A mangrove wetland bird intelligent identification system based on voiceprint

Abstract

The application discloses a mangrove wetland bird intelligent identification system based on a voiceprint, comprising: an original signal acquisition module, including a plurality of signal acquisition units and a tidal synchronous acquisition unit; an effective signal generation module, which extracts and integrates effective bird chirping signals of the mangrove wetland; a data set generation module, which constructs a bird chirping signal data set in a global range of the mangrove wetland; a fusion processing module, which obtains fused audio features; a voiceprint analysis module, which is used for generating a voiceprint energy spectrum according to the fused audio features; according to the voiceprint energy spectrum, combined with time difference of arrival of the same bird chirping signal captured by the plurality of signal acquisition units, three-dimensional sound source positioning and quantity deduplication are performed, and bird population information, bird quantity information and sound source coordinate information in the global range of the mangrove wetland are determined; and a state recognition module, which is used for generating a bird state representation map in the global range of the mangrove wetland according to the bird population information, the bird quantity information and the sound source coordinate information.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring technology, and in particular to an intelligent identification system for birds in mangrove wetlands based on voiceprints. Background Technology

[0002] The continuous expansion of human activity has put pressure on the habitats and activity spaces of birds and other wild animals, resulting in significant spatial overlap between human activities and bird survival, and disrupting normal bird activities. Birds, as a crucial link in the food chain, directly impact ecological cycles through their diversity. Mangrove wetlands are important foraging and breeding grounds for birds and are vital sites for monitoring and assessing bird biodiversity. Current technologies primarily utilize infrared imaging and voiceprint recognition for on-site monitoring and analysis of mangrove wetlands. However, considering the significant tidal fluctuations and dense tree density in mangrove wetlands, infrared imaging cannot comprehensively and accurately identify the presence and activity of birds. Furthermore, while existing voiceprint recognition methods can monitor bird types and behaviors around the clock, current methods mostly rely on single-feature processing, making them susceptible to the influence of the sound signal itself, reducing the accuracy of bird morphological identification, exhibiting low anti-interference performance, and failing to obtain reliable bird identification results across the entire mangrove wetland area. Summary of the Invention

[0003] Given that existing voiceprint recognition methods are easily affected by the components of the sound signal itself, they cannot guarantee the accuracy of bird morphology identification and have low anti-interference performance, and cannot obtain reliable bird identification results for the entire mangrove wetland.

[0004] In view of the above problems, the present invention provides a voiceprint-based intelligent identification system for birds in mangrove wetlands, comprising: The original signal acquisition module includes multiple signal acquisition units and a tidal synchronous acquisition unit. The multiple signal acquisition units are used to acquire acoustic signals in two types of vertical regions within the mangrove wetland. The tidal synchronous acquisition unit acquires tidal water level data of the corresponding geographic grid and obtains acoustic data. The acoustic data is an acoustic signal with time domain, spatial domain and tidal water level data. The effective signal generation module is used to compare the acoustic signals of two types of vertical region sets, combine them with tidal water level data for adaptive background noise suppression, and extract and integrate the effective bird song signals of the mangrove wetland. The dataset generation module is used to preprocess valid bird song signals within a preset time range and construct a global bird song signal dataset for mangrove wetlands. The fusion processing module is used to extract multidimensional features from the bird song signal dataset to obtain multidimensional audio features, and to perform dynamic weight fusion on the multidimensional audio features in the current time window to obtain fused audio features. The voiceprint analysis module is used to generate a voiceprint energy spectrum based on the fused audio features; based on the voiceprint energy spectrum, combined with the arrival time difference of the same bird call signal captured by the multiple signal acquisition units, three-dimensional sound source localization and number deduplication are performed to determine the bird population information, bird quantity information and sound source coordinate information of the entire mangrove wetland. The status recognition module is used to generate a bird status representation map of the entire mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information.

[0005] Preferably, the original signal acquisition module includes a first signal acquisition group, a second signal acquisition group, and a tidal synchronization acquisition unit; The first signal acquisition group includes multiple first signal acquisition units distributed in the first airspace within the mangrove wetland. The multiple first signal acquisition units are placed at different locations within the first airspace and can acquire acoustic signals emitted from any location within the first airspace to obtain the first acoustic signal. The second signal acquisition group includes multiple second signal acquisition units distributed within the second airspace of the mangrove wetland. These multiple second signal acquisition units are placed at different locations within the second airspace and can acquire acoustic signals emitted from any location within the second airspace to obtain second acoustic signals. Multiple first signal acquisition units and multiple second signal acquisition units constitute the multiple signal acquisition units, and the first acoustic signal and the second acoustic signal constitute the acoustic signal; The tidal synchronization acquisition unit includes several water level sensors distributed according to a geographic grid. Each water level sensor covers a preset geographic grid area and is used to collect real-time tidal water level data within that geographic grid. All first signal acquisition units and second signal acquisition units within the same geographic grid share the tidal water level data of that grid. ; The acoustic data is obtained by time-domain labeling the first acoustic signal and the second acoustic signal. airspace tags and corresponding geographic grid tidal water level data The joint calibration was obtained.

