Rare Bird Virtual Digitalization Method and System Based on Implicit Neural Representations

Abstract

The present invention discloses a virtual digitization method and system for rare birds based on implicit neural representation, which includes: Step 1, creating input data; Step 2, mapping the spatial position coordinates X of the reconstruction area of rare birds into a high-dimensional Fourier feature vector γ(X) through Fourier feature mapping, and setting an adaptive learnable filter H B (α(X)) at X, where H B (α(X)) includes B for controlling the dimension of H B (α(X)) and α(X) for controlling the initial filtering position of H B ; Step 3, combining H B (α(X)) with γ(X) to obtain the hidden unit z θ of the first layer of the MLPs network F (1) to obtain the color and volume density at X; Step 4, generating new perspective and new pose images of the static background of rare birds through volume rendering technology. The present invention can quickly render high-quality pictures of rare birds from any perspective only through the data collected by a monocular camera, and performs excellently in dealing with the complex shapes and details of rare birds.

Description

Technical Field

[0001] The present invention relates to the technical fields of computer vision and computer graphics, and particularly to a virtual digitalization method for rare birds based on implicit neural representation. Background Art

[0002] The virtual digitalization of rare birds faces many difficulties, mainly due to the requirements for high precision and authenticity. First of all, rare birds usually have complex feather structures and unique dynamic behaviors, and the detailed capture and accurate reproduction of these characteristics are an indispensable part of the digitalization process. However, traditional 3D modeling methods are difficult to meet the high-precision requirements for appearance, details, and motion at the same time, especially in simulating the texture of feathers, light and shadow changes, and subtle dynamic performances. In addition, the number of rare birds is limited, and it is difficult to obtain high-quality shooting and scanning data, especially in the wild environment. This data scarcity further exacerbates the complexity of digital modeling. In order to achieve a highly realistic representation of the virtual model, a large amount of manual adjustment and data processing must be carried out, and these processes are both time-consuming and laborious, and it is difficult to ensure the adaptability and authenticity of the model in different scenarios.

[0003] Learnable implicit neural representation provides significant advantages for the virtual digitalization of rare birds. First of all, this method can provide a three-dimensional highly realistic representation of rare birds in a high-quality manner, not only accurately capturing the appearance characteristics of the birds, but also dynamically presenting their complex feather structures and natural motion behaviors. Implicit neural representation is different from traditional explicit modeling techniques. It does not rely on specific geometric forms such as meshes or point clouds, but generates high-resolution three-dimensional models by learning the continuous function of the object through a neural network. This makes it perform excellently in dealing with the complex shapes and details of rare birds, and can more naturally simulate the interaction between light and feathers, thus enhancing the overall visual effect. In addition, this representation method can also effectively address the problem of data scarcity. It can generate high-quality models through a small amount of data training, and has good generalization ability, suitable for realistic reproduction in different environments and scenarios. Summary of the Invention

[0004] The purpose of the present invention is to provide a virtual digitalization method for rare birds based on implicit neural representation to meet the demand for high-quality reconstruction of virtual digitalization models of rare birds in current practical applications.

[0005] To achieve the above object, the present invention provides a virtual digitalization method for rare birds based on implicit neural representation, which includes:

[0006] Step 1: Collect images of rare birds from multiple angles using a camera, record the camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel and the ray direction of each pixel relative to the camera in the aligned images as input data;

[0007] Step 2: Determine the reconstruction area of the rare bird on the input data in Step 1, and map the spatial position coordinate X of the reconstruction area of the rare bird to the high-dimensional Fourier feature vector γ(X) shown in Equation (1) through Fourier feature mapping:

[0008] γ(X) = (sin(2 0 πX), cos(2 0 πX), …, sin(2 L-1 πX), cos(2 L-1 πX)) (1)

[0009] In the formula, L is a hyperparameter in the Fourier feature network;

[0010] Allocate a three-dimensional trainable grid with a set resolution to the reconstruction area of the rare bird, and store a one-dimensional learnable variable α(X) at a preset position in the three-dimensional trainable grid as a task parameter. Set an adaptive learnable filter H B (α(X)) at X. The filter H B (α(X)) is used to filter out some redundant information in γ(X). The filter H B (α(X)) contains a hyperparameter B. The hyperparameter B is used to control that the dimension of the filter H B (α(X)) does not exceed the dimension of γ(X), and further controls the filter H B (α(X)) to filter out the redundant information corresponding to the dimension of the filter H B (α(X)) in γ(X). The filter H B (α(X)) also contains the task parameter α(X). The task parameter α(X) is used to control the initial position of the filtering of the filter H B (α(X));

