Carbon sink dynamic prediction method, device and equipment based on multi-source remote sensing space-time fusion and three-dimensional point cloud deep learning, and storage medium

Abstract

The invention discloses a carbon sink dynamic prediction method, device and equipment based on multi-source remote sensing space-time fusion and three-dimensional point cloud deep learning and a storage medium, and relates to the technical field of remote sensing information processing and ecological environment monitoring, and the method comprises the steps: obtaining space-time fusion data, a vegetation point cloud inversion algorithm and a carbon sink prediction hybrid model; extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the space-time fusion data, and determining vegetation parameter information; and predicting a carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model, determining a carbon sink prediction result, and controlling a system to complete disaster assessment and risk early warning based on the carbon sink prediction result. According to the method, multi-source remote sensing space-time fusion is carried out to eliminate the space-time resolution difference, the three-dimensional point cloud biomass is extracted, the carbon sink change trend is predicted to realize disaster assessment and risk early warning, the multi-source data fusion precision and the vegetation parameter inversion accuracy are effectively improved, and the method has remarkable environmental benefits and social values.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing information processing and ecological environment monitoring, and particularly relates to a carbon sink dynamic prediction method, device, equipment and storage medium based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. Background Art

[0002] With the intensification of global climate change, carbon sink monitoring is of great significance for evaluating the ecosystem service function and formulating strategies to address climate change. As important carbon sink resources, the dynamic changes of ecosystems such as forests, farmlands and wetlands will directly affect the balance of carbon absorption and release. Therefore, it is necessary to evaluate the carbon sink capacity of ecosystems in real time for ecological protection and restoration.

[0003] Currently, carbon sink monitoring in existing practices mainly relies on traditional remote sensing technologies and ground observation networks. In terms of remote sensing data processing, multi-source data such as optical images, LiDAR point clouds and radar data are usually fused to improve the monitoring accuracy. In terms of vegetation parameter inversion, the region growing algorithm is used for point cloud segmentation and biomass estimation. At the same time, a carbon sink static model is used to respond to climate change and ecosystem disturbances.

[0004] However, in existing practices, it is difficult to effectively solve the differences in spatio-temporal resolution of multi-source data based on simple interpolation or alignment techniques. The spatio-temporal resolution differences are large, resulting in fusion errors, which cannot meet the requirements of high-precision carbon sink monitoring. Moreover, the vegetation parameter inversion accuracy is low, the traditional point cloud segmentation method has poor adaptability to complex vegetation structures, with large estimation errors. At the same time, the carbon sink model used lacks dynamics and is difficult to reflect the changes of ecosystems in real time. Therefore, how to dynamically perform high-precision carbon sink monitoring through multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning technology has become an urgent problem to be solved.

[0005] The above content is only used to assist in understanding the technical solution of the present application, and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main purpose of the present application is to provide a carbon sink dynamic prediction method, device, equipment and storage medium based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning, aiming to solve the technical problem of how to dynamically perform high-precision carbon sink monitoring through multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning technology.

[0007] To achieve the above purpose, the present application proposes a carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. The method includes:

[0008] Obtain spatio-temporal fusion data, a vegetation point cloud inversion algorithm and a carbon sink prediction hybrid model;

[0009] Extract the three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and determine the vegetation parameter information;

[0010] Predict the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model, determine the carbon sink prediction result, and control the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

[0011] In one embodiment, the steps of obtaining the spatio-temporal fusion data, the vegetation point cloud inversion algorithm, and the carbon sink prediction hybrid model include:

[0012] Obtain optical images, point cloud remote sensing monitoring data, radar data, adaptive sampling algorithm, feature fusion algorithm, biomass constraint loss algorithm, light energy correction model, and meteorological correction model;

[0013] Extract the spatio-temporal fusion details based on the optical image, the point cloud remote sensing monitoring data, and the radar data, and determine the spatio-temporal fusion data;

[0014] Determine the vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass constraint loss algorithm;

[0015] Determine the carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model.

[0016] In one embodiment, the steps of extracting the spatio-temporal fusion details based on the optical image, the point cloud remote sensing monitoring data, and the radar data, and determining the spatio-temporal fusion data include:

[0017] Extract the spatio-temporal fusion details based on the optical image, the point cloud remote sensing monitoring data, and the radar data, and construct a spatio-temporal pyramid, where the spatio-temporal pyramid includes an optical image pyramid, a point cloud remote sensing monitoring data pyramid, and a radar data pyramid;

[0018] Extract the spatio-temporal features based on the optical image pyramid, the point cloud remote sensing monitoring data pyramid, and the radar data pyramid, calculate the attention weight set, align the spatio-temporal coordinate system based on the attention weight set, and calculate the spatio-temporal alignment data;

[0019] Perform super-resolution reconstruction on the spatio-temporal alignment data and calculate the reconstruction loss to obtain the spatio-temporal fusion data.

[0020] In one embodiment, the steps of performing super-resolution reconstruction on the spatio-temporal alignment data and calculating the reconstruction loss to obtain the spatio-temporal fusion data include:

[0021] Obtain the cascaded operation mode and the loss function, where the loss function includes a reconstruction loss function, an alignment loss function, and a total loss function;

[0022] Upsample and extract super-resolution details based on the spatio-temporal alignment data to determine the upsampled data;

[0023] Calculate the cascaded data based on the cascaded operation mode and the upsampled data;

[0024] Calculate the spatio-temporal fusion data based on the reconstruction loss function, the alignment loss function, the total loss function, and the cascaded data.

[0025] In one embodiment, the step of extracting the three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine the vegetation parameter information includes:

[0026] Extract the three-dimensional point cloud biomass based on the adaptive sampling algorithm in the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, calculate the vegetation sampling weight, and determine the key points in the vegetation area. The three-dimensional point cloud biomass includes ground biomass, underground biomass, and litter;

[0027] Construct a nested local area based on the feature fusion algorithm in the vegetation point cloud inversion algorithm and the key points in the vegetation area to determine the vegetation fusion data;

[0028] Calculate the biomass growth loss based on the biomass limitation loss algorithm in the vegetation point cloud inversion algorithm and the vegetation fusion data to obtain the vegetation parameter information.

[0029] In one embodiment, the step of predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result includes:

[0030] Obtain the flux tower data;

[0031] Adjust the vegetation parameter information based on the flux tower data to determine the optimized vegetation parameter information;

[0032] Input the optimized vegetation parameter information into the carbon sink prediction hybrid model for training to determine the carbon sink prediction hybrid optimized model;

[0033] Quantify the defects of the remote sensing data based on the carbon sink prediction hybrid optimized model and predict the carbon sink change trend to obtain the carbon sink prediction result.

[0034] In one embodiment, the step of controlling the system to complete the disaster assessment and risk warning based on the carbon sink prediction result includes:

[0035] Analyze the carbon sink prediction result to identify the disaster risk, and determine the carbon loss map and the recovery strategy report;

[0036] Based on the carbon loss map and the recovery strategy report, the control system completes disaster assessment and risk warning.

