A method for bird species identification based on all-solid-state radar

Abstract

The present invention provides a method for bird species identification based on an all-solid-state radar. This method relates to the field of radar data processing technology. The method uses an all-solid-state radar to identify bird species. The method comprises the following steps: first, acquiring raw echo signals and using multi-channel technology to separate the bird target echo data to obtain an initial dataset; then, using a Kalman filter algorithm combined with speed and direction information to generate continuous motion trajectories and determine a trajectory set; then, extracting target spacing features to distinguish between gregarious and solitary birds based on spacing and trajectory synchronization; then, combining echo signal intensity and frequency features, using a convolutional neural network to extract biological features such as wingspan and body length to obtain a target biological feature set; finally, matching the data with a bird feature database to determine the bird species based on similarity. This method overcomes the inherent limitations of relying on traditional optical equipment and manual monitoring, establishing an all-weather, high-precision dynamic monitoring system for ecological protection.

Description

Technical Field

[0001] The present invention relates to the technical field of radar data processing, and in particular to a method for identifying flying bird species based on an all-solid-state radar. Background Art

[0002] Bird ecological monitoring is an important area of ​​ecological protection and environmental science research, and is of key significance for maintaining biodiversity and assessing ecosystem health. By monitoring bird species, numbers, and their distribution, a scientific basis can be provided for the protection of ecological zones such as wetlands and forests. However, existing bird monitoring methods mostly rely on optical equipment or manual observation, which are easily restricted by environmental factors such as weather and light, and it is difficult to achieve large-scale, real-time and efficient tracking. In particular, there are obvious deficiencies in the accurate distinction between gregarious birds and solitary birds in complex environments. This restricts the comprehensiveness and accuracy of ecological data, making it difficult to meet the high standards of modern ecological monitoring. Summary of the Invention

[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to propose a method for bird species identification based on all-solid-state radar to solve the problems mentioned in the above background technology section.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for identifying bird species based on all-solid-state radar, the method comprising the following steps:

[0006] S1. Obtain the original echo signal through the all-solid-state radar, and use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain the initial target data set;

[0007] S2. For the initial target data set, a motion trajectory prediction algorithm based on Kalman filtering is used to generate a continuous motion trajectory for each target by combining the target speed and direction information, thereby determining the target motion trajectory set.

[0008] S3. Extract target spacing features from the target trajectory set. If the target spacing is less than a preset threshold and the trajectory synchronization is higher than a set standard, the target is judged to be a gregarious bird. Otherwise, the target is judged to be a solitary bird, and the classification results of gregarious birds and solitary birds are obtained;

[0009] S4. Combining the intensity and frequency characteristics of the all-solid-state radar echo signal, a convolutional neural network model is used to extract the biological characteristics of the wingspan length and body length of all solitary birds and several social birds, thereby obtaining a target biological characteristic set for each bird.

[0010] S5. Match the target biological feature set of each bird with a pre-established bird feature database. If the similarity between the biological feature and a bird feature in the database is higher than a preset threshold, the bird species is determined to be that bird, and a bird species identification result is obtained.

[0011] As an embodiment, step S1 specifically includes the following sub-steps:

[0012] S11, dividing the detection range of the all-solid-state radar into grids, and calculating the spatial distribution density and signal overlap of the original echo signal of each grid;

[0013] S12. When the spatial distribution density or signal overlap of the original echo signals of a certain grid is greater than a set threshold, it is preliminarily determined to be a "suspected gregarious bird flock" area;

[0014] S13. Obtain the number of original echo signals within the "suspected flock of gregarious birds" area, allocate a number of channels in proportion to the number, and then use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain an initial target data set.

[0015] As an embodiment, step S2 specifically includes the following sub-steps:

[0016] S21, obtaining speed and direction information of the target from the initial target data set to obtain a first feature set;

[0017] S22. Process the first feature set using a Kalman filter algorithm, fuse speed and direction information, generate an initial motion trajectory for each target, and obtain a first trajectory set;

[0018] S23, if the distance between any track points in the first track set exceeds a preset threshold, interpolation is used to supplement the intermediate points to obtain a second track set;

[0019] S24. Analyze the continuity of the trajectories using a sliding window method based on the second trajectory set, determine the dynamic update frequency of the trajectories, and obtain an update parameter set;

[0020] S25, adjusting the state transfer matrix of the Kalman filter by updating the parameter set to generate a dynamically updated motion trajectory to obtain a third trajectory set;

[0021] S26. For the third trajectory set, calculate the trajectory smoothness of each target, determine whether the trajectory meets the continuity requirement, and obtain the final trajectory set;

[0022] S27. If there are abnormal trajectories in the final trajectory set, correction is performed by re-integrating the first feature set to obtain the target motion trajectory set.

