Wild animal identification and species group quantity statistical method and system based on unmanned aerial vehicle shot image processing

Abstract

The invention provides a wildlife identification and species group quantity statistical method and system based on unmanned aerial vehicle shot image processing, relates to the technical field of wildlife identification, and remarkably improves the accuracy of identification and statistics by fusing image identification and sound identification results and dynamically adjusting the shooting parameters of an unmanned aerial vehicle. The YOLOv5 model, the SORT algorithm, the deep learning model and the short-time Fourier transform are combined to carry out image recognition and sound recognition, and the image recognition result and the sound recognition result are combined, so that the false alarm and the missing alarm are effectively reduced, and the consistency and the reliability of the data are improved. And the shooting parameters of the unmanned aerial vehicle can be dynamically adjusted to find a balance point between the regional resolution and the coverage range, so that the system can effectively work in various environments. Therefore, the recognition capability and the data quality of the system can be improved, the robustness and the universality of the system are enhanced, and the application effect and the practical value of the unmanned aerial vehicle in wild animal monitoring are improved.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a method and system for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles. Background Art

[0002] With the development of drone technology, the use of drones for wildlife monitoring has become increasingly common. Traditional methods mainly rely on single image recognition or sound recognition, which often have problems such as low recognition accuracy, high underreporting and false alarm rates when dealing with complex environments and diverse wildlife species. In addition, single-modality recognition methods are difficult to adapt to different wildlife distribution conditions and environmental changes, and often require frequent adjustments to drone shooting parameters, increasing the workload and complexity. Existing technologies are difficult to ensure high-resolution images while covering a larger area, thereby improving the comprehensiveness and accuracy of data, resulting in limited accuracy in subsequent data analysis and decision-making.

[0003] For example, the method for investigating the population size of wild animals based on image processing taken by drones, as disclosed in Chinese patent application number CN202311830191.X, analyzes aerial images taken by drones to investigate the population size of wild animals in a region. However, it is still a single image recognition method, which is difficult to adapt to different scenarios and reduces the final recognition accuracy.

[0004] Therefore, how to effectively integrate multimodal data in wildlife monitoring, dynamically adjust shooting parameters, and improve the accuracy of recognition and statistics has become a technical problem that needs to be solved urgently.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles, so as to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] Wildlife identification and species population statistics based on drone image processing, the specific steps include:

[0009] S1: collect an image of wild animals to generate a reference image, and use an image processing algorithm to process the reference image to generate a target ratio coefficient for wild animals;

[0010] S2: Adjust the shooting parameters of the drone based on the target ratio coefficient. After the adjustment is completed, perform secondary image and sound collection on the wild animals to generate analysis pictures and analysis sound sources;

[0011] S3: Use the image processing algorithm to process the analysis image to generate an image recognition result, and then use the sound processing algorithm to process the analysis sound source to generate a sound recognition result;

[0012] S4: Perform set division based on the image recognition result and the sound recognition result, and use the intersection of the image recognition result and the sound recognition result as the final recognition result.

[0013] Preferably, the generation logic of the target proportion coefficient is:

[0014] The image processing algorithm uses the Yolov5 model, collects training pictures containing various types of wild animals, divides the training pictures into a training set and a validation set in a ratio of 8:2, and uses the training set and validation set to train and optimize the Yolov5 model respectively;

[0015] Use the optimized Yolov5 model to perform image recognition on the wild animals to be identified in the reference image, and generate the identification area corresponding to each wild animal individual;

[0016] Calculate the average area of ​​the identified regions in the reference image as follows:

[0017]

[0018] In the formula represents the average area, S k represents the area of ​​the kth recognition area, the subscript k represents the number of the recognition area, and n represents the number of recognition areas;

[0019] The target proportion coefficient is calculated based on the average area of ​​the recognition area. The calculation method is: In the formula represents the target proportion coefficient, and Sz represents the total area of ​​the reference image.