[0037] In addition, to achieve the above object, the present application also proposes a carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning. The carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning includes:

[0038] An acquisition module, configured to acquire spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model;

[0039] A processing module, configured to extract 3D point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and determine vegetation parameter information;

[0040] An execution module, configured to predict the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model, determine the carbon sink prediction result, and based on the carbon sink prediction result, the control system completes disaster assessment and risk warning.

[0041] In addition, to achieve the above object, the present application also proposes a carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning. The device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The computer program is configured to implement the steps of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning as described above.

[0042] In addition, to achieve the above object, the present application also proposes a storage medium. The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning as described above.

[0043] One or more technical solutions proposed by the present application have at least the following technical effects:

[0044] A carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning proposed in this embodiment obtains spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; extracts 3D point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; predicts the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and controls the system based on the carbon sink prediction result to complete disaster assessment and risk warning. By obtaining spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model, this application uses the vegetation point cloud inversion algorithm and spatio-temporal fusion data to extract 3D point cloud biomass, determines vegetation parameter information, combines the carbon sink prediction hybrid model to predict the carbon sink change trend, and realizes disaster assessment and risk warning based on the prediction result, effectively improving the accuracy of multi-source data fusion and the accuracy of vegetation parameter inversion, enhancing the carbon sink dynamic prediction ability, quickly responding and giving risk warnings after ecosystem disturbances, and improving the practicality of carbon sink monitoring, with significant environmental benefits and social values. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings here are incorporated into the specification and form a part of this specification, showing embodiments consistent with this application, and are used together with the specification to explain the principles of this application.

[0046] To more clearly illustrate the technical solutions in the embodiments of this application or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0047] Figure 1 It is a schematic flowchart provided for Embodiment 1 of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning of this application;

[0048] Figure 2 It is a schematic flowchart provided for Embodiment 2 of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning of this application;

[0049] Figure 3 It is a schematic module structure diagram of the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning for the embodiments of this application;

[0050] Figure 4 It is a schematic device structure diagram of the hardware operating environment involved in the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning for the embodiments of this application.

[0051] The implementation of the purpose, functional features, and advantages of this application will be further described with reference to the embodiments and the drawings. Specific Embodiments

[0052] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not used to limit this application.

[0053] To better understand the technical solutions of this application, the following will be described in detail in combination with the specification drawings and specific embodiments.

[0054] The main solution of the embodiments of this application is: obtaining spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and controlling the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

[0055] In this embodiment, for the convenience of description, the following will be described with the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning as the execution subject.

[0056] Due to the existing technology based on simple interpolation or alignment techniques, it is difficult to effectively solve the differences in spatio-temporal resolution of multi-source data. The spatio-temporal resolution difference is large, resulting in fusion errors, and it cannot meet the requirements of high-precision carbon sink monitoring. Moreover, the inversion accuracy of vegetation parameters is low, the traditional point cloud segmentation method has poor adaptability to complex vegetation structures, with large estimation errors. At the same time, the carbon sink model used lacks dynamics and is difficult to reflect the changes of the ecosystem in real time.

[0057] This application provides a solution: obtaining spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and controlling the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

[0058] As can be seen from the above embodiments, this application obtains spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model, extracts three-dimensional point cloud biomass using the vegetation point cloud inversion algorithm and spatio-temporal fusion data to determine vegetation parameter information, combines the carbon sink prediction hybrid model to predict the carbon sink change trend, and realizes disaster assessment and risk warning based on the prediction result, effectively improving the fusion accuracy of multi-source data and the accuracy of vegetation parameter inversion, enhancing the carbon sink dynamic prediction ability, making a rapid response and risk warning after the ecosystem is disturbed, and improving the practicality of carbon sink monitoring, with significant environmental benefits and social values.

[0059] Based on this, the embodiments of the present application provide a carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. Referring to Figure 1 , Figure 1 is a schematic flowchart of the first embodiment of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning of the present application.

[0060] In this embodiment, the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning includes steps S10 to S30:

[0061] Step S10, obtain spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model;

[0062] It should be noted that the spatio-temporal fusion data reflects the characteristics of high-precision alignment and fusion of the extracted multi-source remote sensing data in spatio-temporal resolution. The vegetation point cloud inversion algorithm reflects the characteristics of measuring the high-precision segmentation and biomass inversion ability of complex vegetation structures. The carbon sink prediction hybrid model reflects the characteristics of being used for carbon sink dynamic prediction modeling and adapting to environmental changes and ecosystem disturbances.

[0063] It can be understood that the spatio-temporal fusion data is generated by constructing a spatio-temporal pyramid, performing hierarchical sampling and feature extraction on optical images, LiDAR point clouds, and radar data, and using an attention mechanism to align the spatio-temporal coordinate system, thereby generating high-resolution fusion data through super-resolution reconstruction, eliminating the spatio-temporal resolution differences of multi-source data, and significantly improving the accuracy and spatial alignment accuracy of the fusion data. The vegetation point cloud inversion algorithm can efficiently segment the single-tree canopy and extract three-dimensional parameters of vegetation, such as tree height, crown diameter, and volume, through an improved PointNet++ network, combined with hierarchical adaptive sampling, multi-scale feature fusion, and a biomass geometric constraint loss function, significantly improving the processing efficiency and accuracy of vegetation point clouds, adapting to complex canopy structures, and effectively reducing biomass estimation errors. The carbon sink prediction hybrid model couples a physical module based on the principle of light use efficiency and an AI module composed of a long short-term memory network, learns the dynamic non-linear relationship between meteorological factors and carbon sinks in real time, and performs online optimization through flux tower data, and supports online incremental learning and real-time calibration, adapting to environmental changes and ecosystem disturbances, thereby providing high-precision and high-reliability modeling for carbon sink dynamic prediction.

[0064] For the convenience of understanding, taking the acquisition of spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model as an example for illustration, where the information acquisition device is an information acquisition module, and the storage device is a memory.

[0065] The information acquisition module obtains optical images, point cloud remote sensing monitoring data, radar data, adaptive sampling algorithms, feature fusion algorithms, biomass limitation loss algorithms, light energy correction models, and meteorological correction models. Based on the optical images, the point cloud remote sensing monitoring data, and the radar data, it extracts spatio-temporal fusion details, determines spatio-temporal fusion data, determines a vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass limitation loss algorithm, determines a carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model, and performs subsequent processing based on the spatio-temporal fusion data, the vegetation point cloud inversion algorithm, and the carbon sink prediction hybrid model.

[0066] Step S20: Extract three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and determine vegetation parameter information;

[0067] It should be noted that the vegetation parameter information reflects the characteristics of multi-dimensional parameters of the three-dimensional point cloud biomass extracted from the spatio-temporal fusion data through the vegetation point cloud inversion algorithm.

[0068] It can be understood that the three-dimensional point cloud biomass refers to the total amount of biological organisms per unit area or unit volume within a specific time, usually expressed in dry weight, with the unit of tons per hectare, t / ha. It can consist of above-ground biomass, below-ground biomass, and litter. Among them, the above-ground biomass is the above-ground parts such as the stems, leaves, and branches of vegetation, the below-ground biomass is the below-ground parts such as the roots, and the litter is the fallen branches, leaves, and dead plants.