[0023] As an embodiment, step S3 specifically includes the following sub-steps:

[0024] S31, obtaining target motion trajectories from the target motion trajectory set, separating each target trajectory using trajectory segmentation technology, and obtaining a single target trajectory set;

[0025] S32. For a single target trajectory set, calculate the spatial distance between each target, calculate the target spacing feature using the Euclidean distance formula, and obtain a target spacing feature set;

[0026] S33, extracting a distance value from the target distance feature set, and if the distance value is less than a preset threshold, marking it as a potential gregarious bird target, thereby obtaining a potential gregarious bird target set;

[0027] S34. For the potential gregarious bird target set, the Pearson correlation coefficient is used to calculate the synchronization between trajectories and obtain the synchronization analysis results;

[0028] S35. Extracting a synchronization value from the synchronization analysis result. If the synchronization value is higher than a preset standard, the target is determined to be a gregarious bird; otherwise, the target is determined to be a solitary bird.

[0029] As an embodiment, step S4 specifically includes the following sub-steps:

[0030] S41, obtaining an initial signal feature set from the echo intensity and frequency of the all-solid-state radar through signal processing;

[0031] S42, performing frequency domain analysis on the radar echo intensity and frequency using fast Fourier transform to obtain a frequency domain feature set;

[0032] S43. Based on the frequency domain feature set and the behavioral pattern set of social birds and solitary birds, a convolutional neural network is used to extract wingspan length and body length features to generate a preliminary biological feature set;

[0033] S44. If the feature value of the preliminary biological feature set matches the preset threshold, the wingspan length and body length features are subjected to dimensionality reduction processing through principal component analysis to obtain an optimized biological feature set.

[0034] As an embodiment, step S5 specifically includes the following sub-steps:

[0035] S51, extracting feature data from the optimized biometric feature set, generating feature vectors using a standardized feature extraction algorithm, and obtaining a feature vector set;

[0036] S52, matching the feature vector set with a pre-established bird feature database, and calculating the similarity between the feature vectors using a cosine similarity algorithm to obtain a similarity value;

[0037] S53: If the similarity value exceeds the preset threshold, the corresponding bird species information is obtained from the bird feature database, the bird species is determined, a preliminary recognition result is obtained, and the process jumps to step S55;

[0038] S54: If the similarity value is lower than the preset threshold, return to step S4 and loop through steps S41 to S44;

[0039] S55. Based on the preliminary recognition results, a pre-established bird classification model is used to perform secondary classification on the feature vectors using a support vector machine algorithm to obtain an optimized recognition result;

[0040] S56. Based on the optimized recognition result, query the associated species information from the bird feature database, obtain species attribute data, and obtain species detailed information;

[0041] S57. Perform a consistency check between the species detailed information and the feature vector set. If the consistency check passes, output the final bird species identification result.

[0042] As an embodiment, the method further includes the following steps:

[0043] S6. For interference signals in complex environments, extract environmental noise characteristics from all-solid-state radar data processing, use adaptive filtering algorithms to denoise the original echo signals, generate an optimized target data set, and update the bird species recognition results.

[0044] As an embodiment, step S6 specifically includes the following sub-steps:

[0045] S61. Obtaining raw data from the echo signal of the all-solid-state radar, extracting time domain and frequency domain features using signal decomposition technology, and obtaining an environmental noise feature set;

[0046] S62: construct an adaptive filtering model based on the environmental noise feature set, adjust filtering parameters to suppress noise, and generate a first denoised signal;

[0047] S63. If the signal-to-noise ratio of the first denoised signal is lower than a preset threshold, iteratively optimize the filter parameters and reprocess the original data to obtain a second denoised signal;

[0048] S64. Extract target echo features through the second denoised signal to generate an optimized data set;

[0049] S65. Use a classification algorithm to perform pattern recognition on the optimized data set to determine the bird species.

[0050] The beneficial effects of this invention include: through the multi-channel signal processing technology of all-solid-state radar, precise separation of bird targets is achieved, effectively reducing interference caused by signal overlap, improving the signal-to-noise ratio and reliability of the initial data set, and being able to track the motion trajectories of hundreds of bird targets within the detection range in real time. Combined with target spacing and behavioral pattern analysis, it can accurately distinguish between social birds and solitary birds, and determine bird species and numbers by extracting key biological characteristics. By deeply integrating all-solid-state radar technology with intelligent data processing processes, this invention fundamentally overcomes the inherent limitations of relying on traditional optical equipment and manual monitoring, establishing an all-weather, high-precision dynamic monitoring system for ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The present invention is a flow chart of a method for identifying bird species based on all-solid-state radar. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. It should be noted that the description of these embodiments is intended to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0053] By identifying bird species, we can accurately monitor the ecology of wetland and forest ecosystems. Different bird species have varying sensitivities to environmental changes. For example, egrets in wetlands have strict water quality requirements, while woodpeckers in forests rely on healthy trees for food and nesting. If their habitat deteriorates, such as when wetland water is polluted or forest trees are affected by pests and diseases, the species or numbers of the corresponding birds will change. Regular bird species identification can help us detect these changes, promptly identify potential ecosystem problems, and seize opportunities to maintain ecological integrity.

