feature dimension, generating a covariance matrix that reflects the correlation between features (such as the correlation between body shape features and tendon strength). The variance matrix can reveal the internal structure of multimodal features. For example, positive covariance indicates that features increase or decrease together, while negative covariance indicates that the feature change trends are opposite, providing a basis for subsequent feature dimensionality reduction. Secondly, eigenvalue decomposition is performed on the covariance matrix, and multiple eigenvectors corresponding to the top K largest eigenvalues ​​are extracted. Principal component analysis (PCA) is used for dimensionality reduction, eliminating redundant features and retaining the core features that contribute most to species identification, thus improving computational efficiency and model generalization ability. The magnitude of the eigenvalues ​​in the covariance matrix reflects the importance of the features, and the eigenvectors corresponding to the top K largest eigenvalues ​​can capture the information entropy in the original data, filtering out noise and secondary features (such as individual random variations, measurement errors, etc.). For example, bird association matrices typically contain multimodal features such as physiological, mechanical, and magnetic fields (e.g., more than 10 dimensions), and direct processing would require enormous computation. Eigenvalue decomposition can compress the high-dimensional feature space to a low dimension (within 5 dimensions in this scheme), retaining the most discriminative information. Then, based on the feature space formed by the multiple feature vectors, a core factor matrix is ​​generated, where each column of the core factor matrix corresponds to a bird feature. The K orthogonalized feature vectors are arranged column-wise to form a feature space matrix U=[u1,u2,…,uK]. Each column of this matrix corresponds to a core factor (such as the principal factor of body posture, principal factor of mechanics, etc.). Next, the original bird correlation matrix X is projected onto the feature space to obtain the core factor matrix F=XU, where each row of F represents the coordinates of a sample in the K-dimensional core factor space. At the same time, the contribution (loading) of each original feature to the core factor is calculated. U represents the feature space matrix, F represents the core factor matrix, and X represents the original bird correlation matrix. Finally, based on a distributed bird feature library, multiple bird features are identified to obtain bird credentials. The distributed bird feature library uses HDFS to store standard feature templates, and a Spark cluster is used for parallel computation. Each node stores core factor templates for some species (e.g., Passeriformes and Falconiformes are stored separately on different nodes). The identification process involves calculating the cosine similarity between the sample feature vectors in the core factor matrix and the template vectors in the feature library. Once the cosine similarity match reaches a preset threshold, a bird credential containing the species name, similarity score, and feature matching details is generated. This enables bird identification even when bird audio features and image acquisition are affected.

[0031] This application also provides a multimodal bird identification system, including: The first acquisition module is used to acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data. The second acquisition module is used to acquire regional location data and bird biological carrier data based on the biological trace data, and to acquire bird generalization data based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird feature data. The third acquisition module is used to acquire atmospheric ion distribution information and bird landing shock wave spectrum based on the environmental disturbance data, and to acquire bird tendon compressive strength based on the atmospheric ion distribution information and bird landing shock wave spectrum. The fourth acquisition module is used to acquire bird wing-moving magnetic field signals based on the biomagnetic field data, convert the bird wing-moving magnetic field signals into frequency domain eddy current spectra, and acquire the main frequency band energy proportion coefficient based on the frequency domain eddy current spectra. The construction module is used to construct a bird correlation matrix based on the initial bird characteristic data, the compressive strength of the bird tendons, and the energy proportion coefficient of the main frequency band; The identification module is used to identify the bird association matrix based on the distributed bird feature database to obtain bird credentials.

[0032] In one embodiment, the second acquisition module includes: The first acquisition unit is used to acquire bird fecal contaminant information and scattered feather appearance information based on the bird biological carrier data, acquire hormone level information and digestive metabolism information based on the bird fecal contaminant information, and acquire physiological characteristic information based on the hormone level information and digestive metabolism information. The second acquisition unit is used to acquire feather color information and feather shape information based on the scattered feather appearance information, acquire appearance feature information based on the feather color information and feather shape information, and acquire bird generalization data based on the physiological feature information and the appearance feature information. The comparison unit is used to compare the data with a pre-set bird migration route database and regional location data to obtain information on migratory birds. The third acquisition unit is used to acquire multiple bird population information based on the migratory bird information, and to filter the bird generalization data based on the multiple bird population information to obtain initial bird feature data.