[0069] In addition, it should be noted that the three-dimensional point cloud biomass is the main carrier of the carbon sink. Vegetation converts CO2 in the atmosphere into organic carbon through photosynthesis and stores it in the biomass. The carbon content in the biomass usually accounts for 45% - 50% of its dry weight. Therefore, the carbon storage can be estimated through the biomass. For example, if the biomass of a certain forest is 100 t / ha and the carbon content coefficient is 0.47, then the carbon storage is 47 tC / ha. Vegetation growth will increase the carbon sink capacity, such as afforestation and forest restoration, while vegetation destruction will reduce the carbon sink capacity, such as deforestation and fires, and even convert it into a carbon source, such as releasing CO2. Areas with high biomass usually have a higher carbon sink capacity, such as tropical rainforests, and the seasonal changes in biomass will affect the annual fluctuations of the carbon sink, such as the growth season and the dormant season. The biomass can be inverted through vegetation indices, such as NDVI, or through LiDAR point cloud inversion, or calculated through quadrat measurements or allometric equations.

[0070] For easy understanding, taking the determination of vegetation parameter information as an example, the information acquisition device is the information acquisition module, the storage device is the memory, and the processing device is the processing module.

[0071] The information acquisition module obtains the vegetation point cloud inversion algorithm and spatio-temporal fusion data, extracts the three-dimensional point cloud biomass based on the adaptive sampling algorithm in the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, calculates the vegetation sampling weight, and determines the key points in the vegetation area. The three-dimensional point cloud biomass includes ground biomass, underground biomass, and litter. That is, using the adaptive sampling algorithm, hierarchical adaptive sampling HAS, dynamically adjusts the sampling density according to the point cloud density gradient. For example, the sampling interval is small in the dense canopy area and large in the sparse ground area, and combined with LiDAR elevation data, preferentially retains the key points in the high vegetation area, expressed as:

[0072]

[0073] Among them, S i is the sampling weight of the i-th point, α is the weight coefficient of the local density term, σ i is the local density of the i-th point, β is the weight coefficient of the height term, H i is the height value of the i-th point.

[0074] Based on the feature fusion algorithm in the vegetation point cloud inversion algorithm and the key points in the vegetation area, a nested local area is constructed to determine the vegetation fusion data. That is, a nested local area is constructed at each level, such as spheres with radii of 0.5m, 1m, and 2m, and multi-scale features are extracted in parallel. The attention mechanism is used to dynamically weight and fuse different-scale features, multi-scale feature fusion MSFF, expressed as:

[0075] F fusion = Attention(F 0.5m , F 1m , F 2m )

[0076] Among them, F fusion is the fused multi-scale feature, Attention is the attention mechanism for dynamically weighting features of different scales, F 0.5m is the feature at the 0.5m scale, F 1m is the feature at the 1m scale, F 2m is the feature at the 2m scale.

[0077] And replace the MLP with depthwise separable convolution to reduce the computational amount by 30%, introduce skip connections and upsampling layers to restore the point cloud details.

[0078] Based on the biomass limit loss algorithm in the vegetation point cloud inversion algorithm and the vegetation fusion data, calculate the biomass growth loss to obtain the vegetation parameter information. That is, add a biomass geometric limit term to the loss function, biomass geometric limit loss, expressed as:

[0079]

[0080] Among them, is the total loss, is the segmentation loss, used to measure the accuracy of point cloud segmentation. λ is the weight coefficient of the geometric constraint term, H pred is the predicted tree height, k is the empirical coefficient related to the tree species, D pred is the predicted crown diameter.

[0081] And biomass calculation is carried out to obtain vegetation parameter information, expressed as:

[0082] B = ρ·V + ∈

[0083] Among them, ρ is the wood density, V is the volume of a single tree, and ∈ is the foliage biomass correction term.

[0084] Subsequent processing is carried out based on the vegetation parameter information.

[0085] In a feasible implementation manner, step S20 may include steps A11 to A13:

[0086] Step A11, based on the adaptive sampling algorithm in the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, extract the three-dimensional point cloud biomass, calculate the vegetation sampling weight, and determine the key points in the vegetation area. The three-dimensional point cloud biomass includes ground biomass, underground biomass, and litter;

[0087] It should be noted that the key points in the vegetation area reflect the characteristics of the representative positions in the vegetation point cloud data.

[0088] It can be understood that the key points in the vegetation area can effectively capture the morphological characteristics of the vegetation, such as the crown, trunk, and foliage distribution, so as to efficiently represent the three-dimensional structure of the vegetation, while reducing redundant information and improving the calculation efficiency.

[0089] Step A12, based on the feature fusion algorithm in the vegetation point cloud inversion algorithm and the key points in the vegetation area, construct a nested local area and determine the vegetation fusion data;

[0090] It should be noted that the vegetation fusion data reflects the characteristics of the comprehensive information after multi-scale feature fusion of the vegetation point cloud.

[0091] It can be understood that the vegetation fusion data can capture the local details and overall structure of the vegetation point cloud, such as foliage texture, crown, and trunk morphology, and dynamically weight the features of different scales through the attention mechanism to enhance the adaptability to complex vegetation structures.

[0092] Step A13, based on the biomass constraint loss algorithm in the vegetation point cloud inversion algorithm and the vegetation fusion data, calculate the biomass growth loss to obtain the vegetation parameter information.

[0093] It can be understood that the vegetation parameter information may include ground biomass, underground biomass, litter, tree height, crown diameter, and volume, so as to comprehensively describe the physical form and health status of vegetation. And through the biomass limitation loss algorithm, the accuracy of biomass estimation is ensured, which conforms to the physical laws of vegetation growth.

[0094] Step S30: Based on the vegetation parameter information and the carbon sink prediction hybrid model, predict the carbon sink change trend, determine the carbon sink prediction result, and based on the carbon sink prediction result, control the system to complete disaster assessment and risk warning.

[0095] It should be noted that the carbon sink prediction result reflects the characteristics of the dynamic changes of carbon absorption and release in the ecosystem under different time scales and environmental conditions.

[0096] It can be understood that carbon sink refers to a system or process that absorbs and stores carbon dioxide (CO2) from the atmosphere through natural or artificial processes, usually expressed in carbon storage, with the unit of tons of carbon per hectare (tC / ha). And the types of carbon sinks can include forest carbon sink, soil carbon sink, and ocean carbon sink. Among them, forest carbon sink absorbs CO2 through vegetation photosynthesis, soil carbon sink stores carbon through the decomposition and mineralization processes of organic matter, and ocean carbon sink absorbs CO2 through marine organisms and water bodies.