[0054] Identifying bird species can provide guidance for habitat management. Understanding the nesting, foraging, and breeding habits of different bird species can help optimize wetland and forest environments. In wetlands, creating a variety of shallow and deep water areas and reed beds for waterfowl can help. In forests, preserving dead wood provides habitats for woodpeckers and other birds. Appropriate thinning of trees ensures sunlight access to understory vegetation, promoting ecological balance, and enhancing ecological integrity can also be helpful. Following restoration projects in damaged wetlands and forests, observing the recovery of target bird species and populations can help determine the effectiveness of restoration measures.

[0055] Reference Figure 1 As shown, the present invention provides a method for identifying bird species based on an all-solid-state radar, the method comprising the following steps:

[0056] S1. Obtain the original echo signal through the all-solid-state radar, and use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain the initial target data set;

[0057] S2. For the initial target data set, a motion trajectory prediction algorithm based on Kalman filtering is used to generate a continuous motion trajectory for each target by combining the target speed and direction information, thereby determining the target motion trajectory set.

[0058] S3. Extract target spacing features from the target trajectory set. If the target spacing is less than a preset threshold and the trajectory synchronization is higher than a set standard, the target is judged to be a gregarious bird. Otherwise, the target is judged to be a solitary bird, and the classification results of gregarious birds and solitary birds are obtained;

[0059] S4. Combining the intensity and frequency characteristics of the all-solid-state radar echo signal, a convolutional neural network model is used to extract the biological characteristics of the wingspan length and body length of all solitary birds and several social birds, thereby obtaining a target biological characteristic set for each bird.

[0060] S5. Match the target biological feature set of each bird with a pre-established bird feature database. If the similarity between the biological feature and a bird feature in the database is higher than a preset threshold, the bird species is determined to be that bird, and a bird species identification result is obtained.

[0061] The beneficial effects of the above-described embodiments are: through the multi-channel signal processing technology of the all-solid-state radar, precise separation of bird targets is achieved, interference caused by signal overlap is effectively reduced, and the signal-to-noise ratio and reliability of the initial data set are improved. Traditional optical equipment relies on visible light or infrared spectra and is susceptible to environmental interference such as haze, rain, snow, and insufficient nightlight, resulting in blind spots or errors in monitoring data. However, all-solid-state radar, based on electromagnetic wave detection, has the characteristics of penetrating clouds and fog and resisting interference from rain and snow, and can obtain echo signals of bird targets around the clock.

[0062] The system tracks the movements of birds within its detection range in real time, analyzes target spacing and behavioral patterns, and accurately distinguishes between solitary and gregarious birds. It also extracts key biological characteristics to determine bird species and abundance. By integrating all-solid-state radar technology with intelligent data processing, this system fundamentally overcomes the inherent limitations of traditional optical equipment and manual monitoring, establishing a 24 / 7, high-precision dynamic monitoring system for ecological protection.

[0063] As an embodiment, step S1 specifically includes the following sub-steps:

[0064] S11, dividing the detection range of the all-solid-state radar into grids, and calculating the spatial distribution density and signal overlap of the original echo signal of each grid;

[0065] S12. When the spatial distribution density or signal overlap of the original echo signals of a certain grid is greater than a set threshold, it is preliminarily determined to be a "suspected gregarious bird flock" area;

[0066] S13. Obtain the number of original echo signals within the "suspected flock of gregarious birds" area, allocate a number of channels in proportion to the number, and then use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain an initial target data set.

[0067] Specifically, the radar detection range is divided into square grids, and the number of echo signals per unit area within each grid is calculated. The time-frequency overlap ratio of different echo signals within the same grid is also calculated. For example, if a grid detects at least 200 signals and the signal overlap threshold exceeds 30%, it is marked as a "suspected flock" area.