[0033] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0034] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0035] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0036] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0037] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A multimodal bird identification method, characterized in that, include: Acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data; Based on the biological trace data, regional location data and bird biological carrier data are obtained, and bird generalization data is obtained based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird characteristic data. Atmospheric ion distribution information and bird landing shock wave spectrum are obtained based on the environmental disturbance data, and bird tendon compressive strength is obtained based on the atmospheric ion distribution information and bird landing shock wave spectrum. The bird wing kinetic magnetic field signal is obtained based on the biomagnetic field data, and the bird wing kinetic magnetic field signal is converted into a frequency domain eddy current spectrum. The main frequency band energy proportion coefficient is obtained based on the frequency domain eddy current spectrum. A bird correlation matrix is ​​constructed based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy proportion coefficient. The bird association matrix is ​​identified based on a distributed bird feature database to obtain bird credentials.

2. The multimodal bird identification method according to claim 1, characterized in that, The steps of obtaining bird generalization data based on the bird biological carrier data, filtering the bird generalization data based on regional location data, and obtaining initial bird characteristic data include: Based on the bird biological carrier data, information on bird fecal contaminants and the appearance of scattered feathers is obtained; based on the bird fecal contaminant information, information on hormone levels and digestive metabolism is obtained; and based on the hormone levels and digestive metabolism information, information on physiological characteristics is obtained. Based on the scattered feather appearance information, feather color information and feather shape information are obtained, and based on the feather color information and feather shape information, appearance feature information is obtained, and based on the physiological feature information and appearance feature information, bird generalization data is obtained; Information on migratory birds is obtained by comparing a pre-set database of bird migration routes with regional location data. Based on the migratory bird information, multiple bird population information is obtained, and based on the multiple bird population information, bird generalization data is filtered to obtain initial bird characteristic data.

3. The multimodal bird identification method according to claim 1, characterized in that, The step of obtaining the compressive strength of bird tendons based on the atmospheric ion distribution information and the bird landing shock wave spectrum includes: The atmospheric ion distribution information is used to obtain the ion charge gradient and ion mobility, and the atmospheric ion distribution coefficient is obtained based on the ion charge gradient and the ion mobility. The instantaneous peak value and attenuation coefficient of the shock wave are obtained from the spectrum of the bird landing shock wave, and the bird's weight is obtained from the instantaneous peak value and attenuation coefficient of the shock wave. The bird's landing speed is obtained based on the atmospheric ion distribution coefficient and the instantaneous peak value of the shock wave, and the bird's tendon compressive strength is obtained based on the bird's landing speed and the bird's weight.

4. The multimodal bird identification method according to claim 1, characterized in that, The step of converting the bird wing kinetic magnetic field signal into a frequency domain eddy current spectrum and obtaining the main frequency band energy proportion coefficient based on the frequency domain eddy current spectrum includes: The wing-induced geomagnetic field and muscle biomagnetic field were obtained based on the bird wing-induced magnetic field signal. Based on time-frequency transformation technology, an energy distribution characterization map is generated for the wing-moving geomagnetic field and the muscle biomagnetic field. The response frequency for a preset time is obtained based on the bird wing kinetic magnetic field signal, and a frequency-energy distribution map is obtained based on the energy distribution characterization map and the response frequency. The frequency-energy distribution map is defined as a frequency domain eddy current spectrum. The total energy at each frequency point is obtained from the frequency domain eddy current spectrum. Based on the frequency domain eddy current spectrum, frequency band energies greater than a preset threshold are selected and used as the main frequency band energies. The main frequency band energy proportion coefficient is calculated based on the ratio of the main frequency band energy to the total energy.