[0020] Preferably, the shooting parameters include shooting height and shooting angle of view, and the logic for adjusting the shooting parameters is:

[0021] Compare the target proportion coefficient with a preset proportion threshold, generate a first adjustment coefficient according to the comparison result, and then generate a second adjustment coefficient according to the number of recognition areas;

[0022] The first adjustment coefficient and the second adjustment coefficient are weighted to generate a total adjustment coefficient, and the total adjustment coefficient is used to adjust the shooting parameters.

[0023] Preferably, the first adjustment coefficient and the second adjustment coefficient are calculated as follows:

[0024]

[0025] ω2=ln(n+1)

[0026] Where ω1 and ω2 represent the first adjustment coefficient and the second adjustment coefficient respectively. Indicates the preset percentage threshold;

[0027] The total adjustment factor is calculated as:

[0028]

[0029] Where ω z represents the total adjustment coefficient, μ1 and μ2 represent the first weight and the second weight respectively, and μ1+μ2=1, δ1 and δ2 represent the preset first adjustment factor and the second adjustment factor respectively.

[0030] Preferably, the shooting parameters are adjusted according to the total adjustment coefficient, and the shooting height and the shooting angle are adjusted in the following ways:

[0031]

[0032]

[0033] Where H and θ represent the shooting height and shooting angle of view respectively, and c represents the preset adjustment constant.

[0034] Preferably, the logic of using the image processing algorithm to process the analysis image to generate the image recognition result is:

[0035] Load N consecutive frames of analysis pictures into the optimized Yolov5 model in sequence for image recognition, and obtain the corresponding detection results. The detection results are expressed as:

[0036] D i =Yolov5(X i )

[0037] Where D i represents the detection result of the i-th frame analysis picture. The detection result consists of several groups of sub-results. The sub-results correspond to different types of wild animals one by one and include the identified bounding box position, animal category and confidence. The detection result D i ={d i (1),d i (2),…,d i (j),…d i (m)}, where d i(j) represents the sub-result of the jth animal category in the i-th frame analysis picture, j = 1, 2, 3, ..., m, m represents the number of animal categories, X i Indicates the i-th frame analysis picture;

[0038] The SORT algorithm is used to track the detection results of the Yolov5 model to obtain the position information of each individual in the same animal category in different frames. The tracking results are expressed as:

[0039] T i (j) = SORT[d i (j)]

[0040] Where T i Indicates the tracking result of the i-th frame analysis picture;

[0041] The tracking results of N consecutive frames of analysis images are processed, and the same target detected in consecutive frames is matched to reduce repeated counts, and the number of unique individuals of the same animal category appearing in different frames is obtained. Then the number of species of each animal is counted. The calculation method is:

[0042]

[0043] Where N z (j) represents the number of species in the jth animal category, T i ′(j) represents the number of unique individuals of the jth animal category appearing in different frames;

[0044] The animal categories are divided into set A1, and the number of species of each animal category is divided into set A2. The elements in set A1 and set A2 correspond one to one, and set A1 and set A2 are taken together as the image recognition result.

[0045] Preferably, the logic for processing the analyzed sound source to generate the sound recognition result is:

[0046] Build a deep learning model, collect the sounds of various wild animals as training sound sources, divide the training sound sources into training set and validation set in a ratio of 8:2, and use the training set and validation set to train and optimize the deep learning model;

[0047] The frequency domain features of the analyzed sound source are extracted through short-time Fourier transform, and the analyzed sound source is converted into a spectrum diagram. The calculation method is:

[0048] S(f,t)=∫s(τ)·w(t-τ)·e -j′2πfτ dτ

[0049] Where S(f,t) represents the spectrum of the analyzed sound source, s(τ) represents the sound signal of the analyzed sound source at time τ, w(t-τ) represents the time domain window function, f represents the frequency, t represents the reference time, and j′ represents the imaginary unit;

[0050] The spectrum of the analyzed sound source is input into the deep learning model to obtain the animal categories corresponding to different wild animals. The animal categories corresponding to the analyzed sound source are divided into set B and used as the sound recognition results.