[0011] Step 3: Combine the filter H B (α(X)) in Step 2 with the high-dimensional Fourier feature vector γ(X) through the Hadamard product ⊙, as shown in Equation (2), to obtain the hidden unit z θ of the first layer of the MLPs (Chinese: used to learn the mapping from encoded features to signal values) network F (1) , z (1) Subsequently, it is input into the MLPs network F θ to output the color C and volume density σ at X;

[0012]

[0013] Among them, σ is the activation function of the MLPs network F θ and and respectively represent the weight and bias of the i-th layer of the MLPs network F θ where i = 1, …, k - 1, represents the hidden unit of the i-th layer of the MLPs network F θ and f represents the MLPs network F θ ;

[0014] Step 4: Integrate the color C and volume density σ at X obtained in Step 3 through volume rendering technology to generate a new perspective and new pose map of the static background of rare birds.

[0015] Furthermore, the filter H B (α(X)) is composed of its respective components and each is calculated by Equation (3):

[0016]

[0017] where ∈ is a filter performance adjustment parameter whose value can be preset in advance, the value of j is twice the dimension of X, and the hyperparameter B of the filter H B (α(X)) is less than j, thereby controlling the dimension of the filter H B (α(X)) not to exceed the dimension of γ(X).

[0018] Furthermore, ∈ = -1000.

[0019] Furthermore, the method for obtaining the task parameter α(X) includes: The task parameter α(X) is obtained through learning by a trainable grid that co-optimizes with the neural network parameters during the training process. The learning method includes: First, find the grid where X is located in the rare bird reconstruction area, and obtain the value of the task parameter α(X) at X through interpolation.

[0020] Furthermore, when α(X) = -1, the filter H B (α(X)) is transformed into a low-pass filter; when α(X) = 11, the filter H B (α(X)) is transformed into a band-pass filter; when α(X) = 21, the filter H B (α(X)) is transformed into a high-pass filter, and the hyperparameter B takes the value of 15 in all cases.

[0021] The present invention also provides a virtual digital system for rare birds based on implicit neural representation, which includes:

[0022] An input data creation unit, which is configured to acquire images of rare birds from multiple angles by a camera, record the camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel and the ray direction of each pixel relative to the camera in the aligned images as input data;

[0023] A Fourier feature mapping unit, which is configured to determine a reconstruction area of a rare bird on the input data of the input data creation unit, and map the spatial position coordinate X of the reconstruction area of the rare bird into a high-dimensional Fourier feature vector γ(X) shown in Equation (1) through Fourier feature mapping:

[0024] Y(X) = (sin(2 0 πX), cos(2 0 πX), …, sin(2 L-1 πX), cos(2 L-1 πX)) (1)

[0025] In the formula, L is a hyperparameter in the Fourier feature network;

[0026] An adaptive learnable filter setting unit, which is configured to allocate a three-dimensional trainable grid with a set resolution to the reconstruction area of the rare bird, store a one-dimensional learnable variable α(X) at a preset position of the three-dimensional trainable grid as a task parameter, and set an adaptive learnable filter H B (α(X)) at X. The filter H B (α(X)) is used to filter a part of the redundant information in γ(X). The filter H B (α(X)) includes a hyperparameter B. The hyperparameter B is used to control that the dimension of the filter H B (α(X)) does not exceed the dimension of γ(X), and further controls that the filter H B (α(X)) filters the redundant information corresponding to the dimension of the filter H B (α(X)) in γ(X). The filter H B (α(X)) also includes a task parameter α(X). The task parameter α(X) is used to control the initial position of the filtering of the filter H B (α(X));

[0027] A reconstruction parameter acquisition unit, which is configured to combine the filter H B (α(X)) and the high-dimensional Fourier feature vector γ(X) through the Hadamard product ⊙, as shown in Equation (2), to obtain the hidden unit z θ of the first layer of the MLPs network F (1) , z (1) Subsequently, it is input into the MLPs network F θ and outputs the color C and the volume density σ at X;

[0028]

[0029] Among them, σ is the activation function of the MLPs network F θ and and respectively represent the weight and bias of the i-th layer of the MLPs network F θ where i = 1, …, k - 1, represents the hidden unit of the i-th layer of the MLPs network F θ and f represents the MLPs network F θ ;

[0030] The scene reconstruction unit is used to integrate the color C and the volume density σ at X through volume rendering technology to generate a new view and a new pose map of the static background of rare birds.

