A water bird species identification method and system based on dynamic adaptation of distance of a variable-power ball machine

Abstract

This invention discloses a method and system for waterbird species identification based on a dynamically adaptive distance using a PTZ camera, relating to the field of waterbird monitoring technology. The method includes configuring the PTZ camera hardware and monitoring control parameters according to the needs of the waterbird monitoring scenario and the body shape characteristics of the target waterbirds; calculating the effective monitoring range and dividing the monitoring layers; dynamically adjusting the horizontal rotation angle, vertical tilt angle, and lens focal length of the PTZ camera; capturing images layer by layer from far to near, ensuring that the pixel ratio of waterbirds meets the identification requirements and that overlapping pixels are retained in adjacent photos; after preprocessing the captured photos, inputting them into a waterbird identification model for identification and filtering high-confidence results; statistically analyzing the species and quantity based on deduplication rules; evaluating the model accuracy and iteratively optimizing it through supplementary samples to adapt for the identification of new species. The system includes a PTZ camera, control, data processing, and storage interaction units, enabling long-term, all-weather, automated waterbird monitoring, solving problems such as low automation and data lag in traditional monitoring.

Description

Technical Field

[0001] This invention relates to the field of wild waterbird monitoring technology, specifically to a method and system for identifying waterbird species based on the dynamic adaptive distance of a variable magnification PTZ camera. Background Technology

[0002] In the field of wild waterbird ecological monitoring, existing technologies still face several bottlenecks that urgently need to be addressed, making it difficult to meet the needs of long-term, all-weather, and precise monitoring. Current mainstream monitoring methods mainly include traditional manual surveys, infrared camera monitoring, drone aerial photography, and manually controlled PTZ camera capture. However, these methods have several technical drawbacks: traditional manual surveys require long-term field patrols by professionals, which is not only time-consuming and labor-intensive but also inefficient, especially in large-scale or remote wetland environments where applicability is extremely poor. Furthermore, the identification results are highly dependent on the observer's experience and weather conditions, resulting in large errors and difficulty in standardization. Additionally, human intervention can easily disturb birds and disrupt their normal behavior. While infrared camera monitoring can achieve 24-hour monitoring, the complex wetland environment makes the equipment susceptible to flooding, corrosion, or human damage, leading to high deployment and maintenance costs. Moreover, factors such as water surface reflection and wave movement can easily cause errors. Triggering certain factors generates a large amount of invalid data, and the shooting range is only a few meters, making it impossible to achieve full-area coverage monitoring. Although drone aerial monitoring has the advantages of wide coverage and non-contact, the strong winds at the seaside severely affect flight stability, and the battery life is limited (the battery needs to be replaced after only 20-30 minutes of single monitoring), resulting in low monitoring efficiency. At the same time, the massive amount of photos and videos captured require subsequent manual identification of species and counting, leading to a serious lag in monitoring results. Although manually controlled PTZ cameras can acquire high-definition bird photos from long distances, they rely on manual operation of the gimbal rotation, resulting in low automation, making it impossible to maintain long-term operation, and it is difficult to achieve full coverage monitoring of the visible area, with obvious limitations in monitoring.

[0003] In summary, existing monitoring methods generally suffer from technical drawbacks such as low automation, high labor costs, poor environmental adaptability, delayed data statistics, and limited monitoring coverage, failing to meet the demands for intelligent, efficient, and comprehensive waterbird monitoring. Therefore, developing a technology that can dynamically adapt to monitoring distances, achieve automated imaging across the entire area, and accurately identify and count waterbird species is a pressing issue that needs to be addressed. Summary of the Invention

[0004] To address the technical problems existing in the prior art, the first objective of this invention is to provide a method for identifying waterbird species based on the dynamic adaptive distance of a variable magnification PTZ camera. This method achieves full-area, layer-by-layer, full-coverage shooting by dynamically adjusting the PTZ parameters of the PTZ camera. Combined with AI recognition and deduplication statistics, it enables long-term, all-weather, and automated species identification and quantity statistics of waterbirds.

[0005] The second objective of this invention is to provide a waterbird species identification system based on the dynamic adaptive distance of a variable magnification PTZ camera. Through the collaboration of the variable magnification PTZ camera unit, control unit, data processing unit, and storage and interaction unit, the system provides hardware and software support for the above-mentioned identification method, meets the waterbird monitoring needs of wetland scenarios such as islands and tidal flats, and fills the gap in intelligent and full-coverage monitoring of existing monitoring systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A method for identifying waterbird species based on the dynamic adaptive distance of a zoom PTZ camera includes the following steps: Based on the requirements of waterbird monitoring scenarios and the physical characteristics of target waterbirds, the hardware parameters and monitoring control parameters of the zoom PTZ camera are configured, the effective monitoring range of the PTZ camera is calculated and the monitoring layers are divided, and the horizontal rotation angle, vertical pitch angle and lens focal length of the PTZ camera are dynamically adjusted. The cameras are then photographed layer by layer from far end to near end according to the monitoring layers to ensure that the waterbirds have a pixel ratio that meets the recognition requirements in the photos, and that adjacent photos retain preset overlapping pixels. After preprocessing the captured photos, they are input into the trained waterbird recognition model for target detection and species identification. Results with recognition confidence reaching a preset threshold are selected. Based on the overlapping characteristics of the photos and preset deduplication rules, duplicate counts are removed, and the types and numbers of waterbirds are counted to generate a monitoring dataset. The model recognition accuracy is evaluated, and the model is iteratively optimized by supplementing training samples to adapt to the needs of new species recognition.

[0007] According to one example, the hardware parameters include the image sensor size, focal length range, and photo resolution of the zoom PTZ camera, and the monitoring and control parameters include the installation height of the PTZ camera, the minimum pixel ratio of water birds in the photo, and the overlapping pixel value of adjacent photos. When calculating the effective monitoring range of the PTZ camera, the maximum monitoring distance is first determined based on the hardware parameters and the monitoring control parameters. Then, the ground coverage area is calculated in combination with the installation height of the PTZ camera. If a large area needs to be covered, multiple PTZ cameras are deployed and a deduplication rule for the overlapping area of ​​adjacent PTZ cameras is planned.