[0051] Preferably, a set Q of wild animal species finally identified is generated based on the set A1 in the image recognition result and the set B in the sound recognition result. The set Q is expressed as:

[0052] Q=A1∩B

[0053] Based on the elements in set Q, corresponding elements are selected from set A2 to re-form a new set A3. The elements in set A3 correspond one-to-one to the elements in set Q, and set A3 and set Q are used as the final recognition results.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention significantly improves the accuracy of recognition and statistics by fusing image recognition and sound recognition results and dynamically adjusting the shooting parameters of the drone. The image recognition is performed by combining the YOLOv5 model and the SORT algorithm, and the sound recognition is performed by combining the deep learning model and the short-time Fourier transform, thereby ensuring the accuracy of the image and sound recognition results. The image and sound recognition results are combined to effectively reduce false positives and false negatives, and improve the consistency and reliability of the data. The shooting parameters of the drone can be dynamically adjusted to find a balance between regional resolution and coverage, ensuring that the system can work effectively in various environments and complex scenarios. This not only improves the recognition ability and data quality of the system, but also enhances the robustness and versatility of the system, thereby achieving the effect of improving the application effect and practical value of drones in wildlife monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0058] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0059] Example:

[0060] See also Figure 1 , the present invention provides a technical solution:

[0061] Wildlife identification and species population statistics based on drone image processing, the specific steps include:

[0062] S1: Capture an image of wild animals to generate a reference image, and use an image processing algorithm to process the reference image to generate a target ratio coefficient for wild animals.

[0063] The generation logic of the target proportion coefficient is:

[0064] The image processing algorithm uses the Yolov5 model, collects training pictures containing various types of wild animals, divides the training pictures into a training set and a validation set in a ratio of 8:2, and uses the training set and validation set to train and optimize the Yolov5 model respectively;

[0065] Use the optimized Yolov5 model to perform image recognition on the wild animals to be identified in the reference image, and generate the identification area corresponding to each wild animal individual;

[0066] Calculate the average area of ​​the identified regions in the reference image as follows:

[0067]

[0068] In the formula represents the average area, S k represents the area of ​​the kth recognition area, the subscript k represents the number of the recognition area, and n represents the number of recognition areas;

[0069] The target proportion coefficient is calculated based on the average area of ​​the recognition area. The calculation method is: In the formula It represents the target proportion coefficient, and Sz represents the total area of ​​the reference image. It can be seen from the calculation formula of the target proportion coefficient that the target proportion coefficient is proportional to the size of the identification area, reflecting the area proportion of wild animals in the reference image. It can be understood that if the target proportion coefficient is too small, it is considered that the area proportion of wild animals in the reference image is too small, and the image features cannot be accurately identified, which may lead to a decrease in the accuracy of subsequent wild animal species identification. On the contrary, if the target proportion coefficient is too large, it is considered that the area proportion of wild animals in the reference image is too large, and the overall image of the entire population cannot be collected, which may lead to a decrease in the accuracy of subsequent wild animal species identification.

[0070] In this step, the drone's shooting parameters can be precisely adjusted by calculating the target ratio. If the ratio is too small, the shooting parameters are readjusted to improve the resolution of the recognition area; if the ratio is too large, the shooting parameters are adjusted to expand the field of view to ensure that the entire group can be captured. This dynamic adjustment mechanism ensures that the collected image and sound data have the best quality and coverage, which not only improves the accuracy of subsequent recognition, but also enhances the versatility and adaptability of the system, enabling the system to work effectively in different scenarios.

[0071] S2: Adjust the shooting parameters of the drone based on the target ratio coefficient. After the adjustment is completed, perform secondary image and sound collection on the wild animals to generate analysis pictures and analysis sound sources.

