Inversion method of total chlorophyll fluorescence in farmland canopy based on ray tracing technology

Abstract

The present invention provides an inversion method for total chlorophyll fluorescence in a farmland canopy based on ray tracing technology, comprising the following steps: step 1, performing three-dimensional reconstruction of the canopy based on point cloud; step 2, obtaining the proportional distribution of chlorophyll fluorescence in the canopy by forward ray tracing; step 3, obtaining the numerical distribution of chlorophyll fluorescence in the canopy by backward ray tracing. Solar-induced chlorophyll fluorescence has been considered as a powerful way to estimate photosynthesis efficiency and vegetation productivity in recent years because of its direct relationship with photosynthesis. The present invention uses a new idea to realize a method for directly inverting total chlorophyll fluorescence through observation data, laying a foundation for accurately and simply estimating vegetation productivity.

Description

Technical Field

[0001] The invention belongs to the technical field of chlorophyll fluorescence inversion, and in particular relates to an inversion method for total chlorophyll fluorescence in a farmland canopy based on ray tracing technology. Background Art

[0002] In recent years, chlorophyll fluorescence has been proven to be useful for estimating total primary productivity. Based on this discovery, chlorophyll fluorescence inversion technology has begun to be widely used in agriculture, forestry, and environmental monitoring, especially in the assessment of vegetation growth conditions. This technology uses point cloud data obtained from laser scanning to perform three-dimensional reconstruction of the vegetation canopy, and on this basis combines forward and backward ray tracing technology to achieve the inversion of total chlorophyll fluorescence in the vegetation canopy. This is of great significance for the quantitative assessment of farmland and forest resources, biomass estimation, and carbon storage estimation.

[0003] Chlorophyll fluorescence inversion technology can be used in the following fields:

[0004] Agricultural resource management: Accurately assessing crop growth status through chlorophyll fluorescence can help develop better management strategies, such as optimizing irrigation and fertilization plans, monitoring crop health, etc.

[0005] Forestry Resource Management: Similar to agricultural applications, this technology can help accurately assess forest conditions and support forest health monitoring and management.

[0006] Environmental Monitoring and Conservation: The ability to monitor changes in vegetation health and productivity provides valuable data support for assessing the impacts of environmental stressors (such as drought or pollution) and promotes conservation efforts.

[0007] Carbon Stock Assessment: Precise estimation of biomass through chlorophyll fluorescence allows accurate calculation of carbon storage in vegetation.

[0008] Ecological research: Understanding vegetation structure and function is essential for studying ecosystem dynamics and processes. Chlorophyll fluorescence inversion provides an efficient means to obtain detailed information on vegetation health and productivity.

[0009] Solar induced chlorophyll fluorescence (SIF) is a special fluorescence emitted by excited chlorophyll molecules. As an energy expenditure, it shares the energy provided by photosynthetically active radiation with photochemical reactions and heat dissipation. More importantly, it is closely coupled with photosynthesis in basic biochemical and biophysical processes, which indicates that SIF can be used as a way to accurately estimate gross primary productivity. In recent studies, the SIF-GPP mechanistic relationship based on electron transfer rate has been deduced by researchers, greatly improving the versatility of GPP deduction based on SIF. In actual application, it is an important step to improve the accuracy of SIF-GPP deduction results by inverting the observed SIF values ​​to the total SIF value inside the canopy.

[0010] At present, the method of inferring the observed SIF value to the total SIF value mainly introduces a parameter called the escape coefficient, which is used to characterize the ratio of the escape from the leaves to the canopy. The introduction of the escape rate has solved the problem of calculating the total amount from observation to total amount; however, the escape coefficient is a parameter affected by vegetation structure and vegetation type, which can only be obtained through field observation estimation. For large-scale and multi-type vegetation work, it is difficult to have a suitable calculation method to obtain the escape rate.