[0068] For example, in a shallow wetland area, bean geese flock and grey herons solitary. If 200 bean geese gather in a grid, forming a tight formation when flying, the number of signals within a single grid will reach 200, with an overlap of 45%. For grey herons or other solitary birds scattered across other grids, the number of signals per grid will be ≤30, with an overlap of <15%. Each grid containing a solitary bird is assigned one channel, while grids containing social birds can be assigned one channel for every 50 signals. This allows each grey heron to be assigned an independent channel, allowing for continuous tracking of its foraging paths and dwelling times. This is particularly useful when identifying endangered species such as the yellow-billed egret and continuously tracking them, providing precise data for studying individual behavior. By isolating 200 bean geese in six channels, parameters such as flock density and flight synchronization can be calculated, assessing the carrying capacity of wetland habitats for migratory bird flocks and guiding ecological restoration efforts (e.g., increasing the shallows to accommodate more flocks).

[0069] Through steps S11 to S13, using dual thresholds of density (≥200 signals / grid) and overlap (>30%), only approximately 10% of grids (62-75) are identified as "suspected flocks." The remaining 90% of low-density grids (solo birds or targetless areas) do not require complex multi-channel processing. This approach is particularly suitable for large-scale, long-term bird monitoring scenarios. While maintaining identification accuracy, it reduces system energy consumption and computing power requirements to one-fifth of traditional solutions, reduces the number of subsequent Kalman filter iterations by 60%, and eliminates the need to process a large number of low-value trajectory points (such as trees swaying in the wind).

[0070] As an embodiment, step S2 specifically includes the following sub-steps:

[0071] S21, obtaining speed and direction information of the target from the initial target data set to obtain a first feature set;

[0072] S22. Process the first feature set using a Kalman filter algorithm, fuse speed and direction information, generate an initial motion trajectory for each target, and obtain a first trajectory set;

[0073] S23, if the distance between any track points in the first track set exceeds a preset threshold, interpolation is used to supplement the intermediate points to obtain a second track set;

[0074] S24. Analyze the continuity of the trajectories using a sliding window method based on the second trajectory set, determine the dynamic update frequency of the trajectories, and obtain an update parameter set;

[0075] S25, adjusting the state transfer matrix of the Kalman filter by updating the parameter set to generate a dynamically updated motion trajectory to obtain a third trajectory set;

[0076] S26. For the third trajectory set, calculate the trajectory smoothness of each target, determine whether the trajectory meets the continuity requirement, and obtain the final trajectory set;

[0077] S27. If there are abnormal trajectories in the final trajectory set, correction is performed by re-integrating the first feature set to obtain the target motion trajectory set.

[0078] In the above embodiment, step S21 extracts the target's velocity and direction information (e.g., velocity vector and acceleration) from the initial data set to construct a dynamic motion model of the target. Combined with the Kalman filter algorithm in step S22, the system is able to fuse multi-dimensional information (position, velocity, and direction), effectively suppressing measurement noise and predicting the target's continuous motion trajectory. For example, when the target accelerates or turns, the Kalman filter, through iterative correction of the state equation and observations, controls the trajectory prediction error to within 2%, significantly improving the smoothness and reliability of the trajectory.

[0079] Step S23 uses interpolation methods (such as linear or spline interpolation) to supplement missing trajectory points, resolving track fragmentation issues caused by signal loss or uneven sampling intervals. For example, when radar echoes are obscured within a certain time period, the interpolation algorithm generates intermediate points based on the motion trends of preceding and succeeding trajectory points, ensuring the spatiotemporal continuity of the trajectory data. This method increases track coverage from 85% with traditional algorithms to 98%, making it particularly suitable for continuous tracking of high-speed moving targets.

[0080] Step S24 uses a sliding window technique to analyze the local continuity of the trajectory (e.g., velocity change rate, acceleration variance) and dynamically adjust the trajectory update frequency. For example, for birds that rapidly turn (such as bee-eaters), the system can automatically increase the update frequency from 1Hz to 5Hz to avoid trajectory prediction lag. Step S25 further optimizes the model's adaptability to complex motion patterns (e.g., speed changes and circling) by updating the Kalman filter's state transition matrix, reducing trajectory fitting errors by 30% to 50%.

[0081] Step S26 identifies anomalous trajectory points (such as sudden jumps or noise interference) by calculating trajectory smoothness (e.g., the variance of the second-order derivative of acceleration). Step S27 corrects the anomalous data by reintegrating the original feature set (speed, direction). For example, if a bird briefly enters a radar blind spot, causing its trajectory to deviate, the system can backtrack the original motion parameters and use the Kalman filter's prediction-correction mechanism to restore the true path, reducing the false positive rate from 8% to below 2%.

[0082] The above steps form a closed-loop optimization process. Through dynamic parameter adjustment and data verification, the system maintains high robustness in complex environments (such as wind and rain interference and overlapping targets). For example, in high-noise scenarios, trajectory smoothness analysis and anomaly correction ensure target recognition accuracy remains above 95%. Furthermore, the combination of sliding window and interpolation technology reduces redundant calculations, reducing data processing time by 40% and significantly improving resource utilization.