[0006] Preferably, the effective signal generation module is used to compare the acoustic signals of the two types of region sets, extract and integrate them to obtain the effective bird song signal of the mangrove wetland, including: Extract the first background sound components from the first acoustic signal and the second acoustic signal respectively. Second background sound components ; Comparison of the first background sound components Second background sound components The sound characteristics were used to obtain the first background sound component. Second background sound components The similarity relationship S; According to tidal water level data Based on the type of the vertical region where the current signal acquisition unit is located, the adaptive background noise threshold for the corresponding vertical region at the current moment is calculated using the following formula. : (1), in, For airspace type identification, Represents the first airspace range. Represents the second airspace range; For the first Reference background noise threshold for the spatial domain; For the first Tidal influence coefficient of quasi-space, This is the average tidal level; like If so, it is determined that the two acoustic signals come from the same environmental area, and there is a possibility of overlapping bird calls; Based on the similarity coefficient The acoustic data is used to perform temporal alignment and spatial matching on overlapping bird song signals, and then extracted and integrated to obtain effective bird song signals from mangrove wetlands without repetition.

[0007] Preferably, the dataset generation module is used to preprocess valid bird song signals within a preset time range to construct a global bird song signal dataset for the mangrove wetland, including: Distinguish the valid bird calls within a preset time range to obtain several valid bird call signal segments that correspond one-to-one with all intervals within the preset time range; Multimodal preprocessing is performed on valid bird song signal segments to construct a global bird song signal dataset for mangrove wetlands. The multimodal preprocessing includes framing, windowing, and Fourier transform. The bird song signal dataset includes several frequency domain segments of bird song signals that correspond one-to-one with several valid bird song signal segments.

[0008] Preferably, the fusion processing module is used to extract multidimensional features from the bird song signal dataset to obtain multidimensional audio features; and to perform dynamic weight fusion on the multidimensional audio features of the current time window to obtain fused audio features, including: Linear prediction analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the first audio feature. Among them, the first audio feature is the audio feature related to syllables; Power correlation analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the second audio features. The second audio feature is an audio feature relating to the human auditory response. Receive the environmental confidence factor of the previous time window returned by the voiceprint analysis module. The first audio feature of the current time window is calculated according to the following formula. Second audio features Dynamic fusion weights: (2), (3), in, The characteristic robustness coefficient; Using neural networks to analyze the first audio features of the current time window Second audio features After mean and normalization processing, according to dynamic weights and Perform weighted fusion to obtain the fused audio features for the current time window. : (4).

[0009] Preferably, the voiceprint analysis module is used to generate a voiceprint energy spectrum based on the fused audio features; and to determine the bird population information and bird quantity information of the entire mangrove wetland based on the voiceprint energy spectrum, including steps S1-S4: Step S1: Perform dimensional reconstruction, energy calculation, and time-frequency spectrum mapping on the fused audio features to generate an energy spectrogram; Step S2: Process the obtained energy spectrogram using an attention residual network to generate a voiceprint energy spectrogram; Step S3: Extract the spectral segment containing a single bird call sound from the soundprint energy spectrum map, and determine the bird population information based on the matching results of the spectral segment and the preset bird soundprint database. Step S4: Based on the arrival time difference of the same bird call signal captured by multiple first signal acquisition units or second signal acquisition units, perform three-dimensional sound source localization to obtain the sound source coordinates; Step S5: Based on the spatial distance and temporal overlap of the sound source coordinates, deduplicate the bird count to obtain the final bird count information.