[0011] The existing methods for achieving total SIF distribution in the canopy through ray tracing, such as FluorWPS, etc., mainly simulate the distribution of photosynthetically active radiation in the canopy first, and use photosynthetically active radiation and vegetation physiological parameters to estimate SIF. This type of scheme can invert the overall distribution of SIF in the canopy more accurately, but because vegetation physiological parameters are affected by external conditions such as vegetation type and meteorological conditions, its versatility is also limited to a certain extent. Summary of the invention

[0012] Purpose of the invention: The technical problem to be solved by the present invention is to provide an inversion method for total chlorophyll fluorescence in the farmland canopy based on ray tracing technology in view of the deficiencies of the prior art. This solution firstly realizes the three-dimensional structure reconstruction of vegetation through LiDAR point cloud, uses forward ray tracing technology to evaluate the distribution of photosynthetically active radiation in the canopy, and derives the SIF distribution ratio in the canopy based on the correlation between photosynthetically active radiation and SIF. Finally, the total SIF in the canopy is inverted by using backward ray tracing technology combined with the observed SIF results.

[0013] The method of the present invention comprises the following steps:

[0014] Step 1, perform three-dimensional reconstruction of the canopy based on point cloud;

[0015] Step 2, obtaining the chlorophyll fluorescence ratio distribution in the canopy by forward ray tracing;

[0016] Step 3: Obtain the chlorophyll fluorescence value distribution in the canopy through backward ray tracing.

[0017] Step 1 includes:

[0018] Step 1-1, extracting individual plants from point cloud;

[0019] Step 1-2, complete the point cloud;

[0020] Steps 1-3: perform 3D reconstruction based on point cloud.

[0021] Step 1-1 includes: using the yoloV8 deep learning model to segment the full-color overhead image obtained in the process of collecting point clouds, and extracting the point cloud of each crop from the farmland point cloud.

[0022] Step 1-2 includes: selecting the PF-NET (Point Fractal Network) deep learning model for point cloud completion, the PF-NET deep learning model includes a multi-resolution encoder, a point pyramid decoder and a discriminant network, first using iterative farthest point sampling to extract skeleton points from the point cloud to capture shape features, in the multi-resolution encoder, a combined multi-layer perceptron is used to extract features, and the final latent vector F is formed by connecting multi-dimensional feature vectors. The point pyramid decoder uses the final feature vector V as input to generate a point cloud of the missing area, and uses a hierarchical structure to predict point clouds at different resolutions, gradually refining geometric details. The loss function includes multi-stage completion loss and adversarial loss.

[0023] Steps 1-3 include: using the SIREN (Sinusoidal Representation Networks) deep learning model to achieve 3D reconstruction of vegetation, the formula is:

[0024] (1)

[0025] Where x is the three-dimensional coordinate of a point in the input point cloud; is the implicit function that the neural network needs to learn; is an implicit function with x as variable; for The gradient of for The n-order gradient, n is the highest order gradient number, which is determined according to the actual reconstruction accuracy and is usually 2; F is the constraint function; Represents mapping the three-dimensional coordinate x to ;

[0026] The SIREN deep learning model uses a sinusoidal activation function as an implicit function:

[0027] (2)

[0028] in and are the weight matrix and bias vector of the i-th layer of the neural network respectively; n is the total number of layers of the neural network, and i is the current number of layers (the number of layers refers to the number of levels in the network. Each layer usually includes a linear transformation and a nonlinear activation function. The more layers there are, the deeper the network is, and it can usually learn more complex features); is the implicit function to be learned by the 0th layer of the neural network; It is the implicit function to be learned by the n-1th layer of the neural network; in each layer of the neural network, the sine activation function is used ;

[0029] After learning the implicit function, SIREN minimizes the loss function by penalizing deviations from the original constraints on the input domain:

[0030] (3)

[0031] Where L is the loss function; is the currently defined three-dimensional space area; is the mth constraint, is the three-dimensional space area under the mth constraint; a(x) is related to the features of the three-dimensional coordinate x (here is the three-dimensional coordinate of the output point cloud), which is used to define the constraint condition Parameters; M is the total number of constraints.

[0032] Step 2 includes: simulating photosynthetically active radiation in the vegetation canopy, and determining the chlorophyll fluorescence distribution in the canopy based on the distribution of photosynthetically active radiation in the canopy;

[0033] When simulating photosynthetically active radiation, forward ray tracing technology is selected as the simulation method. Forward ray tracing is to emit photons from the light source. When the photons touch the scene, they interact and calculate reflection, absorption and transmission phenomena until the light intersects with the observer or dissipates in the scene.