[0083] Specifically, assume that in wetland ecological monitoring, the initial data of a heron target is obtained through step S1: initial position (0m, 0m), initial velocity 2.5m / s (due east, positive x-axis direction), flight direction angle 0° (uniform motion along the x-axis, no y-axis velocity), and acceleration 0.3m / s² (simulating the speed change when a bird flaps its wings). Initial trajectory points: (0, 0) at second 0, (2.5, 0) at second 1, (5.0, 0) at second 2, and (12.5, 0) at second 3. The distance between the first two points is 2.5 meters (within the limit), but the ideal position at second 3 is 7.5 meters. If the actual measurement point is (10, 0) (due to noise, a jump occurs), the distance is 5 meters (just equal to the threshold), and no interpolation is required. If the measurement point is (15, 0) (a distance of 10 meters > 5 meters), linear interpolation is used to supplement the intermediate points. If the time interval of a trajectory segment fluctuates (such as intervals of 0.5 seconds and 1.5 seconds), the update frequency is adjusted based on the standard deviation of the time interval within the window. If the acceleration variance of a trajectory segment is > 0.5 m / s² (such as sudden crosswind causing speed fluctuations), it is marked as "discontinuous" and the acceleration jump points are filtered out to retain the smooth trajectory. Assume that the trajectory point at second 4 is misidentified as (20, 0) due to radar clutter (actually 10 meters). The distance of 10 meters is > the threshold of 5 meters, triggering a correction.

[0084] As an embodiment, step S3 specifically includes the following sub-steps:

[0085] S31, obtaining target motion trajectories from the target motion trajectory set, separating each target trajectory using trajectory segmentation technology, and obtaining a single target trajectory set;

[0086] S32. For a single target trajectory set, calculate the spatial distance between each target, calculate the target spacing feature using the Euclidean distance formula, and obtain a target spacing feature set;

[0087] S33, extracting a distance value from the target distance feature set, and if the distance value is less than a preset threshold, marking it as a potential gregarious bird target, thereby obtaining a potential gregarious bird target set;

[0088] S34. For the potential gregarious bird target set, the Pearson correlation coefficient is used to calculate the synchronization between trajectories and obtain the synchronization analysis results;

[0089] S35. Extracting a synchronization value from the synchronization analysis result. If the synchronization value is higher than a preset standard, the target is determined to be a gregarious bird; otherwise, the target is determined to be a solitary bird.

[0090] In the above embodiment, distance screening is used to exclude distant targets, thereby avoiding misjudging solitary birds as flocks. Then, synchronization analysis is used to exclude "accidentally close but unsynchronized" targets (such as two solitary birds that briefly approach each other with a distance of less than 3 meters but a synchronization value of less than 0.8). However, "loosely gregarious birds" (such as geese that are not tightly packed in formation at the beginning of migration with a distance of 2-4 meters) can still be accurately identified through synchronization analysis (such as flight direction and speed consistency), thus making up for the shortcomings of single distance judgment.

[0091] The synchronicity value is calculated using the following formula:

[0092] ;

[0093] In the above formula, Indicates the points Deviation of coordinates from the mean (dispersion); Indicates the points The deviation (dispersion) of the coordinate from the mean. By calculating the sum of the products of the deviations of the variables of the two targets, it reflects the trend of coordinated change.

[0094] Specifically, track segmentation technology is used to separate overlapping radar echoes into individual target tracks, effectively addressing the issue of signal aliasing for flocking birds. Target separation is calculated using the Euclidean distance formula, transforming the abstract concept of "gregariousness" into a quantifiable physical metric. For a mixed flock of bean geese (approximately 1 meter in length) and herons (approximately 1.8 meter in wingspan), a 3-meter separation threshold is set to quickly exclude solitary birds at greater physical distances. This separation threshold filtering reduces the number of targets subject to track synchronization analysis (S34) from 625 (full range) to 62 (in suspected flocking areas), reducing the computational load by 90% and shortening the single-frame processing time from 500ms to 50ms, meeting real-time monitoring requirements.

[0095] As an embodiment, step S4 specifically includes the following sub-steps:

[0096] S41, obtaining an initial signal feature set from the echo intensity and frequency of the all-solid-state radar through signal processing;

[0097] S42, performing frequency domain analysis on the radar echo intensity and frequency using fast Fourier transform to obtain a frequency domain feature set;

[0098] S43. Based on the frequency domain feature set and the behavioral pattern set of social birds and solitary birds, a convolutional neural network is used to extract wingspan length and body length features to generate a preliminary biological feature set;

[0099] S44. If the feature value of the preliminary biological feature set matches the preset threshold, the wingspan length and body length features are subjected to dimensionality reduction processing through principal component analysis to obtain an optimized biological feature set.