[0010] Preferably, step S2 can be implemented as follows: S21. According to a preset time window, the fused audio features are segmented into continuous fused audio feature segments; S22. Perform attention residual network processing on each of the fused audio feature segments to generate a sequence of audio energy spectrum maps corresponding to each time window. S23. For each voiceprint energy spectrum map in the sequence, a quality assessment is performed, specifically including the following steps S131-133: S231. Based on the signal-to-noise ratio (SNR), bandwidth richness (B), and temporal continuity score (C) of the current voiceprint energy spectrum, and the personalized noise floor factor (σ) extracted from the current voiceprint energy spectrum, calculate the quality score of the current voiceprint energy spectrum according to the following formula. : (5), Where SNR represents the signal-to-noise ratio of the current voiceprint energy spectrum, σ represents the personalized noise floor factor, B represents the frequency band richness, C represents the time continuity score, and λ represents the frequency band richness attenuation coefficient. Let w1 represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, satisfying w1 + w2 + w3 = 1; S232. Based on the energy distribution of the current and previous voiceprint energy spectrum maps within a preset non-birdsong frequency range, extract the environmental stability factor ρ, and calculate the environmental reliability factor of the current voiceprint energy spectrum map according to the following formula. : (6), Wherein, ρ represents the environmental stability factor of the current voiceprint energy spectrum map. r 0 This indicates the preset stability benchmark threshold. k Indicates the gain factor; S233, Calculate the quality score Compared with the preset quality threshold Compare and use environmental credibility factors Compared with the preset confidence threshold Compare; if and If the current voiceprint energy spectrum is deemed to be of acceptable quality, it is retained and the process proceeds to step S3; otherwise, the current voiceprint energy spectrum is deemed to be of unacceptable quality, triggering enhancement processing on the corresponding original signal segment, and returning to step S22 to regenerate the voiceprint energy spectrum; if the enhancement processing is still unacceptable, the current voiceprint energy spectrum and its corresponding original signal segment are discarded.

[0011] Preferably, step S4 specifically includes the following steps: 1) For the same acoustic signal, record its occurrence by the first... The and the first The time captured by each signal acquisition unit and Calculate the time difference ; 2) Based on the spatial coordinates of the signal acquisition unit and Establish the following hyperbolic positioning equation: (7), in, The speed of sound in air; The coordinates of the sound source to be determined; 3) Following the methods in steps 1) and 2), using three signal acquisition units yields three formulas (7). Solving these three formulas (7) gives the three-dimensional coordinates of the sound source. .

[0012] Preferably, step S5 specifically includes: 1) For all identified sound sources, calculate the spatial distance between any two sound source coordinates. : (8), Where m and n are the m-th and n-th sound sources, respectively. Let be the coordinates of the m-th sound source. Let n be the coordinates of the nth sound source; 2) Calculate the temporal overlap between the two sound sources. (9), in, The time overlap length between the two sound source signals. The total duration of the two sound source signals; 3) Calculate the coefficient of determination for duplication using the following formula. : (10) in, The maximum activity radius of the same bird species. This is the spatial weighting coefficient, with a value ranging from 0.6 to 0.8; 4) If If the two bird calls are from the same bird, they are determined to be from different individuals and deduplication is performed; otherwise, they are determined to be from different individuals. This is the preset repetition threshold.

[0013] Preferably, the state recognition module is used to generate a bird state representation map of the entire mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information, including: 1) Spatial and temporal changes in bird population information, bird quantity information, and sound source coordinate information are calibrated to generate several spatial distribution maps of birds that correspond one-to-one with several time points; 2) Arrange and annotate several bird spatial distribution maps in a time sequence to generate a bird status representation map of the entire mangrove wetland. The information annotation includes at least bird population information, bird quantity information, sound source coordinate information, and tidal water level information.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention sets up multiple signal acquisition units at different locations in the mangrove forest. The acoustic patterns are detected by the multiple signal acquisition units, and the species and number of birds can be automatically identified based on the detection results. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the structure of a voiceprint-based intelligent bird identification system for mangrove wetlands provided in an embodiment of the present invention. Detailed Implementation

[0016] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein.

[0017] A voiceprint-based intelligent bird identification system for mangrove wetlands includes: The original signal acquisition module includes multiple signal acquisition units and a tidal synchronous acquisition unit. The multiple signal acquisition units are used to acquire acoustic signals in two types of vertical regions within the mangrove wetland. The tidal synchronous acquisition unit acquires tidal water level data of the corresponding geographic grid and obtains acoustic data. The acoustic data is an acoustic signal with time domain, spatial domain and tidal water level data. The effective signal generation module is used to compare the acoustic signals of two types of vertical region sets, combine tidal water level data and vertical region stratification coefficients to perform adaptive background noise suppression, and extract and integrate the effective bird song signals of mangrove wetlands. The dataset generation module is used to preprocess valid bird song signals within a preset time range and construct a global bird song signal dataset for mangrove wetlands. The fusion processing module is used to extract multi-dimensional features from the bird song signal dataset to obtain multi-dimensional audio features; it receives the environmental credibility factor of the previous time window from the voiceprint analysis module, and performs dynamic weight fusion on the multi-dimensional audio features of the current time window to obtain fused audio features. The voiceprint analysis module is used to generate a voiceprint energy spectrum based on the fused audio features; based on the voiceprint energy spectrum, and combined with the arrival time difference of the same bird call signal captured by multiple signal acquisition units, it performs three-dimensional sound source localization and deduplication to determine the bird population information, bird quantity information, and sound source coordinate information of the entire mangrove wetland; at the same time, it calculates the quality score and environmental credibility factor of the voiceprint energy spectrum map of the current time window, and sends the environmental credibility factor back to the fusion processing module for feature fusion weight adjustment in the next time window; The status recognition module is used to generate a bird status representation map of the entire mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information.