[0008] According to one example, the deduplication rule for the overlapping area of ​​adjacent PTZ cameras includes calculating the length of the overlapping area in the direction of the line connecting the two PTZ cameras, deriving the overlap angle that the corresponding monitoring layer should ignore, and realizing the deduplication of the overlapping area based on the overlap angle during horizontal rotation shooting.

[0009] According to one example, the monitoring layer is divided with the starting pitch angle corresponding to the maximum monitoring distance of the PTZ camera as the far boundary and the preset ending pitch angle as the near boundary. The initial pitch angle is calculated based on the installation height of the PTZ camera and the maximum monitoring distance to ensure that the PTZ camera can cover the farthest monitoring area, and the final pitch angle is set to an angle value that allows waterbirds to maintain an effective identification pattern.

[0010] According to one example, during the layer-by-layer full-coverage shooting, before shooting each layer, the horizontal rotation angle is first calibrated to the reference angle, and then rotated one revolution according to the horizontal step angle to take a picture. Each rotation takes a picture and is numbered and stored according to preset rules. After completing the shooting of one layer, adjust the pitch angle to the next layer by vertical step angle, and repeat the shooting operation until the pitch angle reaches the termination pitch angle.

[0011] According to one example, when dynamically adjusting the focal length of the PTZ camera lens, the required focal length is calculated by combining the current elevation angle of the monitoring layer, the actual body shape characteristics of the target waterbird, and the image sensor parameters, so as to ensure that the pixel ratio of the waterbird in the photo meets the preset requirements. The horizontal step angle is determined based on the horizontal viewing angle corresponding to the current monitoring layer focal length, the photo size, and the overlapping pixel value of adjacent photos. The vertical step angle is determined based on the vertical viewing angle corresponding to the current monitoring layer focal length, the photo size, and the overlapping pixel value of adjacent photos.

[0012] According to one example, the preprocessing includes noise reduction and size normalization, the recognition model is a target detection model trained based on a deep learning framework, and the preset recognition confidence threshold is set according to the recognition accuracy requirements, and only the recognition results with confidence reaching the threshold are retained for quantity statistics.

[0013] According to one example, the deduplication rule is a rule for eliminating duplicate counts of adjacent photos on the same monitoring layer within a single PTZ camera, including: For adjacent photos in the same monitoring layer, only waterbirds whose target frame intersects with the preset middle area of ​​the photo are counted. The first and last photos of the monitoring layer respectively count waterbirds in the corresponding single-sided non-overlapping areas. Due to the time difference in shooting and the movement characteristics of waterbirds, waterbirds in the overlapping areas of vertically adjacent photos are counted as independent events.

[0014] As an example, the triggering conditions for the model iterative optimization include: The average recognition accuracy of a single species is lower than the preset accuracy threshold, or the number of waterbirds that cannot be identified has accumulated to a preset number; After being triggered, the model is optimized by automatically selecting high-confidence samples for incremental training and by manually labeling and expanding the training set.

[0015] A waterbird species identification system based on a variable magnification PTZ camera with dynamically adaptable distance, applied to the aforementioned identification method, includes: The zoom PTZ camera unit includes a pan-tilt module, a vertical tilt module, and a lens zoom module, and is equipped with an image sensor to adapt to the environmental conditions of waterbird monitoring scenarios and to perform automatic capture with full-area coverage. The control unit is used to receive the monitoring scene requirements and waterbird characteristic parameters, calculate the effective monitoring range of the PTZ camera, the division of monitoring layers and the parameter adjustment instructions for each layer, and drive the variable magnification PTZ camera to complete the capture operation. The data processing unit includes a preprocessing module, an identification and statistics module, and a model optimization module. The preprocessing module is used to perform image optimization processing on the captured photos. The identification and statistics module is used to call the waterbird identification model to perform target detection and species identification, and count the number of duplicates by combining deduplication rules. The model optimization module is used to periodically evaluate the model accuracy and trigger sample supplementation and model training. The storage and interaction unit is used to store original photos, monitoring datasets, and model files, and provides data retrieval and visualization.

[0016] The present invention has the following advantages: This invention deploys a high-definition zoom PTZ camera in flat wetlands such as islands and tidal flats. Based on the core functions of the PTZ camera's pan, tilt, and zoom capabilities, and considering key influencing factors such as the camera's installation height, sensor parameters, and the pixel proportion of the monitored object in the image, the maximum monitoring range is precisely calculated and divided into multiple monitoring layers. Full coverage of the monitoring area is achieved by rotating the camera horizontally around each layer. During shooting, the system automatically and dynamically adjusts the PTZ parameters to ensure that waterbirds occupy at least 300 pixels in the image, fully presenting key morphological features such as body shape, wings, feathers, and color. Simultaneously, by setting a 300-pixel overlap between adjacent images, the system avoids capturing only partial images of birds, ensuring that each waterbird is captured in its entirety. Combining the calculation results of various parameters, the system precisely sets the horizontal rotation step angle and vertical switching step angle for each layer, further improving the comprehensiveness and accuracy of the shooting.

[0017] After the captured photos are transmitted to the edge recognition device, the self-trained AI recognition model completes species identification and quantity statistics, ultimately forming a waterbird monitoring database for the entire area within a given time period. Simultaneously, the photos collected by this monitoring method can serve as a dataset, enabling autonomous labeling and model reinforcement training on the server for continuous iteration and improvement of recognition accuracy. This invention effectively replaces traditional methods such as manual surveys and infrared camera monitoring, completely solving the pain points of low automation, poor environmental adaptability, and lagging data statistics in traditional monitoring. It provides a more intelligent, comprehensive, efficient, and reliable technical solution, successfully achieving long-term, all-weather, automated intelligent monitoring of waterbirds. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for identifying waterbird species based on the dynamic adaptive distance of a variable magnification PTZ camera according to the present invention.

[0019] Figure 2 This is a system block diagram of the waterbird species identification method based on the dynamic adaptive distance of a variable magnification PTZ camera according to the present invention.

[0020] Figure 3 This is a schematic diagram illustrating the operation of the PTZ camera automatically capturing photos of waterbirds according to the present invention.

[0021] Figure 4 This is a schematic diagram illustrating the deduplication of overlapping parts in the complementary monitoring of two PTZ cameras according to the present invention.

[0022] Figure 5 This is a schematic diagram illustrating the deduplication of species counts between horizontally adjacent photos according to the present invention.

