An ai interaction system for ornithology research

Abstract

The present application relates to the technical field of artificial intelligence, and more particularly to an AI interaction system for ornithological scientific research, which comprises a collection unit, a trigger determination unit, a path determination unit, a dominant determination unit, a prompt generation unit and an adjustment unit. The present application constructs a multi-level behavior determination process through joint analysis of bird song spectrum, image motion characteristics, flight trajectory and environmental factors, and through statistical analysis of the time distribution and spatial distribution of these behavior indicators within a preset period, and correlation analysis with environmental factors such as temperature, humidity, wind speed and illumination, the influence of environmental changes on bird activity frequency and spatial aggregation degree can be identified, so that the acoustic trigger threshold can be dynamically adjusted, so that the system can continuously maintain the sensitive recognition ability of bird activity with the change of the habitat environment, and then realize high-reliable behavior recognition and interaction prompt for ornithological scientific research.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an AI interactive system for ornithological research. Background Technology

[0002] With the intensification of global climate change and the increasing urgency of biodiversity conservation, birds, as important indicator species of ecological environment health, have become key observational subjects in ecological research due to their migration patterns, habitat selection, and behavioral response mechanisms. However, traditional ornithological research has long relied on manual field observation and specimen collection, which is not only limited by observation time and manpower coverage, but also makes it difficult to simultaneously collect and correlate multimodal data on birds' calls, movement trajectories, group distribution, and environmental response behaviors in their natural state. Although some automated monitoring equipment has emerged, such as infrared cameras and voiceprint recognition devices, these devices are mostly single-modal data recording tools, only providing raw data collection functions and lacking intelligent understanding and semantic output of bird behavior. Researchers still need to spend a lot of time manually interpreting and labeling massive amounts of raw data. More importantly, existing systems generally adopt a one-way data transmission mode, which cannot provide feedback learning and parameter optimization based on researchers' professional judgment. As a result, the system's recognition ability is difficult to continuously improve with research needs in the long run, and there is a lack of an effective intelligent interaction mechanism between humans and the system, failing to form a closed loop of human-machine collaborative research assistance. Therefore, there is an urgent need for an AI system that can achieve multimodal data fusion perception, intelligent understanding of bird behavior, and deep interaction with researchers to support the efficient conduct of modern avian ecology research.

[0003] Chinese Patent Publication No. CN120727018A discloses a method, apparatus, and terminal device for biodiversity monitoring. The method includes: acquiring biological image information and biological sound information; extracting features from the biological sound information according to multiple preset biological sound feature extraction windows, preset filtering windows, and preset window processing functions to obtain multiple biological audio feature information; extracting features from the biological image information according to multiple preset biological image feature extraction matrices to obtain multiple biological image feature information; aligning the biological audio feature information and the biological image feature information based on preset basis vectors, preset encoding vectors, and preset spatial projection functions to obtain biological audio alignment features and biological image alignment features; enhancing the biological audio alignment features and biological image alignment features respectively to obtain biological audio monitoring features and biological image monitoring features; and monitoring organisms according to the biological audio monitoring features and biological image monitoring features.

[0004] Therefore, the existing technology has the following problems: it relies on fixed feature extraction parameters such as preset sound feature extraction window, filtering window and image feature extraction matrix, which makes it difficult to dynamically adjust the feature extraction process according to real-time environmental changes. It is also prone to feature extraction distortion due to background noise interference or changes in lighting conditions in complex field environments. Furthermore, it relies on aligning and enhancing biological audio features and image features to output monitoring results, which only reaches the level of identifying the presence of organisms and cannot make a refined judgment on the specific behavioral types of the identified organisms. Summary of the Invention

[0005] To address this, the present invention provides an AI interaction system for ornithological research. This system overcomes the technical problem in existing technologies where the accuracy of bird behavior interaction results is poor due to reliance on fixed feature extraction parameters and limitation to species identification. This is achieved by constructing a multi-level behavior determination architecture based on acoustic triggers and dynamically adjusting the acoustic trigger threshold in conjunction with environmental parameters and behavioral statistics.

[0006] To achieve the above objectives, this invention provides an AI interactive system for ornithological research, comprising: The data acquisition unit is used to periodically collect bird activity image sequences, bird song spectrum sequences, flight paths, changes in perching locations, foraging behavior frequency, population dispersal range, habitat vegetation structure, and environmental parameters within the bird observation area. The trigger determination unit is used to extract the frequency band change rate in the bird song spectrum sequence according to the preset AI model, and combine it with the preset acoustic change threshold to determine whether a trigger event corresponding to the activity of the target bird has occurred. The path determination unit is used to determine the path of the triggering event based on the determination result of the occurrence of the triggering event, according to the body shape contour change characteristics in the bird activity image sequence and the spatial turning characteristics of the flight trajectory, so as to obtain the flight path or the perching path. The dominant determination unit is used to determine the dominant factor type of the flight path as either a migratory type or a patrol type based on the flight path, according to the altitude change trend, speed change trend and the population dispersal range of the flight trajectory; and to determine the dominant factor type of the stopover path as either a foraging type or a warning type based on the stopover location change value, the number of foraging behaviors and the habitat vegetation structure. The prompt generation unit is used to generate corresponding interactive prompt information based on the dominant factor type; The adjustment unit is used to adjust the preset acoustic change threshold according to the time distribution characteristics and spatial distribution characteristics of each of the dominant factor types within a preset defined period, as well as the environmental parameters.

[0007] Furthermore, the trigger determination unit includes: An extraction subunit is used to extract acoustic features from the bird song spectrum sequence using a preset AI model, so as to obtain the spectral energy distribution characteristics of each preset frequency band, and calculate the frequency band change rate based on the change in spectral energy of each adjacent period. A trigger determination subunit is used to determine the occurrence of the trigger event based on a comparison result where the frequency band change rate is greater than the preset acoustic change threshold.

[0008] Furthermore, the path determination unit includes: The change feature determination subunit is used to extract the target contour of the bird activity image sequence based on the determination result of the occurrence of the triggering event, and calculate the area change, aspect ratio change and contour centroid displacement of the target contour within adjacent periods to obtain the body shape contour change feature. It also calculates the turning angle between consecutive preset first number of trajectory points, the number of turns per unit time and the trajectory curvature based on the flight trajectory to obtain the spatial turning feature. The path determination subunit is used to determine the behavioral factor path of the triggering event based on the body shape contour change features and the spatial turning features, so as to obtain the flight path or the resting path.

[0009] Furthermore, the path determination subunit includes: The flight index calculation module is used to calculate the flight behavior index based on the image motion index and the trajectory motion index. The image motion index is determined based on the displacement of the outline center of gravity and the change in area, and the trajectory motion index is determined based on the turning angle, the number of turns, and the trajectory curvature. The stopping index calculation module is used to calculate the stopping behavior index based on the outline stability index and the trajectory stability index. The outline stability index is determined based on the change in the aspect ratio and the displacement of the outline center of gravity, and the trajectory stability index is determined based on the number of turns and the trajectory curvature. The path determination module is used to determine the behavioral factor path as the flight path based on the comparison result that the flight behavior index is greater than a preset flight determination threshold, and to determine the behavioral factor path as the stopover path based on the comparison result that the stopover behavior index is greater than a preset stopover determination threshold.

[0010] Furthermore, the dominant decision-making unit includes: The flight characteristic calculation subunit is used to calculate the altitude change, average flight speed, and group diffusion radius within the preset unit time based on the flight path and the flight trajectory, so as to obtain the altitude change trend, the speed change trend, and the group diffusion range, respectively. The stopover feature calculation unit is used to calculate the stopover location movement distance within a preset unit time based on the stopover path and the change in stopover location, so as to obtain the stopover location change value; and to calculate the foraging frequency within a unit time based on the number of foraging behaviors; and to determine the vegetation coverage rate based on the habitat vegetation structure. The dominant determination subunit is used to determine whether the dominant factor type is the migratory type or the patrol type based on the altitude change, the average flight speed and the population diffusion radius, and to determine whether the dominant factor type is the foraging type or the alert type based on the stopover location change, the foraging frequency and the vegetation coverage.

[0011] Furthermore, the dominant determination subunit includes: The flight type index calculation module is used to determine the migratory flight index and the cruise flight index based on the altitude change, the average flight speed, and the group diffusion radius. A stopover type index calculation module is used to determine a foraging behavior index and a vigilance behavior index based on the stopover location change value, the foraging frequency, and the vegetation coverage. The dominant determination module is used to determine whether the dominant factor type is the migration type or the patrol type based on the comparison result of the migration behavior index and the patrol behavior index, and to determine whether the dominant factor type is the foraging type or the vigilance type based on the comparison result of the foraging behavior index and the vigilance behavior index.

[0012] Furthermore, the prompt generation unit includes: The prompt template library stores preset prompt templates corresponding to migration type, patrol type, foraging type and alert type, respectively. The prompt generation subunit is used to call the corresponding preset prompt template from the prompt template library according to the dominant factor type, and fill the preset prompt template with the time information, location information and environmental parameters corresponding to the triggering event, so as to generate interactive prompt information containing behavioral description and context information.