[0072] Shooting parameters include shooting height and shooting angle. The logic for adjusting shooting parameters is:

[0073] Compare the target proportion coefficient with a preset proportion threshold, generate a first adjustment coefficient according to the comparison result, and then generate a second adjustment coefficient according to the number of recognition areas;

[0074] The first adjustment coefficient and the second adjustment coefficient are weighted to generate a total adjustment coefficient, and the total adjustment coefficient is used to adjust the shooting parameters.

[0075] The calculation methods of the first adjustment coefficient and the second adjustment coefficient are respectively:

[0076]

[0077] ω2=ln(n+1)

[0078] Where ω1 and ω2 represent the first adjustment coefficient and the second adjustment coefficient respectively. Indicates the preset percentage threshold.

[0079] The total adjustment factor is calculated as:

[0080]

[0081] Where ω z represents the total adjustment coefficient, μ1 and μ2 represent the first weight and the second weight respectively, and μ1+μ2=1, δ1 and δ2 represent the preset first adjustment factor and the second adjustment factor respectively, the first adjustment factor and the second adjustment factor are used to fine-tune the total adjustment coefficient, and can be adjusted in a user-preset manner.

[0082] Adjust the shooting parameters according to the total adjustment coefficient. The adjustment methods of shooting height and shooting angle are as follows:

[0083]

[0084] Where H and θ represent the shooting height and shooting angle of view respectively, and c represents the preset adjustment constant.

[0085] It can be seen that the first adjustment coefficient is inversely proportional to the size of the target proportion coefficient and is proportional to the size of the total adjustment coefficient. The larger the first adjustment coefficient, the larger the total adjustment coefficient and the smaller the target proportion coefficient, which means that the shooting height of the drone needs to be lowered and the shooting angle needs to be narrowed to obtain a higher regional resolution. Conversely, the smaller the first adjustment coefficient, the higher the shooting height of the drone needs to be raised and the shooting angle needs to be expanded to cover a larger area. The second adjustment coefficient is proportional to the number of identification areas and inversely proportional to the size of the total adjustment coefficient. That is to say, the larger the second adjustment coefficient, the smaller the total adjustment coefficient, the larger the number of identification areas, that is, the more wild animals in the image, so it is necessary to raise the shooting height of the drone and expand the shooting angle to cover a larger area. Conversely, the smaller the number of identification areas, that is, the fewer wild animals in the image, the shooting height of the drone needs to be lowered and the shooting angle needs to be narrowed to obtain a higher regional resolution. The specific values ​​and size relationship of the first weight and the second weight can be adjusted according to the actual needs of the user.

[0086] In this step, by setting the total adjustment coefficient obtained by the first adjustment coefficient and the second adjustment coefficient, a balance can be found between regional resolution and coverage, ensuring that the image has sufficient details and covers a larger area, adapting to the distribution of different wild animals, and being able to adaptively adjust shooting parameters under different environments and conditions, thereby enhancing the universality and robustness of the system and being able to work effectively in various complex scenarios. Compared with the existing technology, this dynamic adjustment mechanism not only improves the accuracy of recognition and statistics, but also reduces the system's errors and workload, effectively improving the performance and application effect of the overall solution.

[0087] S3: Use the image processing algorithm to process the analysis image to generate an image recognition result, and then use the sound processing algorithm to process the analysis sound source to generate a sound recognition result.