[0097] In addition, it should be noted that carbon sink estimation can be carried out by using net primary productivity and carbon storage change. Among them, net primary productivity estimates the carbon absorption of vegetation through a light use efficiency model, such as CASA. Carbon storage change estimates the carbon sink capacity through a biomass dynamics and soil carbon model. For example, if the biomass of a certain forest is 150 t / ha, the carbon content coefficient can be calculated as 0.47, and the carbon storage is 70.5 tC / ha, so as to obtain the carbon sink capacity. Assuming the annual NPP is 10 tC / ha, the annual carbon sink capacity is 10 tC / ha. By monitoring the forest biomass dynamics, the carbon sink capacity can be evaluated to support the REDD+ project, that is, reducing emissions from deforestation and forest degradation. And the farmland carbon sink can be evaluated, that is, by estimating crop biomass, optimizing agricultural management, and increasing soil carbon storage. The regional carbon sink contribution can be estimated through the biomass growth potential, such as afforestation and vegetation restoration. And based on the biomass and carbon sink data, carbon credit projects can be developed, such as forest carbon sink trading.

[0098] For the convenience of understanding, taking the determination of the carbon sink prediction result as an example, the information collection device is the information collection module, the storage device is the memory, and the execution device is the execution module.

[0099] The information acquisition module obtains flux tower data, adjusts the vegetation parameter information based on the flux tower data, determines the vegetation optimization parameter information, inputs the vegetation optimization parameter information into the carbon sink prediction hybrid model for training to determine the carbon sink prediction hybrid optimization model, quantifies the remote sensing data defects and predicts the carbon sink change trend based on the carbon sink prediction hybrid optimization model to obtain the carbon sink prediction result, that is, proposes the CASA-LSTM hybrid model, calculates the net primary productivity NPP based on the principle of light use efficiency, introduces the red edge band to correct the photosynthetically active radiation PAR absorption rate, and uses the long short-term memory network LSTM to learn the dynamic non-linear relationship between meteorological factors such as temperature, precipitation, and CO2 concentration and the carbon sink. Among them, this hybrid model supports online incremental learning, dynamically optimizes parameters through the real-time data of the flux tower to adapt to environmental mutations, calibrates the model in real time with the flux tower data, feeds the error back to the online learning module, analyzes the carbon sink prediction result to identify disaster risks, determines the carbon loss map and the recovery strategy report, and controls the system based on the carbon loss map and the recovery strategy report to complete disaster assessment and risk warning, that is, quantifies the remote sensing data noise and the uncertainty of model parameters, and generates a confidence interval map of carbon sink estimation to support risk warning.

[0100] In a feasible implementation manner, step S30 may include steps B11 to B13:

[0101] Step B11, obtaining flux tower data;

[0102] It should be noted that the flux tower data reflects the real-time dynamic characteristics of the material and energy exchange between the ecosystem and the atmosphere.

[0103] It can be understood that the flux tower data measures the carbon, water, and energy fluxes of the ecosystem at different time scales through high-precision sensors, and can directly quantify the net ecosystem exchange, net primary productivity, and ecosystem respiration of vegetation, with the characteristics of high time resolution and high precision.

[0104] Step B12, adjusting the vegetation parameter information based on the flux tower data to determine the vegetation optimization parameter information;

[0105] It should be noted that the vegetation optimization parameter information reflects the high-precision characteristics of the vegetation biomass, growth state, and carbon absorption ability after being calibrated by the flux tower data.

[0106] It can be understood that the vegetation optimization parameter information is more accurately characterized by the real-time calibration of the flux tower data, representing the actual performance of vegetation under different environmental conditions, and supporting the high-precision operation of the carbon sink prediction model.

[0107] Step B13, inputting the vegetation optimization parameter information into the carbon sink prediction hybrid model for training to determine the carbon sink prediction hybrid optimization model;

[0108] It should be noted that the carbon sink prediction hybrid optimization model reflects the characteristics of an optimized carbon sink dynamic change prediction model that adapts to environmental changes and ecosystem disturbances.

[0109] It can be understood that the carbon sink prediction hybrid optimization model can learn in real time the complex non-linear relationships between ecosystem carbon sinks and environmental factors such as temperature, precipitation, and CO2 concentration, support online incremental learning, dynamically adjust parameters, and thus provide carbon sink prediction results with high spatio-temporal resolution.

[0110] Step B14, quantify the defects of remote sensing data based on the carbon sink prediction hybrid optimization model and predict the carbon sink change trend to obtain the carbon sink prediction result.

[0111] It can be understood that the carbon sink prediction result can characterize the dynamic changes of carbon absorption and release in the ecosystem on a future time scale, reveal the carbon sink capacity of the ecosystem under different environmental conditions, identify the response mechanisms of the ecosystem to climate change, natural disasters, and human activities, and has significant environmental benefits and social value.

[0112] In another feasible implementation, step S30 may include steps C11 - C12:

[0113] Step C11, analyze the carbon sink prediction result to identify disaster risks and determine the carbon loss map and recovery strategy report;

[0114] It should be noted that the carbon loss map reflects the characteristics of the loss distribution of carbon storage in the ecosystem after a specific disturbance event, and the recovery strategy report reflects the characteristics of scientific recovery suggestions and management measures for carbon loss events.

[0115] Step C12, based on the carbon loss map and the recovery strategy report, control the system to complete disaster assessment and risk warning.

[0116] It can be understood that the carbon loss map can visualize the spatial distribution of carbon loss with high spatial resolution, including the regional scope of loss, the degree of loss, and the types of affected ecosystems, quantify the lost carbon storage, while the recovery strategy report details the recovery objectives, priorities, specific technical means such as the selection of tree species for replanting, planting density, and soil improvement measures, as well as the expected recovery schedule and cost estimate, reduce the economic losses caused by ecosystem damage, and at the same time enhance the ecological service function of the ecosystem, and has significant social and economic value.

[0117] A carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning proposed in this embodiment obtains spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; extracts 3D point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; predicts the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and controls the system based on the carbon sink prediction result to complete disaster assessment and risk warning. It solves the technical problem of how to dynamically monitor carbon sinks through multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning technology. Compared with the prior art, this application obtains high-precision spatio-temporal fusion data, an improved vegetation point cloud inversion algorithm, and a dynamic carbon sink prediction hybrid model, and uses a spatio-temporal pyramid and an attention mechanism to achieve high-precision alignment and fusion of multi-source data, significantly improving the alignment accuracy of multi-source remote sensing data and reducing the fusion error. It extracts 3D point cloud biomass and determines vegetation parameter information through an improved PointNet++ network, and at the same time combines a carbon sink prediction hybrid model of a physical model and an AI module to predict the carbon sink change trend in real time, improving the accuracy and efficiency of carbon sink monitoring, and also enhancing the system's rapid response ability to environmental changes and ecosystem disturbances, so as to complete disaster assessment and risk warning, with significant environmental benefits and social value.

[0118] Based on the first embodiment of this application, in the second embodiment of this application, the same or similar content as the above-mentioned embodiment 1 can be referred to the above introduction and will not be repeated hereinafter.

