A raptor identification algorithm and a multifunctional raptor stand

Abstract

This invention proposes a raptor recognition algorithm, comprising the following steps: S001: acquiring raptor video and weight data; S002: scaling, normalizing, and data augmenting the acquired raptor images; normalizing the acquired raptor weight data and pairing it with the raptor image data to form a multimodal dataset; S003: selecting and training a model based on the raptor image and weight data; S004: performing feature fusion and classification on the raptor image and weight data; S005: outputting the results; S006: deploying the trained model in practice. This invention also proposes a multifunctional raptor cage for implementing the aforementioned raptor recognition algorithm.

Description

Technical Field

[0001] This invention relates to the field of grassland rodent control technology, and in particular to a raptor identification algorithm and a multifunctional raptor cage. Background Technology

[0002] Rodent infestation is a major biological disaster affecting grasslands, leading to forage loss, grassland degradation, and the spread of various zoonotic diseases such as plague. Controlling grassland rodent infestations is essential to maintaining grassland ecological balance and ensuring food security, ecological security, and public health. Raptor perches are a green pest control measure that utilizes the natural predators of rodents. The principle is to leverage the lack of trees on grasslands, which leaves raptors with limited habitats. By establishing raptor perches at a certain density in rodent-affected areas, more habitats are provided for raptors, increasing their range of activity on the grassland and thus improving their predation rate on grassland rodents, thereby achieving long-term rodent control. This method has been promoted in Xinjiang, Qinghai, Sichuan, and Inner Mongolia in my country. However, since raptor cages are generally located in remote grassland areas, long-term monitoring is inconvenient. Moreover, during the dry season in grasslands, raptors may not venture into water-scarce areas, which is also detrimental to the control of grassland rodents. Therefore, the actual utilization effect of raptor cages, as well as the relevant information such as the species, number, and weight of raptors, lack sufficient evaluation, and their actual rodent control ability is unclear. Traditional bird identification methods rely on manual classification, which is both time-consuming and prone to errors. With the development of deep learning technology, vision-based bird identification methods have made significant progress. However, relying solely on visual recognition still has certain limitations in complex environments. This method, based on the weight monitoring information obtained by the raptor cage device, can provide additional identification features and improve identification accuracy. Summary of the Invention

[0003] Therefore, there is a need to provide a raptor recognition algorithm and a multifunctional raptor cage to overcome the above problems or at least partially solve or alleviate them.

[0004] To achieve the above objectives, this invention proposes a raptor recognition algorithm, comprising the following steps:

[0005] S001: Acquire video and weight data of the raptor;

[0006] S002: The acquired images of raptors are scaled, normalized, and augmented. The weight data of the acquired raptors is normalized and paired with the image data of the raptors to form a multimodal dataset.

[0007] S003: Model selection and training based on image and weight data of raptors;

[0008] S004: Perform data preprocessing on the image data in the multimodal data, input the preprocessed data into the raptor recognition model, and output raptor species 1 and raptor location information; input the weight data in the multimodal dataset and the raptor location information into the selected raptor classification model, and output raptor species 2; perform scoring calculation based on raptor species 1 and raptor species 2 to obtain the calibrated raptor species;

[0009] S005: Model results output;

[0010] S006: Deploy the trained model in practice.

[0011] In step S003, the total loss function for model training is obtained by weighted summation of classification loss, localization loss, and confidence loss. The localization loss function uses the CIoU method, and the calculation formula is as follows:

[0012] Localization loss function:

[0013] Intersection over Union (IoU) of predicted bounding boxes and ground truth bounding boxes:

[0014] Trade-off parameters:

[0015] Measuring consistency:

[0016] Where A is the predicted bounding box, B is the true bounding box, d is the distance between the center points of the two bounding boxes, c is the length of the maximum diagonal of the minimum bounding rectangle of the two bounding boxes, and w, h, wgt and hgt are the width and height of the predicted bounding box and the true bounding box, respectively.

[0017] In step S004, the scores of different categories obtained from the two models are summed to obtain the calibrated raptor species probability. The raptor species with the highest probability is selected as the final raptor species. The scoring formula is as follows:

[0018]

[0019] Where i represents different species of raptors, Ti represents the final calibration score predicted by the fusion method for a certain species of raptor, Ri represents the raptor species score predicted by the identification model, and Ci represents the raptor species score predicted by the classification model.