[0100] Specifically, the above-mentioned target biological feature set includes wingspan length, body length, wing flapping frequency, foraging time, rest time, group spacing, peak activity time period, etc.

[0101] As an embodiment, step S5 specifically includes the following sub-steps:

[0102] S51, extracting feature data from the optimized biometric feature set, generating feature vectors using a standardized feature extraction algorithm, and obtaining a feature vector set;

[0103] S52, matching the feature vector set with a pre-established bird feature database, and calculating the similarity between the feature vectors using a cosine similarity algorithm to obtain a similarity value;

[0104] S53: If the similarity value exceeds the preset threshold, the corresponding bird species information is obtained from the bird feature database, the bird species is determined, a preliminary recognition result is obtained, and the process jumps to step S55;

[0105] S54: If the similarity value is lower than the preset threshold, return to step S4 and loop through steps S41 to S44;

[0106] S55. Based on the preliminary recognition results, a pre-established bird classification model is used to perform secondary classification on the feature vectors using a support vector machine algorithm to obtain an optimized recognition result;

[0107] S56. Based on the optimized recognition result, query the associated species information from the bird feature database, obtain species attribute data, and obtain species detailed information;

[0108] S57. Perform a consistency check between the species detailed information and the feature vector set. If the consistency check passes, output the final bird species identification result.

[0109] Specifically, in step S52, the feature vector set is matched with a pre-established bird feature database, and the similarity between the feature vectors is calculated using a cosine similarity algorithm to obtain a similarity value. In step S54, if the similarity value is lower than a preset threshold, the process returns to step S4 and loops through steps S41 to S44, specifically including:

[0110] Step S54 specifically includes the following loops: reanalyzing the radar echo intensity and frequency through signal processing to obtain a supplementary signal feature set; performing frequency domain analysis on the radar echo intensity and frequency using fast Fourier transform to obtain a frequency domain feature set; extracting wingspan length and body length features using a convolutional neural network based on the frequency domain feature set and the behavioral pattern set of gregarious birds and solitary birds to generate a preliminary biological feature set; if the feature value of the preliminary biological feature set matches a preset threshold, performing dimensionality reduction processing on the wingspan length and body length features through principal component analysis to obtain an optimized biological feature set; extracting feature data from the optimized biological feature set, generating feature vectors using a standardized feature extraction algorithm to obtain a feature vector set; and comparing the feature vector set with pre-established bird feature data. The database is matched, and the cosine similarity algorithm is used to calculate the similarity between the feature vectors to obtain a similarity value; until the similarity value exceeds a preset threshold, then step S53 is entered to obtain the corresponding bird species information from the bird feature database, determine the bird species, and obtain a preliminary recognition result; based on the preliminary recognition result, a pre-established bird classification model is used to perform secondary classification on the feature vector through the support vector machine algorithm to obtain an optimized recognition result; based on the optimized recognition result, the associated species information is queried from the bird feature database to obtain species attribute data and obtain species detailed information; a consistency check is performed between the species detailed information and the feature vector set, and if the consistency check passes, the final bird species recognition result is output.

[0111] The number of cycles in S54 can be set. When the similarity value is still lower than the preset threshold after a number of cycles, the selected target bird is calibrated and determined to be unknown.

[0112] In substeps S4 and S5, a standardized feature extraction algorithm is used to generate feature vectors from the target biological feature set, resulting in a feature vector set. This feature vector set is then matched against a pre-established bird feature database, and the cosine similarity algorithm is used to calculate the similarity between the feature vectors, ultimately identifying the bird species. The all-solid-state radar of this invention utilizes a multi-dimensional analysis of biomorphic morphology, behavioral patterns, and time series, combined with an AI algorithm, to efficiently and accurately identify bird species.

[0113] Steps S52 and S53 allow for a rapid initial screening of candidate species, addressing the inefficiency of traversing massive amounts of data. Then, in S55, a support vector machine algorithm is used to perform secondary classification of the feature vectors. This resolves the fuzzy boundaries between similar species and compensates for the linear limitations of cosine similarity, improving recognition accuracy for similar species such as herons and great egrets. Species attribute data, including habitat type and activity time, is collected in S56 and S57.

[0114] For example, by inputting the intensity and frequency characteristics of a radar echo signal, the model can output the wingspan and body length of a target organism. Specifically, the convolutional layer uses a 3x3 convolution kernel with a stride of 1 and a Reluctant Unit (ReLU) activation function. The pooling layer uses a 2x2 max pooling, and the fully connected layer outputs the target organism's feature set. This method effectively extracts the target organism's feature set from the radar echo signal, providing data support for subsequent biometric identification and behavioral analysis.