[0013] Furthermore, the adjustment unit includes: The distribution statistics subunit is used to statistically analyze the triggering events corresponding to each dominant factor type within the preset defined period, and record the occurrence time and location of each triggering event in order to calculate the temporal distribution frequency and spatial distribution density of each dominant factor type. An environmental correlation subunit is used to determine the correlation and change relationship between each dominant factor type and the environmental parameters based on the time distribution frequency, the spatial distribution density, and the corresponding environmental parameters. A threshold adjustment subunit is used to adjust the preset acoustic change threshold based on the correlation change relationship.

[0014] Furthermore, the environment-related subunit includes: The parameter extraction module is used to extract temperature parameters, humidity parameters, wind speed parameters, and light intensity parameters from the environmental parameters. The correlation calculation module is used to calculate the correlation coefficients between the temporal distribution frequency and spatial distribution density of each of the dominant factor types and each of the temperature parameters, humidity parameters, wind speed parameters and light intensity parameters. The correlation determination module is used to determine the key environmental parameters corresponding to the dominant factor type based on the comparison result between the correlation coefficient and the preset correlation determination threshold.

[0015] Furthermore, the threshold adjustment subunit includes: The weight determination module is used to determine the parameter weight of each key environmental parameter based on the correlation coefficient corresponding to the key environmental parameter. The threshold update module is used to calculate the acoustic threshold correction amount based on the parameter weights and the changes of the key environmental parameters within the preset defined period, and to adjust the preset acoustic change threshold based on the acoustic threshold correction amount.

[0016] Compared with existing technologies, the advantages of this invention lie in constructing a multi-level behavior determination process through joint analysis of bird song spectra, image motion characteristics, flight trajectories, and environmental factors. First, the rate of change of bird song spectrum energy across different frequency bands is used as a trigger. Since birds exhibit significant changes in the frequency structure and energy distribution of their calls during flight, foraging, or alertness, changes in spectral energy can identify potential activity events in their early stages. Subsequently, changes in image contours and flight trajectory turning characteristics are combined to distinguish behavioral paths. Flight behavior typically manifests as continuous center of gravity displacement, area changes, and trajectory curvature changes, while perching behavior is characterized by stable contours and reduced trajectory changes. These motion characteristics allow for the differentiation between flight paths and perching paths. Within flight paths, changes in altitude, average flight speed, and flock dispersal radius reflect the continuous ascent and dispersal characteristics of long-distance migration, while patrolling flight typically exhibits a more stable altitude and a smaller dispersal range. Within perching paths, changes in perching locations, foraging frequency, and vegetation cover reflect the dependence of foraging activities on resource distribution and the response characteristics of alertness behavior to spatial movement. By statistically analyzing the temporal and spatial distribution of these behavioral indicators within a preset period and correlating them with environmental factors such as temperature, humidity, wind speed, and light, the impact of environmental changes on the frequency and spatial concentration of bird activities can be identified. This allows for dynamic adjustment of the acoustic trigger threshold, enabling the system to maintain a sensitive ability to identify bird activities as the habitat changes, thereby achieving highly reliable behavioral recognition and interactive prompts for ornithological research. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the AI ​​interactive system for ornithological research in this embodiment; Figure 2 This is a schematic diagram of the trigger determination unit in this embodiment; Figure 3 This is a schematic diagram of the path determination unit in this embodiment; Figure 4 This is a schematic diagram of the main determination unit in this embodiment. Detailed Implementation

[0018] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0019] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0020] Please see Figure 1As shown, this is a schematic diagram of an AI interactive system for ornithological research according to this embodiment. This embodiment provides an AI interactive system for ornithological research, including: The data acquisition unit is used to periodically collect bird activity image sequences, bird song spectrum sequences, flight paths, changes in perching locations, foraging behavior frequency, population dispersal range, habitat vegetation structure, and environmental parameters within the bird observation area. The trigger determination unit is connected to the acquisition unit and is used to extract the frequency band change rate in the bird song spectrum sequence according to the preset AI model, and combine it with the preset acoustic change threshold to determine whether a trigger event corresponding to the activity of the target bird has occurred. The path determination unit is connected to the acquisition unit and the trigger determination unit respectively. Based on the determination result of the occurrence of the trigger event, it performs behavioral factor path determination on the trigger event according to the body shape contour change characteristics in the bird activity image sequence and the spatial turning characteristics of the flight trajectory, so as to obtain the flight path or perching path. The dominant determination unit, which is connected to the acquisition unit and the path determination unit respectively, is used to determine the dominant factor type of the flight path as either a migratory type or a patrol type based on the flight path, according to the altitude change trend, speed change trend and the population dispersal range of the flight trajectory; and to determine the dominant factor type of the stopover path as either a foraging type or a warning type based on the stopover location change value, the number of foraging behaviors and the habitat vegetation structure. A prompt generation unit, which is connected to the dominant determination unit, is used to generate corresponding interactive prompt information based on the dominant factor type. An adjustment unit, which is connected to the trigger determination unit and the dominant determination unit respectively, is used to adjust the preset acoustic change threshold according to the time distribution characteristics, spatial distribution characteristics and environmental parameters of each dominant factor type within a preset determination period.

[0021] In this embodiment, the acquisition unit is deployed at fixed observation stations or distributed monitoring nodes in the bird observation area. The bird observation area can be a wetland reserve, a migratory bird flyway, a forest habitat, or an urban ecological green space, or other areas with stable bird activity. Bird activity image sequences are acquired continuously by high-definition visible light cameras or infrared cameras positioned around the observation area. The cameras acquire image frames at a preset acquisition cycle to form a time-continuous image sequence. Bird song spectrum sequences are acquired by a directional microphone array to collect environmental sound signals within the observation area, and the acquired audio signals are processed using Fast Fourier Transform to obtain the spectral energy distribution of different frequency bands, thus forming a bird song spectrum sequence. Flight trajectories are obtained by continuously tracking bird targets identified in the image sequence and calculating the changes in the spatial coordinates of the targets in adjacent frames. Perching position changes are determined by identifying bird perching targets in the image sequence and recording their spatial position within the continuous acquisition cycle. Coordinate changes were statistically analyzed; the frequency of foraging behaviors was obtained by identifying bird head-down feeding or pecking actions in the image sequence and counting the number of occurrences per unit time; the population dispersal range was obtained by identifying the location coordinates of multiple bird targets within the observation area at the same time and calculating their maximum spatial distribution radius relative to the population center point; the habitat vegetation structure was obtained by analyzing the vegetation cover or vegetation type distribution within the observation area through surface imagery; environmental parameters were collected by environmental monitoring sensors installed at the observation station, including temperature sensors for air temperature, humidity sensors for air humidity, wind speed sensors for real-time wind speed, and light sensors for light intensity. The acquisition unit synchronously collected the above types of data at a preset acquisition cycle, which can be set from 1 second to 30 seconds, thus forming multi-source bird behavior observation data arranged in chronological order.

[0022] In this embodiment, a preset AI model is used to extract the frequency band energy variation features of bird calls from the bird call spectrum sequence. Specifically, continuous audio signals are first acquired using a directional sound pickup device placed in the observation area, and the audio signals are segmented into segments with preset time windows, the length of which is set to 0.5 seconds to 2 seconds. Then, a Fast Fourier Transform (FFT) is performed on the audio signal within each time window to obtain the corresponding spectral data. The spectral data is then segmented according to preset frequency intervals, thereby forming spectral energy distribution data for multiple preset frequency bands. The spectral energy distribution matrix formed by multiple consecutive time windows is used as the model input data, and the input data is normalized to ensure that the energy values ​​of each frequency band are within a uniform numerical range.

[0023] Based on this, a pre-defined AI model is constructed. This model employs a convolutional neural network (CNN), whose structure includes an input layer, several convolutional layers, pooling layers, and an output layer. The input layer receives the acoustic spectrum energy matrix composed of continuous time windows. The convolutional layers extract local features from the acoustic spectrum energy matrix to obtain the energy variation characteristics of bird calls across different frequency bands and time dimensions. The pooling layers reduce the dimensionality of the output features from the convolutional layers to reduce feature redundancy and improve feature stability. The output layer outputs the feature response values ​​for the corresponding frequency bands. By calculating the output feature response values, the acoustic spectrum energy change of each pre-defined frequency band between adjacent acquisition periods can be obtained, and the frequency band change rate can be further calculated based on the acoustic spectrum energy change.

[0024] To obtain the preset AI model, training audio data containing bird calls and non-bird calls are collected during the model building phase. The training audio undergoes the aforementioned spectrogram transformation process to form training samples. Samples containing bird calls are labeled as bird acoustic event samples, and environmental noise samples without bird calls are labeled as background sound samples. The training samples are input into a convolutional neural network model, and the model parameters are iteratively updated using the backpropagation algorithm. Model training is completed by minimizing the loss function between the model output and the sample labels, thereby obtaining the preset AI model for extracting the energy variation characteristics of the bird call frequency band. The structural parameters in the convolutional neural network, such as the kernel size, number of convolutional layers, pooling layer type, stride parameter, and number of network channels, can be set according to the spectrogram size, target recognition accuracy, and computational resource conditions. For example, the kernel size can be selected as 3×3 or 5×5, the number of convolutional layers can be set to 2 to 6 layers, and the pooling layers can use max pooling or average pooling. The specific values ​​of the above parameters are conventional structural configurations or performance optimization choices in the neural network model construction process, and those skilled in the art can adjust them appropriately according to specific application scenarios. As a specific implementation, the preset AI model can employ a lightweight convolutional neural network containing three convolutional layers and two fully connected layers. The input is a 128-dimensional Mel-ray spectrogram generated after pre-emphasis, frame segmentation (frame length 25ms), windowing, and Fast Fourier Transform. The first convolutional layer uses 32 3x3 convolutional kernels, the second convolutional layer uses 64 3x3 convolutional kernels, and the third convolutional layer uses 128 3x3 convolutional kernels. Each layer is followed by a ReLU activation function and a 2x2 max-pooling layer. The fully connected layers have 256 and 128 neurons respectively. Finally, the output layer uses the Sigmoid function to output the probability value that the current audio frame belongs to the target bird's song. The frequency band change rate is obtained by integrating the change of this probability value within a continuous time window.