[0020] This invention also proposes a multifunctional raptor cage, including the raptor recognition algorithm described above, and further comprising:

[0021] A pole, the top of which is equipped with a fixed pulley;

[0022] A support is provided with a collar on one side, which is fitted onto the upright. A lifting ring is provided at the top of the support, and a connecting rope is connected to the lifting ring. The connecting rope passes around the fixed pulley and leads to the ground. A support platform is provided at the bottom of the support.

[0023] A horizontal bar is provided on the support platform, with both ends of the horizontal bar extending out of the support platform, and a weight sensor is provided between the horizontal bar and the support platform;

[0024] A water tank, the lower part of which is provided with a water trough and a solenoid valve seat, the solenoid valve seat being connected to the water tank and the water trough;

[0025] A camera is mounted on the support and faces both ends of the horizontal bar.

[0026] The controller includes a processor, a battery module, a memory, and a communication module. One signal receiving terminal of the controller is electrically connected to the weight sensor, and the two signal output terminals of the controller are electrically connected to the solenoid valve seat and the camera.

[0027] The multifunctional raptor frame of the present invention also has the following features.

[0028] It also includes a solar panel, which is disposed on the top of the support and is electrically connected to the battery module via a charging circuit.

[0029] It also includes bird spikes, which are installed at the edges of the solar panel or at the joints on both sides and in the middle of the pole.

[0030] It also includes a GPS module, which is connected to the communication module in the controller via a circuit.

[0031] It also includes a proximity switch or an ultrasonic ranging device, which is mounted on the support or the horizontal bar.

[0032] It also includes a chip sensor, which is disposed at the lower part of the support.

[0033] The raptor recognition algorithm of this invention achieves efficient and accurate bird recognition by combining an improved lightweight YOLOv5s model and a random forest classification model. It not only improves the accuracy and real-time performance of grassland eagle recognition, but also provides strong technical support for bird conservation and ecological research.

[0034] The multifunctional raptor frame of this invention mainly features a height-adjustable support on the upright pole, which facilitates maintenance of the raptor frame by staff on the ground without affecting the height of the upright pole. This attracts raptors to land. The support is equipped with a camera and weight sensor to photograph and weigh the raptors that land on the horizontal pole, monitoring the species, number, and size of the raptors. It can also attract raptors by supplying water through a water tank, allowing them to enter water-scarce areas of the grassland even during the dry season, which is beneficial for the control of grassland rodents. Attached Figure Description

[0035] Figure 1 This is a flowchart of the raptor recognition algorithm of the present invention;

[0036] Figure 2 This is a schematic diagram of the algorithm structure of the raptor recognition algorithm of the present invention;

[0037] Figure 3 This is a diagram of the YOLOv5s-MobileNetV3 network architecture.

[0038] Figure 4 This is a diagram illustrating the recognition effect of the raptor recognition algorithm of the present invention.

[0039] Figure 5 This is a structural diagram of the multifunctional raptor frame of the present invention from one perspective;

[0040] Figure 6 This is a structural diagram of the multifunctional raptor frame of the present invention from another perspective.

[0041] In the above attached figures:

[0042] 1. Upright pole; 2. Fixed pulley; 3. Support; 301. Support platform; 302. Connecting seat; 303. Top frame; 4. Ring; 5. Lifting ring; 6. Horizontal bar; 7. Weight sensor; 8. Water tank; 9. Water trough; 10. Solenoid valve seat; 11. Camera; 12. Controller; 13. Solar panel; 14. GPS module; 15. Ultrasonic ranging device; 16. Chip sensor; 17. Bird spikes. Detailed Implementation

[0043] To explain in detail the technical content, structural features, objectives, and effects of the technical solution, the following description is provided in conjunction with specific embodiments and accompanying drawings.

[0044] like Figure 1 and Figure 2 As shown, an embodiment of the present invention proposes a raptor recognition algorithm, including the following steps:

[0045] S001: Use a multi-functional raptor rack to acquire video and weight data of raptors; collect image data and weight data of prairie eagles under different lighting conditions at different times through camera 11 on pole 1, and acquire the weight data of raptors through weight sensor 7 installed at the bottom of horizontal pole 6.

[0046] S002: The acquired raptor images are scaled, normalized, and augmented. The weight data of the acquired raptors is normalized and paired with the image data to form a multimodal dataset.