[0034] During vegetation modeling, each triangular face is defined as a front leaf or a back leaf, and different faces are defined with different reflectances and transmittances. When a photon interacts with a leaf, the energy of the incident photon is divided into three parts: reflected photon energy, transmitted photon energy, and absorbed photon energy. The calculation formula is:

[0035] (4)

[0036] (5)

[0037] (6)

[0038] in, , , They are respectively the generated photon energy, transmitted photon energy and absorbed photon energy generated when photon e contacts the blade; is the total energy of the incident photon e; and are the reflectivity and transmittance of the current leaf respectively.

[0039] The process of performing forward ray tracing includes: emitting a large number of photons from the sun, and after each photon hits a leaf or the ground, the energy and angle of the transmitted and scattered light are calculated based on the reflectivity and transmittance of the material of the object it hits, and then simulated again until all photons are absorbed, emitted to the sky, or completely dissipated, and the forward ray tracing is considered complete;

[0040] After executing the photosynthetically active radiation simulation, the absorbed photon energy of each leaf is calculated to obtain the photosynthetically active radiation energy distribution absorbed by each leaf in the entire canopy.

[0041] In step 2, the leaf triangle that absorbs the most photosynthetically active radiation in the environment is selected, and the fluorescence energy emitted by the leaf triangle is recorded as 1 standard unit. Then, based on the ratio of the photosynthetically active radiation energy absorbed by each leaf triangle and the maximum value, the proportion of the fluorescence energy emitted by each leaf triangle is confirmed.

[0042] Step 3 includes: using backward ray tracing combined with observed chlorophyll fluorescence to determine the chlorophyll fluorescence value distribution and total chlorophyll fluorescence value within the canopy;

[0043] Backward ray tracing is to send photons from the observer, and the photons interact with the scene when they touch it and calculate reflection, absorption and transmission phenomena until the light dissipates in the scene;

[0044] After receiving photosynthetically active radiation, plants will emit solar-induced chlorophyll fluorescence. The energy and angle of the transmitted and scattered light generated by the solar-induced chlorophyll fluorescence are calculated according to the reflectivity and transmittance of the material of the object that the solar-induced chlorophyll fluorescence contacts, and the simulation is repeated until all photons are absorbed, emitted to the sky, or completely dissipated. This is considered as the completion of the backward ray tracing, and the numerical distribution of chlorophyll fluorescence in the local area is obtained.

[0045] The numerical distribution of chlorophyll fluorescence in the local area is combined with the energy ratio of fluorescence emitted by all leaves in the canopy obtained by forward ray tracing to calculate the numerical distribution of chlorophyll fluorescence inside the entire canopy, and finally the total chlorophyll fluorescence emitted by all vegetation is calculated.

[0046] The present invention also provides an electronic device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the described method.

[0047] The present invention also provides a storage medium storing a computer program or instruction. When the computer program or instruction is run on a computer, the steps of the method described are executed.

[0048] Solar induced chlorophyll fluorescence has a direct relationship with photosynthesis and has been considered in recent years to be a powerful way to estimate photosynthesis efficiency and vegetation productivity. The present invention, through a completely new approach, has realized a method for directly inverting total chlorophyll fluorescence from observational data, laying the foundation for accurate and simple estimation of vegetation productivity.

[0049] Beneficial effects: Compared with the escape coefficient method and the existing ray tracing method commonly used in current inversion, the present invention can more accurately achieve the inversion of total chlorophyll fluorescence in the canopy. The versatility of this solution has been significantly improved compared with previous solutions, and it can be directly applied to various types of vegetation. Unlike previous solutions that require the use of various vegetation physiological parameters during the inversion process, this solution only uses observation data during the inversion process, filling the gap from direct observation to inversion method. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic diagram of the interaction between a single photon and a leaf.

[0051] Figure 2 This is a schematic diagram of a ray tracing simulation.

[0052] Figure 3 This is a schematic diagram of forward ray tracing.

[0053] Figure 4 This is a diagram of backward ray tracing.

[0054] Figure 5 It is a simulation result verification diagram.