[0088] The logic of using image processing algorithms to process analysis images and generate image recognition results is as follows:

[0089] Load N consecutive frames of analysis pictures into the optimized Yolov5 model in sequence for image recognition, and obtain the corresponding detection results. The detection results are expressed as:

[0090] D i =Yolov5(X i )

[0091] Where D i represents the detection result of the i-th frame analysis picture. The detection result consists of several groups of sub-results. The sub-results correspond to different types of wild animals one by one and include the identified bounding box position, animal category and confidence. The detection result D i ={d i (1),d i (2),…,d i (j),…d i (m)}, where d i (j) represents the sub-result of the jth animal category in the i-th frame analysis picture, j = 1, 2, 3, ..., m, m represents the number of animal categories, X i Indicates the i-th frame analysis picture;

[0092] The SORT algorithm is used to track the detection results of the Yolov5 model to obtain the position information of each individual in the same animal category in different frames. The tracking results are expressed as:

[0093] T i (j) = SORT[d i (j)]

[0094] Where T i It represents the tracking result of the i-th frame analysis picture. The SORT algorithm can accurately track the position information of each individual in the same animal category in different frames, reduce repeated counting, and ensure the accuracy of target tracking.

[0095] The tracking results of N consecutive frames of analysis images are processed, and the same target detected in consecutive frames is matched to reduce repeated counts, and the number of unique individuals of the same animal category appearing in different frames is obtained. Then the number of species of each animal is counted. The calculation method is:

[0096]

[0097] Where Nz (j) represents the number of species in the jth animal category, T i ′(j) represents the number of unique individuals of the jth animal category appearing in different frames;

[0098] The animal categories are divided into set A1, and the number of species of each animal category is divided into set A2. The elements in set A1 and set A2 correspond one to one, and set A1 and set A2 are taken together as the image recognition result.

[0099] In this step, the Yolov5 model is efficient and accurate, and can quickly detect wild animals in the image and generate corresponding detection results. The SORT algorithm can track the detection results and accurately track the position information of each individual in the same animal category in different frames, reduce repeated counting, and ensure the accuracy of target tracking. By combining the efficient target detection of the YOLOv5 model and the precise target tracking of the SORT algorithm, accurate image recognition results can be generated to improve the accuracy of recognition and statistics, and reduce repeated counting and errors. The Yolov5 model and the SORT algorithm are existing technologies, so their specific working steps are not described in detail here.

[0100] The logic for processing the analyzed sound source and generating the sound recognition result is:

[0101] Build a deep learning model, collect the sounds of various wild animals as training sound sources, divide the training sound sources into training set and validation set in a ratio of 8:2, and use the training set and validation set to train and optimize the deep learning model;

[0102] The frequency domain features of the analyzed sound source are extracted through short-time Fourier transform, and the analyzed sound source is converted into a spectrum diagram. The calculation method is:

[0103] S(f,t)=∫s(τ)·w(t-τ)·e -j′2πfτ dτ

[0104] Where S(f,t) represents the spectrum of the analyzed sound source, s(τ) represents the sound signal of the analyzed sound source at time τ, w(t-τ) represents the time domain window function, f represents the frequency, t represents the reference time, and j′ represents the imaginary unit;

[0105] The spectrum of the analyzed sound source is input into the deep learning model to obtain the animal categories corresponding to different wild animals. The animal categories corresponding to the analyzed sound source are divided into set B and used as the sound recognition results.

[0106] In this step, the frequency domain features are extracted through deep learning models and short-time Fourier transform, and the sound source is processed to generate sound recognition results. The system can more accurately identify the sounds of different types of wild animals and combine them with the image recognition results to improve the overall recognition and statistical accuracy. This step not only improves the system's recognition ability and data quality, but also enhances the system's robustness and adaptability, which effectively promotes the performance and application value of the overall solution.

[0107] S4: Perform set division based on the image recognition result and the sound recognition result, and use the intersection of the image recognition result and the sound recognition result as the final recognition result.

[0108] According to the set A1 in the image recognition results and the set B in the sound recognition results, the set Q of the finally identified wild animal species is generated. The set Q is expressed as:

[0109] q=A1∩B

[0110] Based on the elements in set Q, corresponding elements are selected from set A2 to re-form a new set A3. The elements in set A3 correspond one-to-one to the elements in set Q, and set A3 and set Q are used as the final recognition results.