[0025] The preset acoustic change threshold is a numerical threshold used to determine whether the energy change in the mid-frequency bands of the bird song spectrum reaches the triggering condition for bird activity. Its value depends on the background environmental noise level of the observation area and the energy amplification characteristics of the bird song in the spectrum. Specifically, environmental sound can be collected in the observation area for a period of time during the initial deployment of the system. The collected background audio is processed by Fast Fourier Transform and the energy change rate of each frequency band is calculated. The threshold range is determined by statistically analyzing the average value and fluctuation range of the energy change rate of the frequency bands under background noise conditions, combined with the energy change amplitude of the bird song samples in the corresponding frequency bands. Typically, the preset acoustic change threshold can be set between 0.15 and 0.40. In this embodiment, it is set to 0.25, so that when bird song occurs and causes a significant change in the energy of the corresponding frequency band spectrum, it can be effectively identified, while avoiding false triggering by short-term energy fluctuations caused by wind, insect sounds, or other environmental noises.

[0026] The preset determination period is a time period used to statistically analyze the temporal and spatial distribution characteristics of each dominant factor type. Its value depends on the temporal variation patterns of bird activity behavior and the data collection frequency in the observation area. Specifically, it can be set based on the sampling period of the collection unit and the statistical results of the duration of bird behavior during foraging, loitering, or migration, ensuring that multiple triggering events are included within this period to form a stable statistical distribution. Typically, the preset determination period can be set between 30 and 180 minutes; in this embodiment, it is set to 60 minutes. This allows for the accumulation of a sufficient number of behavioral events within this time range to calculate the temporal distribution frequency and spatial distribution density, ensuring the stability and representativeness of the statistical results.

[0027] A multi-level behavior determination process was constructed through joint analysis of bird song spectra, image motion characteristics, flight trajectories, and environmental factors. First, the rate of change of bird song spectrum energy across different frequency bands was used as a trigger. Since birds exhibit significant changes in the frequency structure and energy distribution of their calls during flight, foraging, or alertness, changes in spectral energy can identify potential activity events in their early stages. Subsequently, changes in image contours and flight trajectory turning characteristics were combined to distinguish behavioral paths. Flight behavior typically manifests as continuous center of gravity shifts, area changes, and trajectory curvature changes, while perching behavior is characterized by stable contours and reduced trajectory changes. These motion characteristics allow for the differentiation between flight and perching paths. Within flight paths, altitude changes, average flight speed, and flock dispersal radius reflect the continuous ascent and dispersal characteristics of long-distance migration, while patrolling flight typically exhibits a more stable altitude and a smaller dispersal range. Within perching paths, changes in perching locations, foraging frequency, and vegetation cover reflect the dependence of foraging activities on resource distribution and the response characteristics of alertness behavior to spatial movement. By statistically analyzing the temporal and spatial distribution of these behavioral indicators within a preset period and correlating them with environmental factors such as temperature, humidity, wind speed, and light, the impact of environmental changes on the frequency and spatial concentration of bird activities can be identified. This allows for dynamic adjustment of the acoustic trigger threshold, enabling the system to maintain a sensitive ability to identify bird activities as the habitat changes, thereby achieving highly reliable behavioral recognition and interactive prompts for ornithological research.

[0028] Please see Figure 2 As shown, this is a schematic diagram of the trigger determination unit in this embodiment. In this embodiment, the trigger determination unit includes: An extraction subunit is used to extract acoustic features from the bird song spectrum sequence using a preset AI model, so as to obtain the spectral energy distribution characteristics of each preset frequency band, and calculate the frequency band change rate based on the change in spectral energy of each adjacent period. A trigger determination subunit, which is connected to the extraction subunit, is used to determine the occurrence of the trigger event based on the comparison result that the frequency band change rate is greater than the preset acoustic change threshold.

[0029] The preset frequency band is a frequency interval formed by dividing the acoustic spectrum data obtained from the Fast Fourier Transform (FFT) according to fixed frequency intervals. Each frequency interval corresponds to an acoustic spectrum energy statistical unit. Its setting depends on the main frequency distribution range of bird calls and the frequency resolution of the acoustic spectrum analysis. Bird calls are typically concentrated in the range of approximately 1kHz to 8kHz. To ensure that different bird call frequency components can be effectively distinguished, and to avoid feature aliasing due to overly wide frequency band divisions or noise fluctuation amplification due to overly narrow band divisions, the frequency band width can be set according to the FFT spectral resolution and the bird call frequency distribution. Typically, the preset frequency band width can be set between 200Hz and 1000Hz; in this embodiment, it is set to 500Hz. This divides the 1kHz to 8kHz range into multiple continuous frequency intervals, and statistically analyzes the acoustic spectrum energy changes within each frequency interval, enabling stable identification of the concentrated energy changes of bird calls in specific frequency bands.

[0030] By extracting frequency band-level acoustic features from bird song spectral sequences and calculating the spectral energy changes between adjacent acquisition periods, the frequency band variation rate, reflecting the temporal variation characteristics of bird songs, can be obtained. Since bird calls typically possess a relatively stable dominant frequency range and harmonic structure, when bird activity occurs, their calls exhibit significant energy concentration in specific frequency bands, manifesting as rapid changes in energy amplitude within a continuous time window. In contrast, background noise in the natural environment, such as wind, water, or distant ambient sounds, often exhibits more dispersed and slowly changing spectral characteristics. By quantifying the spectral energy changes in each preset frequency band and comparing them with preset acoustic change thresholds, the spectral energy abrupt changes caused by bird songs can be distinguished from random fluctuations caused by background noise, thereby achieving effective trigger identification of bird activity events at the acoustic level.

[0031] Please see Figure 3 As shown, this is a schematic diagram of the path determination unit in this embodiment. In this embodiment, the path determination unit includes: The change feature determination subunit is used to extract the target contour of the bird activity image sequence based on the determination result of the occurrence of the triggering event, and calculate the area change, aspect ratio change and contour centroid displacement of the target contour within adjacent periods to obtain the body shape contour change feature. It also calculates the turning angle between consecutive preset first number of trajectory points, the number of turns per unit time and the trajectory curvature based on the flight trajectory to obtain the spatial turning feature. The path determination subunit, which is connected to the change feature determination subunit, is used to determine the behavioral factor path of the triggering event based on the body shape profile change feature and the spatial turning feature, so as to obtain the flight path or the resting path.

[0032] In this embodiment, the change feature determination subunit first performs target detection and contour extraction processing on the bird activity image sequence. Specifically, it can identify bird targets in the image frames through background modeling and foreground segmentation methods or deep learning-based target detection algorithms, and extract the outer contour of the identified target area after binarization, thereby obtaining the target contour corresponding to each acquisition cycle. Subsequently, the target contour area is calculated based on the contour pixel set, and the lengths of the major and minor axes of the target are determined by the least bounding rectangle or ellipse fitting method, thereby calculating the aspect ratio; at the same time, the contour centroid position is calculated based on the contour pixel coordinates, and the centroid displacement is obtained by the difference in centroid coordinates between adjacent acquisition cycles, thereby obtaining the body shape contour change feature. On the other hand, for the flight trajectory, the trajectory point sequence can be obtained by tracking the position coordinates of the same bird target in consecutive image frames, and a trajectory point set is formed in chronological order; the angle between adjacent trajectory segments is calculated based on a consecutive preset first number of trajectory points to obtain the turning angle, and the number of turns is obtained by counting the number of times the turning angle is greater than a preset angle threshold per unit time, and the trajectory curvature is calculated based on the curvature of the curve formed by the consecutive trajectory points, thereby obtaining the spatial turning feature.

[0033] The preset first quantity is the number of continuous trajectory points selected when calculating the spatial turning characteristics of the flight trajectory. Its value depends on the image sampling frequency of the acquisition unit and the spatial scale of the bird's flight path changes over a short period of time. When the number of trajectory points is too small, the calculated turning angle and trajectory curvature are easily affected by the error of a single sampling point, resulting in large fluctuations. On the other hand, too many trajectory points will cause excessive smoothing of local trajectory changes, thereby reducing the sensitivity to changes in flight direction. Therefore, an appropriate number of trajectory points can be determined according to the image sampling frequency and the typical flight speed of birds, so that the selected trajectory points can cover a complete local flight path. Typically, the preset first quantity can be set between 5 and 15. In this embodiment, it is set to 10, so that the continuous trajectory points can cover the flight path changes within a few seconds, thereby stably reflecting the spatial change characteristics of the bird's flight direction when calculating the turning angle, number of turns, and trajectory curvature.