[0023] Figure 6 This is a system block diagram of the present invention for automatically capturing species photos with full-area coverage.

[0024] Figure 7 This is a system block diagram of species identification and model optimization according to the present invention. Detailed Implementation

[0025] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0026] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0027] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0028] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all promotional information and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0029] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily necessary for implementing this application, and therefore cannot be used to limit the scope of protection of this application.

[0030] Before detailing the embodiments of this application, some terms used in the embodiments of this application will be explained first, so that those skilled in the art can understand them.

[0031] A zoom PTZ camera is a high-definition monitoring device that has the functions of pan, tilt, and zoom. It can achieve target shooting at different distances and angles by adjusting the PTZ (pan-tilt-zoom) parameters, adapting to the waterbird monitoring needs of wetland scenes such as islands and mudflats. It is the core hardware device for achieving full-area coverage automatic capture in this application.

[0032] The effective monitoring range refers to the spatial range within which the PTZ camera can clearly capture waterbirds and meet the requirements for species identification, calculated based on the hardware parameters of the PTZ camera (such as image sensor size and focal length range) and monitoring and control parameters (such as installation height and minimum pixel ratio of waterbirds). This range includes the maximum straight-line distance from the PTZ camera to the monitoring point and the corresponding ground coverage area.

[0033] The monitoring layer refers to several virtual monitoring planes parallel to the ground that divide the effective monitoring range of the zoom PTZ camera from far to near. Each layer corresponds to a specific gimbal tilt angle and lens focal length. By shooting layer by layer, the effective monitoring range is fully covered, ensuring that waterbirds at different distances can be clearly captured.

[0034] PTZ parameters refer to the core control parameters of the pan-tilt head and lens of a zoom PTZ camera. Among them, P (Pan) refers to the horizontal rotation angle of the pan-tilt head, T (Tilt) refers to the vertical tilt angle of the pan-tilt head (0° in the horizontal direction and negative angle in the downward tilt direction), and Z (Zoom) refers to the zoom focal length of the lens. The coordinated adjustment of the three parameters can achieve dynamic adaptation of the PTZ camera's shooting angle and distance.

[0035] Preset overlapping pixels refer to the amount of pixel overlap set between adjacent photos (including horizontally adjacent photos and vertically adjacent photos of the same monitoring layer) to avoid incomplete waterbird images and ensure the accuracy of species identification. This setting ensures that waterbirds in the area can be captured completely and provides a basis for subsequent duplicate counting and removal.

[0036] Waterbird identification models refer to waterbird target detection models trained based on deep learning frameworks (such as YOLOv12). They can identify waterbird species and select target boxes in photos captured by zoom cameras, and output results including recognition confidence scores. The model can be iteratively optimized by supplementing training samples to adapt to the needs of identifying new species.

[0037] The deduplication rules refer to the two types of rules used in this application to remove duplicate counts of waterbirds. One is the deduplication rule within the same monitoring layer of a single PTZ camera, which counts the target frames of waterbirds in the middle area or one side of the non-overlapping area of ​​the photo. The other is the deduplication rule for overlapping areas between multiple PTZ cameras, which calculates the overlap length and angle to crop the shooting range. The two types of rules solve the problems of duplicate counts of photos within a single device and monitoring overlap between multiple devices, respectively.

[0038] Model iterative optimization refers to the process of improving the accuracy of a waterbird recognition model by supplementing training samples based on the actual recognition effect of the model. Triggering conditions include the average recognition accuracy of a single species being lower than a preset threshold and the cumulative number of unrecognizable waterbirds reaching a preset number. Optimization methods include incremental training with high-confidence samples and manual annotation to expand the training set.

[0039] Formula variable definition and explanation: The actual body length of the target waterbird is preset based on the common waterbird size and can be adjusted according to the monitoring object. It is mainly used to calculate the lens focal length that meets the recognition requirements, and the unit is m (meter). The preset value is 0.3m. The minimum pixel ratio required for photos of waterbirds taken with a PTZ camera to ensure that details of species morphology (body size, wings, feathers, color, etc.) can be extracted. The unit is pixels (px), and the default value is 300px. : The preset overlap of pixels between adjacent photos to avoid incomplete images of waterbirds and to provide a basis for subsequent duplicate counting and removal. The unit is pixels (px), and the preset value is 300px. The installation height of the variable magnification PTZ camera is determined based on the actual deployment of the monitoring scenario, and has an impact on the calculation of the maximum monitoring distance and elevation angle. The unit is m (meter). The photosensitive area width of the CMOS image sensor on the zoom PTZ camera is an inherent parameter of the camera hardware, used to calculate the horizontal angle of view and the shooting coverage area, and is measured in mm. The height of the photosensitive surface of the CMOS image sensor in the zoom PTZ camera is an inherent parameter of the camera hardware, used to calculate the vertical viewing angle and shooting coverage, and is measured in mm (millimeters). The width resolution of photos taken by the zoom PTZ camera is a preset parameter of the PTZ camera hardware. It can be adjusted as needed and affects the calculation of the overlap ratio of adjacent photos. The unit is pixels (px). The height resolution of photos taken by the zoom PTZ camera is a preset parameter of the PTZ camera hardware. It can be adjusted as needed and affects the calculation of the vertical shooting coverage and overlap ratio. The unit is pixels (px). The shortest focal length of a zoom PTZ camera is an inherent parameter of the camera's hardware and represents the lower limit of the zoom range, measured in mm (millimeters). The longest focal length of a zoom PTZ camera is an inherent parameter of the camera's hardware. It is the upper limit of the zoom range and determines the maximum monitoring distance of the camera. The unit is mm (millimeters). The effective monitoring coverage area of ​​a single variable magnification PTZ camera is calculated based on the maximum monitoring distance and the ground coverage radius, and the unit is m² (square meters). Gimbal tilt angle, defined as 0° in the horizontal direction, is negative when the lens tilts down and positive when it tilts up. It is used to adjust the vertical shooting angle, and the unit is ° (degree). The starting pitch angle of the PTZ camera is the far-end shooting angle corresponding to the maximum monitoring distance of the PTZ camera. It is calculated from the installation height of the PTZ camera and the maximum monitoring distance, and the unit is ° (degrees). The gimbal's final pitch angle corresponds to the shooting angle at the near end of the monitoring range. It is set to an angle that can effectively identify the waterbird's form and avoid blurry images due to excessive angle. The unit is ° (degrees). The vertical angle of the PTZ camera's viewpoint is calculated from the current focal length and the height of the CMOS sensor, determining the vertical shooting coverage area, and is expressed in degrees (°). : Horizontal rotation angle of the gimbal, defined as 0° as the reference calibration angle, is adjusted by rotating clockwise / counterclockwise along the horizontal direction to achieve full coverage shooting in the horizontal direction, and the unit is ° (degree). The horizontal angle of the PTZ camera's view is calculated from the current focal length and the width of the CMOS sensor. It determines the horizontal shooting coverage area and is used to calculate the horizontal step angle. The unit is ° (degree).