[0119] In this embodiment, refer to Figure 2 , Figure 2 is a schematic flowchart provided for the second embodiment of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning of this application. Step S10 specifically includes steps S11 to S14:

[0120] Step S11, obtain optical images, point cloud remote sensing monitoring data, radar data, an adaptive sampling algorithm, a feature fusion algorithm, a biomass limitation loss algorithm, a light energy correction model, and a meteorological correction model;

[0121] It should be noted that the optical image reflects the characteristics that the optical image obtained by a multi-spectral or hyperspectral sensor can provide the reflection spectral information of vegetation, the point cloud remote sensing monitoring data reflects the characteristics of the three-dimensional spatial structure of vegetation, the radar data reflects the characteristics of the electromagnetic scattering of surface vegetation and other ground objects, the adaptive sampling algorithm reflects the characteristics of efficient processing of complex vegetation point cloud data, the feature fusion algorithm reflects the characteristics of collaborative processing of multi-source data, the biomass limitation loss algorithm reflects the characteristics of optimizing the accuracy of vegetation biomass estimation, the light energy correction model reflects the characteristics of a model for dynamically adjusting the photosynthesis efficiency of vegetation, and the meteorological correction model reflects the characteristics of a model for dynamically compensating the influence of meteorological factors.

[0122] For ease of understanding, taking the acquisition of optical images, point cloud remote sensing monitoring data, radar data, adaptive sampling algorithms, feature fusion algorithms, biomass limitation loss algorithms, light energy correction models, and meteorological correction models as examples for illustration, where the information acquisition device is the information acquisition module and the storage device is the memory.

[0123] The information acquisition module acquires optical images, such as optical images (Sentinel-2, 10m), acquires point cloud remote sensing monitoring data, such as LiDAR point clouds (drone, 0.1m), acquires radar data, such as radar data (Sentinel-1, 20m), acquires adaptive sampling algorithms, feature fusion algorithms, and biomass limitation loss algorithms, acquires a light energy correction model, the light energy correction model calculates the net primary productivity NPP based on the principle of light energy utilization efficiency, introduces the red edge band to correct the absorption rate of photosynthetically active radiation PAR, acquires a meteorological correction model, the meteorological correction model can use the long short-term memory network LSTM to learn the dynamic non-linear relationship between meteorological factors such as temperature, precipitation, and CO2 concentration and carbon sinks, and performs subsequent processing based on the optical images, point cloud remote sensing monitoring data, radar data, adaptive sampling algorithms, feature fusion algorithms, biomass limitation loss algorithms, light energy correction models, and meteorological correction models.

[0124] Step S12, extract spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and determine spatio-temporal fusion data;

[0125] It should be noted that the spatio-temporal fusion data reflects the characteristics of data after high-precision alignment and fusion of multi-source remote sensing data in spatio-temporal resolution.

[0126] For ease of understanding, taking the determination of spatio-temporal fusion data as an example for illustration, where the information acquisition device is the information acquisition module and the storage device is the memory.

[0127] The information acquisition module acquires optical images, such as optical images (Sentinel-2, 10m), acquires point cloud remote sensing monitoring data, such as LiDAR point clouds (drone, 0.1m), acquires radar data, such as radar data (Sentinel-1, 20m), extracts spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and constructs a spatio-temporal pyramid, and the spatio-temporal pyramid includes an optical image pyramid, a point cloud remote sensing monitoring data pyramid, and a radar data pyramid.

[0128] That is, construct a spatio-temporal pyramid, where the optical image pyramid is represented as:

[0129]

[0130] Where, Io is the original optical image, is the optical image after downsampling at the l-th layer, L is the number of pyramid layers, usually L = 3, Downsample is the downsampling operation, which reduces the image resolution proportionally.

[0131] The point cloud remote sensing monitoring data pyramid, that is, the LiDAR point cloud pyramid, is expressed as:

[0132]

[0133] where, is the LiDAR point cloud after voxelization at the l-th layer, r0 is the initial resolution, such as 1m, Voxelize is the voxelization operation, which converts the point cloud into a regular three-dimensional voxel network, resolution = 2 l ·r0 is the voxel resolution, and the resolution of the l-th layer is 2 l ·r0.

[0134] The radar data pyramid is expressed as:

[0135]

[0136] where, I r is the original radar image, scale = 2 l is the downsampling ratio, and the resolution of the l-th layer is 2 l times the original resolution.

[0137] Extract spatio-temporal features based on the optical image pyramid, the point cloud remote sensing monitoring data pyramid, and the radar data pyramid, and calculate the attention weight set, that is, perform spatio-temporal feature extraction. Among them, the extraction of spatio-temporal features of optical images is expressed as:

[0138]

[0139] where, is the optical image feature at the l-th layer, is the optical image after downsampling at the l-th layer, and CNN is a convolutional neural network for extracting the spatial features of the image.

[0140] The extraction of spatio-temporal features of point cloud remote sensing monitoring data is expressed as:

[0141]

[0142] where, is the LiDAR point cloud feature at the l-th layer, is the LiDAR point cloud after voxelization at the l-th layer, and PointNet is a point cloud processing neural network for extracting the three-dimensional structural features of the point cloud.

[0143] The spatio-temporal features of radar data are extracted and represented as:

[0144]

[0145] Among them, is the radar image feature of the l-th layer, is the radar image after downsampling at the l-th layer, and CNN is a convolutional neural network used to extract the spatial features of the radar image.

[0146] The attention weight set is calculated using the extracted spatio-temporal features of the optical image, the spatio-temporal features of the point cloud remote sensing monitoring data, and the spatio-temporal features of the radar data. The attention weight set may include the optical image weight, the point cloud remote sensing monitoring data weight, and the radar data weight.

[0147] Among them, the optical image weight is represented as:

[0148]

[0149] Among them, is the attention weight of the optical image feature of the l-th layer, is the learnable weight matrix of the optical image feature, is the bias top of the optical image feature, and Softmax is a normalization exponential function that maps the weights to the interval [0, 1].

[0150] The point cloud remote sensing monitoring data weight is represented as:

[0151]

[0152] Among them, is the attention weight of the LiDAR point cloud feature of the l-th layer, is the learnable weight matrix of the LiDAR point cloud feature, is the bias top of the LiDAR point cloud feature.

[0153] The radar data weight is represented as:

[0154]

[0155] Among them, is the attention weight of the radar image feature of the l-th layer, is the learnable weight matrix of the radar image feature, is the bias top of the radar image feature.

[0156] Based on the attention weight set, the spatio-temporal coordinate system is aligned to calculate the spatio-temporally aligned data, that is, the spatio-temporal coordinate system is aligned using the attention alignment method, and the attention alignment method is represented as:

[0157]

[0158] Thus, spatio-temporal alignment data can be calculated and expressed as:

[0159]

[0160] Wherein, is the fused feature after alignment at the l-th layer, is the weighted contribution of the optical image feature, is the weighted contribution of the LiDAR point cloud feature, is the weighted contribution of the radar image feature.

[0161] Obtain the cascading operation mode and the loss function. The loss function includes a reconstruction loss function, an alignment loss function, and a total loss function. Based on the spatio-temporal alignment data, perform upsampling to extract super-resolution details, and determine the upsampled data, that is, perform feature upsampling, which is expressed as:

[0162]

[0163] Wherein, is the fused feature after alignment at the l-th layer, is the feature after upsampling at the l-th layer, Deconv is the deconvolution operation, scale = 2 l is the upsampling ratio, and the upsampling multiple at the l-th layer is 2 l .