[0018] The original signal acquisition module includes a first signal acquisition group, a second signal acquisition group, and a tidal synchronous acquisition unit. The first signal acquisition group comprises multiple first signal acquisition units distributed across a first airspace within the mangrove wetland, preferably a mudflat layer with a height of 0-2m. The second signal acquisition group comprises multiple second signal acquisition units distributed across a second airspace within the mangrove wetland, the canopy layer with a height of 5-15m. The multiple signal acquisition units can all originate from the first or second signal acquisition units, or they can originate from the first and second signal acquisition units respectively, with at least three acquisition units for three-dimensional sound source localization. The tidal synchronous acquisition unit includes several water level sensors distributed according to a geographic grid. Each water level sensor covers a preset geographic grid area and is used to collect real-time tidal water level data within that geographic grid. All first and second signal acquisition units within the same geographic grid share the tidal water level data of that grid. .

[0019] Multiple first signal acquisition units are placed at different locations within a first spatial domain, capable of acquiring acoustic signals emitted from any location within the first spatial domain to obtain a first acoustic signal. Multiple second signal acquisition units are placed at different locations within a second spatial domain, capable of acquiring acoustic signals emitted from any location within the second spatial domain to obtain a second acoustic signal. The first and second acoustic signals constitute the acoustic signal. The acoustic data is obtained by time-domain labeling of the first and second acoustic signals. airspace tags and corresponding geographic grid tidal water level data The joint calibration was obtained.

[0020] The effective signal generation module is used to compare the acoustic signals of two types of region sets, extract and integrate them to obtain the effective bird song signals of the mangrove wetland, including: Extract the first background sound components from the first acoustic signal and the second acoustic signal respectively. Second background sound components ; Comparison of the first background sound components Second background sound components The sound characteristics were used to obtain the first background sound component. Second background sound components The similarity relationship S; According to tidal water level data Based on the type of the vertical region where the current signal acquisition unit is located, the adaptive background noise threshold for the corresponding vertical region at the current moment is calculated using the following formula. : (1), in, For airspace type identification, Represents the first airspace range. Represents the second airspace range; For the first The baseline background noise threshold for the vertical region was measured under calm, tidal conditions. For the first The tidal influence coefficient of the quasi-space domain, preferably, The value ranges from 0.3 to 0.5. The value ranges from 0.1 to 0.2. The lower the airspace, the greater the impact of tides on noise. This is the average tidal level; like If so, it is determined that the two acoustic signals come from the same environmental area, and there is a possibility of overlapping bird calls; Based on the similarity coefficient The acoustic data is used to perform temporal alignment and spatial matching on overlapping bird song signals, and then extracted and integrated to obtain effective bird song signals from mangrove wetlands without repetition.

[0021] The dataset generation module is used to preprocess valid bird song signals within a preset time range to construct a global bird song signal dataset for mangrove wetlands, including: Distinguish the valid bird calls within a preset time range to obtain several valid bird call signal segments that correspond one-to-one with all intervals within the preset time range; Multimodal preprocessing is performed on valid bird song signal segments to construct a global bird song signal dataset for mangrove wetlands. The multimodal preprocessing includes framing, windowing, and Fourier transform. The bird song signal dataset includes several frequency domain segments of bird song signals that correspond one-to-one with several valid bird song signal segments.

[0022] The dynamic changes in bird activity in mangrove wetlands over a certain period of time result in different effective bird call signal components for each time interval within that time range. To continuously identify bird states in mangrove wetlands, effective bird call signals within a preset time range are first distinguished using time windows of a predetermined length, resulting in several effective bird call signal segments. Each effective bird call signal segment contains a relatively small amount of bird call signal data, which helps reduce the processing workload for each segment. Then, each effective bird call signal segment undergoes frame segmentation, windowing (i.e., transformation using a windowing function), and short-time Fourier transform sequentially. This multimodal preprocessing transforms the effective bird call signal segments from the time domain to the frequency domain, thus constructing a bird call signal dataset and providing a reliable data foundation for subsequent multidimensional feature extraction.

[0023] The fusion processing module is used to extract multidimensional features from the bird song signal dataset to obtain multidimensional audio features; it receives the environmental credibility factor of the previous time window from the speaker analysis module, and performs dynamic weight fusion on the multidimensional audio features of the current time window to obtain fused audio features, including: Linear prediction analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the first audio feature. Among them, the first audio feature is the audio feature related to syllables; Power correlation analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the second audio features. The second audio feature is an audio feature relating to the human auditory response. Receive the environmental confidence factor of the previous time window returned by the voiceprint analysis module. The first audio feature of the current time window is calculated according to the following formula. Second audio features Dynamic fusion weights: (2), (3), in, The characteristic robustness coefficient ranges from 0.3 to 0.7. The noise immunity is weaker than Therefore When calculating the first time window, and Calculations are performed using predetermined or empirical values; Using neural networks to analyze the first audio features of the current time window Second audio features After mean and normalization processing, according to dynamic weights and Perform weighted fusion to obtain the fused audio features for the current time window. : (4).