[0047] S003: Model selection and training based on image and weight data of raptors;

[0048] like Figure 2 and Figure 3 As shown, to further reduce computational load and save energy, the visual recognition model uses the lightweight YOLOv5s model, and replaces the original CSPDarkNet backbone network in YOLOv5s with MobileNetV3. Through transfer learning, the MobileNetV3 model, pre-trained on ImageNet, is fine-tuned on a bird dataset. After training, the algorithm is deployed on an added edge computing device, enabling rapid and accurate monitoring of wild raptors.

[0049] In step S003, the total loss function for model training is obtained by weighted summation of classification loss, localization loss, and confidence loss. The localization loss function uses the CIoU method, and the calculation formula is as follows:

[0050] Localization loss function: The quality of model training is determined and evaluated by minimizing the localization loss function. Generally, the smaller the localization loss function, the better the predicted bounding box for locating the target matches the ground truth bounding box.

[0051] Intersection over Union (IoU) of predicted bounding boxes and ground truth bounding boxes: Calculate the intersection-union ratio (IUGR) of the predicted bounding box and the ground truth bounding box, which is the ratio of the intersection area to the union area of ​​the two target boxes;

[0052] Trade-off parameters: Used to define the percentage of consistency;

[0053] Measuring consistency: Measure the consistency of the aspect ratio of two target boxes;

[0054] Where A is the predicted bounding box, B is the true bounding box, d is the distance between the center points of the two bounding boxes, c is the length of the maximum diagonal of the minimum bounding rectangle of the two bounding boxes, and w, h, wgt and hgt are the width and height of the predicted bounding box and the true bounding box, respectively.

[0055] S004: Perform data preprocessing on the image data in the multimodal data, input the preprocessed data into the raptor recognition model, and output raptor species 1 and raptor location information; input the weight data in the multimodal dataset and the raptor location information into the selected raptor classification model, and output raptor species 2; perform scoring calculation based on raptor species 1 and raptor species 2 to obtain the calibrated raptor species;

[0056] In step S004, the scale data and weight features of the raptors identified in the image are fused, and the Random Forest (RF) algorithm is used for joint feature extraction and classification to improve the robustness and accuracy of the model. The raptor identification model and the raptor classification model are assigned scores in a 1:1 ratio. The scores from the two models for different categories are summed to obtain the calibrated raptor species probability. The raptor species with the highest probability is selected as the final raptor species. The scoring formula is as follows:

[0057]

[0058] Where i represents different species of raptors, Ti represents the final calibration score predicted by the fusion method for a certain species of raptor, Ri represents the raptor species score predicted by the identification model, and Ci represents the raptor species score predicted by the classification model.

[0059] S005: Model result output; The results obtained after multimodal feature fusion detection are raptor species, raptor location, raptor weight, and raptor size. Among them, raptor species is the calibrated category, raptor location is the location information obtained by the raptor recognition model, raptor weight is obtained by an electronic scale, and raptor size is the diagonal length of the raptor location rectangle.

[0060] S006: Deploy the trained model in practice. The trained model will be deployed to a real-world application, ultimately being used on a hawk-catching platform to achieve real-time recognition of steppe eagles. The recognition results will be displayed through a front-end interface, providing relevant bird information to assist researchers and birdwatchers in bird identification. The recognition results are as follows: Figure 4 As shown.

[0061] refer to Figure 5 and Figure 6The present invention also proposes a multifunctional raptor frame for implementing the raptor recognition algorithm described above. The multifunctional raptor frame includes: a vertical pole 1, a support 3, a horizontal bar 6, a water tank 8, a camera 11, and a controller 12. A fixed pulley 2 is provided at the top of the vertical pole 1. A collar 4 is provided on one side of the support 3, and the collar 4 is fitted onto the vertical pole 1. A hanging ring 5 is provided at the top of the support 3, and a connecting rope is connected to the hanging ring 5. The connecting rope passes around the fixed pulley 2 and leads to the ground. A support platform 301 is provided at the lower part of the support 3. The horizontal bar 6 is mounted on the support platform 301. The horizontal bar 6 extends outward from the support platform 301 at both ends, and a weight sensor 7 is installed between the horizontal bar 6 and the support platform 301; a water tank 9 and a solenoid valve seat 10 are installed at the lower part of the water tank 8, and the solenoid valve seat 10 is connected to the water tank 8 and the water tank 9; two cameras 11 are installed on the support 3, and the two cameras 11 face the two ends of the horizontal bar 6 respectively; the controller 12 is equipped with a processor, a battery module, a memory and a communication module, one signal receiving end of the controller 12 is electrically connected to the weight sensor 7, and the two signal output ends of the controller 12 are electrically connected to the solenoid valve seat 10 and the camera 11.