[0115] As an embodiment, the method further includes the following steps:

[0116] S6. For interference signals in complex environments, extract environmental noise characteristics from all-solid-state radar data processing, use adaptive filtering algorithms to denoise the original echo signals, generate an optimized target data set, and update the bird species recognition results.

[0117] As an embodiment, step S6 specifically includes the following sub-steps:

[0118] S61. Obtaining raw data from the echo signal of the all-solid-state radar, extracting time domain and frequency domain features using signal decomposition technology, and obtaining an environmental noise feature set;

[0119] S62: construct an adaptive filtering model based on the environmental noise feature set, adjust filtering parameters to suppress noise, and generate a first denoised signal;

[0120] S63. If the signal-to-noise ratio of the first denoised signal is lower than a preset threshold, iteratively optimize the filter parameters and reprocess the original data to obtain a second denoised signal;

[0121] S64. Extract target echo features through the second denoised signal to generate an optimized data set;

[0122] S65. Use a classification algorithm to perform pattern recognition on the optimized data set to determine the bird species.

[0123] Specifically, in complex environments, the echo signals received by radar systems are often affected by various interfering signals, such as ground clutter and weather clutter. To extract the characteristics of environmental noise, the original echo signal is first subjected to spectral analysis. Fast Fourier Transform (FFT) is used to convert the time domain signal into a frequency domain signal, and the noise frequency band is identified.

[0124] For example, frequency domain analysis revealed that noise was primarily concentrated in the 0-500Hz and 2000-2500Hz ranges. Based on this, an adaptive filtering algorithm, such as the least mean square error (LMS) algorithm, was used to denoise the signal. The LMS algorithm used a step size of 0.01 and 1000 iterations, continuously adjusting the filter coefficients to minimize the mean square error between the output signal and the desired signal. After filtering, the signal-to-noise ratio (SNR) increased from the original 10dB to 25dB, significantly improving signal quality. Next, feature extraction was performed on the optimized target dataset. Principal component analysis (PCA) was used to reduce the high-dimensional data to three dimensions, preserving 95% of the original information. Based on the reduced data, a support vector machine (SVM) classifier was used to identify bird species and abundance. The SVM kernel function used was the radial basis function (RBF), with a penalty parameter C set to 1.0 and a kernel parameter γ set to 0.1. After training and testing, the recognition accuracy reached 92%, a significant improvement from the 85% before denoising. Finally, the identification results of bird species and numbers are updated to provide reliable data support for subsequent ecological monitoring and environmental protection.

[0125] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0126] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, features specified as "first" or "second" may explicitly or implicitly include one or more of the features.

[0127] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for identifying bird species based on all-solid-state radar, characterized in that: The method comprises the following steps: S1. Obtain the original echo signal through the all-solid-state radar, and use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain the initial target data set; S2. For the initial target data set, a motion trajectory prediction algorithm based on Kalman filtering is used to generate a continuous motion trajectory for each target by combining the target speed and direction information, thereby determining the target motion trajectory set. S3. Extract target spacing features from the target trajectory set. If the target spacing is less than a preset threshold and the trajectory synchronization is higher than a set standard, the target is judged to be a gregarious bird. Otherwise, the target is judged to be a solitary bird, and the classification results of gregarious birds and solitary birds are obtained; The step S3 specifically includes the following sub-steps: S31, obtaining target motion trajectories from the target motion trajectory set, separating each target trajectory using trajectory segmentation technology, and obtaining a single target trajectory set; S32. For a single target trajectory set, calculate the spatial distance between each target, calculate the target spacing feature using the Euclidean distance formula, and obtain a target spacing feature set; S33, extracting a distance value from the target distance feature set, and if the distance value is less than a preset threshold, marking it as a potential gregarious bird target, thereby obtaining a potential gregarious bird target set; S34. For the potential gregarious bird target set, the Pearson correlation coefficient is used to calculate the synchronization between trajectories and obtain the synchronization analysis results; S35. Extracting a synchronization value from the synchronization analysis results. If the synchronization value is higher than a preset standard, the target is determined to be a gregarious bird; otherwise, the target is determined to be a solitary bird. S4. Combining the intensity and frequency characteristics of the all-solid-state radar echo signal, a convolutional neural network model is used to extract the biological characteristics of the wingspan length and body length of all solitary birds and several social birds, thereby obtaining a target biological characteristic set for each bird. S5. Match the target biological feature set of each bird with a pre-established bird feature database. If the similarity between the biological feature and a bird feature in the database is higher than a preset threshold, the bird species is determined to be that bird, and a bird species identification result is obtained.