[0024] In practice, linear predictive cepstral calculations are performed on all frequency domain segments of the bird song signal dataset to obtain the first audio feature (i.e., the linear predictive cepstral audio feature), which accurately represents the vowels and other syllables of all frequency domain segments of the bird song signal. Power regularization calculations are then performed on all frequency domain segments of the bird song signal dataset to obtain the second audio feature (i.e., the power-regularized cepstral audio feature), which accurately represents the power state of all frequency segments of the bird song signal. A convolutional neural network is then used to perform mean processing, normalization, and concatenation and fusion of the first and second audio features to obtain the fused audio feature. It can be understood that the fused audio feature is used to represent the spectral characteristics of the entire bird song signal corresponding to different feature energies.

[0025] The voiceprint analysis module is used to generate a voiceprint energy spectrum based on fused audio features; based on the voiceprint energy spectrum, it determines bird population information and bird quantity information for the entire mangrove wetland, including steps S1-S4: Step S1: Perform dimensional reconstruction, energy calculation, and time-frequency spectrum mapping on the fused audio features to generate an energy spectrogram; Step S2: Process the obtained energy spectrogram using an attention residual network to generate a voiceprint energy spectrogram; Step S3: Extract the spectral segment containing a single bird call sound from the soundprint energy spectrum map, and determine the bird population information based on the matching results of the spectral segment and the preset bird soundprint database. Step S4: Based on the arrival time difference of the same bird call signal captured by multiple first signal acquisition units or second signal acquisition units, perform three-dimensional sound source localization to obtain the sound source coordinates; Step S5: Based on the spatial distance and temporal overlap of the sound source coordinates, deduplicate the bird count to obtain the final bird count information.

[0026] Step S2 can be specifically implemented as follows: S21. According to a preset time window, the fused audio features are segmented into continuous fused audio feature segments; S22. Perform attention residual network processing on each of the fused audio feature segments to generate a sequence of audio energy spectrum maps corresponding to each time window. S23. For each voiceprint energy spectrum map in the sequence, a quality assessment is performed, specifically including the following steps S131-133: S231. Based on the signal-to-noise ratio (SNR), bandwidth richness (B), and temporal continuity score (C) of the current voiceprint energy spectrum, and the personalized noise floor factor (σ) extracted from the current voiceprint energy spectrum, calculate the quality score of the current voiceprint energy spectrum according to the following formula. : (5), Wherein, SNR represents the signal-to-noise ratio of the current audioprint energy spectrum, and its value is a real number greater than or equal to 0. It is obtained by dividing the average energy of the current audioprint energy spectrum within a preset bird song frequency range by the average energy of the background noise frequency range; σ represents the personalized noise basis factor, and its value range is (0,1]. It is obtained by selecting a preset non-bird song frequency range in the current audioprint energy spectrum, calculating the average energy of all time frames within that range, and normalizing it to the (0,1) interval; B represents the frequency band richness, and its value range is [0,1]. It is obtained by dividing the current audioprint energy spectrum into several frequency bands. The number of frequency bands with energy exceeding a preset threshold is divided by the total number of frequency bands; C represents the time continuity score, with a value range of [0,1], which is obtained by calculating the Pearson correlation coefficient between two adjacent frames in the current voiceprint energy spectrum map and taking the average of the correlation coefficients of all adjacent frames; λ represents the frequency band richness attenuation coefficient, with a value range of (0,1], which is set based on experience or obtained by fitting calibration data; w1, w2, and w3 represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, respectively, with a value range of [0,1] and satisfying w1+w2+w3=1, which are dynamically adjusted according to the monitoring scenario or determined by expert scoring through the analytic hierarchy process; S232. Based on the energy distribution of the current and previous voiceprint energy spectrum maps within a preset non-birdsong frequency range, extract the environmental stability factor ρ, and calculate the environmental reliability factor of the current voiceprint energy spectrum map according to the following formula. : (6), Wherein, ρ represents the environmental stability factor of the current voiceprint energy spectrum map, with a value range of [0,1]. It is obtained by calculating the Pearson correlation coefficient between the energy distribution of the current voiceprint energy spectrum map and the previous voiceprint energy spectrum map within the preset non-birdsong frequency range. If the current voiceprint energy spectrum map is the first one in the sequence, the default value of 1 is used. ρ0 represents the preset stability benchmark threshold, with a value range of (0,1), which is set according to historical data statistics or expert experience. κ represents the gain factor, with a value range of (1,10), which is used to adjust the intensity of the influence of stability on credibility, and is preset according to the environmental complexity. S233, Calculate the quality score Compared with the preset quality threshold Compare and use environmental credibility factors Compared with the preset confidence threshold Compare; if and If the current voiceprint energy spectrum is deemed to be of acceptable quality, it is retained and proceeds to the next step S3; otherwise, the current voiceprint energy spectrum is deemed to be of unacceptable quality, triggering enhancement processing on the corresponding original signal segment, and returning to step S22 to regenerate the voiceprint energy spectrum; if the enhancement processing is still unacceptable, the current voiceprint energy spectrum and its corresponding original signal segment are discarded.