5. The multimodal bird identification method according to claim 1, characterized in that, The step of constructing a bird correlation matrix based on the initial bird characteristic data, the bird tendon compressive strength, and the dominant frequency band energy proportion coefficient includes: Based on the initial bird feature data, the initial bird feature data is mapped to the coding space according to a preset reception time sequence for encoding to obtain the initial bird feature code, wherein the initial bird feature code includes body shape code, biomarker code and spatial behavior feature code; The body shape encoding, the biomarker encoding, and the spatial behavior feature encoding are vector-mapped to obtain the body shape encoding vector, the biomarker encoding vector, and the spatial behavior feature encoding vector; By performing vector mapping between the compressive strength of the bird tendon and the energy proportion coefficient of the main frequency band, we can obtain the bird tendon compressive strength vector and the main frequency band energy proportion coefficient vector. An initial coordination matrix is ​​formed by combining the body shape encoding vector, the biomarker encoding vector, the spatial behavior feature encoding vector, the bird tendon compressive strength vector, and the main frequency band energy proportion coefficient vector in a column-wise combination. The initial coordination matrix is ​​standardized to eliminate the dimensional differences of different parameters, thus obtaining the bird correlation matrix.

6. The multimodal bird identification method according to claim 1, characterized in that, The step of identifying the bird association matrix based on a distributed bird feature database to obtain bird credentials includes: The covariance matrix is ​​obtained by identifying the bird correlation matrix. The covariance matrix is ​​decomposed into eigenvalues, and multiple eigenvectors corresponding to the top K largest eigenvalues ​​are extracted. Based on the feature space formed by the multiple feature vectors, a core factor matrix is ​​generated, wherein each column of the core factor matrix corresponds to a bird feature; Based on a distributed bird feature database, multiple bird features are identified to obtain bird credentials.

7. A multimodal bird recognition system, characterized in that, include: The first acquisition module is used to acquire multimodal recognition data of birds, including biological trace data, environmental disturbance data, and biomagnetic field data. The second acquisition module is used to acquire regional location data and bird biological carrier data based on the biological trace data, and to acquire bird generalization data based on the bird biological carrier data. The bird generalization data is then filtered based on the regional location data to obtain initial bird feature data. The third acquisition module is used to acquire atmospheric ion distribution information and bird landing shock wave spectrum based on the environmental disturbance data, and to acquire bird tendon compressive strength based on the atmospheric ion distribution information and bird landing shock wave spectrum. The fourth acquisition module is used to acquire bird wing-moving magnetic field signals based on the biomagnetic field data, convert the bird wing-moving magnetic field signals into frequency domain eddy current spectra, and acquire the main frequency band energy proportion coefficient based on the frequency domain eddy current spectra. The construction module is used to construct a bird correlation matrix based on the initial bird characteristic data, the compressive strength of the bird tendons, and the energy proportion coefficient of the main frequency band; The identification module is used to identify the bird association matrix based on the distributed bird feature database to obtain bird credentials.

8. A multimodal bird identification system according to claim 7, characterized in that, The second acquisition module includes: The first acquisition unit is used to acquire bird fecal contaminant information and scattered feather appearance information based on the bird biological carrier data, acquire hormone level information and digestive metabolism information based on the bird fecal contaminant information, and acquire physiological characteristic information based on the hormone level information and digestive metabolism information. The second acquisition unit is used to acquire feather color information and feather shape information based on the scattered feather appearance information, acquire appearance feature information based on the feather color information and feather shape information, and acquire bird generalization data based on the physiological feature information and the appearance feature information. The comparison unit is used to compare the data with a pre-set bird migration route database and regional location data to obtain information on migratory birds. The third acquisition unit is used to acquire multiple bird population information based on the migratory bird information, and to filter the bird generalization data based on the multiple bird population information to obtain initial bird feature data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images