[0034] By extracting the target contour from bird activity image sequences and calculating the changes in contour area, aspect ratio, and center of gravity displacement between adjacent acquisition periods, we can reflect the bird's posture changes and spatial displacement characteristics during movement. In flight, the wing deployment and folding cause periodic changes in contour area, and the target's center of gravity shifts significantly across consecutive image frames. In perching, the bird's posture is relatively stable, with smaller changes in contour area and center of gravity. Simultaneously, by calculating the turning angles between consecutive trajectory points, the number of turns per unit time, and trajectory curvature, we can reflect the bird's spatial movement patterns. Continuous flight typically involves continuous trajectory extension and large spatial distance traveled, while perching or short-distance movement results in limited trajectory changes and smaller curvature variations. By comprehensively analyzing the morphological movement characteristics reflected by contour changes and the spatial movement characteristics reflected by trajectory changes, we can distinguish bird behavior states from two levels: target morphological changes and spatial trajectory changes.

[0035] Specifically, the path determination subunit includes: The flight index calculation module is used to calculate the flight behavior index based on the image motion index and the trajectory motion index. The image motion index is determined based on the displacement of the outline center of gravity and the change in area, and the trajectory motion index is determined based on the turning angle, the number of turns, and the trajectory curvature. In this embodiment, the image motion index is calculated by performing maximum-minimum normalization on the contour centroid displacement and area change based on a preset historical sliding window, followed by weighted summation to obtain the image motion index. The preset image centroid weight corresponding to the contour centroid displacement depends on its ability to represent flight behavior. That is, the centroid displacement directly reflects the continuous movement range of birds on the image plane and is the core indicator for distinguishing between flight and perching. Usually, based on experience or statistical data analysis, its weight is set between 0.5 and 0.8. In this embodiment, it is set to 0.6 to ensure that displacement change dominates the index. The preset image area weight corresponding to the area change depends on its supplementary representation of flight action details. That is, the area change can reflect wing flapping or body posture adjustment and help determine whether it is in flight. However, it is easily affected by the shooting angle, so the weight is relatively low and is usually set between 0.2 and 0.5. In this embodiment, it is set to 0.4, which can highlight the dominant role of center of gravity displacement while taking into account the detailed representation of flight behavior by area change, thereby improving the accuracy and robustness of image motion index in identifying flight behavior. The trajectory motion index is calculated by performing maximum-minimum normalization on the steering angle, number of steering turns, and trajectory curvature, followed by weighted summation to obtain the trajectory motion index. The weight of the preset trajectory angle corresponding to the turning angle depends on its ability to represent the magnitude of changes in flight direction. That is, the turning angle can directly reflect the degree of change in direction of birds between continuous trajectory segments. When birds fly in the air, their paths are often accompanied by continuous direction adjustments. Therefore, this parameter can reflect the dynamic change characteristics of the flight path. Usually, its weight is set between 0.3 and 0.5 according to the statistical results of trajectory data. In this embodiment, it is set to 0.4. The weight of the preset trajectory number corresponding to the number of turns depends on the frequency of flight direction changes per unit time. That is, the number of turns can reflect the frequency of flight path direction changes in a short period of time. When birds are loitering or moving in space, they usually produce more turning behaviors. Therefore, this parameter can supplement the reflection of the dynamics of the trajectory. It is usually set between 0.2 and 0.4. In this embodiment, it is set to 0.3. The preset trajectory curvature weight corresponding to the trajectory curvature depends on the ability to represent the overall curvature of the flight trajectory. That is, the trajectory curvature can reflect the degree to which the trajectory line segment deviates from a straight line, thereby reflecting the spatial curvature characteristics of the flight path. It is usually set between 0.2 and 0.4. In this embodiment, it is set to 0.3, so that the trajectory angle change, direction change frequency and path curvature form complementary representations in the trajectory motion index, thereby improving the ability of the trajectory motion index to describe the motion state of the flight trajectory. The calculation process of the Flight Behavior Index involves a weighted summation of the image motion index and the trajectory motion index to obtain the Flight Behavior Index. The preset image motion weight corresponding to the image motion index depends on the bird's actual displacement and posture changes on the image plane and its ability to directly reflect flight behavior. That is, when the bird is in flight, its position and outline changes in consecutive image frames are more obvious. Therefore, the image motion features can directly reflect the existence of flight behavior. Usually, this weight is set between 0.5 and 0.7. In this embodiment, it is set to 0.6. The preset trajectory motion weight corresponding to the trajectory motion index depends on its supplementary representation of the changes in the spatial movement path of birds. That is, the trajectory motion index can reflect the changes in direction and trajectory curvature during flight from the perspective of spatial movement path. However, this feature is relatively greatly affected by factors such as trajectory sampling accuracy and environmental occlusion. Therefore, the weight is usually set between 0.3 and 0.5. In this embodiment, it is set to 0.4, so that the image motion feature plays a major role in the flight behavior index. At the same time, the trajectory motion feature supplements the description of the spatial movement state, so that the flight behavior index can more stably reflect the flight behavior characteristics of birds. The stopping index calculation module is used to calculate the stopping behavior index based on the outline stability index and the trajectory stability index. The outline stability index is determined based on the change in the aspect ratio and the displacement of the outline center of gravity, and the trajectory stability index is determined based on the number of turns and the trajectory curvature. In this embodiment, the stopover behavior index is calculated by weighted summation of the outline stability index and the trajectory stability index. Specifically, the outline stability index is calculated by weighted summation of the aspect ratio change and the outline centroid displacement after normalization based on a preset historical sliding window. The preset contour ratio weight corresponding to the aspect ratio change depends on its ability to represent the stability of the target posture. That is, when a bird is in a perching state, its body posture is basically stable and the aspect ratio change is small. Therefore, this parameter can reflect the stability of the target's external structure in continuous images. Usually, its weight is set between 0.5 and 0.7 based on the statistical results of historical samples. In this embodiment, it is set to 0.6. The preset contour displacement weight corresponding to the contour centroid displacement depends on its ability to characterize the spatial position stability of the target. That is, the overall position of the target changes little in the resting state, but it may still be slightly displaced due to minor posture adjustments or environmental disturbances. Therefore, its weight is relatively lower than the aspect ratio change, and is usually set between 0.3 and 0.5. In this embodiment, it is set to 0.4 so that the contour stability index can reflect both posture stability and spatial position stability.

[0036] The trajectory stability index is calculated by performing maximum-minimum normalization on the number of turns and trajectory curvature, followed by weighted summation. The preset trajectory number stability weight corresponding to the number of turns depends on the frequency of trajectory direction changes per unit time. When birds are in a resting state, their spatial movement is minimal, and the number of trajectory direction changes is typically low. Therefore, this parameter reflects the overall static characteristics of the trajectory, and its weight is usually set between 0.5 and 0.7; in this embodiment, it is set to 0.6. The preset trajectory curvature stability weight corresponding to the trajectory curvature depends on the ability of the overall curvature of the trajectory to characterize spatial motion stability. In a resting state, the trajectory length is shorter and the curvature change is smaller; therefore, this parameter reflects the spatial stability of the trajectory, and its weight is usually set between 0.3 and 0.5; in this embodiment, it is set to 0.4. This allows the trajectory stability index to comprehensively characterize the spatial stability characteristics of a resting state from both the frequency of trajectory direction changes and the changes in trajectory morphology, thus providing a stable and reliable basis for determining the resting behavior index.

[0037] The path determination module is connected to the flight index calculation module and the stopover index calculation module, respectively, and is used to determine the behavioral factor path as the flight path based on the comparison result that the flight behavior index is greater than the preset flight determination threshold, and to determine the behavioral factor path as the stopover path based on the comparison result that the stopover behavior index is greater than the preset stopover determination threshold.

[0038] The preset historical sliding window refers to the number of consecutive historical data periods selected when performing maximum-minimum normalization on the displacement of the centroid of the contour, the change in area, the change in aspect ratio, the turning angle, the number of turns, and the trajectory curvature. It is used to determine the reference range of the maximum and minimum values ​​in the normalization calculation. It depends on the temporal stability of the target motion feature changes and the video sampling frame rate. That is, the window length needs to be able to cover a complete or nearly complete typical behavior change cycle to avoid short-term outliers from having an excessive impact on the normalization results, while not being too long to reduce the sensitivity to behavior changes. In practical applications, it can be determined according to the video frame rate (e.g., 20 frames / second to 30 frames / second) and the duration of typical bird flight or posture changes (usually 2 to 5 seconds). Therefore, the preset historical sliding window is usually set between 40 and 150 frames. In this embodiment, it is set to 90 frames, so that the normalization calculation can maintain the stability of historical statistics while being sufficiently sensitive to behavior changes, thereby ensuring that the motion feature indicators are comparable across different time periods.

[0039] The preset flight judgment threshold refers to the numerical limit used to determine whether the flight behavior index has reached the flight state. It depends on the statistical distribution difference of the flight behavior index between flight samples and resting samples. That is, it is necessary to statistically analyze the value range of the flight behavior index in flight and non-flight states based on historical annotation data so that the threshold can be located in the boundary area of ​​the two types of sample distribution, thereby reducing misjudgment. After statistical analysis of historical video samples containing flight and resting behaviors, the flight behavior index threshold is usually set between 0.55 and 0.75. In this embodiment, it is set to 0.65 so that when the image motion features and trajectory motion features show obvious dynamic changes at the same time, it can be stably judged as a flight path, thereby improving the accuracy of flight behavior recognition.