[0062] The upright pole 1 can be made of wood or metal and is erected on the ground. A fixed pulley 2 is installed at the top of the upright pole 1. The support 3 is a vertical metal plate or wooden board. A collar 4 is installed on the back of the support 3. The collar 4 is slidably fitted onto the upright pole 1. A lifting ring 5 is installed at the top of the support 3. The lifting ring 5 is connected to a cable that goes around the fixed pulley 2 and is wound and anchored to the ground. By unwinding the wound cable, the support 3 can be lowered from the top of the upright pole 1 to the ground through the fixed pulley 2, which makes it convenient for staff to maintain the raptor frame.

[0063] A horizontal support platform 301 is provided at the lower end of the support 3. A connecting seat 302 is horizontally provided on the support platform 301. The middle part of the horizontal bar 6 is fixed in the connecting seat 302. A weight sensor 7 is provided between the lower part of the connecting seat 302 and the support platform 301. When the raptor lands on the horizontal bar 6, the weight sensor 7 will automatically weigh the raptor.

[0064] One side of the solenoid valve seat 10 is fixedly connected to the support 3. The upper port of the solenoid valve seat 10 is connected to the water tank 8. A water trough 9 is provided on one side of the solenoid valve seat 10. When the solenoid valve seat 10 is energized, it can release water from the water tank 8 into the water trough 9 for the birds of prey to drink.

[0065] The support 3 is also equipped with a top frame 303, and the camera 11 is located on the lower side of the top frame 303, which can be used to film the raptors that fall on the horizontal bar 6.

[0066] The controller 12 is fixed inside the support 3. The controller 12 contains a processor, a battery module, a memory, and a communication module. When the raptor lands on the horizontal bar 6, the controller 12 receives the weight measurement signal from the weight sensor 7 and simultaneously controls the camera 11 to film the raptor. The controller 12 also opens the electromagnetic switch in the solenoid valve seat 10 to fill the water tank 9 for the raptor to drink. The processor in the controller 12 processes the received weight measurement signal and video signal and stores the video information in the memory. Finally, the weight information and video information of the raptor are sent to the network through the communication module, and the staff can receive them through the communication equipment.

[0067] like Figure 5 and Figure 6 According to one embodiment of the present invention, a solar panel 13 is also included. The solar panel 13 is disposed on the top of the support 3 and is electrically connected to the battery module through a charging circuit.

[0068] The solar panel 13 is fixed above the top frame 303. The solar panel 13 is electrically connected to the battery module through a charging circuit, which can charge the battery and improve the raptor frame's endurance.

[0069] like Figure 5 and Figure 6 It also includes bird spikes 17, which are installed at the edges of the solar panel 13 or at the joints on both sides and in the middle of the pole 1.

[0070] Bird spikes prevent birds of prey from landing directly on the solar panel 13, thus preventing the collection of bird data, and instead allow birds of prey to land on the horizontal bar 6.

[0071] refer to Figure 5 and Figure 6 According to one embodiment of the present invention, it further includes a GPS module 14, which is connected to the communication module in the controller 12 via a circuit.

[0072] The GPS module 14 is electrically connected to the communication module, enabling wireless signal transmission and facilitating remote reception of image and video data by staff.

[0073] refer to Figure 5 and Figure 6 According to one embodiment of the present invention, it further includes a proximity switch or an ultrasonic ranging device 15, wherein the proximity switch 15 is disposed on the support 3 or the horizontal bar 6.

[0074] To save power, a proximity switch or ultrasonic ranging device 15 can be set up with the detection end of the proximity switch or ultrasonic ranging device 15 facing both ends of the horizontal bar 6. When the raptor lands on the horizontal bar 6, the proximity switch or ultrasonic ranging device 15 is triggered. At this time, the controller 12 starts the camera 11 to take pictures of the raptor and at the same time starts the weight sensor 7 to measure the weight of the raptor.

[0075] like Figure 5 and Figure 6 According to one embodiment of the present invention, a chip sensor 16 is also included, which is disposed at the lower part of the support 3.

[0076] After a raptor is fitted with a sensor chip and released, when the raptor lands on the horizontal bar 6, the chip sensor 16 can detect the chip fixed to the raptor and thus immediately identify the raptor's information.