[0111] Here is a simple example. Suppose that after image recognition, three types of birds are identified, bird a, bird b and bird c. The corresponding animal categories are numbered 1, 2, and 3 respectively. The number of the three birds is N. z (1) = 10, N z (2) = 8, N z (3) = 6, then the corresponding image recognition result set A1 = {bird a, bird b, bird c}, set A2 = {N z (1),N z (2),N z (3)}. If three types of birds are identified after sound recognition, bird b, bird c and bird d, then the corresponding set B = {bird b, bird c, bird d}. Therefore, the set Q of wild animal species finally identified is the intersection of set A1 and set B, set Q = {bird b, bird c}. Then, according to the elements in set Q, the corresponding elements are selected from set A2 to re-form a new set A3, A3 = {N z (2),N z (3)}, and finally set A3 and set Q are used as the final recognition results to obtain the types and quantities of wild animals.

[0112] In this step, by combining the image recognition results with the sound recognition results, a multimodal data fusion method is provided, which effectively reduces the misidentification caused by single modality recognition, improves the precision and accuracy of recognition, and thus provides reliable data on the species and quantity of wild animals.

[0113] In summary, the present invention significantly improves the accuracy of recognition and statistics by fusing image recognition and sound recognition results and dynamically adjusting the shooting parameters of the drone. Image recognition is performed in combination with the YOLOv5 model and the SORT algorithm, and sound recognition is performed in combination with the deep learning model and the short-time Fourier transform to ensure the accuracy of the image and sound recognition results. The image and sound recognition results are combined to effectively reduce false positives and false negatives, and improve the consistency and reliability of the data. The shooting parameters of the drone can be dynamically adjusted to find a balance between regional resolution and coverage to ensure that the system can work effectively in various environments and complex scenarios. Thereby not only improving the recognition ability and data quality of the system, but also enhancing the robustness and versatility of the system, achieving the effect of improving the application effect and practical value of drones in wildlife monitoring.

[0114] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0115] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0116] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0117] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for identifying wild animals and counting species based on image processing taken by drones, characterized in that: The specific steps include: S1: collect an image of wild animals to generate a reference image, and use an image processing algorithm to process the reference image to generate a target ratio coefficient for wild animals; S2: Adjust the shooting parameters of the drone based on the target ratio coefficient. After the adjustment is completed, perform secondary image and sound collection on the wild animals to generate analysis pictures and analysis sound sources; S3: Use the image processing algorithm to process the analysis image to generate an image recognition result, and then use the sound processing algorithm to process the analysis sound source to generate a sound recognition result; S4: Perform set division based on the image recognition result and the sound recognition result, and use the intersection of the image recognition result and the sound recognition result as the final recognition result.

2. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 1 is characterized by: The generation logic of the target proportion coefficient is: The image processing algorithm uses the Yolov5 model, collects training pictures containing various types of wild animals, divides the training pictures into a training set and a validation set in a ratio of 8:2, and uses the training set and validation set to train and optimize the Yolov5 model respectively; Use the optimized Yolov5 model to perform image recognition on the wild animals to be identified in the reference image, and generate the identification area corresponding to each wild animal individual; Calculate the average area of ​​the identified regions in the reference image as follows: In the formula represents the average area, S k represents the area of ​​the kth recognition area, the subscript k represents the number of the recognition area, and n represents the number of recognition areas; The target proportion coefficient is calculated based on the average area of ​​the recognition area. The calculation method is: In the formula represents the target proportion coefficient, and Sz represents the total area of ​​the reference image.

3. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 2 is characterized by: The shooting parameters include shooting height and shooting angle of view. The logic for adjusting the shooting parameters is: Compare the target proportion coefficient with a preset proportion threshold, generate a first adjustment coefficient according to the comparison result, and then generate a second adjustment coefficient according to the number of recognition areas; The first adjustment coefficient and the second adjustment coefficient are weighted to generate a total adjustment coefficient, and the total adjustment coefficient is used to adjust the shooting parameters.

4. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 3 is characterized by: The calculation methods of the first adjustment coefficient and the second adjustment coefficient are respectively: ω2=ln(n+1) Where ω1 and ω2 represent the first adjustment coefficient and the second adjustment coefficient respectively. Indicates the preset percentage threshold; The total adjustment factor is calculated as: Where ω z represents the total adjustment coefficient, μ1 and μ2 represent the first weight and the second weight respectively, and μ1+μ2=1, δ1 and δ2 represent the preset first adjustment factor and the second adjustment factor respectively.

5. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 4 is characterized by: Adjust the shooting parameters according to the total adjustment coefficient. The adjustment methods of shooting height and shooting angle are as follows: Where H and θ represent the shooting height and shooting angle of view respectively, and c represents the preset adjustment constant.

6. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 1, characterized in that: The logic of using image processing algorithms to process analysis images and generate image recognition results is as follows: Load N consecutive frames of analysis pictures into the optimized Yolov5 model in sequence for image recognition, and obtain the corresponding detection results. The detection results are expressed as: D i =Yolov5(X i ) Where D i represents the detection result of the i-th frame analysis picture. The detection result consists of several groups of sub-results. The sub-results correspond to different types of wild animals one by one and include the identified bounding box position, animal category and confidence. The detection result D i ={d i (1),d i (2),…,d i (j),…d i (m)}, where d i (j) represents the sub-result of the jth animal category in the i-th frame analysis picture, j = 1, 2, 3, ..., m, m represents the number of animal categories, X i Indicates the i-th frame analysis picture; The SORT algorithm is used to track the detection results of the Yolov5 model to obtain the position information of each individual in the same animal category in different frames. The tracking results are expressed as: T i (j)=SORT[d i (j)] Where T i Indicates the tracking result of the i-th frame analysis picture; The tracking results of N consecutive frames of analysis images are processed, and the same target detected in consecutive frames is matched to reduce repeated counts, and the number of unique individuals of the same animal category appearing in different frames is obtained. Then the number of species of each animal is counted. The calculation method is: Where N z (j) represents the number of species in the jth animal category, T i ′(j) represents the number of unique individuals of the jth animal category appearing in different frames; The animal categories are divided into set A1, and the number of species of each animal category is divided into set A2. The elements in set A1 and set A2 correspond one to one, and set A1 and set A2 are taken together as the image recognition result.

7. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 6, characterized in that: The logic for processing the analyzed sound source and generating the sound recognition result is: Build a deep learning model, collect the sounds of various wild animals as training sound sources, divide the training sound sources into training set and validation set in a ratio of 8:2, and use the training set and validation set to train and optimize the deep learning model; The frequency domain features of the analyzed sound source are extracted through short-time Fourier transform, and the analyzed sound source is converted into a spectrum diagram. The calculation method is: S(f,t)=∫s(τ)·w(t-τ)·e -j′2πfτ dτ Where S(f,t) represents the spectrum of the analyzed sound source, s(τ) represents the sound signal of the analyzed sound source at time τ, w(t-τ) represents the time domain window function, f represents the frequency, t represents the reference time, and j′ represents the imaginary unit; The spectrum of the analyzed sound source is input into the deep learning model to obtain the animal categories corresponding to different wild animals. The animal categories corresponding to the analyzed sound source are divided into set B and used as the sound recognition results.

8. The method for identifying wild animals and counting species populations based on image processing taken by unmanned aerial vehicles according to claim 7, characterized in that: According to the set A1 in the image recognition results and the set B in the sound recognition results, the set Q of the finally identified wild animal species is generated. The set Q is expressed as: Q=A1∩B Based on the elements in set Q, corresponding elements are selected from set A2 to re-form a new set A3. The elements in set A3 correspond one-to-one to the elements in set Q, and set A3 and set Q are used as the final recognition results.

Images