[0040] The preset perching judgment threshold refers to the numerical limit used to determine whether the perching behavior index has reached the perching state. It depends on the statistical stability of the contour stability index and the trajectory stability index in the perching state. That is, it is necessary to analyze the typical range of contour changes and trajectory changes in the perching state based on historical sample data so that the threshold can effectively distinguish between continuous stable state and short-term movement state. After statistical analysis of video samples containing perching behavior, the perching behavior index threshold is usually set between 0.6 and 0.8. In this embodiment, it is set to 0.7 so that when the contour shape changes little and the trajectory movement is low, it can be stably judged as a perching path, thereby improving the reliability of perching behavior recognition.

[0041] By constructing image motion indices and trajectory motion indices to quantify flight behavior, and by constructing contour stability indices and trajectory stability indices to quantify perching behavior, bird behavior can be differentiated from two dimensions: morphological changes and spatial movement. In flight, the bird's center of gravity shifts continuously across consecutive image frames, and the flapping and folding of its wings causes significant changes in its contour area. Spatial movement manifests as continuous trajectory extension, directional adjustments, and changes in trajectory curvature. Therefore, by comprehensively calculating the contour center of gravity displacement, area change, turning angle, number of turns, and trajectory curvature, a flight behavior index reflecting continuous flight characteristics can be formed. In perching, the bird's position changes less, its body contour is relatively stable, and the aspect ratio changes less. Spatial trajectory points are mainly concentrated in local areas, and the number of turns and trajectory curvature changes less. Therefore, by constructing a contour stability index using aspect ratio changes and center of gravity displacement, and combining this with the number of turns and trajectory curvature to form a trajectory stability index, a perching behavior index reflecting the stability of the perching state can be formed. By comparing the flight behavior index and the stopover behavior index with their corresponding judgment thresholds, flight paths formed by continuous spatial movement and stopover paths formed by local stable activities can be effectively distinguished.

[0042] Please see Figure 4 As shown, this is a schematic diagram of the dominant determination unit in this embodiment. In this embodiment, the dominant determination unit includes: The flight characteristic calculation subunit is used to calculate the altitude change, average flight speed, and group diffusion radius within the preset unit time based on the flight path and the flight trajectory, so as to obtain the altitude change trend, the speed change trend, and the group diffusion range, respectively. The stopover feature calculation unit is used to calculate the stopover location movement distance within a preset unit time based on the stopover path and the change in stopover location, so as to obtain the stopover location change value; and to calculate the foraging frequency within a unit time based on the number of foraging behaviors; and to determine the vegetation coverage rate based on the habitat vegetation structure. The dominant determination subunit is connected to the flight characteristic calculation subunit and the perching characteristic calculation subunit, respectively, and is used to determine the dominant factor type as the migration type or the patrol type based on the altitude change, the average flight speed and the population diffusion radius, and to determine the dominant factor type as the foraging type or the alert type based on the perching location change value, the foraging frequency and the vegetation coverage.

[0043] In this embodiment, the calculation process for altitude change is as follows: Based on continuously collected flight trajectory points along the flight path, corresponding spatial coordinate data is obtained. Each trajectory point contains horizontal and altitude coordinate information. The altitude change value between adjacent sampling times is obtained by differential calculation of the altitude coordinates of continuous trajectory points. The altitude change values ​​are accumulated or averaged within the preset unit time period to obtain the altitude change within the preset unit time period. The calculation process for average flight speed is as follows: Calculation is performed based on the spatial distance between adjacent trajectory points along the flight path and the corresponding time interval. The spatial distance is determined using the three-dimensional Euclidean distance formula. The flight speeds of each time period within the preset unit time period are averaged to obtain the average flight speed. The calculation process for the group diffusion radius is as follows: Spatial aggregation analysis is performed on the trajectory points of multiple individual birds detected at the same collection time. First, the geometric center position of the group of all trajectory points is calculated. Then, the spatial distance between each trajectory point and the geometric center is calculated, and the mean or maximum value of this distance is selected as the group diffusion radius. By continuously statistically analyzing the group diffusion radius within the preset unit time period, the altitude change trend, the speed change trend, and the group diffusion range can be obtained respectively.

[0044] In this embodiment, the calculation process for the perching location movement distance is as follows: based on the continuously collected perching location coordinate data in the perching path, the spatial displacement distance between adjacent sampling times is calculated, and all displacement distances are accumulated and summed within the preset unit time to obtain the perching location movement distance, thereby forming the perching location change value; the calculation process for the foraging frequency is as follows: by performing behavioral recognition analysis on the perching image sequence, behaviors such as birds pecking at food with their heads down, picking up food from the ground, or jumping short distances are determined as one foraging behavior, and the number of foraging behaviors identified within the preset unit time is counted, and the statistical result is divided by the preset unit time to obtain the foraging frequency per unit time; the calculation process for the vegetation coverage rate is as follows: based on the image data of the perching area, the surface vegetation area is processed by image segmentation, and a vegetation mask image is formed by extracting green vegetation areas or texture feature areas, and the proportion of vegetation pixels to the total number of pixels in the perching area is calculated, and this proportion is used as the habitat vegetation coverage rate.

[0045] The preset unit duration refers to the length of the time interval used for statistical calculation of flight trajectory characteristics or perching behavior characteristics. Within this time interval, parameters such as altitude change, average flight speed, flock dispersal radius, perching location movement distance, and foraging behavior frequency are uniformly statistically analyzed to form comparable behavioral characteristic indicators. This depends on the typical timescale of bird behavior changes and the video sampling frequency of the monitoring system. When the time interval is too short, it is easily affected by instantaneous fluctuations or short-term individual actions, resulting in large random errors. Conversely, a time interval that is too long may mask the behavioral change process. Therefore, it is usually set according to the duration of bird flight or perching behavior and the frame rate of the monitoring video. In practical applications, it is generally set between 5 and 20 seconds. In this embodiment, it is set to 10 seconds to ensure that several continuous behavioral actions are fully covered within this time interval, thereby guaranteeing that the statistically obtained flight and perching characteristic parameters have good stability and representativeness.

[0046] By comprehensively analyzing three parameters in the flight path—altitude variation, average flight speed, and flock dispersal radius—bird flight behavior can be characterized from three dimensions: vertical motion characteristics, horizontal motion intensity, and flock spatial distribution pattern. Altitude variation reflects the continuous ascent or descent trend of birds adapting to airflow or terrain during migration, while average flight speed reflects the stable and high motion intensity during long-distance migration. Flock dispersal radius further reflects the spatial organization of the flock, maintaining a relatively tight formation or gradually spreading out during migration. When all three parameters simultaneously represent continuous altitude variation, high speed, and a relatively stable or gradually expanding flock distribution, migratory behavior can be effectively characterized. Conversely, when the altitude variation is small, the flight speed is moderate, and the flock dispersal radius is large, it is more likely to correspond to patrolling behavior within a local area. In stopover routes, stopover activities are comprehensively characterized by three parameters: stopover location movement distance, foraging frequency, and vegetation coverage. Stopover location movement distance reflects the extent of an individual's movement within a short range within the stopover area; foraging frequency reflects the activity level of an individual's foraging behavior per unit time; and vegetation coverage reflects habitat resource conditions and concealment conditions. When the stopover location movement distance is small but the foraging frequency and vegetation coverage are high, it usually corresponds to a foraging-oriented behavior state. Conversely, when the foraging frequency is low and the stopover location changes little in a high-vegetation-coverage environment, it is more likely to exhibit an alert or environmental observation state. By combining these parameters, bird activity characteristics can be cross-judged from both the perspective of movement behavior and environmental conditions, thereby improving the accuracy and stability of identifying dominant behavior types.