[0027] In the above technical solution, the quality scoring formula quantifies and evaluates the voiceprint energy spectrum from three dimensions: signal-to-noise ratio, frequency band richness, and temporal continuity. It automatically filters out clear, complete, and stable high-quality signals, eliminating low-quality signals contaminated by noise, ensuring that the data source for identification is clean and reliable. The environmental reliability factor dynamically senses environmental stability by comparing the energy distribution similarity of adjacent spectrum maps in non-birdsong frequency bands. When there are sudden environmental changes (such as tides or waves), this factor decreases, triggering signal enhancement or discarding, avoiding invalid identification during periods of severe interference. These mechanisms work together to ensure the input quality for bird identification from the source, improving the system's anti-interference capability and the reliability of identification results in complex mangrove wetland environments.

[0028] Step S4 specifically includes the following steps: 1) For the same bird call signal, record the number of times it is heard. The and the first The time captured by each signal acquisition unit and Calculate the time difference ; 2) Based on the spatial coordinates of the signal acquisition unit and Establish the following hyperbolic positioning equation: (7), in, The speed of sound in air is taken as 340 m / s, which can be corrected according to the ambient temperature. The coordinates of the sound source to be determined; 3) Following the methods in steps 1) and 2), using three signal acquisition units yields three formulas (7). Solving these three formulas (7) gives the three-dimensional coordinates of the sound source. .

[0029] Preferably, step S5 specifically includes: 1) For all identified sound sources, calculate the spatial distance between any two sound source coordinates. : (8), Where m and n are the m-th and n-th sound sources, respectively. Let be the coordinates of the m-th sound source. Let n be the coordinates of the nth sound source; 2) Calculate the temporal overlap between the two sound sources. (9), in, The time overlap length between the two sound source signals. The total duration of the two sound source signals; 4) Calculate the coefficient of determination for duplication using the following formula. : (10) in, The maximum activity radius of the same bird species. This is the spatial weighting coefficient, with a value ranging from 0.6 to 0.8; 4) If If the two bird calls are from the same bird, they are determined to be from different individuals and deduplication is performed; otherwise, they are determined to be from different individuals. The preset repetition threshold can be set manually, and preferably, the value is 0.7.

[0030] The state recognition module is used to generate a global bird state representation map of the mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information, including: 1) Spatial and temporal changes in bird population information, bird quantity information, and sound source coordinate information are calibrated to generate several spatial distribution maps of birds that correspond one-to-one with several time points; 2) Arrange and annotate several bird spatial distribution maps in a time sequence to generate a bird status representation map of the entire mangrove wetland. The information annotation includes at least bird population information, bird quantity information, sound source coordinate information, and tidal water level information.

[0031] Birds inhabit and forage in mangrove wetlands, and their activities are dynamic. To characterize the ecological activities of birds in mangrove wetlands, spatial and temporal changes in bird population and quantity information are calibrated to generate several spatial distribution maps of birds corresponding to several time points. Then, information is labeled and time-series arranged on several bird spatial distribution maps to generate a global bird status representation map of mangrove wetlands. By consulting the above bird status representation map, a real-time and reliable data foundation can be provided for subsequent bird ecological research.

[0032] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. This disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims. Thus, if these modifications and variations of the invention fall within the scope of the claims of the invention and their equivalents, the invention is also intended to include these modifications and variations.

Claims

1. A voiceprint-based intelligent bird identification system for mangrove wetlands, comprising: The original signal acquisition module includes multiple signal acquisition units and a tidal synchronous acquisition unit. The multiple signal acquisition units are used to acquire acoustic signals in two types of vertical regions within the mangrove wetland. The tidal synchronous acquisition unit acquires tidal water level data of the corresponding geographic grid and obtains acoustic data. The acoustic data is an acoustic signal with time domain, spatial domain and tidal water level data. The effective signal generation module is used to compare the acoustic signals of two types of vertical region sets, combine them with tidal water level data for adaptive background noise suppression, and extract and integrate the effective bird song signals of the mangrove wetland. The dataset generation module is used to preprocess valid bird song signals within a preset time range and construct a global bird song signal dataset for mangrove wetlands. The fusion processing module is used to extract multidimensional features from the bird song signal dataset to obtain multidimensional audio features, and to perform dynamic weight fusion on the multidimensional audio features in the current time window to obtain fused audio features. The voiceprint analysis module is used to generate a voiceprint energy spectrum based on the fused audio features; based on the voiceprint energy spectrum, combined with the arrival time difference of the same bird call signal captured by the multiple signal acquisition units, three-dimensional sound source localization and number deduplication are performed to determine the bird population information, bird quantity information and sound source coordinate information of the entire mangrove wetland. The status recognition module is used to generate a bird status representation map of the entire mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information.