[0031] Furthermore, the filter H B (α(X)) is composed of its respective components and each is calculated by Equation (3):

[0032]

[0033] where ∈ is a filter performance adjustment parameter whose value can be preset in advance, the value of j is twice the dimension of X, and the hyperparameter B of the filter H B (α(X)) is less than j, thereby controlling the dimension of the filter H B (α(X)) not to exceed the dimension of γ(X).

[0034] Furthermore, ∈ = -1000.

[0035] Furthermore, the method for obtaining the task parameter α(X) includes: The task parameter α(X) is learned through a trainable grid that co-optimizes with the neural network parameters during the training process. The learning method includes: First, find the grid where X is located in the rare bird reconstruction area, and obtain the value of the task parameter α(X) at X through interpolation.

[0036] Furthermore, when α(X) = -1, the filter H B (α(X)) is transformed into a low-pass filter; when α(X) = 11, the filter H B (α(X)) is transformed into a band-pass filter; when α(X) = 21, the filter H B (α(X)) is transformed into a high-pass filter, and the hyperparameter B takes a value of 15 in all cases.

[0037] The present invention can quickly render high-quality images of rare birds from any perspective only through the data collected by a monocular camera, and performs excellently in dealing with the complex shapes and details of rare birds. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is a schematic flowchart of a virtual digitization method for rare birds based on implicit neural representation provided by an embodiment of the present invention.

[0039] Figure 2 、 Figure 3 and Figure 4 are respectively visual schematic diagrams of the conversion of a learnable filter into a low-pass filter when α(X) takes different values according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] In the drawings, the same or similar reference numerals are used to represent the same or similar elements or elements with the same or similar functions. The embodiments of the present invention will be described in detail below with reference to the drawings.

[0041] In the description of the present invention, the orientation or positional relationship indicated by the terms "center", "longitudinal", "transverse", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the protection scope of the present invention.

[0042] As Figure 1 shown, the virtual digitization method for rare birds based on implicit neural representation provided by an embodiment of the present invention includes:

[0043] Step 1, collect images of rare birds from multiple angles through a camera, record the camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel and the ray direction of each pixel relative to the camera in the aligned images as input data.

[0044] Among them, the sampling angles are required to be as sparse as possible and cover a large range. The camera parameters include the camera internal parameters and external parameters. The camera internal parameters include the focal length, and the camera external parameters include the rotation matrix and the translation vector. The ray direction defines the direction of each pixel in the image in 3D space starting from the viewpoint of the camera. Each ray is calculated through the camera internal parameters and external parameters, points to the pixel on the image plane, and then extends to a specific position in the 3D scene.

[0045] Step 2, determine the reconstruction area of the rare bird on the input data of Step 1, and this reconstruction area of the rare bird is often determined by continuously adjusting parameters.

[0046] The spatial position coordinates X of the rare bird reconstruction area are Fourier feature mapped into the high-dimensional Fourier feature vector γ(X) shown in Equation (1), combined with Figure 1 , where X represents the three-dimensional coordinate vector at position x in the world coordinate system xyz0, corresponding to InputPosition in Figure 1 in the text:

[0047] γ(X) = (sin(2 0 πX), cos(2 0 πX), …, sin(2 L-1 πX), cos(2 L-1 πX)) (1)

[0048] In the formula, L is a hyperparameter in the Fourier feature network, and its value can be preset for different tasks to avoid the problem of reduced running speed of the MLPs network Fθ caused by being too large.

[0049] Allocate a three-dimensional trainable grid with a set resolution for the rare bird reconstruction area, and store a one-dimensional learnable variable α(X) at a preset position of the three-dimensional trainable grid, such as the spatial position of the corner point of the three-dimensional trainable grid, as a task parameter.

[0050] Set an adaptive learnable filter H B (α(X)) at X. The filter H B (α(X)) is used to filter out some redundant information in γ(X) to reduce the training time of the MLPs network Fθ.