[0055] Figure 6 It is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0056] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments, and the above and / or other advantages of the present invention will become more clear.

[0057] like Figure 6 As shown, the embodiment of the present invention provides an inversion method for total chlorophyll fluorescence in a farmland canopy based on ray tracing technology, comprising the following steps:

[0058] Step 1, perform three-dimensional reconstruction of the canopy based on point cloud;

[0059] In order to obtain the triangular vegetation model containing material information required for ray tracing simulation, it is necessary to perform single plant extraction, point cloud completion, and three-dimensional reconstruction on the pre-processed vegetation point cloud data obtained through actual scanning, and recombine the reconstructed three-dimensional modeling to reconstruct the three-dimensional structure in the area.

[0060] Step 2, extracting individual plants from point clouds;

[0061] In order to extract the point cloud of each crop from the farmland point cloud, the YOLOV8 deep learning model is first used to segment the full-color overhead image obtained during the point cloud collection process. The architecture of YOLOv8 consists of a backbone network (Backbone), a feature pyramid network (FPN) and a detection head (Head), and each part works together to achieve target detection. The backbone network is responsible for extracting features from the input image. It is usually composed of multiple convolutional layers, gradually reducing the spatial dimension to obtain multi-level information of the image. The feature pyramid network fuses features of different scales, combines high-level abstract features and low-level detail features, so that the network can better handle targets of different sizes. On this basis, the path aggregation network (PANet) enhances the bottom-up propagation of features, ensures that low-level features can be effectively transmitted to the output layer, and improves the accuracy of detection. The detection head decodes these fused features and outputs the category and location of the target. The detection head can adopt an anchor frame or an anchor-free frame design to perform target classification and regression operations according to specific needs.

[0062] Step 3: complete the point cloud;

[0063] Since the point cloud data obtained from equipment such as LiDAR is often missing, for vegetation, the point cloud obtained by airborne LiDAR such as drones and airplanes often lacks point cloud data below the canopy. In order to obtain the complete point cloud required for 3D reconstruction, the PF-NET deep learning model is selected as a means to achieve vegetation point cloud completion. Figure 1As shown in the figure, PF-Net aims to predict missing point clouds through a multi-scale generative network. Its architecture includes a multi-resolution encoder (MRE), a point pyramid decoder (PPD), and a discriminative network. First, iterative farthest point sampling (IFPS) is used to extract skeleton points from the point cloud to capture shape features, which is more efficient than random sampling and can better represent the distribution of point sets. In the multi-resolution encoder, a combined multi-layer perceptron (CMLP) is used to extract features, and the final latent vector F is formed by connecting multi-dimensional feature vectors, which contains high- and low-level information. The point pyramid decoder takes the final feature vector V as input to generate a point cloud of the missing area, and uses a hierarchical structure to predict point clouds at different resolutions, gradually refining geometric details. The loss function consists of a multi-stage completion loss and an adversarial loss, where the completion loss measures the difference between the predicted and true point clouds through the chamfer distance (CD), and the adversarial loss draws on the generative adversarial network (GAN) to improve the authenticity of the predicted point cloud.

[0064] Step 4: Perform 3D reconstruction based on point cloud;

[0065] Point cloud data cannot be directly applied to ray tracing. Therefore, the SIREN deep learning model is selected to realize the 3D reconstruction of vegetation. The core idea of ​​SIREN can be expressed as follows:

[0066] (1)

[0067] Where x is the 3D coordinate of a point in the input point cloud. It is an implicit function that the neural network needs to learn. In the work of point cloud 3D reconstruction, it can be understood as the distance from point x to the surface of the object. Through this function, the position of the object surface can be obtained. for The gradient of can be understood as a function that characterizes the curvature of the object surface in point cloud 3D reconstruction, which is usually related to the normal vector of the object surface. F is the constraint function.

[0068] In order to better capture the complex surfaces, holes and details of objects, SIREN uses the sine activation function as the implicit function:

[0069] (2)

[0070] In the formula, and are the weight matrix and bias vector of the network respectively. n is the total number of layers in the neural network, and i is the current number of layers. In each layer, a sinusoidal activation function is used. , which enables the network to effectively capture high-frequency details in point clouds, such as small details and complex shapes on the surface of objects.