[0047] Specifically, the dominant determination subunit includes: The flight type index calculation module is used to determine the migratory flight index and the cruise flight index based on the altitude change, the average flight speed, and the group diffusion radius. In this embodiment, the calculation process for the migration behavior index and the cruise behavior index is as follows: First, the altitude change, average flight speed, and group diffusion radius are normalized to their maximum and minimum values ​​to map each parameter to a uniform numerical range. Then, a weighted combination calculation is performed based on different flight behavior characteristics. The migration behavior index is obtained by weighted summation of the normalized altitude change, average flight speed, and group diffusion radius. The migration altitude weight corresponding to the altitude change depends on its ability to characterize the continuous climb or descent trend during long-distance flight, and is usually set between 0.3 and 0.5; in this embodiment, it is set to 0.4. The migration speed weight corresponding to the average flight speed depends on the characteristics of high overall motion intensity and stable speed during migration, and is usually set between 0.3 and 0.5; in this embodiment, it is set to 0.4. The migration diffusion weight corresponding to the group diffusion radius depends on the changes in the spatial distribution of the group during migration. The ability of the metric to characterize the state of group cooperative flight is typically set between 0.1 and 0.3, and in this embodiment, it is set to 0.2. The loitering behavior index is obtained by another set of weighted summations of the normalized altitude change, average flight speed, and group diffusion radius. The loitering speed weight corresponding to the average flight speed depends on the characteristic of moderate speed but continuous movement during loitering, and is typically set between 0.3 and 0.5, and in this embodiment, it is set to 0.4. The loitering diffusion weight corresponding to the group diffusion radius depends on the spatial distribution characteristics of individuals dispersed in local areas during loitering, and is typically set between 0.3 and 0.5, and in this embodiment, it is set to 0.4. The loitering altitude weight corresponding to the altitude change depends on the characteristic of relatively small altitude change during loitering, and is typically set between 0.1 and 0.3, and in this embodiment, it is set to 0.2. Thus, different weight combinations form a distinguishing index between migratory flight behavior and loitering flight behavior. A stopover type index calculation module is used to determine a foraging behavior index and a vigilance behavior index based on the stopover location change value, the foraging frequency, and the vegetation coverage. In this embodiment, the calculation process for the foraging behavior index and the vigilance behavior index is as follows: First, the perching location change value, foraging frequency, and vegetation coverage are normalized by maximum-minimum values. Then, the foraging behavior index and the vigilance behavior index are calculated by weighted summation. The foraging behavior index is obtained by weighted summation of the normalized perching location change value, foraging frequency, and vegetation coverage. The foraging weight corresponding to the foraging frequency depends on the direct representation ability of foraging behavior per unit time and is usually set between 0.4 and 0.6. In this embodiment, it is set to 0.5. The foraging displacement weight corresponding to the perching location change value depends on the behavior characteristic of an individual continuously moving within a small range to find food during foraging. It is usually set between 0.2 and 0.4. In this embodiment, it is set to 0.3. The foraging environment weight corresponding to the vegetation coverage depends on the degree of influence of vegetation conditions on the distribution of food resources and is usually set between 0.1 and 0. The threshold values ​​are between 3 and 3, and in this embodiment, they are set to 0.2. The vigilance behavior index is obtained by another set of weighted summations of the normalized perching location change value, foraging frequency, and vegetation coverage. The vigilance environment weight corresponding to vegetation coverage depends on the characteristic that higher vegetation coverage can provide birds with concealment and form observation and vigilance positions, and is usually set between 0.4 and 0.6, and in this embodiment, it is set to 0.5. The vigilance displacement weight corresponding to perching location change value depends on the behavioral characteristic that individuals usually maintain a small range of movement but have slight position adjustments when vigilant, and is usually set between 0.2 and 0.4, and in this embodiment, it is set to 0.3. The vigilance foraging weight corresponding to foraging frequency depends on the characteristic that foraging behavior is significantly reduced when vigilant, and is usually set between 0.1 and 0.3, and in this embodiment, it is set to 0.2. Thus, the distinction between foraging behavior and vigilance behavior is achieved through the combination of behavioral activity characteristics and environmental conditions. The dominant determination module is used to determine whether the dominant factor type is the migration type or the patrol type based on the comparison result of the migration behavior index and the patrol behavior index, and to determine whether the dominant factor type is the foraging type or the vigilance type based on the comparison result of the foraging behavior index and the vigilance behavior index.

[0048] By normalizing altitude change, average flight speed, and flock dispersal radius, and constructing migration behavior indices and patrol behavior indices, bird flight status can be comprehensively characterized from three aspects: flight altitude change trend, movement intensity, and flock spatial organization morphology. Among them, migration is usually accompanied by continuous altitude adjustment and high and stable flight speed. At the same time, flocks maintain a relatively coordinated spatial distribution during long-distance migration. Therefore, increasing the weight of altitude change and flight speed in the migration index can more accurately reflect the dynamic characteristics of the migration process. Patrol behavior mostly occurs in the local area patrol or pre-foraging search stage. Its flight altitude change is small, but its spatial dispersion is high. Increasing the weight of flock dispersal radius in the patrol index can highlight this type of spatial distribution characteristic. Meanwhile, in the analysis of pausing behavior, foraging behavior index and vigilance behavior index are constructed by using pausing location change value, foraging frequency and vegetation coverage. This allows for a joint analysis of individual activity level and environmental resource conditions: foraging behavior is usually characterized by a high foraging frequency accompanied by small-scale location movement, while vigilance behavior is characterized by a lower individual activity frequency but often located in areas with high vegetation coverage to obtain concealment and observation conditions. By combining different weights, the two types of pausing behavior can be effectively distinguished.

[0049] Specifically, the prompt generation unit includes: The prompt template library stores preset prompt templates corresponding to migration type, patrol type, foraging type and alert type, respectively. The prompt generation subunit is used to call the corresponding preset prompt template from the prompt template library according to the dominant factor type, and fill the preset prompt template with the time information, location information and environmental parameters corresponding to the triggering event, so as to generate interactive prompt information containing behavioral description and context information.

[0050] In this embodiment, the preset prompt template refers to a structured text template used to generate bird behavior prompts. It is predefined using a fixed sentence structure combined with replaceable fields, and is independently set for behavior categories such as migration, patrol, foraging, and alert. Each preset prompt template includes a behavior description field, a time placeholder field, a location placeholder field, and an environmental parameter placeholder field. The behavior description field describes the basic characteristics of the corresponding behavior type; for example, migration corresponds to "a flock of birds is detected migrating in the monitoring area," and patrol corresponds to "a flock of birds is detected patrolling within the monitoring area." The time placeholder field is used to fill in the time information of the triggering event, such as the specific moment or time interval. The location placeholder field is used to fill in the spatial location information collected by the monitoring equipment, such as latitude and longitude coordinates or the monitoring area number. The environmental parameter placeholder field is used to fill in the environmental parameter information collected at the corresponding time point, including at least one of temperature, wind speed, humidity, or light intensity. Each preset prompt template is stored in a text structure with placeholders. For example, the corresponding fill fields are identified by "{time}", "{location}", "{environmental parameters}", etc. When the prompt information is generated, the corresponding data is replaced in the placeholder position to form a complete behavior prompt statement.

[0051] By associating dominant factor types with a cue template library and incorporating the time, location, and environmental parameters of the triggering event when generating cue information, the behavior determination results can form a complete description of the specific observation context. The dominant factor type reflects the bird's current behavioral state characteristics, while the time information reflects the time-series patterns of the behavior, such as migration activities typically concentrated in specific seasons or during the day. The location information reflects the spatial location of the behavior, thus determining whether it occurs in a migration corridor, habitat, or foraging area. Environmental parameters further describe the impact of climate, light, or habitat conditions on the behavior. By jointly expressing the behavior type with time, space, and environmental information, behavior cue information with a complete context can be formed, ensuring that the cue content not only explains what behavior occurred but also reflects the specific conditions and environmental background of the behavior.

[0052] In this embodiment, the prompt generation unit further includes an interactive interface module and an interactive feedback module, wherein, The interactive interface module is used to present interactive prompts in a visual form on scientific research terminal devices, including mobile terminals, computer workstations, or large screens in monitoring centers. The interactive interface module provides a graphical user interface, including a behavior timeline view, a spatial distribution heatmap, and a real-time prompt message list. The behavior timeline view displays the trigger times and durations of each dominant factor type (migration, loitering, foraging, and alert) in a timeline format, and supports time interval zooming and positioning functions, enabling researchers to quickly trace behavioral events within a specific time period. The spatial distribution heatmap maps the locations of each triggering event to the observation area map, distinguishing different dominant factor types with different colors or icons, allowing researchers to intuitively identify spatial clusters and activity hotspots of bird behavior. The real-time prompt message list displays generated interactive prompts sequentially in a message stream format, sorted and filtered by behavior type, occurrence time, or spatial location, supporting researchers in marking, annotating, or categorizing the prompts.

[0053] The interactive feedback module receives feedback on interactive prompts and dynamically adjusts subsequent prompt generation strategies based on this feedback. Feedback actions include researchers' confirmation, correction, ignoring, or annotation of prompts. For example, researchers might confirm the accuracy of a migration prompt, correct the actual behavior type of a foraging prompt, or annotate a warning prompt as belonging to a specific bird species. The interactive feedback module records these feedback actions and associates them with the original data of the corresponding triggering events (including bird song spectrogram sequences, bird activity image sequences, flight trajectories, environmental parameters, etc.) to form a calibration sample dataset with researcher annotations. Based on this, the interactive feedback module transmits the calibration sample dataset to the adjustment unit to optimize various judgment parameters in the preset AI model, preset acoustic change thresholds, or path determination unit and the dominant determination unit. This allows the system to gradually improve the accuracy of behavior recognition and prompt generation based on researchers' professional feedback, forming a continuous optimization mechanism through human-machine collaboration.

[0054] In this embodiment, the interactive interface module also provides a data export interface, which is used to export the interactive prompts, trigger event raw data and the researchers' feedback annotation information corresponding to the specified time period or the specified dominant factor type in a structured format according to the researchers' selection instructions. The structured format includes comma-separated value file format or geographic information system compatible vector data format, which can be used by researchers to carry out subsequent statistical analysis, behavior spectrum construction or academic paper writing.

[0055] Through the aforementioned interactive interface and feedback modules, the system can not only automatically identify bird behavior and generate prompts, but also provide researchers with an intuitive visual interface, flexible data retrieval and management functions, and a model optimization mechanism based on professional feedback. This transforms the system from a one-way data output tool into a two-way scientific research collaboration platform. Researchers can quickly grasp the temporal patterns and spatial distribution of bird behavior through the interactive interface, and use the feedback function to correct and supplement the system's identification results. The system, in turn, utilizes this feedback data to continuously improve its identification capabilities, providing efficient, accurate, and sustainable technical support for ornithological research.