2. The system according to claim 1, characterized in that, The original signal acquisition module includes a first signal acquisition group, a second signal acquisition group, and a tidal synchronization acquisition unit; The first signal acquisition group includes multiple first signal acquisition units distributed in the first airspace within the mangrove wetland. The multiple first signal acquisition units are placed at different locations within the first airspace and can acquire acoustic signals emitted from any location within the first airspace to obtain the first acoustic signal. The second signal acquisition group includes multiple second signal acquisition units distributed within the second airspace of the mangrove wetland. These multiple second signal acquisition units are placed at different locations within the second airspace and can acquire acoustic signals emitted from any location within the second airspace to obtain second acoustic signals. Multiple first signal acquisition units and multiple second signal acquisition units constitute the multiple signal acquisition units, and the first acoustic signal and the second acoustic signal constitute the acoustic signal; The tidal synchronization acquisition unit includes several water level sensors distributed according to a geographic grid. Each water level sensor covers a preset geographic grid area and is used to collect real-time tidal water level data within that geographic grid. All first signal acquisition units and second signal acquisition units within the same geographic grid share the tidal water level data of that grid. ; The acoustic data is obtained by time-domain labeling the first acoustic signal and the second acoustic signal. airspace tags and corresponding geographic grid tidal water level data The joint calibration was obtained.

3. The system according to claim 2, characterized in that, The effective signal generation module is used to compare the acoustic signals of two types of region sets, extract and integrate them to obtain the effective bird song signals of the mangrove wetland, including: Extract the first background sound components from the first acoustic signal and the second acoustic signal respectively. Second background sound components ; Comparison of the first background sound components Second background sound components The sound characteristics were used to obtain the first background sound component. Second background sound components The similarity relationship S; According to tidal water level data Based on the type of the vertical region where the current signal acquisition unit is located, the adaptive background noise threshold for the corresponding vertical region at the current moment is calculated using the following formula. : （1）， in, For airspace type identification, Represents the first airspace range. Represents the second airspace range; For the first Reference background noise threshold for the spatial domain; For the first Tidal influence coefficient of quasi-space, This is the average tidal level; like If so, it is determined that the two acoustic signals come from the same environmental area, and there is a possibility of overlapping bird calls; Based on the similarity coefficient The acoustic data is used to perform temporal alignment and spatial matching on overlapping bird song signals, and then extracted and integrated to obtain effective bird song signals from mangrove wetlands without repetition.

4. The system according to claim 3, characterized in that, The dataset generation module is used to preprocess valid bird song signals within a preset time range to construct a global bird song signal dataset for mangrove wetlands, including: Distinguish the valid bird calls within a preset time range to obtain several valid bird call signal segments that correspond one-to-one with all intervals within the preset time range; Multimodal preprocessing is performed on valid bird song signal segments to construct a global bird song signal dataset for mangrove wetlands. The multimodal preprocessing includes framing, windowing, and Fourier transform. The bird song signal dataset includes several frequency domain segments of bird song signals that correspond one-to-one with several valid bird song signal segments.

5. The system according to claim 4, characterized in that, The fusion processing module is used to extract multidimensional features from the bird song signal dataset to obtain multidimensional audio features; Dynamic weighted fusion is performed on the multidimensional audio features of the current time window to obtain fused audio features, including: Linear prediction analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the first audio feature. Among them, the first audio feature is the audio feature related to syllables; Power correlation analysis was performed on all frequency domain segments of the bird song signal dataset to obtain the second audio features. The second audio feature is an audio feature relating to the human auditory response. Receive the environmental confidence factor of the previous time window returned by the voiceprint analysis module. The first audio feature of the current time window is calculated according to the following formula. Second audio features Dynamic fusion weights: （2）， （3）， in, The characteristic robustness coefficient; Using neural networks to analyze the first audio features of the current time window Second audio features After mean and normalization processing, according to dynamic weights and Perform weighted fusion to obtain the fused audio features for the current time window. : （4）。 6. The system according to claim 5, characterized in that, The voiceprint analysis module is used to generate a voiceprint energy spectrum based on fused audio features; based on the voiceprint energy spectrum, it determines the bird population information and bird quantity information of the entire mangrove wetland, including steps S1-S5: Step S1: Perform dimensional reconstruction, energy calculation, and time-frequency spectrum mapping on the fused audio features to generate an energy spectrogram; Step S2: Process the obtained energy spectrogram using an attention residual network to generate a voiceprint energy spectrogram; Step S3: Extract the spectral segment containing a single bird call sound from the soundprint energy spectrum map, and determine the bird population information based on the matching results of the spectral segment and the preset bird soundprint database. Step S4: Based on the arrival time difference of the same bird call signal captured by multiple first signal acquisition units or second signal acquisition units, perform three-dimensional sound source localization to obtain the sound source coordinates; Step S5: Based on the spatial distance and temporal overlap of the sound source coordinates, deduplicate the bird count to obtain the final bird count information.