[0071] After learning the implicit function, SIREN tries to minimize the loss function, which penalizes the This is done by deviating from the original constraints:

[0072] (3)

[0073] Where L is the loss function, and the goal of grid training is to minimize this function so that the learned Able to represent the surface of an object as accurately as possible. The currently defined three-dimensional space area. is the mth constraint.

[0074] Step 5, obtaining the chlorophyll fluorescence ratio distribution in the canopy by forward ray tracing;

[0075] Different from the existing photosynthetically active radiation distribution inversion algorithm based on ray tracing, since chlorophyll fluorescence is emitted by all plants, and actual observation can only obtain the fluorescence emission of some areas, it is very difficult to obtain the fluorescence distribution by tracing rays in only one direction. Therefore, based on the principle that chlorophyll fluorescence is excited by the photosynthetic system and there is a direct relationship between the photosynthetic system and photosynthetically active radiation, this method first simulates the photosynthetically active radiation in the vegetation canopy, and determines the chlorophyll fluorescence distribution in the canopy based on the distribution of photosynthetically active radiation in the canopy.

[0076] When simulating photosynthetically active radiation, forward ray tracing technology is selected as the simulation method. Forward ray tracing is to emit photons from the light source (the sun in this invention), and when the photons touch the scene, they interact and calculate reflection, absorption, and transmission phenomena until the light intersects with the observer or dissipates in the scene. Usually, ray tracing simulates the emission of a very large number of photons, thereby achieving high-precision simulation in the environment.

[0077] like Figure 1 As shown in the figure, during the vegetation modeling process, each triangular face will be defined as a front leaf or a back leaf, and different faces will be defined with different reflectivity and transmittance. When a photon interacts with a leaf, the energy of the incident photon will be divided into three parts: reflected photon energy, transmitted photon energy, and absorbed photon energy. The calculation method of the three is shown in the following formula:

[0078] (4)

[0079] (5)

[0080] (6)

[0081] in, , , The energy of generated photons, transmitted photons and absorbed photons generated when photons e contact the blades; is the total energy of the incident photon e; and is the reflectivity and transmittance of the current leaf.

[0082] The logic diagram for performing forward ray tracing is as follows: Figure 2 and Figure 3 As shown. First, a large number of photons are emitted from the sun. After each photon contacts a leaf or the ground, the energy and angle of the transmitted and scattered light generated are calculated according to the reflectivity and transmittance of the material of the object it contacts, and the simulation is repeated until all photons are absorbed, emitted to the sky, or completely dissipated, and the forward ray tracing is considered complete. After the photosynthetically active radiation simulation is completed, the energy of the absorbed photons of each leaf is calculated to obtain the distribution of photosynthetically active radiation energy absorbed by each leaf in the entire canopy. Existing studies have shown that there is a significant positive correlation between photosynthetically active radiation and solar-induced chlorophyll fluorescence. Based on this theory, this method selects the leaf triangle that absorbs the most photosynthetically active radiation in the environment, records its emitted fluorescence energy as 1 standard unit, and then confirms the ratio of its emitted fluorescence energy based on the ratio of the photosynthetically active radiation energy absorbed by each leaf triangle to the maximum value.

[0083] Step 6, obtaining the chlorophyll fluorescence value distribution in the canopy by backward ray tracing;

[0084] Since the positive correlation between photosynthetically active radiation and solar-induced chlorophyll fluorescence is affected by plant type and meteorological conditions, directly using the correlation to calculate the fluorescence value will cause large errors. To overcome this difficulty, this method uses backward ray tracing combined with chlorophyll fluorescence observation to determine the chlorophyll fluorescence value distribution and total chlorophyll fluorescence value in the canopy.

[0085] Backward ray tracing is to emit photons from the observer (the observation probe in this invention), and the photons interact and calculate reflection, absorption, and transmission phenomena when they touch the scene until the light dissipates in the scene.