[0051] In one embodiment, the filter H B (α(X)) includes a hyperparameter B. The hyperparameter B is used to control the dimension of the filter H B (α(X)) not to exceed the dimension of γ(X), thereby controlling the filter H B (α(X)) to filter the redundant information in γ(X) corresponding to the dimension of the filter H B (α(X)).

[0052] In another embodiment, the filter H B (α(X)) further includes the task parameter α(X). The task parameter α(X) is used to control the filter H B(α(X)) The initial position of the filter is adjusted to meet the requirements of different tasks. The method for obtaining the task parameter α(X) includes: the task parameter α(X) is learned during the training process through a trainable grid that is co-optimized with the neural network parameters. The learning method includes: first, find the grid where X is located in the rare bird reconstruction area, and obtain the value of the task parameter α(X) at X through interpolation. Therefore, in this embodiment, by adjusting the resolution of the grid where X is located, the value of the task parameter α(X) can be adjusted, thereby controlling the initial position of the filter and further meeting the needs of different tasks. For example: by reducing the task parameter α(X), a filter H B (α(X)) is assigned to a region in the rare bird reconstruction area, which can remove the noise and outlier points in the region. Generally, in the 2D image fitting task, the resolution of the trainable grid of α(X) is set to the image resolution, which significantly improves the ability to accurately represent the natural scene in the image.

[0053] Step 3, combine the filter H B (α(X)) with the high-dimensional Fourier feature vector γ(X) through the Hadamard product ⊙, as shown in Equation (2), to obtain the hidden unit z θ of the first layer of the MLPs network F (1) , z (1) is then input into the MLPs network F θ and the color C and volume density σ at X are output.

[0054]

[0055] where σ is the activation function of the MLPs network F θ , and respectively represent the weight and bias of the i-th layer of the MLPs network F θ , i = 1,..., k - 1, represents the hidden unit of the i-th layer of the MLPs network F θ , and f represents the MLPs network F θ .

[0056] In other embodiments, other network structures other than the MLPs network can also be integrated with the filter H B (α(X)). The ReLU activation function in the network can be replaced with a sine activation function, and better performance than the original structure can be obtained.

[0057] In the fields of computer vision and computer graphics, the radiance field represents the radiance emitted at each point in space. Through the neural network of the radiance field and with the help of deep learning techniques, the color information and volume density information of each point are learned, and then the scene is reconstructed through volume rendering.

[0058] Step 4: Through the volume rendering technique, integrate the color C and volume density σ at X obtained in Step 3, as shown in Equation (4), to generate a new view and new pose map of the static background of rare birds:

[0059]

[0060] In the formula, C(r) is the color information of the pixel point in the new view under the camera ray r(t) at any view angle, t n and t f are the near and far boundaries where r(t) passes through the object respectively, T(t) is the cumulative transmittance of the camera ray r(t) from t n to t f , which is described by Equation (5), σ(r(t)) is the volume density of the three-dimensional point in the camera space under r(t), c(r(t)) is the color of the three-dimensional point in the camera space under r(t), r(t) = o + td, d represents the direction, o represents the optical center, and t represents the distance:

[0061]

[0062] In one embodiment, the filter H B (α(X)) is composed of its respective components , and each is calculated through Equation (3):

[0063]

[0064] Among them, for the filter H B (α(X)), its dimension is the same as that of the formula γ(X). When ∈ takes a larger value, the filter performance deviates from the ideal state; the smaller the value of ∈, the closer the filter is to the ideal performance. Through experimental verification, ∈ = -1000 can meet the design requirements. Therefore, ∈ = -1000 is selected. Equation (3) designs an adaptive differentiable filter, which can achieve various filtering functions such as low-pass, band-pass, and high-pass through the adjustment of a single parameter. Of course, according to the actual filtering requirements, those skilled in the art can also adjust in form.

[0065] Figure 2 , Figure 3 and Figure 4 are the visualization diagrams of the learnable filter being transformed into a low-pass filter when α(X) takes different numerical values according to the embodiments of the present invention respectively.Figure 2 It shows the visualization that when α(X) = -1, the learnable filter is transformed into a low-pass filter. Figure 3 It shows the visualization that when α(X) = 11, the learnable filter is transformed into a band-pass filter. Figure 4 It shows the visualization that when α(X) = 21, the learnable filter is transformed into a high-pass filter. In the figure, the horizontal axis j represents the index value of the learnable filter component, and the vertical axis H represents the response value. Among them, the parameter B controls the width of the filter, and the value is 15 in Figure 1 , 2, and 3.