[0040] The waterbird species identification method and system based on dynamic adaptive distance of a zoom PTZ camera provided in this application relates to wild waterbird monitoring. The specific implementation is described in the embodiments below. In practical applications, if related variables in the calculation process are calculated together, or the calculation results of each step are retained to at least five decimal places to improve the accuracy of the results. This embodiment requires detailed step-by-step explanation, which may result in some loss of accuracy.

[0041] Reference Figure 1-3 This paper presents a waterbird species identification method based on a zoom PTZ camera with dynamic adaptive distance. The method includes a full-area coverage automatic image capture step and a species identification and quantity counting step. In the full-area coverage automatic capture step, the zoom PTZ camera first calculates the monitoring range to determine the number of layers and the PTZ (horizontal rotation, vertical tilt, lens zoom) transformation parameters for each layer. Then, starting from the farthest monitoring layer, it continuously captures images at a set horizontal rotation step angle until full coverage of that layer is achieved. Subsequently, it switches to the next layer by using a vertical tilt step angle, repeating the shooting process until all monitoring layers are covered. During the shooting process, it ensures that the waterbird's pixel proportion in the image is ≥300px and that adjacent images overlap by ≥300px, achieving full-area, blind-spot-free, and non-duplicative waterbird image acquisition. In the species identification and quantity counting step, after the captured images are stored, if used as a dataset, they are labeled and used for model training to iteratively optimize the identification model. Then, AI visual recognition technology is used to determine the waterbird species, and combined with preset deduplication rules, species and quantities are counted, ultimately achieving long-term, automated waterbird species identification and quantity monitoring.

[0042] S1. Steps for Automatic Species Photo Capture with Full Area Coverage Reference Figure 4-6 This step uses a zoom PTZ camera as the core hardware. Through a closed-loop process of parameter configuration, range calculation, layer division, dynamic parameter adjustment, and layer-by-layer shooting, it achieves full coverage, high definition, and non-duplication of waterbird images within the monitoring area, providing high-quality data support for subsequent species identification and population statistics.

[0043] (1) Parameter configuration and PTZ camera selection The selection of the PTZ camera is determined based on the actual body length of the waterbirds to be monitored, the on-site environment of the installation location, and the monitoring range. Based on the selection results, the hardware parameters and monitoring control parameters of the variable magnification PTZ camera are configured. Hardware parameters include the CMOS image sensor's width and height, focal length range, and maximum image resolution. Monitoring control parameters include the PTZ camera's installation height above the ground, the minimum pixel ratio required for waterbirds in the images, and the preset overlap pixel value between adjacent images. The PTZ camera selection is completed by combining the monitoring range and the on-site environment. If coverage of a large area is required, multiple PTZ cameras can be deployed to form a complementary monitoring network.

[0044] (2) Calculation of effective monitoring range Based on hardware parameters and monitoring and control parameters, the effective monitoring range of the PTZ camera is calculated sequentially. The formula for calculating the maximum monitoring distance (i.e., the straight-line distance from the PTZ camera lens to the farthest monitoring point, in meters) is as follows:

[0045] in, This refers to the actual body length of the waterbird (unit: m). This is the longest focal length of the PTZ camera. The height of the photo taken by the PTZ camera (in pixels). This is the minimum pixel percentage for waterbirds in photos taken with a PTZ camera (generally set to 300px, but can be customized depending on the monitored object). The height of the photosensitive surface of the CMOS image sensor for the PTZ camera (unit: mm).

[0046] In this embodiment, the actual body length of the waterbird is 0.3m, the longest focal length of the PTZ camera is 320mm, the height pixel value of the photo taken by the PTZ camera is 2160px, the minimum pixel ratio of the waterbird in the photo taken by the PTZ camera is 300px, and the height of the CMOS photosensitive surface is 5.4mm. Substituting these values ​​into the formula, the maximum monitoring distance can be obtained.

[0047] The maximum monitoring distance was ultimately determined to be 128 meters. This means that, under these parameters, the PTZ camera can clearly capture images of waterbirds with a body length of 0.3 meters at a maximum distance of 128 meters, and the bird's pixel ratio in the photo is no less than 300 pixels, which meets the requirements for morphological feature extraction for species identification.

[0048] The formula for calculating the ground coverage radius (i.e., the straight-line distance from the point where the PTZ camera is installed perpendicular to the ground to the farthest monitoring point, in meters) is:

[0049] in, This refers to the maximum monitoring distance of the PTZ camera (the straight-line distance from the PTZ camera to the monitoring point, in meters). Installation height of the PTZ camera (unit: m).

[0050] In this embodiment, the PTZ camera installation height is 10m, and the maximum monitoring distance is 128m. Substituting these values ​​into the formula, the ground coverage radius can be obtained:

[0051] With an installation height of 10m and a maximum monitoring distance of 128m, the farthest boundary of the PTZ camera's ground coverage area is 127.6 meters from the vertical projection point of the camera. Combined with the PTZ camera's 360° horizontal rotation capability, a circular monitoring area with the vertical projection of the camera's installation point as the center and a radius of 127.6 meters can be formed, providing a spatial range basis for subsequent monitoring layer division and full-area coverage shooting.

[0052] The formula for calculating the coverage area (unit: m²) monitored by a PTZ camera is as follows:

[0053] in, Ground coverage radius, which is the straight-line distance from the point where the PTZ camera is installed perpendicular to the ground to the farthest monitoring point, in meters.

[0054] In this embodiment, the ground coverage radius is 127.6 meters. Substituting this into the formula, the coverage area is 51,125 square meters, demonstrating the spatial coverage capability of a single PTZ camera.