[0164] Calculate the cascaded data based on the cascading operation mode and the upsampled data, that is, perform multi-level cascading to calculate the cascaded data, which is expressed as:

[0165]

[0166] Wherein, F final is the final high-resolution fused feature, W l is the learnable weight matrix of the l-th layer feature, is the feature after upsampling at the l-th layer, b is the bias term, and L is the total number of pyramid layers, usually L = 3.

[0167] Calculate the spatio-temporal fusion data based on the reconstruction loss function, the alignment loss function, the total loss function, and the cascaded data, that is, use the reconstruction loss function to calculate the reconstruction loss, which is expressed as:

[0168]

[0169] Wherein, is the reconstruction loss, F final is the final high-resolution fused feature, F gt is the high-resolution ground truth data, formula It is the square of the L2 norm and is used to measure the difference between the predicted value and the true value.

[0170] The alignment loss is calculated using the alignment loss function and is expressed as:

[0171]

[0172] Where, is the alignment loss, is the fused feature after alignment at the l-th layer, is the true value data at the l-th layer.

[0173] The total loss is calculated using the total loss function and is expressed as:

[0174]

[0175] Where, λ1 is the weight coefficient of the reconstruction loss, and λ2 is the weight coefficient of the alignment loss.

[0176] The spatio-temporal fusion data can be obtained by subtracting the total loss from the cascaded data, and subsequent processing is performed based on the spatio-temporal fusion data.

[0177] In a feasible implementation manner, step S10 may include steps D11 to D13:

[0178] Step D11, extracting spatio-temporal fusion details based on the optical image, the point cloud remote sensing monitoring data, and the radar data, and constructing a spatio-temporal pyramid, where the spatio-temporal pyramid includes an optical image pyramid, a point cloud remote sensing monitoring data pyramid, and a radar data pyramid;

[0179] It should be noted that the spatio-temporal pyramid reflects the characteristics of spatio-temporal feature extraction and hierarchical organization of multi-source remote sensing data at different resolution levels.

[0180] It can be understood that the spatio-temporal pyramid constructs the optical image, the LiDAR point cloud, and the radar data into a multi-resolution pyramid structure through hierarchical sampling. Each layer corresponds to a different resolution, thereby capturing multi-scale spatio-temporal information from the global to the local. It not only preserves the original features of the data but also enhances the expression ability of detailed information through hierarchical processing, significantly improving the accuracy of the fused data and the spatial alignment accuracy, and enhancing the performance and practicality of the system.

[0181] Step D12, extracting spatio-temporal features based on the optical image pyramid, the point cloud remote sensing monitoring data pyramid, and the radar data pyramid, calculating the attention weight set, aligning the spatio-temporal coordinate system based on the attention weight set, and calculating the spatio-temporal alignment data;

[0182] It should be noted that the spatio-temporal alignment data reflects the characteristics of multi-source remote sensing data after high-precision alignment in spatio-temporal coordinates.

[0183] It can be understood that the spatio-temporal alignment data can effectively eliminate the spatio-temporal deviations between different data sources, ensure the consistency of multi-source data in spatial position and time dimension, retain the original information of each data source, and dynamically adjust the weights of different features through the attention mechanism to achieve high-precision spatio-temporal alignment.

[0184] Step D13, perform super-resolution reconstruction based on the spatio-temporal alignment data and calculate the reconstruction loss to obtain spatio-temporal fusion data.

[0185] It can be understood that the spatio-temporal fusion data eliminates the differences in spatio-temporal resolution of multi-source data, and significantly improves the accuracy and spatial alignment accuracy of the fusion data.

[0186] In a feasible implementation manner, step D13 may include steps E11 to E14:

[0187] Step E11, obtain the cascade operation mode and the loss function, where the loss function includes a reconstruction loss function, an alignment loss function, and a total loss function;

[0188] It should be noted that the cascade operation mode reflects the characteristics of retaining the detailed information from low resolution to high resolution and gradually performing optimization operations, and the loss function reflects the characteristics of quantitatively evaluating the quality loss of the fusion data.

[0189] It can be understood that the cascade operation mode is suitable for processing the complex structure of multi-source remote sensing data, and can gradually improve the resolution and accuracy of the data at different levels, while the loss function is used to measure the difference between the fusion data and the real data, the accuracy of spatio-temporal alignment, and the overall fusion effect. By minimizing the loss, it can ensure that the spatio-temporal fusion data reaches the optimal in terms of accuracy and alignment accuracy.

[0190] Step E12, perform upsampling on the spatio-temporal alignment data to extract super-resolution details and determine the upsampled data;

[0191] It should be noted that the upsampled data reflects the characteristics of extracting and restoring higher-resolution detailed data from the low-resolution spatio-temporal alignment data using upsampling technology.

[0192] It can be understood that the upsampled data can make the detailed information more abundant and clear, significantly improve the spatial resolution of the spatio-temporal fusion data, thus more accurately reflecting the vegetation structure and terrain features, and improving the accuracy of vegetation parameter inversion and the reliability of carbon sink prediction.

[0193] Step E13, calculate the cascaded data based on the cascaded operation mode and the upsampled data;

[0194] It should be noted that the cascaded data reflects the characteristics of the data obtained by gradually fusing the upsampled data at different levels.

[0195] It can be understood that the cascaded data can effectively reduce the information loss caused by single-scale processing, while ensuring the spatio-temporal consistency of the data at different levels, improving the resolution and detail representation ability of the final fused data, and significantly enhancing the accuracy and reliability of the carbon sink monitoring system.

[0196] Step E14, calculate the spatio-temporal fused data based on the reconstruction loss function, the alignment loss function, the total loss function, and the cascaded data.

[0197] It can be understood that based on the spatio-temporal fused data, the resolution can be increased to 0.1 meters, and the spatial alignment error can be controlled within 5%, better supporting high-precision carbon sink monitoring and rapid response after ecosystem disturbance.

[0198] Step S13, determine the vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass constraint loss algorithm;

[0199] It should be noted that the vegetation point cloud inversion algorithm reflects the characteristics of high-precision segmentation of complex vegetation structures and biomass inversion capabilities.

[0200] It can be understood that the vegetation point cloud inversion algorithm can efficiently segment the single-tree canopy and extract three-dimensional parameters of the vegetation, such as tree height, crown diameter, and volume, through an improved PointNet++ network, combined with hierarchical adaptive sampling, multi-scale feature fusion, and a biomass geometric constraint loss function. At the same time, it adapts to complex canopy structures, significantly reducing the biomass estimation error and improving the processing efficiency and accuracy of the vegetation point cloud.

[0201] For ease of understanding, taking the determination of the vegetation point cloud inversion algorithm as an example, the information acquisition device is the information acquisition module, and the storage device is the memory.

[0202] The information acquisition module obtains the adaptive sampling algorithm, that is, using the adaptive sampling algorithm, hierarchical adaptive sampling HAS, dynamically adjusts the sampling density according to the point cloud density gradient, such as a small sampling interval in the dense canopy area and a large interval in the sparse ground area, and combines LiDAR elevation data to preferentially retain the key points in the high-vegetation area, expressed as:

[0203]

[0204] Among them, S iis the sampling weight of the i-th point, α is the weight coefficient of the local density term, σ i is the local density of the i-th point, β is the weight coefficient of the height term, H i is the height value of the i-th point.