[0066] The parameters of the filter provided in the above embodiments can be automatically adjusted through the learning process of the neural network. This method allows the network to dynamically adjust the response of the filter to frequency components according to different task requirements.

[0067] In a natural scene containing rare birds, there are obvious differences in the spectrum of local areas. First of all, rare birds usually have complex feather structures and unique dynamic behaviors. The careful capture and accurate reproduction of these features are an indispensable part of the digitalization process. Generally, the main part presents low frequency, and the edge part presents high frequency. Fixed coding frequency Fourier features will lead to redundant input information and reduce the running speed of the model. Sine coding is the main factor for obtaining high-quality signal reconstruction, aiming to embed multiple orthogonal Fourier bases into the subsequent network. Although Fourier coding has achieved success, there are still some unresolved problems. The selection of the correct frequency scale often requires manual adjustment and usually involves a complex parameter adjustment process. This method selects frequencies for the entire input in a global and spatially invariant manner, thus missing the opportunity to better adapt to local high-frequency features, and may ultimately lead to deficiencies in the model when processing areas with rich details, manifested as unsatisfactory representation effects such as missing details and blurred edges. The learnable filter network proposed in this embodiment performs low-pass, band-pass, and high-pass operations on the Fourier feature vector by a learnable filter controlled by a learnable parameter. The Fourier features after filtering can better represent local high-frequency details. Adaptive filtering of Fourier features with different frequencies through a differentiable filter with learnable parameters significantly accelerates the convergence speed of MLPs.

[0068] The present invention also provides a virtual digitalization system for rare birds based on implicit neural representation, which includes an input data creation unit, a Fourier feature mapping unit, an adaptive learnable filter setting unit, a reconstruction parameter acquisition unit, and a scene reconstruction unit, wherein:

[0069] The input data creation unit is used to collect images of rare birds from multiple angles by a camera, record the camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel and the ray direction of each pixel relative to the camera in the aligned images as input data.

[0070] The Fourier feature mapping unit is used to determine the reconstruction area of rare birds on the input data of the input data creation unit, and map the spatial position coordinates X of the reconstruction area of rare birds into the high-dimensional Fourier feature vector γ(X) shown in Equation (1) through Fourier feature mapping.

[0071] The adaptive learnable filter setting unit is used to allocate a three-dimensional trainable grid with a set resolution for the reconstruction area of rare birds, store a one-dimensional learnable variable α(X) at a preset position in the three-dimensional trainable grid as a task parameter, and set the adaptive learnable filter H B (α(X)), and the filter H B (α(X)) is used to filter some redundant information in γ(X), and the filter H B (α(X)) contains a hyperparameter B, and the hyperparameter B is used to control the dimension of the filter H B (α(X)) not exceeding the dimension of γ(X), thereby controlling the filter H B (α(X)) to filter the redundant information corresponding to the dimension of γ(X) in γ(X), and the filter H B (α(X)) also contains the task parameter α(X), and the task parameter α(X) is used to control the initial position of the filter H B (α(X)) for filtering. B (α(X)) filtering.

[0072] The reconstruction parameter acquisition unit is used to combine the filter H B (α(X)) and the high-dimensional Fourier feature vector γ(X) through the Hadamard product ⊙, as shown in Equation (2), to obtain the hidden unit z( 1) of the first layer of the MLPs network Fθ, and z(1) is then input into the MLPs network Fθ to output the color C and volume density σ at X.

[0073] The scene reconstruction unit is used to integrate the color C and volume density σ at X through volume rendering technology to generate new perspective and new pose maps of the static background of rare birds.

[0074] The present invention provides a framework for learnable implicit neural representation, aiming to address the limitations of traditional implicit neural representation methods (INR) in expressing high-frequency information and convergence speed. The core idea is to introduce an adaptive learnable filter, which controls the filter through a learnable parameter to achieve functions such as low-pass, band-pass, and high-pass, thereby realizing spatial adaptive filtering of Fourier features. By performing spatial adaptive filtering on Fourier features, the expression ability of the network for natural scenes is enhanced, and the convergence speed is significantly improved.