[0055] Deduplication planning for overlapping areas of multi-PTZ cameras Reference Figure 4 When using multiple PTZ cameras for complementary monitoring of large wetlands (such as contiguous mudflats or large migratory bird habitats), overlapping areas between cameras are inevitable to ensure comprehensive coverage. Directly counting waterbirds repeatedly in these overlapping areas would lead to severe data distortion. Therefore, it is necessary to analyze and calculate the overlapping areas between pairs of cameras and crop the overlapping area of ​​one camera during the image capture process to remove duplicates. The specific implementation method involves the following two steps: First, calculate the length of the overlapping portion of each monitoring layer along the line connecting the two centers. For each monitoring layer, the length L (unit: m) of its overlapping area along the line connecting the two PTZ cameras must be calculated first. The calculation formula is as follows:

[0056] in, Installation height of the PTZ camera (unit: m). The distance between the two PTZ cameras (unit: m). The pitch angle of the PTZ camera is denoted by L. When the monitoring layer of the two PTZ cameras is L≤0, it means that the two PTZ cameras have no overlapping area within the current monitoring layer, and there is no need to consider deduplication.

[0057] In this embodiment, taking the calculation of the outermost monitoring layer as an example, the installation height of the PTZ camera is 10m, the distance between the two PTZ cameras is assumed to be 200m, and the pitch angle of the PTZ camera's pan-tilt unit at the outermost monitoring layer is -4.48°. Substituting into the formula, we can obtain L=55.26m.

[0058] Second, derive the angle to be ignored in the overlapping area and perform deduplication. After obtaining the length L of the overlapping part of each monitoring layer in the direction of the line connecting the two PTZ cameras, it is necessary to derive the overlapping angle α (unit: °) that should be ignored for that monitoring layer, and combine it with the zero-degree horizontal rotation angle when taking pictures to crop and remove duplicates in the overlapping area. The calculation formula is as follows:

[0059] Where L is the length of the overlapping portion of each monitoring layer along the line connecting the centers of the two circles, and R is the radius of each monitoring layer.

[0060] In this embodiment, with L = 55.26 meters and R = 127.6 meters, substituting into the formula yields α ≈ 76.84°. During actual shooting, when the PTZ camera rotates horizontally to the interval corresponding to the overlap angle α, it will automatically stop shooting in that area or mark the photos of that area as overlapping and requiring deduplication. Ultimately, only waterbirds in non-overlapping areas are counted, thereby avoiding duplicate counting among multiple PTZ cameras and ensuring the accuracy of waterbird population statistics.

[0061] (3) Monitoring layer division and pitch angle setting After calculating the effective monitoring range of the PTZ camera, the monitoring layers need to be divided based on the maximum monitoring distance and installation height, and the boundary values ​​of the PTZ tilt angle need to be set to provide an angle basis for subsequent full-coverage shooting layer by layer.

[0062] After calculating the maximum monitoring distance of the PTZ camera (i.e., the straight-line distance from the camera to the monitoring point, in meters), and combining this with the camera's installation height, the initial value of the pan-tilt angle of the PTZ unit when the camera is shooting the farthest area can be calculated. This initial value is the Tilt angle in the PTZ parameters, denoted as . Define the vertical pitch angle when the PTZ camera's field of view is completely parallel to the ground. 0° is considered a negative angle when the lens is tilted downwards, and a positive angle when it is tilted upwards.

[0063] The formula for calculating the initial pitch angle is:

[0064] in Installation height of the PTZ camera (unit: m). The maximum monitoring distance of the PTZ camera (unit: m).

[0065] In this embodiment, taking a PTZ camera installation height of 10 m and a maximum monitoring distance of 128 m as an example, substituting into the formula yields:

[0066] This angle marks the far boundary of the monitoring layer, ensuring that the PTZ camera can accurately cover the farthest waterbird monitoring area.

[0067] To ensure that key morphological features of waterbirds (body outline, wing texture, feather color, leg structure, beak shape, etc.) can be clearly captured, thereby achieving accurate species identification, a termination pitch angle (denoted as ) needs to be set. This serves as the near-end boundary of the monitoring layer. When the pitch angle exceeds -75°, waterbirds appear as near-dot images in photographs due to the close viewing angle, resulting in significant loss of key morphological features and a substantial decrease in species identification accuracy. Therefore, the termination pitch angle is generally set to a uniform value. =−75°, ensuring that the waterbirds captured always retain recognizable morphological details. Based on the initial pitch angle. With the final pitch angle The effective monitoring range of the PTZ camera can be divided into several monitoring layers along the vertical direction, with each layer corresponding to a specific pitch angle range, providing a spatial division basis for subsequent layer-by-layer adjustment of focus and horizontal shooting.

[0068] (4) Dynamically adjust focal length parameters After determining the pitch angle of the PTZ camera (obtained through the initial pitch angle or tiered pitch angles), it is necessary to calculate the lens focal length that meets the recognition requirements at the current angle, i.e., the Z (Zoom value) in the PTZ parameters, based on the PTZ camera's installation height, sensor hardware parameters, and the actual body length of the target waterbird. This focal length is denoted as Z. This ensures that the number of pixels occupied by waterbirds in the photo is not less than a preset threshold (300px), providing a clear morphological feature basis for species identification.

[0069] The formula for calculating focal length is as follows:

[0070] in, Installation height of the PTZ camera (unit: m). The current pan-tilt angle (unit: °) is the gimbal tilt angle of the current monitoring layer. This is the minimum number of pixels a waterbird occupies in a photo taken with a PTZ camera (generally set to 300 for waterbirds, unit: px). The height of the photosensitive surface of the CMOS image sensor for the PTZ camera (unit: mm). The actual body length of the monitored object (waterbird) is used (generally set to 0.3 for waterbirds, unit: m). The height of the captured photo (PTZ camera parameters, unit: px).

[0071] In this embodiment, the PTZ camera is installed at a height of 10 m, with a pitch angle of -30°, a minimum pixel ratio of 300 px for waterbirds, a CMOS sensor height of 5.4 mm, an actual waterbird length of 0.3 m, and a photo height pixel value of 2160 px.

[0072] Substitute into the formula to calculate:

[0073] The dynamic calculation of focal length using the formula ensures that waterbirds at different distances and angles maintain a sufficient pixel ratio in the photos, avoiding image blurring or feature loss due to unsuitable focal length, and providing high-quality input data for subsequent species identification models.