[0086] like Figure 4As shown in the figure, after receiving photosynthetically active radiation, plants will emit solar-induced chlorophyll fluorescence, as shown by the red line, and part of the fluorescence is captured by the sensor and becomes the observed SIF. Since the light path is reversible, photons with the same energy and opposite directions are emitted from the sensor. The energy and angle of the generated transmitted light and scattered light are calculated based on the reflectivity and transmittance of the material of the object in contact and simulated again until all photons are absorbed, emitted to the sky, or completely dissipated, and the backward ray tracing is considered complete. It is worth noting that due to the weak fluorescence energy, the energy absorbed by the leaves can be regarded as dissipated.

[0087] After the execution is completed, the numerical distribution of chlorophyll fluorescence in the local area can be obtained. Since the vegetation types in the farmland are exactly the same, it can be assumed that the correlation between photosynthetically active radiation and fluorescence of each crop is consistent. Combined with the energy ratio of fluorescence emitted by all leaves in the canopy obtained by forward ray tracing, the numerical distribution of chlorophyll fluorescence inside the entire canopy is obtained, and the total chlorophyll fluorescence emitted by all vegetation is finally calculated.

[0088] Result evaluation: To verify the simulation results, a test field was selected as the verification site. First, the point cloud and full-color data of the sample plot were captured using a drone-mounted laser radar. The point cloud data was preprocessed and the yoloV8, PF-NET, and SIREN models were used to gradually realize the three-dimensional reconstruction of the point cloud. After simulating the fluorescence distribution in the sample plot through the ray tracing algorithm, fixed sample points were selected to compare the simulation and measured results. Figure 5 As shown ( Figure 5 The letters on the vertical axis are solar-induced chlorophyll fluorescence units, which means milliwatts per square meter per nanometer per solid angle in Chinese. It describes the radiation power per unit area, per unit wavelength, and per unit solid angle). The simulation results of the method in this embodiment are highly consistent with the observation results, and can better simulate the fluorescence distribution in the canopy.

[0089] The present invention provides an inversion method for total chlorophyll fluorescence in a farmland canopy based on ray tracing technology. There are many methods and ways to implement the technical solution. The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A method for inverting total chlorophyll fluorescence in a farmland canopy based on ray tracing technology, characterized in that: The following steps are involved: Step 1, perform three-dimensional reconstruction of the canopy based on point cloud; Step 2, obtaining the chlorophyll fluorescence ratio distribution in the canopy by forward ray tracing; Step 3, obtaining the chlorophyll fluorescence value distribution in the canopy by backward ray tracing; Step 2 includes: simulating photosynthetically active radiation in the vegetation canopy, and determining the chlorophyll fluorescence distribution in the canopy based on the distribution of photosynthetically active radiation in the canopy; When simulating photosynthetically active radiation, forward ray tracing technology is selected as the simulation method. Forward ray tracing is to emit photons from the light source. When the photons touch the scene, they interact and calculate reflection, absorption and transmission phenomena until the light intersects with the observer or dissipates in the scene. During vegetation modeling, each triangular face is defined as a front leaf or a back leaf, and different faces are defined with different reflectances and transmittances. When a photon interacts with a leaf, the energy of the incident photon is divided into three parts: reflected photon energy, transmitted photon energy, and absorbed photon energy. The calculation formula is: Q ref (and)=Q in (e)·ρ(e), Q tran (and)=Q in (e)·τ(e), Q abs (and)=Q in (e)·(1-ρ(e)-τ(e)), Among them, Q ref (e), Q tran (e), Q abs (e) are the energy of generated photon, transmitted photon and absorbed photon when photon e contacts the blade; Q in (e) is the total energy of photon e when it is incident; ρ(e) and τ(e) are the reflectivity and transmittance of the current leaf, respectively; The process of performing forward ray tracing includes: emitting a large number of photons from the sun, and after each photon hits a leaf or the ground, the energy and angle of the transmitted and scattered light are calculated based on the reflectivity and transmittance of the material of the object it hits, and then simulated again until all photons are absorbed, emitted to the sky, or completely dissipated, and the forward ray tracing is considered complete; After executing the photosynthetically active radiation simulation, the absorbed photon energy of each leaf is calculated to obtain the photosynthetically active radiation energy distribution absorbed by each leaf in the entire canopy; In step 2, the leaf triangle that absorbs the most photosynthetically active radiation in the environment is selected, and the fluorescence energy emitted by the leaf triangle is recorded as 1 standard unit. Then, according to the ratio of the photosynthetically active radiation energy absorbed by each leaf triangle and the maximum value, the ratio of the fluorescence energy emitted by each leaf triangle is determined; Step 3 includes: using backward ray tracing combined with observed chlorophyll fluorescence to determine the chlorophyll fluorescence value distribution and total chlorophyll fluorescence value within the canopy; Backward ray tracing is to send photons from the observer, and the photons interact with the scene when they touch it and calculate reflection, absorption and transmission phenomena until the light dissipates in the scene; After receiving photosynthetically active radiation, plants will emit solar-induced chlorophyll fluorescence. The energy and angle of the transmitted and scattered light generated by the solar-induced chlorophyll fluorescence are calculated according to the reflectivity and transmittance of the material of the object that the solar-induced chlorophyll fluorescence contacts, and the simulation is repeated until all photons are absorbed, emitted to the sky, or completely dissipated. This is considered as the completion of the backward ray tracing, and the numerical distribution of chlorophyll fluorescence in the local area is obtained. The numerical distribution of chlorophyll fluorescence in the local area is combined with the energy ratio of fluorescence emitted by all leaves in the canopy obtained by forward ray tracing to calculate the numerical distribution of chlorophyll fluorescence inside the entire canopy, and finally the total chlorophyll fluorescence emitted by all vegetation is calculated.