[0056] Specifically, the adjustment unit includes: The distribution statistics subunit is used to statistically analyze the triggering events corresponding to each dominant factor type within the preset defined period, and record the occurrence time and location of each triggering event in order to calculate the temporal distribution frequency and spatial distribution density of each dominant factor type. In this embodiment, all trigger events recorded by the monitoring system within the preset defined period are summarized and classified according to the corresponding dominant factor type. Each trigger event records its occurrence timestamp and corresponding spatial location information. The time information is obtained through the clock module of the monitoring system, and the spatial location information is determined by the installation location of the monitoring equipment or image positioning results, such as recording in the form of latitude and longitude coordinates or monitoring area number. Then, the preset defined period is used as the statistical interval, and the period is divided into several equal-length segments. The number of trigger events of each dominant factor type occurring in each time segment is counted, and the time distribution frequency is obtained by dividing the counted number of events by the length of the corresponding time segment. At the same time, based on the location data corresponding to each trigger event, a spatial grid or area division is established in the monitoring area. Each trigger event is mapped to the corresponding grid or area according to its occurrence location, and the number of trigger events per unit area is counted. The spatial distribution density of each dominant factor type in the monitoring area is calculated by the ratio of the number of events to the corresponding spatial area, thereby forming statistical results for describing the distribution characteristics of various behavioral activities in the time and spatial dimensions.

[0057] An environmental correlation subunit is used to determine the correlation and change relationship between each dominant factor type and the environmental parameters based on the time distribution frequency, the spatial distribution density, and the corresponding environmental parameters. A threshold adjustment subunit is used to adjust the preset acoustic change threshold based on the correlation change relationship.

[0058] By conducting statistical analysis of the triggering events corresponding to different dominant factor types within a preset period in both temporal and spatial dimensions, and further combining this with correlation analysis using environmental parameters, the relationship between changes in bird behavior and external environmental conditions can be quantitatively described. Temporal distribution frequency reflects the occurrence pattern of specific behaviors over different time periods, spatial distribution density reflects the degree of aggregation or diffusion of this behavior within the monitoring area, and environmental parameters reflect the influence of external conditions such as temperature, wind speed, or light on bird activity. When a certain behavior type exhibits a higher temporal frequency or more pronounced spatial aggregation characteristics under specific environmental conditions, it indicates that the environmental conditions have a significant impact on this type of acoustic or behavioral activity. By comprehensively analyzing the above statistical relationships and adjusting the preset acoustic change threshold accordingly, the acoustic triggering conditions can be dynamically corrected according to actual environmental changes. This allows the system to maintain a stable ability to identify bird activity under different environmental conditions, improving overall monitoring accuracy and environmental adaptability.

[0059] Specifically, the environment association subunit includes: The parameter extraction module is used to extract temperature parameters, humidity parameters, wind speed parameters, and light intensity parameters from the environmental parameters. The correlation calculation module is used to calculate the correlation coefficients between the temporal distribution frequency and spatial distribution density of each of the dominant factor types and each of the temperature parameters, humidity parameters, wind speed parameters and light intensity parameters. The correlation determination module is used to determine the key environmental parameters corresponding to the dominant factor type based on the comparison result between the correlation coefficient and the preset correlation determination threshold.

[0060] In this embodiment, the parameter extraction module reads environmental parameter records corresponding to the trigger event time from the environmental data collected by the monitoring system, and extracts temperature, humidity, wind speed, and light intensity parameters from them. Temperature is expressed in degrees Celsius, humidity in relative humidity percentage, wind speed in meters per second, and light intensity in lux. An environmental parameter sequence is then established according to the chronological order of the trigger events. The correlation calculation module, within the preset defined period, calculates the correlation between the time distribution frequency sequence and spatial distribution density sequence corresponding to each dominant factor type and the corresponding temperature, humidity, wind speed, and light intensity parameter sequences. The correlation coefficient is calculated using the Pearson correlation coefficient, which is calculated by multiplying the behavioral statistical sequence and the environmental parameter sequence by their respective means, summing the products, and dividing by the product of the standard deviations of the two sequences. This yields a correlation coefficient ranging from -1 to 1, representing the degree of correlation between changes in behavioral activity and changes in environmental parameters. The correlation determination module compares the absolute value of the calculated correlation coefficient with the preset correlation determination threshold. When the absolute value of the correlation coefficient corresponding to a certain environmental parameter is greater than the preset correlation determination threshold, the environmental parameter is determined to be a key environmental parameter of the corresponding dominant factor type, and the correlation between the environmental parameter and the corresponding behavior type is recorded for subsequent threshold adjustment process.

[0061] The preset correlation threshold is a numerical limit used to determine whether there is a significant correlation between environmental parameters and bird behavior. Its value depends on the significance level required for the correlation coefficient in the sample statistics and the sensitivity of bird behavior to environmental changes. In other words, it is necessary to determine, based on statistical principles, what correlation coefficient value is needed to consider a real, rather than accidental, correlation between changes in environmental parameters and changes in behavioral frequency or spatial distribution. In behavioral ecology research, the impact of environmental factors on bird activity usually exhibits a gradual change. A correlation coefficient below 0.2 is generally considered to be extremely weak or meaningless; a value between 0.2 and 0.3 indicates a weak correlation that is easily affected by sample fluctuations; and a value above 0.3 begins to have a relatively clear indicative significance and can reflect... To reflect the stable influence trend of environmental conditions on bird behavior, and considering the limited number of triggering events participating in the statistics within a preset period, the statistical significance of the correlation coefficient will be constrained by the sample size. Therefore, it is necessary to select a threshold range that can eliminate random noise interference without being overly strict and causing omission of true correlations. Taking into account the above factors, the preset correlation judgment threshold is usually set between 0.25 and 0.35. In this embodiment, it is set to 0.3, so that only when the absolute value of the correlation coefficient between the environmental parameter and the behavioral statistics exceeds 0.3 is the environmental parameter judged as a key environmental parameter of the corresponding dominant factor type. This ensures that the environmental factors participating in the threshold adjustment have a real and reliable statistical correlation basis, and avoids the improper adjustment of the acoustic threshold due to weak or spurious correlations.

[0062] By conducting quantitative correlation analysis on the temporal distribution frequency and spatial distribution density of each dominant factor type with environmental parameters such as temperature, humidity, wind speed, and light intensity, the key environmental conditions affecting the behavioral patterns of different birds can be accurately identified. This allows for a clear characterization of the intrinsic connection between triggering events and environmental changes, thereby enabling dynamic adjustment of the acoustic judgment threshold to adapt to environmental changes and improving the accuracy and stability of behavior recognition. The beneficial effect of this embodiment is that it enhances the system's real-time adaptability to bird activities and environmental sensitivity.

[0063] Specifically, the threshold adjustment subunit includes: The weight determination module is used to determine the parameter weight of each key environmental parameter based on the correlation coefficient corresponding to the key environmental parameter. In this embodiment, the process of determining the parameter weights of each key environmental parameter includes: First, obtaining the correlation coefficients between each key environmental parameter and its corresponding dominant factor type, calculated by the correlation determination module. These correlation coefficients reflect the degree of correlation between changes in environmental parameters and changes in behavioral frequency or spatial distribution. For each key environmental parameter, the absolute value of its correlation coefficient is taken as a measure of the intensity of its influence on behavior. Subsequently, the absolute values ​​of the correlation coefficients of all key environmental parameters are summed to obtain the total correlation strength. Based on this, the absolute value of the correlation coefficient of each key environmental parameter is divided by the total correlation strength; the resulting ratio is the parameter weight of that key environmental parameter. This weight reflects the proportion of contribution that environmental parameter should have to the acoustic threshold adjustment in the overall environmental change. If no key environmental parameter is identified within a preset defined period, i.e., the absolute values ​​of the correlation coefficients of all environmental parameters are lower than the preset correlation determination threshold, the system maintains the original preset acoustic change threshold unchanged and does not perform weight calculation or threshold adjustment to ensure stable operation of the system in the absence of significant environmental correlations.

[0064] The threshold update module is used to calculate the acoustic threshold correction amount based on the parameter weights and the changes of the key environmental parameters within the preset determined period, and adjust the preset acoustic change threshold S based on the acoustic threshold correction amount Δy, S'=S×(1+μ×Δy), where S' is the adjusted preset acoustic change threshold and μ is the preset correction coefficient. In this embodiment, the process of calculating the acoustic threshold correction includes: First, for each key environmental parameter with a determined parameter weight, its average value within the current preset determination period is calculated and compared with the average value within the previous preset determination period to obtain the change in the parameter. The change reflects the fluctuation amplitude and direction of the environmental parameter relative to historical levels. Then, the change in each key environmental parameter is multiplied by its corresponding parameter weight, and all weighted changes are summed to obtain a comprehensive environmental change index. This index comprehensively represents the overall change trend of multiple environmental factors within the current period. Subsequently, the comprehensive environmental change index is scaled according to a preset correction coefficient. The correction coefficient is preset based on the acoustic change threshold's response sensitivity to environmental changes. Multiplying the comprehensive environmental change index by the correction coefficient yields the acoustic threshold correction. Finally, the acoustic threshold correction is added to the current preset acoustic change threshold to obtain the updated threshold. The updated threshold is then subjected to boundary verification to ensure it is within a preset reasonable value range. If it exceeds the range, its boundary value is taken. The threshold that passes the verification is used as the new preset acoustic change threshold for triggering the next period.