[0074] (5) Full coverage shooting layer by layer Reference Figure 5 After calculating the focal length parameters of the current monitoring layer, it is necessary to determine the horizontal viewing angle and vertical field of view by combining the parameters of the PTZ camera's CMOS image sensor. By controlling the horizontal rotation step angle and the vertical pitch step angle, it is possible to achieve blind spot shooting within the monitoring layer and switching between layers, while ensuring the overlap of photos and the integrity of waterbird images.

[0075] Focal length value based on monitoring layer Based on the width of the PTZ camera's CMOS image sensor, the horizontal angle of view of the PTZ camera at the current focal length is calculated (denoted as...). The perspective determines the horizontal coverage of a single photograph, calculated using the following formula:

[0076] in, Width of the photosensitive surface of the CMOS image sensor for PTZ cameras (unit: mm). The focal length is calculated in real time (unit: mm).

[0077] In this embodiment, the CMOS photosensitive surface width is 7.2 mm, and the focal length of the monitoring layer is 50 mm.

[0078] Substitute into the formula to calculate:

[0079] At the current focal length, the horizontal coverage angle of a single photo taken by the PTZ camera is approximately 8.25°.

[0080] To avoid image fragmentation of waterbirds (e.g., capturing only the head or feet), a 300-pixel overlap should be maintained between adjacent horizontally captured images (corresponding to the smallest pixel percentage of the waterbird). Consistency), ensuring the central area of ​​the photo is consistent during AI recognition ( The waterbird's form is complete (300px to the left and right). Based on the real-time horizontal viewpoint and photo width, the horizontal rotation step angle (denoted as ) is calculated. The formula is as follows:

[0081] in, The horizontal viewing angle is measured in degrees. Preset the overlap pixel value (unit: px, set to 300px) for adjacent photos. This refers to the width of the photo taken by the PTZ camera. The PTZ parameters control the degree of horizontal rotation for each layer. That is, P(Pan) = Starting from 0°, the horizontal rotation angle is used to take one photo at a time and save the photo to local storage until it overlaps with the first photo.

[0082] In this embodiment, the horizontal viewing angle is 8.25°, the overlapping pixel value is 300px, and the width of the photo taken by the PTZ camera is 3840px.

[0083] Substitute into the formula to calculate:

[0084] Before shooting, calibrate the horizontal rotation angle to the 0° reference angle to ensure a consistent starting position; starting from 0°, press [the button] each time. Rotate horizontally (approximately 7.6°), take a photo after rotating to the correct position, and store the photos according to the monitoring layer number and shooting sequence number. Repeat the rotation and shooting operation until the horizontal rotation angle overlaps with the first photo (to complete 360° full coverage). When counting, only retain the results of the intersection between the waterbird target frame and the middle area of ​​the photo (or the non-overlapping area on the right side of the first photo and the non-overlapping area on the left side of the last photo) to avoid double counting.

[0085] After completing the horizontal shooting of the current layer, the vertical field of view (denoted as ) is calculated based on the focal length and the height of the CMOS image sensor's photosensitive surface. This angle determines the vertical coverage area of ​​a single photo, and the calculation formula is as follows:

[0086] in, The height of the photosensitive surface of the CMOS image sensor for the PTZ camera (unit: mm). The focal length is calculated in real time (unit: mm).

[0087] In this embodiment, the height of the CMOS photosensitive surface is 5.4 mm, and the focal length of the current monitoring layer is 50 mm.

[0088] Substitute into the formula to calculate:

[0089] To ensure that no images of adjacent monitoring layers in the vertical direction are missed and that there is no excessive repetition, the vertical pitch step angle (denoted as ) is calculated by combining the vertical field of view, image height, and overlapping pixel values. The formula is as follows:

[0090] in, This is the real-time vertical field of view angle for the current layer (unit: °). Preset the overlap pixel value (unit: px, set to 300px) for adjacent photos. The height of the photo taken by the PTZ camera (unit: px).

[0091] In this embodiment, the real-time vertical field of view is 6.19°, the overlap pixel value is 300px, and the photo height is 2160px.

[0092] Substitute into the formula to calculate:

[0093] After the current layer is captured, press (Approximately 5.48°) Adjust the gimbal tilt angle downwards to obtain the next tilt angle (i.e., the current tilt angle); recalibrate the horizontal rotation angle to 0°, repeat the aforementioned focal length calculation and horizontal shooting operation in this step, and complete the next layer of full coverage shooting; cycle through the above layer switching and shooting process until the tilt angle exceeds the termination tilt angle of -75°, then stop all shooting operations and end one round of monitoring.

[0094] S2. Species identification and population statistics steps Reference Figure 7 Based on automatically captured waterbird photos covering the entire area, the system achieves accurate identification and population statistics of waterbird species through sample training, intelligent recognition, and autonomous iteration processes. It also has the ability to autonomously optimize the model and expand to new species.

[0095] (1) Training sample collection and model training In the early stages of PTZ camera installation, when machine learning samples are scarce, the PTZ camera can be used to collect species photos in a loop. Pre-processing such as initial screening and photo cropping using the YOLOv12 general model can save a lot of labor costs. After manual inspection, screening, and species labeling, a YOLOv12 training dataset is created to train the waterbird recognition model and obtain a relatively rough version of the recognition model, which is then used for species recognition based on the photo data collected by the PTZ camera.

[0096] YOLOv12 was used for training the species identification model and object detection. YOLOv12 is the first object detection framework in the YOLO series to fully integrate an attention mechanism. Through region attention, R-ELAN network and FlashAttention technology, it achieves breakthroughs in accuracy and speed, and can quickly identify species from a large number of photos taken by the PTZ camera to achieve real-time inference and statistical effects.

[0097] When the initial sample size is small, at least 200 photos of each waterbird species are required. This is because the initial recognition model is relatively coarse and does not have requirements regarding different postures or environments of the waterbirds. The number of training epochs is set to 100-300, generally starting from 100 epochs and gradually increasing to 300 epochs based on the performance on the validation set. At the same time, the mAP metric is closely monitored during the process to avoid overfitting or underfitting.

[0098] As the number of species photos collected in the later stages increased, the photos were divided into 24-hour time periods based on the time of shooting. At least 500 photos were extracted from each time period according to the proportion of each time period to reduce errors caused by different lighting and environmental conditions.