2. The method according to claim 1, characterized in that Step 1 includes: Step 1-1, extracting individual plants from point clouds; Step 1-2, complete the point cloud; Steps 1-3: perform 3D reconstruction based on point cloud.

3. The method according to claim 2, characterized in that Step 1-1 includes: using the yoloV8 deep learning model to segment the full-color overhead image obtained in the process of collecting point clouds, and extracting the point cloud of each crop from the farmland point cloud.

4. The method according to claim 3, characterized in that Step 1-2 includes: selecting the PF-NET deep learning model for point cloud completion, the PF-NET deep learning model includes a multi-resolution encoder, a point pyramid decoder and a discriminant network, first using iterative farthest point sampling to extract skeleton points from the point cloud to capture shape features, in the multi-resolution encoder, using a combined multi-layer perceptron to extract features, and forming a final latent vector F by connecting multi-dimensional feature vectors, the point pyramid decoder uses the final feature vector V as input to generate a point cloud of the missing area, and uses a hierarchical structure to predict point clouds at different resolutions, gradually refining geometric details, and the loss function includes multi-stage completion loss and adversarial loss.

5. The method according to claim 4, characterized in that Steps 1-3 include: using the SIREN deep learning model to achieve 3D reconstruction of vegetation, the formula is: Where x is the three-dimensional coordinate of a point in the input point cloud; φ is the implicit function that the neural network needs to learn; φ(x) is the implicit function with x as the variable; is the gradient of φ(x); is the nth order gradient of φ(x), n is the highest order gradient number, which is determined according to the actual reconstruction accuracy; F is the constraint function; φ:x→φ(x) means mapping the three-dimensional coordinate x to φ(x); The SIREN deep learning model uses a sinusoidal activation function as an implicit function: φ(x)=W n (f n-1 …φ2…φ0)(x)+b n , Where W n and b n are the weight matrix and bias vector of the i-th layer of the neural network respectively; n is the total number of layers of the neural network, i is the current number of layers; φ0 is the implicit function to be learned by the 0th layer of the neural network; φ n-1 is the implicit function to be learned by the n-1th layer of the neural network; in each layer of the neural network, the sinusoidal activation function sin(W i x i +b i ); After learning the implicit function, SIREN minimizes the loss function by penalizing deviations from the original constraints on the input domain: Where L is the loss function; Ω is the currently defined three-dimensional space area; C m is the mth constraint, Ω m is the three-dimensional space region under the mth constraint; a(x) is related to the characteristics of the three-dimensional coordinate x and is used to define the constraint condition C m Parameters; M is the total number of constraints.

6. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.

7. A storage medium, characterized in that: A computer program or instruction is stored, and when the computer program or instruction is run on a computer, the steps of the method according to any one of claims 1 to 5 are executed.

Images