2. The method for bird species identification based on all-solid-state radar according to claim 1, characterized in that: The step S1 specifically includes the following sub-steps: S11, dividing the detection range of the all-solid-state radar into grids, and calculating the spatial distribution density and signal overlap of the original echo signal of each grid; S12. When the spatial distribution density or signal overlap of the original echo signals of a certain grid is greater than a set threshold, it is preliminarily determined to be a "suspected gregarious bird flock" area; S13. Obtain the number of original echo signals within the "suspected flock of gregarious birds" area, allocate a number of channels in proportion to the number, and then use multi-channel signal processing technology to separate the echo data of bird targets within the detection range to obtain an initial target data set.

3. The method for bird species identification based on all-solid-state radar according to claim 1, characterized in that: The step S2 specifically includes the following sub-steps: S21, obtaining speed and direction information of the target from the initial target data set to obtain a first feature set; S22. Process the first feature set using a Kalman filter algorithm, fuse speed and direction information, generate an initial motion trajectory for each target, and obtain a first trajectory set; S23, if the distance between any track points in the first track set exceeds a preset threshold, interpolation is used to supplement the intermediate points to obtain a second track set; S24. Analyze the continuity of the trajectories using a sliding window method based on the second trajectory set, determine the dynamic update frequency of the trajectories, and obtain an update parameter set; S25, adjusting the state transfer matrix of the Kalman filter by updating the parameter set to generate a dynamically updated motion trajectory to obtain a third trajectory set; S26. For the third trajectory set, calculate the trajectory smoothness of each target, determine whether the trajectory meets the continuity requirement, and obtain the final trajectory set; S27. If there are abnormal trajectories in the final trajectory set, correction is performed by re-integrating the first feature set to obtain the target motion trajectory set.

4. The method for bird species identification based on all-solid-state radar according to claim 1, characterized in that: The step S4 specifically includes the following sub-steps: S41, obtaining an initial signal feature set from the echo intensity and frequency of the all-solid-state radar through signal processing; S42, performing frequency domain analysis on the radar echo intensity and frequency using fast Fourier transform to obtain a frequency domain feature set; S43. Based on the frequency domain feature set and the behavioral pattern set of social birds and solitary birds, a convolutional neural network is used to extract wingspan length and body length features to generate a preliminary biological feature set; S44. If the feature value of the preliminary biological feature set matches the preset threshold, the wingspan length and body length features are subjected to dimensionality reduction processing through principal component analysis to obtain an optimized biological feature set.

5. The method for bird species identification based on all-solid-state radar according to claim 4, characterized in that: The step S5 specifically includes the following sub-steps: S51, extracting feature data from the optimized biometric feature set, generating feature vectors using a standardized feature extraction algorithm, and obtaining a feature vector set; S52, matching the feature vector set with a pre-established bird feature database, and calculating the similarity between the feature vectors using a cosine similarity algorithm to obtain a similarity value; S53: If the similarity value exceeds the preset threshold, the corresponding bird species information is obtained from the bird feature database, the bird species is determined, a preliminary recognition result is obtained, and the process jumps to step S55; S54: If the similarity value is lower than the preset threshold, return to step S4 and loop through steps S41 to S44; S55. Based on the preliminary recognition results, a pre-established bird classification model is used to perform secondary classification on the feature vectors using a support vector machine algorithm to obtain an optimized recognition result; S56. Based on the optimized recognition result, query the associated species information from the bird feature database, obtain species attribute data, and obtain species detailed information; S57. Perform a consistency check between the species detailed information and the feature vector set. If the consistency check passes, output the final bird species identification result.

6. The method for bird species identification based on all-solid-state radar according to claim 1, characterized in that: The method further comprises the steps of: S6. For interference signals in complex environments, extract environmental noise characteristics from all-solid-state radar data processing, use adaptive filtering algorithms to denoise the original echo signals, generate an optimized target data set, and update the bird species recognition results.

7. The method for bird species identification based on all-solid-state radar according to claim 6, characterized in that: The step S6 specifically includes the following sub-steps: S61. Obtaining raw data from the echo signal of the all-solid-state radar, extracting time domain and frequency domain features using signal decomposition technology, and obtaining an environmental noise feature set; S62: construct an adaptive filtering model based on the environmental noise feature set, adjust filtering parameters to suppress noise, and generate a first denoised signal; S63. If the signal-to-noise ratio of the first denoised signal is lower than a preset threshold, iteratively optimize the filter parameters and reprocess the original data to obtain a second denoised signal; S64. Extract target echo features through the second denoised signal to generate an optimized data set; S65. Use a classification algorithm to perform pattern recognition on the optimized data set to determine the bird species.

Images