[0205] Obtain the feature fusion algorithm, that is, construct nested local regions at each level, such as spheres with radii of 0.5m, 1m, and 2m, extract multi-scale features in parallel, use the attention mechanism to dynamically weight and fuse different-scale features, and the multi-scale feature fusion MSFF is expressed as:

[0206] F fusion = Attention(F 0.5m , F 1m , F 2m )

[0207] Among them, F fusion is the fused multi-scale feature, Attention is the attention mechanism used to dynamically weight features of different scales, F 0.5m is the feature at the 0.5m scale, F 1m is the feature at the 1m scale, F 2m is the feature at the 2m scale.

[0208] Obtain the biomass limit loss algorithm, that is, add a biomass geometric limit term to the loss function, and the biomass geometric limit loss is expressed as:

[0209]

[0210] Among them, is the total loss, is the segmentation loss used to measure the accuracy of point cloud segmentation, λ is the weight coefficient of the geometric limit term, H pred is the predicted tree height, k is an empirical coefficient related to the tree species, D pred is the predicted crown diameter.

[0211] Determine the vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass limit loss algorithm, and perform subsequent processing based on the vegetation point cloud inversion algorithm.

[0212] Step S14, determine the carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model.

[0213] It can be understood that the carbon sink prediction hybrid model couples the physical model with the AI module, which can not only accurately reflect the dynamic relationship between the carbon sink and environmental factors, but also support real-time optimization and online learning, can quickly adapt to environmental changes and ecosystem disturbances, and significantly improves the accuracy and adaptability of carbon sink dynamic prediction.

[0214] For ease of understanding, an example of determining a carbon sink prediction hybrid model is used for illustration, where the information collection device is an information collection module and the storage device is a memory.

[0215] The information collection module obtains a light energy correction model, that is, calculates the net primary productivity NPP based on the principle of light energy utilization rate, introduces the red edge band to correct the absorption rate of photosynthetically active radiation PAR, and obtains a meteorological correction model, that is, uses the long short-term memory network LSTM to learn the dynamic non-linear relationship between meteorological factors such as temperature, precipitation, and CO2 concentration and the carbon sink. Among them, the light energy correction model and the meteorological correction model support online incremental learning, dynamically optimize parameters through real-time data of the flux tower, adapt to environmental mutations, and the flux tower data calibrates the model in real time, and the error is fed back to the online learning module to obtain a carbon sink prediction hybrid model, and subsequent processing is performed based on the carbon sink prediction hybrid model.

[0216] A carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning proposed in this embodiment obtains optical images, point cloud remote sensing monitoring data, radar data, an adaptive sampling algorithm, a feature fusion algorithm, a biomass limit loss algorithm, a light energy correction model, and a meteorological correction model; extracts spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data to determine spatio-temporal fusion data; determines a vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass limit loss algorithm; determines a carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model. It solves the technical problem of how to perform high-precision carbon sink monitoring through multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning technology. Compared with the prior art, this application integrates optical images, point cloud data, radar data, and related algorithms, realizes high-precision spatio-temporal fusion, vegetation parameter inversion, and carbon sink dynamic prediction, greatly improves the spatio-temporal fusion accuracy, eliminates the resolution differences of multi-source data, significantly reduces the vegetation parameter inversion error, improves the accuracy of biomass estimation, and can respond to environmental changes in real time, complete disaster assessment and risk warning, and has significant environmental benefits and social value.

[0217] This application also provides a carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. Please refer to Figure 3 , the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning includes:

[0218] An acquisition module 10 for acquiring spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model;

[0219] A processing module 20 for extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information;

[0220] The execution module 30 is configured to predict the changing trend of carbon sinks based on the vegetation parameter information and the carbon sink prediction hybrid model, determine the carbon sink prediction result, and control the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

[0221] The acquisition module 10 is further configured to acquire optical images, point cloud remote sensing monitoring data, radar data, an adaptive sampling algorithm, a feature fusion algorithm, a biomass limit loss algorithm, a light energy correction model, and a meteorological correction model;

[0222] Extract spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data to determine spatio-temporal fusion data;

[0223] Determine a vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass limit loss algorithm;

[0224] Determine a carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model.

[0225] The acquisition module 10 is further configured to extract spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and construct a spatio-temporal pyramid, where the spatio-temporal pyramid includes an optical image pyramid, a point cloud remote sensing monitoring data pyramid, and a radar data pyramid;

[0226] Extract spatio-temporal features based on the optical image pyramid, the point cloud remote sensing monitoring data pyramid, and the radar data pyramid, calculate an attention weight set, align the spatio-temporal coordinate system based on the attention weight set, and calculate spatio-temporal alignment data;

[0227] Perform super-resolution reconstruction on the spatio-temporal alignment data and calculate a reconstruction loss to obtain spatio-temporal fusion data.

[0228] The acquisition module 10 is further configured to acquire a cascaded operation mode and a loss function, where the loss function includes a reconstruction loss function, an alignment loss function, and a total loss function;

[0229] Perform upsampling on the spatio-temporal alignment data to extract super-resolution details to determine upsampled data;

[0230] Calculate cascaded data based on the cascaded operation mode and the upsampled data;

[0231] Calculate spatio-temporal fusion data based on the reconstruction loss function, the alignment loss function, the total loss function, and the cascaded data.

[0232] The processing module 20 is further configured to retrieve three-dimensional point cloud biomass based on the adaptive sampling algorithm in the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, calculate the vegetation sampling weights, and determine the key points in the vegetation area, where the three-dimensional point cloud biomass includes ground biomass, underground biomass, and litter;

[0233] Construct a nested local area based on the feature fusion algorithm in the vegetation point cloud inversion algorithm and the key points in the vegetation area, and determine the vegetation fusion data;

[0234] Calculate the biomass growth loss based on the biomass limitation loss algorithm in the vegetation point cloud inversion algorithm and the vegetation fusion data to obtain the vegetation parameter information.

[0235] The execution module 30 is further configured to obtain the flux tower data;

[0236] Adjust the vegetation parameter information based on the flux tower data to determine the optimized vegetation parameter information;

[0237] Input the optimized vegetation parameter information into the carbon sink prediction hybrid model for training to determine the carbon sink prediction hybrid optimization model;

[0238] Quantify the defects in the remote sensing data and predict the carbon sink change trend based on the carbon sink prediction hybrid optimization model to obtain the carbon sink prediction result.

[0239] The execution module 30 is further configured to analyze the carbon sink prediction result to identify the disaster risk, and determine the carbon loss map and the restoration strategy report;

[0240] Control the system based on the carbon loss map and the restoration strategy report to complete the disaster assessment and risk warning.

[0241] The carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning provided by this application adopts the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning in the above embodiment, and can solve the technical problem of how to dynamically perform high-precision carbon sink monitoring through multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning technology. Compared with the prior art, the beneficial effects of the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning provided by this application are the same as those of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning provided by the above embodiment, and other technical features in the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning are the same as the features disclosed in the method of the above embodiment, and will not be elaborated here.