[0099] (2) Species identification and quantity statistics Species photos automatically captured across the entire area are numbered according to layer number and the shooting sequence number of each layer and saved to a designated photo storage database. During species identification, photos are extracted sequentially from the photo storage database, and a trained waterbird species model is used for inference to identify the waterbird species in the photos and record the pixel coordinates of target boxes with a species identification rate higher than 85% in the photos.

[0100] Because a 300-pixel overlap was set between adjacent photos during the capture process, only the waterbird target bounding boxes intersecting the middle portion (after subtracting 300 pixels from the left and right sides) of the photos in other photos, except for the first and last photos of each monitoring layer, were counted. This avoids double counting of species in overlapping areas of the photos, achieving accurate species counting.

[0101] A 300-pixel overlap was also set between vertically adjacent photos. However, since taking photos of each monitoring layer takes time and the waterbirds are moving, each species in the overlapping area of ​​vertically adjacent photos can be counted as an independent event in its own photo, with minimal impact on the final statistical results, and therefore no deduplication is required. Finally, the species photos and the counts are saved to a database to document the identification process and results.

[0102] (3) Machine autonomous training In the initial stage of equipment installation, after species identification, if the species recognition rate in the photos is below 85%, the program will be triggered to calculate the average recognition rate of a single species. If the recognition rate is below 70%, the machine will automatically select photos with a recognition rate above 85% from the recognition result database, merge the YOLOv12 recognition results and bounding boxes with the old model's dataset to generate a new dataset, and train a new recognition model until the recognition model has a stable recognition effect.

[0103] Furthermore, as the seasons change, the migration times of various migratory birds differ, and the PTZ camera will capture different species, including some previously unrecorded species, in different seasons and months. Since the AI ​​recognition model lacks support for identifying these species, it cannot recognize them. When the number of unidentifiable species exceeds 1000, the system will repeat the process of collecting training samples and training the model. After machine and manual screening, preprocessing, and target labeling, the system will expand and train the model based on the old one, and then deploy the new model to the recognition program to support the identification of new species.

[0104] The waterbird species identification system based on the dynamic adaptive distance of a zoom PTZ camera, applied to the aforementioned waterbird species identification method based on the dynamic adaptive distance of a zoom PTZ camera, mainly includes a zoom PTZ camera unit, a control unit, a data processing unit, and a storage and interaction unit. The zoom PTZ camera unit includes a pan-tilt module for horizontal rotation, a vertical tilt module, and a lens zoom module, and is equipped with an image sensor. The horizontal rotation module supports 360° rotation, the vertical tilt module can achieve tilt adjustment from 0° to -75°, and the lens zoom module covers a wide range of zoom capabilities from the shortest to the longest focal length, ensuring that high-definition photos of waterbirds meeting identification requirements can be captured at different distances and angles, completing the automatic capture task with full-area coverage.

[0105] The control unit receives the monitoring scene requirements and waterbird characteristic parameters, and reads the hardware parameters of the zoom PTZ camera unit, such as sensor size, focal length range, and image resolution. Based on these parameters, a preset algorithm sequentially completes the calculation of the effective monitoring range, the division of monitoring layers, and the generation of PTZ parameter adjustment instructions for each layer. The calculation of the effective monitoring range includes the maximum monitoring distance, ground coverage radius, and monitoring area; the division of monitoring layers determines the starting and ending pitch angles; and the PTZ parameter adjustment includes the real-time focal length, horizontal step angle, and vertical step angle. Drive signals are output to the zoom PTZ camera unit to control the horizontal rotation module, vertical pitch module, and lens zoom module to operate according to preset logic, synchronously coordinating the timing of image capture and the rhythm of data transmission to ensure an orderly and efficient capture process.

[0106] The data processing unit includes a preprocessing module, a recognition and statistics module, and a model optimization module. The preprocessing module optimizes the raw captured images transmitted by the PTZ camera unit, including noise reduction, size normalization, invalid image removal, and target region cropping, improving image quality and subsequent recognition efficiency. The recognition and statistics module uses a pre-trained waterbird recognition model (based on the YOLOv12 deep learning framework) to perform species identification and target detection, filtering valid results with recognition confidence levels reaching preset thresholds. It then combines preset deduplication rules to accurately count waterbird species and numbers, outputting structured monitoring data. The model optimization module periodically evaluates the recognition model's accuracy. When optimization conditions are triggered, it automatically filters high-confidence samples or triggers manual annotation, supplements the training dataset, and initiates iterative model training to continuously improve the model's recognition accuracy and its ability to adapt to new species.

[0107] The storage and interaction unit is used to store raw photos, monitoring datasets, and model files, providing data retrieval and visualization. It employs a distributed storage architecture, categorizing and storing raw captured photos (metadata such as shooting time, PTZ parameters, and monitoring layer number), structured monitoring datasets (species type, quantity, statistical time, monitoring area, etc.), and various versions of recognition model files, supporting long-term data storage and rapid retrieval. It provides a visual user interface and data display platform, allowing users to query historical monitoring data, view real-time monitoring footage, and export monitoring reports. It also offers an open model parameter configuration interface, allowing users to adjust key parameters such as recognition thresholds and training epochs according to changes in the monitoring scenario.

[0108] It should be noted that the above hardware design is merely a specific example illustrating the specific structure and appearance of the waterbird species identification system based on dynamic adaptive distance of a variable magnification PTZ camera provided by this invention in practical applications, the installation positions of each module, and the specific hardware selection in each module. The system's software settings are matched to the hardware design. The specific settings of the hardware and software designs can be adjusted according to actual needs; this embodiment does not limit this.

[0109] The waterbird species identification method and system proposed in this invention, based on the dynamic adaptive distance of a variable magnification PTZ camera, can be used to accurately identify waterbirds. By dynamically calculating parameters such as the PTZ camera's installation height, sensor parameters, and the pixel count of the monitored object in the photograph, high-definition bird photos are automatically captured, automatically identifying waterbird species and counting their numbers, thus achieving intelligent monitoring of waterbirds across the entire area.

[0110] The embodiments and examples presented herein are provided to best illustrate embodiments of this application and its particular applications, thereby enabling those skilled in the art to implement and use this application. However, those skilled in the art will understand that the above description and examples are provided for ease of illustration and example only. The descriptions presented are not intended to cover all aspects of this application or to limit this application to the precise forms disclosed.