[0065] The preset correction coefficient is a proportional factor used to map the comprehensive environmental change index to the acoustic threshold adjustment range. Its value depends on the sensitivity requirements of the acoustic threshold to environmental fluctuations and the statistical distribution range of the comprehensive environmental change index itself. In other words, a reasonable linear mapping relationship needs to be established between the environmental change range and the threshold adjustment amount, so that the adjusted threshold can effectively follow environmental changes without excessive sensitivity leading to frequent fluctuations. Specifically, the comprehensive environmental change index is obtained by summing the weighted relative changes of various key environmental parameters. Its numerical range is usually constrained by the fluctuation characteristics of the environmental parameters themselves. Under natural conditions, the periodic changes of parameters such as temperature, humidity, and light intensity relative to historical averages are mostly concentrated within ±0.5%. Although wind speed changes may be larger, after correlation screening, their... The weights and changes are comprehensively balanced, so the typical fluctuation range of the comprehensive environmental change index is usually between ±0.6. If the correction coefficient is set too small, environmental changes cannot effectively adjust the threshold, and the system is not adaptable enough to the environment. If the correction coefficient is set too large, even small environmental fluctuations will cause drastic changes in the threshold, leading to instability in the trigger judgment criteria and increasing the risk of misjudgment. Taking the above factors into consideration, the preset correction coefficient is usually set between 0.05 and 0.2. In this embodiment, it is set to 0.1, so that when the comprehensive environmental change index is ±0.5, the corresponding threshold adjustment range is ±5%. This allows the threshold to respond effectively to environmental changes while ensuring that the adjustment process is stable and controllable, thereby enhancing the system's adaptability to environmental conditions while maintaining the stability of trigger event recognition.

[0066] By dynamically allocating weights based on the correlation coefficients of key environmental parameters and calculating the acoustic threshold correction amount in combination with the changes of each environmental parameter within a preset period, the threshold can be automatically adjusted with environmental changes such as temperature, humidity, wind speed and light intensity. When the environment has a significant impact on bird behavior, the threshold is strengthened or weakened accordingly, thereby ensuring that the triggering judgment does not miss important behaviors and suppresses false triggering caused by environmental fluctuations.

[0067] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An AI interactive system for ornithological research, characterized in that, include: The data acquisition unit is used to periodically collect bird activity image sequences, bird song spectrum sequences, flight paths, changes in perching locations, foraging behavior frequency, population dispersal range, habitat vegetation structure, and environmental parameters within the bird observation area. The trigger determination unit is used to extract the frequency band change rate in the bird song spectrum sequence according to the preset AI model, and combine it with the preset acoustic change threshold to determine whether a trigger event corresponding to the activity of the target bird has occurred. The path determination unit is used to determine the path of the triggering event based on the determination result of the occurrence of the triggering event, according to the body shape contour change characteristics in the bird activity image sequence and the spatial turning characteristics of the flight trajectory, so as to obtain the flight path or the perching path. The dominant determination unit is used to determine the dominant factor type of the flight path as either a migratory type or a patrol type based on the flight path, according to the altitude change trend, speed change trend and the population dispersal range of the flight trajectory; and to determine the dominant factor type of the stopover path as either a foraging type or a warning type based on the stopover location change value, the number of foraging behaviors and the habitat vegetation structure. The prompt generation unit is used to generate corresponding interactive prompt information based on the dominant factor type; The adjustment unit is used to adjust the preset acoustic change threshold according to the time distribution characteristics and spatial distribution characteristics of each of the dominant factor types within a preset defined period, as well as the environmental parameters.

2. The AI ​​interactive system for ornithological research according to claim 1, characterized in that, The trigger determination unit includes: An extraction subunit is used to extract acoustic features from the bird song spectrum sequence using a preset AI model, so as to obtain the spectral energy distribution characteristics of each preset frequency band, and calculate the frequency band change rate based on the change in spectral energy of each adjacent period. A trigger determination subunit is used to determine the occurrence of the trigger event based on a comparison result where the frequency band change rate is greater than the preset acoustic change threshold.

3. The AI ​​interactive system for ornithological research according to claim 2, characterized in that, The path determination unit includes: The change feature determination subunit is used to extract the target contour of the bird activity image sequence based on the determination result of the occurrence of the triggering event, and calculate the area change, aspect ratio change and contour centroid displacement of the target contour within adjacent periods to obtain the body shape contour change feature. It also calculates the turning angle between consecutive preset first number of trajectory points, the number of turns per unit time and the trajectory curvature based on the flight trajectory to obtain the spatial turning feature. The path determination subunit is used to determine the behavioral factor path of the triggering event based on the body shape contour change features and the spatial turning features, so as to obtain the flight path or the resting path.

4. The AI ​​interactive system for ornithological research according to claim 3, characterized in that, The path determination subunit includes: The flight index calculation module is used to calculate the flight behavior index based on the image motion index and the trajectory motion index. The image motion index is determined based on the displacement of the outline center of gravity and the change in area, and the trajectory motion index is determined based on the turning angle, the number of turns, and the trajectory curvature. The stopping index calculation module is used to calculate the stopping behavior index based on the outline stability index and the trajectory stability index. The outline stability index is determined based on the change in the aspect ratio and the displacement of the outline center of gravity, and the trajectory stability index is determined based on the number of turns and the trajectory curvature. The path determination module is used to determine the behavioral factor path as the flight path based on the comparison result that the flight behavior index is greater than a preset flight determination threshold, and to determine the behavioral factor path as the stopover path based on the comparison result that the stopover behavior index is greater than a preset stopover determination threshold.

5. The AI ​​interactive system for ornithological research according to claim 4, characterized in that, The dominant decision-making unit includes: The flight characteristic calculation subunit is used to calculate the altitude change, average flight speed, and group diffusion radius within the preset unit time based on the flight path and the flight trajectory, so as to obtain the altitude change trend, the speed change trend, and the group diffusion range, respectively. The stopover feature calculation unit is used to calculate the stopover location movement distance within a preset unit time based on the stopover path and the change in stopover location, so as to obtain the stopover location change value; and to calculate the foraging frequency within a unit time based on the number of foraging behaviors; and to determine the vegetation coverage rate based on the habitat vegetation structure. The dominant determination subunit is used to determine whether the dominant factor type is the migratory type or the patrol type based on the altitude change, the average flight speed and the population diffusion radius, and to determine whether the dominant factor type is the foraging type or the alert type based on the stopover location change, the foraging frequency and the vegetation coverage.

6. The AI ​​interactive system for ornithological research according to claim 5, characterized in that, The dominant determination subunit includes: The flight type index calculation module is used to determine the migratory flight index and the cruise flight index based on the altitude change, the average flight speed, and the group diffusion radius. A stopover type index calculation module is used to determine a foraging behavior index and a vigilance behavior index based on the stopover location change value, the foraging frequency, and the vegetation coverage. The dominant determination module is used to determine whether the dominant factor type is the migration type or the patrol type based on the comparison result of the migration behavior index and the patrol behavior index, and to determine whether the dominant factor type is the foraging type or the vigilance type based on the comparison result of the foraging behavior index and the vigilance behavior index.

7. The AI ​​interactive system for ornithological research according to claim 6, characterized in that, The prompt generation unit includes: The prompt template library stores preset prompt templates corresponding to migration type, patrol type, foraging type and alert type, respectively. The prompt generation subunit is used to call the corresponding preset prompt template from the prompt template library according to the dominant factor type, and fill the preset prompt template with the time information, location information and environmental parameters corresponding to the triggering event, so as to generate interactive prompt information containing behavioral description and context information.

8. The AI ​​interactive system for ornithological research according to claim 7, characterized in that, The adjustment unit includes: The distribution statistics subunit is used to statistically analyze the triggering events corresponding to each dominant factor type within the preset defined period, and record the occurrence time and location of each triggering event in order to calculate the temporal distribution frequency and spatial distribution density of each dominant factor type. An environmental correlation subunit is used to determine the correlation and change relationship between each dominant factor type and the environmental parameters based on the time distribution frequency, the spatial distribution density, and the corresponding environmental parameters. A threshold adjustment subunit is used to adjust the preset acoustic change threshold based on the correlation change relationship.

9. The AI ​​interactive system for ornithological research according to claim 8, characterized in that, The environment-related subunit includes: The parameter extraction module is used to extract temperature parameters, humidity parameters, wind speed parameters, and light intensity parameters from the environmental parameters. The correlation calculation module is used to calculate the correlation coefficients between the temporal distribution frequency and spatial distribution density of each of the dominant factor types and each of the temperature parameters, humidity parameters, wind speed parameters and light intensity parameters. The correlation determination module is used to determine the key environmental parameters corresponding to the dominant factor type based on the comparison result between the correlation coefficient and the preset correlation determination threshold.

10. The AI ​​interactive system for ornithological research according to claim 9, characterized in that, The threshold adjustment subunit includes: The weight determination module is used to determine the parameter weight of each key environmental parameter based on the correlation coefficient corresponding to the key environmental parameter. The threshold update module is used to calculate the acoustic threshold correction amount based on the parameter weights and the changes of the key environmental parameters within the preset defined period, and to adjust the preset acoustic change threshold based on the acoustic threshold correction amount.

Images