[0242] The present application provides a carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. The carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning in Embodiment 1 above.

[0243] Reference is made below Figure 4 , which shows a schematic structural diagram of a carbon sink dynamic prediction device suitable for implementing the embodiments of the present application based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning. The carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 4 The carbon sink dynamic prediction device shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present application.

[0244] As Figure 4As shown, the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the ROM (Read Only Memory) 1002 or the program loaded from the storage device 1003 into the RAM (Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning are also stored. The processing device 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. The input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems may be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a carbon sink dynamic prediction device with various systems, it should be understood that it is not required to implement or have all the shown systems. Instead, more or fewer systems may be implemented or had.

[0245] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0246] The carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning provided by this application adopts the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning in the above-mentioned embodiment, and can solve the technical problem of how to dynamically monitor carbon sinks with high precision through multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning technologies. Compared with the prior art, the beneficial effects of the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning provided by this application are the same as those of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning provided by the above-mentioned embodiment, and other technical features in the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0247] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0248] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all of them should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0249] This application provides a computer-readable storage medium with computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning in the above-mentioned embodiment.

[0250] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or components, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM) or flash memory, optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device, or component. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0251] The above computer-readable storage medium can be included in the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning; it can also exist independently without being assembled into the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning.

[0252] The above computer-readable storage medium carries one or more programs. When the above one or more programs are executed by the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning, the carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning is enabled to: obtain spatio-temporal fusion data, vegetation point cloud inversion algorithm, and carbon sink prediction hybrid model; extract 3D point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; predict the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and control the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

[0253] Computer program code for performing the operations of this application can be written in one or more programming languages or combinations thereof. The above-mentioned programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0254] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0255] The modules described in the embodiments of this application can be implemented in software or in hardware. Among them, the name of the module does not constitute a limitation on the unit itself in some cases.

[0256] The readable storage medium provided by this application is a computer-readable storage medium. The computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning, and can solve the technical problem of how to dynamically monitor carbon sinks with high precision through multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning technologies. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by this application are the same as those of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning provided by the above embodiments, and will not be elaborated here.

[0257] The above are only some embodiments of this application, and thus do not limit the patent scope of this application. Any equivalent structural transformation made under the technical concept of this application by using the content of the specification and drawings of this application, or any direct / indirect application in other related technical fields, is included in the patent protection scope of this application.

Claims

1. A dynamic carbon sink prediction method based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning, characterized in that, The method described above includes: Obtaining spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; Extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and determining vegetation parameter information; Predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model, determining the carbon sink prediction result, and controlling the system based on the carbon sink prediction result to complete disaster assessment and risk warning.

2. The method according to claim 1, wherein The step of obtaining the spatio-temporal fusion data, the vegetation point cloud inversion algorithm, and the carbon sink prediction hybrid model includes: Obtaining optical images, point cloud remote sensing monitoring data, radar data, an adaptive sampling algorithm, a feature fusion algorithm, a biomass constraint loss algorithm, a light energy correction model, and a meteorological correction model; Extracting spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and determining spatio-temporal fusion data; Determining the vegetation point cloud inversion algorithm based on the adaptive sampling algorithm, the feature fusion algorithm, and the biomass constraint loss algorithm; Determining the carbon sink prediction hybrid model based on the light energy correction model and the meteorological correction model.

3. The method according to claim 2, characterized in that The step of extracting spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and determining spatio-temporal fusion data includes: Extracting spatio-temporal fusion details based on the optical images, the point cloud remote sensing monitoring data, and the radar data, and constructing a spatio-temporal pyramid, where the spatio-temporal pyramid includes an optical image pyramid, a point cloud remote sensing monitoring data pyramid, and a radar data pyramid; Extracting spatio-temporal features based on the optical image pyramid, the point cloud remote sensing monitoring data pyramid, and the radar data pyramid, calculating an attention weight set, aligning the spatio-temporal coordinate system based on the attention weight set, and calculating spatio-temporally aligned data; Performing super-resolution reconstruction on the spatio-temporally aligned data and calculating the reconstruction loss to obtain spatio-temporal fusion data.

4. The method according to claim 3, characterized in that, The step of performing super-resolution reconstruction on the spatio-temporally aligned data and calculating the reconstruction loss to obtain spatio-temporal fusion data includes: Obtaining a cascaded operation method and a loss function, where the loss function includes a reconstruction loss function, an alignment loss function, and a total loss function; Performing upsampling on the spatio-temporally aligned data to extract super-resolution details, and determining upsampled data; Calculating cascaded data based on the cascaded operation method and the upsampled data; Calculating spatio-temporal fusion data based on the reconstruction loss function, the alignment loss function, the total loss function, and the cascaded data.

5. The method according to claim 1, wherein The step of extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and determining vegetation parameter information includes: Extracting three-dimensional point cloud biomass based on the adaptive sampling algorithm in the vegetation point cloud inversion algorithm and the spatio-temporal fusion data, and calculating the vegetation sampling weight to determine the key points in the vegetation area, where the three-dimensional point cloud biomass includes ground biomass, underground biomass, and litter; Constructing a nested local area based on the feature fusion algorithm in the vegetation point cloud inversion algorithm and the key points in the vegetation area, and determining vegetation fusion data; Calculate the biomass growth loss based on the biomass limitation loss algorithm in the vegetation point cloud inversion algorithm and the vegetation fusion data to obtain vegetation parameter information.

6. The method according to claim 1, characterized in that The steps of predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result include: Obtain flux tower data; Adjust the vegetation parameter information based on the flux tower data to determine the optimized vegetation parameter information; Input the optimized vegetation parameter information into the carbon sink prediction hybrid model for training to determine the carbon sink prediction hybrid optimization model; Quantify the remote sensing data defects based on the carbon sink prediction hybrid optimization model and predict the carbon sink change trend to obtain the carbon sink prediction result.

7. The method according to claim 1, characterized in that The steps of controlling the system to complete disaster assessment and risk warning based on the carbon sink prediction result include: Analyze the carbon sink prediction result to identify disaster risks and determine the carbon loss map and recovery strategy report; Based on the carbon loss map and the recovery strategy report, control the system to complete disaster assessment and risk warning.

8. A carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning, characterized in that, The device includes: An acquisition module for acquiring spatio-temporal fusion data, a vegetation point cloud inversion algorithm, and a carbon sink prediction hybrid model; A processing module for extracting three-dimensional point cloud biomass based on the vegetation point cloud inversion algorithm and the spatio-temporal fusion data to determine vegetation parameter information; An execution module for predicting the carbon sink change trend based on the vegetation parameter information and the carbon sink prediction hybrid model to determine the carbon sink prediction result, and controlling the system to complete disaster assessment and risk warning based on the carbon sink prediction result.

9. A carbon sink dynamic prediction device based on multi-source remote sensing spatio-temporal fusion and 3D point cloud deep learning, characterized in that, The device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium, and when the computer program is executed by the processor, it implements the steps of the carbon sink dynamic prediction method based on multi-source remote sensing spatio-temporal fusion and three-dimensional point cloud deep learning as described in any one of claims 1 to 7.

Images