Claims

1. A method for identifying waterbird species based on dynamic adaptive distance using a variable magnification PTZ camera, characterized in that, Includes the following steps: Based on the requirements of waterbird monitoring scenarios and the body shape characteristics of target waterbirds, the hardware parameters and monitoring control parameters of the zoom PTZ camera are configured, the effective monitoring range of the PTZ camera is calculated, and monitoring layers are divided. The monitoring layer division is based on the starting pitch angle corresponding to the maximum monitoring distance of the PTZ camera as the far boundary and the preset ending pitch angle as the near boundary. Starting from the farthest monitoring layer, the horizontal rotation angle, vertical pitch angle, and lens focal length of the PTZ camera are dynamically adjusted. The camera is cyclically photographed according to the set horizontal rotation step angle until the current monitoring layer is fully covered. Then, the camera switches to the next layer by the vertical pitch step angle. The camera is photographed layer by layer from far to near to ensure that the waterbirds have a sufficient pixel ratio in the photos to meet the recognition requirements, and that adjacent photos retain a preset overlapping pixel. After preprocessing the captured photos, they are input into the trained waterbird recognition model for target detection and species identification. Results with recognition confidence reaching a preset threshold are selected. Duplicate counts are removed based on photo overlap characteristics and preset deduplication rules. The deduplication rules include: for adjacent photos in the same monitoring layer, only waterbirds whose target boxes intersect with the preset middle area of ​​the photo are counted. The species and number of waterbirds are counted and a monitoring dataset is generated. The model's recognition accuracy is evaluated, and iterative optimization is achieved by supplementing training samples. The triggering conditions for model iterative optimization include: the average recognition accuracy of a single species is lower than a preset accuracy threshold, or the number of unrecognizable waterbirds reaches a preset number. After triggering, the model is optimized by automatically selecting high-confidence samples for incremental training and manually labeling to expand the training set, respectively, to adapt to the needs of new species recognition.

2. The identification method according to claim 1, characterized in that, The hardware parameters include the image sensor size, focal length range, and photo resolution of the zoom PTZ camera; the monitoring and control parameters include the installation height of the PTZ camera, the minimum pixel ratio of water birds in the photo, and the overlapping pixel value of adjacent photos. When calculating the effective monitoring range of the PTZ camera, the maximum monitoring distance is first determined based on the hardware parameters and the monitoring control parameters. Then, the ground coverage area is calculated in combination with the installation height of the PTZ camera. If a large area needs to be covered, multiple PTZ cameras are deployed and a deduplication rule for the overlapping area of ​​adjacent PTZ cameras is planned.

3. The identification method according to claim 2, characterized in that, The deduplication rule for the overlapping area of ​​adjacent PTZ cameras includes calculating the length of the overlapping area in the direction of the line connecting the two PTZ cameras, deriving the overlap angle that the corresponding monitoring layer needs to ignore, and realizing the deduplication of the overlapping area based on the overlap angle during horizontal rotation shooting.

4. The identification method according to claim 2, characterized in that, The initial pitch angle is calculated based on the installation height of the PTZ camera and the maximum monitoring distance to ensure that the PTZ camera can cover the farthest monitoring area, and the final pitch angle is set to an angle value that allows waterbirds to maintain an effective identification pattern.

5. The identification method according to claim 1, characterized in that, During the layer-by-layer full-coverage shooting, before shooting each layer, the horizontal rotation angle is first calibrated to the reference angle, and then rotated one revolution according to the horizontal step angle to take a picture. Each rotation takes a picture and is numbered and stored according to preset rules. After completing the shooting of one layer, adjust the pitch angle to the next layer by vertical step angle, and repeat the shooting operation until the pitch angle reaches the termination pitch angle.

6. The identification method according to claim 5, characterized in that, When dynamically adjusting the focal length of the PTZ camera lens, the required focal length is calculated by combining the current elevation angle of the monitoring layer, the actual body shape characteristics of the target waterbird, and the image sensor parameters to ensure that the pixel ratio of the waterbird in the photo meets the preset requirements. The horizontal step angle is determined based on the horizontal viewing angle corresponding to the current monitoring layer focal length, the photo size, and the overlapping pixel value of adjacent photos. The vertical step angle is determined based on the vertical viewing angle corresponding to the current monitoring layer focal length, the photo size, and the overlapping pixel value of adjacent photos.

7. The identification method according to claim 1, characterized in that, The preprocessing includes noise reduction and size normalization. The recognition model is a target detection model trained based on a deep learning framework. The preset threshold is set according to the recognition accuracy requirements, and only the recognition results with confidence levels reaching the threshold are retained for statistical purposes.

8. The identification method according to claim 1, characterized in that, The deduplication rule is the rule for counting and removing duplicate images of adjacent images on the same monitoring layer within a single PTZ camera, including: The first and last photos of the monitoring layer respectively count the waterbirds in the corresponding single-sided non-overlapping areas; Due to the time difference in shooting and the movement characteristics of waterbirds, waterbirds in the overlapping areas of vertically adjacent photos are counted as independent events.

9. A waterbird species identification system based on a variable magnification PTZ camera with dynamically adaptable distance, characterized in that, The identification method applied to any one of claims 1-8 includes: The zoom PTZ camera unit includes a pan-tilt module, a vertical tilt module, and a lens zoom module, and is equipped with an image sensor to adapt to the environmental conditions of waterbird monitoring scenarios and to perform automatic capture with full-area coverage. The control unit is used to receive the monitoring scene requirements and waterbird characteristic parameters, calculate the effective monitoring range of the PTZ camera, the division of monitoring layers and the parameter adjustment instructions for each layer, and drive the variable magnification PTZ camera to complete the capture operation. The data processing unit includes a preprocessing module, an identification and statistics module, and a model optimization module. The preprocessing module is used to perform image optimization processing on the captured photos. The identification and statistics module is used to call the waterbird identification model to perform target detection and species identification, and count the number of duplicates by combining deduplication rules. The model optimization module is used to periodically evaluate the model accuracy and trigger sample supplementation and model training. The storage and interaction unit is used to store original photos, monitoring datasets, and model files, and provides data retrieval and visualization.

Images