[0075] The framework of learnable implicit neural representation provided by the invention can also be applied to other tasks. This framework can not only efficiently capture complex structures in data, but also has extremely high flexibility and is suitable for various types of data processing tasks. By training network parameters, the network can implicitly represent the features of input data, thereby achieving efficient learning and expression of data.

[0076] Taking rare birds as an example, the application prospect of this framework is particularly broad. First, in the aspect of one-dimensional audio signal processing, this framework can be used to fit the calls of rare birds. By learning a large number of bird call samples, the neural network can capture the unique features of the sounds of different species of birds, and then achieve high-precision recognition and classification of the calls of specific birds. Second, in the field of two-dimensional image signal processing, it can be used to represent high-definition pictures of rare birds. By learning visual features such as textures and colors in the images, the neural network can reconstruct high-quality bird images. Finally, it can also play a role in text information processing. By implicitly storing the text information of birds in the parameters and weights of the network, rapid retrieval and understanding of bird-related literature can be achieved. It is worth mentioning that all of the above functions can be realized through a lightweight neural network architecture. This means that while ensuring performance, this method also has low computational costs and storage requirements, and is very suitable for resource-limited environments, such as field investigation sites or deployment on mobile devices.

[0077] The present invention explores the frequency domain features of local regions (usually the main part presents low frequency and the edge part presents high frequency), and filters the Fourier feature encoding of local regions through a learnable filter, effectively avoiding information redundancy caused by fixed encoding frequencies and improving the operating efficiency of the model.

[0078] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Those of ordinary skill in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some of the technical features can be equivalently replaced; these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A virtual digitization method for rare birds based on implicit neural representation, characterized in that: include: Step 1, collect images of rare birds from multiple angles through a camera, record the camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel in the aligned image and the ray direction of each pixel relative to the camera as input data; Step 2: Determine the reconstruction area of ​​rare birds based on the input data of step 1, and transform the spatial position coordinate X of the reconstruction area of ​​rare birds into a high-dimensional Fourier feature vector γ(X) shown in formula (1) through Fourier feature mapping: γ(X)=(sin(2 0 πX),cos(2 0 πX),…,sin(2 L-1 πX),cos(2 L-1 πX)) (1) Where L is the hyperparameter in the Fourier feature network; Assign a three-dimensional trainable grid of set resolution to the reconstruction area of ​​rare birds, and store a one-dimensional learnable variable α(X) at the preset position of the three-dimensional trainable grid as a task parameter, and set an adaptive learnable filter H at X. B (α(X)), filter H B (α(X)) is used to filter some redundant information in γ(X). The filter H B (α(X)) contains the hyperparameter B, which is used to control the filter H B The dimension of (α(X)) does not exceed the dimension of γ(X), thus controlling the filter H B (α(X)) corresponds to the filter H in γ(X) B The redundant information of the dimension (α(X)) is filtered, and the filter H B (α(X)) also contains the task parameter α(X), which is used to control the filter H B (α(X)) Initial position of the filter; Step 3: Replace the filter H in step 2 B (α(X)) is combined with the high-dimensional Fourier eigenvector γ(X) through the Hadamard product ⊙, as shown in formula (2), to obtain the MLPs network F θ The first layer of hidden units z (1) , z (1) It is then input into the MLPs network F θ In the output, the color C and volume density σ at X are output; Where σ is the MLPs network F θ The activation function, and Represents the MLPs network F θ The weights and biases of the i-th layer, i = 1, ..., k-1, Represents the MLPs network F θ The hidden unit of the i-th layer, f represents the MLPs network F θ ; Step 4, using volume rendering technology, the color C and volume density σ at X obtained in step 3 are integrated to generate a new perspective and new posture map of the static background of the rare bird.

2. The method for virtual digitization of rare birds based on implicit neural representation according to claim 1, characterized in that: Filter H B (α(X)) is composed of its components Composition, each Calculated by formula (3): Among them, ∈ is a filter performance adjustment parameter whose value can be preset, the value of j is twice the dimension of X, and the filter H B The hyperparameter B of (α(X)) is less than j, thus controlling the filter H B The dimension of (α(X)) does not exceed the dimension of γ(X).