7. The system according to claim 6, characterized in that, Step S2 can be specifically implemented as follows: S21. According to a preset time window, the fused audio features are segmented into continuous fused audio feature segments; S22. Perform attention residual network processing on each of the fused audio feature segments to generate a sequence of audio energy spectrum maps corresponding to each time window. S23. For each voiceprint energy spectrum map in the sequence, a quality assessment is performed, specifically including the following steps S131-133: S231. Based on the signal-to-noise ratio (SNR), bandwidth richness (B), and temporal continuity score (C) of the current voiceprint energy spectrum, and the personalized noise floor factor (σ) extracted from the current voiceprint energy spectrum, calculate the quality score of the current voiceprint energy spectrum according to the following formula. : （5）， Where SNR represents the signal-to-noise ratio of the current voiceprint energy spectrum, σ represents the personalized noise floor factor, B represents the frequency band richness, C represents the time continuity score, and λ represents the frequency band richness attenuation coefficient. Let w1 represent the first weight coefficient, the second weight coefficient, and the third weight coefficient, satisfying w1 + w2 + w3 = 1; S232. Based on the energy distribution of the current and previous voiceprint energy spectrum maps within a preset non-birdsong frequency range, extract the environmental stability factor ρ, and calculate the environmental reliability factor of the current voiceprint energy spectrum map according to the following formula. : （6）， Wherein, ρ represents the environmental stability factor of the current voiceprint energy spectrum map. ρ 0 This indicates the preset stability benchmark threshold. κ Indicates the gain factor; S233, Calculate the quality score Compared with the preset quality threshold Compare and use environmental credibility factors Compared with the preset confidence threshold Compare; if and If the current voiceprint energy spectrum is deemed to be of acceptable quality, it is retained and the process proceeds to step S3; otherwise, the current voiceprint energy spectrum is deemed to be of unacceptable quality, triggering enhancement processing on the corresponding original signal segment, and returning to step S22 to regenerate the voiceprint energy spectrum; if the enhancement processing is still unacceptable, the current voiceprint energy spectrum and its corresponding original signal segment are discarded.

8. The system according to claim 7, characterized in that, Step S4 specifically includes the following steps: 1) For the same acoustic signal, record its occurrence by the first... The and the first The time captured by each signal acquisition unit and Calculate the time difference ; 2) Based on the spatial coordinates of the signal acquisition unit and Establish the following hyperbolic positioning equation: (7), in, The speed of sound in air; The coordinates of the sound source to be determined; 3) Following the methods in steps 1) and 2), using three signal acquisition units yields three formulas (7). Solving these three formulas (7) gives the three-dimensional coordinates of the sound source. .

9. The system according to claim 8, characterized in that, Step S5 specifically includes: 1) For all identified sound sources, calculate the spatial distance between any two sound source coordinates. : （8）， Where m and n are the m-th and n-th sound sources, respectively. Let be the coordinates of the m-th sound source. Let n be the coordinates of the nth sound source; 2) Calculate the temporal overlap between the two sound sources. （9）， in, The time overlap length between the two sound source signals. The total duration of the two sound source signals; 3) Calculate the coefficient of determination for duplication using the following formula. : (10) in, The maximum activity radius of the same bird species. This is the spatial weighting coefficient, with a value ranging from 0.6 to 0.8; 4) If If the two bird calls are from the same bird, they are determined to be from different individuals and deduplication is performed; otherwise, they are determined to be from different individuals. This is the preset repetition threshold.

10. The system according to claim 9, characterized in that, The state recognition module is used to generate a global bird state representation map of the mangrove wetland based on bird population information, bird quantity information, and sound source coordinate information, including: 1) Spatial and temporal changes in bird population information, bird quantity information, and sound source coordinate information are calibrated to generate several spatial distribution maps of birds that correspond one-to-one with several time points; 2) Arrange and annotate several bird spatial distribution maps in a time sequence to generate a bird status representation map of the entire mangrove wetland. The information annotation includes at least bird population information, bird quantity information, sound source coordinate information, and tidal water level information.

Images