[0077] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising..." or "including..." does not exclude the presence of additional elements in the process, method, article, or terminal device that includes said element. Additionally, in this document, "greater than," "less than," "exceeding," etc., are understood to exclude the stated number; "above," "below," "within," etc., are understood to include the stated number.

[0078] Although the above embodiments have been described, those skilled in the art, once they understand the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the above descriptions are merely embodiments of the present invention and do not limit the scope of patent protection of the present invention. Any equivalent structural or procedural transformations made using the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A raptor recognition algorithm, characterized in that, Includes the following steps: S001: Acquire video and weight data of the raptor; S002: The acquired raptor images are scaled, normalized, and augmented. The weight data of the acquired raptors is normalized and paired with the image data of the raptors to form a multimodal dataset. S003: Model selection and training based on image and weight data of raptors; S004: Perform data preprocessing on the image data in the multimodal data, input the preprocessed data into the raptor recognition model, and output raptor species 1 and raptor location information; input the weight data in the multimodal dataset and the raptor location information into the selected raptor classification model, and output raptor species 2; Based on the scoring calculations of raptor species 1 and raptor species 2, the calibrated raptor species are obtained; S005: Model results output; S006: Deploy the trained model in practice.

2. The raptor recognition algorithm according to claim 1, characterized in that, In step S003, the total loss function for model training is obtained by weighted summation of classification loss, localization loss, and confidence loss. The localization loss function uses the CIoU method, and the calculation formula is as follows: Localization loss function Intersection over Union (IoU) of predicted bounding boxes and ground truth bounding boxes: Trade-off parameters: Measuring consistency: Where A is the predicted bounding box, B is the true bounding box, d is the distance between the center points of the two bounding boxes, c is the length of the maximum diagonal of the minimum bounding rectangle of the two bounding boxes, and w, h, and w gt and h gt These represent the width and height of the predicted bounding box and the ground truth bounding box, respectively.

3. The raptor recognition algorithm according to claim 2, characterized in that, In step S004, the scores of different categories obtained from the two models are summed to obtain the calibrated raptor species probability. The raptor species with the highest probability is selected as the final raptor species. The scoring formula is as follows: Where i represents different species of birds of prey, and T i To predict the final calibration score for a certain raptor species using the fusion method, R i To identify the model that predicts raptor species scores, C i Predict raptor species scores for the classification model.

4. A multifunctional raptor cage, comprising the raptor recognition algorithm as described in any one of claims 1 to 3, characterized in that, Also includes: A pole (1) is provided with a fixed pulley (2) at the top of the pole (1); Support (3); a collar (4) is provided on one side of the support (3), the collar (4) is fitted onto the upright (1), a lifting ring (5) is provided at the top of the support (3), a connecting rope is connected to the lifting ring (5), the connecting rope passes around the fixed pulley (2) and leads to the ground, and a support platform (301) is provided at the lower part of the support (3). A horizontal rod (6) is provided on the support platform (301), with both ends of the horizontal rod (6) extending out of the support platform (301), and a weight sensor (7) is provided between the horizontal rod (6) and the support platform (301); Water tank (8), the lower part of which is provided with water tank (9) and solenoid valve seat (10), the solenoid valve seat (10) being connected to the water tank (8) and the water tank (9); A camera (11) is mounted on the support (3) and faces both ends of the horizontal bar (6); The controller (12) is equipped with a processor, a battery module, a memory and a communication module. One signal receiving end of the controller (12) is electrically connected to the weight sensor (7), and the two signal output ends of the controller (12) are electrically connected to the solenoid valve seat (10) and the camera (11).

5. The multifunctional raptor cage as described in claim 4, characterized in that, It also includes a solar panel (13), which is disposed on the top of the support (3) and is connected to the battery module circuit via a charging circuit.

6. The multifunctional raptor cage as described in claim 4, characterized in that, It also includes bird spikes (17), which are provided at the edge of the solar panel (13) or at the joints on both sides and in the middle of the pole (1).

7. The multifunctional raptor cage as described in claim 4, characterized in that, It also includes a GPS module (14), which is connected to the communication module in the controller (12) via a circuit.

8. The multifunctional raptor cage as described in claim 4, characterized in that, It also includes a proximity switch, which or an ultrasonic ranging device (15) is mounted on the support (3) or the horizontal bar (6).

9. The multifunctional raptor cage as described in claim 4, characterized in that, It also includes a chip sensor (16), which is disposed at the lower part of the support (3).

Images