3. The method for virtual digitization of rare birds based on implicit neural representation according to claim 2, characterized in that: ∈＝-1000。 4. The method for virtual digitization of rare birds based on implicit neural representation according to any one of claims 1 to 3, characterized in that: The method for obtaining the task parameter α(X) includes: the task parameter α(X) is learned and obtained through a trainable grid that is coordinated with the neural network parameters during the training process. The learning method includes: first finding the grid where X is located in the rare bird reconstruction area, and obtaining the value of the task parameter α(X) at X through an interpolation method.

5. The method for virtual digitization of rare birds based on implicit neural representation according to claim 4, characterized in that: When α(X) = -1, the filter H B (α(X)) is converted into a low-pass filter; when α(X) = 11, the filter H B (α(X)) is converted into a bandpass filter; when α(X) = 21, the filter H B (α(X)) is transformed into a high-pass filter, and the hyperparameter B is set to 15.

6. A rare bird virtual digitization system based on implicit neural representation, characterized in that: include: An input data creation unit, which is used to collect images of rare birds from multiple angles according to a camera, record camera parameters of each image, align the images according to the camera parameters, and use the color value of each pixel in the aligned image and the ray direction of each pixel relative to the camera as input data; The Fourier feature mapping unit is used to determine the rare bird reconstruction area on the input data of the input data creation unit, and transform the spatial position coordinate X of the rare bird reconstruction area into a high-dimensional Fourier feature vector γ(X) shown in formula (1) through Fourier feature mapping: γ(X)=(sin(2 0 πX),cos(2 0 πX),…,sin(2 L-1 πX),cos(2 L-1 πX)) (1) Where L is the hyperparameter in the Fourier feature network; The adaptive learnable filter setting unit is used to allocate a three-dimensional trainable grid with a set resolution to the rare bird reconstruction area, and store a one-dimensional learnable variable α(X) at a preset position of the three-dimensional trainable grid as a task parameter, and set the adaptive learnable filter H at X. B (α(X)), filter H B (α(X)) is used to filter some redundant information in γ(X). The filter H B (α(X)) contains the hyperparameter B, which is used to control the filter H B The dimension of (α(X)) does not exceed the dimension of γ(X), thus controlling the filter H B (α(X)) corresponds to the filter H in γ(X) B The redundant information of the dimension (α(X)) is filtered, and the filter H B (α(X)) also contains the task parameter α(X), which is used to control the filter H B (α(X)) Initial position of the filter; The reconstruction parameter acquisition unit is used to transform the filter H B (α(X)) is combined with the high-dimensional Fourier eigenvector γ(X) through the Hadamard product ⊙, as shown in formula (2), to obtain the MLPs network F θ The first layer of hidden units z( 1 0,z( 1 ) is then input into the MLPs network F θ In the output, the color C and volume density σ at X are output; Where σ is the MLPs network F θ The activation function, and Represents the MLPs network F θ The weights and biases of the i-th layer, i = 1, ..., k-1, Represents the MLPs network F θ The hidden unit of the i-th layer, f represents the MLPs network F θ ; The scene reconstruction unit is used to integrate the color C and the volume density σ at X through volume rendering technology to generate a new perspective and a new posture map of the static background of the rare bird.

7. The rare bird virtual digitization system based on implicit neural representation according to claim 6, characterized in that: Filter H B (α(X)) is composed of its components Composition, each Calculated by formula (3): Among them, ∈ is a filter performance adjustment parameter whose value can be preset, the value of j is twice the dimension of X, and the filter H B The hyperparameter B of (α(X)) is less than j, thus controlling the filter H B The dimension of (α(X)) does not exceed the dimension of γ(X).

8. The rare bird virtual digitization system based on implicit neural representation as claimed in claim 7, characterized in that: ∈＝-1000。 9. The rare bird virtual digitization system based on implicit neural representation according to any one of claims 6 to 8, characterized in that: The method for obtaining the task parameter α(X) includes: the task parameter α(X) is learned and obtained through a trainable grid that is coordinated with the neural network parameters during the training process. The learning method includes: first finding the grid where X is located in the rare bird reconstruction area, and obtaining the value of the task parameter α(X) at X through an interpolation method.

10. The rare bird virtual digitization system based on implicit neural representation according to claim 9, characterized in that: When α(X) = -1, the filter H B (α(X)) is converted into a low-pass filter; when α(X) = 11, the filter H B (α(X)) is converted into a bandpass filter; when α(X) = 21, the filter H B (α*X)) is transformed into a high-pass filter, and the hyperparameter B is set to 15.

Images